summaryrefslogtreecommitdiffstats
path: root/src/nspawn
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-07 15:35:18 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-07 15:35:18 +0000
commitb750101eb236130cf056c675997decbac904cc49 (patch)
treea5df1a06754bdd014cb975c051c83b01c9a97532 /src/nspawn
parentInitial commit. (diff)
downloadsystemd-b750101eb236130cf056c675997decbac904cc49.tar.xz
systemd-b750101eb236130cf056c675997decbac904cc49.zip
Adding upstream version 252.22.upstream/252.22upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to '')
-rw-r--r--src/nspawn/fuzz-nspawn-oci.c26
-rw-r--r--src/nspawn/fuzz-nspawn-oci.options2
-rw-r--r--src/nspawn/fuzz-nspawn-settings.c26
-rw-r--r--src/nspawn/fuzz-nspawn-settings.options2
-rw-r--r--src/nspawn/meson.build85
-rw-r--r--src/nspawn/nspawn-bind-user.c474
-rw-r--r--src/nspawn/nspawn-bind-user.h29
-rw-r--r--src/nspawn/nspawn-cgroup.c606
-rw-r--r--src/nspawn/nspawn-cgroup.h14
-rw-r--r--src/nspawn/nspawn-creds.c25
-rw-r--r--src/nspawn/nspawn-creds.h12
-rw-r--r--src/nspawn/nspawn-def.h9
-rw-r--r--src/nspawn/nspawn-expose-ports.c220
-rw-r--r--src/nspawn/nspawn-expose-ports.h27
-rw-r--r--src/nspawn/nspawn-gperf.gperf82
-rw-r--r--src/nspawn/nspawn-mount.c1338
-rw-r--r--src/nspawn/nspawn-mount.h69
-rw-r--r--src/nspawn/nspawn-network.c745
-rw-r--r--src/nspawn/nspawn-network.h23
-rw-r--r--src/nspawn/nspawn-oci.c2207
-rw-r--r--src/nspawn/nspawn-oci.h6
-rw-r--r--src/nspawn/nspawn-patch-uid.c477
-rw-r--r--src/nspawn/nspawn-patch-uid.h7
-rw-r--r--src/nspawn/nspawn-register.c361
-rw-r--r--src/nspawn/nspawn-register.h14
-rw-r--r--src/nspawn/nspawn-seccomp.c260
-rw-r--r--src/nspawn/nspawn-seccomp.h6
-rw-r--r--src/nspawn/nspawn-settings.c957
-rw-r--r--src/nspawn/nspawn-settings.h287
-rw-r--r--src/nspawn/nspawn-setuid.c245
-rw-r--r--src/nspawn/nspawn-setuid.h5
-rw-r--r--src/nspawn/nspawn-stub-pid1.c198
-rw-r--r--src/nspawn/nspawn-stub-pid1.h6
-rw-r--r--src/nspawn/nspawn-util.c72
-rw-r--r--src/nspawn/nspawn-util.h4
-rw-r--r--src/nspawn/nspawn.c5913
-rw-r--r--src/nspawn/nspawn.h7
-rw-r--r--src/nspawn/test-nspawn-tables.c11
-rw-r--r--src/nspawn/test-nspawn-util.c22
-rw-r--r--src/nspawn/test-patch-uid.c43
40 files changed, 14922 insertions, 0 deletions
diff --git a/src/nspawn/fuzz-nspawn-oci.c b/src/nspawn/fuzz-nspawn-oci.c
new file mode 100644
index 0000000..5d03834
--- /dev/null
+++ b/src/nspawn/fuzz-nspawn-oci.c
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "alloc-util.h"
+#include "fd-util.h"
+#include "fuzz.h"
+#include "nspawn-oci.h"
+
+int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
+ _cleanup_fclose_ FILE *f = NULL;
+ _cleanup_(settings_freep) Settings *s = NULL;
+
+ if (outside_size_range(size, 0, 65536))
+ return 0;
+
+ f = data_to_file(data, size);
+ assert_se(f);
+
+ /* We don't want to fill the logs with messages about parse errors.
+ * Disable most logging if not running standalone */
+ if (!getenv("SYSTEMD_LOG_LEVEL"))
+ log_set_max_level(LOG_CRIT);
+
+ (void) oci_load(f, "/dev/null", &s);
+
+ return 0;
+}
diff --git a/src/nspawn/fuzz-nspawn-oci.options b/src/nspawn/fuzz-nspawn-oci.options
new file mode 100644
index 0000000..678d526
--- /dev/null
+++ b/src/nspawn/fuzz-nspawn-oci.options
@@ -0,0 +1,2 @@
+[libfuzzer]
+max_len = 65536
diff --git a/src/nspawn/fuzz-nspawn-settings.c b/src/nspawn/fuzz-nspawn-settings.c
new file mode 100644
index 0000000..40ee784
--- /dev/null
+++ b/src/nspawn/fuzz-nspawn-settings.c
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "alloc-util.h"
+#include "fd-util.h"
+#include "fuzz.h"
+#include "nspawn-settings.h"
+
+int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
+ _cleanup_fclose_ FILE *f = NULL;
+ _cleanup_(settings_freep) Settings *s = NULL;
+
+ if (outside_size_range(size, 0, 65536))
+ return 0;
+
+ f = data_to_file(data, size);
+ assert_se(f);
+
+ /* We don't want to fill the logs with messages about parse errors.
+ * Disable most logging if not running standalone */
+ if (!getenv("SYSTEMD_LOG_LEVEL"))
+ log_set_max_level(LOG_CRIT);
+
+ (void) settings_load(f, "/dev/null", &s);
+
+ return 0;
+}
diff --git a/src/nspawn/fuzz-nspawn-settings.options b/src/nspawn/fuzz-nspawn-settings.options
new file mode 100644
index 0000000..678d526
--- /dev/null
+++ b/src/nspawn/fuzz-nspawn-settings.options
@@ -0,0 +1,2 @@
+[libfuzzer]
+max_len = 65536
diff --git a/src/nspawn/meson.build b/src/nspawn/meson.build
new file mode 100644
index 0000000..f6cca70
--- /dev/null
+++ b/src/nspawn/meson.build
@@ -0,0 +1,85 @@
+# SPDX-License-Identifier: LGPL-2.1-or-later
+
+libnspawn_core_sources = files(
+ 'nspawn-bind-user.c',
+ 'nspawn-bind-user.h',
+ 'nspawn-cgroup.c',
+ 'nspawn-cgroup.h',
+ 'nspawn-creds.c',
+ 'nspawn-creds.h',
+ 'nspawn-def.h',
+ 'nspawn-expose-ports.c',
+ 'nspawn-expose-ports.h',
+ 'nspawn-mount.c',
+ 'nspawn-mount.h',
+ 'nspawn-network.c',
+ 'nspawn-network.h',
+ 'nspawn-oci.c',
+ 'nspawn-oci.h',
+ 'nspawn-patch-uid.c',
+ 'nspawn-patch-uid.h',
+ 'nspawn-register.c',
+ 'nspawn-register.h',
+ 'nspawn-seccomp.c',
+ 'nspawn-seccomp.h',
+ 'nspawn-settings.c',
+ 'nspawn-settings.h',
+ 'nspawn-setuid.c',
+ 'nspawn-setuid.h',
+ 'nspawn-stub-pid1.c',
+ 'nspawn-stub-pid1.h',
+ 'nspawn-util.c',
+ 'nspawn-util.h',
+ 'nspawn.h',
+)
+
+nspawn_gperf_c = custom_target(
+ 'nspawn-gperf.c',
+ input : 'nspawn-gperf.gperf',
+ output : 'nspawn-gperf.c',
+ command : [gperf, '@INPUT@', '--output-file', '@OUTPUT@'])
+
+libnspawn_core_sources += [nspawn_gperf_c]
+
+libnspawn_core = static_library(
+ 'nspawn-core',
+ libnspawn_core_sources,
+ include_directories : includes,
+ dependencies : [libacl,
+ libseccomp,
+ libselinux],
+ build_by_default : false)
+
+systemd_nspawn_sources = files('nspawn.c')
+
+############################################################
+
+tests += [
+ [files('test-nspawn-tables.c'),
+ [libnspawn_core,
+ libshared],
+ [libseccomp]],
+
+ [files('test-nspawn-util.c'),
+ [libnspawn_core,
+ libshared],
+ [libseccomp]],
+
+ [files('test-patch-uid.c'),
+ [libnspawn_core,
+ libshared],
+ [libacl],
+ [], '', 'manual'],
+]
+
+fuzzers += [
+ [files('fuzz-nspawn-settings.c'),
+ [libshared,
+ libnspawn_core],
+ [libseccomp]],
+
+ [files('fuzz-nspawn-oci.c'),
+ [libshared,
+ libnspawn_core],
+ [libseccomp]],
+]
diff --git a/src/nspawn/nspawn-bind-user.c b/src/nspawn/nspawn-bind-user.c
new file mode 100644
index 0000000..7a7a19f
--- /dev/null
+++ b/src/nspawn/nspawn-bind-user.c
@@ -0,0 +1,474 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "chase-symlinks.h"
+#include "fd-util.h"
+#include "fileio.h"
+#include "format-util.h"
+#include "nspawn-bind-user.h"
+#include "nspawn.h"
+#include "path-util.h"
+#include "user-util.h"
+#include "userdb.h"
+
+static int check_etc_passwd_collisions(
+ const char *directory,
+ const char *name,
+ uid_t uid) {
+
+ _cleanup_fclose_ FILE *f = NULL;
+ int r;
+
+ assert(directory);
+ assert(name || uid_is_valid(uid));
+
+ r = chase_symlinks_and_fopen_unlocked("/etc/passwd", directory, CHASE_PREFIX_ROOT, "re", NULL, &f);
+ if (r == -ENOENT)
+ return 0; /* no user database? then no user, hence no collision */
+ if (r < 0)
+ return log_error_errno(r, "Failed to open /etc/passwd of container: %m");
+
+ for (;;) {
+ struct passwd *pw;
+
+ r = fgetpwent_sane(f, &pw);
+ if (r < 0)
+ return log_error_errno(r, "Failed to iterate through /etc/passwd of container: %m");
+ if (r == 0) /* EOF */
+ return 0; /* no collision */
+
+ if (name && streq_ptr(pw->pw_name, name))
+ return 1; /* name collision */
+ if (uid_is_valid(uid) && pw->pw_uid == uid)
+ return 1; /* UID collision */
+ }
+}
+
+static int check_etc_group_collisions(
+ const char *directory,
+ const char *name,
+ gid_t gid) {
+
+ _cleanup_fclose_ FILE *f = NULL;
+ int r;
+
+ assert(directory);
+ assert(name || gid_is_valid(gid));
+
+ r = chase_symlinks_and_fopen_unlocked("/etc/group", directory, CHASE_PREFIX_ROOT, "re", NULL, &f);
+ if (r == -ENOENT)
+ return 0; /* no group database? then no group, hence no collision */
+ if (r < 0)
+ return log_error_errno(r, "Failed to open /etc/group of container: %m");
+
+ for (;;) {
+ struct group *gr;
+
+ r = fgetgrent_sane(f, &gr);
+ if (r < 0)
+ return log_error_errno(r, "Failed to iterate through /etc/group of container: %m");
+ if (r == 0)
+ return 0; /* no collision */
+
+ if (name && streq_ptr(gr->gr_name, name))
+ return 1; /* name collision */
+ if (gid_is_valid(gid) && gr->gr_gid == gid)
+ return 1; /* gid collision */
+ }
+}
+
+static int convert_user(
+ const char *directory,
+ UserRecord *u,
+ GroupRecord *g,
+ uid_t allocate_uid,
+ UserRecord **ret_converted_user,
+ GroupRecord **ret_converted_group) {
+
+ _cleanup_(group_record_unrefp) GroupRecord *converted_group = NULL;
+ _cleanup_(user_record_unrefp) UserRecord *converted_user = NULL;
+ _cleanup_free_ char *h = NULL;
+ JsonVariant *p, *hp = NULL;
+ int r;
+
+ assert(u);
+ assert(g);
+ assert(u->gid == g->gid);
+
+ r = check_etc_passwd_collisions(directory, u->user_name, UID_INVALID);
+ if (r < 0)
+ return r;
+ if (r > 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EBUSY),
+ "Sorry, the user '%s' already exists in the container.", u->user_name);
+
+ r = check_etc_group_collisions(directory, g->group_name, GID_INVALID);
+ if (r < 0)
+ return r;
+ if (r > 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EBUSY),
+ "Sorry, the group '%s' already exists in the container.", g->group_name);
+
+ h = path_join("/run/host/home/", u->user_name);
+ if (!h)
+ return log_oom();
+
+ /* Acquire the source hashed password array as-is, so that it retains the JSON_VARIANT_SENSITIVE flag */
+ p = json_variant_by_key(u->json, "privileged");
+ if (p)
+ hp = json_variant_by_key(p, "hashedPassword");
+
+ r = user_record_build(
+ &converted_user,
+ JSON_BUILD_OBJECT(
+ JSON_BUILD_PAIR("userName", JSON_BUILD_STRING(u->user_name)),
+ JSON_BUILD_PAIR("uid", JSON_BUILD_UNSIGNED(allocate_uid)),
+ JSON_BUILD_PAIR("gid", JSON_BUILD_UNSIGNED(allocate_uid)),
+ JSON_BUILD_PAIR_CONDITION(u->disposition >= 0, "disposition", JSON_BUILD_STRING(user_disposition_to_string(u->disposition))),
+ JSON_BUILD_PAIR("homeDirectory", JSON_BUILD_STRING(h)),
+ JSON_BUILD_PAIR("service", JSON_BUILD_CONST_STRING("io.systemd.NSpawn")),
+ JSON_BUILD_PAIR_CONDITION(!strv_isempty(u->hashed_password), "privileged", JSON_BUILD_OBJECT(
+ JSON_BUILD_PAIR("hashedPassword", JSON_BUILD_VARIANT(hp))))));
+ if (r < 0)
+ return log_error_errno(r, "Failed to build container user record: %m");
+
+ r = group_record_build(
+ &converted_group,
+ JSON_BUILD_OBJECT(
+ JSON_BUILD_PAIR("groupName", JSON_BUILD_STRING(g->group_name)),
+ JSON_BUILD_PAIR("gid", JSON_BUILD_UNSIGNED(allocate_uid)),
+ JSON_BUILD_PAIR_CONDITION(g->disposition >= 0, "disposition", JSON_BUILD_STRING(user_disposition_to_string(g->disposition))),
+ JSON_BUILD_PAIR("service", JSON_BUILD_CONST_STRING("io.systemd.NSpawn"))));
+ if (r < 0)
+ return log_error_errno(r, "Failed to build container group record: %m");
+
+ *ret_converted_user = TAKE_PTR(converted_user);
+ *ret_converted_group = TAKE_PTR(converted_group);
+
+ return 0;
+}
+
+static int find_free_uid(const char *directory, uid_t max_uid, uid_t *current_uid) {
+ int r;
+
+ assert(directory);
+ assert(current_uid);
+
+ for (;; (*current_uid) ++) {
+ if (*current_uid > MAP_UID_MAX || *current_uid > max_uid)
+ return log_error_errno(
+ SYNTHETIC_ERRNO(EBUSY),
+ "No suitable available UID in range " UID_FMT "…" UID_FMT " in container detected, can't map user.",
+ MAP_UID_MIN, MAP_UID_MAX);
+
+ r = check_etc_passwd_collisions(directory, NULL, *current_uid);
+ if (r < 0)
+ return r;
+ if (r > 0) /* already used */
+ continue;
+
+ /* We want to use the UID also as GID, hence check for it in /etc/group too */
+ r = check_etc_group_collisions(directory, NULL, (gid_t) *current_uid);
+ if (r <= 0)
+ return r;
+ }
+}
+
+BindUserContext* bind_user_context_free(BindUserContext *c) {
+ if (!c)
+ return NULL;
+
+ assert(c->n_data == 0 || c->data);
+
+ for (size_t i = 0; i < c->n_data; i++) {
+ user_record_unref(c->data[i].host_user);
+ group_record_unref(c->data[i].host_group);
+ user_record_unref(c->data[i].payload_user);
+ group_record_unref(c->data[i].payload_group);
+ }
+
+ return mfree(c);
+}
+
+int bind_user_prepare(
+ const char *directory,
+ char **bind_user,
+ uid_t uid_shift,
+ uid_t uid_range,
+ CustomMount **custom_mounts,
+ size_t *n_custom_mounts,
+ BindUserContext **ret) {
+
+ _cleanup_(bind_user_context_freep) BindUserContext *c = NULL;
+ uid_t current_uid = MAP_UID_MIN;
+ int r;
+
+ assert(custom_mounts);
+ assert(n_custom_mounts);
+ assert(ret);
+
+ /* This resolves the users specified in 'bind_user', generates a minimalized JSON user + group record
+ * for it to stick in the container, allocates a UID/GID for it, and updates the custom mount table,
+ * to include an appropriate bind mount mapping.
+ *
+ * This extends the passed custom_mounts/n_custom_mounts with the home directories, and allocates a
+ * new BindUserContext for the user records */
+
+ if (strv_isempty(bind_user)) {
+ *ret = NULL;
+ return 0;
+ }
+
+ c = new0(BindUserContext, 1);
+ if (!c)
+ return log_oom();
+
+ STRV_FOREACH(n, bind_user) {
+ _cleanup_(user_record_unrefp) UserRecord *u = NULL, *cu = NULL;
+ _cleanup_(group_record_unrefp) GroupRecord *g = NULL, *cg = NULL;
+ _cleanup_free_ char *sm = NULL, *sd = NULL;
+ CustomMount *cm;
+
+ r = userdb_by_name(*n, USERDB_DONT_SYNTHESIZE, &u);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve user '%s': %m", *n);
+
+ /* For now, let's refuse mapping the root/nobody users explicitly. The records we generate
+ * are strictly additive, nss-systemd is typically placed last in /etc/nsswitch.conf. Thus
+ * even if we wanted, we couldn't override the root or nobody user records. Note we also
+ * check for name conflicts in /etc/passwd + /etc/group later on, which would usually filter
+ * out root/nobody too, hence these checks might appear redundant — but they actually are
+ * not, as we want to support environments where /etc/passwd and /etc/group are non-existent,
+ * and the user/group databases fully synthesized at runtime. Moreover, the name of the
+ * user/group name of the "nobody" account differs between distros, hence a check by numeric
+ * UID is safer. */
+ if (u->uid == 0 || streq(u->user_name, "root"))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Mapping 'root' user not supported, sorry.");
+ if (u->uid == UID_NOBODY || STR_IN_SET(u->user_name, NOBODY_USER_NAME, "nobody"))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Mapping 'nobody' user not supported, sorry.");
+
+ if (u->uid >= uid_shift && u->uid < uid_shift + uid_range)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID of user '%s' to map is already in container UID range, refusing.", u->user_name);
+
+ r = groupdb_by_gid(u->gid, USERDB_DONT_SYNTHESIZE, &g);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve group of user '%s': %m", u->user_name);
+
+ if (g->gid >= uid_shift && g->gid < uid_shift + uid_range)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "GID of group '%s' to map is already in container GID range, refusing.", g->group_name);
+
+ /* We want to synthesize exactly one user + group from the host into the container. This only
+ * makes sense if the user on the host has its own private group. We can't reasonably check
+ * this, so we just check of the name of user and group match.
+ *
+ * One of these days we might want to support users in a shared/common group too, but it's
+ * not clear to me how this would have to be mapped, precisely given that the common group
+ * probably already exists in the container. */
+ if (!streq(u->user_name, g->group_name))
+ return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Sorry, mapping users without private groups is currently not supported.");
+
+ r = find_free_uid(directory, uid_range, &current_uid);
+ if (r < 0)
+ return r;
+
+ r = convert_user(directory, u, g, current_uid, &cu, &cg);
+ if (r < 0)
+ return r;
+
+ if (!GREEDY_REALLOC(c->data, c->n_data + 1))
+ return log_oom();
+
+ sm = strdup(u->home_directory);
+ if (!sm)
+ return log_oom();
+
+ sd = strdup(cu->home_directory);
+ if (!sd)
+ return log_oom();
+
+ cm = reallocarray(*custom_mounts, *n_custom_mounts + 1, sizeof(CustomMount));
+ if (!cm)
+ return log_oom();
+
+ *custom_mounts = cm;
+
+ (*custom_mounts)[(*n_custom_mounts)++] = (CustomMount) {
+ .type = CUSTOM_MOUNT_BIND,
+ .source = TAKE_PTR(sm),
+ .destination = TAKE_PTR(sd),
+ };
+
+ c->data[c->n_data++] = (BindUserData) {
+ .host_user = TAKE_PTR(u),
+ .host_group = TAKE_PTR(g),
+ .payload_user = TAKE_PTR(cu),
+ .payload_group = TAKE_PTR(cg),
+ };
+
+ current_uid++;
+ }
+
+ *ret = TAKE_PTR(c);
+ return 1;
+}
+
+static int write_and_symlink(
+ const char *root,
+ JsonVariant *v,
+ const char *name,
+ uid_t uid,
+ const char *suffix,
+ WriteStringFileFlags extra_flags) {
+
+ _cleanup_free_ char *j = NULL, *f = NULL, *p = NULL, *q = NULL;
+ int r;
+
+ assert(root);
+ assert(v);
+ assert(name);
+ assert(uid_is_valid(uid));
+ assert(suffix);
+
+ r = json_variant_format(v, JSON_FORMAT_NEWLINE, &j);
+ if (r < 0)
+ return log_error_errno(r, "Failed to format user record JSON: %m");
+
+ f = strjoin(name, suffix);
+ if (!f)
+ return log_oom();
+
+ p = path_join(root, "/run/host/userdb/", f);
+ if (!p)
+ return log_oom();
+
+ if (asprintf(&q, "%s/run/host/userdb/" UID_FMT "%s", root, uid, suffix) < 0)
+ return log_oom();
+
+ if (symlink(f, q) < 0)
+ return log_error_errno(errno, "Failed to create symlink '%s': %m", q);
+
+ r = userns_lchown(q, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to adjust access mode of '%s': %m", q);
+
+ r = write_string_file(p, j, WRITE_STRING_FILE_CREATE|extra_flags);
+ if (r < 0)
+ return log_error_errno(r, "Failed to write %s: %m", p);
+
+ r = userns_lchown(p, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to adjust access mode of '%s': %m", p);
+
+ return 0;
+}
+
+int bind_user_setup(
+ const BindUserContext *c,
+ const char *root) {
+
+ static const UserRecordLoadFlags strip_flags = /* Removes privileged info */
+ USER_RECORD_REQUIRE_REGULAR|
+ USER_RECORD_STRIP_PRIVILEGED|
+ USER_RECORD_ALLOW_PER_MACHINE|
+ USER_RECORD_ALLOW_BINDING|
+ USER_RECORD_ALLOW_SIGNATURE|
+ USER_RECORD_PERMISSIVE;
+ static const UserRecordLoadFlags shadow_flags = /* Extracts privileged info */
+ USER_RECORD_STRIP_REGULAR|
+ USER_RECORD_ALLOW_PRIVILEGED|
+ USER_RECORD_STRIP_PER_MACHINE|
+ USER_RECORD_STRIP_BINDING|
+ USER_RECORD_STRIP_SIGNATURE|
+ USER_RECORD_EMPTY_OK|
+ USER_RECORD_PERMISSIVE;
+ int r;
+
+ assert(root);
+
+ if (!c || c->n_data == 0)
+ return 0;
+
+ r = userns_mkdir(root, "/run/host", 0755, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create /run/host: %m");
+
+ r = userns_mkdir(root, "/run/host/home", 0755, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create /run/host/home: %m");
+
+ r = userns_mkdir(root, "/run/host/userdb", 0755, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create /run/host/userdb: %m");
+
+ for (size_t i = 0; i < c->n_data; i++) {
+ _cleanup_(group_record_unrefp) GroupRecord *stripped_group = NULL, *shadow_group = NULL;
+ _cleanup_(user_record_unrefp) UserRecord *stripped_user = NULL, *shadow_user = NULL;
+ const BindUserData *d = c->data + i;
+
+ /* First, write shadow (i.e. privileged) data for group record */
+ r = group_record_clone(d->payload_group, shadow_flags, &shadow_group);
+ if (r < 0)
+ return log_error_errno(r, "Failed to extract privileged information from group record: %m");
+
+ if (!json_variant_is_blank_object(shadow_group->json)) {
+ r = write_and_symlink(
+ root,
+ shadow_group->json,
+ d->payload_group->group_name,
+ d->payload_group->gid,
+ ".group-privileged",
+ WRITE_STRING_FILE_MODE_0600);
+ if (r < 0)
+ return r;
+ }
+
+ /* Second, write main part of group record. */
+ r = group_record_clone(d->payload_group, strip_flags, &stripped_group);
+ if (r < 0)
+ return log_error_errno(r, "Failed to strip privileged information from group record: %m");
+
+ r = write_and_symlink(
+ root,
+ stripped_group->json,
+ d->payload_group->group_name,
+ d->payload_group->gid,
+ ".group",
+ 0);
+ if (r < 0)
+ return r;
+
+ /* Third, write out user shadow data. i.e. extract privileged info from user record */
+ r = user_record_clone(d->payload_user, shadow_flags, &shadow_user);
+ if (r < 0)
+ return log_error_errno(r, "Failed to extract privileged information from user record: %m");
+
+ if (!json_variant_is_blank_object(shadow_user->json)) {
+ r = write_and_symlink(
+ root,
+ shadow_user->json,
+ d->payload_user->user_name,
+ d->payload_user->uid,
+ ".user-privileged",
+ WRITE_STRING_FILE_MODE_0600);
+ if (r < 0)
+ return r;
+ }
+
+ /* Finally write out the main part of the user record */
+ r = user_record_clone(d->payload_user, strip_flags, &stripped_user);
+ if (r < 0)
+ return log_error_errno(r, "Failed to strip privileged information from user record: %m");
+
+ r = write_and_symlink(
+ root,
+ stripped_user->json,
+ d->payload_user->user_name,
+ d->payload_user->uid,
+ ".user",
+ 0);
+ if (r < 0)
+ return r;
+ }
+
+ return 1;
+}
diff --git a/src/nspawn/nspawn-bind-user.h b/src/nspawn/nspawn-bind-user.h
new file mode 100644
index 0000000..4352ce0
--- /dev/null
+++ b/src/nspawn/nspawn-bind-user.h
@@ -0,0 +1,29 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include "user-record.h"
+#include "group-record.h"
+#include "nspawn-mount.h"
+
+typedef struct BindUserData {
+ /* The host's user/group records */
+ UserRecord *host_user;
+ GroupRecord *host_group;
+
+ /* The mapped records to place into the container */
+ UserRecord *payload_user;
+ GroupRecord *payload_group;
+} BindUserData;
+
+typedef struct BindUserContext {
+ BindUserData *data;
+ size_t n_data;
+} BindUserContext;
+
+BindUserContext* bind_user_context_free(BindUserContext *c);
+
+DEFINE_TRIVIAL_CLEANUP_FUNC(BindUserContext*, bind_user_context_free);
+
+int bind_user_prepare(const char *directory, char **bind_user, uid_t uid_shift, uid_t uid_range, CustomMount **custom_mounts, size_t *n_custom_mounts, BindUserContext **ret);
+
+int bind_user_setup(const BindUserContext *c, const char *root);
diff --git a/src/nspawn/nspawn-cgroup.c b/src/nspawn/nspawn-cgroup.c
new file mode 100644
index 0000000..6c47ab6
--- /dev/null
+++ b/src/nspawn/nspawn-cgroup.c
@@ -0,0 +1,606 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <sys/mount.h>
+
+#include "alloc-util.h"
+#include "cgroup-setup.h"
+#include "fd-util.h"
+#include "fileio.h"
+#include "format-util.h"
+#include "fs-util.h"
+#include "mkdir.h"
+#include "mount-util.h"
+#include "mountpoint-util.h"
+#include "nspawn-cgroup.h"
+#include "nspawn-mount.h"
+#include "path-util.h"
+#include "rm-rf.h"
+#include "string-util.h"
+#include "strv.h"
+#include "user-util.h"
+#include "util.h"
+
+static int chown_cgroup_path(const char *path, uid_t uid_shift) {
+ _cleanup_close_ int fd = -1;
+
+ fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY);
+ if (fd < 0)
+ return -errno;
+
+ FOREACH_STRING(fn,
+ ".",
+ "cgroup.clone_children",
+ "cgroup.controllers",
+ "cgroup.events",
+ "cgroup.procs",
+ "cgroup.stat",
+ "cgroup.subtree_control",
+ "cgroup.threads",
+ "memory.oom.group",
+ "memory.reclaim",
+ "notify_on_release",
+ "tasks")
+ if (fchownat(fd, fn, uid_shift, uid_shift, 0) < 0)
+ log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno,
+ "Failed to chown \"%s/%s\", ignoring: %m", path, fn);
+
+ return 0;
+}
+
+int chown_cgroup(pid_t pid, CGroupUnified unified_requested, uid_t uid_shift) {
+ _cleanup_free_ char *path = NULL, *fs = NULL;
+ int r;
+
+ r = cg_pid_get_path(NULL, pid, &path);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get container cgroup path: %m");
+
+ r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, path, NULL, &fs);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get file system path for container cgroup: %m");
+
+ r = chown_cgroup_path(fs, uid_shift);
+ if (r < 0)
+ return log_error_errno(r, "Failed to chown() cgroup %s: %m", fs);
+
+ if (unified_requested == CGROUP_UNIFIED_SYSTEMD || (unified_requested == CGROUP_UNIFIED_NONE && cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0)) {
+ _cleanup_free_ char *lfs = NULL;
+ /* Always propagate access rights from unified to legacy controller */
+
+ r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, NULL, &lfs);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get file system path for container cgroup: %m");
+
+ r = chown_cgroup_path(lfs, uid_shift);
+ if (r < 0)
+ return log_error_errno(r, "Failed to chown() cgroup %s: %m", lfs);
+ }
+
+ return 0;
+}
+
+int sync_cgroup(pid_t pid, CGroupUnified unified_requested, uid_t uid_shift) {
+ _cleanup_free_ char *cgroup = NULL;
+ char tree[] = "/tmp/unifiedXXXXXX", pid_string[DECIMAL_STR_MAX(pid) + 1];
+ bool undo_mount = false;
+ const char *fn;
+ int r, unified_controller;
+
+ unified_controller = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER);
+ if (unified_controller < 0)
+ return log_error_errno(unified_controller, "Failed to determine whether the systemd hierarchy is unified: %m");
+ if ((unified_controller > 0) == (unified_requested >= CGROUP_UNIFIED_SYSTEMD))
+ return 0;
+
+ /* When the host uses the legacy cgroup setup, but the
+ * container shall use the unified hierarchy, let's make sure
+ * we copy the path from the name=systemd hierarchy into the
+ * unified hierarchy. Similar for the reverse situation. */
+
+ r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get control group of " PID_FMT ": %m", pid);
+
+ /* In order to access the unified hierarchy we need to mount it */
+ if (!mkdtemp(tree))
+ return log_error_errno(errno, "Failed to generate temporary mount point for unified hierarchy: %m");
+
+ if (unified_controller > 0)
+ r = mount_nofollow_verbose(LOG_ERR, "cgroup", tree, "cgroup",
+ MS_NOSUID|MS_NOEXEC|MS_NODEV, "none,name=systemd,xattr");
+ else
+ r = mount_nofollow_verbose(LOG_ERR, "cgroup", tree, "cgroup2",
+ MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
+ if (r < 0)
+ goto finish;
+
+ undo_mount = true;
+
+ /* If nspawn dies abruptly the cgroup hierarchy created below
+ * its unit isn't cleaned up. So, let's remove it
+ * https://github.com/systemd/systemd/pull/4223#issuecomment-252519810 */
+ fn = strjoina(tree, cgroup);
+ (void) rm_rf(fn, REMOVE_ROOT|REMOVE_ONLY_DIRECTORIES);
+
+ fn = strjoina(tree, cgroup, "/cgroup.procs");
+
+ sprintf(pid_string, PID_FMT, pid);
+ r = write_string_file(fn, pid_string, WRITE_STRING_FILE_DISABLE_BUFFER|WRITE_STRING_FILE_MKDIR_0755);
+ if (r < 0) {
+ log_error_errno(r, "Failed to move process: %m");
+ goto finish;
+ }
+
+ fn = strjoina(tree, cgroup);
+ r = chown_cgroup_path(fn, uid_shift);
+ if (r < 0)
+ log_error_errno(r, "Failed to chown() cgroup %s: %m", fn);
+finish:
+ if (undo_mount)
+ (void) umount_verbose(LOG_ERR, tree, UMOUNT_NOFOLLOW);
+
+ (void) rmdir(tree);
+ return r;
+}
+
+int create_subcgroup(pid_t pid, bool keep_unit, CGroupUnified unified_requested) {
+ _cleanup_free_ char *cgroup = NULL;
+ CGroupMask supported;
+ const char *payload;
+ int r;
+
+ assert(pid > 1);
+
+ /* In the unified hierarchy inner nodes may only contain subgroups, but not processes. Hence, if we running in
+ * the unified hierarchy and the container does the same, and we did not create a scope unit for the container
+ * move us and the container into two separate subcgroups.
+ *
+ * Moreover, container payloads such as systemd try to manage the cgroup they run in full (i.e. including
+ * its attributes), while the host systemd will only delegate cgroups for children of the cgroup created for a
+ * delegation unit, instead of the cgroup itself. This means, if we'd pass on the cgroup allocated from the
+ * host systemd directly to the payload, the host and payload systemd might fight for the cgroup
+ * attributes. Hence, let's insert an intermediary cgroup to cover that case too.
+ *
+ * Note that we only bother with the main hierarchy here, not with any secondary ones. On the unified setup
+ * that's fine because there's only one hierarchy anyway and controllers are enabled directly on it. On the
+ * legacy setup, this is fine too, since delegation of controllers is generally not safe there, hence we won't
+ * do it. */
+
+ r = cg_mask_supported(&supported);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine supported controllers: %m");
+
+ if (keep_unit)
+ r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
+ else
+ r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get our control group: %m");
+
+ payload = strjoina(cgroup, "/payload");
+ r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, payload, pid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create %s subcgroup: %m", payload);
+
+ if (keep_unit) {
+ const char *supervisor;
+
+ supervisor = strjoina(cgroup, "/supervisor");
+ r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, supervisor, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create %s subcgroup: %m", supervisor);
+ }
+
+ /* Try to enable as many controllers as possible for the new payload. */
+ (void) cg_enable_everywhere(supported, supported, cgroup, NULL);
+ return 0;
+}
+
+/* Retrieve existing subsystems. This function is called in a new cgroup
+ * namespace.
+ */
+static int get_process_controllers(Set **ret) {
+ _cleanup_set_free_ Set *controllers = NULL;
+ _cleanup_fclose_ FILE *f = NULL;
+ int r;
+
+ assert(ret);
+
+ f = fopen("/proc/self/cgroup", "re");
+ if (!f)
+ return errno == ENOENT ? -ESRCH : -errno;
+
+ for (;;) {
+ _cleanup_free_ char *line = NULL;
+ char *e, *l;
+
+ r = read_line(f, LONG_LINE_MAX, &line);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ break;
+
+ l = strchr(line, ':');
+ if (!l)
+ continue;
+
+ l++;
+ e = strchr(l, ':');
+ if (!e)
+ continue;
+
+ *e = 0;
+
+ if (STR_IN_SET(l, "", "name=systemd", "name=unified"))
+ continue;
+
+ r = set_put_strdup(&controllers, l);
+ if (r < 0)
+ return r;
+ }
+
+ *ret = TAKE_PTR(controllers);
+
+ return 0;
+}
+
+static int mount_legacy_cgroup_hierarchy(
+ const char *dest,
+ const char *controller,
+ const char *hierarchy,
+ bool read_only) {
+
+ const char *to, *fstype, *opts;
+ int r;
+
+ to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy);
+
+ r = path_is_mount_point(to, dest, 0);
+ if (r < 0 && r != -ENOENT)
+ return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to);
+ if (r > 0)
+ return 0;
+
+ (void) mkdir_p(to, 0755);
+
+ /* The superblock mount options of the mount point need to be
+ * identical to the hosts', and hence writable... */
+ if (streq(controller, SYSTEMD_CGROUP_CONTROLLER_HYBRID)) {
+ fstype = "cgroup2";
+ opts = NULL;
+ } else if (streq(controller, SYSTEMD_CGROUP_CONTROLLER_LEGACY)) {
+ fstype = "cgroup";
+ opts = "none,name=systemd,xattr";
+ } else {
+ fstype = "cgroup";
+ opts = controller;
+ }
+
+ r = mount_nofollow_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts);
+ if (r < 0)
+ return r;
+
+ /* ... hence let's only make the bind mount read-only, not the superblock. */
+ if (read_only) {
+ r = mount_nofollow_verbose(LOG_ERR, NULL, to, NULL,
+ MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
+ if (r < 0)
+ return r;
+ }
+
+ return 1;
+}
+
+/* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
+static int mount_legacy_cgns_supported(
+ const char *dest,
+ CGroupUnified unified_requested,
+ bool userns,
+ uid_t uid_shift,
+ uid_t uid_range,
+ const char *selinux_apifs_context) {
+
+ _cleanup_set_free_ Set *controllers = NULL;
+ const char *cgroup_root = "/sys/fs/cgroup", *c;
+ int r;
+
+ (void) mkdir_p(cgroup_root, 0755);
+
+ /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
+ r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
+ if (r == 0) {
+ _cleanup_free_ char *options = NULL;
+
+ /* When cgroup namespaces are enabled and user namespaces are
+ * used then the mount of the cgroupfs is done *inside* the new
+ * user namespace. We're root in the new user namespace and the
+ * kernel will happily translate our uid/gid to the correct
+ * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
+ * pass uid 0 and not uid_shift to tmpfs_patch_options().
+ */
+ r = tmpfs_patch_options("mode=755" TMPFS_LIMITS_SYS_FS_CGROUP, 0, selinux_apifs_context, &options);
+ if (r < 0)
+ return log_oom();
+
+ r = mount_nofollow_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
+ MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
+ if (r < 0)
+ return r;
+ }
+
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0)
+ goto skip_controllers;
+
+ r = get_process_controllers(&controllers);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine cgroup controllers: %m");
+
+ for (;;) {
+ _cleanup_free_ const char *controller = NULL;
+
+ controller = set_steal_first(controllers);
+ if (!controller)
+ break;
+
+ r = mount_legacy_cgroup_hierarchy("", controller, controller, !userns);
+ if (r < 0)
+ return r;
+
+ /* When multiple hierarchies are co-mounted, make their
+ * constituting individual hierarchies a symlink to the
+ * co-mount.
+ */
+ c = controller;
+ for (;;) {
+ _cleanup_free_ char *target = NULL, *tok = NULL;
+
+ r = extract_first_word(&c, &tok, ",", 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
+ if (r == 0)
+ break;
+
+ if (streq(controller, tok))
+ break;
+
+ target = path_join("/sys/fs/cgroup/", tok);
+ if (!target)
+ return log_oom();
+
+ r = symlink_idempotent(controller, target, false);
+ if (r == -EINVAL)
+ return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
+ if (r < 0)
+ return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
+ }
+ }
+
+skip_controllers:
+ if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
+ r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
+ if (r < 0)
+ return r;
+ }
+
+ r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
+ if (r < 0)
+ return r;
+
+ if (!userns)
+ return mount_nofollow_verbose(LOG_ERR, NULL, cgroup_root, NULL,
+ MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
+
+ return 0;
+}
+
+/* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
+static int mount_legacy_cgns_unsupported(
+ const char *dest,
+ CGroupUnified unified_requested,
+ bool userns,
+ uid_t uid_shift,
+ uid_t uid_range,
+ const char *selinux_apifs_context) {
+
+ _cleanup_set_free_ Set *controllers = NULL;
+ const char *cgroup_root;
+ int r;
+
+ cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
+
+ (void) mkdir_p(cgroup_root, 0755);
+
+ /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
+ r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
+ if (r == 0) {
+ _cleanup_free_ char *options = NULL;
+
+ r = tmpfs_patch_options("mode=755" TMPFS_LIMITS_SYS_FS_CGROUP, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &options);
+ if (r < 0)
+ return log_oom();
+
+ r = mount_nofollow_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
+ MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
+ if (r < 0)
+ return r;
+ }
+
+ r = cg_all_unified();
+ if (r < 0)
+ return r;
+ if (r > 0)
+ goto skip_controllers;
+
+ r = cg_kernel_controllers(&controllers);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine cgroup controllers: %m");
+
+ for (;;) {
+ _cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
+
+ controller = set_steal_first(controllers);
+ if (!controller)
+ break;
+
+ origin = path_join("/sys/fs/cgroup/", controller);
+ if (!origin)
+ return log_oom();
+
+ r = readlink_malloc(origin, &combined);
+ if (r == -EINVAL) {
+ /* Not a symbolic link, but directly a single cgroup hierarchy */
+
+ r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true);
+ if (r < 0)
+ return r;
+
+ } else if (r < 0)
+ return log_error_errno(r, "Failed to read link %s: %m", origin);
+ else {
+ _cleanup_free_ char *target = NULL;
+
+ target = path_join(dest, origin);
+ if (!target)
+ return log_oom();
+
+ /* A symbolic link, a combination of controllers in one hierarchy */
+
+ if (!filename_is_valid(combined)) {
+ log_warning("Ignoring invalid combined hierarchy %s.", combined);
+ continue;
+ }
+
+ r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true);
+ if (r < 0)
+ return r;
+
+ r = symlink_idempotent(combined, target, false);
+ if (r == -EINVAL)
+ return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
+ if (r < 0)
+ return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
+ }
+ }
+
+skip_controllers:
+ if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
+ r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
+ if (r < 0)
+ return r;
+ }
+
+ r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
+ if (r < 0)
+ return r;
+
+ return mount_nofollow_verbose(LOG_ERR, NULL, cgroup_root, NULL,
+ MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
+}
+
+static int mount_unified_cgroups(const char *dest) {
+ const char *p;
+ int r;
+
+ assert(dest);
+
+ p = prefix_roota(dest, "/sys/fs/cgroup");
+
+ (void) mkdir_p(p, 0755);
+
+ r = path_is_mount_point(p, dest, AT_SYMLINK_FOLLOW);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p);
+ if (r > 0) {
+ p = prefix_roota(dest, "/sys/fs/cgroup/cgroup.procs");
+ if (access(p, F_OK) >= 0)
+ return 0;
+ if (errno != ENOENT)
+ return log_error_errno(errno, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p);
+
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "%s is already mounted but not a unified cgroup hierarchy. Refusing.", p);
+ }
+
+ return mount_nofollow_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
+}
+
+int mount_cgroups(
+ const char *dest,
+ CGroupUnified unified_requested,
+ bool userns,
+ uid_t uid_shift,
+ uid_t uid_range,
+ const char *selinux_apifs_context,
+ bool use_cgns) {
+
+ if (unified_requested >= CGROUP_UNIFIED_ALL)
+ return mount_unified_cgroups(dest);
+ if (use_cgns)
+ return mount_legacy_cgns_supported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
+
+ return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context);
+}
+
+static int mount_systemd_cgroup_writable_one(const char *root, const char *own) {
+ int r;
+
+ assert(root);
+ assert(own);
+
+ /* Make our own cgroup a (writable) bind mount */
+ r = mount_nofollow_verbose(LOG_ERR, own, own, NULL, MS_BIND, NULL);
+ if (r < 0)
+ return r;
+
+ /* And then remount the systemd cgroup root read-only */
+ return mount_nofollow_verbose(LOG_ERR, NULL, root, NULL,
+ MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL);
+}
+
+int mount_systemd_cgroup_writable(
+ const char *dest,
+ CGroupUnified unified_requested) {
+
+ _cleanup_free_ char *own_cgroup_path = NULL;
+ const char *root, *own;
+ int r;
+
+ assert(dest);
+
+ r = cg_pid_get_path(NULL, 0, &own_cgroup_path);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine our own cgroup path: %m");
+
+ /* If we are living in the top-level, then there's nothing to do... */
+ if (path_equal(own_cgroup_path, "/"))
+ return 0;
+
+ if (unified_requested >= CGROUP_UNIFIED_ALL) {
+
+ root = prefix_roota(dest, "/sys/fs/cgroup");
+ own = strjoina(root, own_cgroup_path);
+
+ } else {
+
+ if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
+ root = prefix_roota(dest, "/sys/fs/cgroup/unified");
+ own = strjoina(root, own_cgroup_path);
+
+ r = mount_systemd_cgroup_writable_one(root, own);
+ if (r < 0)
+ return r;
+ }
+
+ root = prefix_roota(dest, "/sys/fs/cgroup/systemd");
+ own = strjoina(root, own_cgroup_path);
+ }
+
+ return mount_systemd_cgroup_writable_one(root, own);
+}
diff --git a/src/nspawn/nspawn-cgroup.h b/src/nspawn/nspawn-cgroup.h
new file mode 100644
index 0000000..3f5ba62
--- /dev/null
+++ b/src/nspawn/nspawn-cgroup.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <stdbool.h>
+#include <sys/types.h>
+
+#include "cgroup-util.h"
+
+int chown_cgroup(pid_t pid, CGroupUnified unified_requested, uid_t uid_shift);
+int sync_cgroup(pid_t pid, CGroupUnified unified_requested, uid_t uid_shift);
+int create_subcgroup(pid_t pid, bool keep_unit, CGroupUnified unified_requested);
+
+int mount_cgroups(const char *dest, CGroupUnified unified_requested, bool userns, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context, bool use_cgns);
+int mount_systemd_cgroup_writable(const char *dest, CGroupUnified unified_requested);
diff --git a/src/nspawn/nspawn-creds.c b/src/nspawn/nspawn-creds.c
new file mode 100644
index 0000000..0900d8c
--- /dev/null
+++ b/src/nspawn/nspawn-creds.c
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "alloc-util.h"
+#include "macro.h"
+#include "memory-util.h"
+#include "nspawn-creds.h"
+
+static void credential_free(Credential *cred) {
+ assert(cred);
+
+ cred->id = mfree(cred->id);
+ cred->data = erase_and_free(cred->data);
+ cred->size = 0;
+}
+
+void credential_free_all(Credential *creds, size_t n) {
+ size_t i;
+
+ assert(creds || n == 0);
+
+ for (i = 0; i < n; i++)
+ credential_free(creds + i);
+
+ free(creds);
+}
diff --git a/src/nspawn/nspawn-creds.h b/src/nspawn/nspawn-creds.h
new file mode 100644
index 0000000..de0661b
--- /dev/null
+++ b/src/nspawn/nspawn-creds.h
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <sys/types.h>
+
+typedef struct Credential {
+ char *id;
+ void *data;
+ size_t size;
+} Credential;
+
+void credential_free_all(Credential *creds, size_t n);
diff --git a/src/nspawn/nspawn-def.h b/src/nspawn/nspawn-def.h
new file mode 100644
index 0000000..32a20aa
--- /dev/null
+++ b/src/nspawn/nspawn-def.h
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <sys/types.h>
+
+/* While we are chmod()ing a directory tree, we set the top-level UID base to this "busy" base, so that we can always
+ * recognize trees we are were chmod()ing recursively and got interrupted in */
+#define UID_BUSY_BASE ((uid_t) UINT32_C(0xFFFE0000))
+#define UID_BUSY_MASK ((uid_t) UINT32_C(0xFFFF0000))
diff --git a/src/nspawn/nspawn-expose-ports.c b/src/nspawn/nspawn-expose-ports.c
new file mode 100644
index 0000000..bb54f8c
--- /dev/null
+++ b/src/nspawn/nspawn-expose-ports.c
@@ -0,0 +1,220 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "sd-netlink.h"
+
+#include "af-list.h"
+#include "alloc-util.h"
+#include "fd-util.h"
+#include "firewall-util.h"
+#include "in-addr-util.h"
+#include "local-addresses.h"
+#include "netlink-util.h"
+#include "nspawn-expose-ports.h"
+#include "parse-util.h"
+#include "socket-util.h"
+#include "string-util.h"
+#include "util.h"
+
+int expose_port_parse(ExposePort **l, const char *s) {
+ const char *split, *e;
+ uint16_t container_port, host_port;
+ ExposePort *port;
+ int protocol;
+ int r;
+
+ assert(l);
+ assert(s);
+
+ if ((e = startswith(s, "tcp:")))
+ protocol = IPPROTO_TCP;
+ else if ((e = startswith(s, "udp:")))
+ protocol = IPPROTO_UDP;
+ else {
+ e = s;
+ protocol = IPPROTO_TCP;
+ }
+
+ split = strchr(e, ':');
+ if (split) {
+ char v[split - e + 1];
+
+ memcpy(v, e, split - e);
+ v[split - e] = 0;
+
+ r = parse_ip_port(v, &host_port);
+ if (r < 0)
+ return -EINVAL;
+
+ r = parse_ip_port(split + 1, &container_port);
+ } else {
+ r = parse_ip_port(e, &container_port);
+ host_port = container_port;
+ }
+
+ if (r < 0)
+ return r;
+
+ LIST_FOREACH(ports, p, *l)
+ if (p->protocol == protocol && p->host_port == host_port)
+ return -EEXIST;
+
+ port = new(ExposePort, 1);
+ if (!port)
+ return -ENOMEM;
+
+ *port = (ExposePort) {
+ .protocol = protocol,
+ .host_port = host_port,
+ .container_port = container_port,
+ };
+
+ LIST_PREPEND(ports, *l, port);
+
+ return 0;
+}
+
+void expose_port_free_all(ExposePort *p) {
+
+ while (p) {
+ ExposePort *q = p;
+ LIST_REMOVE(ports, p, q);
+ free(q);
+ }
+}
+
+int expose_port_flush(FirewallContext **fw_ctx, ExposePort* l, int af, union in_addr_union *exposed) {
+ int r;
+
+ assert(exposed);
+
+ if (!l)
+ return 0;
+
+ if (!in_addr_is_set(af, exposed))
+ return 0;
+
+ log_debug("Lost IP address.");
+
+ LIST_FOREACH(ports, p, l) {
+ r = fw_add_local_dnat(fw_ctx,
+ false,
+ af,
+ p->protocol,
+ p->host_port,
+ exposed,
+ p->container_port,
+ NULL);
+ if (r < 0)
+ log_warning_errno(r, "Failed to modify %s firewall: %m", af_to_name(af));
+ }
+
+ *exposed = IN_ADDR_NULL;
+ return 0;
+}
+
+int expose_port_execute(sd_netlink *rtnl, FirewallContext **fw_ctx, ExposePort *l, int af, union in_addr_union *exposed) {
+ _cleanup_free_ struct local_address *addresses = NULL;
+ union in_addr_union new_exposed;
+ bool add;
+ int r;
+
+ assert(exposed);
+
+ /* Invoked each time an address is added or removed inside the
+ * container */
+
+ if (!l)
+ return 0;
+
+ r = local_addresses(rtnl, 0, af, &addresses);
+ if (r < 0)
+ return log_error_errno(r, "Failed to enumerate local addresses: %m");
+
+ add = r > 0 &&
+ addresses[0].family == af &&
+ addresses[0].scope < RT_SCOPE_LINK;
+
+ if (!add)
+ return expose_port_flush(fw_ctx, l, af, exposed);
+
+ new_exposed = addresses[0].address;
+ if (in_addr_equal(af, exposed, &new_exposed))
+ return 0;
+
+ log_debug("New container IP is %s.", IN_ADDR_TO_STRING(af, &new_exposed));
+
+ LIST_FOREACH(ports, p, l) {
+ r = fw_add_local_dnat(fw_ctx,
+ true,
+ af,
+ p->protocol,
+ p->host_port,
+ &new_exposed,
+ p->container_port,
+ in_addr_is_set(af, exposed) ? exposed : NULL);
+ if (r < 0)
+ log_warning_errno(r, "Failed to modify %s firewall: %m", af_to_name(af));
+ }
+
+ *exposed = new_exposed;
+ return 0;
+}
+
+int expose_port_send_rtnl(int send_fd) {
+ _cleanup_close_ int fd = -1;
+ int r;
+
+ assert(send_fd >= 0);
+
+ fd = socket(AF_NETLINK, SOCK_RAW|SOCK_CLOEXEC|SOCK_NONBLOCK, NETLINK_ROUTE);
+ if (fd < 0)
+ return log_error_errno(errno, "Failed to allocate container netlink: %m");
+
+ /* Store away the fd in the socket, so that it stays open as
+ * long as we run the child */
+ r = send_one_fd(send_fd, fd, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to send netlink fd: %m");
+
+ return 0;
+}
+
+int expose_port_watch_rtnl(
+ sd_event *event,
+ int recv_fd,
+ sd_netlink_message_handler_t handler,
+ void *userdata,
+ sd_netlink **ret) {
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ int fd, r;
+
+ assert(event);
+ assert(recv_fd >= 0);
+ assert(ret);
+
+ fd = receive_one_fd(recv_fd, 0);
+ if (fd < 0)
+ return log_error_errno(fd, "Failed to recv netlink fd: %m");
+
+ r = sd_netlink_open_fd(&rtnl, fd);
+ if (r < 0) {
+ safe_close(fd);
+ return log_error_errno(r, "Failed to create rtnl object: %m");
+ }
+
+ r = sd_netlink_add_match(rtnl, NULL, RTM_NEWADDR, handler, NULL, userdata, "nspawn-NEWADDR");
+ if (r < 0)
+ return log_error_errno(r, "Failed to subscribe to RTM_NEWADDR messages: %m");
+
+ r = sd_netlink_add_match(rtnl, NULL, RTM_DELADDR, handler, NULL, userdata, "nspawn-DELADDR");
+ if (r < 0)
+ return log_error_errno(r, "Failed to subscribe to RTM_DELADDR messages: %m");
+
+ r = sd_netlink_attach_event(rtnl, event, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add to event loop: %m");
+
+ *ret = TAKE_PTR(rtnl);
+
+ return 0;
+}
diff --git a/src/nspawn/nspawn-expose-ports.h b/src/nspawn/nspawn-expose-ports.h
new file mode 100644
index 0000000..27cfccf
--- /dev/null
+++ b/src/nspawn/nspawn-expose-ports.h
@@ -0,0 +1,27 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <inttypes.h>
+
+#include "sd-event.h"
+#include "sd-netlink.h"
+
+#include "firewall-util.h"
+#include "in-addr-util.h"
+#include "list.h"
+
+typedef struct ExposePort {
+ int protocol;
+ uint16_t host_port;
+ uint16_t container_port;
+ LIST_FIELDS(struct ExposePort, ports);
+} ExposePort;
+
+void expose_port_free_all(ExposePort *p);
+int expose_port_parse(ExposePort **l, const char *s);
+
+int expose_port_watch_rtnl(sd_event *event, int recv_fd, sd_netlink_message_handler_t handler, void *userdata, sd_netlink **ret);
+int expose_port_send_rtnl(int send_fd);
+
+int expose_port_execute(sd_netlink *rtnl, FirewallContext **fw_ctx, ExposePort *l, int af, union in_addr_union *exposed);
+int expose_port_flush(FirewallContext **fw_ctx, ExposePort* l, int af, union in_addr_union *exposed);
diff --git a/src/nspawn/nspawn-gperf.gperf b/src/nspawn/nspawn-gperf.gperf
new file mode 100644
index 0000000..a93b8c3
--- /dev/null
+++ b/src/nspawn/nspawn-gperf.gperf
@@ -0,0 +1,82 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+%{
+#if __GNUC__ >= 7
+_Pragma("GCC diagnostic ignored \"-Wimplicit-fallthrough\"")
+#endif
+#include <stddef.h>
+#include "conf-parser.h"
+#include "nspawn-settings.h"
+#include "nspawn-expose-ports.h"
+%}
+struct ConfigPerfItem;
+%null_strings
+%language=ANSI-C
+%define slot-name section_and_lvalue
+%define hash-function-name nspawn_gperf_hash
+%define lookup-function-name nspawn_gperf_lookup
+%readonly-tables
+%omit-struct-type
+%struct-type
+%includes
+%%
+Exec.Boot, config_parse_boot, 0, 0
+Exec.Ephemeral, config_parse_tristate, 0, offsetof(Settings, ephemeral)
+Exec.ProcessTwo, config_parse_pid2, 0, 0
+Exec.Parameters, config_parse_strv, 0, offsetof(Settings, parameters)
+Exec.Environment, config_parse_strv, 0, offsetof(Settings, environment)
+Exec.User, config_parse_string, CONFIG_PARSE_STRING_SAFE, offsetof(Settings, user)
+Exec.Capability, config_parse_capability, 0, offsetof(Settings, capability)
+Exec.AmbientCapability, config_parse_capability, 0, offsetof(Settings, ambient_capability)
+Exec.DropCapability, config_parse_capability, 0, offsetof(Settings, drop_capability)
+Exec.KillSignal, config_parse_signal, 0, offsetof(Settings, kill_signal)
+Exec.Personality, config_parse_personality, 0, offsetof(Settings, personality)
+Exec.MachineID, config_parse_id128, 0, offsetof(Settings, machine_id)
+Exec.WorkingDirectory, config_parse_path, 0, offsetof(Settings, working_directory)
+Exec.PivotRoot, config_parse_pivot_root, 0, 0
+Exec.PrivateUsers, config_parse_private_users, 0, 0
+Exec.NotifyReady, config_parse_tristate, 0, offsetof(Settings, notify_ready)
+Exec.SystemCallFilter, config_parse_syscall_filter, 0, 0
+Exec.LimitCPU, config_parse_rlimit, RLIMIT_CPU, offsetof(Settings, rlimit)
+Exec.LimitFSIZE, config_parse_rlimit, RLIMIT_FSIZE, offsetof(Settings, rlimit)
+Exec.LimitDATA, config_parse_rlimit, RLIMIT_DATA, offsetof(Settings, rlimit)
+Exec.LimitSTACK, config_parse_rlimit, RLIMIT_STACK, offsetof(Settings, rlimit)
+Exec.LimitCORE, config_parse_rlimit, RLIMIT_CORE, offsetof(Settings, rlimit)
+Exec.LimitRSS, config_parse_rlimit, RLIMIT_RSS, offsetof(Settings, rlimit)
+Exec.LimitNOFILE, config_parse_rlimit, RLIMIT_NOFILE, offsetof(Settings, rlimit)
+Exec.LimitAS, config_parse_rlimit, RLIMIT_AS, offsetof(Settings, rlimit)
+Exec.LimitNPROC, config_parse_rlimit, RLIMIT_NPROC, offsetof(Settings, rlimit)
+Exec.LimitMEMLOCK, config_parse_rlimit, RLIMIT_MEMLOCK, offsetof(Settings, rlimit)
+Exec.LimitLOCKS, config_parse_rlimit, RLIMIT_LOCKS, offsetof(Settings, rlimit)
+Exec.LimitSIGPENDING, config_parse_rlimit, RLIMIT_SIGPENDING, offsetof(Settings, rlimit)
+Exec.LimitMSGQUEUE, config_parse_rlimit, RLIMIT_MSGQUEUE, offsetof(Settings, rlimit)
+Exec.LimitNICE, config_parse_rlimit, RLIMIT_NICE, offsetof(Settings, rlimit)
+Exec.LimitRTPRIO, config_parse_rlimit, RLIMIT_RTPRIO, offsetof(Settings, rlimit)
+Exec.LimitRTTIME, config_parse_rlimit, RLIMIT_RTTIME, offsetof(Settings, rlimit)
+Exec.Hostname, config_parse_hostname, 0, offsetof(Settings, hostname)
+Exec.NoNewPrivileges, config_parse_tristate, 0, offsetof(Settings, no_new_privileges)
+Exec.OOMScoreAdjust, config_parse_oom_score_adjust, 0, 0
+Exec.CPUAffinity, config_parse_cpu_affinity, 0, 0
+Exec.ResolvConf, config_parse_resolv_conf, 0, offsetof(Settings, resolv_conf)
+Exec.LinkJournal, config_parse_link_journal, 0, 0
+Exec.Timezone, config_parse_timezone, 0, offsetof(Settings, timezone)
+Exec.SuppressSync, config_parse_tristate, 0, offsetof(Settings, suppress_sync)
+Files.ReadOnly, config_parse_tristate, 0, offsetof(Settings, read_only)
+Files.Volatile, config_parse_volatile_mode, 0, offsetof(Settings, volatile_mode)
+Files.Bind, config_parse_bind, 0, 0
+Files.BindReadOnly, config_parse_bind, 1, 0
+Files.TemporaryFileSystem, config_parse_tmpfs, 0, 0
+Files.Inaccessible, config_parse_inaccessible, 0, 0
+Files.Overlay, config_parse_overlay, 0, 0
+Files.OverlayReadOnly, config_parse_overlay, 1, 0
+Files.PrivateUsersChown, config_parse_userns_chown, 0, offsetof(Settings, userns_ownership)
+Files.PrivateUsersOwnership, config_parse_userns_ownership, 0, offsetof(Settings, userns_ownership)
+Files.BindUser, config_parse_bind_user, 0, offsetof(Settings, bind_user)
+Network.Private, config_parse_tristate, 0, offsetof(Settings, private_network)
+Network.Interface, config_parse_strv, 0, offsetof(Settings, network_interfaces)
+Network.MACVLAN, config_parse_strv, 0, offsetof(Settings, network_macvlan)
+Network.IPVLAN, config_parse_strv, 0, offsetof(Settings, network_ipvlan)
+Network.VirtualEthernet, config_parse_tristate, 0, offsetof(Settings, network_veth)
+Network.VirtualEthernetExtra, config_parse_veth_extra, 0, 0
+Network.Bridge, config_parse_ifname, 0, offsetof(Settings, network_bridge)
+Network.Zone, config_parse_network_zone, 0, 0
+Network.Port, config_parse_expose_port, 0, 0
diff --git a/src/nspawn/nspawn-mount.c b/src/nspawn/nspawn-mount.c
new file mode 100644
index 0000000..63f59fe
--- /dev/null
+++ b/src/nspawn/nspawn-mount.c
@@ -0,0 +1,1338 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <sys/mount.h>
+#include <linux/magic.h>
+
+#include "alloc-util.h"
+#include "chase-symlinks.h"
+#include "escape.h"
+#include "fd-util.h"
+#include "format-util.h"
+#include "fs-util.h"
+#include "label.h"
+#include "mkdir-label.h"
+#include "mount-util.h"
+#include "mountpoint-util.h"
+#include "nspawn-mount.h"
+#include "parse-util.h"
+#include "path-util.h"
+#include "rm-rf.h"
+#include "set.h"
+#include "sort-util.h"
+#include "stat-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "tmpfile-util.h"
+#include "user-util.h"
+
+CustomMount* custom_mount_add(CustomMount **l, size_t *n, CustomMountType t) {
+ CustomMount *c, *ret;
+
+ assert(l);
+ assert(n);
+ assert(t >= 0);
+ assert(t < _CUSTOM_MOUNT_TYPE_MAX);
+
+ c = reallocarray(*l, *n + 1, sizeof(CustomMount));
+ if (!c)
+ return NULL;
+
+ *l = c;
+ ret = *l + *n;
+ (*n)++;
+
+ *ret = (CustomMount) {
+ .type = t
+ };
+
+ return ret;
+}
+
+void custom_mount_free_all(CustomMount *l, size_t n) {
+ for (size_t i = 0; i < n; i++) {
+ CustomMount *m = l + i;
+
+ free(m->source);
+ free(m->destination);
+ free(m->options);
+
+ if (m->work_dir) {
+ (void) rm_rf(m->work_dir, REMOVE_ROOT|REMOVE_PHYSICAL);
+ free(m->work_dir);
+ }
+
+ if (m->rm_rf_tmpdir) {
+ (void) rm_rf(m->rm_rf_tmpdir, REMOVE_ROOT|REMOVE_PHYSICAL);
+ free(m->rm_rf_tmpdir);
+ }
+
+ strv_free(m->lower);
+ free(m->type_argument);
+ }
+
+ free(l);
+}
+
+static int custom_mount_compare(const CustomMount *a, const CustomMount *b) {
+ int r;
+
+ r = path_compare(a->destination, b->destination);
+ if (r != 0)
+ return r;
+
+ return CMP(a->type, b->type);
+}
+
+static int source_path_parse(const char *p, char **ret) {
+ assert(p);
+ assert(ret);
+
+ if (isempty(p))
+ return -EINVAL;
+
+ if (*p == '+') {
+ if (!path_is_absolute(p + 1))
+ return -EINVAL;
+
+ char *s = strdup(p);
+ if (!s)
+ return -ENOMEM;
+
+ *ret = TAKE_PTR(s);
+ return 0;
+ }
+
+ return path_make_absolute_cwd(p, ret);
+}
+
+static int source_path_parse_nullable(const char *p, char **ret) {
+ assert(p);
+ assert(ret);
+
+ if (isempty(p)) {
+ *ret = NULL;
+ return 0;
+ }
+
+ return source_path_parse(p, ret);
+}
+
+static char *resolve_source_path(const char *dest, const char *source) {
+ if (!source)
+ return NULL;
+
+ if (source[0] == '+')
+ return path_join(dest, source + 1);
+
+ return strdup(source);
+}
+
+static int allocate_temporary_source(CustomMount *m) {
+ assert(m);
+ assert(!m->source);
+ assert(!m->rm_rf_tmpdir);
+
+ m->rm_rf_tmpdir = strdup("/var/tmp/nspawn-temp-XXXXXX");
+ if (!m->rm_rf_tmpdir)
+ return log_oom();
+
+ if (!mkdtemp(m->rm_rf_tmpdir)) {
+ m->rm_rf_tmpdir = mfree(m->rm_rf_tmpdir);
+ return log_error_errno(errno, "Failed to acquire temporary directory: %m");
+ }
+
+ m->source = path_join(m->rm_rf_tmpdir, "src");
+ if (!m->source)
+ return log_oom();
+
+ if (mkdir(m->source, 0755) < 0)
+ return log_error_errno(errno, "Failed to create %s: %m", m->source);
+
+ return 0;
+}
+
+int custom_mount_prepare_all(const char *dest, CustomMount *l, size_t n) {
+ int r;
+
+ /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the
+ * parent process, so that we know the temporary directories to remove on exit before we fork off the
+ * children. */
+
+ assert(l || n == 0);
+
+ /* Order the custom mounts, and make sure we have a working directory */
+ typesafe_qsort(l, n, custom_mount_compare);
+
+ for (size_t i = 0; i < n; i++) {
+ CustomMount *m = l + i;
+
+ /* /proc we mount in the inner child, i.e. when we acquired CLONE_NEWPID. All other mounts we mount
+ * already in the outer child, so that the mounts are already established before CLONE_NEWPID and in
+ * particular CLONE_NEWUSER. This also means any custom mounts below /proc also need to be mounted in
+ * the inner child, not the outer one. Determine this here. */
+ m->in_userns = path_startswith(m->destination, "/proc");
+
+ if (m->type == CUSTOM_MOUNT_BIND) {
+ if (m->source) {
+ char *s;
+
+ s = resolve_source_path(dest, m->source);
+ if (!s)
+ return log_oom();
+
+ free_and_replace(m->source, s);
+ } else {
+ /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */
+
+ r = allocate_temporary_source(m);
+ if (r < 0)
+ return r;
+ }
+ }
+
+ if (m->type == CUSTOM_MOUNT_OVERLAY) {
+ STRV_FOREACH(j, m->lower) {
+ char *s;
+
+ s = resolve_source_path(dest, *j);
+ if (!s)
+ return log_oom();
+
+ free_and_replace(*j, s);
+ }
+
+ if (m->source) {
+ char *s;
+
+ s = resolve_source_path(dest, m->source);
+ if (!s)
+ return log_oom();
+
+ free_and_replace(m->source, s);
+ } else {
+ r = allocate_temporary_source(m);
+ if (r < 0)
+ return r;
+ }
+
+ if (m->work_dir) {
+ char *s;
+
+ s = resolve_source_path(dest, m->work_dir);
+ if (!s)
+ return log_oom();
+
+ free_and_replace(m->work_dir, s);
+ } else {
+ r = tempfn_random(m->source, NULL, &m->work_dir);
+ if (r < 0)
+ return log_error_errno(r, "Failed to acquire working directory: %m");
+ }
+
+ (void) mkdir_label(m->work_dir, 0700);
+ }
+ }
+
+ return 0;
+}
+
+int bind_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
+ _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL, *p = NULL;
+ CustomMount *m;
+ int r;
+
+ assert(l);
+ assert(n);
+
+ r = extract_many_words(&s, ":", EXTRACT_DONT_COALESCE_SEPARATORS, &source, &destination, NULL);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ return -EINVAL;
+ if (r == 1) {
+ destination = strdup(source[0] == '+' ? source+1 : source);
+ if (!destination)
+ return -ENOMEM;
+ }
+ if (r == 2 && !isempty(s)) {
+ opts = strdup(s);
+ if (!opts)
+ return -ENOMEM;
+ }
+
+ r = source_path_parse_nullable(source, &p);
+ if (r < 0)
+ return r;
+
+ if (!path_is_absolute(destination))
+ return -EINVAL;
+
+ m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND);
+ if (!m)
+ return -ENOMEM;
+
+ m->source = TAKE_PTR(p);
+ m->destination = TAKE_PTR(destination);
+ m->read_only = read_only;
+ m->options = TAKE_PTR(opts);
+
+ return 0;
+}
+
+int tmpfs_mount_parse(CustomMount **l, size_t *n, const char *s) {
+ _cleanup_free_ char *path = NULL, *opts = NULL;
+ const char *p = ASSERT_PTR(s);
+ CustomMount *m;
+ int r;
+
+ assert(l);
+ assert(n);
+
+ r = extract_first_word(&p, &path, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ return -EINVAL;
+
+ if (isempty(p))
+ opts = strdup("mode=0755");
+ else
+ opts = strdup(p);
+ if (!opts)
+ return -ENOMEM;
+
+ if (!path_is_absolute(path))
+ return -EINVAL;
+
+ m = custom_mount_add(l, n, CUSTOM_MOUNT_TMPFS);
+ if (!m)
+ return -ENOMEM;
+
+ m->destination = TAKE_PTR(path);
+ m->options = TAKE_PTR(opts);
+
+ return 0;
+}
+
+int overlay_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only) {
+ _cleanup_free_ char *upper = NULL, *destination = NULL;
+ _cleanup_strv_free_ char **lower = NULL;
+ CustomMount *m;
+ int r, k;
+
+ k = strv_split_full(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
+ if (k < 0)
+ return k;
+ if (k < 2)
+ return -EADDRNOTAVAIL;
+ if (k == 2) {
+ _cleanup_free_ char *p = NULL;
+
+ /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
+ * we'll also define the destination mount point the same as the upper. */
+
+ r = source_path_parse(lower[0], &p);
+ if (r < 0)
+ return r;
+
+ free_and_replace(lower[0], p);
+
+ r = source_path_parse(lower[1], &p);
+ if (r < 0)
+ return r;
+
+ free_and_replace(lower[1], p);
+
+ upper = TAKE_PTR(lower[1]);
+
+ destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */
+ if (!destination)
+ return -ENOMEM;
+ } else {
+ _cleanup_free_ char *p = NULL;
+
+ /* If more than two parameters are specified, the last one is the destination, the second to last one
+ * the "upper", and all before that the "lower" directories. */
+
+ destination = lower[k - 1];
+ upper = TAKE_PTR(lower[k - 2]);
+
+ STRV_FOREACH(i, lower) {
+ r = source_path_parse(*i, &p);
+ if (r < 0)
+ return r;
+
+ free_and_replace(*i, p);
+ }
+
+ /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory
+ * in /var/tmp */
+ r = source_path_parse_nullable(upper, &p);
+ if (r < 0)
+ return r;
+
+ free_and_replace(upper, p);
+
+ if (!path_is_absolute(destination))
+ return -EINVAL;
+ }
+
+ m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY);
+ if (!m)
+ return -ENOMEM;
+
+ m->destination = TAKE_PTR(destination);
+ m->source = TAKE_PTR(upper);
+ m->lower = TAKE_PTR(lower);
+ m->read_only = read_only;
+
+ return 0;
+}
+
+int inaccessible_mount_parse(CustomMount **l, size_t *n, const char *s) {
+ _cleanup_free_ char *path = NULL;
+ CustomMount *m;
+
+ assert(l);
+ assert(n);
+ assert(s);
+
+ if (!path_is_absolute(s))
+ return -EINVAL;
+
+ path = strdup(s);
+ if (!path)
+ return -ENOMEM;
+
+ m = custom_mount_add(l, n, CUSTOM_MOUNT_INACCESSIBLE);
+ if (!m)
+ return -ENOMEM;
+
+ m->destination = TAKE_PTR(path);
+ return 0;
+}
+
+int tmpfs_patch_options(
+ const char *options,
+ uid_t uid_shift,
+ const char *selinux_apifs_context,
+ char **ret) {
+
+ _cleanup_free_ char *buf = NULL;
+
+ assert(ret);
+
+ if (options) {
+ buf = strdup(options);
+ if (!buf)
+ return -ENOMEM;
+ }
+
+ if (uid_shift != UID_INVALID)
+ if (strextendf_with_separator(&buf, ",", "uid=" UID_FMT ",gid=" UID_FMT, uid_shift, uid_shift) < 0)
+ return -ENOMEM;
+
+#if HAVE_SELINUX
+ if (selinux_apifs_context)
+ if (strextendf_with_separator(&buf, ",", "context=\"%s\"", selinux_apifs_context) < 0)
+ return -ENOMEM;
+#endif
+
+ *ret = TAKE_PTR(buf);
+ return !!*ret;
+}
+
+int mount_sysfs(const char *dest, MountSettingsMask mount_settings) {
+ const char *full, *top;
+ int r;
+ unsigned long extra_flags = 0;
+
+ top = prefix_roota(dest, "/sys");
+ r = path_is_fs_type(top, SYSFS_MAGIC);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine filesystem type of %s: %m", top);
+ /* /sys might already be mounted as sysfs by the outer child in the
+ * !netns case. In this case, it's all good. Don't touch it because we
+ * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
+ */
+ if (r > 0)
+ return 0;
+
+ full = prefix_roota(top, "/full");
+
+ (void) mkdir(full, 0755);
+
+ if (FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO))
+ extra_flags |= MS_RDONLY;
+
+ r = mount_nofollow_verbose(LOG_ERR, "sysfs", full, "sysfs",
+ MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL);
+ if (r < 0)
+ return r;
+
+ FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") {
+ _cleanup_free_ char *from = NULL, *to = NULL;
+
+ from = path_join(full, x);
+ if (!from)
+ return log_oom();
+
+ to = path_join(top, x);
+ if (!to)
+ return log_oom();
+
+ (void) mkdir(to, 0755);
+
+ r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
+ if (r < 0)
+ return r;
+
+ r = mount_nofollow_verbose(LOG_ERR, NULL, to, NULL,
+ MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
+ if (r < 0)
+ return r;
+ }
+
+ r = umount_verbose(LOG_ERR, full, UMOUNT_NOFOLLOW);
+ if (r < 0)
+ return r;
+
+ if (rmdir(full) < 0)
+ return log_error_errno(errno, "Failed to remove %s: %m", full);
+
+ /* Create mountpoint for cgroups. Otherwise we are not allowed since we
+ * remount /sys read-only.
+ */
+ const char *x = prefix_roota(top, "/fs/cgroup");
+ (void) mkdir_p(x, 0755);
+
+ return mount_nofollow_verbose(LOG_ERR, NULL, top, NULL,
+ MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL);
+}
+
+int mount_all(const char *dest,
+ MountSettingsMask mount_settings,
+ uid_t uid_shift,
+ const char *selinux_apifs_context) {
+
+#define PROC_INACCESSIBLE_REG(path) \
+ { "/run/systemd/inaccessible/reg", (path), NULL, NULL, MS_BIND, \
+ MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
+ { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
+ MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
+
+#define PROC_READ_ONLY(path) \
+ { (path), (path), NULL, NULL, MS_BIND, \
+ MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
+ { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
+ MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
+
+ typedef struct MountPoint {
+ const char *what;
+ const char *where;
+ const char *type;
+ const char *options;
+ unsigned long flags;
+ MountSettingsMask mount_settings;
+ } MountPoint;
+
+ static const MountPoint mount_table[] = {
+ /* First we list inner child mounts (i.e. mounts applied *after* entering user namespacing) */
+ { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_MKDIR|MOUNT_FOLLOW_SYMLINKS }, /* we follow symlinks here since not following them requires /proc/ already being mounted, which we don't have here. */
+
+ { "/proc/sys", "/proc/sys", NULL, NULL, MS_BIND,
+ MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */
+
+ { "/proc/sys/net", "/proc/sys/net", NULL, NULL, MS_BIND,
+ MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */
+
+ { NULL, "/proc/sys", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
+ MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* ... then, make it r/o */
+
+ /* Make these files inaccessible to container payloads: they potentially leak information about kernel
+ * internals or the host's execution environment to the container */
+ PROC_INACCESSIBLE_REG("/proc/kallsyms"),
+ PROC_INACCESSIBLE_REG("/proc/kcore"),
+ PROC_INACCESSIBLE_REG("/proc/keys"),
+ PROC_INACCESSIBLE_REG("/proc/sysrq-trigger"),
+ PROC_INACCESSIBLE_REG("/proc/timer_list"),
+
+ /* Make these directories read-only to container payloads: they show hardware information, and in some
+ * cases contain tunables the container really shouldn't have access to. */
+ PROC_READ_ONLY("/proc/acpi"),
+ PROC_READ_ONLY("/proc/apm"),
+ PROC_READ_ONLY("/proc/asound"),
+ PROC_READ_ONLY("/proc/bus"),
+ PROC_READ_ONLY("/proc/fs"),
+ PROC_READ_ONLY("/proc/irq"),
+ PROC_READ_ONLY("/proc/scsi"),
+
+ { "mqueue", "/dev/mqueue", "mqueue", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ MOUNT_IN_USERNS|MOUNT_MKDIR },
+
+ /* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing) */
+ { "tmpfs", "/tmp", "tmpfs", "mode=1777" NESTED_TMPFS_LIMITS, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ MOUNT_FATAL|MOUNT_APPLY_TMPFS_TMP|MOUNT_MKDIR },
+ { "tmpfs", "/sys", "tmpfs", "mode=555" TMPFS_LIMITS_SYS, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS|MOUNT_MKDIR },
+ { "sysfs", "/sys", "sysfs", NULL, MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO|MOUNT_MKDIR }, /* skipped if above was mounted */
+ { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ MOUNT_FATAL|MOUNT_MKDIR }, /* skipped if above was mounted */
+ { "tmpfs", "/dev", "tmpfs", "mode=755" TMPFS_LIMITS_PRIVATE_DEV, MS_NOSUID|MS_STRICTATIME,
+ MOUNT_FATAL|MOUNT_MKDIR },
+ { "tmpfs", "/dev/shm", "tmpfs", "mode=1777" NESTED_TMPFS_LIMITS, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ MOUNT_FATAL|MOUNT_MKDIR },
+ { "tmpfs", "/run", "tmpfs", "mode=755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ MOUNT_FATAL|MOUNT_MKDIR },
+ { "/run/host", "/run/host", NULL, NULL, MS_BIND,
+ MOUNT_FATAL|MOUNT_MKDIR|MOUNT_PREFIX_ROOT }, /* Prepare this so that we can make it read-only when we are done */
+ { "/etc/os-release", "/run/host/os-release", NULL, NULL, MS_BIND,
+ MOUNT_TOUCH }, /* As per kernel interface requirements, bind mount first (creating mount points) and make read-only later */
+ { "/usr/lib/os-release", "/run/host/os-release", NULL, NULL, MS_BIND,
+ MOUNT_FATAL }, /* If /etc/os-release doesn't exist use the version in /usr/lib as fallback */
+ { NULL, "/run/host/os-release", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
+ MOUNT_FATAL },
+ { NULL, "/run/host", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
+ MOUNT_FATAL|MOUNT_IN_USERNS },
+#if HAVE_SELINUX
+ { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND,
+ MOUNT_MKDIR }, /* Bind mount first (mkdir/chown the mount point in case /sys/ is mounted as minimal skeleton tmpfs) */
+ { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT,
+ 0 }, /* Then, make it r/o (don't mkdir/chown the mount point here, the previous entry already did that) */
+#endif
+ };
+
+ bool use_userns = FLAGS_SET(mount_settings, MOUNT_USE_USERNS);
+ bool netns = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_NETNS);
+ bool ro = FLAGS_SET(mount_settings, MOUNT_APPLY_APIVFS_RO);
+ bool in_userns = FLAGS_SET(mount_settings, MOUNT_IN_USERNS);
+ bool tmpfs_tmp = FLAGS_SET(mount_settings, MOUNT_APPLY_TMPFS_TMP);
+ int r;
+
+ for (size_t k = 0; k < ELEMENTSOF(mount_table); k++) {
+ _cleanup_free_ char *where = NULL, *options = NULL, *prefixed = NULL;
+ bool fatal = FLAGS_SET(mount_table[k].mount_settings, MOUNT_FATAL);
+ const char *o;
+
+ if (in_userns != FLAGS_SET(mount_table[k].mount_settings, MOUNT_IN_USERNS))
+ continue;
+
+ if (!netns && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_NETNS))
+ continue;
+
+ if (!ro && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_APIVFS_RO))
+ continue;
+
+ if (!tmpfs_tmp && FLAGS_SET(mount_table[k].mount_settings, MOUNT_APPLY_TMPFS_TMP))
+ continue;
+
+ r = chase_symlinks(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve %s%s: %m", strempty(dest), mount_table[k].where);
+
+ /* Skip this entry if it is not a remount. */
+ if (mount_table[k].what) {
+ r = path_is_mount_point(where, NULL, 0);
+ if (r < 0 && r != -ENOENT)
+ return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where);
+ if (r > 0)
+ continue;
+ }
+
+ if ((mount_table[k].mount_settings & (MOUNT_MKDIR|MOUNT_TOUCH)) != 0) {
+ uid_t u = (use_userns && !in_userns) ? uid_shift : UID_INVALID;
+
+ if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH))
+ r = mkdir_parents_safe(dest, where, 0755, u, u, 0);
+ else
+ r = mkdir_p_safe(dest, where, 0755, u, u, 0);
+ if (r < 0 && r != -EEXIST) {
+ if (fatal && r != -EROFS)
+ return log_error_errno(r, "Failed to create directory %s: %m", where);
+
+ log_debug_errno(r, "Failed to create directory %s: %m", where);
+
+ /* If we failed mkdir() or chown() due to the root directory being read only,
+ * attempt to mount this fs anyway and let mount_verbose log any errors */
+ if (r != -EROFS)
+ continue;
+ }
+ }
+
+ if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_TOUCH)) {
+ r = touch(where);
+ if (r < 0 && r != -EEXIST) {
+ if (fatal && r != -EROFS)
+ return log_error_errno(r, "Failed to create file %s: %m", where);
+
+ log_debug_errno(r, "Failed to create file %s: %m", where);
+ if (r != -EROFS)
+ continue;
+ }
+ }
+
+ o = mount_table[k].options;
+ if (streq_ptr(mount_table[k].type, "tmpfs")) {
+ r = tmpfs_patch_options(o, in_userns ? 0 : uid_shift, selinux_apifs_context, &options);
+ if (r < 0)
+ return log_oom();
+ if (r > 0)
+ o = options;
+ }
+
+ if (FLAGS_SET(mount_table[k].mount_settings, MOUNT_PREFIX_ROOT)) {
+ /* Optionally prefix the mount source with the root dir. This is useful in bind
+ * mounts to be created within the container image before we transition into it. Note
+ * that MOUNT_IN_USERNS is run after we transitioned hence prefixing is not ncessary
+ * for those. */
+ r = chase_symlinks(mount_table[k].what, dest, CHASE_PREFIX_ROOT, &prefixed, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve %s%s: %m", strempty(dest), mount_table[k].what);
+ }
+
+ r = mount_verbose_full(
+ fatal ? LOG_ERR : LOG_DEBUG,
+ prefixed ?: mount_table[k].what,
+ where,
+ mount_table[k].type,
+ mount_table[k].flags,
+ o,
+ FLAGS_SET(mount_table[k].mount_settings, MOUNT_FOLLOW_SYMLINKS));
+ if (r < 0 && fatal)
+ return r;
+ }
+
+ return 0;
+}
+
+static int parse_mount_bind_options(const char *options, unsigned long *mount_flags, char **mount_opts, RemountIdmapping *idmapping) {
+ unsigned long flags = *mount_flags;
+ char *opts = NULL;
+ RemountIdmapping new_idmapping = *idmapping;
+ int r;
+
+ assert(options);
+
+ for (;;) {
+ _cleanup_free_ char *word = NULL;
+
+ r = extract_first_word(&options, &word, ",", 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to extract mount option: %m");
+ if (r == 0)
+ break;
+
+ if (streq(word, "rbind"))
+ flags |= MS_REC;
+ else if (streq(word, "norbind"))
+ flags &= ~MS_REC;
+ else if (streq(word, "idmap"))
+ new_idmapping = REMOUNT_IDMAPPING_HOST_ROOT;
+ else if (streq(word, "noidmap"))
+ new_idmapping = REMOUNT_IDMAPPING_NONE;
+ else if (streq(word, "rootidmap"))
+ new_idmapping = REMOUNT_IDMAPPING_HOST_OWNER;
+ else
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Invalid bind mount option: %s", word);
+ }
+
+ *mount_flags = flags;
+ *idmapping = new_idmapping;
+ /* in the future mount_opts will hold string options for mount(2) */
+ *mount_opts = opts;
+
+ return 0;
+}
+
+static int mount_bind(const char *dest, CustomMount *m, uid_t uid_shift, uid_t uid_range) {
+ _cleanup_free_ char *mount_opts = NULL, *where = NULL;
+ unsigned long mount_flags = MS_BIND | MS_REC;
+ struct stat source_st, dest_st;
+ int r;
+ RemountIdmapping idmapping = REMOUNT_IDMAPPING_NONE;
+
+ assert(dest);
+ assert(m);
+
+ if (m->options) {
+ r = parse_mount_bind_options(m->options, &mount_flags, &mount_opts, &idmapping);
+ if (r < 0)
+ return r;
+ }
+
+ /* If this is a bind mount from a temporary sources change ownership of the source to the container's
+ * root UID. Otherwise it would always show up as "nobody" if user namespacing is used. */
+ if (m->rm_rf_tmpdir && chown(m->source, uid_shift, uid_shift) < 0)
+ return log_error_errno(errno, "Failed to chown %s: %m", m->source);
+
+ if (stat(m->source, &source_st) < 0)
+ return log_error_errno(errno, "Failed to stat %s: %m", m->source);
+
+ r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
+ if (r > 0) { /* Path exists already? */
+
+ if (stat(where, &dest_st) < 0)
+ return log_error_errno(errno, "Failed to stat %s: %m", where);
+
+ if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Cannot bind mount directory %s on file %s.",
+ m->source, where);
+
+ if (!S_ISDIR(source_st.st_mode) && S_ISDIR(dest_st.st_mode))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Cannot bind mount file %s on directory %s.",
+ m->source, where);
+
+ } else { /* Path doesn't exist yet? */
+ r = mkdir_parents_label(where, 0755);
+ if (r < 0)
+ return log_error_errno(r, "Failed to make parents of %s: %m", where);
+
+ /* Create the mount point. Any non-directory file can be
+ * mounted on any non-directory file (regular, fifo, socket,
+ * char, block).
+ */
+ if (S_ISDIR(source_st.st_mode))
+ r = mkdir_label(where, 0755);
+ else
+ r = touch(where);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create mount point %s: %m", where);
+ }
+
+ r = mount_nofollow_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts);
+ if (r < 0)
+ return r;
+
+ if (m->read_only) {
+ r = bind_remount_recursive(where, MS_RDONLY, MS_RDONLY, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Read-only bind mount failed: %m");
+ }
+
+ if (idmapping != REMOUNT_IDMAPPING_NONE) {
+ r = remount_idmap(where, uid_shift, uid_range, source_st.st_uid, idmapping);
+ if (r < 0)
+ return log_error_errno(r, "Failed to map ids for bind mount %s: %m", where);
+ }
+
+ return 0;
+}
+
+static int mount_tmpfs(const char *dest, CustomMount *m, uid_t uid_shift, const char *selinux_apifs_context) {
+ const char *options;
+ _cleanup_free_ char *buf = NULL, *where = NULL;
+ int r;
+
+ assert(dest);
+ assert(m);
+
+ r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
+ if (r == 0) { /* Doesn't exist yet? */
+ r = mkdir_p_label(where, 0755);
+ if (r < 0)
+ return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where);
+ }
+
+ r = tmpfs_patch_options(m->options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
+ if (r < 0)
+ return log_oom();
+ options = r > 0 ? buf : m->options;
+
+ return mount_nofollow_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options);
+}
+
+static char *joined_and_escaped_lower_dirs(char **lower) {
+ _cleanup_strv_free_ char **sv = NULL;
+
+ sv = strv_copy(lower);
+ if (!sv)
+ return NULL;
+
+ strv_reverse(sv);
+
+ if (!strv_shell_escape(sv, ",:"))
+ return NULL;
+
+ return strv_join(sv, ":");
+}
+
+static int mount_overlay(const char *dest, CustomMount *m) {
+ _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL;
+ const char *options;
+ int r;
+
+ assert(dest);
+ assert(m);
+
+ r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
+ if (r == 0) { /* Doesn't exist yet? */
+ r = mkdir_label(where, 0755);
+ if (r < 0)
+ return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where);
+ }
+
+ (void) mkdir_p_label(m->source, 0755);
+
+ lower = joined_and_escaped_lower_dirs(m->lower);
+ if (!lower)
+ return log_oom();
+
+ escaped_source = shell_escape(m->source, ",:");
+ if (!escaped_source)
+ return log_oom();
+
+ if (m->read_only)
+ options = strjoina("lowerdir=", escaped_source, ":", lower);
+ else {
+ _cleanup_free_ char *escaped_work_dir = NULL;
+
+ escaped_work_dir = shell_escape(m->work_dir, ",:");
+ if (!escaped_work_dir)
+ return log_oom();
+
+ options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir);
+ }
+
+ return mount_nofollow_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options);
+}
+
+static int mount_inaccessible(const char *dest, CustomMount *m) {
+ _cleanup_free_ char *where = NULL, *source = NULL;
+ struct stat st;
+ int r;
+
+ assert(dest);
+ assert(m);
+
+ r = chase_symlinks_and_stat(m->destination, dest, CHASE_PREFIX_ROOT, &where, &st, NULL);
+ if (r < 0) {
+ log_full_errno(m->graceful ? LOG_DEBUG : LOG_ERR, r, "Failed to resolve %s/%s: %m", dest, m->destination);
+ return m->graceful ? 0 : r;
+ }
+
+ r = mode_to_inaccessible_node(NULL, st.st_mode, &source);
+ if (r < 0)
+ return m->graceful ? 0 : r;
+
+ r = mount_nofollow_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, source, where, NULL, MS_BIND, NULL);
+ if (r < 0)
+ return m->graceful ? 0 : r;
+
+ r = mount_nofollow_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, NULL, where, NULL, MS_BIND|MS_RDONLY|MS_REMOUNT, NULL);
+ if (r < 0) {
+ (void) umount_verbose(m->graceful ? LOG_DEBUG : LOG_ERR, where, UMOUNT_NOFOLLOW);
+ return m->graceful ? 0 : r;
+ }
+
+ return 0;
+}
+
+static int mount_arbitrary(const char *dest, CustomMount *m) {
+ _cleanup_free_ char *where = NULL;
+ int r;
+
+ assert(dest);
+ assert(m);
+
+ r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination);
+ if (r == 0) { /* Doesn't exist yet? */
+ r = mkdir_p_label(where, 0755);
+ if (r < 0)
+ return log_error_errno(r, "Creating mount point for mount %s failed: %m", where);
+ }
+
+ return mount_nofollow_verbose(LOG_ERR, m->source, where, m->type_argument, 0, m->options);
+}
+
+int mount_custom(
+ const char *dest,
+ CustomMount *mounts, size_t n,
+ uid_t uid_shift,
+ uid_t uid_range,
+ const char *selinux_apifs_context,
+ MountSettingsMask mount_settings) {
+ int r;
+
+ assert(dest);
+
+ for (size_t i = 0; i < n; i++) {
+ CustomMount *m = mounts + i;
+
+ if (FLAGS_SET(mount_settings, MOUNT_IN_USERNS) != m->in_userns)
+ continue;
+
+ if (FLAGS_SET(mount_settings, MOUNT_ROOT_ONLY) && !path_equal(m->destination, "/"))
+ continue;
+
+ if (FLAGS_SET(mount_settings, MOUNT_NON_ROOT_ONLY) && path_equal(m->destination, "/"))
+ continue;
+
+ switch (m->type) {
+
+ case CUSTOM_MOUNT_BIND:
+ r = mount_bind(dest, m, uid_shift, uid_range);
+ break;
+
+ case CUSTOM_MOUNT_TMPFS:
+ r = mount_tmpfs(dest, m, uid_shift, selinux_apifs_context);
+ break;
+
+ case CUSTOM_MOUNT_OVERLAY:
+ r = mount_overlay(dest, m);
+ break;
+
+ case CUSTOM_MOUNT_INACCESSIBLE:
+ r = mount_inaccessible(dest, m);
+ break;
+
+ case CUSTOM_MOUNT_ARBITRARY:
+ r = mount_arbitrary(dest, m);
+ break;
+
+ default:
+ assert_not_reached();
+ }
+
+ if (r < 0)
+ return r;
+ }
+
+ return 0;
+}
+
+bool has_custom_root_mount(const CustomMount *mounts, size_t n) {
+ for (size_t i = 0; i < n; i++)
+ if (path_equal(mounts[i].destination, "/"))
+ return true;
+
+ return false;
+}
+
+static int setup_volatile_state(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
+ _cleanup_free_ char *buf = NULL;
+ const char *p, *options;
+ int r;
+
+ assert(directory);
+
+ /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */
+
+ r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to remount %s read-only: %m", directory);
+
+ p = prefix_roota(directory, "/var");
+ r = mkdir(p, 0755);
+ if (r < 0 && errno != EEXIST)
+ return log_error_errno(errno, "Failed to create %s: %m", directory);
+
+ options = "mode=755" TMPFS_LIMITS_VOLATILE_STATE;
+ r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
+ if (r < 0)
+ return log_oom();
+ if (r > 0)
+ options = buf;
+
+ return mount_nofollow_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options);
+}
+
+static int setup_volatile_yes(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
+ bool tmpfs_mounted = false, bind_mounted = false;
+ char template[] = "/tmp/nspawn-volatile-XXXXXX";
+ _cleanup_free_ char *buf = NULL, *bindir = NULL;
+ const char *f, *t, *options;
+ struct stat st;
+ int r;
+
+ assert(directory);
+
+ /* --volatile=yes means we mount a tmpfs to the root dir, and the original /usr to use inside it, and
+ * that read-only. Before we start setting this up let's validate if the image has the /usr merge
+ * implemented, and let's output a friendly log message if it hasn't. */
+
+ bindir = path_join(directory, "/bin");
+ if (!bindir)
+ return log_oom();
+ if (lstat(bindir, &st) < 0) {
+ if (errno != ENOENT)
+ return log_error_errno(errno, "Failed to stat /bin directory below image: %m");
+
+ /* ENOENT is fine, just means the image is probably just a naked /usr and we can create the
+ * rest. */
+ } else if (S_ISDIR(st.st_mode))
+ return log_error_errno(SYNTHETIC_ERRNO(EISDIR),
+ "Sorry, --volatile=yes mode is not supported with OS images that have not merged /bin/, /sbin/, /lib/, /lib64/ into /usr/. "
+ "Please work with your distribution and help them adopt the merged /usr scheme.");
+ else if (!S_ISLNK(st.st_mode))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Error starting image: if --volatile=yes is used /bin must be a symlink (for merged /usr support) or non-existent (in which case a symlink is created automatically).");
+
+ if (!mkdtemp(template))
+ return log_error_errno(errno, "Failed to create temporary directory: %m");
+
+ options = "mode=755" TMPFS_LIMITS_ROOTFS;
+ r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
+ if (r < 0)
+ goto fail;
+ if (r > 0)
+ options = buf;
+
+ r = mount_nofollow_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
+ if (r < 0)
+ goto fail;
+
+ tmpfs_mounted = true;
+
+ f = prefix_roota(directory, "/usr");
+ t = prefix_roota(template, "/usr");
+
+ r = mkdir(t, 0755);
+ if (r < 0 && errno != EEXIST) {
+ r = log_error_errno(errno, "Failed to create %s: %m", t);
+ goto fail;
+ }
+
+ r = mount_nofollow_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL);
+ if (r < 0)
+ goto fail;
+
+ bind_mounted = true;
+
+ r = bind_remount_recursive(t, MS_RDONLY, MS_RDONLY, NULL);
+ if (r < 0) {
+ log_error_errno(r, "Failed to remount %s read-only: %m", t);
+ goto fail;
+ }
+
+ r = mount_nofollow_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL);
+ if (r < 0)
+ goto fail;
+
+ (void) rmdir(template);
+
+ return 0;
+
+fail:
+ if (bind_mounted)
+ (void) umount_verbose(LOG_ERR, t, UMOUNT_NOFOLLOW);
+
+ if (tmpfs_mounted)
+ (void) umount_verbose(LOG_ERR, template, UMOUNT_NOFOLLOW);
+
+ (void) rmdir(template);
+ return r;
+}
+
+static int setup_volatile_overlay(const char *directory, uid_t uid_shift, const char *selinux_apifs_context) {
+ _cleanup_free_ char *buf = NULL, *escaped_directory = NULL, *escaped_upper = NULL, *escaped_work = NULL;
+ char template[] = "/tmp/nspawn-volatile-XXXXXX";
+ const char *upper, *work, *options;
+ bool tmpfs_mounted = false;
+ int r;
+
+ assert(directory);
+
+ /* --volatile=overlay means we mount an overlayfs to the root dir. */
+
+ if (!mkdtemp(template))
+ return log_error_errno(errno, "Failed to create temporary directory: %m");
+
+ options = "mode=755" TMPFS_LIMITS_ROOTFS;
+ r = tmpfs_patch_options(options, uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &buf);
+ if (r < 0)
+ goto finish;
+ if (r > 0)
+ options = buf;
+
+ r = mount_nofollow_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options);
+ if (r < 0)
+ goto finish;
+
+ tmpfs_mounted = true;
+
+ upper = strjoina(template, "/upper");
+ work = strjoina(template, "/work");
+
+ if (mkdir(upper, 0755) < 0) {
+ r = log_error_errno(errno, "Failed to create %s: %m", upper);
+ goto finish;
+ }
+ if (mkdir(work, 0755) < 0) {
+ r = log_error_errno(errno, "Failed to create %s: %m", work);
+ goto finish;
+ }
+
+ /* And now, let's overmount the root dir with an overlayfs that uses the root dir as lower dir. It's kinda nice
+ * that the kernel allows us to do that without going through some mount point rearrangements. */
+
+ escaped_directory = shell_escape(directory, ",:");
+ escaped_upper = shell_escape(upper, ",:");
+ escaped_work = shell_escape(work, ",:");
+ if (!escaped_directory || !escaped_upper || !escaped_work) {
+ r = -ENOMEM;
+ goto finish;
+ }
+
+ options = strjoina("lowerdir=", escaped_directory, ",upperdir=", escaped_upper, ",workdir=", escaped_work);
+ r = mount_nofollow_verbose(LOG_ERR, "overlay", directory, "overlay", 0, options);
+
+finish:
+ if (tmpfs_mounted)
+ (void) umount_verbose(LOG_ERR, template, UMOUNT_NOFOLLOW);
+
+ (void) rmdir(template);
+ return r;
+}
+
+int setup_volatile_mode(
+ const char *directory,
+ VolatileMode mode,
+ uid_t uid_shift,
+ const char *selinux_apifs_context) {
+
+ switch (mode) {
+
+ case VOLATILE_YES:
+ return setup_volatile_yes(directory, uid_shift, selinux_apifs_context);
+
+ case VOLATILE_STATE:
+ return setup_volatile_state(directory, uid_shift, selinux_apifs_context);
+
+ case VOLATILE_OVERLAY:
+ return setup_volatile_overlay(directory, uid_shift, selinux_apifs_context);
+
+ default:
+ return 0;
+ }
+}
+
+/* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
+int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s) {
+ _cleanup_free_ char *root_new = NULL, *root_old = NULL;
+ const char *p = s;
+ int r;
+
+ assert(pivot_root_new);
+ assert(pivot_root_old);
+
+ r = extract_first_word(&p, &root_new, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
+ if (r < 0)
+ return r;
+ if (r == 0)
+ return -EINVAL;
+
+ if (isempty(p))
+ root_old = NULL;
+ else {
+ root_old = strdup(p);
+ if (!root_old)
+ return -ENOMEM;
+ }
+
+ if (!path_is_absolute(root_new))
+ return -EINVAL;
+ if (root_old && !path_is_absolute(root_old))
+ return -EINVAL;
+
+ free_and_replace(*pivot_root_new, root_new);
+ free_and_replace(*pivot_root_old, root_old);
+
+ return 0;
+}
+
+int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old) {
+ _cleanup_free_ char *directory_pivot_root_new = NULL;
+ _cleanup_free_ char *pivot_tmp_pivot_root_old = NULL;
+ char pivot_tmp[] = "/tmp/nspawn-pivot-XXXXXX";
+ bool remove_pivot_tmp = false;
+ int r;
+
+ assert(directory);
+
+ if (!pivot_root_new)
+ return 0;
+
+ /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
+ * If pivot_root_old is NULL, the existing / disappears.
+ * This requires a temporary directory, pivot_tmp, which is
+ * not a child of either.
+ *
+ * This is typically used for OSTree-style containers, where the root partition contains several
+ * sysroots which could be run. Normally, one would be chosen by the bootloader and pivoted to / by
+ * initrd.
+ *
+ * For example, for an OSTree deployment, pivot_root_new
+ * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
+ * code doesn’t do the /var mount which OSTree expects: use
+ * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
+ *
+ * So in the OSTree case, we’ll end up with something like:
+ * - directory = /tmp/nspawn-root-123456
+ * - pivot_root_new = /ostree/deploy/os/deploy/123abc
+ * - pivot_root_old = /sysroot
+ * - directory_pivot_root_new =
+ * /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
+ * - pivot_tmp = /tmp/nspawn-pivot-123456
+ * - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
+ *
+ * Requires all file systems at directory and below to be mounted
+ * MS_PRIVATE or MS_SLAVE so they can be moved.
+ */
+ directory_pivot_root_new = path_join(directory, pivot_root_new);
+ if (!directory_pivot_root_new)
+ return log_oom();
+
+ /* Remount directory_pivot_root_new to make it movable. */
+ r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, directory_pivot_root_new, NULL, MS_BIND, NULL);
+ if (r < 0)
+ goto done;
+
+ if (pivot_root_old) {
+ if (!mkdtemp(pivot_tmp)) {
+ r = log_error_errno(errno, "Failed to create temporary directory: %m");
+ goto done;
+ }
+
+ remove_pivot_tmp = true;
+ pivot_tmp_pivot_root_old = path_join(pivot_tmp, pivot_root_old);
+ if (!pivot_tmp_pivot_root_old) {
+ r = log_oom();
+ goto done;
+ }
+
+ r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, pivot_tmp, NULL, MS_MOVE, NULL);
+ if (r < 0)
+ goto done;
+
+ r = mount_nofollow_verbose(LOG_ERR, directory, pivot_tmp_pivot_root_old, NULL, MS_MOVE, NULL);
+ if (r < 0)
+ goto done;
+
+ r = mount_nofollow_verbose(LOG_ERR, pivot_tmp, directory, NULL, MS_MOVE, NULL);
+ if (r < 0)
+ goto done;
+ } else {
+ r = mount_nofollow_verbose(LOG_ERR, directory_pivot_root_new, directory, NULL, MS_MOVE, NULL);
+ if (r < 0)
+ goto done;
+ }
+
+done:
+ if (remove_pivot_tmp)
+ (void) rmdir(pivot_tmp);
+
+ return r;
+}
diff --git a/src/nspawn/nspawn-mount.h b/src/nspawn/nspawn-mount.h
new file mode 100644
index 0000000..6bedbf9
--- /dev/null
+++ b/src/nspawn/nspawn-mount.h
@@ -0,0 +1,69 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <stdbool.h>
+
+#include "cgroup-util.h"
+#include "volatile-util.h"
+
+typedef enum MountSettingsMask {
+ MOUNT_FATAL = 1 << 0, /* if set, a mount error is considered fatal */
+ MOUNT_USE_USERNS = 1 << 1, /* if set, mounts are patched considering uid/gid shifts in a user namespace */
+ MOUNT_IN_USERNS = 1 << 2, /* if set, the mount is executed in the inner child, otherwise in the outer child */
+ MOUNT_APPLY_APIVFS_RO = 1 << 3, /* if set, /proc/sys, and /sys will be mounted read-only, otherwise read-write. */
+ MOUNT_APPLY_APIVFS_NETNS = 1 << 4, /* if set, /proc/sys/net will be mounted read-write.
+ Works only if MOUNT_APPLY_APIVFS_RO is also set. */
+ MOUNT_APPLY_TMPFS_TMP = 1 << 5, /* if set, /tmp will be mounted as tmpfs */
+ MOUNT_ROOT_ONLY = 1 << 6, /* if set, only root mounts are mounted */
+ MOUNT_NON_ROOT_ONLY = 1 << 7, /* if set, only non-root mounts are mounted */
+ MOUNT_MKDIR = 1 << 8, /* if set, make directory to mount over first */
+ MOUNT_TOUCH = 1 << 9, /* if set, touch file to mount over first */
+ MOUNT_PREFIX_ROOT = 1 << 10,/* if set, prefix the source path with the container's root directory */
+ MOUNT_FOLLOW_SYMLINKS = 1 << 11,/* if set, we'll follow symlinks for the mount target */
+} MountSettingsMask;
+
+typedef enum CustomMountType {
+ CUSTOM_MOUNT_BIND,
+ CUSTOM_MOUNT_TMPFS,
+ CUSTOM_MOUNT_OVERLAY,
+ CUSTOM_MOUNT_INACCESSIBLE,
+ CUSTOM_MOUNT_ARBITRARY,
+ _CUSTOM_MOUNT_TYPE_MAX,
+ _CUSTOM_MOUNT_TYPE_INVALID = -EINVAL,
+} CustomMountType;
+
+typedef struct CustomMount {
+ CustomMountType type;
+ bool read_only;
+ char *source; /* for overlayfs this is the upper directory */
+ char *destination;
+ char *options;
+ char *work_dir;
+ char **lower;
+ char *rm_rf_tmpdir;
+ char *type_argument; /* only for CUSTOM_MOUNT_ARBITRARY */
+ bool graceful;
+ bool in_userns;
+} CustomMount;
+
+CustomMount* custom_mount_add(CustomMount **l, size_t *n, CustomMountType t);
+void custom_mount_free_all(CustomMount *l, size_t n);
+int custom_mount_prepare_all(const char *dest, CustomMount *l, size_t n);
+
+int bind_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only);
+int tmpfs_mount_parse(CustomMount **l, size_t *n, const char *s);
+int overlay_mount_parse(CustomMount **l, size_t *n, const char *s, bool read_only);
+int inaccessible_mount_parse(CustomMount **l, size_t *n, const char *s);
+
+int mount_all(const char *dest, MountSettingsMask mount_settings, uid_t uid_shift, const char *selinux_apifs_context);
+int mount_sysfs(const char *dest, MountSettingsMask mount_settings);
+
+int mount_custom(const char *dest, CustomMount *mounts, size_t n, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context, MountSettingsMask mount_settings);
+bool has_custom_root_mount(const CustomMount *mounts, size_t n);
+
+int setup_volatile_mode(const char *directory, VolatileMode mode, uid_t uid_shift, const char *selinux_apifs_context);
+
+int pivot_root_parse(char **pivot_root_new, char **pivot_root_old, const char *s);
+int setup_pivot_root(const char *directory, const char *pivot_root_new, const char *pivot_root_old);
+
+int tmpfs_patch_options(const char *options,uid_t uid_shift, const char *selinux_apifs_context, char **ret);
diff --git a/src/nspawn/nspawn-network.c b/src/nspawn/nspawn-network.c
new file mode 100644
index 0000000..536e369
--- /dev/null
+++ b/src/nspawn/nspawn-network.c
@@ -0,0 +1,745 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <net/if.h>
+#include <linux/if.h>
+#include <linux/veth.h>
+#include <sys/file.h>
+
+#include "sd-device.h"
+#include "sd-id128.h"
+#include "sd-netlink.h"
+
+#include "alloc-util.h"
+#include "ether-addr-util.h"
+#include "hexdecoct.h"
+#include "lockfile-util.h"
+#include "missing_network.h"
+#include "netif-naming-scheme.h"
+#include "netlink-util.h"
+#include "nspawn-network.h"
+#include "parse-util.h"
+#include "siphash24.h"
+#include "socket-netlink.h"
+#include "socket-util.h"
+#include "stat-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "udev-util.h"
+#include "util.h"
+
+#define HOST_HASH_KEY SD_ID128_MAKE(1a,37,6f,c7,46,ec,45,0b,ad,a3,d5,31,06,60,5d,b1)
+#define CONTAINER_HASH_KEY SD_ID128_MAKE(c3,c4,f9,19,b5,57,b2,1c,e6,cf,14,27,03,9c,ee,a2)
+#define VETH_EXTRA_HOST_HASH_KEY SD_ID128_MAKE(48,c7,f6,b7,ea,9d,4c,9e,b7,28,d4,de,91,d5,bf,66)
+#define VETH_EXTRA_CONTAINER_HASH_KEY SD_ID128_MAKE(af,50,17,61,ce,f9,4d,35,84,0d,2b,20,54,be,ce,59)
+#define MACVLAN_HASH_KEY SD_ID128_MAKE(00,13,6d,bc,66,83,44,81,bb,0c,f9,51,1f,24,a6,6f)
+#define SHORTEN_IFNAME_HASH_KEY SD_ID128_MAKE(e1,90,a4,04,a8,ef,4b,51,8c,cc,c3,3a,9f,11,fc,a2)
+
+static int remove_one_link(sd_netlink *rtnl, const char *name) {
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
+ int r;
+
+ if (isempty(name))
+ return 0;
+
+ r = sd_rtnl_message_new_link(rtnl, &m, RTM_DELLINK, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate netlink message: %m");
+
+ r = sd_netlink_message_append_string(m, IFLA_IFNAME, name);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink interface name: %m");
+
+ r = sd_netlink_call(rtnl, m, 0, NULL);
+ if (r == -ENODEV) /* Already gone */
+ return 0;
+ if (r < 0)
+ return log_error_errno(r, "Failed to remove interface %s: %m", name);
+
+ return 1;
+}
+
+static int generate_mac(
+ const char *machine_name,
+ struct ether_addr *mac,
+ sd_id128_t hash_key,
+ uint64_t idx) {
+
+ uint64_t result;
+ size_t l, sz;
+ uint8_t *v, *i;
+ int r;
+
+ l = strlen(machine_name);
+ sz = sizeof(sd_id128_t) + l;
+ if (idx > 0)
+ sz += sizeof(idx);
+
+ v = newa(uint8_t, sz);
+
+ /* fetch some persistent data unique to the host */
+ r = sd_id128_get_machine((sd_id128_t*) v);
+ if (r < 0)
+ return r;
+
+ /* combine with some data unique (on this host) to this
+ * container instance */
+ i = mempcpy(v + sizeof(sd_id128_t), machine_name, l);
+ if (idx > 0) {
+ idx = htole64(idx);
+ memcpy(i, &idx, sizeof(idx));
+ }
+
+ /* Let's hash the host machine ID plus the container name. We
+ * use a fixed, but originally randomly created hash key here. */
+ result = htole64(siphash24(v, sz, hash_key.bytes));
+
+ assert_cc(ETH_ALEN <= sizeof(result));
+ memcpy(mac->ether_addr_octet, &result, ETH_ALEN);
+
+ /* see eth_random_addr in the kernel */
+ mac->ether_addr_octet[0] &= 0xfe; /* clear multicast bit */
+ mac->ether_addr_octet[0] |= 0x02; /* set local assignment bit (IEEE802) */
+
+ return 0;
+}
+
+static int set_alternative_ifname(sd_netlink *rtnl, const char *ifname, const char *altifname) {
+ int r;
+
+ assert(rtnl);
+ assert(ifname);
+
+ if (!altifname)
+ return 0;
+
+ if (strlen(altifname) >= ALTIFNAMSIZ)
+ return log_warning_errno(SYNTHETIC_ERRNO(ERANGE),
+ "Alternative interface name '%s' for '%s' is too long, ignoring",
+ altifname, ifname);
+
+ r = rtnl_set_link_alternative_names_by_ifname(&rtnl, ifname, STRV_MAKE(altifname));
+ if (r < 0)
+ return log_warning_errno(r,
+ "Failed to set alternative interface name '%s' to '%s', ignoring: %m",
+ altifname, ifname);
+
+ return 0;
+}
+
+static int add_veth(
+ sd_netlink *rtnl,
+ pid_t pid,
+ const char *ifname_host,
+ const char *altifname_host,
+ const struct ether_addr *mac_host,
+ const char *ifname_container,
+ const struct ether_addr *mac_container) {
+
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
+ int r;
+
+ assert(rtnl);
+ assert(ifname_host);
+ assert(mac_host);
+ assert(ifname_container);
+ assert(mac_container);
+
+ r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate netlink message: %m");
+
+ r = sd_netlink_message_append_string(m, IFLA_IFNAME, ifname_host);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink interface name: %m");
+
+ r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, mac_host);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink MAC address: %m");
+
+ r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open netlink container: %m");
+
+ r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "veth");
+ if (r < 0)
+ return log_error_errno(r, "Failed to open netlink container: %m");
+
+ r = sd_netlink_message_open_container(m, VETH_INFO_PEER);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open netlink container: %m");
+
+ r = sd_netlink_message_append_string(m, IFLA_IFNAME, ifname_container);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink interface name: %m");
+
+ r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, mac_container);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink MAC address: %m");
+
+ r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink namespace field: %m");
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return log_error_errno(r, "Failed to close netlink container: %m");
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return log_error_errno(r, "Failed to close netlink container: %m");
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return log_error_errno(r, "Failed to close netlink container: %m");
+
+ r = sd_netlink_call(rtnl, m, 0, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add new veth interfaces (%s:%s): %m", ifname_host, ifname_container);
+
+ (void) set_alternative_ifname(rtnl, ifname_host, altifname_host);
+
+ return 0;
+}
+
+static int shorten_ifname(char *ifname) {
+ char new_ifname[IFNAMSIZ];
+
+ assert(ifname);
+
+ if (strlen(ifname) < IFNAMSIZ) /* Name is short enough */
+ return 0;
+
+ if (naming_scheme_has(NAMING_NSPAWN_LONG_HASH)) {
+ uint64_t h;
+
+ /* Calculate 64bit hash value */
+ h = siphash24(ifname, strlen(ifname), SHORTEN_IFNAME_HASH_KEY.bytes);
+
+ /* Set the final four bytes (i.e. 32bit) to the lower 24bit of the hash, encoded in url-safe base64 */
+ memcpy(new_ifname, ifname, IFNAMSIZ - 5);
+ new_ifname[IFNAMSIZ - 5] = urlsafe_base64char(h >> 18);
+ new_ifname[IFNAMSIZ - 4] = urlsafe_base64char(h >> 12);
+ new_ifname[IFNAMSIZ - 3] = urlsafe_base64char(h >> 6);
+ new_ifname[IFNAMSIZ - 2] = urlsafe_base64char(h);
+ } else
+ /* On old nspawn versions we just truncated the name, provide compatibility */
+ memcpy(new_ifname, ifname, IFNAMSIZ-1);
+
+ new_ifname[IFNAMSIZ - 1] = 0;
+
+ /* Log the incident to make it more discoverable */
+ log_warning("Network interface name '%s' has been changed to '%s' to fit length constraints.", ifname, new_ifname);
+
+ strcpy(ifname, new_ifname);
+ return 1;
+}
+
+int setup_veth(const char *machine_name,
+ pid_t pid,
+ char iface_name[IFNAMSIZ],
+ bool bridge) {
+
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ struct ether_addr mac_host, mac_container;
+ unsigned u;
+ char *n, *a = NULL;
+ int r;
+
+ assert(machine_name);
+ assert(pid > 0);
+ assert(iface_name);
+
+ /* Use two different interface name prefixes depending whether
+ * we are in bridge mode or not. */
+ n = strjoina(bridge ? "vb-" : "ve-", machine_name);
+ r = shorten_ifname(n);
+ if (r > 0)
+ a = strjoina(bridge ? "vb-" : "ve-", machine_name);
+
+ r = generate_mac(machine_name, &mac_container, CONTAINER_HASH_KEY, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to generate predictable MAC address for container side: %m");
+
+ r = generate_mac(machine_name, &mac_host, HOST_HASH_KEY, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to generate predictable MAC address for host side: %m");
+
+ r = sd_netlink_open(&rtnl);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to netlink: %m");
+
+ r = add_veth(rtnl, pid, n, a, &mac_host, "host0", &mac_container);
+ if (r < 0)
+ return r;
+
+ u = if_nametoindex(n); /* We don't need to use rtnl_resolve_ifname() here because the
+ * name we assigned is always the main name. */
+ if (u == 0)
+ return log_error_errno(errno, "Failed to resolve interface %s: %m", n);
+
+ strcpy(iface_name, n);
+ return (int) u;
+}
+
+int setup_veth_extra(
+ const char *machine_name,
+ pid_t pid,
+ char **pairs) {
+
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ uint64_t idx = 0;
+ int r;
+
+ assert(machine_name);
+ assert(pid > 0);
+
+ if (strv_isempty(pairs))
+ return 0;
+
+ r = sd_netlink_open(&rtnl);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to netlink: %m");
+
+ STRV_FOREACH_PAIR(a, b, pairs) {
+ struct ether_addr mac_host, mac_container;
+
+ r = generate_mac(machine_name, &mac_container, VETH_EXTRA_CONTAINER_HASH_KEY, idx);
+ if (r < 0)
+ return log_error_errno(r, "Failed to generate predictable MAC address for container side of extra veth link: %m");
+
+ r = generate_mac(machine_name, &mac_host, VETH_EXTRA_HOST_HASH_KEY, idx);
+ if (r < 0)
+ return log_error_errno(r, "Failed to generate predictable MAC address for host side of extra veth link: %m");
+
+ r = add_veth(rtnl, pid, *a, NULL, &mac_host, *b, &mac_container);
+ if (r < 0)
+ return r;
+
+ idx++;
+ }
+
+ return 0;
+}
+
+static int join_bridge(sd_netlink *rtnl, const char *veth_name, const char *bridge_name) {
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
+ int r, bridge_ifi;
+
+ assert(rtnl);
+ assert(veth_name);
+ assert(bridge_name);
+
+ bridge_ifi = rtnl_resolve_interface(&rtnl, bridge_name);
+ if (bridge_ifi < 0)
+ return bridge_ifi;
+
+ r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, 0);
+ if (r < 0)
+ return r;
+
+ r = sd_rtnl_message_link_set_flags(m, IFF_UP, IFF_UP);
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_message_append_string(m, IFLA_IFNAME, veth_name);
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_message_append_u32(m, IFLA_MASTER, bridge_ifi);
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_call(rtnl, m, 0, NULL);
+ if (r < 0)
+ return r;
+
+ return bridge_ifi;
+}
+
+static int create_bridge(sd_netlink *rtnl, const char *bridge_name) {
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
+ int r;
+
+ r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_message_append_string(m, IFLA_IFNAME, bridge_name);
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "bridge");
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return r;
+
+ r = sd_netlink_call(rtnl, m, 0, NULL);
+ if (r < 0)
+ return r;
+
+ return 0;
+}
+
+int setup_bridge(const char *veth_name, const char *bridge_name, bool create) {
+ _cleanup_(release_lock_file) LockFile bridge_lock = LOCK_FILE_INIT;
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ int r, bridge_ifi;
+ unsigned n = 0;
+
+ assert(veth_name);
+ assert(bridge_name);
+
+ r = sd_netlink_open(&rtnl);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to netlink: %m");
+
+ if (create) {
+ /* We take a system-wide lock here, so that we can safely check whether there's still a member in the
+ * bridge before removing it, without risking interference from other nspawn instances. */
+
+ r = make_lock_file("/run/systemd/nspawn-network-zone", LOCK_EX, &bridge_lock);
+ if (r < 0)
+ return log_error_errno(r, "Failed to take network zone lock: %m");
+ }
+
+ for (;;) {
+ bridge_ifi = join_bridge(rtnl, veth_name, bridge_name);
+ if (bridge_ifi >= 0)
+ return bridge_ifi;
+ if (bridge_ifi != -ENODEV || !create || n > 10)
+ return log_error_errno(bridge_ifi, "Failed to add interface %s to bridge %s: %m", veth_name, bridge_name);
+
+ /* Count attempts, so that we don't enter an endless loop here. */
+ n++;
+
+ /* The bridge doesn't exist yet. Let's create it */
+ r = create_bridge(rtnl, bridge_name);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create bridge interface %s: %m", bridge_name);
+
+ /* Try again, now that the bridge exists */
+ }
+}
+
+int remove_bridge(const char *bridge_name) {
+ _cleanup_(release_lock_file) LockFile bridge_lock = LOCK_FILE_INIT;
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ const char *path;
+ int r;
+
+ /* Removes the specified bridge, but only if it is currently empty */
+
+ if (isempty(bridge_name))
+ return 0;
+
+ r = make_lock_file("/run/systemd/nspawn-network-zone", LOCK_EX, &bridge_lock);
+ if (r < 0)
+ return log_error_errno(r, "Failed to take network zone lock: %m");
+
+ path = strjoina("/sys/class/net/", bridge_name, "/brif");
+
+ r = dir_is_empty(path, /* ignore_hidden_or_backup= */ false);
+ if (r == -ENOENT) /* Already gone? */
+ return 0;
+ if (r < 0)
+ return log_error_errno(r, "Can't detect if bridge %s is empty: %m", bridge_name);
+ if (r == 0) /* Still populated, leave it around */
+ return 0;
+
+ r = sd_netlink_open(&rtnl);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to netlink: %m");
+
+ return remove_one_link(rtnl, bridge_name);
+}
+
+int test_network_interface_initialized(const char *name) {
+ _cleanup_(sd_device_unrefp) sd_device *d = NULL;
+ int r;
+
+ if (!udev_available())
+ return 0;
+
+ /* udev should be around. */
+
+ r = sd_device_new_from_ifname(&d, name);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get device %s: %m", name);
+
+ r = sd_device_get_is_initialized(d);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine whether interface %s is initialized: %m", name);
+ if (r == 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EBUSY), "Network interface %s is not initialized yet.", name);
+
+ r = device_is_renaming(d);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine the interface %s is being renamed: %m", name);
+ if (r > 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EBUSY), "Interface %s is being renamed.", name);
+
+ return 0;
+}
+
+int move_network_interfaces(int netns_fd, char **ifaces) {
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ int r;
+
+ if (strv_isempty(ifaces))
+ return 0;
+
+ r = sd_netlink_open(&rtnl);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to netlink: %m");
+
+ STRV_FOREACH(i, ifaces) {
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
+ int ifi;
+
+ ifi = rtnl_resolve_interface_or_warn(&rtnl, *i);
+ if (ifi < 0)
+ return ifi;
+
+ r = sd_rtnl_message_new_link(rtnl, &m, RTM_SETLINK, ifi);
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate netlink message: %m");
+
+ r = sd_netlink_message_append_u32(m, IFLA_NET_NS_FD, netns_fd);
+ if (r < 0)
+ return log_error_errno(r, "Failed to append namespace fd to netlink message: %m");
+
+ r = sd_netlink_call(rtnl, m, 0, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to move interface %s to namespace: %m", *i);
+ }
+
+ return 0;
+}
+
+int setup_macvlan(const char *machine_name, pid_t pid, char **ifaces) {
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ unsigned idx = 0;
+ int r;
+
+ if (strv_isempty(ifaces))
+ return 0;
+
+ r = sd_netlink_open(&rtnl);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to netlink: %m");
+
+ STRV_FOREACH(i, ifaces) {
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
+ _cleanup_free_ char *n = NULL, *a = NULL;
+ struct ether_addr mac;
+ int ifi;
+
+ ifi = rtnl_resolve_interface_or_warn(&rtnl, *i);
+ if (ifi < 0)
+ return ifi;
+
+ r = generate_mac(machine_name, &mac, MACVLAN_HASH_KEY, idx++);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create MACVLAN MAC address: %m");
+
+ r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate netlink message: %m");
+
+ r = sd_netlink_message_append_u32(m, IFLA_LINK, ifi);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink interface index: %m");
+
+ n = strjoin("mv-", *i);
+ if (!n)
+ return log_oom();
+
+ r = shorten_ifname(n);
+ if (r > 0) {
+ a = strjoin("mv-", *i);
+ if (!a)
+ return log_oom();
+ }
+
+ r = sd_netlink_message_append_string(m, IFLA_IFNAME, n);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink interface name: %m");
+
+ r = sd_netlink_message_append_ether_addr(m, IFLA_ADDRESS, &mac);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink MAC address: %m");
+
+ r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink namespace field: %m");
+
+ r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open netlink container: %m");
+
+ r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "macvlan");
+ if (r < 0)
+ return log_error_errno(r, "Failed to open netlink container: %m");
+
+ r = sd_netlink_message_append_u32(m, IFLA_MACVLAN_MODE, MACVLAN_MODE_BRIDGE);
+ if (r < 0)
+ return log_error_errno(r, "Failed to append macvlan mode: %m");
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return log_error_errno(r, "Failed to close netlink container: %m");
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return log_error_errno(r, "Failed to close netlink container: %m");
+
+ r = sd_netlink_call(rtnl, m, 0, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add new macvlan interfaces: %m");
+
+ (void) set_alternative_ifname(rtnl, n, a);
+ }
+
+ return 0;
+}
+
+int setup_ipvlan(const char *machine_name, pid_t pid, char **ifaces) {
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ int r;
+
+ if (strv_isempty(ifaces))
+ return 0;
+
+ r = sd_netlink_open(&rtnl);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to netlink: %m");
+
+ STRV_FOREACH(i, ifaces) {
+ _cleanup_(sd_netlink_message_unrefp) sd_netlink_message *m = NULL;
+ _cleanup_free_ char *n = NULL, *a = NULL;
+ int ifi;
+
+ ifi = rtnl_resolve_interface_or_warn(&rtnl, *i);
+ if (ifi < 0)
+ return ifi;
+
+ r = sd_rtnl_message_new_link(rtnl, &m, RTM_NEWLINK, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate netlink message: %m");
+
+ r = sd_netlink_message_append_u32(m, IFLA_LINK, ifi);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink interface index: %m");
+
+ n = strjoin("iv-", *i);
+ if (!n)
+ return log_oom();
+
+ r = shorten_ifname(n);
+ if (r > 0) {
+ a = strjoin("iv-", *i);
+ if (!a)
+ return log_oom();
+ }
+
+ r = sd_netlink_message_append_string(m, IFLA_IFNAME, n);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink interface name: %m");
+
+ r = sd_netlink_message_append_u32(m, IFLA_NET_NS_PID, pid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add netlink namespace field: %m");
+
+ r = sd_netlink_message_open_container(m, IFLA_LINKINFO);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open netlink container: %m");
+
+ r = sd_netlink_message_open_container_union(m, IFLA_INFO_DATA, "ipvlan");
+ if (r < 0)
+ return log_error_errno(r, "Failed to open netlink container: %m");
+
+ r = sd_netlink_message_append_u16(m, IFLA_IPVLAN_MODE, IPVLAN_MODE_L2);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add ipvlan mode: %m");
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return log_error_errno(r, "Failed to close netlink container: %m");
+
+ r = sd_netlink_message_close_container(m);
+ if (r < 0)
+ return log_error_errno(r, "Failed to close netlink container: %m");
+
+ r = sd_netlink_call(rtnl, m, 0, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add new ipvlan interfaces: %m");
+
+ (void) set_alternative_ifname(rtnl, n, a);
+ }
+
+ return 0;
+}
+
+int veth_extra_parse(char ***l, const char *p) {
+ _cleanup_free_ char *a = NULL, *b = NULL;
+ int r;
+
+ r = extract_first_word(&p, &a, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
+ if (r < 0)
+ return r;
+ if (r == 0 || !ifname_valid(a))
+ return -EINVAL;
+
+ r = extract_first_word(&p, &b, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
+ if (r < 0)
+ return r;
+ if (r == 0 || !ifname_valid(b)) {
+ free(b);
+ b = strdup(a);
+ if (!b)
+ return -ENOMEM;
+ }
+
+ if (p)
+ return -EINVAL;
+
+ r = strv_push_pair(l, a, b);
+ if (r < 0)
+ return -ENOMEM;
+
+ a = b = NULL;
+ return 0;
+}
+
+int remove_veth_links(const char *primary, char **pairs) {
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ int r;
+
+ /* In some cases the kernel might pin the veth links between host and container even after the namespace
+ * died. Hence, let's better remove them explicitly too. */
+
+ if (isempty(primary) && strv_isempty(pairs))
+ return 0;
+
+ r = sd_netlink_open(&rtnl);
+ if (r < 0)
+ return log_error_errno(r, "Failed to connect to netlink: %m");
+
+ remove_one_link(rtnl, primary);
+
+ STRV_FOREACH_PAIR(a, b, pairs)
+ remove_one_link(rtnl, *a);
+
+ return 0;
+}
diff --git a/src/nspawn/nspawn-network.h b/src/nspawn/nspawn-network.h
new file mode 100644
index 0000000..5c2d983
--- /dev/null
+++ b/src/nspawn/nspawn-network.h
@@ -0,0 +1,23 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <net/if.h>
+#include <stdbool.h>
+#include <sys/types.h>
+
+int test_network_interface_initialized(const char *name);
+
+int setup_veth(const char *machine_name, pid_t pid, char iface_name[IFNAMSIZ], bool bridge);
+int setup_veth_extra(const char *machine_name, pid_t pid, char **pairs);
+
+int setup_bridge(const char *veth_name, const char *bridge_name, bool create);
+int remove_bridge(const char *bridge_name);
+
+int setup_macvlan(const char *machine_name, pid_t pid, char **ifaces);
+int setup_ipvlan(const char *machine_name, pid_t pid, char **ifaces);
+
+int move_network_interfaces(int netns_fd, char **ifaces);
+
+int veth_extra_parse(char ***l, const char *p);
+
+int remove_veth_links(const char *primary, char **pairs);
diff --git a/src/nspawn/nspawn-oci.c b/src/nspawn/nspawn-oci.c
new file mode 100644
index 0000000..c8ed3f1
--- /dev/null
+++ b/src/nspawn/nspawn-oci.c
@@ -0,0 +1,2207 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <linux/oom.h>
+#if HAVE_SECCOMP
+#include <seccomp.h>
+#endif
+
+#include "bus-util.h"
+#include "cap-list.h"
+#include "cpu-set-util.h"
+#include "device-util.h"
+#include "devnum-util.h"
+#include "env-util.h"
+#include "format-util.h"
+#include "fs-util.h"
+#include "hostname-util.h"
+#include "json.h"
+#include "missing_sched.h"
+#include "nspawn-oci.h"
+#include "path-util.h"
+#include "rlimit-util.h"
+#if HAVE_SECCOMP
+#include "seccomp-util.h"
+#endif
+#include "stdio-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "user-util.h"
+
+/* TODO:
+ * OCI runtime tool implementation
+ * hooks
+ *
+ * Spec issues:
+ *
+ * How is RLIM_INFINITY supposed to be encoded?
+ * configured effective caps is bullshit, as execv() corrupts it anyway
+ * pipes bind mounted is *very* different from pipes newly created, comments regarding bind mount or not are bogus
+ * annotation values structured? or string?
+ * configurable file system namespace path, but then also root path? wtf?
+ * apply sysctl inside of the container? or outside?
+ * how is unlimited pids tasks limit to be encoded?
+ * what are the defaults for caps if not specified?
+ * what are the default uid/gid mappings if one is missing but the other set, or when user ns is on but no namespace configured
+ * the source field of "mounts" is really weird, as it cannot realistically be relative to the bundle, since we never know if that's what the fs wants
+ * spec contradicts itself on the mount "type" field, as the example uses "bind" as type, but it's not listed in /proc/filesystem, and is something made up by /bin/mount
+ * if type of mount is left out, what shall be assumed? "bind"?
+ * readonly mounts is entirely redundant?
+ * should escaping be applied when joining mount options with ","?
+ * devices cgroup support is bogus, "allow" and "deny" on the kernel level is about adding/removing entries, not about access
+ * spec needs to say that "rwm" devices cgroup combination can't be the empty string
+ * cgrouspv1 crap: kernel, kernelTCP, swappiness, disableOOMKiller, swap, devices, leafWeight
+ * general: it shouldn't leak lower level abstractions this obviously
+ * unmanagable cgroups stuff: realtimeRuntime/realtimePeriod
+ * needs to say what happense when some option is not specified, i.e. which defaults apply
+ * no architecture? no personality?
+ * seccomp example and logic is simply broken: there's no constant "SCMP_ACT_ERRNO".
+ * spec should say what to do with unknown props
+ * /bin/mount regarding NFS and FUSE required?
+ * what does terminal=false mean?
+ * sysctl inside or outside? allow-listing?
+ * swapiness typo -> swappiness
+ *
+ * Unsupported:
+ *
+ * apparmorProfile
+ * selinuxLabel + mountLabel
+ * hugepageLimits
+ * network
+ * rdma
+ * intelRdt
+ * swappiness, disableOOMKiller, kernel, kernelTCP, leafWeight (because it's dead, cgroupsv2 can't do it and hence systemd neither)
+ *
+ * Non-slice cgroup paths
+ * Propagation that is not slave + shared
+ * more than one uid/gid mapping, mappings with a container base != 0, or non-matching uid/gid mappings
+ * device cgroups access = false items that are not catchall
+ * device cgroups matches where minor is specified, but major isn't. similar where major is specified but char/block is not. also, any match that only has a type set that has less than "rwm" set. also, any entry that has none of rwm set.
+ *
+ */
+
+static int oci_unexpected(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Unexpected OCI element '%s' of type '%s'.", name, json_variant_type_to_string(json_variant_type(v)));
+}
+
+static int oci_unsupported(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Unsupported OCI element '%s' of type '%s'.", name, json_variant_type_to_string(json_variant_type(v)));
+}
+
+static int oci_terminal(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = userdata;
+
+ /* If not specified, or set to true, we'll default to either an interactive or a read-only
+ * console. If specified as false, we'll forcibly move to "pipe" mode though. */
+ s->console_mode = json_variant_boolean(v) ? _CONSOLE_MODE_INVALID : CONSOLE_PIPE;
+ return 0;
+}
+
+static int oci_console_dimension(const char *name, JsonVariant *variant, JsonDispatchFlags flags, void *userdata) {
+ unsigned *u = ASSERT_PTR(userdata);
+ uint64_t k;
+
+ k = json_variant_unsigned(variant);
+ if (k == 0)
+ return json_log(variant, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Console size field '%s' is too small.", strna(name));
+ if (k > USHRT_MAX) /* TIOCSWINSZ's struct winsize uses "unsigned short" for width and height */
+ return json_log(variant, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Console size field '%s' is too large.", strna(name));
+
+ *u = (unsigned) k;
+ return 0;
+}
+
+static int oci_console_size(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "height", JSON_VARIANT_UNSIGNED, oci_console_dimension, offsetof(Settings, console_height), JSON_MANDATORY },
+ { "width", JSON_VARIANT_UNSIGNED, oci_console_dimension, offsetof(Settings, console_width), JSON_MANDATORY },
+ {}
+ };
+
+ return json_dispatch(v, table, oci_unexpected, flags, userdata);
+}
+
+static int oci_absolute_path(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ char **p = ASSERT_PTR(userdata);
+ const char *n;
+
+ n = json_variant_string(v);
+
+ if (!path_is_absolute(n))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Path in JSON field '%s' is not absolute: %s", strna(name), n);
+
+ return free_and_strdup_warn(p, n);
+}
+
+static int oci_env(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ char ***l = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ const char *n;
+
+ if (!json_variant_is_string(e))
+ return json_log(e, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Environment array contains non-string.");
+
+ assert_se(n = json_variant_string(e));
+
+ if (!env_assignment_is_valid(n))
+ return json_log(e, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Environment assignment not valid: %s", n);
+
+ r = strv_extend(l, n);
+ if (r < 0)
+ return log_oom();
+ }
+
+ return 0;
+}
+
+static int oci_args(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ _cleanup_strv_free_ char **l = NULL;
+ char ***value = ASSERT_PTR(userdata);
+ int r;
+
+ r = json_variant_strv(v, &l);
+ if (r < 0)
+ return json_log(v, flags, r, "Cannot parse arguments as list of strings: %m");
+
+ if (strv_isempty(l))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Argument list empty, refusing.");
+
+ if (isempty(l[0]))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Executable name is empty, refusing.");
+
+ return strv_free_and_replace(*value, l);
+}
+
+static int oci_rlimit_type(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ const char *z;
+ int t, *type = userdata;
+
+ assert_se(type);
+
+ z = startswith(json_variant_string(v), "RLIMIT_");
+ if (!z)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "rlimit entry's name does not begin with 'RLIMIT_', refusing: %s",
+ json_variant_string(v));
+
+ t = rlimit_from_string(z);
+ if (t < 0)
+ return json_log(v, flags, t,
+ "rlimit name unknown: %s", json_variant_string(v));
+
+ *type = t;
+ return 0;
+}
+
+static int oci_rlimit_value(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ rlim_t z, *value = ASSERT_PTR(userdata);
+
+ if (json_variant_is_negative(v))
+ z = RLIM_INFINITY;
+ else {
+ if (!json_variant_is_unsigned(v))
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "rlimits limit not unsigned, refusing.");
+
+ z = (rlim_t) json_variant_unsigned(v);
+
+ if ((uint64_t) z != json_variant_unsigned(v))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "rlimits limit out of range, refusing.");
+ }
+
+ *value = z;
+ return 0;
+}
+
+static int oci_rlimits(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+
+ struct rlimit_data {
+ int type;
+ rlim_t soft;
+ rlim_t hard;
+ } data = {
+ .type = -1,
+ .soft = RLIM_INFINITY,
+ .hard = RLIM_INFINITY,
+ };
+
+ static const JsonDispatch table[] = {
+ { "soft", JSON_VARIANT_NUMBER, oci_rlimit_value, offsetof(struct rlimit_data, soft), JSON_MANDATORY },
+ { "hard", JSON_VARIANT_NUMBER, oci_rlimit_value, offsetof(struct rlimit_data, hard), JSON_MANDATORY },
+ { "type", JSON_VARIANT_STRING, oci_rlimit_type, offsetof(struct rlimit_data, type), JSON_MANDATORY },
+ {}
+ };
+
+ r = json_dispatch(e, table, oci_unexpected, flags, &data);
+ if (r < 0)
+ return r;
+
+ assert(data.type >= 0);
+ assert(data.type < _RLIMIT_MAX);
+
+ if (s->rlimit[data.type])
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "rlimits array contains duplicate entry, refusing.");
+
+ s->rlimit[data.type] = new(struct rlimit, 1);
+ if (!s->rlimit[data.type])
+ return log_oom();
+
+ *s->rlimit[data.type] = (struct rlimit) {
+ .rlim_cur = data.soft,
+ .rlim_max = data.hard,
+ };
+
+ }
+ return 0;
+}
+
+static int oci_capability_array(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ uint64_t *mask = userdata, m = 0;
+ JsonVariant *e;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ const char *n;
+ int cap;
+
+ if (!json_variant_is_string(e))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Entry in capabilities array is not a string.");
+
+ assert_se(n = json_variant_string(e));
+
+ cap = capability_from_name(n);
+ if (cap < 0)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Unknown capability: %s", n);
+
+ m |= UINT64_C(1) << cap;
+ }
+
+ if (*mask == UINT64_MAX)
+ *mask = m;
+ else
+ *mask |= m;
+
+ return 0;
+}
+
+static int oci_capabilities(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "effective", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, effective) },
+ { "bounding", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, bounding) },
+ { "inheritable", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, inheritable) },
+ { "permitted", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, permitted) },
+ { "ambient", JSON_VARIANT_ARRAY, oci_capability_array, offsetof(CapabilityQuintet, ambient) },
+ {}
+ };
+
+ Settings *s = ASSERT_PTR(userdata);
+ int r;
+
+ r = json_dispatch(v, table, oci_unexpected, flags, &s->full_capabilities);
+ if (r < 0)
+ return r;
+
+ if (s->full_capabilities.bounding != UINT64_MAX) {
+ s->capability = s->full_capabilities.bounding;
+ s->drop_capability = ~s->full_capabilities.bounding;
+ }
+
+ return 0;
+}
+
+static int oci_oom_score_adj(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ int64_t k;
+
+ k = json_variant_integer(v);
+ if (k < OOM_SCORE_ADJ_MIN || k > OOM_SCORE_ADJ_MAX)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "oomScoreAdj value out of range: %" PRIi64, k);
+
+ s->oom_score_adjust = (int) k;
+ s->oom_score_adjust_set = true;
+
+ return 0;
+}
+
+static int oci_uid_gid(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ uid_t *uid = userdata, u;
+ uint64_t k;
+
+ assert(uid);
+ assert_cc(sizeof(uid_t) == sizeof(gid_t));
+
+ k = json_variant_unsigned(v);
+ u = (uid_t) k;
+ if ((uint64_t) u != k)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "UID/GID out of range: %" PRIu64, k);
+
+ if (!uid_is_valid(u))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "UID/GID is not valid: " UID_FMT, u);
+
+ *uid = u;
+ return 0;
+}
+
+static int oci_supplementary_gids(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ gid_t gid, *a;
+
+ if (!json_variant_is_unsigned(e))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Supplementary GID entry is not a UID.");
+
+ r = oci_uid_gid(name, e, flags, &gid);
+ if (r < 0)
+ return r;
+
+ a = reallocarray(s->supplementary_gids, s->n_supplementary_gids + 1, sizeof(gid_t));
+ if (!a)
+ return log_oom();
+
+ s->supplementary_gids = a;
+ s->supplementary_gids[s->n_supplementary_gids++] = gid;
+ }
+
+ return 0;
+}
+
+static int oci_user(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ static const JsonDispatch table[] = {
+ { "uid", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(Settings, uid), JSON_MANDATORY },
+ { "gid", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(Settings, gid), JSON_MANDATORY },
+ { "additionalGids", JSON_VARIANT_ARRAY, oci_supplementary_gids, 0, 0 },
+ {}
+ };
+
+ return json_dispatch(v, table, oci_unexpected, flags, userdata);
+}
+
+static int oci_process(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "terminal", JSON_VARIANT_BOOLEAN, oci_terminal, 0, 0 },
+ { "consoleSize", JSON_VARIANT_OBJECT, oci_console_size, 0, 0 },
+ { "cwd", JSON_VARIANT_STRING, oci_absolute_path, offsetof(Settings, working_directory), 0 },
+ { "env", JSON_VARIANT_ARRAY, oci_env, offsetof(Settings, environment), 0 },
+ { "args", JSON_VARIANT_ARRAY, oci_args, offsetof(Settings, parameters), 0 },
+ { "rlimits", JSON_VARIANT_ARRAY, oci_rlimits, 0, 0 },
+ { "apparmorProfile", JSON_VARIANT_STRING, oci_unsupported, 0, JSON_PERMISSIVE },
+ { "capabilities", JSON_VARIANT_OBJECT, oci_capabilities, 0, 0 },
+ { "noNewPrivileges", JSON_VARIANT_BOOLEAN, json_dispatch_boolean, offsetof(Settings, no_new_privileges), 0 },
+ { "oomScoreAdj", JSON_VARIANT_INTEGER, oci_oom_score_adj, 0, 0 },
+ { "selinuxLabel", JSON_VARIANT_STRING, oci_unsupported, 0, JSON_PERMISSIVE },
+ { "user", JSON_VARIANT_OBJECT, oci_user, 0, 0 },
+ {}
+ };
+
+ return json_dispatch(v, table, oci_unexpected, flags, userdata);
+}
+
+static int oci_root(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = userdata;
+ int r;
+
+ static const JsonDispatch table[] = {
+ { "path", JSON_VARIANT_STRING, json_dispatch_string, offsetof(Settings, root) },
+ { "readonly", JSON_VARIANT_BOOLEAN, json_dispatch_boolean, offsetof(Settings, read_only) },
+ {}
+ };
+
+ r = json_dispatch(v, table, oci_unexpected, flags, s);
+ if (r < 0)
+ return r;
+
+ if (s->root && !path_is_absolute(s->root)) {
+ char *joined;
+
+ joined = path_join(s->bundle, s->root);
+ if (!joined)
+ return log_oom();
+
+ free_and_replace(s->root, joined);
+ }
+
+ return 0;
+}
+
+static int oci_hostname(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ const char *n;
+
+ assert_se(n = json_variant_string(v));
+
+ if (!hostname_is_valid(n, 0))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Hostname string is not a valid hostname: %s", n);
+
+ return free_and_strdup_warn(&s->hostname, n);
+}
+
+static bool oci_exclude_mount(const char *path) {
+
+ /* Returns "true" for all mounts we insist to mount on our own, and hence ignore the OCI data. */
+
+ if (PATH_IN_SET(path,
+ "/dev",
+ "/dev/mqueue",
+ "/dev/pts",
+ "/dev/shm",
+ "/proc",
+ "/proc/acpi",
+ "/proc/apm",
+ "/proc/asound",
+ "/proc/bus",
+ "/proc/fs",
+ "/proc/irq",
+ "/proc/kallsyms",
+ "/proc/kcore",
+ "/proc/keys",
+ "/proc/scsi",
+ "/proc/sys",
+ "/proc/sys/net",
+ "/proc/sysrq-trigger",
+ "/proc/timer_list",
+ "/run",
+ "/sys",
+ "/sys",
+ "/sys/fs/selinux",
+ "/tmp"))
+ return true;
+
+ /* Similar, skip the whole /sys/fs/cgroups subtree */
+ if (path_startswith(path, "/sys/fs/cgroup"))
+ return true;
+
+ return false;
+}
+
+typedef struct oci_mount_data {
+ char *destination;
+ char *source;
+ char *type;
+ char **options;
+} oci_mount_data;
+
+static void cleanup_oci_mount_data(oci_mount_data *data) {
+ free(data->destination);
+ free(data->source);
+ strv_free(data->options);
+ free(data->type);
+}
+
+static int oci_mounts(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ static const JsonDispatch table[] = {
+ { "destination", JSON_VARIANT_STRING, oci_absolute_path, offsetof(oci_mount_data, destination), JSON_MANDATORY },
+ { "source", JSON_VARIANT_STRING, json_dispatch_string, offsetof(oci_mount_data, source), 0 },
+ { "options", JSON_VARIANT_ARRAY, json_dispatch_strv, offsetof(oci_mount_data, options), 0, },
+ { "type", JSON_VARIANT_STRING, json_dispatch_string, offsetof(oci_mount_data, type), 0 },
+ {}
+ };
+
+ _cleanup_free_ char *joined_options = NULL;
+ CustomMount *m;
+ _cleanup_(cleanup_oci_mount_data) oci_mount_data data = {};
+
+ r = json_dispatch(e, table, oci_unexpected, flags, &data);
+ if (r < 0)
+ return r;
+
+ if (!path_is_absolute(data.destination))
+ return json_log(e, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Mount destination not an absolute path: %s", data.destination);
+
+ if (oci_exclude_mount(data.destination))
+ continue;
+
+ if (data.options) {
+ joined_options = strv_join(data.options, ",");
+ if (!joined_options)
+ return log_oom();
+ }
+
+ if (!data.type || streq(data.type, "bind")) {
+ if (data.source && !path_is_absolute(data.source)) {
+ char *joined;
+
+ joined = path_join(s->bundle, data.source);
+ if (!joined)
+ return log_oom();
+
+ free_and_replace(data.source, joined);
+ }
+
+ data.type = mfree(data.type);
+
+ m = custom_mount_add(&s->custom_mounts, &s->n_custom_mounts, CUSTOM_MOUNT_BIND);
+ } else
+ m = custom_mount_add(&s->custom_mounts, &s->n_custom_mounts, CUSTOM_MOUNT_ARBITRARY);
+ if (!m)
+ return log_oom();
+
+ m->destination = TAKE_PTR(data.destination);
+ m->source = TAKE_PTR(data.source);
+ m->options = TAKE_PTR(joined_options);
+ m->type_argument = TAKE_PTR(data.type);
+ }
+
+ return 0;
+}
+
+static int oci_namespace_type(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ unsigned long *nsflags = ASSERT_PTR(userdata);
+ const char *n;
+
+ assert_se(n = json_variant_string(v));
+
+ /* We don't use namespace_flags_from_string() here, as the OCI spec uses slightly different names than the
+ * kernel here. */
+ if (streq(n, "pid"))
+ *nsflags = CLONE_NEWPID;
+ else if (streq(n, "network"))
+ *nsflags = CLONE_NEWNET;
+ else if (streq(n, "mount"))
+ *nsflags = CLONE_NEWNS;
+ else if (streq(n, "ipc"))
+ *nsflags = CLONE_NEWIPC;
+ else if (streq(n, "uts"))
+ *nsflags = CLONE_NEWUTS;
+ else if (streq(n, "user"))
+ *nsflags = CLONE_NEWUSER;
+ else if (streq(n, "cgroup"))
+ *nsflags = CLONE_NEWCGROUP;
+ else
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Unknown namespace type, refusing: %s", n);
+
+ return 0;
+}
+
+static int oci_namespaces(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = userdata;
+ unsigned long n = 0;
+ JsonVariant *e;
+ int r;
+
+ assert_se(s);
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+
+ struct namespace_data {
+ unsigned long type;
+ char *path;
+ } data = {};
+
+ static const JsonDispatch table[] = {
+ { "type", JSON_VARIANT_STRING, oci_namespace_type, offsetof(struct namespace_data, type), JSON_MANDATORY },
+ { "path", JSON_VARIANT_STRING, oci_absolute_path, offsetof(struct namespace_data, path), 0 },
+ {}
+ };
+
+ r = json_dispatch(e, table, oci_unexpected, flags, &data);
+ if (r < 0) {
+ free(data.path);
+ return r;
+ }
+
+ if (data.path) {
+ if (data.type != CLONE_NEWNET) {
+ free(data.path);
+ return json_log(e, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Specifying namespace path for non-network namespace is not supported.");
+ }
+
+ if (s->network_namespace_path) {
+ free(data.path);
+ return json_log(e, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Network namespace path specified more than once, refusing.");
+ }
+
+ free(s->network_namespace_path);
+ s->network_namespace_path = data.path;
+ }
+
+ if (FLAGS_SET(n, data.type))
+ return json_log(e, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Duplicate namespace specification, refusing.");
+
+ n |= data.type;
+ }
+
+ if (!FLAGS_SET(n, CLONE_NEWNS))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Containers without a mount namespace aren't supported.");
+
+ s->private_network = FLAGS_SET(n, CLONE_NEWNET);
+ s->userns_mode = FLAGS_SET(n, CLONE_NEWUSER) ? USER_NAMESPACE_FIXED : USER_NAMESPACE_NO;
+ s->use_cgns = FLAGS_SET(n, CLONE_NEWCGROUP);
+
+ s->clone_ns_flags = n & (CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
+
+ return 0;
+}
+
+static int oci_uid_gid_range(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ uid_t *uid = userdata, u;
+ uint64_t k;
+
+ assert(uid);
+ assert_cc(sizeof(uid_t) == sizeof(gid_t));
+
+ /* This is very much like oci_uid_gid(), except the checks are a bit different, as this is a UID range rather
+ * than a specific UID, and hence UID_INVALID has no special significance. OTOH a range of zero makes no
+ * sense. */
+
+ k = json_variant_unsigned(v);
+ u = (uid_t) k;
+ if ((uint64_t) u != k)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "UID/GID out of range: %" PRIu64, k);
+ if (u == 0)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "UID/GID range can't be zero.");
+
+ *uid = u;
+ return 0;
+}
+
+static int oci_uid_gid_mappings(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ struct mapping_data {
+ uid_t host_id;
+ uid_t container_id;
+ uid_t range;
+ } data = {
+ .host_id = UID_INVALID,
+ .container_id = UID_INVALID,
+ .range = 0,
+ };
+
+ static const JsonDispatch table[] = {
+ { "containerID", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(struct mapping_data, container_id), JSON_MANDATORY },
+ { "hostID", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(struct mapping_data, host_id), JSON_MANDATORY },
+ { "size", JSON_VARIANT_UNSIGNED, oci_uid_gid_range, offsetof(struct mapping_data, range), JSON_MANDATORY },
+ {}
+ };
+
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ if (json_variant_elements(v) == 0)
+ return 0;
+
+ if (json_variant_elements(v) > 1)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "UID/GID mappings with more than one entry are not supported.");
+
+ assert_se(e = json_variant_by_index(v, 0));
+
+ r = json_dispatch(e, table, oci_unexpected, flags, &data);
+ if (r < 0)
+ return r;
+
+ if (data.host_id + data.range < data.host_id ||
+ data.container_id + data.range < data.container_id)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "UID/GID range goes beyond UID/GID validity range, refusing.");
+
+ if (data.container_id != 0)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "UID/GID mappings with a non-zero container base are not supported.");
+
+ if (data.range < 0x10000)
+ json_log(v, flags|JSON_WARNING, 0,
+ "UID/GID mapping with less than 65536 UID/GIDS set up, you are looking for trouble.");
+
+ if (s->uid_range != UID_INVALID &&
+ (s->uid_shift != data.host_id || s->uid_range != data.range))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Non-matching UID and GID mappings are not supported.");
+
+ s->uid_shift = data.host_id;
+ s->uid_range = data.range;
+
+ return 0;
+}
+
+static int oci_device_type(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ mode_t *mode = ASSERT_PTR(userdata);
+ const char *t;
+
+ assert_se(t = json_variant_string(v));
+
+ if (STR_IN_SET(t, "c", "u"))
+ *mode = (*mode & ~S_IFMT) | S_IFCHR;
+ else if (streq(t, "b"))
+ *mode = (*mode & ~S_IFMT) | S_IFBLK;
+ else if (streq(t, "p"))
+ *mode = (*mode & ~S_IFMT) | S_IFIFO;
+ else
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Unknown device type: %s", t);
+
+ return 0;
+}
+
+static int oci_device_major(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ unsigned *u = userdata;
+ uint64_t k;
+
+ assert_se(u);
+
+ k = json_variant_unsigned(v);
+ if (!DEVICE_MAJOR_VALID(k))
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Device major %" PRIu64 " out of range.", k);
+
+ *u = (unsigned) k;
+ return 0;
+}
+
+static int oci_device_minor(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ unsigned *u = userdata;
+ uint64_t k;
+
+ assert_se(u);
+
+ k = json_variant_unsigned(v);
+ if (!DEVICE_MINOR_VALID(k))
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Device minor %" PRIu64 " out of range.", k);
+
+ *u = (unsigned) k;
+ return 0;
+}
+
+static int oci_device_file_mode(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ mode_t *mode = userdata, m;
+ uint64_t k;
+
+ assert(mode);
+
+ k = json_variant_unsigned(v);
+ m = (mode_t) k;
+
+ if ((m & ~07777) != 0 || (uint64_t) m != k)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "fileMode out of range, refusing.");
+
+ *mode = (*mode & ~07777) | m;
+ return 0;
+}
+
+static int oci_devices(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+
+ static const JsonDispatch table[] = {
+ { "type", JSON_VARIANT_STRING, oci_device_type, offsetof(DeviceNode, mode), JSON_MANDATORY },
+ { "path", JSON_VARIANT_STRING, oci_absolute_path, offsetof(DeviceNode, path), JSON_MANDATORY },
+ { "major", JSON_VARIANT_UNSIGNED, oci_device_major, offsetof(DeviceNode, major), 0 },
+ { "minor", JSON_VARIANT_UNSIGNED, oci_device_minor, offsetof(DeviceNode, minor), 0 },
+ { "fileMode", JSON_VARIANT_UNSIGNED, oci_device_file_mode, offsetof(DeviceNode, mode), 0 },
+ { "uid", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(DeviceNode, uid), 0 },
+ { "gid", JSON_VARIANT_UNSIGNED, oci_uid_gid, offsetof(DeviceNode, gid), 0 },
+ {}
+ };
+
+ DeviceNode *node, *nodes;
+
+ nodes = reallocarray(s->extra_nodes, s->n_extra_nodes + 1, sizeof(DeviceNode));
+ if (!nodes)
+ return log_oom();
+
+ s->extra_nodes = nodes;
+
+ node = nodes + s->n_extra_nodes;
+ *node = (DeviceNode) {
+ .uid = UID_INVALID,
+ .gid = GID_INVALID,
+ .major = UINT_MAX,
+ .minor = UINT_MAX,
+ .mode = 0644,
+ };
+
+ r = json_dispatch(e, table, oci_unexpected, flags, node);
+ if (r < 0)
+ goto fail_element;
+
+ if (S_ISCHR(node->mode) || S_ISBLK(node->mode)) {
+ _cleanup_free_ char *path = NULL;
+
+ if (node->major == UINT_MAX || node->minor == UINT_MAX) {
+ r = json_log(e, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Major/minor required when device node is device node");
+ goto fail_element;
+ }
+
+ /* Suppress a couple of implicit device nodes */
+ r = devname_from_devnum(node->mode, makedev(node->major, node->minor), &path);
+ if (r < 0)
+ json_log(e, flags|JSON_DEBUG, r, "Failed to resolve device node %u:%u, ignoring: %m", node->major, node->minor);
+ else {
+ if (PATH_IN_SET(path,
+ "/dev/null",
+ "/dev/zero",
+ "/dev/full",
+ "/dev/random",
+ "/dev/urandom",
+ "/dev/tty",
+ "/dev/net/tun",
+ "/dev/ptmx",
+ "/dev/pts/ptmx",
+ "/dev/console")) {
+
+ json_log(e, flags|JSON_DEBUG, 0, "Ignoring devices item for device '%s', as it is implicitly created anyway.", path);
+ free(node->path);
+ continue;
+ }
+ }
+ }
+
+ s->n_extra_nodes++;
+ continue;
+
+ fail_element:
+ free(node->path);
+ return r;
+ }
+
+ return 0;
+}
+
+static int oci_cgroups_path(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ _cleanup_free_ char *slice = NULL, *backwards = NULL;
+ Settings *s = ASSERT_PTR(userdata);
+ const char *p;
+ int r;
+
+ assert_se(p = json_variant_string(v));
+
+ r = cg_path_get_slice(p, &slice);
+ if (r < 0)
+ return json_log(v, flags, r, "Couldn't derive slice unit name from path '%s': %m", p);
+
+ r = cg_slice_to_path(slice, &backwards);
+ if (r < 0)
+ return json_log(v, flags, r, "Couldn't convert slice unit name '%s' back to path: %m", slice);
+
+ if (!path_equal(backwards, p))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Control group path '%s' does not refer to slice unit, refusing.", p);
+
+ free_and_replace(s->slice, slice);
+ return 0;
+}
+
+static int oci_cgroup_device_type(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ mode_t *mode = userdata;
+ const char *n;
+
+ assert_se(n = json_variant_string(v));
+
+ if (streq(n, "c"))
+ *mode = S_IFCHR;
+ else if (streq(n, "b"))
+ *mode = S_IFBLK;
+ else
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Control group device type unknown: %s", n);
+
+ return 0;
+}
+
+struct device_data {
+ bool allow;
+ bool r;
+ bool w;
+ bool m;
+ mode_t type;
+ unsigned major;
+ unsigned minor;
+};
+
+static int oci_cgroup_device_access(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ struct device_data *d = userdata;
+ bool r = false, w = false, m = false;
+ const char *s;
+ size_t i;
+
+ assert_se(s = json_variant_string(v));
+
+ for (i = 0; s[i]; i++)
+ if (s[i] == 'r')
+ r = true;
+ else if (s[i] == 'w')
+ w = true;
+ else if (s[i] == 'm')
+ m = true;
+ else
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Unknown device access character '%c'.", s[i]);
+
+ d->r = r;
+ d->w = w;
+ d->m = m;
+
+ return 0;
+}
+
+static int oci_cgroup_devices(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ _cleanup_free_ struct device_data *list = NULL;
+ Settings *s = ASSERT_PTR(userdata);
+ size_t n_list = 0, i;
+ bool noop = false;
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+
+ struct device_data data = {
+ .major = UINT_MAX,
+ .minor = UINT_MAX,
+ }, *a;
+
+ static const JsonDispatch table[] = {
+ { "allow", JSON_VARIANT_BOOLEAN, json_dispatch_boolean, offsetof(struct device_data, allow), JSON_MANDATORY },
+ { "type", JSON_VARIANT_STRING, oci_cgroup_device_type, offsetof(struct device_data, type), 0 },
+ { "major", JSON_VARIANT_UNSIGNED, oci_device_major, offsetof(struct device_data, major), 0 },
+ { "minor", JSON_VARIANT_UNSIGNED, oci_device_minor, offsetof(struct device_data, minor), 0 },
+ { "access", JSON_VARIANT_STRING, oci_cgroup_device_access, 0, 0 },
+ {}
+ };
+
+ r = json_dispatch(e, table, oci_unexpected, flags, &data);
+ if (r < 0)
+ return r;
+
+ if (!data.allow) {
+ /* The fact that OCI allows 'deny' entries makes really no sense, as 'allow'
+ * vs. 'deny' for the devices cgroup controller is really not about allow-listing and
+ * deny-listing but about adding and removing entries from the allow list. Since we
+ * always start out with an empty allow list we hence ignore the whole thing, as
+ * removing entries which don't exist make no sense. We'll log about this, since this
+ * is really borked in the spec, with one exception: the entry that's supposed to
+ * drop the kernel's default we ignore silently */
+
+ if (!data.r || !data.w || !data.m || data.type != 0 || data.major != UINT_MAX || data.minor != UINT_MAX)
+ json_log(v, flags|JSON_WARNING, 0, "Devices cgroup allow list with arbitrary 'allow' entries not supported, ignoring.");
+
+ /* We ignore the 'deny' entry as for us that's implied */
+ continue;
+ }
+
+ if (!data.r && !data.w && !data.m) {
+ json_log(v, flags|LOG_WARNING, 0, "Device cgroup allow list entry with no effect found, ignoring.");
+ continue;
+ }
+
+ if (data.minor != UINT_MAX && data.major == UINT_MAX)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Device cgroup allow list entries with minors but no majors not supported.");
+
+ if (data.major != UINT_MAX && data.type == 0)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Device cgroup allow list entries with majors but no device node type not supported.");
+
+ if (data.type == 0) {
+ if (data.r && data.w && data.m) /* a catchall allow list entry means we are looking at a noop */
+ noop = true;
+ else
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Device cgroup allow list entries with no type not supported.");
+ }
+
+ a = reallocarray(list, n_list + 1, sizeof(struct device_data));
+ if (!a)
+ return log_oom();
+
+ list = a;
+ list[n_list++] = data;
+ }
+
+ if (noop)
+ return 0;
+
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_open_container(s->properties, 'r', "sv");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_append(s->properties, "s", "DeviceAllow");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_open_container(s->properties, 'v', "a(ss)");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_open_container(s->properties, 'a', "(ss)");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ for (i = 0; i < n_list; i++) {
+ _cleanup_free_ char *pattern = NULL;
+ char access[4];
+ size_t n = 0;
+
+ if (list[i].minor == UINT_MAX) {
+ const char *t;
+
+ if (list[i].type == S_IFBLK)
+ t = "block";
+ else {
+ assert(list[i].type == S_IFCHR);
+ t = "char";
+ }
+
+ if (list[i].major == UINT_MAX) {
+ pattern = strjoin(t, "-*");
+ if (!pattern)
+ return log_oom();
+ } else {
+ if (asprintf(&pattern, "%s-%u", t, list[i].major) < 0)
+ return log_oom();
+ }
+
+ } else {
+ assert(list[i].major != UINT_MAX); /* If a minor is specified, then a major also needs to be specified */
+
+ r = device_path_make_major_minor(list[i].type, makedev(list[i].major, list[i].minor), &pattern);
+ if (r < 0)
+ return log_oom();
+ }
+
+ if (list[i].r)
+ access[n++] = 'r';
+ if (list[i].w)
+ access[n++] = 'w';
+ if (list[i].m)
+ access[n++] = 'm';
+ access[n] = 0;
+
+ assert(n > 0);
+
+ r = sd_bus_message_append(s->properties, "(ss)", pattern, access);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ r = sd_bus_message_close_container(s->properties);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_close_container(s->properties);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_close_container(s->properties);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ return 0;
+}
+
+static int oci_cgroup_memory_limit(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ uint64_t *m = ASSERT_PTR(userdata);
+ uint64_t k;
+
+ if (json_variant_is_negative(v)) {
+ *m = UINT64_MAX;
+ return 0;
+ }
+
+ if (!json_variant_is_unsigned(v))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Memory limit is not an unsigned integer");
+
+ k = json_variant_unsigned(v);
+ if (k >= UINT64_MAX)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Memory limit too large: %" PRIu64, k);
+
+ *m = (uint64_t) k;
+ return 0;
+}
+
+static int oci_cgroup_memory(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ struct memory_data {
+ uint64_t limit;
+ uint64_t reservation;
+ uint64_t swap;
+ } data = {
+ .limit = UINT64_MAX,
+ .reservation = UINT64_MAX,
+ .swap = UINT64_MAX,
+ };
+
+ static const JsonDispatch table[] = {
+ { "limit", JSON_VARIANT_NUMBER, oci_cgroup_memory_limit, offsetof(struct memory_data, limit), 0 },
+ { "reservation", JSON_VARIANT_NUMBER, oci_cgroup_memory_limit, offsetof(struct memory_data, reservation), 0 },
+ { "swap", JSON_VARIANT_NUMBER, oci_cgroup_memory_limit, offsetof(struct memory_data, swap), 0 },
+ { "kernel", JSON_VARIANT_NUMBER, oci_unsupported, 0, JSON_PERMISSIVE },
+ { "kernelTCP", JSON_VARIANT_NUMBER, oci_unsupported, 0, JSON_PERMISSIVE },
+ { "swapiness", JSON_VARIANT_NUMBER, oci_unsupported, 0, JSON_PERMISSIVE },
+ { "disableOOMKiller", JSON_VARIANT_NUMBER, oci_unsupported, 0, JSON_PERMISSIVE },
+ {}
+ };
+
+ Settings *s = userdata;
+ int r;
+
+ r = json_dispatch(v, table, oci_unexpected, flags, &data);
+ if (r < 0)
+ return r;
+
+ if (data.swap != UINT64_MAX) {
+ if (data.limit == UINT64_MAX)
+ json_log(v, flags|LOG_WARNING, 0, "swap limit without memory limit is not supported, ignoring.");
+ else if (data.swap < data.limit)
+ json_log(v, flags|LOG_WARNING, 0, "swap limit is below memory limit, ignoring.");
+ else {
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", "MemorySwapMax", "t", data.swap - data.limit);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+ }
+
+ if (data.limit != UINT64_MAX) {
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", "MemoryMax", "t", data.limit);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ if (data.reservation != UINT64_MAX) {
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", "MemoryLow", "t", data.reservation);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ return 0;
+}
+
+struct cpu_data {
+ uint64_t shares;
+ uint64_t quota;
+ uint64_t period;
+ CPUSet cpu_set;
+};
+
+static int oci_cgroup_cpu_shares(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ uint64_t *u = ASSERT_PTR(userdata);
+ uint64_t k;
+
+ k = json_variant_unsigned(v);
+ if (k < CGROUP_CPU_SHARES_MIN || k > CGROUP_CPU_SHARES_MAX)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "shares value out of range.");
+
+ *u = (uint64_t) k;
+ return 0;
+}
+
+static int oci_cgroup_cpu_quota(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ uint64_t *u = ASSERT_PTR(userdata);
+ uint64_t k;
+
+ k = json_variant_unsigned(v);
+ if (k <= 0 || k >= UINT64_MAX)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "period/quota value out of range.");
+
+ *u = (uint64_t) k;
+ return 0;
+}
+
+static int oci_cgroup_cpu_cpus(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ struct cpu_data *data = ASSERT_PTR(userdata);
+ CPUSet set;
+ const char *n;
+ int r;
+
+ assert_se(n = json_variant_string(v));
+
+ r = parse_cpu_set(n, &set);
+ if (r < 0)
+ return json_log(v, flags, r, "Failed to parse CPU set specification: %s", n);
+
+ cpu_set_reset(&data->cpu_set);
+ data->cpu_set = set;
+
+ return 0;
+}
+
+static int oci_cgroup_cpu(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "shares", JSON_VARIANT_UNSIGNED, oci_cgroup_cpu_shares, offsetof(struct cpu_data, shares), 0 },
+ { "quota", JSON_VARIANT_UNSIGNED, oci_cgroup_cpu_quota, offsetof(struct cpu_data, quota), 0 },
+ { "period", JSON_VARIANT_UNSIGNED, oci_cgroup_cpu_quota, offsetof(struct cpu_data, period), 0 },
+ { "realtimeRuntime", JSON_VARIANT_UNSIGNED, oci_unsupported, 0, 0 },
+ { "realtimePeriod", JSON_VARIANT_UNSIGNED, oci_unsupported, 0, 0 },
+ { "cpus", JSON_VARIANT_STRING, oci_cgroup_cpu_cpus, 0, 0 },
+ { "mems", JSON_VARIANT_STRING, oci_unsupported, 0, 0 },
+ {}
+ };
+
+ struct cpu_data data = {
+ .shares = UINT64_MAX,
+ .quota = UINT64_MAX,
+ .period = UINT64_MAX,
+ };
+
+ Settings *s = userdata;
+ int r;
+
+ r = json_dispatch(v, table, oci_unexpected, flags, &data);
+ if (r < 0) {
+ cpu_set_reset(&data.cpu_set);
+ return r;
+ }
+
+ cpu_set_reset(&s->cpu_set);
+ s->cpu_set = data.cpu_set;
+
+ if (data.shares != UINT64_MAX) {
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", "CPUShares", "t", data.shares);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ if (data.quota != UINT64_MAX && data.period != UINT64_MAX) {
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", "CPUQuotaPerSecUSec", "t", (uint64_t) (data.quota * USEC_PER_SEC / data.period));
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ } else if ((data.quota != UINT64_MAX) != (data.period != UINT64_MAX))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "CPU quota and period not used together.");
+
+ return 0;
+}
+
+static int oci_cgroup_block_io_weight(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ uint64_t k;
+ int r;
+
+ k = json_variant_unsigned(v);
+ if (k < CGROUP_BLKIO_WEIGHT_MIN || k > CGROUP_BLKIO_WEIGHT_MAX)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Block I/O weight out of range.");
+
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", "BlockIOWeight", "t", (uint64_t) k);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ return 0;
+}
+
+static int oci_cgroup_block_io_weight_device(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ struct device_data {
+ unsigned major;
+ unsigned minor;
+ uint64_t weight;
+ } data = {
+ .major = UINT_MAX,
+ .minor = UINT_MAX,
+ .weight = UINT64_MAX,
+ };
+
+ static const JsonDispatch table[] = {
+ { "major", JSON_VARIANT_UNSIGNED, oci_device_major, offsetof(struct device_data, major), JSON_MANDATORY },
+ { "minor", JSON_VARIANT_UNSIGNED, oci_device_minor, offsetof(struct device_data, minor), JSON_MANDATORY },
+ { "weight", JSON_VARIANT_UNSIGNED, json_dispatch_uint64, offsetof(struct device_data, weight), 0 },
+ { "leafWeight", JSON_VARIANT_INTEGER, oci_unsupported, 0, JSON_PERMISSIVE },
+ {}
+ };
+
+ _cleanup_free_ char *path = NULL;
+
+ r = json_dispatch(e, table, oci_unexpected, flags, &data);
+ if (r < 0)
+ return r;
+
+ if (data.weight == UINT64_MAX)
+ continue;
+
+ if (data.weight < CGROUP_BLKIO_WEIGHT_MIN || data.weight > CGROUP_BLKIO_WEIGHT_MAX)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Block I/O device weight out of range.");
+
+ r = device_path_make_major_minor(S_IFBLK, makedev(data.major, data.minor), &path);
+ if (r < 0)
+ return json_log(v, flags, r, "Failed to build device path: %m");
+
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", "BlockIODeviceWeight", "a(st)", 1, path, (uint64_t) data.weight);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ return 0;
+}
+
+static int oci_cgroup_block_io_throttle(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ const char *pname;
+ JsonVariant *e;
+ int r;
+
+ pname = streq(name, "throttleReadBpsDevice") ? "IOReadBandwidthMax" :
+ streq(name, "throttleWriteBpsDevice") ? "IOWriteBandwidthMax" :
+ streq(name, "throttleReadIOPSDevice") ? "IOReadIOPSMax" :
+ "IOWriteIOPSMax";
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ struct device_data {
+ unsigned major;
+ unsigned minor;
+ uint64_t rate;
+ } data = {
+ .major = UINT_MAX,
+ .minor = UINT_MAX,
+ };
+
+ static const JsonDispatch table[] = {
+ { "major", JSON_VARIANT_UNSIGNED, oci_device_major, offsetof(struct device_data, major), JSON_MANDATORY },
+ { "minor", JSON_VARIANT_UNSIGNED, oci_device_minor, offsetof(struct device_data, minor), JSON_MANDATORY },
+ { "rate", JSON_VARIANT_UNSIGNED, json_dispatch_uint64, offsetof(struct device_data, rate), JSON_MANDATORY },
+ {}
+ };
+
+ _cleanup_free_ char *path = NULL;
+
+ r = json_dispatch(e, table, oci_unexpected, flags, &data);
+ if (r < 0)
+ return r;
+
+ if (data.rate >= UINT64_MAX)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Block I/O device rate out of range.");
+
+ r = device_path_make_major_minor(S_IFBLK, makedev(data.major, data.minor), &path);
+ if (r < 0)
+ return json_log(v, flags, r, "Failed to build device path: %m");
+
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", pname, "a(st)", 1, path, (uint64_t) data.rate);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ return 0;
+}
+
+static int oci_cgroup_block_io(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "weight", JSON_VARIANT_UNSIGNED, oci_cgroup_block_io_weight, 0, 0 },
+ { "leafWeight", JSON_VARIANT_UNSIGNED, oci_unsupported, 0, JSON_PERMISSIVE },
+ { "weightDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_weight_device, 0, 0 },
+ { "throttleReadBpsDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_throttle, 0, 0 },
+ { "throttleWriteBpsDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_throttle, 0, 0 },
+ { "throttleReadIOPSDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_throttle, 0, 0 },
+ { "throttleWriteIOPSDevice", JSON_VARIANT_ARRAY, oci_cgroup_block_io_throttle, 0, 0 },
+ {}
+ };
+
+ return json_dispatch(v, table, oci_unexpected, flags, userdata);
+}
+
+static int oci_cgroup_pids(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "limit", JSON_VARIANT_NUMBER, json_dispatch_variant, 0, JSON_MANDATORY },
+ {}
+ };
+
+ _cleanup_(json_variant_unrefp) JsonVariant *k = NULL;
+ Settings *s = ASSERT_PTR(userdata);
+ uint64_t m;
+ int r;
+
+ r = json_dispatch(v, table, oci_unexpected, flags, &k);
+ if (r < 0)
+ return r;
+
+ if (json_variant_is_negative(k))
+ m = UINT64_MAX;
+ else {
+ if (!json_variant_is_unsigned(k))
+ return json_log(k, flags, SYNTHETIC_ERRNO(EINVAL),
+ "pids limit not unsigned integer, refusing.");
+
+ m = (uint64_t) json_variant_unsigned(k);
+
+ if ((uint64_t) m != json_variant_unsigned(k))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "pids limit out of range, refusing.");
+ }
+
+ r = settings_allocate_properties(s);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_append(s->properties, "(sv)", "TasksMax", "t", m);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ return 0;
+}
+
+static int oci_resources(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "devices", JSON_VARIANT_ARRAY, oci_cgroup_devices, 0, 0 },
+ { "memory", JSON_VARIANT_OBJECT, oci_cgroup_memory, 0, 0 },
+ { "cpu", JSON_VARIANT_OBJECT, oci_cgroup_cpu, 0, 0 },
+ { "blockIO", JSON_VARIANT_OBJECT, oci_cgroup_block_io, 0, 0 },
+ { "hugepageLimits", JSON_VARIANT_ARRAY, oci_unsupported, 0, 0 },
+ { "network", JSON_VARIANT_OBJECT, oci_unsupported, 0, 0 },
+ { "pids", JSON_VARIANT_OBJECT, oci_cgroup_pids, 0, 0 },
+ { "rdma", JSON_VARIANT_OBJECT, oci_unsupported, 0, 0 },
+ {}
+ };
+
+ return json_dispatch(v, table, oci_unexpected, flags, userdata);
+}
+
+static bool sysctl_key_valid(const char *s) {
+ bool dot = true;
+
+ /* Note that we are a bit stricter here than in systemd-sysctl, as that inherited semantics from the old sysctl
+ * tool, which were really weird (as it swaps / and . in both ways) */
+
+ if (isempty(s))
+ return false;
+
+ for (; *s; s++) {
+
+ if (*s <= ' ' || *s >= 127)
+ return false;
+ if (*s == '/')
+ return false;
+ if (*s == '.') {
+
+ if (dot) /* Don't allow two dots next to each other (or at the beginning) */
+ return false;
+
+ dot = true;
+ } else
+ dot = false;
+ }
+
+ if (dot) /* don't allow a dot at the end */
+ return false;
+
+ return true;
+}
+
+static int oci_sysctl(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *w;
+ const char *k;
+ int r;
+
+ JSON_VARIANT_OBJECT_FOREACH(k, w, v) {
+ const char *m;
+
+ if (!json_variant_is_string(w))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "sysctl parameter is not a string, refusing.");
+
+ assert_se(m = json_variant_string(w));
+
+ if (!sysctl_key_valid(k))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "sysctl key invalid, refusing: %s", k);
+
+ r = strv_extend_strv(&s->sysctl, STRV_MAKE(k, m), false);
+ if (r < 0)
+ return log_oom();
+ }
+
+ return 0;
+}
+
+#if HAVE_SECCOMP
+static int oci_seccomp_action_from_string(const char *name, uint32_t *ret) {
+
+ static const struct {
+ const char *name;
+ uint32_t action;
+ } table[] = {
+ { "SCMP_ACT_ALLOW", SCMP_ACT_ALLOW },
+ { "SCMP_ACT_ERRNO", SCMP_ACT_ERRNO(EPERM) }, /* the OCI spec doesn't document the error, but it appears EPERM is supposed to be used */
+ { "SCMP_ACT_KILL", SCMP_ACT_KILL },
+#ifdef SCMP_ACT_KILL_PROCESS
+ { "SCMP_ACT_KILL_PROCESS", SCMP_ACT_KILL_PROCESS },
+#endif
+#ifdef SCMP_ACT_KILL_THREAD
+ { "SCMP_ACT_KILL_THREAD", SCMP_ACT_KILL_THREAD },
+#endif
+#ifdef SCMP_ACT_LOG
+ { "SCMP_ACT_LOG", SCMP_ACT_LOG },
+#endif
+ { "SCMP_ACT_TRAP", SCMP_ACT_TRAP },
+
+ /* We don't support SCMP_ACT_TRACE because that requires a tracer, and that doesn't really make sense
+ * here */
+ };
+
+ size_t i;
+
+ for (i = 0; i < ELEMENTSOF(table); i++)
+ if (streq_ptr(name, table[i].name)) {
+ *ret = table[i].action;
+ return 0;
+ }
+
+ return -EINVAL;
+}
+
+static int oci_seccomp_arch_from_string(const char *name, uint32_t *ret) {
+
+ static const struct {
+ const char *name;
+ uint32_t arch;
+ } table[] = {
+ { "SCMP_ARCH_AARCH64", SCMP_ARCH_AARCH64 },
+ { "SCMP_ARCH_ARM", SCMP_ARCH_ARM },
+ { "SCMP_ARCH_MIPS", SCMP_ARCH_MIPS },
+ { "SCMP_ARCH_MIPS64", SCMP_ARCH_MIPS64 },
+ { "SCMP_ARCH_MIPS64N32", SCMP_ARCH_MIPS64N32 },
+ { "SCMP_ARCH_MIPSEL", SCMP_ARCH_MIPSEL },
+ { "SCMP_ARCH_MIPSEL64", SCMP_ARCH_MIPSEL64 },
+ { "SCMP_ARCH_MIPSEL64N32", SCMP_ARCH_MIPSEL64N32 },
+ { "SCMP_ARCH_NATIVE", SCMP_ARCH_NATIVE },
+#ifdef SCMP_ARCH_PARISC
+ { "SCMP_ARCH_PARISC", SCMP_ARCH_PARISC },
+#endif
+#ifdef SCMP_ARCH_PARISC64
+ { "SCMP_ARCH_PARISC64", SCMP_ARCH_PARISC64 },
+#endif
+ { "SCMP_ARCH_PPC", SCMP_ARCH_PPC },
+ { "SCMP_ARCH_PPC64", SCMP_ARCH_PPC64 },
+ { "SCMP_ARCH_PPC64LE", SCMP_ARCH_PPC64LE },
+#ifdef SCMP_ARCH_RISCV64
+ { "SCMP_ARCH_RISCV64", SCMP_ARCH_RISCV64 },
+#endif
+ { "SCMP_ARCH_S390", SCMP_ARCH_S390 },
+ { "SCMP_ARCH_S390X", SCMP_ARCH_S390X },
+ { "SCMP_ARCH_X32", SCMP_ARCH_X32 },
+ { "SCMP_ARCH_X86", SCMP_ARCH_X86 },
+ { "SCMP_ARCH_X86_64", SCMP_ARCH_X86_64 },
+ };
+
+ size_t i;
+
+ for (i = 0; i < ELEMENTSOF(table); i++)
+ if (streq_ptr(table[i].name, name)) {
+ *ret = table[i].arch;
+ return 0;
+ }
+
+ return -EINVAL;
+}
+
+static int oci_seccomp_compare_from_string(const char *name, enum scmp_compare *ret) {
+
+ static const struct {
+ const char *name;
+ enum scmp_compare op;
+ } table[] = {
+ { "SCMP_CMP_NE", SCMP_CMP_NE },
+ { "SCMP_CMP_LT", SCMP_CMP_LT },
+ { "SCMP_CMP_LE", SCMP_CMP_LE },
+ { "SCMP_CMP_EQ", SCMP_CMP_EQ },
+ { "SCMP_CMP_GE", SCMP_CMP_GE },
+ { "SCMP_CMP_GT", SCMP_CMP_GT },
+ { "SCMP_CMP_MASKED_EQ", SCMP_CMP_MASKED_EQ },
+ };
+
+ size_t i;
+
+ for (i = 0; i < ELEMENTSOF(table); i++)
+ if (streq_ptr(table[i].name, name)) {
+ *ret = table[i].op;
+ return 0;
+ }
+
+ return -EINVAL;
+}
+
+static int oci_seccomp_archs(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ scmp_filter_ctx *sc = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ uint32_t a;
+
+ if (!json_variant_is_string(e))
+ return json_log(e, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Architecture entry is not a string");
+
+ r = oci_seccomp_arch_from_string(json_variant_string(e), &a);
+ if (r < 0)
+ return json_log(e, flags, r, "Unknown architecture: %s", json_variant_string(e));
+
+ r = seccomp_arch_add(sc, a);
+ if (r == -EEXIST)
+ continue;
+ if (r < 0)
+ return json_log(e, flags, r, "Failed to add architecture to seccomp filter: %m");
+ }
+
+ return 0;
+}
+
+struct syscall_rule {
+ char **names;
+ uint32_t action;
+ struct scmp_arg_cmp *arguments;
+ size_t n_arguments;
+};
+
+static void syscall_rule_free(struct syscall_rule *rule) {
+ assert(rule);
+
+ strv_free(rule->names);
+ free(rule->arguments);
+};
+
+static int oci_seccomp_action(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ uint32_t *action = ASSERT_PTR(userdata);
+ int r;
+
+ r = oci_seccomp_action_from_string(json_variant_string(v), action);
+ if (r < 0)
+ return json_log(v, flags, r, "Unknown system call action '%s': %m", json_variant_string(v));
+
+ return 0;
+}
+
+static int oci_seccomp_op(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ enum scmp_compare *op = ASSERT_PTR(userdata);
+ int r;
+
+ r = oci_seccomp_compare_from_string(json_variant_string(v), op);
+ if (r < 0)
+ return json_log(v, flags, r, "Unknown seccomp operator '%s': %m", json_variant_string(v));
+
+ return 0;
+}
+
+static int oci_seccomp_args(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ struct syscall_rule *rule = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ static const struct JsonDispatch table[] = {
+ { "index", JSON_VARIANT_UNSIGNED, json_dispatch_uint32, offsetof(struct scmp_arg_cmp, arg), JSON_MANDATORY },
+ { "value", JSON_VARIANT_UNSIGNED, json_dispatch_uint64, offsetof(struct scmp_arg_cmp, datum_a), JSON_MANDATORY },
+ { "valueTwo", JSON_VARIANT_UNSIGNED, json_dispatch_uint64, offsetof(struct scmp_arg_cmp, datum_b), 0 },
+ { "op", JSON_VARIANT_STRING, oci_seccomp_op, offsetof(struct scmp_arg_cmp, op), JSON_MANDATORY },
+ {},
+ };
+
+ struct scmp_arg_cmp *a, *p;
+ int expected;
+
+ a = reallocarray(rule->arguments, rule->n_arguments + 1, sizeof(struct syscall_rule));
+ if (!a)
+ return log_oom();
+
+ rule->arguments = a;
+ p = rule->arguments + rule->n_arguments;
+
+ *p = (struct scmp_arg_cmp) {
+ .arg = 0,
+ .datum_a = 0,
+ .datum_b = 0,
+ .op = 0,
+ };
+
+ r = json_dispatch(e, table, oci_unexpected, flags, p);
+ if (r < 0)
+ return r;
+
+ expected = p->op == SCMP_CMP_MASKED_EQ ? 4 : 3;
+ if (r != expected)
+ json_log(e, flags|JSON_WARNING, 0, "Wrong number of system call arguments for JSON data data, ignoring.");
+
+ /* Note that we are a bit sloppy here and do not insist that SCMP_CMP_MASKED_EQ gets two datum values,
+ * and the other only one. That's because buildah for example by default calls things with
+ * SCMP_CMP_MASKED_EQ but only one argument. We use 0 when the value is not specified. */
+
+ rule->n_arguments++;
+ }
+
+ return 0;
+}
+
+static int oci_seccomp_syscalls(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ scmp_filter_ctx *sc = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ static const JsonDispatch table[] = {
+ { "names", JSON_VARIANT_ARRAY, json_dispatch_strv, offsetof(struct syscall_rule, names), JSON_MANDATORY },
+ { "action", JSON_VARIANT_STRING, oci_seccomp_action, offsetof(struct syscall_rule, action), JSON_MANDATORY },
+ { "args", JSON_VARIANT_ARRAY, oci_seccomp_args, 0, 0 },
+ {}
+ };
+ struct syscall_rule rule = {
+ .action = UINT32_MAX,
+ };
+
+ r = json_dispatch(e, table, oci_unexpected, flags, &rule);
+ if (r < 0)
+ goto fail_rule;
+
+ if (strv_isempty(rule.names)) {
+ json_log(e, flags, 0, "System call name list is empty.");
+ r = -EINVAL;
+ goto fail_rule;
+ }
+
+ STRV_FOREACH(i, rule.names) {
+ int nr;
+
+ nr = seccomp_syscall_resolve_name(*i);
+ if (nr == __NR_SCMP_ERROR) {
+ log_debug("Unknown syscall %s, skipping.", *i);
+ continue;
+ }
+
+ r = seccomp_rule_add_array(sc, rule.action, nr, rule.n_arguments, rule.arguments);
+ if (r < 0)
+ goto fail_rule;
+ }
+
+ syscall_rule_free(&rule);
+ continue;
+
+ fail_rule:
+ syscall_rule_free(&rule);
+ return r;
+ }
+
+ return 0;
+}
+#endif
+
+static int oci_seccomp(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+#if HAVE_SECCOMP
+ static const JsonDispatch table[] = {
+ { "defaultAction", JSON_VARIANT_STRING, NULL, 0, JSON_MANDATORY },
+ { "architectures", JSON_VARIANT_ARRAY, oci_seccomp_archs, 0, 0 },
+ { "syscalls", JSON_VARIANT_ARRAY, oci_seccomp_syscalls, 0, 0 },
+ {}
+ };
+
+ _cleanup_(seccomp_releasep) scmp_filter_ctx sc = NULL;
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *def;
+ uint32_t d;
+ int r;
+
+ def = json_variant_by_key(v, "defaultAction");
+ if (!def)
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL), "defaultAction element missing.");
+
+ if (!json_variant_is_string(def))
+ return json_log(def, flags, SYNTHETIC_ERRNO(EINVAL), "defaultAction is not a string.");
+
+ r = oci_seccomp_action_from_string(json_variant_string(def), &d);
+ if (r < 0)
+ return json_log(def, flags, r, "Unknown default action: %s", json_variant_string(def));
+
+ sc = seccomp_init(d);
+ if (!sc)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ENOMEM), "Couldn't allocate seccomp object.");
+
+ r = json_dispatch(v, table, oci_unexpected, flags, sc);
+ if (r < 0)
+ return r;
+
+ seccomp_release(s->seccomp);
+ s->seccomp = TAKE_PTR(sc);
+ return 0;
+#else
+ return json_log(v, flags, SYNTHETIC_ERRNO(EOPNOTSUPP), "libseccomp support not enabled, can't parse seccomp object.");
+#endif
+}
+
+static int oci_rootfs_propagation(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ const char *s;
+
+ s = json_variant_string(v);
+
+ if (streq(s, "shared"))
+ return 0;
+
+ json_log(v, flags|JSON_DEBUG, 0, "Ignoring rootfsPropagation setting '%s'.", s);
+ return 0;
+}
+
+static int oci_masked_paths(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ _cleanup_free_ char *destination = NULL;
+ CustomMount *m;
+ const char *p;
+
+ if (!json_variant_is_string(e))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Path is not a string, refusing.");
+
+ assert_se(p = json_variant_string(e));
+
+ if (!path_is_absolute(p))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Path is not absolute, refusing: %s", p);
+
+ if (oci_exclude_mount(p))
+ continue;
+
+ destination = strdup(p);
+ if (!destination)
+ return log_oom();
+
+ m = custom_mount_add(&s->custom_mounts, &s->n_custom_mounts, CUSTOM_MOUNT_INACCESSIBLE);
+ if (!m)
+ return log_oom();
+
+ m->destination = TAKE_PTR(destination);
+
+ /* The spec doesn't say this, but apparently pre-existing implementations are lenient towards
+ * non-existing paths to mask. Let's hence be too. */
+ m->graceful = true;
+ }
+
+ return 0;
+}
+
+static int oci_readonly_paths(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+ _cleanup_free_ char *source = NULL, *destination = NULL;
+ CustomMount *m;
+ const char *p;
+
+ if (!json_variant_is_string(e))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Path is not a string, refusing.");
+
+ assert_se(p = json_variant_string(e));
+
+ if (!path_is_absolute(p))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Path is not absolute, refusing: %s", p);
+
+ if (oci_exclude_mount(p))
+ continue;
+
+ source = strjoin("+", p);
+ if (!source)
+ return log_oom();
+
+ destination = strdup(p);
+ if (!destination)
+ return log_oom();
+
+ m = custom_mount_add(&s->custom_mounts, &s->n_custom_mounts, CUSTOM_MOUNT_BIND);
+ if (!m)
+ return log_oom();
+
+ m->source = TAKE_PTR(source);
+ m->destination = TAKE_PTR(destination);
+ m->read_only = true;
+ }
+
+ return 0;
+}
+
+static int oci_linux(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "namespaces", JSON_VARIANT_ARRAY, oci_namespaces, 0, 0 },
+ { "uidMappings", JSON_VARIANT_ARRAY, oci_uid_gid_mappings, 0, 0 },
+ { "gidMappings", JSON_VARIANT_ARRAY, oci_uid_gid_mappings, 0, 0 },
+ { "devices", JSON_VARIANT_ARRAY, oci_devices, 0, 0 },
+ { "cgroupsPath", JSON_VARIANT_STRING, oci_cgroups_path, 0, 0 },
+ { "resources", JSON_VARIANT_OBJECT, oci_resources, 0, 0 },
+ { "intelRdt", JSON_VARIANT_OBJECT, oci_unsupported, 0, JSON_PERMISSIVE },
+ { "sysctl", JSON_VARIANT_OBJECT, oci_sysctl, 0, 0 },
+ { "seccomp", JSON_VARIANT_OBJECT, oci_seccomp, 0, 0 },
+ { "rootfsPropagation", JSON_VARIANT_STRING, oci_rootfs_propagation, 0, 0 },
+ { "maskedPaths", JSON_VARIANT_ARRAY, oci_masked_paths, 0, 0 },
+ { "readonlyPaths", JSON_VARIANT_ARRAY, oci_readonly_paths, 0, 0 },
+ { "mountLabel", JSON_VARIANT_STRING, oci_unsupported, 0, JSON_PERMISSIVE },
+ {}
+ };
+
+ return json_dispatch(v, table, oci_unexpected, flags, userdata);
+}
+
+static int oci_hook_timeout(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ usec_t *u = userdata;
+ uint64_t k;
+
+ k = json_variant_unsigned(v);
+ if (k == 0 || k > (UINT64_MAX-1)/USEC_PER_SEC)
+ return json_log(v, flags, SYNTHETIC_ERRNO(ERANGE),
+ "Hook timeout value out of range.");
+
+ *u = k * USEC_PER_SEC;
+ return 0;
+}
+
+static int oci_hooks_array(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ Settings *s = ASSERT_PTR(userdata);
+ JsonVariant *e;
+ int r;
+
+ JSON_VARIANT_ARRAY_FOREACH(e, v) {
+
+ static const JsonDispatch table[] = {
+ { "path", JSON_VARIANT_STRING, oci_absolute_path, offsetof(OciHook, path), JSON_MANDATORY },
+ { "args", JSON_VARIANT_ARRAY, oci_args, offsetof(OciHook, args), 0 },
+ { "env", JSON_VARIANT_ARRAY, oci_env, offsetof(OciHook, env), 0 },
+ { "timeout", JSON_VARIANT_UNSIGNED, oci_hook_timeout, offsetof(OciHook, timeout), 0 },
+ {}
+ };
+
+ OciHook *a, **array, *new_item;
+ size_t *n_array;
+
+ if (streq(name, "prestart")) {
+ array = &s->oci_hooks_prestart;
+ n_array = &s->n_oci_hooks_prestart;
+ } else if (streq(name, "poststart")) {
+ array = &s->oci_hooks_poststart;
+ n_array = &s->n_oci_hooks_poststart;
+ } else {
+ assert(streq(name, "poststop"));
+ array = &s->oci_hooks_poststop;
+ n_array = &s->n_oci_hooks_poststop;
+ }
+
+ a = reallocarray(*array, *n_array + 1, sizeof(OciHook));
+ if (!a)
+ return log_oom();
+
+ *array = a;
+ new_item = a + *n_array;
+
+ *new_item = (OciHook) {
+ .timeout = USEC_INFINITY,
+ };
+
+ r = json_dispatch(e, table, oci_unexpected, flags, new_item);
+ if (r < 0) {
+ free(new_item->path);
+ strv_free(new_item->args);
+ strv_free(new_item->env);
+ return r;
+ }
+
+ (*n_array) ++;
+ }
+
+ return 0;
+}
+
+static int oci_hooks(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+
+ static const JsonDispatch table[] = {
+ { "prestart", JSON_VARIANT_OBJECT, oci_hooks_array, 0, 0 },
+ { "poststart", JSON_VARIANT_OBJECT, oci_hooks_array, 0, 0 },
+ { "poststop", JSON_VARIANT_OBJECT, oci_hooks_array, 0, 0 },
+ {}
+ };
+
+ return json_dispatch(v, table, oci_unexpected, flags, userdata);
+}
+
+static int oci_annotations(const char *name, JsonVariant *v, JsonDispatchFlags flags, void *userdata) {
+ JsonVariant *w;
+ const char *k;
+
+ JSON_VARIANT_OBJECT_FOREACH(k, w, v) {
+
+ if (isempty(k))
+ return json_log(v, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Annotation with empty key, refusing.");
+
+ if (!json_variant_is_string(w))
+ return json_log(w, flags, SYNTHETIC_ERRNO(EINVAL),
+ "Annotation has non-string value, refusing.");
+
+ json_log(w, flags|JSON_DEBUG, 0, "Ignoring annotation '%s' with value '%s'.", k, json_variant_string(w));
+ }
+
+ return 0;
+}
+
+int oci_load(FILE *f, const char *bundle, Settings **ret) {
+
+ static const JsonDispatch table[] = {
+ { "ociVersion", JSON_VARIANT_STRING, NULL, 0, JSON_MANDATORY },
+ { "process", JSON_VARIANT_OBJECT, oci_process, 0, 0 },
+ { "root", JSON_VARIANT_OBJECT, oci_root, 0, 0 },
+ { "hostname", JSON_VARIANT_STRING, oci_hostname, 0, 0 },
+ { "mounts", JSON_VARIANT_ARRAY, oci_mounts, 0, 0 },
+ { "linux", JSON_VARIANT_OBJECT, oci_linux, 0, 0 },
+ { "hooks", JSON_VARIANT_OBJECT, oci_hooks, 0, 0 },
+ { "annotations", JSON_VARIANT_OBJECT, oci_annotations, 0, 0 },
+ {}
+ };
+
+ _cleanup_(json_variant_unrefp) JsonVariant *oci = NULL;
+ _cleanup_(settings_freep) Settings *s = NULL;
+ unsigned line = 0, column = 0;
+ JsonVariant *v;
+ const char *path;
+ int r;
+
+ assert_se(bundle);
+
+ path = strjoina(bundle, "/config.json");
+
+ r = json_parse_file(f, path, 0, &oci, &line, &column);
+ if (r < 0) {
+ if (line != 0 && column != 0)
+ return log_error_errno(r, "Failed to parse '%s' at %u:%u: %m", path, line, column);
+ else
+ return log_error_errno(r, "Failed to parse '%s': %m", path);
+ }
+
+ v = json_variant_by_key(oci, "ociVersion");
+ if (!v)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "JSON file '%s' is not an OCI bundle configuration file. Refusing.",
+ path);
+ if (!streq_ptr(json_variant_string(v), "1.0.0"))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "OCI bundle version not supported: %s",
+ strna(json_variant_string(v)));
+
+ // {
+ // _cleanup_free_ char *formatted = NULL;
+ // assert_se(json_variant_format(oci, JSON_FORMAT_PRETTY|JSON_FORMAT_COLOR, &formatted) >= 0);
+ // fputs(formatted, stdout);
+ // }
+
+ s = settings_new();
+ if (!s)
+ return log_oom();
+
+ s->start_mode = START_PID1;
+ s->resolv_conf = RESOLV_CONF_OFF;
+ s->link_journal = LINK_NO;
+ s->timezone = TIMEZONE_OFF;
+
+ s->bundle = strdup(bundle);
+ if (!s->bundle)
+ return log_oom();
+
+ r = json_dispatch(oci, table, oci_unexpected, 0, s);
+ if (r < 0)
+ return r;
+
+ if (s->properties) {
+ r = sd_bus_message_seal(s->properties, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Cannot seal properties bus message: %m");
+ }
+
+ *ret = TAKE_PTR(s);
+ return 0;
+}
diff --git a/src/nspawn/nspawn-oci.h b/src/nspawn/nspawn-oci.h
new file mode 100644
index 0000000..ee72c91
--- /dev/null
+++ b/src/nspawn/nspawn-oci.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include "nspawn-settings.h"
+
+int oci_load(FILE *f, const char *path, Settings **ret);
diff --git a/src/nspawn/nspawn-patch-uid.c b/src/nspawn/nspawn-patch-uid.c
new file mode 100644
index 0000000..75fa931
--- /dev/null
+++ b/src/nspawn/nspawn-patch-uid.c
@@ -0,0 +1,477 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <fcntl.h>
+#include <sys/statvfs.h>
+#include <sys/vfs.h>
+#include <unistd.h>
+
+#include "acl-util.h"
+#include "dirent-util.h"
+#include "fd-util.h"
+#include "fileio.h"
+#include "fs-util.h"
+#include "missing_magic.h"
+#include "nspawn-def.h"
+#include "nspawn-patch-uid.h"
+#include "stat-util.h"
+#include "stdio-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "user-util.h"
+
+#if HAVE_ACL
+
+static int get_acl(int fd, const char *name, acl_type_t type, acl_t *ret) {
+ acl_t acl;
+
+ assert(fd >= 0);
+ assert(ret);
+
+ if (name) {
+ _cleanup_close_ int child_fd = -1;
+
+ child_fd = openat(fd, name, O_PATH|O_CLOEXEC|O_NOFOLLOW);
+ if (child_fd < 0)
+ return -errno;
+
+ acl = acl_get_file(FORMAT_PROC_FD_PATH(child_fd), type);
+ } else if (type == ACL_TYPE_ACCESS)
+ acl = acl_get_fd(fd);
+ else
+ acl = acl_get_file(FORMAT_PROC_FD_PATH(fd), type);
+ if (!acl)
+ return -errno;
+
+ *ret = acl;
+ return 0;
+}
+
+static int set_acl(int fd, const char *name, acl_type_t type, acl_t acl) {
+ int r;
+
+ assert(fd >= 0);
+ assert(acl);
+
+ if (name) {
+ _cleanup_close_ int child_fd = -1;
+
+ child_fd = openat(fd, name, O_PATH|O_CLOEXEC|O_NOFOLLOW);
+ if (child_fd < 0)
+ return -errno;
+
+ r = acl_set_file(FORMAT_PROC_FD_PATH(child_fd), type, acl);
+ } else if (type == ACL_TYPE_ACCESS)
+ r = acl_set_fd(fd, acl);
+ else
+ r = acl_set_file(FORMAT_PROC_FD_PATH(fd), type, acl);
+ if (r < 0)
+ return -errno;
+
+ return 0;
+}
+
+static int shift_acl(acl_t acl, uid_t shift, acl_t *ret) {
+ _cleanup_(acl_freep) acl_t copy = NULL;
+ acl_entry_t i;
+ int r;
+
+ assert(acl);
+ assert(ret);
+
+ r = acl_get_entry(acl, ACL_FIRST_ENTRY, &i);
+ if (r < 0)
+ return -errno;
+ while (r > 0) {
+ uid_t *old_uid, new_uid;
+ bool modify = false;
+ acl_tag_t tag;
+
+ if (acl_get_tag_type(i, &tag) < 0)
+ return -errno;
+
+ if (IN_SET(tag, ACL_USER, ACL_GROUP)) {
+
+ /* We don't distinguish here between uid_t and gid_t, let's make sure the compiler checks that
+ * this is actually OK */
+ assert_cc(sizeof(uid_t) == sizeof(gid_t));
+
+ old_uid = acl_get_qualifier(i);
+ if (!old_uid)
+ return -errno;
+
+ new_uid = shift | (*old_uid & UINT32_C(0xFFFF));
+ if (!uid_is_valid(new_uid))
+ return -EINVAL;
+
+ modify = new_uid != *old_uid;
+ if (modify && !copy) {
+ int n;
+
+ /* There's no copy of the ACL yet? if so, let's create one, and start the loop from the
+ * beginning, so that we copy all entries, starting from the first, this time. */
+
+ n = acl_entries(acl);
+ if (n < 0)
+ return -errno;
+
+ copy = acl_init(n);
+ if (!copy)
+ return -errno;
+
+ /* Seek back to the beginning */
+ r = acl_get_entry(acl, ACL_FIRST_ENTRY, &i);
+ if (r < 0)
+ return -errno;
+ continue;
+ }
+ }
+
+ if (copy) {
+ acl_entry_t new_entry;
+
+ if (acl_create_entry(&copy, &new_entry) < 0)
+ return -errno;
+
+ if (acl_copy_entry(new_entry, i) < 0)
+ return -errno;
+
+ if (modify)
+ if (acl_set_qualifier(new_entry, &new_uid) < 0)
+ return -errno;
+ }
+
+ r = acl_get_entry(acl, ACL_NEXT_ENTRY, &i);
+ if (r < 0)
+ return -errno;
+ }
+
+ *ret = TAKE_PTR(copy);
+
+ return !!*ret;
+}
+
+static int patch_acls(int fd, const char *name, const struct stat *st, uid_t shift) {
+ _cleanup_(acl_freep) acl_t acl = NULL, shifted = NULL;
+ bool changed = false;
+ int r;
+
+ assert(fd >= 0);
+ assert(st);
+
+ /* ACLs are not supported on symlinks, there's no point in trying */
+ if (S_ISLNK(st->st_mode))
+ return 0;
+
+ r = get_acl(fd, name, ACL_TYPE_ACCESS, &acl);
+ if (r == -EOPNOTSUPP)
+ return 0;
+ if (r < 0)
+ return r;
+
+ r = shift_acl(acl, shift, &shifted);
+ if (r < 0)
+ return r;
+ if (r > 0) {
+ r = set_acl(fd, name, ACL_TYPE_ACCESS, shifted);
+ if (r < 0)
+ return r;
+
+ changed = true;
+ }
+
+ if (S_ISDIR(st->st_mode)) {
+ acl_free(acl);
+
+ if (shifted)
+ acl_free(shifted);
+
+ acl = shifted = NULL;
+
+ r = get_acl(fd, name, ACL_TYPE_DEFAULT, &acl);
+ if (r < 0)
+ return r;
+
+ r = shift_acl(acl, shift, &shifted);
+ if (r < 0)
+ return r;
+ if (r > 0) {
+ r = set_acl(fd, name, ACL_TYPE_DEFAULT, shifted);
+ if (r < 0)
+ return r;
+
+ changed = true;
+ }
+ }
+
+ return changed;
+}
+
+#else
+
+static int patch_acls(int fd, const char *name, const struct stat *st, uid_t shift) {
+ return 0;
+}
+
+#endif
+
+static int patch_fd(int fd, const char *name, const struct stat *st, uid_t shift) {
+ uid_t new_uid;
+ gid_t new_gid;
+ bool changed = false;
+ int r;
+
+ assert(fd >= 0);
+ assert(st);
+
+ new_uid = shift | (st->st_uid & UINT32_C(0xFFFF));
+ new_gid = (gid_t) shift | (st->st_gid & UINT32_C(0xFFFF));
+
+ if (!uid_is_valid(new_uid) || !gid_is_valid(new_gid))
+ return -EINVAL;
+
+ if (st->st_uid != new_uid || st->st_gid != new_gid) {
+ if (name)
+ r = fchownat(fd, name, new_uid, new_gid, AT_SYMLINK_NOFOLLOW);
+ else
+ r = fchown(fd, new_uid, new_gid);
+ if (r < 0)
+ return -errno;
+
+ /* The Linux kernel alters the mode in some cases of chown(). Let's undo this. */
+ if (name) {
+ if (!S_ISLNK(st->st_mode))
+ r = fchmodat(fd, name, st->st_mode, 0);
+ else /* AT_SYMLINK_NOFOLLOW is not available for fchmodat() */
+ r = 0;
+ } else
+ r = fchmod(fd, st->st_mode);
+ if (r < 0)
+ return -errno;
+
+ changed = true;
+ }
+
+ r = patch_acls(fd, name, st, shift);
+ if (r < 0)
+ return r;
+
+ return r > 0 || changed;
+}
+
+/*
+ * Check if the filesystem is fully compatible with user namespaces or
+ * UID/GID patching. Some filesystems in this list can be fully mounted inside
+ * user namespaces, however their inodes may relate to host resources or only
+ * valid in the global user namespace, therefore no patching should be applied.
+ */
+static int is_fs_fully_userns_compatible(const struct statfs *sfs) {
+
+ assert(sfs);
+
+ return F_TYPE_EQUAL(sfs->f_type, BINFMTFS_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, CGROUP_SUPER_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, CGROUP2_SUPER_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, DEBUGFS_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, DEVPTS_SUPER_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, EFIVARFS_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, HUGETLBFS_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, MQUEUE_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, PROC_SUPER_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, PSTOREFS_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, SELINUX_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, SMACK_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, SECURITYFS_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, BPF_FS_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, TRACEFS_MAGIC) ||
+ F_TYPE_EQUAL(sfs->f_type, SYSFS_MAGIC);
+}
+
+static int recurse_fd(int fd, bool donate_fd, const struct stat *st, uid_t shift, bool is_toplevel) {
+ _cleanup_closedir_ DIR *d = NULL;
+ bool changed = false;
+ struct statfs sfs;
+ int r;
+
+ assert(fd >= 0);
+
+ if (fstatfs(fd, &sfs) < 0)
+ return -errno;
+
+ /* We generally want to permit crossing of mount boundaries when patching the UIDs/GIDs. However, we probably
+ * shouldn't do this for /proc and /sys if that is already mounted into place. Hence, let's stop the recursion
+ * when we hit procfs, sysfs or some other special file systems. */
+
+ r = is_fs_fully_userns_compatible(&sfs);
+ if (r < 0)
+ goto finish;
+ if (r > 0) {
+ r = 0; /* don't recurse */
+ goto finish;
+ }
+
+ /* Also, if we hit a read-only file system, then don't bother, skip the whole subtree */
+ if ((sfs.f_flags & ST_RDONLY) ||
+ access_fd(fd, W_OK) == -EROFS)
+ goto read_only;
+
+ if (S_ISDIR(st->st_mode)) {
+ if (!donate_fd) {
+ int copy;
+
+ copy = fcntl(fd, F_DUPFD_CLOEXEC, 3);
+ if (copy < 0) {
+ r = -errno;
+ goto finish;
+ }
+
+ fd = copy;
+ donate_fd = true;
+ }
+
+ d = take_fdopendir(&fd);
+ if (!d) {
+ r = -errno;
+ goto finish;
+ }
+
+ FOREACH_DIRENT_ALL(de, d, r = -errno; goto finish) {
+ struct stat fst;
+
+ if (dot_or_dot_dot(de->d_name))
+ continue;
+
+ if (fstatat(dirfd(d), de->d_name, &fst, AT_SYMLINK_NOFOLLOW) < 0) {
+ r = -errno;
+ goto finish;
+ }
+
+ if (S_ISDIR(fst.st_mode)) {
+ int subdir_fd;
+
+ subdir_fd = openat(dirfd(d), de->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC|O_NOFOLLOW|O_NOATIME);
+ if (subdir_fd < 0) {
+ r = -errno;
+ goto finish;
+
+ }
+
+ r = recurse_fd(subdir_fd, true, &fst, shift, false);
+ if (r < 0)
+ goto finish;
+ if (r > 0)
+ changed = true;
+
+ } else {
+ r = patch_fd(dirfd(d), de->d_name, &fst, shift);
+ if (r < 0)
+ goto finish;
+ if (r > 0)
+ changed = true;
+ }
+ }
+ }
+
+ /* After we descended, also patch the directory itself. It's key to do this in this order so that the top-level
+ * directory is patched as very last object in the tree, so that we can use it as quick indicator whether the
+ * tree is properly chown()ed already. */
+ r = patch_fd(d ? dirfd(d) : fd, NULL, st, shift);
+ if (r == -EROFS)
+ goto read_only;
+ if (r > 0)
+ changed = true;
+
+ r = changed;
+ goto finish;
+
+read_only:
+ if (!is_toplevel) {
+ _cleanup_free_ char *name = NULL;
+
+ /* When we hit a ready-only subtree we simply skip it, but log about it. */
+ (void) fd_get_path(fd, &name);
+ log_debug("Skipping read-only file or directory %s.", strna(name));
+ r = changed;
+ }
+
+finish:
+ if (donate_fd)
+ safe_close(fd);
+
+ return r;
+}
+
+static int fd_patch_uid_internal(int fd, bool donate_fd, uid_t shift, uid_t range) {
+ struct stat st;
+ int r;
+
+ assert(fd >= 0);
+
+ /* Recursively adjusts the UID/GIDs of all files of a directory tree. This is used to automatically fix up an
+ * OS tree to the used user namespace UID range. Note that this automatic adjustment only works for UID ranges
+ * following the concept that the upper 16bit of a UID identify the container, and the lower 16bit are the actual
+ * UID within the container. */
+
+ if ((shift & 0xFFFF) != 0) {
+ /* We only support containers where the shift starts at a 2^16 boundary */
+ r = -EOPNOTSUPP;
+ goto finish;
+ }
+
+ if (shift == UID_BUSY_BASE) {
+ r = -EINVAL;
+ goto finish;
+ }
+
+ if (range != 0x10000) {
+ /* We only support containers with 16bit UID ranges for the patching logic */
+ r = -EOPNOTSUPP;
+ goto finish;
+ }
+
+ if (fstat(fd, &st) < 0) {
+ r = -errno;
+ goto finish;
+ }
+
+ if ((uint32_t) st.st_uid >> 16 != (uint32_t) st.st_gid >> 16) {
+ /* We only support containers where the uid/gid container ID match */
+ r = -EBADE;
+ goto finish;
+ }
+
+ /* Try to detect if the range is already right. Of course, this a pretty drastic optimization, as we assume
+ * that if the top-level dir has the right upper 16bit assigned, then everything below will have too... */
+ if (((uint32_t) (st.st_uid ^ shift) >> 16) == 0)
+ return 0;
+
+ /* Before we start recursively chowning, mark the top-level dir as "busy" by chowning it to the "busy"
+ * range. Should we be interrupted in the middle of our work, we'll see it owned by this user and will start
+ * chown()ing it again, unconditionally, as the busy UID is not a valid UID we'd everpick for ourselves. */
+
+ if ((st.st_uid & UID_BUSY_MASK) != UID_BUSY_BASE) {
+ if (fchown(fd,
+ UID_BUSY_BASE | (st.st_uid & ~UID_BUSY_MASK),
+ (gid_t) UID_BUSY_BASE | (st.st_gid & ~(gid_t) UID_BUSY_MASK)) < 0) {
+ r = -errno;
+ goto finish;
+ }
+ }
+
+ return recurse_fd(fd, donate_fd, &st, shift, true);
+
+finish:
+ if (donate_fd)
+ safe_close(fd);
+
+ return r;
+}
+
+int path_patch_uid(const char *path, uid_t shift, uid_t range) {
+ int fd;
+
+ fd = open(path, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC|O_NOFOLLOW|O_NOATIME);
+ if (fd < 0)
+ return -errno;
+
+ return fd_patch_uid_internal(fd, true, shift, range);
+}
diff --git a/src/nspawn/nspawn-patch-uid.h b/src/nspawn/nspawn-patch-uid.h
new file mode 100644
index 0000000..5c7349b
--- /dev/null
+++ b/src/nspawn/nspawn-patch-uid.h
@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#pragma once
+
+#include <sys/types.h>
+
+int path_patch_uid(const char *path, uid_t shift, uid_t range);
diff --git a/src/nspawn/nspawn-register.c b/src/nspawn/nspawn-register.c
new file mode 100644
index 0000000..c78bead
--- /dev/null
+++ b/src/nspawn/nspawn-register.c
@@ -0,0 +1,361 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "sd-bus.h"
+
+#include "bus-error.h"
+#include "bus-locator.h"
+#include "bus-unit-util.h"
+#include "bus-util.h"
+#include "bus-wait-for-jobs.h"
+#include "nspawn-register.h"
+#include "special.h"
+#include "stat-util.h"
+#include "strv.h"
+#include "util.h"
+
+static int append_machine_properties(
+ sd_bus_message *m,
+ CustomMount *mounts,
+ unsigned n_mounts,
+ int kill_signal) {
+
+ unsigned j;
+ int r;
+
+ assert(m);
+
+ r = sd_bus_message_append(m, "(sv)", "DevicePolicy", "s", "closed");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ /* If you make changes here, also make sure to update systemd-nspawn@.service, to keep the device policies in
+ * sync regardless if we are run with or without the --keep-unit switch. */
+ r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 2,
+ /* Allow the container to
+ * access and create the API
+ * device nodes, so that
+ * PrivateDevices= in the
+ * container can work
+ * fine */
+ "/dev/net/tun", "rwm",
+ /* Allow the container
+ * access to ptys. However,
+ * do not permit the
+ * container to ever create
+ * these device nodes. */
+ "char-pts", "rw");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ for (j = 0; j < n_mounts; j++) {
+ CustomMount *cm = mounts + j;
+
+ if (cm->type != CUSTOM_MOUNT_BIND)
+ continue;
+
+ r = is_device_node(cm->source);
+ if (r == -ENOENT) {
+ /* The bind source might only appear as the image is put together, hence don't complain */
+ log_debug_errno(r, "Bind mount source %s not found, ignoring: %m", cm->source);
+ continue;
+ }
+ if (r < 0)
+ return log_error_errno(r, "Failed to stat %s: %m", cm->source);
+
+ if (r) {
+ r = sd_bus_message_append(m, "(sv)", "DeviceAllow", "a(ss)", 1,
+ cm->source, cm->read_only ? "r" : "rw");
+ if (r < 0)
+ return log_error_errno(r, "Failed to append message arguments: %m");
+ }
+ }
+
+ if (kill_signal != 0) {
+ r = sd_bus_message_append(m, "(sv)", "KillSignal", "i", kill_signal);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_append(m, "(sv)", "KillMode", "s", "mixed");
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ return 0;
+}
+
+static int append_controller_property(sd_bus *bus, sd_bus_message *m) {
+ const char *unique;
+ int r;
+
+ assert(bus);
+ assert(m);
+
+ r = sd_bus_get_unique_name(bus, &unique);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get unique name: %m");
+
+ r = sd_bus_message_append(m, "(sv)", "Controller", "s", unique);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ return 0;
+}
+
+int register_machine(
+ sd_bus *bus,
+ const char *machine_name,
+ pid_t pid,
+ const char *directory,
+ sd_id128_t uuid,
+ int local_ifindex,
+ const char *slice,
+ CustomMount *mounts,
+ unsigned n_mounts,
+ int kill_signal,
+ char **properties,
+ sd_bus_message *properties_message,
+ bool keep_unit,
+ const char *service) {
+
+ _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
+ int r;
+
+ assert(bus);
+
+ if (keep_unit) {
+ r = bus_call_method(
+ bus,
+ bus_machine_mgr,
+ "RegisterMachineWithNetwork",
+ &error,
+ NULL,
+ "sayssusai",
+ machine_name,
+ SD_BUS_MESSAGE_APPEND_ID128(uuid),
+ service,
+ "container",
+ (uint32_t) pid,
+ strempty(directory),
+ local_ifindex > 0 ? 1 : 0, local_ifindex);
+ } else {
+ _cleanup_(sd_bus_message_unrefp) sd_bus_message *m = NULL;
+
+ r = bus_message_new_method_call(bus, &m, bus_machine_mgr, "CreateMachineWithNetwork");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_append(
+ m,
+ "sayssusai",
+ machine_name,
+ SD_BUS_MESSAGE_APPEND_ID128(uuid),
+ service,
+ "container",
+ (uint32_t) pid,
+ strempty(directory),
+ local_ifindex > 0 ? 1 : 0, local_ifindex);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_open_container(m, 'a', "(sv)");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ if (!isempty(slice)) {
+ r = sd_bus_message_append(m, "(sv)", "Slice", "s", slice);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ r = append_controller_property(bus, m);
+ if (r < 0)
+ return r;
+
+ r = append_machine_properties(
+ m,
+ mounts,
+ n_mounts,
+ kill_signal);
+ if (r < 0)
+ return r;
+
+ if (properties_message) {
+ r = sd_bus_message_copy(m, properties_message, true);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ r = bus_append_unit_property_assignment_many(m, UNIT_SERVICE, properties);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_close_container(m);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_call(bus, m, 0, &error, NULL);
+ }
+
+ if (r < 0)
+ return log_error_errno(r, "Failed to register machine: %s", bus_error_message(&error, r));
+
+ return 0;
+}
+
+int unregister_machine(
+ sd_bus *bus,
+ const char *machine_name) {
+
+ _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
+ int r;
+
+ assert(bus);
+
+ r = bus_call_method(bus, bus_machine_mgr, "UnregisterMachine", &error, NULL, "s", machine_name);
+ if (r < 0)
+ log_debug("Failed to unregister machine: %s", bus_error_message(&error, r));
+
+ return 0;
+}
+
+int allocate_scope(
+ sd_bus *bus,
+ const char *machine_name,
+ pid_t pid,
+ const char *slice,
+ CustomMount *mounts,
+ unsigned n_mounts,
+ int kill_signal,
+ char **properties,
+ sd_bus_message *properties_message) {
+
+ _cleanup_(sd_bus_message_unrefp) sd_bus_message *m = NULL, *reply = NULL;
+ _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
+ _cleanup_(bus_wait_for_jobs_freep) BusWaitForJobs *w = NULL;
+ _cleanup_free_ char *scope = NULL;
+ const char *description, *object;
+ int r;
+
+ assert(bus);
+
+ r = bus_wait_for_jobs_new(bus, &w);
+ if (r < 0)
+ return log_error_errno(r, "Could not watch job: %m");
+
+ r = unit_name_mangle_with_suffix(machine_name, "as machine name", 0, ".scope", &scope);
+ if (r < 0)
+ return log_error_errno(r, "Failed to mangle scope name: %m");
+
+ r = bus_message_new_method_call(bus, &m, bus_systemd_mgr, "StartTransientUnit");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_message_append(m, "ss", scope, "fail");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ /* Properties */
+ r = sd_bus_message_open_container(m, 'a', "(sv)");
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ description = strjoina("Container ", machine_name);
+
+ r = sd_bus_message_append(m, "(sv)(sv)(sv)(sv)(sv)(sv)",
+ "PIDs", "au", 1, pid,
+ "Description", "s", description,
+ "Delegate", "b", 1,
+ "CollectMode", "s", "inactive-or-failed",
+ "AddRef", "b", 1,
+ "Slice", "s", isempty(slice) ? SPECIAL_MACHINE_SLICE : slice);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = append_controller_property(bus, m);
+ if (r < 0)
+ return r;
+
+ if (properties_message) {
+ r = sd_bus_message_copy(m, properties_message, true);
+ if (r < 0)
+ return bus_log_create_error(r);
+ }
+
+ r = append_machine_properties(
+ m,
+ mounts,
+ n_mounts,
+ kill_signal);
+ if (r < 0)
+ return r;
+
+ r = bus_append_unit_property_assignment_many(m, UNIT_SCOPE, properties);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_close_container(m);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ /* No auxiliary units */
+ r = sd_bus_message_append(
+ m,
+ "a(sa(sv))",
+ 0);
+ if (r < 0)
+ return bus_log_create_error(r);
+
+ r = sd_bus_call(bus, m, 0, &error, &reply);
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate scope: %s", bus_error_message(&error, r));
+
+ r = sd_bus_message_read(reply, "o", &object);
+ if (r < 0)
+ return bus_log_parse_error(r);
+
+ r = bus_wait_for_jobs_one(w, object, false, NULL);
+ if (r < 0)
+ return r;
+
+ return 0;
+}
+
+int terminate_scope(
+ sd_bus *bus,
+ const char *machine_name) {
+
+ _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
+ _cleanup_free_ char *scope = NULL;
+ int r;
+
+ r = unit_name_mangle_with_suffix(machine_name, "to terminate", 0, ".scope", &scope);
+ if (r < 0)
+ return log_error_errno(r, "Failed to mangle scope name: %m");
+
+ r = bus_call_method(bus, bus_systemd_mgr, "AbandonScope", &error, NULL, "s", scope);
+ if (r < 0) {
+ log_debug_errno(r, "Failed to abandon scope '%s', ignoring: %s", scope, bus_error_message(&error, r));
+ sd_bus_error_free(&error);
+ }
+
+ r = bus_call_method(
+ bus,
+ bus_systemd_mgr,
+ "KillUnit",
+ &error,
+ NULL,
+ "ssi",
+ scope,
+ "all",
+ (int32_t) SIGKILL);
+ if (r < 0) {
+ log_debug_errno(r, "Failed to SIGKILL scope '%s', ignoring: %s", scope, bus_error_message(&error, r));
+ sd_bus_error_free(&error);
+ }
+
+ r = bus_call_method(bus, bus_systemd_mgr, "UnrefUnit", &error, NULL, "s", scope);
+ if (r < 0)
+ log_debug_errno(r, "Failed to drop reference to scope '%s', ignoring: %s", scope, bus_error_message(&error, r));
+
+ return 0;
+}
diff --git a/src/nspawn/nspawn-register.h b/src/nspawn/nspawn-register.h
new file mode 100644
index 0000000..59fdd1b
--- /dev/null
+++ b/src/nspawn/nspawn-register.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <sys/types.h>
+
+#include "sd-id128.h"
+
+#include "nspawn-mount.h"
+
+int register_machine(sd_bus *bus, const char *machine_name, pid_t pid, const char *directory, sd_id128_t uuid, int local_ifindex, const char *slice, CustomMount *mounts, unsigned n_mounts, int kill_signal, char **properties, sd_bus_message *properties_message, bool keep_unit, const char *service);
+int unregister_machine(sd_bus *bus, const char *machine_name);
+
+int allocate_scope(sd_bus *bus, const char *machine_name, pid_t pid, const char *slice, CustomMount *mounts, unsigned n_mounts, int kill_signal, char **properties, sd_bus_message *properties_message);
+int terminate_scope(sd_bus *bus, const char *machine_name);
diff --git a/src/nspawn/nspawn-seccomp.c b/src/nspawn/nspawn-seccomp.c
new file mode 100644
index 0000000..27044fa
--- /dev/null
+++ b/src/nspawn/nspawn-seccomp.c
@@ -0,0 +1,260 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <errno.h>
+#include <linux/netlink.h>
+#include <sys/capability.h>
+#include <sys/socket.h>
+#include <sys/types.h>
+
+#if HAVE_SECCOMP
+#include <seccomp.h>
+#endif
+
+#include "alloc-util.h"
+#include "log.h"
+#include "nspawn-seccomp.h"
+#if HAVE_SECCOMP
+#include "seccomp-util.h"
+#endif
+#include "string-util.h"
+#include "strv.h"
+
+#if HAVE_SECCOMP
+
+static int add_syscall_filters(
+ scmp_filter_ctx ctx,
+ uint32_t arch,
+ uint64_t cap_list_retain,
+ char **syscall_allow_list,
+ char **syscall_deny_list) {
+
+ static const struct {
+ uint64_t capability;
+ const char* name;
+ } allow_list[] = {
+ /* Let's use set names where we can */
+ { 0, "@aio" },
+ { 0, "@basic-io" },
+ { 0, "@chown" },
+ { 0, "@default" },
+ { 0, "@file-system" },
+ { 0, "@io-event" },
+ { 0, "@ipc" },
+ { 0, "@mount" },
+ { 0, "@network-io" },
+ { 0, "@process" },
+ { 0, "@resources" },
+ { 0, "@setuid" },
+ { 0, "@signal" },
+ { 0, "@sync" },
+ { 0, "@timer" },
+
+ /* The following four are sets we optionally enable, in case the caps have been configured for it */
+ { CAP_SYS_TIME, "@clock" },
+ { CAP_SYS_MODULE, "@module" },
+ { CAP_SYS_RAWIO, "@raw-io" },
+ { CAP_IPC_LOCK, "@memlock" },
+
+ /* Plus a good set of additional syscalls which are not part of any of the groups above */
+ { 0, "brk" },
+ { 0, "capget" },
+ { 0, "capset" },
+ { 0, "copy_file_range" },
+ { 0, "fadvise64" },
+ { 0, "fadvise64_64" },
+ { 0, "flock" },
+ { 0, "get_mempolicy" },
+ { 0, "getcpu" },
+ { 0, "getpriority" },
+ { 0, "getrandom" },
+ { 0, "ioctl" },
+ { 0, "ioprio_get" },
+ { 0, "kcmp" },
+ { 0, "madvise" },
+ { 0, "mincore" },
+ { 0, "mprotect" },
+ { 0, "mremap" },
+ { 0, "name_to_handle_at" },
+ { 0, "oldolduname" },
+ { 0, "olduname" },
+ { 0, "personality" },
+ { 0, "readahead" },
+ { 0, "readdir" },
+ { 0, "remap_file_pages" },
+ { 0, "sched_get_priority_max" },
+ { 0, "sched_get_priority_min" },
+ { 0, "sched_getaffinity" },
+ { 0, "sched_getattr" },
+ { 0, "sched_getparam" },
+ { 0, "sched_getscheduler" },
+ { 0, "sched_rr_get_interval" },
+ { 0, "sched_rr_get_interval_time64" },
+ { 0, "sched_yield" },
+ { 0, "seccomp" },
+ { 0, "sendfile" },
+ { 0, "sendfile64" },
+ { 0, "setdomainname" },
+ { 0, "setfsgid" },
+ { 0, "setfsgid32" },
+ { 0, "setfsuid" },
+ { 0, "setfsuid32" },
+ { 0, "sethostname" },
+ { 0, "setpgid" },
+ { 0, "setsid" },
+ { 0, "splice" },
+ { 0, "sysinfo" },
+ { 0, "tee" },
+ { 0, "umask" },
+ { 0, "uname" },
+ { 0, "userfaultfd" },
+ { 0, "vmsplice" },
+
+ /* The following individual syscalls are added depending on specified caps */
+ { CAP_SYS_PACCT, "acct" },
+ { CAP_SYS_PTRACE, "process_vm_readv" },
+ { CAP_SYS_PTRACE, "process_vm_writev" },
+ { CAP_SYS_PTRACE, "ptrace" },
+ { CAP_SYS_BOOT, "reboot" },
+ { CAP_SYSLOG, "syslog" },
+ { CAP_SYS_TTY_CONFIG, "vhangup" },
+
+ /*
+ * The following syscalls and groups are knowingly excluded:
+ *
+ * @cpu-emulation
+ * @keyring (NB: keyring is not namespaced!)
+ * @obsolete
+ * @pkey
+ * @swap
+ *
+ * bpf
+ * fanotify_init
+ * fanotify_mark
+ * kexec_file_load
+ * kexec_load
+ * lookup_dcookie
+ * nfsservctl
+ * open_by_handle_at
+ * perf_event_open
+ * quotactl
+ */
+ };
+
+ _cleanup_strv_free_ char **added = NULL;
+ int r;
+
+ for (size_t i = 0; i < ELEMENTSOF(allow_list); i++) {
+ if (allow_list[i].capability != 0 && (cap_list_retain & (1ULL << allow_list[i].capability)) == 0)
+ continue;
+
+ r = seccomp_add_syscall_filter_item(ctx,
+ allow_list[i].name,
+ SCMP_ACT_ALLOW,
+ syscall_deny_list,
+ false,
+ &added);
+ if (r < 0)
+ return log_error_errno(r, "Failed to add syscall filter item %s: %m", allow_list[i].name);
+ }
+
+ STRV_FOREACH(p, syscall_allow_list) {
+ r = seccomp_add_syscall_filter_item(ctx, *p, SCMP_ACT_ALLOW, syscall_deny_list, true, &added);
+ if (r < 0)
+ log_warning_errno(r, "Failed to add rule for system call %s on %s, ignoring: %m",
+ *p, seccomp_arch_to_string(arch));
+ }
+
+ /* The default action is ENOSYS. Respond with EPERM to all other "known" but not allow-listed
+ * syscalls. */
+ r = seccomp_add_syscall_filter_item(ctx, "@known", SCMP_ACT_ERRNO(EPERM), added, true, NULL);
+ if (r < 0)
+ log_warning_errno(r, "Failed to add rule for @known set on %s, ignoring: %m",
+ seccomp_arch_to_string(arch));
+
+#if (SCMP_VER_MAJOR == 2 && SCMP_VER_MINOR >= 5) || SCMP_VER_MAJOR > 2
+ /* We have a large filter here, so let's turn on the binary tree mode if possible. */
+ r = seccomp_attr_set(ctx, SCMP_FLTATR_CTL_OPTIMIZE, 2);
+ if (r < 0)
+ return r;
+#endif
+
+ return 0;
+}
+
+int setup_seccomp(uint64_t cap_list_retain, char **syscall_allow_list, char **syscall_deny_list) {
+ uint32_t arch;
+ int r;
+
+ if (!is_seccomp_available()) {
+ log_debug("SECCOMP features not detected in the kernel or disabled at runtime, disabling SECCOMP filtering");
+ return 0;
+ }
+
+ SECCOMP_FOREACH_LOCAL_ARCH(arch) {
+ _cleanup_(seccomp_releasep) scmp_filter_ctx seccomp = NULL;
+
+ log_debug("Applying allow list on architecture: %s", seccomp_arch_to_string(arch));
+
+ /* We install ENOSYS as the default action, but it will only apply to syscalls which are not
+ * in the @known set, see above. */
+ r = seccomp_init_for_arch(&seccomp, arch, SCMP_ACT_ERRNO(ENOSYS));
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate seccomp object: %m");
+
+ r = add_syscall_filters(seccomp, arch, cap_list_retain, syscall_allow_list, syscall_deny_list);
+ if (r < 0)
+ return r;
+
+ r = seccomp_load(seccomp);
+ if (ERRNO_IS_SECCOMP_FATAL(r))
+ return log_error_errno(r, "Failed to install seccomp filter: %m");
+ if (r < 0)
+ log_debug_errno(r, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch));
+ }
+
+ SECCOMP_FOREACH_LOCAL_ARCH(arch) {
+ _cleanup_(seccomp_releasep) scmp_filter_ctx seccomp = NULL;
+
+ log_debug("Applying NETLINK_AUDIT mask on architecture: %s", seccomp_arch_to_string(arch));
+
+ r = seccomp_init_for_arch(&seccomp, arch, SCMP_ACT_ALLOW);
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate seccomp object: %m");
+
+ /*
+ Audit is broken in containers, much of the userspace audit hookup will fail if running inside a
+ container. We don't care and just turn off creation of audit sockets.
+
+ This will make socket(AF_NETLINK, *, NETLINK_AUDIT) fail with EAFNOSUPPORT which audit userspace uses
+ as indication that audit is disabled in the kernel.
+ */
+
+ r = seccomp_rule_add_exact(
+ seccomp,
+ SCMP_ACT_ERRNO(EAFNOSUPPORT),
+ SCMP_SYS(socket),
+ 2,
+ SCMP_A0(SCMP_CMP_EQ, AF_NETLINK),
+ SCMP_A2(SCMP_CMP_EQ, NETLINK_AUDIT));
+ if (r < 0) {
+ log_debug_errno(r, "Failed to add audit seccomp rule, ignoring: %m");
+ continue;
+ }
+
+ r = seccomp_load(seccomp);
+ if (ERRNO_IS_SECCOMP_FATAL(r))
+ return log_error_errno(r, "Failed to install seccomp audit filter: %m");
+ if (r < 0)
+ log_debug_errno(r, "Failed to install filter set for architecture %s, skipping: %m", seccomp_arch_to_string(arch));
+ }
+
+ return 0;
+}
+
+#else
+
+int setup_seccomp(uint64_t cap_list_retain, char **syscall_allow_list, char **syscall_deny_list) {
+ return 0;
+}
+
+#endif
diff --git a/src/nspawn/nspawn-seccomp.h b/src/nspawn/nspawn-seccomp.h
new file mode 100644
index 0000000..2690fba
--- /dev/null
+++ b/src/nspawn/nspawn-seccomp.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <sys/types.h>
+
+int setup_seccomp(uint64_t cap_list_retain, char **syscall_allow_list, char **syscall_deny_list);
diff --git a/src/nspawn/nspawn-settings.c b/src/nspawn/nspawn-settings.c
new file mode 100644
index 0000000..08f22be
--- /dev/null
+++ b/src/nspawn/nspawn-settings.c
@@ -0,0 +1,957 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "alloc-util.h"
+#include "cap-list.h"
+#include "conf-parser.h"
+#include "cpu-set-util.h"
+#include "hostname-util.h"
+#include "namespace-util.h"
+#include "nspawn-network.h"
+#include "nspawn-settings.h"
+#include "parse-util.h"
+#include "process-util.h"
+#include "rlimit-util.h"
+#include "socket-util.h"
+#include "string-table.h"
+#include "string-util.h"
+#include "strv.h"
+#include "user-util.h"
+#include "util.h"
+
+Settings *settings_new(void) {
+ Settings *s;
+
+ s = new(Settings, 1);
+ if (!s)
+ return NULL;
+
+ *s = (Settings) {
+ .start_mode = _START_MODE_INVALID,
+ .ephemeral = -1,
+ .personality = PERSONALITY_INVALID,
+
+ .resolv_conf = _RESOLV_CONF_MODE_INVALID,
+ .link_journal = _LINK_JOURNAL_INVALID,
+ .timezone = _TIMEZONE_MODE_INVALID,
+
+ .userns_mode = _USER_NAMESPACE_MODE_INVALID,
+ .userns_ownership = _USER_NAMESPACE_OWNERSHIP_INVALID,
+ .uid_shift = UID_INVALID,
+ .uid_range = UID_INVALID,
+
+ .no_new_privileges = -1,
+
+ .read_only = -1,
+ .volatile_mode = _VOLATILE_MODE_INVALID,
+
+ .private_network = -1,
+ .network_veth = -1,
+
+ .full_capabilities = CAPABILITY_QUINTET_NULL,
+
+ .uid = UID_INVALID,
+ .gid = GID_INVALID,
+
+ .console_mode = _CONSOLE_MODE_INVALID,
+ .console_width = UINT_MAX,
+ .console_height = UINT_MAX,
+
+ .clone_ns_flags = ULONG_MAX,
+ .use_cgns = -1,
+
+ .notify_ready = -1,
+ .suppress_sync = -1,
+ };
+
+ return s;
+}
+
+int settings_load(FILE *f, const char *path, Settings **ret) {
+ _cleanup_(settings_freep) Settings *s = NULL;
+ int r;
+
+ assert(path);
+ assert(ret);
+
+ s = settings_new();
+ if (!s)
+ return -ENOMEM;
+
+ r = config_parse(NULL, path, f,
+ "Exec\0"
+ "Network\0"
+ "Files\0",
+ config_item_perf_lookup, nspawn_gperf_lookup,
+ CONFIG_PARSE_WARN,
+ s, NULL);
+ if (r < 0)
+ return r;
+
+ /* Make sure that if userns_mode is set, userns_chown is set to something appropriate, and vice versa. Either
+ * both fields shall be initialized or neither. */
+ if (s->userns_mode >= 0 && s->userns_ownership < 0)
+ s->userns_ownership = s->userns_mode == USER_NAMESPACE_PICK ? USER_NAMESPACE_OWNERSHIP_CHOWN : USER_NAMESPACE_OWNERSHIP_OFF;
+ if (s->userns_ownership >= 0 && s->userns_mode < 0)
+ s->userns_mode = USER_NAMESPACE_NO;
+
+ *ret = TAKE_PTR(s);
+ return 0;
+}
+
+static void free_oci_hooks(OciHook *h, size_t n) {
+ size_t i;
+
+ assert(h || n == 0);
+
+ for (i = 0; i < n; i++) {
+ free(h[i].path);
+ strv_free(h[i].args);
+ strv_free(h[i].env);
+ }
+
+ free(h);
+}
+
+void device_node_array_free(DeviceNode *node, size_t n) {
+ size_t i;
+
+ assert(node || n == 0);
+
+ for (i = 0; i < n; i++)
+ free(node[i].path);
+
+ free(node);
+}
+
+Settings* settings_free(Settings *s) {
+ if (!s)
+ return NULL;
+
+ strv_free(s->parameters);
+ strv_free(s->environment);
+ free(s->user);
+ free(s->pivot_root_new);
+ free(s->pivot_root_old);
+ free(s->working_directory);
+ strv_free(s->syscall_allow_list);
+ strv_free(s->syscall_deny_list);
+ rlimit_free_all(s->rlimit);
+ free(s->hostname);
+ cpu_set_reset(&s->cpu_set);
+ strv_free(s->bind_user);
+
+ strv_free(s->network_interfaces);
+ strv_free(s->network_macvlan);
+ strv_free(s->network_ipvlan);
+ strv_free(s->network_veth_extra);
+ free(s->network_bridge);
+ free(s->network_zone);
+ expose_port_free_all(s->expose_ports);
+
+ custom_mount_free_all(s->custom_mounts, s->n_custom_mounts);
+
+ free(s->bundle);
+ free(s->root);
+
+ free_oci_hooks(s->oci_hooks_prestart, s->n_oci_hooks_prestart);
+ free_oci_hooks(s->oci_hooks_poststart, s->n_oci_hooks_poststart);
+ free_oci_hooks(s->oci_hooks_poststop, s->n_oci_hooks_poststop);
+
+ free(s->slice);
+ sd_bus_message_unref(s->properties);
+
+ free(s->supplementary_gids);
+ device_node_array_free(s->extra_nodes, s->n_extra_nodes);
+ free(s->network_namespace_path);
+
+ strv_free(s->sysctl);
+
+#if HAVE_SECCOMP
+ seccomp_release(s->seccomp);
+#endif
+
+ return mfree(s);
+}
+
+bool settings_private_network(Settings *s) {
+ assert(s);
+
+ /* Determines whether we shall open up our own private network */
+
+ return
+ s->private_network > 0 ||
+ s->network_veth > 0 ||
+ s->network_bridge ||
+ s->network_zone ||
+ s->network_interfaces ||
+ s->network_macvlan ||
+ s->network_ipvlan ||
+ s->network_veth_extra;
+}
+
+bool settings_network_veth(Settings *s) {
+ assert(s);
+
+ return
+ s->network_veth > 0 ||
+ s->network_bridge ||
+ s->network_zone;
+}
+
+bool settings_network_configured(Settings *s) {
+ assert(s);
+
+ /* Determines whether any network configuration setting was used. (i.e. in contrast to
+ * settings_private_network() above this might also indicate if private networking was explicitly
+ * turned off.) */
+
+ return
+ s->private_network >= 0 ||
+ s->network_veth >= 0 ||
+ s->network_bridge ||
+ s->network_zone ||
+ s->network_interfaces ||
+ s->network_macvlan ||
+ s->network_ipvlan ||
+ s->network_veth_extra ||
+ s->network_namespace_path;
+}
+
+int settings_allocate_properties(Settings *s) {
+ _cleanup_(sd_bus_unrefp) sd_bus *bus = NULL;
+ int r;
+
+ assert(s);
+
+ if (s->properties)
+ return 0;
+
+ r = sd_bus_default_system(&bus);
+ if (r < 0)
+ return r;
+
+ r = sd_bus_message_new(bus, &s->properties, SD_BUS_MESSAGE_METHOD_CALL);
+ if (r < 0)
+ return r;
+
+ return 0;
+}
+
+DEFINE_CONFIG_PARSE_ENUM(config_parse_volatile_mode, volatile_mode, VolatileMode, "Failed to parse volatile mode");
+
+int config_parse_expose_port(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *s = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = expose_port_parse(&s->expose_ports, rvalue);
+ if (r == -EEXIST)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Duplicate port specification, ignoring: %s", rvalue);
+ else if (r < 0)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse host port %s: %m", rvalue);
+
+ return 0;
+}
+
+int config_parse_capability(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ uint64_t u = 0, *result = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ for (;;) {
+ _cleanup_free_ char *word = NULL;
+
+ r = extract_first_word(&rvalue, &word, NULL, 0);
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to extract capability string, ignoring: %s", rvalue);
+ return 0;
+ }
+ if (r == 0)
+ break;
+
+ if (streq(word, "all"))
+ u = UINT64_MAX;
+ else {
+ r = capability_from_name(word);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse capability, ignoring: %s", word);
+ continue;
+ }
+
+ u |= UINT64_C(1) << r;
+ }
+ }
+
+ *result |= u;
+ return 0;
+}
+
+int config_parse_pivot_root(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = pivot_root_parse(&settings->pivot_root_new, &settings->pivot_root_old, rvalue);
+ if (r < 0)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Invalid pivot root mount specification %s: %m", rvalue);
+
+ return 0;
+}
+
+int config_parse_bind(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = bind_mount_parse(&settings->custom_mounts, &settings->n_custom_mounts, rvalue, ltype);
+ if (r < 0)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Invalid bind mount specification %s: %m", rvalue);
+
+ return 0;
+}
+
+int config_parse_tmpfs(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = tmpfs_mount_parse(&settings->custom_mounts, &settings->n_custom_mounts, rvalue);
+ if (r < 0)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Invalid temporary file system specification %s: %m", rvalue);
+
+ return 0;
+}
+
+int config_parse_inaccessible(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = inaccessible_mount_parse(&settings->custom_mounts, &settings->n_custom_mounts, rvalue);
+ if (r < 0)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Invalid inaccessible file system specification %s: %m", rvalue);
+
+ return 0;
+}
+
+int config_parse_overlay(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = overlay_mount_parse(&settings->custom_mounts, &settings->n_custom_mounts, rvalue, ltype);
+ if (r < 0)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Invalid overlay file system specification %s, ignoring: %m", rvalue);
+
+ return 0;
+}
+
+int config_parse_veth_extra(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = veth_extra_parse(&settings->network_veth_extra, rvalue);
+ if (r < 0)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Invalid extra virtual Ethernet link specification %s: %m", rvalue);
+
+ return 0;
+}
+
+int config_parse_network_zone(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ _cleanup_free_ char *j = NULL;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ j = strjoin("vz-", rvalue);
+ if (!ifname_valid(j)) {
+ log_syntax(unit, LOG_WARNING, filename, line, 0, "Invalid network zone name, ignoring: %s", rvalue);
+ return 0;
+ }
+
+ return free_and_replace(settings->network_zone, j);
+}
+
+int config_parse_boot(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = parse_boolean(rvalue);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse Boot= parameter %s, ignoring: %m", rvalue);
+ return 0;
+ }
+
+ if (r) {
+ if (settings->start_mode == START_PID2)
+ goto conflict;
+
+ settings->start_mode = START_BOOT;
+ } else {
+ if (settings->start_mode == START_BOOT)
+ goto conflict;
+
+ if (settings->start_mode < 0)
+ settings->start_mode = START_PID1;
+ }
+
+ return 0;
+
+conflict:
+ log_syntax(unit, LOG_WARNING, filename, line, 0, "Conflicting Boot= or ProcessTwo= setting found. Ignoring.");
+ return 0;
+}
+
+int config_parse_pid2(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = parse_boolean(rvalue);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse ProcessTwo= parameter %s, ignoring: %m", rvalue);
+ return 0;
+ }
+
+ if (r) {
+ if (settings->start_mode == START_BOOT)
+ goto conflict;
+
+ settings->start_mode = START_PID2;
+ } else {
+ if (settings->start_mode == START_PID2)
+ goto conflict;
+
+ if (settings->start_mode < 0)
+ settings->start_mode = START_PID1;
+ }
+
+ return 0;
+
+conflict:
+ log_syntax(unit, LOG_WARNING, filename, line, 0, "Conflicting Boot= or ProcessTwo= setting found. Ignoring.");
+ return 0;
+}
+
+int config_parse_private_users(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ r = parse_boolean(rvalue);
+ if (r == 0) {
+ /* no: User namespacing off */
+ settings->userns_mode = USER_NAMESPACE_NO;
+ settings->uid_shift = UID_INVALID;
+ settings->uid_range = UINT32_C(0x10000);
+ } else if (r > 0) {
+ /* yes: User namespacing on, UID range is read from root dir */
+ settings->userns_mode = USER_NAMESPACE_FIXED;
+ settings->uid_shift = UID_INVALID;
+ settings->uid_range = UINT32_C(0x10000);
+ } else if (streq(rvalue, "pick")) {
+ /* pick: User namespacing on, UID range is picked randomly */
+ settings->userns_mode = USER_NAMESPACE_PICK;
+ settings->uid_shift = UID_INVALID;
+ settings->uid_range = UINT32_C(0x10000);
+ } else if (streq(rvalue, "identity")) {
+ /* identity: User namespacing on, UID range is 0:65536 */
+ settings->userns_mode = USER_NAMESPACE_FIXED;
+ settings->uid_shift = 0;
+ settings->uid_range = UINT32_C(0x10000);
+ } else {
+ const char *range, *shift;
+ uid_t sh, rn;
+
+ /* anything else: User namespacing on, UID range is explicitly configured */
+
+ range = strchr(rvalue, ':');
+ if (range) {
+ shift = strndupa_safe(rvalue, range - rvalue);
+ range++;
+
+ r = safe_atou32(range, &rn);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "UID/GID range invalid, ignoring: %s", range);
+ return 0;
+ }
+ } else {
+ shift = rvalue;
+ rn = UINT32_C(0x10000);
+ }
+
+ r = parse_uid(shift, &sh);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "UID/GID shift invalid, ignoring: %s", range);
+ return 0;
+ }
+
+ if (!userns_shift_range_valid(sh, rn)) {
+ log_syntax(unit, LOG_WARNING, filename, line, 0, "UID/GID shift and range combination invalid, ignoring: %s", range);
+ return 0;
+ }
+
+ settings->userns_mode = USER_NAMESPACE_FIXED;
+ settings->uid_shift = sh;
+ settings->uid_range = rn;
+ }
+
+ return 0;
+}
+
+int config_parse_syscall_filter(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = data;
+ bool negative;
+ const char *items;
+ int r;
+
+ assert(filename);
+ assert(lvalue);
+ assert(rvalue);
+
+ negative = rvalue[0] == '~';
+ items = negative ? rvalue + 1 : rvalue;
+
+ for (;;) {
+ _cleanup_free_ char *word = NULL;
+
+ r = extract_first_word(&items, &word, NULL, 0);
+ if (r == 0)
+ return 0;
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r,
+ "Failed to parse SystemCallFilter= parameter %s, ignoring: %m", rvalue);
+ return 0;
+ }
+
+ if (negative)
+ r = strv_extend(&settings->syscall_deny_list, word);
+ else
+ r = strv_extend(&settings->syscall_allow_list, word);
+ if (r < 0)
+ return log_oom();
+ }
+}
+
+int config_parse_oom_score_adjust(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = ASSERT_PTR(data);
+ int oa, r;
+
+ assert(rvalue);
+
+ if (isempty(rvalue)) {
+ settings->oom_score_adjust_set = false;
+ return 0;
+ }
+
+ r = parse_oom_score_adjust(rvalue, &oa);
+ if (r == -ERANGE) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "OOM score adjust value out of range, ignoring: %s", rvalue);
+ return 0;
+ }
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse the OOM score adjust value, ignoring: %s", rvalue);
+ return 0;
+ }
+
+ settings->oom_score_adjust = oa;
+ settings->oom_score_adjust_set = true;
+
+ return 0;
+}
+
+int config_parse_cpu_affinity(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = ASSERT_PTR(data);
+
+ assert(rvalue);
+
+ return parse_cpu_set_extend(rvalue, &settings->cpu_set, true, unit, filename, line, lvalue);
+}
+
+DEFINE_CONFIG_PARSE_ENUM(config_parse_resolv_conf, resolv_conf_mode, ResolvConfMode, "Failed to parse resolv.conf mode");
+
+static const char *const resolv_conf_mode_table[_RESOLV_CONF_MODE_MAX] = {
+ [RESOLV_CONF_OFF] = "off",
+ [RESOLV_CONF_COPY_HOST] = "copy-host",
+ [RESOLV_CONF_COPY_STATIC] = "copy-static",
+ [RESOLV_CONF_COPY_UPLINK] = "copy-uplink",
+ [RESOLV_CONF_COPY_STUB] = "copy-stub",
+ [RESOLV_CONF_REPLACE_HOST] = "replace-host",
+ [RESOLV_CONF_REPLACE_STATIC] = "replace-static",
+ [RESOLV_CONF_REPLACE_UPLINK] = "replace-uplink",
+ [RESOLV_CONF_REPLACE_STUB] = "replace-stub",
+ [RESOLV_CONF_BIND_HOST] = "bind-host",
+ [RESOLV_CONF_BIND_STATIC] = "bind-static",
+ [RESOLV_CONF_BIND_UPLINK] = "bind-uplink",
+ [RESOLV_CONF_BIND_STUB] = "bind-stub",
+ [RESOLV_CONF_DELETE] = "delete",
+ [RESOLV_CONF_AUTO] = "auto",
+};
+
+DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(resolv_conf_mode, ResolvConfMode, RESOLV_CONF_AUTO);
+
+int parse_link_journal(const char *s, LinkJournal *ret_mode, bool *ret_try) {
+ int r;
+
+ assert(s);
+ assert(ret_mode);
+ assert(ret_try);
+
+ if (streq(s, "auto")) {
+ *ret_mode = LINK_AUTO;
+ *ret_try = false;
+ } else if (streq(s, "guest")) {
+ *ret_mode = LINK_GUEST;
+ *ret_try = false;
+ } else if (streq(s, "host")) {
+ *ret_mode = LINK_HOST;
+ *ret_try = false;
+ } else if (streq(s, "try-guest")) {
+ *ret_mode = LINK_GUEST;
+ *ret_try = true;
+ } else if (streq(s, "try-host")) {
+ *ret_mode = LINK_HOST;
+ *ret_try = true;
+ } else {
+ /* Also support boolean values, to make things less confusing. */
+ r = parse_boolean(s);
+ if (r < 0)
+ return r;
+
+ /* Let's consider "true" to be equivalent to "auto". */
+ *ret_mode = r ? LINK_AUTO : LINK_NO;
+ *ret_try = false;
+ }
+
+ return 0;
+}
+
+int config_parse_link_journal(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ Settings *settings = ASSERT_PTR(data);
+ int r;
+
+ assert(rvalue);
+
+ r = parse_link_journal(rvalue, &settings->link_journal, &settings->link_journal_try);
+ if (r < 0)
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse link journal mode, ignoring: %s", rvalue);
+
+ return 0;
+}
+
+DEFINE_CONFIG_PARSE_ENUM(config_parse_timezone, timezone_mode, TimezoneMode, "Failed to parse timezone mode");
+
+static const char *const timezone_mode_table[_TIMEZONE_MODE_MAX] = {
+ [TIMEZONE_OFF] = "off",
+ [TIMEZONE_COPY] = "copy",
+ [TIMEZONE_BIND] = "bind",
+ [TIMEZONE_SYMLINK] = "symlink",
+ [TIMEZONE_DELETE] = "delete",
+ [TIMEZONE_AUTO] = "auto",
+};
+
+DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(timezone_mode, TimezoneMode, TIMEZONE_AUTO);
+
+DEFINE_CONFIG_PARSE_ENUM(config_parse_userns_ownership, user_namespace_ownership, UserNamespaceOwnership, "Failed to parse user namespace ownership mode");
+
+static const char *const user_namespace_ownership_table[_USER_NAMESPACE_OWNERSHIP_MAX] = {
+ [USER_NAMESPACE_OWNERSHIP_OFF] = "off",
+ [USER_NAMESPACE_OWNERSHIP_CHOWN] = "chown",
+ [USER_NAMESPACE_OWNERSHIP_MAP] = "map",
+ [USER_NAMESPACE_OWNERSHIP_AUTO] = "auto",
+};
+
+DEFINE_STRING_TABLE_LOOKUP(user_namespace_ownership, UserNamespaceOwnership);
+
+int config_parse_userns_chown(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ UserNamespaceOwnership *ownership = ASSERT_PTR(data);
+ int r;
+
+ assert(rvalue);
+
+ /* Compatibility support for UserNamespaceChown=, whose job has been taken over by UserNamespaceOwnership= */
+
+ r = parse_boolean(rvalue);
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse user namespace ownership mode, ignoring: %s", rvalue);
+ return 0;
+ }
+
+ *ownership = r ? USER_NAMESPACE_OWNERSHIP_CHOWN : USER_NAMESPACE_OWNERSHIP_OFF;
+ return 0;
+}
+
+int config_parse_bind_user(
+ const char *unit,
+ const char *filename,
+ unsigned line,
+ const char *section,
+ unsigned section_line,
+ const char *lvalue,
+ int ltype,
+ const char *rvalue,
+ void *data,
+ void *userdata) {
+
+ char ***bind_user = ASSERT_PTR(data);
+ int r;
+
+ assert(rvalue);
+
+ if (isempty(rvalue)) {
+ *bind_user = strv_free(*bind_user);
+ return 0;
+ }
+
+ for (const char* p = rvalue;;) {
+ _cleanup_free_ char *word = NULL;
+
+ r = extract_first_word(&p, &word, NULL, 0);
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r < 0) {
+ log_syntax(unit, LOG_WARNING, filename, line, r, "Failed to parse BindUser= list, ignoring: %s", rvalue);
+ return 0;
+ }
+ if (r == 0)
+ break;
+
+ if (!valid_user_group_name(word, 0)) {
+ log_syntax(unit, LOG_WARNING, filename, line, 0, "User name '%s' not valid, ignoring.", word);
+ return 0;
+ }
+
+ if (strv_consume(bind_user, TAKE_PTR(word)) < 0)
+ return log_oom();
+ }
+
+ return 0;
+}
diff --git a/src/nspawn/nspawn-settings.h b/src/nspawn/nspawn-settings.h
new file mode 100644
index 0000000..004b663
--- /dev/null
+++ b/src/nspawn/nspawn-settings.h
@@ -0,0 +1,287 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <sched.h>
+#include <stdio.h>
+
+#if HAVE_SECCOMP
+#include <seccomp.h>
+#endif
+
+#include "sd-bus.h"
+#include "sd-id128.h"
+
+#include "capability-util.h"
+#include "conf-parser.h"
+#include "cpu-set-util.h"
+#include "macro.h"
+#include "missing_resource.h"
+#include "nspawn-expose-ports.h"
+#include "nspawn-mount.h"
+#include "time-util.h"
+
+typedef enum StartMode {
+ START_PID1, /* Run parameters as command line as process 1 */
+ START_PID2, /* Use stub init process as PID 1, run parameters as command line as process 2 */
+ START_BOOT, /* Search for init system, pass arguments as parameters */
+ _START_MODE_MAX,
+ _START_MODE_INVALID = -EINVAL,
+} StartMode;
+
+typedef enum UserNamespaceMode {
+ USER_NAMESPACE_NO,
+ USER_NAMESPACE_FIXED,
+ USER_NAMESPACE_PICK,
+ _USER_NAMESPACE_MODE_MAX,
+ _USER_NAMESPACE_MODE_INVALID = -EINVAL,
+} UserNamespaceMode;
+
+typedef enum UserNamespaceOwnership {
+ USER_NAMESPACE_OWNERSHIP_OFF,
+ USER_NAMESPACE_OWNERSHIP_CHOWN,
+ USER_NAMESPACE_OWNERSHIP_MAP,
+ USER_NAMESPACE_OWNERSHIP_AUTO,
+ _USER_NAMESPACE_OWNERSHIP_MAX,
+ _USER_NAMESPACE_OWNERSHIP_INVALID = -1,
+} UserNamespaceOwnership;
+
+typedef enum ResolvConfMode {
+ RESOLV_CONF_OFF,
+ RESOLV_CONF_COPY_HOST, /* /etc/resolv.conf */
+ RESOLV_CONF_COPY_STATIC, /* /usr/lib/systemd/resolv.conf */
+ RESOLV_CONF_COPY_UPLINK, /* /run/systemd/resolve/resolv.conf */
+ RESOLV_CONF_COPY_STUB, /* /run/systemd/resolve/stub-resolv.conf */
+ RESOLV_CONF_REPLACE_HOST,
+ RESOLV_CONF_REPLACE_STATIC,
+ RESOLV_CONF_REPLACE_UPLINK,
+ RESOLV_CONF_REPLACE_STUB,
+ RESOLV_CONF_BIND_HOST,
+ RESOLV_CONF_BIND_STATIC,
+ RESOLV_CONF_BIND_UPLINK,
+ RESOLV_CONF_BIND_STUB,
+ RESOLV_CONF_DELETE,
+ RESOLV_CONF_AUTO,
+ _RESOLV_CONF_MODE_MAX,
+ _RESOLV_CONF_MODE_INVALID = -EINVAL,
+} ResolvConfMode;
+
+typedef enum LinkJournal {
+ LINK_NO,
+ LINK_AUTO,
+ LINK_HOST,
+ LINK_GUEST,
+ _LINK_JOURNAL_MAX,
+ _LINK_JOURNAL_INVALID = -EINVAL,
+} LinkJournal;
+
+typedef enum TimezoneMode {
+ TIMEZONE_OFF,
+ TIMEZONE_COPY,
+ TIMEZONE_BIND,
+ TIMEZONE_SYMLINK,
+ TIMEZONE_DELETE,
+ TIMEZONE_AUTO,
+ _TIMEZONE_MODE_MAX,
+ _TIMEZONE_MODE_INVALID = -EINVAL,
+} TimezoneMode;
+
+typedef enum ConsoleMode {
+ CONSOLE_INTERACTIVE,
+ CONSOLE_READ_ONLY,
+ CONSOLE_PASSIVE,
+ CONSOLE_PIPE,
+ _CONSOLE_MODE_MAX,
+ _CONSOLE_MODE_INVALID = -EINVAL,
+} ConsoleMode;
+
+typedef enum SettingsMask {
+ SETTING_START_MODE = UINT64_C(1) << 0,
+ SETTING_ENVIRONMENT = UINT64_C(1) << 1,
+ SETTING_USER = UINT64_C(1) << 2,
+ SETTING_CAPABILITY = UINT64_C(1) << 3,
+ SETTING_KILL_SIGNAL = UINT64_C(1) << 4,
+ SETTING_PERSONALITY = UINT64_C(1) << 5,
+ SETTING_MACHINE_ID = UINT64_C(1) << 6,
+ SETTING_NETWORK = UINT64_C(1) << 7,
+ SETTING_EXPOSE_PORTS = UINT64_C(1) << 8,
+ SETTING_READ_ONLY = UINT64_C(1) << 9,
+ SETTING_VOLATILE_MODE = UINT64_C(1) << 10,
+ SETTING_CUSTOM_MOUNTS = UINT64_C(1) << 11,
+ SETTING_WORKING_DIRECTORY = UINT64_C(1) << 12,
+ SETTING_USERNS = UINT64_C(1) << 13,
+ SETTING_NOTIFY_READY = UINT64_C(1) << 14,
+ SETTING_PIVOT_ROOT = UINT64_C(1) << 15,
+ SETTING_SYSCALL_FILTER = UINT64_C(1) << 16,
+ SETTING_HOSTNAME = UINT64_C(1) << 17,
+ SETTING_NO_NEW_PRIVILEGES = UINT64_C(1) << 18,
+ SETTING_OOM_SCORE_ADJUST = UINT64_C(1) << 19,
+ SETTING_CPU_AFFINITY = UINT64_C(1) << 20,
+ SETTING_RESOLV_CONF = UINT64_C(1) << 21,
+ SETTING_LINK_JOURNAL = UINT64_C(1) << 22,
+ SETTING_TIMEZONE = UINT64_C(1) << 23,
+ SETTING_EPHEMERAL = UINT64_C(1) << 24,
+ SETTING_SLICE = UINT64_C(1) << 25,
+ SETTING_DIRECTORY = UINT64_C(1) << 26,
+ SETTING_USE_CGNS = UINT64_C(1) << 27,
+ SETTING_CLONE_NS_FLAGS = UINT64_C(1) << 28,
+ SETTING_CONSOLE_MODE = UINT64_C(1) << 29,
+ SETTING_CREDENTIALS = UINT64_C(1) << 30,
+ SETTING_BIND_USER = UINT64_C(1) << 31,
+ SETTING_SUPPRESS_SYNC = UINT64_C(1) << 32,
+ SETTING_RLIMIT_FIRST = UINT64_C(1) << 33, /* we define one bit per resource limit here */
+ SETTING_RLIMIT_LAST = UINT64_C(1) << (33 + _RLIMIT_MAX - 1),
+ _SETTINGS_MASK_ALL = (UINT64_C(1) << (33 + _RLIMIT_MAX)) -1,
+ _SETTING_FORCE_ENUM_WIDTH = UINT64_MAX
+} SettingsMask;
+
+/* We want to use SETTING_RLIMIT_FIRST in shifts, so make sure it is really 64 bits
+ * when used in expressions. */
+#define SETTING_RLIMIT_FIRST ((uint64_t) SETTING_RLIMIT_FIRST)
+#define SETTING_RLIMIT_LAST ((uint64_t) SETTING_RLIMIT_LAST)
+
+assert_cc(sizeof(SettingsMask) == 8);
+assert_cc(sizeof(SETTING_RLIMIT_FIRST) == 8);
+assert_cc(sizeof(SETTING_RLIMIT_LAST) == 8);
+
+typedef struct DeviceNode {
+ char *path;
+ unsigned major;
+ unsigned minor;
+ mode_t mode;
+ uid_t uid;
+ gid_t gid;
+} DeviceNode;
+
+typedef struct OciHook {
+ char *path;
+ char **args;
+ char **env;
+ usec_t timeout;
+} OciHook;
+
+typedef struct Settings {
+ /* [Exec] */
+ StartMode start_mode;
+ int ephemeral;
+ char **parameters;
+ char **environment;
+ char *user;
+ uint64_t capability;
+ uint64_t drop_capability;
+ uint64_t ambient_capability;
+ int kill_signal;
+ unsigned long personality;
+ sd_id128_t machine_id;
+ char *working_directory;
+ char *pivot_root_new;
+ char *pivot_root_old;
+ UserNamespaceMode userns_mode;
+ uid_t uid_shift, uid_range;
+ int notify_ready;
+ char **syscall_allow_list;
+ char **syscall_deny_list;
+ struct rlimit *rlimit[_RLIMIT_MAX];
+ char *hostname;
+ int no_new_privileges;
+ int oom_score_adjust;
+ bool oom_score_adjust_set;
+ CPUSet cpu_set;
+ ResolvConfMode resolv_conf;
+ LinkJournal link_journal;
+ bool link_journal_try;
+ TimezoneMode timezone;
+ int suppress_sync;
+
+ /* [Files] */
+ int read_only;
+ VolatileMode volatile_mode;
+ CustomMount *custom_mounts;
+ size_t n_custom_mounts;
+ UserNamespaceOwnership userns_ownership;
+ char **bind_user;
+
+ /* [Network] */
+ int private_network;
+ int network_veth;
+ char *network_bridge;
+ char *network_zone;
+ char **network_interfaces;
+ char **network_macvlan;
+ char **network_ipvlan;
+ char **network_veth_extra;
+ ExposePort *expose_ports;
+
+ /* Additional fields, that are specific to OCI runtime case */
+ char *bundle;
+ char *root;
+ OciHook *oci_hooks_prestart, *oci_hooks_poststart, *oci_hooks_poststop;
+ size_t n_oci_hooks_prestart, n_oci_hooks_poststart, n_oci_hooks_poststop;
+ char *slice;
+ sd_bus_message *properties;
+ CapabilityQuintet full_capabilities;
+ uid_t uid;
+ gid_t gid;
+ gid_t *supplementary_gids;
+ size_t n_supplementary_gids;
+ unsigned console_width, console_height;
+ ConsoleMode console_mode;
+ DeviceNode *extra_nodes;
+ size_t n_extra_nodes;
+ unsigned long clone_ns_flags;
+ char *network_namespace_path;
+ int use_cgns;
+ char **sysctl;
+#if HAVE_SECCOMP
+ scmp_filter_ctx seccomp;
+#endif
+} Settings;
+
+Settings *settings_new(void);
+int settings_load(FILE *f, const char *path, Settings **ret);
+Settings* settings_free(Settings *s);
+
+bool settings_network_veth(Settings *s);
+bool settings_private_network(Settings *s);
+bool settings_network_configured(Settings *s);
+
+int settings_allocate_properties(Settings *s);
+
+DEFINE_TRIVIAL_CLEANUP_FUNC(Settings*, settings_free);
+
+const struct ConfigPerfItem* nspawn_gperf_lookup(const char *key, GPERF_LEN_TYPE length);
+
+CONFIG_PARSER_PROTOTYPE(config_parse_capability);
+CONFIG_PARSER_PROTOTYPE(config_parse_expose_port);
+CONFIG_PARSER_PROTOTYPE(config_parse_volatile_mode);
+CONFIG_PARSER_PROTOTYPE(config_parse_pivot_root);
+CONFIG_PARSER_PROTOTYPE(config_parse_bind);
+CONFIG_PARSER_PROTOTYPE(config_parse_tmpfs);
+CONFIG_PARSER_PROTOTYPE(config_parse_overlay);
+CONFIG_PARSER_PROTOTYPE(config_parse_inaccessible);
+CONFIG_PARSER_PROTOTYPE(config_parse_veth_extra);
+CONFIG_PARSER_PROTOTYPE(config_parse_network_zone);
+CONFIG_PARSER_PROTOTYPE(config_parse_boot);
+CONFIG_PARSER_PROTOTYPE(config_parse_pid2);
+CONFIG_PARSER_PROTOTYPE(config_parse_private_users);
+CONFIG_PARSER_PROTOTYPE(config_parse_syscall_filter);
+CONFIG_PARSER_PROTOTYPE(config_parse_oom_score_adjust);
+CONFIG_PARSER_PROTOTYPE(config_parse_cpu_affinity);
+CONFIG_PARSER_PROTOTYPE(config_parse_resolv_conf);
+CONFIG_PARSER_PROTOTYPE(config_parse_link_journal);
+CONFIG_PARSER_PROTOTYPE(config_parse_timezone);
+CONFIG_PARSER_PROTOTYPE(config_parse_userns_chown);
+CONFIG_PARSER_PROTOTYPE(config_parse_userns_ownership);
+CONFIG_PARSER_PROTOTYPE(config_parse_bind_user);
+
+const char *resolv_conf_mode_to_string(ResolvConfMode a) _const_;
+ResolvConfMode resolv_conf_mode_from_string(const char *s) _pure_;
+
+const char *timezone_mode_to_string(TimezoneMode a) _const_;
+TimezoneMode timezone_mode_from_string(const char *s) _pure_;
+
+const char *user_namespace_ownership_to_string(UserNamespaceOwnership a) _const_;
+UserNamespaceOwnership user_namespace_ownership_from_string(const char *s) _pure_;
+
+int parse_link_journal(const char *s, LinkJournal *ret_mode, bool *ret_try);
+
+void device_node_array_free(DeviceNode *node, size_t n);
diff --git a/src/nspawn/nspawn-setuid.c b/src/nspawn/nspawn-setuid.c
new file mode 100644
index 0000000..34758d8
--- /dev/null
+++ b/src/nspawn/nspawn-setuid.c
@@ -0,0 +1,245 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <fcntl.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "alloc-util.h"
+#include "def.h"
+#include "errno.h"
+#include "fd-util.h"
+#include "fileio.h"
+#include "mkdir.h"
+#include "nspawn-setuid.h"
+#include "process-util.h"
+#include "rlimit-util.h"
+#include "signal-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "user-util.h"
+#include "util.h"
+
+static int spawn_getent(const char *database, const char *key, pid_t *rpid) {
+ int pipe_fds[2], r;
+ pid_t pid;
+
+ assert(database);
+ assert(key);
+ assert(rpid);
+
+ if (pipe2(pipe_fds, O_CLOEXEC) < 0)
+ return log_error_errno(errno, "Failed to allocate pipe: %m");
+
+ r = safe_fork("(getent)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_LOG, &pid);
+ if (r < 0) {
+ safe_close_pair(pipe_fds);
+ return r;
+ }
+ if (r == 0) {
+ char *empty_env = NULL;
+
+ pipe_fds[0] = safe_close(pipe_fds[0]);
+
+ if (rearrange_stdio(-1, TAKE_FD(pipe_fds[1]), -1) < 0)
+ _exit(EXIT_FAILURE);
+
+ (void) close_all_fds(NULL, 0);
+
+ (void) rlimit_nofile_safe();
+
+ execle("/usr/bin/getent", "getent", database, key, NULL, &empty_env);
+ execle("/bin/getent", "getent", database, key, NULL, &empty_env);
+ _exit(EXIT_FAILURE);
+ }
+
+ pipe_fds[1] = safe_close(pipe_fds[1]);
+
+ *rpid = pid;
+
+ return pipe_fds[0];
+}
+
+int change_uid_gid_raw(
+ uid_t uid,
+ gid_t gid,
+ const gid_t *supplementary_gids,
+ size_t n_supplementary_gids,
+ bool chown_stdio) {
+
+ if (!uid_is_valid(uid))
+ uid = 0;
+ if (!gid_is_valid(gid))
+ gid = 0;
+
+ if (chown_stdio) {
+ (void) fchown(STDIN_FILENO, uid, gid);
+ (void) fchown(STDOUT_FILENO, uid, gid);
+ (void) fchown(STDERR_FILENO, uid, gid);
+ }
+
+ if (setgroups(n_supplementary_gids, supplementary_gids) < 0)
+ return log_error_errno(errno, "Failed to set auxiliary groups: %m");
+
+ if (setresgid(gid, gid, gid) < 0)
+ return log_error_errno(errno, "setresgid() failed: %m");
+
+ if (setresuid(uid, uid, uid) < 0)
+ return log_error_errno(errno, "setresuid() failed: %m");
+
+ return 0;
+}
+
+int change_uid_gid(const char *user, bool chown_stdio, char **ret_home) {
+ char *x, *u, *g, *h;
+ _cleanup_free_ gid_t *gids = NULL;
+ _cleanup_free_ char *home = NULL, *line = NULL;
+ _cleanup_fclose_ FILE *f = NULL;
+ _cleanup_close_ int fd = -1;
+ unsigned n_gids = 0;
+ uid_t uid;
+ gid_t gid;
+ pid_t pid;
+ int r;
+
+ assert(ret_home);
+
+ if (!user || STR_IN_SET(user, "root", "0")) {
+ /* Reset everything fully to 0, just in case */
+
+ r = reset_uid_gid();
+ if (r < 0)
+ return log_error_errno(r, "Failed to become root: %m");
+
+ *ret_home = NULL;
+ return 0;
+ }
+
+ /* First, get user credentials */
+ fd = spawn_getent("passwd", user, &pid);
+ if (fd < 0)
+ return fd;
+
+ f = take_fdopen(&fd, "r");
+ if (!f)
+ return log_oom();
+
+ r = read_line(f, LONG_LINE_MAX, &line);
+ if (r == 0)
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
+ "Failed to resolve user %s.", user);
+ if (r < 0)
+ return log_error_errno(r, "Failed to read from getent: %m");
+
+ (void) wait_for_terminate_and_check("getent passwd", pid, WAIT_LOG);
+
+ x = strchr(line, ':');
+ if (!x)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "/etc/passwd entry has invalid user field.");
+
+ u = strchr(x+1, ':');
+ if (!u)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "/etc/passwd entry has invalid password field.");
+
+ u++;
+ g = strchr(u, ':');
+ if (!g)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "/etc/passwd entry has invalid UID field.");
+
+ *g = 0;
+ g++;
+ x = strchr(g, ':');
+ if (!x)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "/etc/passwd entry has invalid GID field.");
+
+ *x = 0;
+ h = strchr(x+1, ':');
+ if (!h)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "/etc/passwd entry has invalid GECOS field.");
+
+ h++;
+ x = strchr(h, ':');
+ if (!x)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "/etc/passwd entry has invalid home directory field.");
+
+ *x = 0;
+
+ r = parse_uid(u, &uid);
+ if (r < 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Failed to parse UID of user.");
+
+ r = parse_gid(g, &gid);
+ if (r < 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Failed to parse GID of user.");
+
+ home = strdup(h);
+ if (!home)
+ return log_oom();
+
+ f = safe_fclose(f);
+ line = mfree(line);
+
+ /* Second, get group memberships */
+ fd = spawn_getent("initgroups", user, &pid);
+ if (fd < 0)
+ return fd;
+
+ f = take_fdopen(&fd, "r");
+ if (!f)
+ return log_oom();
+
+ r = read_line(f, LONG_LINE_MAX, &line);
+ if (r == 0)
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
+ "Failed to resolve user %s.", user);
+ if (r < 0)
+ return log_error_errno(r, "Failed to read from getent: %m");
+
+ (void) wait_for_terminate_and_check("getent initgroups", pid, WAIT_LOG);
+
+ /* Skip over the username and subsequent separator whitespace */
+ x = line;
+ x += strcspn(x, WHITESPACE);
+ x += strspn(x, WHITESPACE);
+
+ for (const char *p = x;;) {
+ _cleanup_free_ char *word = NULL;
+
+ r = extract_first_word(&p, &word, NULL, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse group data from getent: %m");
+ if (r == 0)
+ break;
+
+ if (!GREEDY_REALLOC(gids, n_gids+1))
+ return log_oom();
+
+ r = parse_gid(word, &gids[n_gids++]);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse group data from getent: %m");
+ }
+
+ r = mkdir_parents(home, 0775);
+ if (r < 0)
+ return log_error_errno(r, "Failed to make home root directory: %m");
+
+ r = mkdir_safe(home, 0755, uid, gid, 0);
+ if (r < 0 && !IN_SET(r, -EEXIST, -ENOTDIR))
+ return log_error_errno(r, "Failed to make home directory: %m");
+
+ r = change_uid_gid_raw(uid, gid, gids, n_gids, chown_stdio);
+ if (r < 0)
+ return r;
+
+ if (ret_home)
+ *ret_home = TAKE_PTR(home);
+
+ return 0;
+}
diff --git a/src/nspawn/nspawn-setuid.h b/src/nspawn/nspawn-setuid.h
new file mode 100644
index 0000000..1924711
--- /dev/null
+++ b/src/nspawn/nspawn-setuid.h
@@ -0,0 +1,5 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+int change_uid_gid_raw(uid_t uid, gid_t gid, const gid_t *supplementary_gids, size_t n_supplementary_gids, bool chown_stdio);
+int change_uid_gid(const char *user, bool chown_stdio, char **ret_home);
diff --git a/src/nspawn/nspawn-stub-pid1.c b/src/nspawn/nspawn-stub-pid1.c
new file mode 100644
index 0000000..85c4398
--- /dev/null
+++ b/src/nspawn/nspawn-stub-pid1.c
@@ -0,0 +1,198 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <sys/ioctl.h>
+#include <sys/reboot.h>
+#include <sys/wait.h>
+#include <sys/prctl.h>
+#include <unistd.h>
+
+#include "def.h"
+#include "exit-status.h"
+#include "fd-util.h"
+#include "log.h"
+#include "nspawn-stub-pid1.h"
+#include "process-util.h"
+#include "signal-util.h"
+#include "time-util.h"
+
+static int reset_environ(const char *new_environment, size_t length) {
+ unsigned long start, end;
+
+ start = (unsigned long) new_environment;
+ end = start + length;
+
+ if (prctl(PR_SET_MM, PR_SET_MM_ENV_START, start, 0, 0) < 0)
+ return -errno;
+
+ if (prctl(PR_SET_MM, PR_SET_MM_ENV_END, end, 0, 0) < 0)
+ return -errno;
+
+ return 0;
+}
+
+int stub_pid1(sd_id128_t uuid) {
+ enum {
+ STATE_RUNNING,
+ STATE_REBOOT,
+ STATE_POWEROFF,
+ } state = STATE_RUNNING;
+
+ sigset_t fullmask, oldmask, waitmask;
+ usec_t quit_usec = USEC_INFINITY;
+ pid_t pid;
+ int r;
+
+ /* The new environment we set up, on the stack. */
+ char new_environment[] =
+ "container=systemd-nspawn\0"
+ "container_uuid=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX";
+
+ /* Implements a stub PID 1, that reaps all processes and processes a couple of standard signals. This is useful
+ * for allowing arbitrary processes run in a container, and still have all zombies reaped. */
+
+ assert_se(sigfillset(&fullmask) >= 0);
+ assert_se(sigprocmask(SIG_BLOCK, &fullmask, &oldmask) >= 0);
+
+ pid = fork();
+ if (pid < 0)
+ return log_error_errno(errno, "Failed to fork child pid: %m");
+
+ if (pid == 0) {
+ /* Return in the child */
+ assert_se(sigprocmask(SIG_SETMASK, &oldmask, NULL) >= 0);
+
+ if (setsid() < 0)
+ return log_error_errno(errno, "Failed to become session leader in payload process: %m");
+
+ return 0;
+ }
+
+ reset_all_signal_handlers();
+
+ log_close();
+ (void) close_all_fds(NULL, 0);
+ log_open();
+
+ if (ioctl(STDIN_FILENO, TIOCNOTTY) < 0) {
+ if (errno != ENOTTY)
+ log_warning_errno(errno, "Unexpected error from TIOCNOTTY ioctl in init stub process, ignoring: %m");
+ } else
+ log_warning("Expected TIOCNOTTY to fail, but it succeeded in init stub process, ignoring.");
+
+ /* Flush out /proc/self/environ, so that we don't leak the environment from the host into the container. Also,
+ * set $container= and $container_uuid= so that clients in the container that query it from /proc/1/environ
+ * find them set. */
+ sd_id128_to_string(uuid, new_environment + sizeof(new_environment) - SD_ID128_STRING_MAX);
+ reset_environ(new_environment, sizeof(new_environment));
+
+ (void) rename_process("(sd-stubinit)");
+
+ assert_se(sigemptyset(&waitmask) >= 0);
+ assert_se(sigset_add_many(&waitmask,
+ SIGCHLD, /* posix: process died */
+ SIGINT, /* sysv: ctrl-alt-del */
+ SIGRTMIN+3, /* systemd: halt */
+ SIGRTMIN+4, /* systemd: poweroff */
+ SIGRTMIN+5, /* systemd: reboot */
+ SIGRTMIN+6, /* systemd: kexec */
+ SIGRTMIN+13, /* systemd: halt */
+ SIGRTMIN+14, /* systemd: poweroff */
+ SIGRTMIN+15, /* systemd: reboot */
+ SIGRTMIN+16, /* systemd: kexec */
+ -1) >= 0);
+
+ /* Note that we ignore SIGTERM (sysv's reexec), SIGHUP (reload), and all other signals here, since we don't
+ * support reexec/reloading in this stub process. */
+
+ for (;;) {
+ siginfo_t si;
+ usec_t current_usec;
+
+ si.si_pid = 0;
+ r = waitid(P_ALL, 0, &si, WEXITED|WNOHANG);
+ if (r < 0) {
+ r = log_error_errno(errno, "Failed to reap children: %m");
+ goto finish;
+ }
+
+ current_usec = now(CLOCK_MONOTONIC);
+
+ if (si.si_pid == pid || current_usec >= quit_usec) {
+
+ /* The child we started ourselves died or we reached a timeout. */
+
+ if (state == STATE_REBOOT) { /* dispatch a queued reboot */
+ (void) reboot(RB_AUTOBOOT);
+ r = log_error_errno(errno, "Failed to reboot: %m");
+ goto finish;
+
+ } else if (state == STATE_POWEROFF)
+ (void) reboot(RB_POWER_OFF); /* if this fails, fall back to normal exit. */
+
+ if (si.si_pid == pid && si.si_code == CLD_EXITED)
+ r = si.si_status; /* pass on exit code */
+ else
+ r = EXIT_EXCEPTION; /* signal, coredump, timeout, … */
+
+ goto finish;
+ }
+ if (si.si_pid != 0)
+ /* We reaped something. Retry until there's nothing more to reap. */
+ continue;
+
+ if (quit_usec == USEC_INFINITY)
+ r = sigwaitinfo(&waitmask, &si);
+ else
+ r = sigtimedwait(&waitmask, &si, TIMESPEC_STORE(quit_usec - current_usec));
+ if (r < 0) {
+ if (errno == EINTR) /* strace -p attach can result in EINTR, let's handle this nicely. */
+ continue;
+ if (errno == EAGAIN) /* timeout reached */
+ continue;
+
+ r = log_error_errno(errno, "Failed to wait for signal: %m");
+ goto finish;
+ }
+
+ if (si.si_signo == SIGCHLD)
+ continue; /* Let's reap this */
+
+ if (state != STATE_RUNNING)
+ continue;
+
+ /* Would love to use a switch() statement here, but SIGRTMIN is actually a function call, not a
+ * constant… */
+
+ if (si.si_signo == SIGRTMIN+3 ||
+ si.si_signo == SIGRTMIN+4 ||
+ si.si_signo == SIGRTMIN+13 ||
+ si.si_signo == SIGRTMIN+14)
+
+ state = STATE_POWEROFF;
+
+ else if (si.si_signo == SIGINT ||
+ si.si_signo == SIGRTMIN+5 ||
+ si.si_signo == SIGRTMIN+6 ||
+ si.si_signo == SIGRTMIN+15 ||
+ si.si_signo == SIGRTMIN+16)
+
+ state = STATE_REBOOT;
+ else
+ assert_not_reached();
+
+ r = kill_and_sigcont(pid, SIGTERM);
+
+ /* Let's send a SIGHUP after the SIGTERM, as shells tend to ignore SIGTERM but do react to SIGHUP. We
+ * do it strictly in this order, so that the SIGTERM is dispatched first, and SIGHUP second for those
+ * processes which handle both. That's because services tend to bind configuration reload or something
+ * else to SIGHUP. */
+
+ if (r != -ESRCH)
+ (void) kill(pid, SIGHUP);
+
+ quit_usec = now(CLOCK_MONOTONIC) + DEFAULT_TIMEOUT_USEC;
+ }
+
+finish:
+ _exit(r < 0 ? EXIT_FAILURE : r);
+}
diff --git a/src/nspawn/nspawn-stub-pid1.h b/src/nspawn/nspawn-stub-pid1.h
new file mode 100644
index 0000000..e0810fe
--- /dev/null
+++ b/src/nspawn/nspawn-stub-pid1.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include "sd-id128.h"
+
+int stub_pid1(sd_id128_t uuid);
diff --git a/src/nspawn/nspawn-util.c b/src/nspawn/nspawn-util.c
new file mode 100644
index 0000000..830ac39
--- /dev/null
+++ b/src/nspawn/nspawn-util.c
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "alloc-util.h"
+#include "glob-util.h"
+#include "log.h"
+#include "nspawn-util.h"
+#include "parse-util.h"
+#include "path-util.h"
+#include "string-util.h"
+
+int systemd_installation_has_version(const char *root, const char *minimal_version) {
+ int r;
+
+ /* Try to guess if systemd installation is later than the specified version. This
+ * is hacky and likely to yield false negatives, particularly if the installation
+ * is non-standard. False positives should be relatively rare.
+ */
+
+ FOREACH_STRING(pattern,
+ /* /lib works for systems without usr-merge, and for systems with a sane
+ * usr-merge, where /lib is a symlink to /usr/lib. /usr/lib is necessary
+ * for Gentoo which does a merge without making /lib a symlink.
+ * Also support multiarch paths von Debian/Ubuntu; *-linux-* is a small
+ * optimization based on the naming scheme of existing multiarch tuples.
+ */
+ "/lib/systemd/libsystemd-shared-*.so",
+ "/lib64/systemd/libsystemd-shared-*.so",
+ "/usr/lib/*-linux-*/systemd/libsystemd-shared-*.so",
+ "/usr/lib/systemd/libsystemd-shared-*.so",
+ "/usr/lib64/systemd/libsystemd-shared-*.so") {
+
+ _cleanup_strv_free_ char **names = NULL;
+ _cleanup_free_ char *path = NULL;
+ char *c;
+
+ path = path_join(root, pattern);
+ if (!path)
+ return -ENOMEM;
+
+ r = glob_extend(&names, path, 0);
+ if (r == -ENOENT)
+ continue;
+ if (r < 0)
+ return r;
+
+ assert_se(c = endswith(path, "*.so"));
+ *c = '\0'; /* truncate the glob part */
+
+ STRV_FOREACH(name, names) {
+ /* This is most likely to run only once, hence let's not optimize anything. */
+ char *t, *t2;
+
+ t = startswith(basename(*name), "libsystemd-shared-");
+ if (!t)
+ continue;
+
+ t2 = endswith(t, ".so");
+ if (!t2)
+ continue;
+ *t2 = '\0';
+
+ r = strverscmp_improved(t, minimal_version);
+ log_debug("Found libsystemd shared at \"%s.so\", version %s (%s).",
+ *name, t,
+ r >= 0 ? "OK" : "too old");
+ if (r >= 0)
+ return true;
+ }
+ }
+
+ return false;
+}
diff --git a/src/nspawn/nspawn-util.h b/src/nspawn/nspawn-util.h
new file mode 100644
index 0000000..e83cd56
--- /dev/null
+++ b/src/nspawn/nspawn-util.h
@@ -0,0 +1,4 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+int systemd_installation_has_version(const char *root, const char *minimal_version);
diff --git a/src/nspawn/nspawn.c b/src/nspawn/nspawn.c
new file mode 100644
index 0000000..6c9f084
--- /dev/null
+++ b/src/nspawn/nspawn.c
@@ -0,0 +1,5913 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#if HAVE_BLKID
+#endif
+#include <errno.h>
+#include <getopt.h>
+#include <linux/fs.h>
+#include <linux/loop.h>
+#if HAVE_SELINUX
+#include <selinux/selinux.h>
+#endif
+#include <stdlib.h>
+#include <sys/file.h>
+#include <sys/ioctl.h>
+#include <sys/personality.h>
+#include <sys/prctl.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <termios.h>
+#include <unistd.h>
+
+#include "sd-bus.h"
+#include "sd-daemon.h"
+#include "sd-id128.h"
+
+#include "alloc-util.h"
+#include "barrier.h"
+#include "base-filesystem.h"
+#include "blkid-util.h"
+#include "btrfs-util.h"
+#include "bus-error.h"
+#include "bus-util.h"
+#include "cap-list.h"
+#include "capability-util.h"
+#include "cgroup-util.h"
+#include "chase-symlinks.h"
+#include "copy.h"
+#include "cpu-set-util.h"
+#include "creds-util.h"
+#include "dev-setup.h"
+#include "discover-image.h"
+#include "dissect-image.h"
+#include "env-util.h"
+#include "escape.h"
+#include "fd-util.h"
+#include "fdset.h"
+#include "fileio.h"
+#include "format-util.h"
+#include "fs-util.h"
+#include "gpt.h"
+#include "hexdecoct.h"
+#include "hostname-setup.h"
+#include "hostname-util.h"
+#include "id128-util.h"
+#include "io-util.h"
+#include "log.h"
+#include "loop-util.h"
+#include "loopback-setup.h"
+#include "macro.h"
+#include "main-func.h"
+#include "missing_sched.h"
+#include "mkdir.h"
+#include "mount-util.h"
+#include "mountpoint-util.h"
+#include "namespace-util.h"
+#include "netlink-util.h"
+#include "nspawn-bind-user.h"
+#include "nspawn-cgroup.h"
+#include "nspawn-creds.h"
+#include "nspawn-def.h"
+#include "nspawn-expose-ports.h"
+#include "nspawn-mount.h"
+#include "nspawn-network.h"
+#include "nspawn-oci.h"
+#include "nspawn-patch-uid.h"
+#include "nspawn-register.h"
+#include "nspawn-seccomp.h"
+#include "nspawn-settings.h"
+#include "nspawn-setuid.h"
+#include "nspawn-stub-pid1.h"
+#include "nspawn-util.h"
+#include "nspawn.h"
+#include "nulstr-util.h"
+#include "os-util.h"
+#include "pager.h"
+#include "parse-argument.h"
+#include "parse-util.h"
+#include "pretty-print.h"
+#include "process-util.h"
+#include "ptyfwd.h"
+#include "random-util.h"
+#include "raw-clone.h"
+#include "resolve-util.h"
+#include "rlimit-util.h"
+#include "rm-rf.h"
+#if HAVE_SECCOMP
+#include "seccomp-util.h"
+#endif
+#include "selinux-util.h"
+#include "signal-util.h"
+#include "socket-util.h"
+#include "stat-util.h"
+#include "stdio-util.h"
+#include "string-table.h"
+#include "string-util.h"
+#include "strv.h"
+#include "sysctl-util.h"
+#include "terminal-util.h"
+#include "tmpfile-util.h"
+#include "umask-util.h"
+#include "unit-name.h"
+#include "user-util.h"
+#include "util.h"
+
+/* The notify socket inside the container it can use to talk to nspawn using the sd_notify(3) protocol */
+#define NSPAWN_NOTIFY_SOCKET_PATH "/run/host/notify"
+
+#define EXIT_FORCE_RESTART 133
+
+typedef enum ContainerStatus {
+ CONTAINER_TERMINATED,
+ CONTAINER_REBOOTED,
+} ContainerStatus;
+
+static char *arg_directory = NULL;
+static char *arg_template = NULL;
+static char *arg_chdir = NULL;
+static char *arg_pivot_root_new = NULL;
+static char *arg_pivot_root_old = NULL;
+static char *arg_user = NULL;
+static uid_t arg_uid = UID_INVALID;
+static gid_t arg_gid = GID_INVALID;
+static gid_t* arg_supplementary_gids = NULL;
+static size_t arg_n_supplementary_gids = 0;
+static sd_id128_t arg_uuid = {};
+static char *arg_machine = NULL; /* The name used by the host to refer to this */
+static char *arg_hostname = NULL; /* The name the payload sees by default */
+static const char *arg_selinux_context = NULL;
+static const char *arg_selinux_apifs_context = NULL;
+static char *arg_slice = NULL;
+static bool arg_private_network = false;
+static bool arg_read_only = false;
+static StartMode arg_start_mode = START_PID1;
+static bool arg_ephemeral = false;
+static LinkJournal arg_link_journal = LINK_AUTO;
+static bool arg_link_journal_try = false;
+static uint64_t arg_caps_retain =
+ (1ULL << CAP_AUDIT_CONTROL) |
+ (1ULL << CAP_AUDIT_WRITE) |
+ (1ULL << CAP_CHOWN) |
+ (1ULL << CAP_DAC_OVERRIDE) |
+ (1ULL << CAP_DAC_READ_SEARCH) |
+ (1ULL << CAP_FOWNER) |
+ (1ULL << CAP_FSETID) |
+ (1ULL << CAP_IPC_OWNER) |
+ (1ULL << CAP_KILL) |
+ (1ULL << CAP_LEASE) |
+ (1ULL << CAP_LINUX_IMMUTABLE) |
+ (1ULL << CAP_MKNOD) |
+ (1ULL << CAP_NET_BIND_SERVICE) |
+ (1ULL << CAP_NET_BROADCAST) |
+ (1ULL << CAP_NET_RAW) |
+ (1ULL << CAP_SETFCAP) |
+ (1ULL << CAP_SETGID) |
+ (1ULL << CAP_SETPCAP) |
+ (1ULL << CAP_SETUID) |
+ (1ULL << CAP_SYS_ADMIN) |
+ (1ULL << CAP_SYS_BOOT) |
+ (1ULL << CAP_SYS_CHROOT) |
+ (1ULL << CAP_SYS_NICE) |
+ (1ULL << CAP_SYS_PTRACE) |
+ (1ULL << CAP_SYS_RESOURCE) |
+ (1ULL << CAP_SYS_TTY_CONFIG);
+static uint64_t arg_caps_ambient = 0;
+static CapabilityQuintet arg_full_capabilities = CAPABILITY_QUINTET_NULL;
+static CustomMount *arg_custom_mounts = NULL;
+static size_t arg_n_custom_mounts = 0;
+static char **arg_setenv = NULL;
+static bool arg_quiet = false;
+static bool arg_register = true;
+static bool arg_keep_unit = false;
+static char **arg_network_interfaces = NULL;
+static char **arg_network_macvlan = NULL;
+static char **arg_network_ipvlan = NULL;
+static bool arg_network_veth = false;
+static char **arg_network_veth_extra = NULL;
+static char *arg_network_bridge = NULL;
+static char *arg_network_zone = NULL;
+static char *arg_network_namespace_path = NULL;
+static PagerFlags arg_pager_flags = 0;
+static unsigned long arg_personality = PERSONALITY_INVALID;
+static char *arg_image = NULL;
+static char *arg_oci_bundle = NULL;
+static VolatileMode arg_volatile_mode = VOLATILE_NO;
+static ExposePort *arg_expose_ports = NULL;
+static char **arg_property = NULL;
+static sd_bus_message *arg_property_message = NULL;
+static UserNamespaceMode arg_userns_mode = USER_NAMESPACE_NO;
+static uid_t arg_uid_shift = UID_INVALID, arg_uid_range = 0x10000U;
+static UserNamespaceOwnership arg_userns_ownership = _USER_NAMESPACE_OWNERSHIP_INVALID;
+static int arg_kill_signal = 0;
+static CGroupUnified arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_UNKNOWN;
+static SettingsMask arg_settings_mask = 0;
+static int arg_settings_trusted = -1;
+static char **arg_parameters = NULL;
+static const char *arg_container_service_name = "systemd-nspawn";
+static bool arg_notify_ready = false;
+static bool arg_use_cgns = true;
+static unsigned long arg_clone_ns_flags = CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS;
+static MountSettingsMask arg_mount_settings = MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_TMPFS_TMP;
+static VeritySettings arg_verity_settings = VERITY_SETTINGS_DEFAULT;
+static char **arg_syscall_allow_list = NULL;
+static char **arg_syscall_deny_list = NULL;
+#if HAVE_SECCOMP
+static scmp_filter_ctx arg_seccomp = NULL;
+#endif
+static struct rlimit *arg_rlimit[_RLIMIT_MAX] = {};
+static bool arg_no_new_privileges = false;
+static int arg_oom_score_adjust = 0;
+static bool arg_oom_score_adjust_set = false;
+static CPUSet arg_cpu_set = {};
+static ResolvConfMode arg_resolv_conf = RESOLV_CONF_AUTO;
+static TimezoneMode arg_timezone = TIMEZONE_AUTO;
+static unsigned arg_console_width = UINT_MAX, arg_console_height = UINT_MAX;
+static DeviceNode* arg_extra_nodes = NULL;
+static size_t arg_n_extra_nodes = 0;
+static char **arg_sysctl = NULL;
+static ConsoleMode arg_console_mode = _CONSOLE_MODE_INVALID;
+static Credential *arg_credentials = NULL;
+static size_t arg_n_credentials = 0;
+static char **arg_bind_user = NULL;
+static bool arg_suppress_sync = false;
+static char *arg_settings_filename = NULL;
+
+STATIC_DESTRUCTOR_REGISTER(arg_directory, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_template, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_chdir, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_new, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_pivot_root_old, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_user, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_supplementary_gids, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_machine, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_hostname, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_slice, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_setenv, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_network_interfaces, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_network_macvlan, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_network_ipvlan, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_network_veth_extra, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_network_bridge, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_network_zone, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_network_namespace_path, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_image, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_oci_bundle, freep);
+STATIC_DESTRUCTOR_REGISTER(arg_property, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_property_message, sd_bus_message_unrefp);
+STATIC_DESTRUCTOR_REGISTER(arg_parameters, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_verity_settings, verity_settings_done);
+STATIC_DESTRUCTOR_REGISTER(arg_syscall_allow_list, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_syscall_deny_list, strv_freep);
+#if HAVE_SECCOMP
+STATIC_DESTRUCTOR_REGISTER(arg_seccomp, seccomp_releasep);
+#endif
+STATIC_DESTRUCTOR_REGISTER(arg_cpu_set, cpu_set_reset);
+STATIC_DESTRUCTOR_REGISTER(arg_sysctl, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_bind_user, strv_freep);
+STATIC_DESTRUCTOR_REGISTER(arg_settings_filename, freep);
+
+static int handle_arg_console(const char *arg) {
+ if (streq(arg, "help")) {
+ puts("autopipe\n"
+ "interactive\n"
+ "passive\n"
+ "pipe\n"
+ "read-only");
+ return 0;
+ }
+
+ if (streq(arg, "interactive"))
+ arg_console_mode = CONSOLE_INTERACTIVE;
+ else if (streq(arg, "read-only"))
+ arg_console_mode = CONSOLE_READ_ONLY;
+ else if (streq(arg, "passive"))
+ arg_console_mode = CONSOLE_PASSIVE;
+ else if (streq(arg, "pipe")) {
+ if (isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0)
+ log_full(arg_quiet ? LOG_DEBUG : LOG_NOTICE,
+ "Console mode 'pipe' selected, but standard input/output are connected to an interactive TTY. "
+ "Most likely you want to use 'interactive' console mode for proper interactivity and shell job control. "
+ "Proceeding anyway.");
+
+ arg_console_mode = CONSOLE_PIPE;
+ } else if (streq(arg, "autopipe")) {
+ if (isatty(STDIN_FILENO) > 0 && isatty(STDOUT_FILENO) > 0)
+ arg_console_mode = CONSOLE_INTERACTIVE;
+ else
+ arg_console_mode = CONSOLE_PIPE;
+ } else
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Unknown console mode: %s", optarg);
+
+ arg_settings_mask |= SETTING_CONSOLE_MODE;
+ return 1;
+}
+
+static int help(void) {
+ _cleanup_free_ char *link = NULL;
+ int r;
+
+ pager_open(arg_pager_flags);
+
+ r = terminal_urlify_man("systemd-nspawn", "1", &link);
+ if (r < 0)
+ return log_oom();
+
+ printf("%1$s [OPTIONS...] [PATH] [ARGUMENTS...]\n\n"
+ "%5$sSpawn a command or OS in a light-weight container.%6$s\n\n"
+ " -h --help Show this help\n"
+ " --version Print version string\n"
+ " -q --quiet Do not show status information\n"
+ " --no-pager Do not pipe output into a pager\n"
+ " --settings=BOOLEAN Load additional settings from .nspawn file\n\n"
+ "%3$sImage:%4$s\n"
+ " -D --directory=PATH Root directory for the container\n"
+ " --template=PATH Initialize root directory from template directory,\n"
+ " if missing\n"
+ " -x --ephemeral Run container with snapshot of root directory, and\n"
+ " remove it after exit\n"
+ " -i --image=PATH Root file system disk image (or device node) for\n"
+ " the container\n"
+ " --oci-bundle=PATH OCI bundle directory\n"
+ " --read-only Mount the root directory read-only\n"
+ " --volatile[=MODE] Run the system in volatile mode\n"
+ " --root-hash=HASH Specify verity root hash for root disk image\n"
+ " --root-hash-sig=SIG Specify pkcs7 signature of root hash for verity\n"
+ " as a DER encoded PKCS7, either as a path to a file\n"
+ " or as an ASCII base64 encoded string prefixed by\n"
+ " 'base64:'\n"
+ " --verity-data=PATH Specify hash device for verity\n"
+ " --pivot-root=PATH[:PATH]\n"
+ " Pivot root to given directory in the container\n\n"
+ "%3$sExecution:%4$s\n"
+ " -a --as-pid2 Maintain a stub init as PID1, invoke binary as PID2\n"
+ " -b --boot Boot up full system (i.e. invoke init)\n"
+ " --chdir=PATH Set working directory in the container\n"
+ " -E --setenv=NAME[=VALUE] Pass an environment variable to PID 1\n"
+ " -u --user=USER Run the command under specified user or UID\n"
+ " --kill-signal=SIGNAL Select signal to use for shutting down PID 1\n"
+ " --notify-ready=BOOLEAN Receive notifications from the child init process\n"
+ " --suppress-sync=BOOLEAN\n"
+ " Suppress any form of disk data synchronization\n\n"
+ "%3$sSystem Identity:%4$s\n"
+ " -M --machine=NAME Set the machine name for the container\n"
+ " --hostname=NAME Override the hostname for the container\n"
+ " --uuid=UUID Set a specific machine UUID for the container\n\n"
+ "%3$sProperties:%4$s\n"
+ " -S --slice=SLICE Place the container in the specified slice\n"
+ " --property=NAME=VALUE Set scope unit property\n"
+ " --register=BOOLEAN Register container as machine\n"
+ " --keep-unit Do not register a scope for the machine, reuse\n"
+ " the service unit nspawn is running in\n\n"
+ "%3$sUser Namespacing:%4$s\n"
+ " --private-users=no Run without user namespacing\n"
+ " --private-users=yes|pick|identity\n"
+ " Run within user namespace, autoselect UID/GID range\n"
+ " --private-users=UIDBASE[:NUIDS]\n"
+ " Similar, but with user configured UID/GID range\n"
+ " --private-users-ownership=MODE\n"
+ " Adjust ('chown') or map ('map') OS tree ownership\n"
+ " to private UID/GID range\n"
+ " -U Equivalent to --private-users=pick and\n"
+ " --private-users-ownership=auto\n\n"
+ "%3$sNetworking:%4$s\n"
+ " --private-network Disable network in container\n"
+ " --network-interface=INTERFACE\n"
+ " Assign an existing network interface to the\n"
+ " container\n"
+ " --network-macvlan=INTERFACE\n"
+ " Create a macvlan network interface based on an\n"
+ " existing network interface to the container\n"
+ " --network-ipvlan=INTERFACE\n"
+ " Create an ipvlan network interface based on an\n"
+ " existing network interface to the container\n"
+ " -n --network-veth Add a virtual Ethernet connection between host\n"
+ " and container\n"
+ " --network-veth-extra=HOSTIF[:CONTAINERIF]\n"
+ " Add an additional virtual Ethernet link between\n"
+ " host and container\n"
+ " --network-bridge=INTERFACE\n"
+ " Add a virtual Ethernet connection to the container\n"
+ " and attach it to an existing bridge on the host\n"
+ " --network-zone=NAME Similar, but attach the new interface to an\n"
+ " an automatically managed bridge interface\n"
+ " --network-namespace-path=PATH\n"
+ " Set network namespace to the one represented by\n"
+ " the specified kernel namespace file node\n"
+ " -p --port=[PROTOCOL:]HOSTPORT[:CONTAINERPORT]\n"
+ " Expose a container IP port on the host\n\n"
+ "%3$sSecurity:%4$s\n"
+ " --capability=CAP In addition to the default, retain specified\n"
+ " capability\n"
+ " --drop-capability=CAP Drop the specified capability from the default set\n"
+ " --ambient-capability=CAP\n"
+ " Sets the specified capability for the started\n"
+ " process. Not useful if booting a machine.\n"
+ " --no-new-privileges Set PR_SET_NO_NEW_PRIVS flag for container payload\n"
+ " --system-call-filter=LIST|~LIST\n"
+ " Permit/prohibit specific system calls\n"
+ " -Z --selinux-context=SECLABEL\n"
+ " Set the SELinux security context to be used by\n"
+ " processes in the container\n"
+ " -L --selinux-apifs-context=SECLABEL\n"
+ " Set the SELinux security context to be used by\n"
+ " API/tmpfs file systems in the container\n\n"
+ "%3$sResources:%4$s\n"
+ " --rlimit=NAME=LIMIT Set a resource limit for the payload\n"
+ " --oom-score-adjust=VALUE\n"
+ " Adjust the OOM score value for the payload\n"
+ " --cpu-affinity=CPUS Adjust the CPU affinity of the container\n"
+ " --personality=ARCH Pick personality for this container\n\n"
+ "%3$sIntegration:%4$s\n"
+ " --resolv-conf=MODE Select mode of /etc/resolv.conf initialization\n"
+ " --timezone=MODE Select mode of /etc/localtime initialization\n"
+ " --link-journal=MODE Link up guest journal, one of no, auto, guest, \n"
+ " host, try-guest, try-host\n"
+ " -j Equivalent to --link-journal=try-guest\n\n"
+ "%3$sMounts:%4$s\n"
+ " --bind=PATH[:PATH[:OPTIONS]]\n"
+ " Bind mount a file or directory from the host into\n"
+ " the container\n"
+ " --bind-ro=PATH[:PATH[:OPTIONS]\n"
+ " Similar, but creates a read-only bind mount\n"
+ " --inaccessible=PATH Over-mount file node with inaccessible node to mask\n"
+ " it\n"
+ " --tmpfs=PATH:[OPTIONS] Mount an empty tmpfs to the specified directory\n"
+ " --overlay=PATH[:PATH...]:PATH\n"
+ " Create an overlay mount from the host to \n"
+ " the container\n"
+ " --overlay-ro=PATH[:PATH...]:PATH\n"
+ " Similar, but creates a read-only overlay mount\n"
+ " --bind-user=NAME Bind user from host to container\n\n"
+ "%3$sInput/Output:%4$s\n"
+ " --console=MODE Select how stdin/stdout/stderr and /dev/console are\n"
+ " set up for the container.\n"
+ " -P --pipe Equivalent to --console=pipe\n\n"
+ "%3$sCredentials:%4$s\n"
+ " --set-credential=ID:VALUE\n"
+ " Pass a credential with literal value to container.\n"
+ " --load-credential=ID:PATH\n"
+ " Load credential to pass to container from file or\n"
+ " AF_UNIX stream socket.\n"
+ "\nSee the %2$s for details.\n",
+ program_invocation_short_name,
+ link,
+ ansi_underline(),
+ ansi_normal(),
+ ansi_highlight(),
+ ansi_normal());
+
+ return 0;
+}
+
+static int custom_mount_check_all(void) {
+ size_t i;
+
+ for (i = 0; i < arg_n_custom_mounts; i++) {
+ CustomMount *m = &arg_custom_mounts[i];
+
+ if (path_equal(m->destination, "/") && arg_userns_mode != USER_NAMESPACE_NO) {
+ if (arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_OFF)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "--private-users-ownership=own may not be combined with custom root mounts.");
+ if (arg_uid_shift == UID_INVALID)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "--private-users with automatic UID shift may not be combined with custom root mounts.");
+ }
+ }
+
+ return 0;
+}
+
+static int detect_unified_cgroup_hierarchy_from_environment(void) {
+ const char *e, *var = "SYSTEMD_NSPAWN_UNIFIED_HIERARCHY";
+ int r;
+
+ /* Allow the user to control whether the unified hierarchy is used */
+
+ e = getenv(var);
+ if (!e) {
+ /* $UNIFIED_CGROUP_HIERARCHY has been renamed to $SYSTEMD_NSPAWN_UNIFIED_HIERARCHY. */
+ var = "UNIFIED_CGROUP_HIERARCHY";
+ e = getenv(var);
+ }
+
+ if (!isempty(e)) {
+ r = parse_boolean(e);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse $%s: %m", var);
+ if (r > 0)
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
+ else
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
+ }
+
+ return 0;
+}
+
+static int detect_unified_cgroup_hierarchy_from_image(const char *directory) {
+ int r;
+
+ /* Let's inherit the mode to use from the host system, but let's take into consideration what systemd
+ * in the image actually supports. */
+ r = cg_all_unified();
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
+ if (r > 0) {
+ /* Unified cgroup hierarchy support was added in 230. Unfortunately the detection
+ * routine only detects 231, so we'll have a false negative here for 230. */
+ r = systemd_installation_has_version(directory, "230");
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine systemd version in container: %m");
+ if (r > 0)
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
+ else
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
+ } else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) {
+ /* Mixed cgroup hierarchy support was added in 233 */
+ r = systemd_installation_has_version(directory, "233");
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine systemd version in container: %m");
+ if (r > 0)
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
+ else
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
+ } else
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
+
+ log_debug("Using %s hierarchy for container.",
+ arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_NONE ? "legacy" :
+ arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_SYSTEMD ? "hybrid" : "unified");
+
+ return 0;
+}
+
+static int parse_capability_spec(const char *spec, uint64_t *ret_mask) {
+ uint64_t mask = 0;
+ int r;
+
+ for (;;) {
+ _cleanup_free_ char *t = NULL;
+
+ r = extract_first_word(&spec, &t, ",", 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse capability %s.", t);
+ if (r == 0)
+ break;
+
+ if (streq(t, "help")) {
+ for (int i = 0; i < capability_list_length(); i++) {
+ const char *name;
+
+ name = capability_to_name(i);
+ if (name)
+ puts(name);
+ }
+
+ return 0; /* quit */
+ }
+
+ if (streq(t, "all"))
+ mask = UINT64_MAX;
+ else {
+ r = capability_from_name(t);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse capability %s.", t);
+
+ mask |= 1ULL << r;
+ }
+ }
+
+ *ret_mask = mask;
+ return 1; /* continue */
+}
+
+static int parse_share_ns_env(const char *name, unsigned long ns_flag) {
+ int r;
+
+ r = getenv_bool(name);
+ if (r == -ENXIO)
+ return 0;
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse $%s: %m", name);
+
+ arg_clone_ns_flags = (arg_clone_ns_flags & ~ns_flag) | (r > 0 ? 0 : ns_flag);
+ arg_settings_mask |= SETTING_CLONE_NS_FLAGS;
+ return 0;
+}
+
+static int parse_mount_settings_env(void) {
+ const char *e;
+ int r;
+
+ r = getenv_bool("SYSTEMD_NSPAWN_TMPFS_TMP");
+ if (r < 0 && r != -ENXIO)
+ return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_TMPFS_TMP: %m");
+ if (r >= 0)
+ SET_FLAG(arg_mount_settings, MOUNT_APPLY_TMPFS_TMP, r > 0);
+
+ e = getenv("SYSTEMD_NSPAWN_API_VFS_WRITABLE");
+ if (streq_ptr(e, "network"))
+ arg_mount_settings |= MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS;
+
+ else if (e) {
+ r = parse_boolean(e);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_API_VFS_WRITABLE: %m");
+
+ SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_RO, r == 0);
+ SET_FLAG(arg_mount_settings, MOUNT_APPLY_APIVFS_NETNS, false);
+ }
+
+ return 0;
+}
+
+static int parse_environment(void) {
+ const char *e;
+ int r;
+
+ r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_IPC", CLONE_NEWIPC);
+ if (r < 0)
+ return r;
+ r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_PID", CLONE_NEWPID);
+ if (r < 0)
+ return r;
+ r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_NS_UTS", CLONE_NEWUTS);
+ if (r < 0)
+ return r;
+ r = parse_share_ns_env("SYSTEMD_NSPAWN_SHARE_SYSTEM", CLONE_NEWIPC|CLONE_NEWPID|CLONE_NEWUTS);
+ if (r < 0)
+ return r;
+
+ r = parse_mount_settings_env();
+ if (r < 0)
+ return r;
+
+ /* SYSTEMD_NSPAWN_USE_CGNS=0 can be used to disable CLONE_NEWCGROUP use,
+ * even if it is supported. If not supported, it has no effect. */
+ if (!cg_ns_supported())
+ arg_use_cgns = false;
+ else {
+ r = getenv_bool("SYSTEMD_NSPAWN_USE_CGNS");
+ if (r < 0) {
+ if (r != -ENXIO)
+ return log_error_errno(r, "Failed to parse $SYSTEMD_NSPAWN_USE_CGNS: %m");
+
+ arg_use_cgns = true;
+ } else {
+ arg_use_cgns = r > 0;
+ arg_settings_mask |= SETTING_USE_CGNS;
+ }
+ }
+
+ e = getenv("SYSTEMD_NSPAWN_CONTAINER_SERVICE");
+ if (e)
+ arg_container_service_name = e;
+
+ r = getenv_bool("SYSTEMD_SUPPRESS_SYNC");
+ if (r >= 0)
+ arg_suppress_sync = r;
+ else if (r != -ENXIO)
+ log_debug_errno(r, "Failed to parse $SYSTEMD_SUPPRESS_SYNC, ignoring: %m");
+
+ return detect_unified_cgroup_hierarchy_from_environment();
+}
+
+static int parse_argv(int argc, char *argv[]) {
+ enum {
+ ARG_VERSION = 0x100,
+ ARG_PRIVATE_NETWORK,
+ ARG_UUID,
+ ARG_READ_ONLY,
+ ARG_CAPABILITY,
+ ARG_AMBIENT_CAPABILITY,
+ ARG_DROP_CAPABILITY,
+ ARG_LINK_JOURNAL,
+ ARG_BIND,
+ ARG_BIND_RO,
+ ARG_TMPFS,
+ ARG_OVERLAY,
+ ARG_OVERLAY_RO,
+ ARG_INACCESSIBLE,
+ ARG_SHARE_SYSTEM,
+ ARG_REGISTER,
+ ARG_KEEP_UNIT,
+ ARG_NETWORK_INTERFACE,
+ ARG_NETWORK_MACVLAN,
+ ARG_NETWORK_IPVLAN,
+ ARG_NETWORK_BRIDGE,
+ ARG_NETWORK_ZONE,
+ ARG_NETWORK_VETH_EXTRA,
+ ARG_NETWORK_NAMESPACE_PATH,
+ ARG_PERSONALITY,
+ ARG_VOLATILE,
+ ARG_TEMPLATE,
+ ARG_PROPERTY,
+ ARG_PRIVATE_USERS,
+ ARG_KILL_SIGNAL,
+ ARG_SETTINGS,
+ ARG_CHDIR,
+ ARG_PIVOT_ROOT,
+ ARG_PRIVATE_USERS_CHOWN,
+ ARG_PRIVATE_USERS_OWNERSHIP,
+ ARG_NOTIFY_READY,
+ ARG_ROOT_HASH,
+ ARG_ROOT_HASH_SIG,
+ ARG_VERITY_DATA,
+ ARG_SYSTEM_CALL_FILTER,
+ ARG_RLIMIT,
+ ARG_HOSTNAME,
+ ARG_NO_NEW_PRIVILEGES,
+ ARG_OOM_SCORE_ADJUST,
+ ARG_CPU_AFFINITY,
+ ARG_RESOLV_CONF,
+ ARG_TIMEZONE,
+ ARG_CONSOLE,
+ ARG_PIPE,
+ ARG_OCI_BUNDLE,
+ ARG_NO_PAGER,
+ ARG_SET_CREDENTIAL,
+ ARG_LOAD_CREDENTIAL,
+ ARG_BIND_USER,
+ ARG_SUPPRESS_SYNC,
+ };
+
+ static const struct option options[] = {
+ { "help", no_argument, NULL, 'h' },
+ { "version", no_argument, NULL, ARG_VERSION },
+ { "directory", required_argument, NULL, 'D' },
+ { "template", required_argument, NULL, ARG_TEMPLATE },
+ { "ephemeral", no_argument, NULL, 'x' },
+ { "user", required_argument, NULL, 'u' },
+ { "private-network", no_argument, NULL, ARG_PRIVATE_NETWORK },
+ { "as-pid2", no_argument, NULL, 'a' },
+ { "boot", no_argument, NULL, 'b' },
+ { "uuid", required_argument, NULL, ARG_UUID },
+ { "read-only", no_argument, NULL, ARG_READ_ONLY },
+ { "capability", required_argument, NULL, ARG_CAPABILITY },
+ { "ambient-capability", required_argument, NULL, ARG_AMBIENT_CAPABILITY },
+ { "drop-capability", required_argument, NULL, ARG_DROP_CAPABILITY },
+ { "no-new-privileges", required_argument, NULL, ARG_NO_NEW_PRIVILEGES },
+ { "link-journal", required_argument, NULL, ARG_LINK_JOURNAL },
+ { "bind", required_argument, NULL, ARG_BIND },
+ { "bind-ro", required_argument, NULL, ARG_BIND_RO },
+ { "tmpfs", required_argument, NULL, ARG_TMPFS },
+ { "overlay", required_argument, NULL, ARG_OVERLAY },
+ { "overlay-ro", required_argument, NULL, ARG_OVERLAY_RO },
+ { "inaccessible", required_argument, NULL, ARG_INACCESSIBLE },
+ { "machine", required_argument, NULL, 'M' },
+ { "hostname", required_argument, NULL, ARG_HOSTNAME },
+ { "slice", required_argument, NULL, 'S' },
+ { "setenv", required_argument, NULL, 'E' },
+ { "selinux-context", required_argument, NULL, 'Z' },
+ { "selinux-apifs-context", required_argument, NULL, 'L' },
+ { "quiet", no_argument, NULL, 'q' },
+ { "share-system", no_argument, NULL, ARG_SHARE_SYSTEM }, /* not documented */
+ { "register", required_argument, NULL, ARG_REGISTER },
+ { "keep-unit", no_argument, NULL, ARG_KEEP_UNIT },
+ { "network-interface", required_argument, NULL, ARG_NETWORK_INTERFACE },
+ { "network-macvlan", required_argument, NULL, ARG_NETWORK_MACVLAN },
+ { "network-ipvlan", required_argument, NULL, ARG_NETWORK_IPVLAN },
+ { "network-veth", no_argument, NULL, 'n' },
+ { "network-veth-extra", required_argument, NULL, ARG_NETWORK_VETH_EXTRA },
+ { "network-bridge", required_argument, NULL, ARG_NETWORK_BRIDGE },
+ { "network-zone", required_argument, NULL, ARG_NETWORK_ZONE },
+ { "network-namespace-path", required_argument, NULL, ARG_NETWORK_NAMESPACE_PATH },
+ { "personality", required_argument, NULL, ARG_PERSONALITY },
+ { "image", required_argument, NULL, 'i' },
+ { "volatile", optional_argument, NULL, ARG_VOLATILE },
+ { "port", required_argument, NULL, 'p' },
+ { "property", required_argument, NULL, ARG_PROPERTY },
+ { "private-users", optional_argument, NULL, ARG_PRIVATE_USERS },
+ { "private-users-chown", optional_argument, NULL, ARG_PRIVATE_USERS_CHOWN }, /* obsolete */
+ { "private-users-ownership",required_argument, NULL, ARG_PRIVATE_USERS_OWNERSHIP},
+ { "kill-signal", required_argument, NULL, ARG_KILL_SIGNAL },
+ { "settings", required_argument, NULL, ARG_SETTINGS },
+ { "chdir", required_argument, NULL, ARG_CHDIR },
+ { "pivot-root", required_argument, NULL, ARG_PIVOT_ROOT },
+ { "notify-ready", required_argument, NULL, ARG_NOTIFY_READY },
+ { "root-hash", required_argument, NULL, ARG_ROOT_HASH },
+ { "root-hash-sig", required_argument, NULL, ARG_ROOT_HASH_SIG },
+ { "verity-data", required_argument, NULL, ARG_VERITY_DATA },
+ { "system-call-filter", required_argument, NULL, ARG_SYSTEM_CALL_FILTER },
+ { "rlimit", required_argument, NULL, ARG_RLIMIT },
+ { "oom-score-adjust", required_argument, NULL, ARG_OOM_SCORE_ADJUST },
+ { "cpu-affinity", required_argument, NULL, ARG_CPU_AFFINITY },
+ { "resolv-conf", required_argument, NULL, ARG_RESOLV_CONF },
+ { "timezone", required_argument, NULL, ARG_TIMEZONE },
+ { "console", required_argument, NULL, ARG_CONSOLE },
+ { "pipe", no_argument, NULL, ARG_PIPE },
+ { "oci-bundle", required_argument, NULL, ARG_OCI_BUNDLE },
+ { "no-pager", no_argument, NULL, ARG_NO_PAGER },
+ { "set-credential", required_argument, NULL, ARG_SET_CREDENTIAL },
+ { "load-credential", required_argument, NULL, ARG_LOAD_CREDENTIAL },
+ { "bind-user", required_argument, NULL, ARG_BIND_USER },
+ { "suppress-sync", required_argument, NULL, ARG_SUPPRESS_SYNC },
+ {}
+ };
+
+ int c, r;
+ uint64_t plus = 0, minus = 0;
+ bool mask_all_settings = false, mask_no_settings = false;
+
+ assert(argc >= 0);
+ assert(argv);
+
+ /* Resetting to 0 forces the invocation of an internal initialization routine of getopt_long()
+ * that checks for GNU extensions in optstring ('-' or '+' at the beginning). */
+ optind = 0;
+ while ((c = getopt_long(argc, argv, "+hD:u:abL:M:jS:Z:qi:xp:nUE:P", options, NULL)) >= 0)
+ switch (c) {
+
+ case 'h':
+ return help();
+
+ case ARG_VERSION:
+ return version();
+
+ case 'D':
+ r = parse_path_argument(optarg, false, &arg_directory);
+ if (r < 0)
+ return r;
+
+ arg_settings_mask |= SETTING_DIRECTORY;
+ break;
+
+ case ARG_TEMPLATE:
+ r = parse_path_argument(optarg, false, &arg_template);
+ if (r < 0)
+ return r;
+
+ arg_settings_mask |= SETTING_DIRECTORY;
+ break;
+
+ case 'i':
+ r = parse_path_argument(optarg, false, &arg_image);
+ if (r < 0)
+ return r;
+
+ arg_settings_mask |= SETTING_DIRECTORY;
+ break;
+
+ case ARG_OCI_BUNDLE:
+ r = parse_path_argument(optarg, false, &arg_oci_bundle);
+ if (r < 0)
+ return r;
+
+ break;
+
+ case 'x':
+ arg_ephemeral = true;
+ arg_settings_mask |= SETTING_EPHEMERAL;
+ break;
+
+ case 'u':
+ r = free_and_strdup(&arg_user, optarg);
+ if (r < 0)
+ return log_oom();
+
+ arg_settings_mask |= SETTING_USER;
+ break;
+
+ case ARG_NETWORK_ZONE: {
+ char *j;
+
+ j = strjoin("vz-", optarg);
+ if (!j)
+ return log_oom();
+
+ if (!ifname_valid(j)) {
+ log_error("Network zone name not valid: %s", j);
+ free(j);
+ return -EINVAL;
+ }
+
+ free_and_replace(arg_network_zone, j);
+
+ arg_network_veth = true;
+ arg_private_network = true;
+ arg_settings_mask |= SETTING_NETWORK;
+ break;
+ }
+
+ case ARG_NETWORK_BRIDGE:
+
+ if (!ifname_valid(optarg))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Bridge interface name not valid: %s", optarg);
+
+ r = free_and_strdup(&arg_network_bridge, optarg);
+ if (r < 0)
+ return log_oom();
+
+ _fallthrough_;
+ case 'n':
+ arg_network_veth = true;
+ arg_private_network = true;
+ arg_settings_mask |= SETTING_NETWORK;
+ break;
+
+ case ARG_NETWORK_VETH_EXTRA:
+ r = veth_extra_parse(&arg_network_veth_extra, optarg);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --network-veth-extra= parameter: %s", optarg);
+
+ arg_private_network = true;
+ arg_settings_mask |= SETTING_NETWORK;
+ break;
+
+ case ARG_NETWORK_INTERFACE:
+ if (!ifname_valid(optarg))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Network interface name not valid: %s", optarg);
+
+ r = test_network_interface_initialized(optarg);
+ if (r < 0)
+ return r;
+
+ if (strv_extend(&arg_network_interfaces, optarg) < 0)
+ return log_oom();
+
+ arg_private_network = true;
+ arg_settings_mask |= SETTING_NETWORK;
+ break;
+
+ case ARG_NETWORK_MACVLAN:
+
+ if (!ifname_valid(optarg))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "MACVLAN network interface name not valid: %s", optarg);
+
+ r = test_network_interface_initialized(optarg);
+ if (r < 0)
+ return r;
+
+ if (strv_extend(&arg_network_macvlan, optarg) < 0)
+ return log_oom();
+
+ arg_private_network = true;
+ arg_settings_mask |= SETTING_NETWORK;
+ break;
+
+ case ARG_NETWORK_IPVLAN:
+
+ if (!ifname_valid(optarg))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "IPVLAN network interface name not valid: %s", optarg);
+
+ r = test_network_interface_initialized(optarg);
+ if (r < 0)
+ return r;
+
+ if (strv_extend(&arg_network_ipvlan, optarg) < 0)
+ return log_oom();
+
+ _fallthrough_;
+ case ARG_PRIVATE_NETWORK:
+ arg_private_network = true;
+ arg_settings_mask |= SETTING_NETWORK;
+ break;
+
+ case ARG_NETWORK_NAMESPACE_PATH:
+ r = parse_path_argument(optarg, false, &arg_network_namespace_path);
+ if (r < 0)
+ return r;
+
+ arg_settings_mask |= SETTING_NETWORK;
+ break;
+
+ case 'b':
+ if (arg_start_mode == START_PID2)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "--boot and --as-pid2 may not be combined.");
+
+ arg_start_mode = START_BOOT;
+ arg_settings_mask |= SETTING_START_MODE;
+ break;
+
+ case 'a':
+ if (arg_start_mode == START_BOOT)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "--boot and --as-pid2 may not be combined.");
+
+ arg_start_mode = START_PID2;
+ arg_settings_mask |= SETTING_START_MODE;
+ break;
+
+ case ARG_UUID:
+ r = sd_id128_from_string(optarg, &arg_uuid);
+ if (r < 0)
+ return log_error_errno(r, "Invalid UUID: %s", optarg);
+
+ if (sd_id128_is_null(arg_uuid))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Machine UUID may not be all zeroes.");
+
+ arg_settings_mask |= SETTING_MACHINE_ID;
+ break;
+
+ case 'S': {
+ _cleanup_free_ char *mangled = NULL;
+
+ r = unit_name_mangle_with_suffix(optarg, NULL, UNIT_NAME_MANGLE_WARN, ".slice", &mangled);
+ if (r < 0)
+ return log_oom();
+
+ free_and_replace(arg_slice, mangled);
+ arg_settings_mask |= SETTING_SLICE;
+ break;
+ }
+
+ case 'M':
+ if (isempty(optarg))
+ arg_machine = mfree(arg_machine);
+ else {
+ if (!hostname_is_valid(optarg, 0))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Invalid machine name: %s", optarg);
+
+ r = free_and_strdup(&arg_machine, optarg);
+ if (r < 0)
+ return log_oom();
+ }
+ break;
+
+ case ARG_HOSTNAME:
+ if (isempty(optarg))
+ arg_hostname = mfree(arg_hostname);
+ else {
+ if (!hostname_is_valid(optarg, 0))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Invalid hostname: %s", optarg);
+
+ r = free_and_strdup(&arg_hostname, optarg);
+ if (r < 0)
+ return log_oom();
+ }
+
+ arg_settings_mask |= SETTING_HOSTNAME;
+ break;
+
+ case 'Z':
+ arg_selinux_context = optarg;
+ break;
+
+ case 'L':
+ arg_selinux_apifs_context = optarg;
+ break;
+
+ case ARG_READ_ONLY:
+ arg_read_only = true;
+ arg_settings_mask |= SETTING_READ_ONLY;
+ break;
+
+ case ARG_AMBIENT_CAPABILITY: {
+ uint64_t m;
+ r = parse_capability_spec(optarg, &m);
+ if (r <= 0)
+ return r;
+ arg_caps_ambient |= m;
+ arg_settings_mask |= SETTING_CAPABILITY;
+ break;
+ }
+ case ARG_CAPABILITY:
+ case ARG_DROP_CAPABILITY: {
+ uint64_t m;
+ r = parse_capability_spec(optarg, &m);
+ if (r <= 0)
+ return r;
+
+ if (c == ARG_CAPABILITY)
+ plus |= m;
+ else
+ minus |= m;
+ arg_settings_mask |= SETTING_CAPABILITY;
+ break;
+ }
+ case ARG_NO_NEW_PRIVILEGES:
+ r = parse_boolean(optarg);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --no-new-privileges= argument: %s", optarg);
+
+ arg_no_new_privileges = r;
+ arg_settings_mask |= SETTING_NO_NEW_PRIVILEGES;
+ break;
+
+ case 'j':
+ arg_link_journal = LINK_GUEST;
+ arg_link_journal_try = true;
+ arg_settings_mask |= SETTING_LINK_JOURNAL;
+ break;
+
+ case ARG_LINK_JOURNAL:
+ r = parse_link_journal(optarg, &arg_link_journal, &arg_link_journal_try);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse link journal mode %s", optarg);
+
+ arg_settings_mask |= SETTING_LINK_JOURNAL;
+ break;
+
+ case ARG_BIND:
+ case ARG_BIND_RO:
+ r = bind_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_BIND_RO);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --bind(-ro)= argument %s: %m", optarg);
+
+ arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
+ break;
+
+ case ARG_TMPFS:
+ r = tmpfs_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --tmpfs= argument %s: %m", optarg);
+
+ arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
+ break;
+
+ case ARG_OVERLAY:
+ case ARG_OVERLAY_RO:
+ r = overlay_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg, c == ARG_OVERLAY_RO);
+ if (r == -EADDRNOTAVAIL)
+ return log_error_errno(r, "--overlay(-ro)= needs at least two colon-separated directories specified.");
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --overlay(-ro)= argument %s: %m", optarg);
+
+ arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
+ break;
+
+ case ARG_INACCESSIBLE:
+ r = inaccessible_mount_parse(&arg_custom_mounts, &arg_n_custom_mounts, optarg);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --inaccessible= argument %s: %m", optarg);
+
+ arg_settings_mask |= SETTING_CUSTOM_MOUNTS;
+ break;
+
+ case 'E':
+ r = strv_env_replace_strdup_passthrough(&arg_setenv, optarg);
+ if (r < 0)
+ return log_error_errno(r, "Cannot assign environment variable %s: %m", optarg);
+
+ arg_settings_mask |= SETTING_ENVIRONMENT;
+ break;
+
+ case 'q':
+ arg_quiet = true;
+ break;
+
+ case ARG_SHARE_SYSTEM:
+ /* We don't officially support this anymore, except for compat reasons. People should use the
+ * $SYSTEMD_NSPAWN_SHARE_* environment variables instead. */
+ log_warning("Please do not use --share-system anymore, use $SYSTEMD_NSPAWN_SHARE_* instead.");
+ arg_clone_ns_flags = 0;
+ break;
+
+ case ARG_REGISTER:
+ r = parse_boolean(optarg);
+ if (r < 0) {
+ log_error("Failed to parse --register= argument: %s", optarg);
+ return r;
+ }
+
+ arg_register = r;
+ break;
+
+ case ARG_KEEP_UNIT:
+ arg_keep_unit = true;
+ break;
+
+ case ARG_PERSONALITY:
+
+ arg_personality = personality_from_string(optarg);
+ if (arg_personality == PERSONALITY_INVALID)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Unknown or unsupported personality '%s'.", optarg);
+
+ arg_settings_mask |= SETTING_PERSONALITY;
+ break;
+
+ case ARG_VOLATILE:
+
+ if (!optarg)
+ arg_volatile_mode = VOLATILE_YES;
+ else if (streq(optarg, "help")) {
+ DUMP_STRING_TABLE(volatile_mode, VolatileMode, _VOLATILE_MODE_MAX);
+ return 0;
+ } else {
+ VolatileMode m;
+
+ m = volatile_mode_from_string(optarg);
+ if (m < 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Failed to parse --volatile= argument: %s", optarg);
+ else
+ arg_volatile_mode = m;
+ }
+
+ arg_settings_mask |= SETTING_VOLATILE_MODE;
+ break;
+
+ case 'p':
+ r = expose_port_parse(&arg_expose_ports, optarg);
+ if (r == -EEXIST)
+ return log_error_errno(r, "Duplicate port specification: %s", optarg);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse host port %s: %m", optarg);
+
+ arg_settings_mask |= SETTING_EXPOSE_PORTS;
+ break;
+
+ case ARG_PROPERTY:
+ if (strv_extend(&arg_property, optarg) < 0)
+ return log_oom();
+
+ break;
+
+ case ARG_PRIVATE_USERS: {
+ int boolean;
+
+ if (!optarg)
+ boolean = true;
+ else if (!in_charset(optarg, DIGITS))
+ /* do *not* parse numbers as booleans */
+ boolean = parse_boolean(optarg);
+ else
+ boolean = -1;
+
+ if (boolean == 0) {
+ /* no: User namespacing off */
+ arg_userns_mode = USER_NAMESPACE_NO;
+ arg_uid_shift = UID_INVALID;
+ arg_uid_range = UINT32_C(0x10000);
+ } else if (boolean > 0) {
+ /* yes: User namespacing on, UID range is read from root dir */
+ arg_userns_mode = USER_NAMESPACE_FIXED;
+ arg_uid_shift = UID_INVALID;
+ arg_uid_range = UINT32_C(0x10000);
+ } else if (streq(optarg, "pick")) {
+ /* pick: User namespacing on, UID range is picked randomly */
+ arg_userns_mode = USER_NAMESPACE_PICK; /* Note that arg_userns_ownership is
+ * implied by USER_NAMESPACE_PICK
+ * further down. */
+ arg_uid_shift = UID_INVALID;
+ arg_uid_range = UINT32_C(0x10000);
+
+ } else if (streq(optarg, "identity")) {
+ /* identitiy: User namespaces on, UID range is map the 0…0xFFFF range to
+ * itself, i.e. we don't actually map anything, but do take benefit of
+ * isolation of capability sets. */
+ arg_userns_mode = USER_NAMESPACE_FIXED;
+ arg_uid_shift = 0;
+ arg_uid_range = UINT32_C(0x10000);
+ } else {
+ _cleanup_free_ char *buffer = NULL;
+ const char *range, *shift;
+
+ /* anything else: User namespacing on, UID range is explicitly configured */
+
+ range = strchr(optarg, ':');
+ if (range) {
+ buffer = strndup(optarg, range - optarg);
+ if (!buffer)
+ return log_oom();
+ shift = buffer;
+
+ range++;
+ r = safe_atou32(range, &arg_uid_range);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse UID range \"%s\": %m", range);
+ } else
+ shift = optarg;
+
+ r = parse_uid(shift, &arg_uid_shift);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse UID \"%s\": %m", optarg);
+
+ arg_userns_mode = USER_NAMESPACE_FIXED;
+
+ if (!userns_shift_range_valid(arg_uid_shift, arg_uid_range))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID range cannot be empty or go beyond " UID_FMT ".", UID_INVALID);
+ }
+
+ arg_settings_mask |= SETTING_USERNS;
+ break;
+ }
+
+ case 'U':
+ if (userns_supported()) {
+ arg_userns_mode = USER_NAMESPACE_PICK; /* Note that arg_userns_ownership is
+ * implied by USER_NAMESPACE_PICK
+ * further down. */
+ arg_uid_shift = UID_INVALID;
+ arg_uid_range = UINT32_C(0x10000);
+
+ arg_settings_mask |= SETTING_USERNS;
+ }
+
+ break;
+
+ case ARG_PRIVATE_USERS_CHOWN:
+ arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
+
+ arg_settings_mask |= SETTING_USERNS;
+ break;
+
+ case ARG_PRIVATE_USERS_OWNERSHIP:
+ if (streq(optarg, "help")) {
+ DUMP_STRING_TABLE(user_namespace_ownership, UserNamespaceOwnership, _USER_NAMESPACE_OWNERSHIP_MAX);
+ return 0;
+ }
+
+ arg_userns_ownership = user_namespace_ownership_from_string(optarg);
+ if (arg_userns_ownership < 0)
+ return log_error_errno(arg_userns_ownership, "Cannot parse --user-namespace-ownership= value: %s", optarg);
+
+ arg_settings_mask |= SETTING_USERNS;
+ break;
+
+ case ARG_KILL_SIGNAL:
+ if (streq(optarg, "help")) {
+ DUMP_STRING_TABLE(signal, int, _NSIG);
+ return 0;
+ }
+
+ arg_kill_signal = signal_from_string(optarg);
+ if (arg_kill_signal < 0)
+ return log_error_errno(arg_kill_signal, "Cannot parse signal: %s", optarg);
+
+ arg_settings_mask |= SETTING_KILL_SIGNAL;
+ break;
+
+ case ARG_SETTINGS:
+
+ /* no → do not read files
+ * yes → read files, do not override cmdline, trust only subset
+ * override → read files, override cmdline, trust only subset
+ * trusted → read files, do not override cmdline, trust all
+ */
+
+ r = parse_boolean(optarg);
+ if (r < 0) {
+ if (streq(optarg, "trusted")) {
+ mask_all_settings = false;
+ mask_no_settings = false;
+ arg_settings_trusted = true;
+
+ } else if (streq(optarg, "override")) {
+ mask_all_settings = false;
+ mask_no_settings = true;
+ arg_settings_trusted = -1;
+ } else
+ return log_error_errno(r, "Failed to parse --settings= argument: %s", optarg);
+ } else if (r > 0) {
+ /* yes */
+ mask_all_settings = false;
+ mask_no_settings = false;
+ arg_settings_trusted = -1;
+ } else {
+ /* no */
+ mask_all_settings = true;
+ mask_no_settings = false;
+ arg_settings_trusted = false;
+ }
+
+ break;
+
+ case ARG_CHDIR:
+ if (!path_is_absolute(optarg))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Working directory %s is not an absolute path.", optarg);
+
+ r = free_and_strdup(&arg_chdir, optarg);
+ if (r < 0)
+ return log_oom();
+
+ arg_settings_mask |= SETTING_WORKING_DIRECTORY;
+ break;
+
+ case ARG_PIVOT_ROOT:
+ r = pivot_root_parse(&arg_pivot_root_new, &arg_pivot_root_old, optarg);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --pivot-root= argument %s: %m", optarg);
+
+ arg_settings_mask |= SETTING_PIVOT_ROOT;
+ break;
+
+ case ARG_NOTIFY_READY:
+ r = parse_boolean(optarg);
+ if (r < 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "%s is not a valid notify mode. Valid modes are: yes, no, and ready.", optarg);
+ arg_notify_ready = r;
+ arg_settings_mask |= SETTING_NOTIFY_READY;
+ break;
+
+ case ARG_ROOT_HASH: {
+ _cleanup_free_ void *k = NULL;
+ size_t l;
+
+ r = unhexmem(optarg, strlen(optarg), &k, &l);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse root hash: %s", optarg);
+ if (l < sizeof(sd_id128_t))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Root hash must be at least 128bit long: %s", optarg);
+
+ free_and_replace(arg_verity_settings.root_hash, k);
+ arg_verity_settings.root_hash_size = l;
+ break;
+ }
+
+ case ARG_ROOT_HASH_SIG: {
+ char *value;
+ size_t l;
+ void *p;
+
+ if ((value = startswith(optarg, "base64:"))) {
+ r = unbase64mem(value, strlen(value), &p, &l);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse root hash signature '%s': %m", optarg);
+
+ } else {
+ r = read_full_file(optarg, (char**) &p, &l);
+ if (r < 0)
+ return log_error_errno(r, "Failed parse root hash signature file '%s': %m", optarg);
+ }
+
+ free_and_replace(arg_verity_settings.root_hash_sig, p);
+ arg_verity_settings.root_hash_sig_size = l;
+ break;
+ }
+
+ case ARG_VERITY_DATA:
+ r = parse_path_argument(optarg, false, &arg_verity_settings.data_path);
+ if (r < 0)
+ return r;
+ break;
+
+ case ARG_SYSTEM_CALL_FILTER: {
+ bool negative;
+ const char *items;
+
+ negative = optarg[0] == '~';
+ items = negative ? optarg + 1 : optarg;
+
+ for (;;) {
+ _cleanup_free_ char *word = NULL;
+
+ r = extract_first_word(&items, &word, NULL, 0);
+ if (r == 0)
+ break;
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse system call filter: %m");
+
+ if (negative)
+ r = strv_extend(&arg_syscall_deny_list, word);
+ else
+ r = strv_extend(&arg_syscall_allow_list, word);
+ if (r < 0)
+ return log_oom();
+ }
+
+ arg_settings_mask |= SETTING_SYSCALL_FILTER;
+ break;
+ }
+
+ case ARG_RLIMIT: {
+ const char *eq;
+ _cleanup_free_ char *name = NULL;
+ int rl;
+
+ if (streq(optarg, "help")) {
+ DUMP_STRING_TABLE(rlimit, int, _RLIMIT_MAX);
+ return 0;
+ }
+
+ eq = strchr(optarg, '=');
+ if (!eq)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "--rlimit= expects an '=' assignment.");
+
+ name = strndup(optarg, eq - optarg);
+ if (!name)
+ return log_oom();
+
+ rl = rlimit_from_string_harder(name);
+ if (rl < 0)
+ return log_error_errno(rl, "Unknown resource limit: %s", name);
+
+ if (!arg_rlimit[rl]) {
+ arg_rlimit[rl] = new0(struct rlimit, 1);
+ if (!arg_rlimit[rl])
+ return log_oom();
+ }
+
+ r = rlimit_parse(rl, eq + 1, arg_rlimit[rl]);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse resource limit: %s", eq + 1);
+
+ arg_settings_mask |= SETTING_RLIMIT_FIRST << rl;
+ break;
+ }
+
+ case ARG_OOM_SCORE_ADJUST:
+ r = parse_oom_score_adjust(optarg, &arg_oom_score_adjust);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --oom-score-adjust= parameter: %s", optarg);
+
+ arg_oom_score_adjust_set = true;
+ arg_settings_mask |= SETTING_OOM_SCORE_ADJUST;
+ break;
+
+ case ARG_CPU_AFFINITY: {
+ CPUSet cpuset;
+
+ r = parse_cpu_set(optarg, &cpuset);
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse CPU affinity mask %s: %m", optarg);
+
+ cpu_set_reset(&arg_cpu_set);
+ arg_cpu_set = cpuset;
+ arg_settings_mask |= SETTING_CPU_AFFINITY;
+ break;
+ }
+
+ case ARG_RESOLV_CONF:
+ if (streq(optarg, "help")) {
+ DUMP_STRING_TABLE(resolv_conf_mode, ResolvConfMode, _RESOLV_CONF_MODE_MAX);
+ return 0;
+ }
+
+ arg_resolv_conf = resolv_conf_mode_from_string(optarg);
+ if (arg_resolv_conf < 0)
+ return log_error_errno(arg_resolv_conf,
+ "Failed to parse /etc/resolv.conf mode: %s", optarg);
+
+ arg_settings_mask |= SETTING_RESOLV_CONF;
+ break;
+
+ case ARG_TIMEZONE:
+ if (streq(optarg, "help")) {
+ DUMP_STRING_TABLE(timezone_mode, TimezoneMode, _TIMEZONE_MODE_MAX);
+ return 0;
+ }
+
+ arg_timezone = timezone_mode_from_string(optarg);
+ if (arg_timezone < 0)
+ return log_error_errno(arg_timezone,
+ "Failed to parse /etc/localtime mode: %s", optarg);
+
+ arg_settings_mask |= SETTING_TIMEZONE;
+ break;
+
+ case ARG_CONSOLE:
+ r = handle_arg_console(optarg);
+ if (r <= 0)
+ return r;
+ break;
+
+ case 'P':
+ case ARG_PIPE:
+ r = handle_arg_console("pipe");
+ if (r <= 0)
+ return r;
+ break;
+
+ case ARG_NO_PAGER:
+ arg_pager_flags |= PAGER_DISABLE;
+ break;
+
+ case ARG_SET_CREDENTIAL: {
+ _cleanup_free_ char *word = NULL, *data = NULL;
+ const char *p = optarg;
+ Credential *a;
+ ssize_t l;
+
+ r = extract_first_word(&p, &word, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --set-credential= parameter: %m");
+ if (r == 0 || !p)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Missing value for --set-credential=: %s", optarg);
+
+ if (!credential_name_valid(word))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential name is not valid: %s", word);
+
+ for (size_t i = 0; i < arg_n_credentials; i++)
+ if (streq(arg_credentials[i].id, word))
+ return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Duplicate credential '%s', refusing.", word);
+
+ l = cunescape(p, UNESCAPE_ACCEPT_NUL, &data);
+ if (l < 0)
+ return log_error_errno(l, "Failed to unescape credential data: %s", p);
+
+ a = reallocarray(arg_credentials, arg_n_credentials + 1, sizeof(Credential));
+ if (!a)
+ return log_oom();
+
+ a[arg_n_credentials++] = (Credential) {
+ .id = TAKE_PTR(word),
+ .data = TAKE_PTR(data),
+ .size = l,
+ };
+
+ arg_credentials = a;
+
+ arg_settings_mask |= SETTING_CREDENTIALS;
+ break;
+ }
+
+ case ARG_LOAD_CREDENTIAL: {
+ ReadFullFileFlags flags = READ_FULL_FILE_SECURE;
+ _cleanup_(erase_and_freep) char *data = NULL;
+ _cleanup_free_ char *word = NULL, *j = NULL;
+ const char *p = optarg;
+ Credential *a;
+ size_t size, i;
+
+ r = extract_first_word(&p, &word, ":", EXTRACT_DONT_COALESCE_SEPARATORS);
+ if (r == -ENOMEM)
+ return log_oom();
+ if (r < 0)
+ return log_error_errno(r, "Failed to parse --load-credential= parameter: %m");
+ if (r == 0 || !p)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Missing value for --load-credential=: %s", optarg);
+
+ if (!credential_name_valid(word))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Credential name is not valid: %s", word);
+
+ for (i = 0; i < arg_n_credentials; i++)
+ if (streq(arg_credentials[i].id, word))
+ return log_error_errno(SYNTHETIC_ERRNO(EEXIST), "Duplicate credential '%s', refusing.", word);
+
+ if (path_is_absolute(p))
+ flags |= READ_FULL_FILE_CONNECT_SOCKET;
+ else {
+ const char *e;
+
+ r = get_credentials_dir(&e);
+ if (r < 0)
+ return log_error_errno(r, "Credential not available (no credentials passed at all): %s", word);
+
+ j = path_join(e, p);
+ if (!j)
+ return log_oom();
+ }
+
+ r = read_full_file_full(AT_FDCWD, j ?: p, UINT64_MAX, SIZE_MAX,
+ flags,
+ NULL,
+ &data, &size);
+ if (r < 0)
+ return log_error_errno(r, "Failed to read credential '%s': %m", j ?: p);
+
+ a = reallocarray(arg_credentials, arg_n_credentials + 1, sizeof(Credential));
+ if (!a)
+ return log_oom();
+
+ a[arg_n_credentials++] = (Credential) {
+ .id = TAKE_PTR(word),
+ .data = TAKE_PTR(data),
+ .size = size,
+ };
+
+ arg_credentials = a;
+
+ arg_settings_mask |= SETTING_CREDENTIALS;
+ break;
+ }
+
+ case ARG_BIND_USER:
+ if (!valid_user_group_name(optarg, 0))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid user name to bind: %s", optarg);
+
+ if (strv_extend(&arg_bind_user, optarg) < 0)
+ return log_oom();
+
+ arg_settings_mask |= SETTING_BIND_USER;
+ break;
+
+ case ARG_SUPPRESS_SYNC:
+ r = parse_boolean_argument("--suppress-sync=", optarg, &arg_suppress_sync);
+ if (r < 0)
+ return r;
+
+ arg_settings_mask |= SETTING_SUPPRESS_SYNC;
+ break;
+
+ case '?':
+ return -EINVAL;
+
+ default:
+ assert_not_reached();
+ }
+
+ if (argc > optind) {
+ strv_free(arg_parameters);
+ arg_parameters = strv_copy(argv + optind);
+ if (!arg_parameters)
+ return log_oom();
+
+ arg_settings_mask |= SETTING_START_MODE;
+ }
+
+ if (arg_ephemeral && arg_template && !arg_directory)
+ /* User asked for ephemeral execution but specified --template= instead of --directory=. Semantically
+ * such an invocation makes some sense, see https://github.com/systemd/systemd/issues/3667. Let's
+ * accept this here, and silently make "--ephemeral --template=" equivalent to "--ephemeral
+ * --directory=". */
+ arg_directory = TAKE_PTR(arg_template);
+
+ arg_caps_retain |= plus;
+ arg_caps_retain |= arg_private_network ? UINT64_C(1) << CAP_NET_ADMIN : 0;
+ arg_caps_retain &= ~minus;
+
+ /* Make sure to parse environment before we reset the settings mask below */
+ r = parse_environment();
+ if (r < 0)
+ return r;
+
+ /* Load all settings from .nspawn files */
+ if (mask_no_settings)
+ arg_settings_mask = 0;
+
+ /* Don't load any settings from .nspawn files */
+ if (mask_all_settings)
+ arg_settings_mask = _SETTINGS_MASK_ALL;
+
+ return 1;
+}
+
+static int verify_arguments(void) {
+ int r;
+
+ if (arg_start_mode == START_PID2 && arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
+ /* If we are running the stub init in the container, we don't need to look at what the init
+ * in the container supports, because we are not using it. Let's immediately pick the right
+ * setting based on the host system configuration.
+ *
+ * We only do this, if the user didn't use an environment variable to override the detection.
+ */
+
+ r = cg_all_unified();
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine whether we are in all unified mode.");
+ if (r > 0)
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_ALL;
+ else if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0)
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_SYSTEMD;
+ else
+ arg_unified_cgroup_hierarchy = CGROUP_UNIFIED_NONE;
+ }
+
+ if (arg_userns_mode != USER_NAMESPACE_NO)
+ arg_mount_settings |= MOUNT_USE_USERNS;
+
+ if (arg_private_network)
+ arg_mount_settings |= MOUNT_APPLY_APIVFS_NETNS;
+
+ if (!(arg_clone_ns_flags & CLONE_NEWPID) ||
+ !(arg_clone_ns_flags & CLONE_NEWUTS)) {
+ arg_register = false;
+ if (arg_start_mode != START_PID1)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--boot cannot be used without namespacing.");
+ }
+
+ if (arg_userns_ownership < 0)
+ arg_userns_ownership =
+ arg_userns_mode == USER_NAMESPACE_PICK ? USER_NAMESPACE_OWNERSHIP_AUTO :
+ USER_NAMESPACE_OWNERSHIP_OFF;
+
+ if (arg_start_mode == START_BOOT && arg_kill_signal <= 0)
+ arg_kill_signal = SIGRTMIN+3;
+
+ if (arg_volatile_mode != VOLATILE_NO) /* Make sure all file systems contained in the image are mounted read-only if we are in volatile mode */
+ arg_read_only = true;
+
+ if (has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts))
+ arg_read_only = true;
+
+ if (arg_keep_unit && arg_register && cg_pid_get_owner_uid(0, NULL) >= 0)
+ /* Save the user from accidentally registering either user-$SESSION.scope or user@.service.
+ * The latter is not technically a user session, but we don't need to labour the point. */
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--keep-unit --register=yes may not be used when invoked from a user session.");
+
+ if (arg_directory && arg_image)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--directory= and --image= may not be combined.");
+
+ if (arg_template && arg_image)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= and --image= may not be combined.");
+
+ if (arg_template && !(arg_directory || arg_machine))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--template= needs --directory= or --machine=.");
+
+ if (arg_ephemeral && arg_template)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --template= may not be combined.");
+
+ if (arg_ephemeral && !IN_SET(arg_link_journal, LINK_NO, LINK_AUTO))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--ephemeral and --link-journal= may not be combined.");
+
+ if (arg_userns_mode != USER_NAMESPACE_NO && !userns_supported())
+ return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "--private-users= is not supported, kernel compiled without user namespace support.");
+
+ if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_CHOWN && arg_read_only)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "--read-only and --private-users-ownership=chown may not be combined.");
+
+ /* We don't support --private-users-ownership=chown together with any of the volatile modes since we
+ * couldn't change the read-only part of the tree (i.e. /usr) anyway, or because it would trigger a
+ * massive copy-up (in case of overlay) making the entire exercise pointless. */
+ if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_CHOWN && arg_volatile_mode != VOLATILE_NO)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--volatile= and --private-users-ownership=chown may not be combined.");
+
+ /* If --network-namespace-path is given with any other network-related option (except --private-network),
+ * we need to error out, to avoid conflicts between different network options. */
+ if (arg_network_namespace_path &&
+ (arg_network_interfaces || arg_network_macvlan ||
+ arg_network_ipvlan || arg_network_veth_extra ||
+ arg_network_bridge || arg_network_zone ||
+ arg_network_veth))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--network-namespace-path= cannot be combined with other network options.");
+
+ if (arg_network_bridge && arg_network_zone)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "--network-bridge= and --network-zone= may not be combined.");
+
+ if (arg_userns_mode != USER_NAMESPACE_NO && (arg_mount_settings & MOUNT_APPLY_APIVFS_NETNS) && !arg_private_network)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Invalid namespacing settings. Mounting sysfs with --private-users requires --private-network.");
+
+ if (arg_userns_mode != USER_NAMESPACE_NO && !(arg_mount_settings & MOUNT_APPLY_APIVFS_RO))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot combine --private-users with read-write mounts.");
+
+ if (arg_expose_ports && !arg_private_network)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Cannot use --port= without private networking.");
+
+ if (arg_caps_ambient) {
+ if (arg_caps_ambient == UINT64_MAX)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= does not support the value all.");
+
+ if ((arg_caps_ambient & arg_caps_retain) != arg_caps_ambient)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not fully covered by Capability= setting.");
+
+ if (arg_start_mode == START_BOOT)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "AmbientCapability= setting is not useful for boot mode.");
+ }
+
+ if (arg_userns_mode == USER_NAMESPACE_NO && !strv_isempty(arg_bind_user))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "--bind-user= requires --private-users");
+
+ /* Drop duplicate --bind-user= entries */
+ strv_uniq(arg_bind_user);
+
+ r = custom_mount_check_all();
+ if (r < 0)
+ return r;
+
+ return 0;
+}
+
+int userns_lchown(const char *p, uid_t uid, gid_t gid) {
+ assert(p);
+
+ if (arg_userns_mode == USER_NAMESPACE_NO)
+ return 0;
+
+ if (uid == UID_INVALID && gid == GID_INVALID)
+ return 0;
+
+ if (uid != UID_INVALID) {
+ uid += arg_uid_shift;
+
+ if (uid < arg_uid_shift || uid >= arg_uid_shift + arg_uid_range)
+ return -EOVERFLOW;
+ }
+
+ if (gid != GID_INVALID) {
+ gid += (gid_t) arg_uid_shift;
+
+ if (gid < (gid_t) arg_uid_shift || gid >= (gid_t) (arg_uid_shift + arg_uid_range))
+ return -EOVERFLOW;
+ }
+
+ return RET_NERRNO(lchown(p, uid, gid));
+}
+
+int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid) {
+ const char *q;
+ int r;
+
+ q = prefix_roota(root, path);
+ r = RET_NERRNO(mkdir(q, mode));
+ if (r == -EEXIST)
+ return 0;
+ if (r < 0)
+ return r;
+
+ return userns_lchown(q, uid, gid);
+}
+
+static const char *timezone_from_path(const char *path) {
+ return PATH_STARTSWITH_SET(
+ path,
+ "../usr/share/zoneinfo/",
+ "/usr/share/zoneinfo/");
+}
+
+static bool etc_writable(void) {
+ return !arg_read_only || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_OVERLAY);
+}
+
+static int setup_timezone(const char *dest) {
+ _cleanup_free_ char *p = NULL, *etc = NULL;
+ const char *where, *check;
+ TimezoneMode m;
+ int r;
+
+ assert(dest);
+
+ if (IN_SET(arg_timezone, TIMEZONE_AUTO, TIMEZONE_SYMLINK)) {
+ r = readlink_malloc("/etc/localtime", &p);
+ if (r == -ENOENT && arg_timezone == TIMEZONE_AUTO)
+ m = etc_writable() ? TIMEZONE_DELETE : TIMEZONE_OFF;
+ else if (r == -EINVAL && arg_timezone == TIMEZONE_AUTO) /* regular file? */
+ m = etc_writable() ? TIMEZONE_COPY : TIMEZONE_BIND;
+ else if (r < 0) {
+ log_warning_errno(r, "Failed to read host's /etc/localtime symlink, not updating container timezone: %m");
+ /* To handle warning, delete /etc/localtime and replace it with a symbolic link to a time zone data
+ * file.
+ *
+ * Example:
+ * ln -s /usr/share/zoneinfo/UTC /etc/localtime
+ */
+ return 0;
+ } else if (arg_timezone == TIMEZONE_AUTO)
+ m = etc_writable() ? TIMEZONE_SYMLINK : TIMEZONE_BIND;
+ else
+ m = arg_timezone;
+ } else
+ m = arg_timezone;
+
+ if (m == TIMEZONE_OFF)
+ return 0;
+
+ r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
+ if (r < 0) {
+ log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
+ return 0;
+ }
+
+ where = strjoina(etc, "/localtime");
+
+ switch (m) {
+
+ case TIMEZONE_DELETE:
+ if (unlink(where) < 0)
+ log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
+
+ return 0;
+
+ case TIMEZONE_SYMLINK: {
+ _cleanup_free_ char *q = NULL;
+ const char *z, *what;
+
+ z = timezone_from_path(p);
+ if (!z) {
+ log_warning("/etc/localtime does not point into /usr/share/zoneinfo/, not updating container timezone.");
+ return 0;
+ }
+
+ r = readlink_malloc(where, &q);
+ if (r >= 0 && streq_ptr(timezone_from_path(q), z))
+ return 0; /* Already pointing to the right place? Then do nothing .. */
+
+ check = strjoina(dest, "/usr/share/zoneinfo/", z);
+ r = chase_symlinks(check, dest, 0, NULL, NULL);
+ if (r < 0)
+ log_debug_errno(r, "Timezone %s does not exist (or is not accessible) in container, not creating symlink: %m", z);
+ else {
+ if (unlink(where) < 0 && errno != ENOENT) {
+ log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING, /* Don't complain on read-only images */
+ errno, "Failed to remove existing timezone info %s in container, ignoring: %m", where);
+ return 0;
+ }
+
+ what = strjoina("../usr/share/zoneinfo/", z);
+ if (symlink(what, where) < 0) {
+ log_full_errno(IN_SET(errno, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING,
+ errno, "Failed to correct timezone of container, ignoring: %m");
+ return 0;
+ }
+
+ break;
+ }
+
+ _fallthrough_;
+ }
+
+ case TIMEZONE_BIND: {
+ _cleanup_free_ char *resolved = NULL;
+ int found;
+
+ found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
+ if (found < 0) {
+ log_warning_errno(found, "Failed to resolve /etc/localtime path in container, ignoring: %m");
+ return 0;
+ }
+
+ if (found == 0) /* missing? */
+ (void) touch(resolved);
+
+ r = mount_nofollow_verbose(LOG_WARNING, "/etc/localtime", resolved, NULL, MS_BIND, NULL);
+ if (r >= 0)
+ return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
+
+ _fallthrough_;
+ }
+
+ case TIMEZONE_COPY:
+ /* If mounting failed, try to copy */
+ r = copy_file_atomic("/etc/localtime", where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
+ if (r < 0) {
+ log_full_errno(IN_SET(r, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
+ "Failed to copy /etc/localtime to %s, ignoring: %m", where);
+ return 0;
+ }
+
+ break;
+
+ default:
+ assert_not_reached();
+ }
+
+ /* Fix permissions of the symlink or file copy we just created */
+ r = userns_lchown(where, 0, 0);
+ if (r < 0)
+ log_warning_errno(r, "Failed to chown /etc/localtime, ignoring: %m");
+
+ return 0;
+}
+
+static int have_resolv_conf(const char *path) {
+ assert(path);
+
+ if (access(path, F_OK) < 0) {
+ if (errno == ENOENT)
+ return 0;
+
+ return log_debug_errno(errno, "Failed to determine whether '%s' is available: %m", path);
+ }
+
+ return 1;
+}
+
+static int resolved_listening(void) {
+ _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
+ _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
+ _cleanup_free_ char *dns_stub_listener_mode = NULL;
+ int r;
+
+ /* Check if resolved is listening */
+
+ r = sd_bus_open_system(&bus);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to open system bus: %m");
+
+ r = bus_name_has_owner(bus, "org.freedesktop.resolve1", NULL);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to check whether the 'org.freedesktop.resolve1' bus name is taken: %m");
+ if (r == 0)
+ return 0;
+
+ r = sd_bus_get_property_string(bus,
+ "org.freedesktop.resolve1",
+ "/org/freedesktop/resolve1",
+ "org.freedesktop.resolve1.Manager",
+ "DNSStubListener",
+ &error,
+ &dns_stub_listener_mode);
+ if (r < 0)
+ return log_debug_errno(r, "Failed to query DNSStubListener property: %s", bus_error_message(&error, r));
+
+ return STR_IN_SET(dns_stub_listener_mode, "udp", "yes");
+}
+
+static int setup_resolv_conf(const char *dest) {
+ _cleanup_free_ char *etc = NULL;
+ const char *where, *what;
+ ResolvConfMode m;
+ int r;
+
+ assert(dest);
+
+ if (arg_resolv_conf == RESOLV_CONF_AUTO) {
+ if (arg_private_network)
+ m = RESOLV_CONF_OFF;
+ else if (have_resolv_conf(PRIVATE_STUB_RESOLV_CONF) > 0 && resolved_listening() > 0)
+ m = etc_writable() ? RESOLV_CONF_COPY_STUB : RESOLV_CONF_BIND_STUB;
+ else if (have_resolv_conf("/etc/resolv.conf") > 0)
+ m = etc_writable() ? RESOLV_CONF_COPY_HOST : RESOLV_CONF_BIND_HOST;
+ else
+ m = etc_writable() ? RESOLV_CONF_DELETE : RESOLV_CONF_OFF;
+
+ } else
+ m = arg_resolv_conf;
+
+ if (m == RESOLV_CONF_OFF)
+ return 0;
+
+ r = chase_symlinks("/etc", dest, CHASE_PREFIX_ROOT, &etc, NULL);
+ if (r < 0) {
+ log_warning_errno(r, "Failed to resolve /etc path in container, ignoring: %m");
+ return 0;
+ }
+
+ where = strjoina(etc, "/resolv.conf");
+
+ if (m == RESOLV_CONF_DELETE) {
+ if (unlink(where) < 0)
+ log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno, "Failed to remove '%s', ignoring: %m", where);
+
+ return 0;
+ }
+
+ if (IN_SET(m, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_COPY_STATIC))
+ what = PRIVATE_STATIC_RESOLV_CONF;
+ else if (IN_SET(m, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_COPY_UPLINK))
+ what = PRIVATE_UPLINK_RESOLV_CONF;
+ else if (IN_SET(m, RESOLV_CONF_BIND_STUB, RESOLV_CONF_REPLACE_STUB, RESOLV_CONF_COPY_STUB))
+ what = PRIVATE_STUB_RESOLV_CONF;
+ else
+ what = "/etc/resolv.conf";
+
+ if (IN_SET(m, RESOLV_CONF_BIND_HOST, RESOLV_CONF_BIND_STATIC, RESOLV_CONF_BIND_UPLINK, RESOLV_CONF_BIND_STUB)) {
+ _cleanup_free_ char *resolved = NULL;
+ int found;
+
+ found = chase_symlinks(where, dest, CHASE_NONEXISTENT, &resolved, NULL);
+ if (found < 0) {
+ log_warning_errno(found, "Failed to resolve /etc/resolv.conf path in container, ignoring: %m");
+ return 0;
+ }
+
+ if (found == 0) /* missing? */
+ (void) touch(resolved);
+
+ r = mount_nofollow_verbose(LOG_WARNING, what, resolved, NULL, MS_BIND, NULL);
+ if (r >= 0)
+ return mount_nofollow_verbose(LOG_ERR, NULL, resolved, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NODEV, NULL);
+
+ /* If that didn't work, let's copy the file */
+ }
+
+ if (IN_SET(m, RESOLV_CONF_REPLACE_HOST, RESOLV_CONF_REPLACE_STATIC, RESOLV_CONF_REPLACE_UPLINK, RESOLV_CONF_REPLACE_STUB))
+ r = copy_file_atomic(what, where, 0644, 0, 0, COPY_REFLINK|COPY_REPLACE);
+ else
+ r = copy_file(what, where, O_TRUNC|O_NOFOLLOW, 0644, 0, 0, COPY_REFLINK);
+ if (r < 0) {
+ /* If the file already exists as symlink, let's suppress the warning, under the assumption that
+ * resolved or something similar runs inside and the symlink points there.
+ *
+ * If the disk image is read-only, there's also no point in complaining.
+ */
+ log_full_errno(!IN_SET(RESOLV_CONF_COPY_HOST, RESOLV_CONF_COPY_STATIC, RESOLV_CONF_COPY_UPLINK, RESOLV_CONF_COPY_STUB) &&
+ IN_SET(r, -ELOOP, -EROFS, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r,
+ "Failed to copy /etc/resolv.conf to %s, ignoring: %m", where);
+ return 0;
+ }
+
+ r = userns_lchown(where, 0, 0);
+ if (r < 0)
+ log_warning_errno(r, "Failed to chown /etc/resolv.conf, ignoring: %m");
+
+ return 0;
+}
+
+static int setup_boot_id(void) {
+ _cleanup_(unlink_and_freep) char *from = NULL;
+ _cleanup_free_ char *path = NULL;
+ sd_id128_t rnd = SD_ID128_NULL;
+ const char *to;
+ int r;
+
+ /* Generate a new randomized boot ID, so that each boot-up of the container gets a new one */
+
+ r = tempfn_random_child("/run", "proc-sys-kernel-random-boot-id", &path);
+ if (r < 0)
+ return log_error_errno(r, "Failed to generate random boot ID path: %m");
+
+ r = sd_id128_randomize(&rnd);
+ if (r < 0)
+ return log_error_errno(r, "Failed to generate random boot id: %m");
+
+ r = id128_write(path, ID128_FORMAT_UUID, rnd, false);
+ if (r < 0)
+ return log_error_errno(r, "Failed to write boot id: %m");
+
+ from = TAKE_PTR(path);
+ to = "/proc/sys/kernel/random/boot_id";
+
+ r = mount_nofollow_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL);
+ if (r < 0)
+ return r;
+
+ return mount_nofollow_verbose(LOG_ERR, NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
+}
+
+static int copy_devnodes(const char *dest) {
+ static const char devnodes[] =
+ "null\0"
+ "zero\0"
+ "full\0"
+ "random\0"
+ "urandom\0"
+ "tty\0"
+ "net/tun\0";
+
+ const char *d;
+ int r = 0;
+
+ assert(dest);
+
+ BLOCK_WITH_UMASK(0000);
+
+ /* Create /dev/net, so that we can create /dev/net/tun in it */
+ if (userns_mkdir(dest, "/dev/net", 0755, 0, 0) < 0)
+ return log_error_errno(r, "Failed to create /dev/net directory: %m");
+
+ NULSTR_FOREACH(d, devnodes) {
+ _cleanup_free_ char *from = NULL, *to = NULL;
+ struct stat st;
+
+ from = path_join("/dev/", d);
+ if (!from)
+ return log_oom();
+
+ to = path_join(dest, from);
+ if (!to)
+ return log_oom();
+
+ if (stat(from, &st) < 0) {
+
+ if (errno != ENOENT)
+ return log_error_errno(errno, "Failed to stat %s: %m", from);
+
+ } else if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "%s is not a char or block device, cannot copy.", from);
+ else {
+ _cleanup_free_ char *sl = NULL, *prefixed = NULL, *dn = NULL, *t = NULL;
+
+ if (mknod(to, st.st_mode, st.st_rdev) < 0) {
+ /* Explicitly warn the user when /dev is already populated. */
+ if (errno == EEXIST)
+ log_notice("%s/dev is pre-mounted and pre-populated. If a pre-mounted /dev is provided it needs to be an unpopulated file system.", dest);
+ if (errno != EPERM)
+ return log_error_errno(errno, "mknod(%s) failed: %m", to);
+
+ /* Some systems abusively restrict mknod but allow bind mounts. */
+ r = touch(to);
+ if (r < 0)
+ return log_error_errno(r, "touch (%s) failed: %m", to);
+ r = mount_nofollow_verbose(LOG_DEBUG, from, to, NULL, MS_BIND, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Both mknod and bind mount (%s) failed: %m", to);
+ }
+
+ r = userns_lchown(to, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "chown() of device node %s failed: %m", to);
+
+ dn = path_join("/dev", S_ISCHR(st.st_mode) ? "char" : "block");
+ if (!dn)
+ return log_oom();
+
+ r = userns_mkdir(dest, dn, 0755, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create '%s': %m", dn);
+
+ if (asprintf(&sl, "%s/%u:%u", dn, major(st.st_rdev), minor(st.st_rdev)) < 0)
+ return log_oom();
+
+ prefixed = path_join(dest, sl);
+ if (!prefixed)
+ return log_oom();
+
+ t = path_join("..", d);
+ if (!t)
+ return log_oom();
+
+ if (symlink(t, prefixed) < 0)
+ log_debug_errno(errno, "Failed to symlink '%s' to '%s': %m", t, prefixed);
+ }
+ }
+
+ return r;
+}
+
+static int make_extra_nodes(const char *dest) {
+ size_t i;
+ int r;
+
+ BLOCK_WITH_UMASK(0000);
+
+ for (i = 0; i < arg_n_extra_nodes; i++) {
+ _cleanup_free_ char *path = NULL;
+ DeviceNode *n = arg_extra_nodes + i;
+
+ path = path_join(dest, n->path);
+ if (!path)
+ return log_oom();
+
+ if (mknod(path, n->mode, S_ISCHR(n->mode) || S_ISBLK(n->mode) ? makedev(n->major, n->minor) : 0) < 0)
+ return log_error_errno(errno, "Failed to create device node '%s': %m", path);
+
+ r = chmod_and_chown(path, n->mode, n->uid, n->gid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to adjust device node ownership of '%s': %m", path);
+ }
+
+ return 0;
+}
+
+static int setup_pts(const char *dest) {
+ _cleanup_free_ char *options = NULL;
+ const char *p;
+ int r;
+
+#if HAVE_SELINUX
+ if (arg_selinux_apifs_context)
+ (void) asprintf(&options,
+ "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT ",context=\"%s\"",
+ arg_uid_shift + TTY_GID,
+ arg_selinux_apifs_context);
+ else
+#endif
+ (void) asprintf(&options,
+ "newinstance,ptmxmode=0666,mode=620,gid=" GID_FMT,
+ arg_uid_shift + TTY_GID);
+
+ if (!options)
+ return log_oom();
+
+ /* Mount /dev/pts itself */
+ p = prefix_roota(dest, "/dev/pts");
+ r = RET_NERRNO(mkdir(p, 0755));
+ if (r < 0)
+ return log_error_errno(r, "Failed to create /dev/pts: %m");
+
+ r = mount_nofollow_verbose(LOG_ERR, "devpts", p, "devpts", MS_NOSUID|MS_NOEXEC, options);
+ if (r < 0)
+ return r;
+ r = userns_lchown(p, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to chown /dev/pts: %m");
+
+ /* Create /dev/ptmx symlink */
+ p = prefix_roota(dest, "/dev/ptmx");
+ if (symlink("pts/ptmx", p) < 0)
+ return log_error_errno(errno, "Failed to create /dev/ptmx symlink: %m");
+ r = userns_lchown(p, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to chown /dev/ptmx: %m");
+
+ /* And fix /dev/pts/ptmx ownership */
+ p = prefix_roota(dest, "/dev/pts/ptmx");
+ r = userns_lchown(p, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to chown /dev/pts/ptmx: %m");
+
+ return 0;
+}
+
+static int setup_stdio_as_dev_console(void) {
+ _cleanup_close_ int terminal = -1;
+ int r;
+
+ /* We open the TTY in O_NOCTTY mode, so that we do not become controller yet. We'll do that later
+ * explicitly, if we are configured to. */
+ terminal = open_terminal("/dev/console", O_RDWR|O_NOCTTY);
+ if (terminal < 0)
+ return log_error_errno(terminal, "Failed to open console: %m");
+
+ /* Make sure we can continue logging to the original stderr, even if
+ * stderr points elsewhere now */
+ r = log_dup_console();
+ if (r < 0)
+ return log_error_errno(r, "Failed to duplicate stderr: %m");
+
+ /* invalidates 'terminal' on success and failure */
+ r = rearrange_stdio(terminal, terminal, terminal);
+ TAKE_FD(terminal);
+ if (r < 0)
+ return log_error_errno(r, "Failed to move console to stdin/stdout/stderr: %m");
+
+ return 0;
+}
+
+static int setup_dev_console(const char *console) {
+ _cleanup_free_ char *p = NULL;
+ int r;
+
+ /* Create /dev/console symlink */
+ r = path_make_relative("/dev", console, &p);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create relative path: %m");
+
+ if (symlink(p, "/dev/console") < 0)
+ return log_error_errno(errno, "Failed to create /dev/console symlink: %m");
+
+ return 0;
+}
+
+static int setup_keyring(void) {
+ key_serial_t keyring;
+
+ /* Allocate a new session keyring for the container. This makes sure the keyring of the session
+ * systemd-nspawn was invoked from doesn't leak into the container. Note that by default we block
+ * keyctl() and request_key() anyway via seccomp so doing this operation isn't strictly necessary,
+ * but in case people explicitly allow-list these system calls let's make sure we don't leak anything
+ * into the container. */
+
+ keyring = keyctl(KEYCTL_JOIN_SESSION_KEYRING, 0, 0, 0, 0);
+ if (keyring == -1) {
+ if (errno == ENOSYS)
+ log_debug_errno(errno, "Kernel keyring not supported, ignoring.");
+ else if (ERRNO_IS_PRIVILEGE(errno))
+ log_debug_errno(errno, "Kernel keyring access prohibited, ignoring.");
+ else
+ return log_error_errno(errno, "Setting up kernel keyring failed: %m");
+ }
+
+ return 0;
+}
+
+static int setup_credentials(const char *root) {
+ const char *q;
+ int r;
+
+ if (arg_n_credentials <= 0)
+ return 0;
+
+ r = userns_mkdir(root, "/run/host", 0755, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create /run/host: %m");
+
+ r = userns_mkdir(root, "/run/host/credentials", 0700, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create /run/host/credentials: %m");
+
+ q = prefix_roota(root, "/run/host/credentials");
+ r = mount_nofollow_verbose(LOG_ERR, NULL, q, "ramfs", MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0700");
+ if (r < 0)
+ return r;
+
+ for (size_t i = 0; i < arg_n_credentials; i++) {
+ _cleanup_free_ char *j = NULL;
+ _cleanup_close_ int fd = -1;
+
+ j = path_join(q, arg_credentials[i].id);
+ if (!j)
+ return log_oom();
+
+ fd = open(j, O_CREAT|O_EXCL|O_WRONLY|O_CLOEXEC|O_NOFOLLOW, 0600);
+ if (fd < 0)
+ return log_error_errno(errno, "Failed to create credential file %s: %m", j);
+
+ r = loop_write(fd, arg_credentials[i].data, arg_credentials[i].size, /* do_poll= */ false);
+ if (r < 0)
+ return log_error_errno(r, "Failed to write credential to file %s: %m", j);
+
+ if (fchmod(fd, 0400) < 0)
+ return log_error_errno(errno, "Failed to adjust access mode of %s: %m", j);
+
+ if (arg_userns_mode != USER_NAMESPACE_NO) {
+ if (fchown(fd, arg_uid_shift, arg_uid_shift) < 0)
+ return log_error_errno(errno, "Failed to adjust ownership of %s: %m", j);
+ }
+ }
+
+ if (chmod(q, 0500) < 0)
+ return log_error_errno(errno, "Failed to adjust access mode of %s: %m", q);
+
+ r = userns_lchown(q, 0, 0);
+ if (r < 0)
+ return r;
+
+ /* Make both mount and superblock read-only now */
+ r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL);
+ if (r < 0)
+ return r;
+
+ return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_REMOUNT|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, "mode=0500");
+}
+
+static int setup_kmsg(int kmsg_socket) {
+ _cleanup_(unlink_and_freep) char *from = NULL;
+ _cleanup_free_ char *fifo = NULL;
+ _cleanup_close_ int fd = -1;
+ int r;
+
+ assert(kmsg_socket >= 0);
+
+ BLOCK_WITH_UMASK(0000);
+
+ /* We create the kmsg FIFO as as temporary file in /run, but immediately delete it after bind mounting it to
+ * /proc/kmsg. While FIFOs on the reading side behave very similar to /proc/kmsg, their writing side behaves
+ * differently from /dev/kmsg in that writing blocks when nothing is reading. In order to avoid any problems
+ * with containers deadlocking due to this we simply make /dev/kmsg unavailable to the container. */
+
+ r = tempfn_random_child("/run", "proc-kmsg", &fifo);
+ if (r < 0)
+ return log_error_errno(r, "Failed to generate kmsg path: %m");
+
+ if (mkfifo(fifo, 0600) < 0)
+ return log_error_errno(errno, "mkfifo() for /run/kmsg failed: %m");
+
+ from = TAKE_PTR(fifo);
+
+ r = mount_nofollow_verbose(LOG_ERR, from, "/proc/kmsg", NULL, MS_BIND, NULL);
+ if (r < 0)
+ return r;
+
+ fd = open(from, O_RDWR|O_NONBLOCK|O_CLOEXEC);
+ if (fd < 0)
+ return log_error_errno(errno, "Failed to open fifo: %m");
+
+ /* Store away the fd in the socket, so that it stays open as long as we run the child */
+ r = send_one_fd(kmsg_socket, fd, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to send FIFO fd: %m");
+
+ return 0;
+}
+
+struct ExposeArgs {
+ union in_addr_union address4;
+ union in_addr_union address6;
+ struct FirewallContext *fw_ctx;
+};
+
+static int on_address_change(sd_netlink *rtnl, sd_netlink_message *m, void *userdata) {
+ struct ExposeArgs *args = ASSERT_PTR(userdata);
+
+ assert(rtnl);
+ assert(m);
+
+ (void) expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, AF_INET, &args->address4);
+ (void) expose_port_execute(rtnl, &args->fw_ctx, arg_expose_ports, AF_INET6, &args->address6);
+ return 0;
+}
+
+static int setup_hostname(void) {
+ int r;
+
+ if ((arg_clone_ns_flags & CLONE_NEWUTS) == 0)
+ return 0;
+
+ r = sethostname_idempotent(arg_hostname ?: arg_machine);
+ if (r < 0)
+ return log_error_errno(r, "Failed to set hostname: %m");
+
+ return 0;
+}
+
+static int setup_journal(const char *directory) {
+ _cleanup_free_ char *d = NULL;
+ const char *p, *q;
+ sd_id128_t this_id;
+ bool try;
+ int r;
+
+ /* Don't link journals in ephemeral mode */
+ if (arg_ephemeral)
+ return 0;
+
+ if (arg_link_journal == LINK_NO)
+ return 0;
+
+ try = arg_link_journal_try || arg_link_journal == LINK_AUTO;
+
+ r = sd_id128_get_machine(&this_id);
+ if (r < 0)
+ return log_error_errno(r, "Failed to retrieve machine ID: %m");
+
+ if (sd_id128_equal(arg_uuid, this_id)) {
+ log_full(try ? LOG_WARNING : LOG_ERR,
+ "Host and machine ids are equal (%s): refusing to link journals", SD_ID128_TO_STRING(arg_uuid));
+ if (try)
+ return 0;
+ return -EEXIST;
+ }
+
+ FOREACH_STRING(dirname, "/var", "/var/log", "/var/log/journal") {
+ r = userns_mkdir(directory, dirname, 0755, 0, 0);
+ if (r < 0) {
+ bool ignore = r == -EROFS && try;
+ log_full_errno(ignore ? LOG_DEBUG : LOG_ERR, r,
+ "Failed to create %s%s: %m", dirname, ignore ? ", ignoring" : "");
+ return ignore ? 0 : r;
+ }
+ }
+
+ p = strjoina("/var/log/journal/", SD_ID128_TO_STRING(arg_uuid));
+ q = prefix_roota(directory, p);
+
+ if (path_is_mount_point(p, NULL, 0) > 0) {
+ if (try)
+ return 0;
+
+ return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
+ "%s: already a mount point, refusing to use for journal", p);
+ }
+
+ if (path_is_mount_point(q, NULL, 0) > 0) {
+ if (try)
+ return 0;
+
+ return log_error_errno(SYNTHETIC_ERRNO(EEXIST),
+ "%s: already a mount point, refusing to use for journal", q);
+ }
+
+ r = readlink_and_make_absolute(p, &d);
+ if (r >= 0) {
+ if (IN_SET(arg_link_journal, LINK_GUEST, LINK_AUTO) &&
+ path_equal(d, q)) {
+
+ r = userns_mkdir(directory, p, 0755, 0, 0);
+ if (r < 0)
+ log_warning_errno(r, "Failed to create directory %s: %m", q);
+ return 0;
+ }
+
+ if (unlink(p) < 0)
+ return log_error_errno(errno, "Failed to remove symlink %s: %m", p);
+ } else if (r == -EINVAL) {
+
+ if (arg_link_journal == LINK_GUEST &&
+ rmdir(p) < 0) {
+
+ if (errno == ENOTDIR) {
+ log_error("%s already exists and is neither a symlink nor a directory", p);
+ return r;
+ } else
+ return log_error_errno(errno, "Failed to remove %s: %m", p);
+ }
+ } else if (r != -ENOENT)
+ return log_error_errno(r, "readlink(%s) failed: %m", p);
+
+ if (arg_link_journal == LINK_GUEST) {
+
+ if (symlink(q, p) < 0) {
+ if (try) {
+ log_debug_errno(errno, "Failed to symlink %s to %s, skipping journal setup: %m", q, p);
+ return 0;
+ } else
+ return log_error_errno(errno, "Failed to symlink %s to %s: %m", q, p);
+ }
+
+ r = userns_mkdir(directory, p, 0755, 0, 0);
+ if (r < 0)
+ log_warning_errno(r, "Failed to create directory %s: %m", q);
+ return 0;
+ }
+
+ if (arg_link_journal == LINK_HOST) {
+ /* don't create parents here — if the host doesn't have
+ * permanent journal set up, don't force it here */
+
+ r = RET_NERRNO(mkdir(p, 0755));
+ if (r < 0 && r != -EEXIST) {
+ if (try) {
+ log_debug_errno(r, "Failed to create %s, skipping journal setup: %m", p);
+ return 0;
+ } else
+ return log_error_errno(r, "Failed to create %s: %m", p);
+ }
+
+ } else if (access(p, F_OK) < 0)
+ return 0;
+
+ if (dir_is_empty(q, /* ignore_hidden_or_backup= */ false) == 0)
+ log_warning("%s is not empty, proceeding anyway.", q);
+
+ r = userns_mkdir(directory, p, 0755, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create %s: %m", q);
+
+ r = mount_nofollow_verbose(LOG_DEBUG, p, q, NULL, MS_BIND, NULL);
+ if (r < 0)
+ return log_error_errno(errno, "Failed to bind mount journal from host into guest: %m");
+
+ return 0;
+}
+
+static int drop_capabilities(uid_t uid) {
+ CapabilityQuintet q;
+
+ /* Let's initialize all five capability sets to something valid. If the quintet was configured via
+ * OCI use that, but fill in missing bits. If it wasn't then derive the quintet in full from
+ * arg_caps_retain. */
+
+ if (capability_quintet_is_set(&arg_full_capabilities)) {
+ q = arg_full_capabilities;
+
+ if (q.bounding == UINT64_MAX)
+ q.bounding = uid == 0 ? arg_caps_retain : 0;
+
+ if (q.effective == UINT64_MAX)
+ q.effective = uid == 0 ? q.bounding : 0;
+
+ if (q.inheritable == UINT64_MAX)
+ q.inheritable = uid == 0 ? q.bounding : arg_caps_ambient;
+
+ if (q.permitted == UINT64_MAX)
+ q.permitted = uid == 0 ? q.bounding : arg_caps_ambient;
+
+ if (q.ambient == UINT64_MAX && ambient_capabilities_supported())
+ q.ambient = arg_caps_ambient;
+
+ if (capability_quintet_mangle(&q))
+ return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Cannot set capabilities that are not in the current bounding set.");
+
+ } else {
+ q = (CapabilityQuintet) {
+ .bounding = arg_caps_retain,
+ .effective = uid == 0 ? arg_caps_retain : 0,
+ .inheritable = uid == 0 ? arg_caps_retain : arg_caps_ambient,
+ .permitted = uid == 0 ? arg_caps_retain : arg_caps_ambient,
+ .ambient = ambient_capabilities_supported() ? arg_caps_ambient : UINT64_MAX,
+ };
+
+ /* If we're not using OCI, proceed with mangled capabilities (so we don't error out)
+ * in order to maintain the same behavior as systemd < 242. */
+ if (capability_quintet_mangle(&q))
+ log_full(arg_quiet ? LOG_DEBUG : LOG_WARNING,
+ "Some capabilities will not be set because they are not in the current bounding set.");
+
+ }
+
+ return capability_quintet_enforce(&q);
+}
+
+static int reset_audit_loginuid(void) {
+ _cleanup_free_ char *p = NULL;
+ int r;
+
+ if ((arg_clone_ns_flags & CLONE_NEWPID) == 0)
+ return 0;
+
+ r = read_one_line_file("/proc/self/loginuid", &p);
+ if (r == -ENOENT)
+ return 0;
+ if (r < 0)
+ return log_error_errno(r, "Failed to read /proc/self/loginuid: %m");
+
+ /* Already reset? */
+ if (streq(p, "4294967295"))
+ return 0;
+
+ r = write_string_file("/proc/self/loginuid", "4294967295", WRITE_STRING_FILE_DISABLE_BUFFER);
+ if (r < 0) {
+ log_error_errno(r,
+ "Failed to reset audit login UID. This probably means that your kernel is too\n"
+ "old and you have audit enabled. Note that the auditing subsystem is known to\n"
+ "be incompatible with containers on old kernels. Please make sure to upgrade\n"
+ "your kernel or to off auditing with 'audit=0' on the kernel command line before\n"
+ "using systemd-nspawn. Sleeping for 5s... (%m)");
+
+ sleep(5);
+ }
+
+ return 0;
+}
+
+static int setup_propagate(const char *root) {
+ const char *p, *q;
+ int r;
+
+ (void) mkdir_p("/run/systemd/nspawn/", 0755);
+ (void) mkdir_p("/run/systemd/nspawn/propagate", 0600);
+ p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
+ (void) mkdir_p(p, 0600);
+
+ r = userns_mkdir(root, "/run/host", 0755, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create /run/host: %m");
+
+ r = userns_mkdir(root, "/run/host/incoming", 0600, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create /run/host/incoming: %m");
+
+ q = prefix_roota(root, "/run/host/incoming");
+ r = mount_nofollow_verbose(LOG_ERR, p, q, NULL, MS_BIND, NULL);
+ if (r < 0)
+ return r;
+
+ r = mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
+ if (r < 0)
+ return r;
+
+ /* machined will MS_MOVE into that directory, and that's only supported for non-shared mounts. */
+ return mount_nofollow_verbose(LOG_ERR, NULL, q, NULL, MS_SLAVE, NULL);
+}
+
+static int setup_machine_id(const char *directory) {
+ const char *etc_machine_id;
+ sd_id128_t id;
+ int r;
+
+ /* If the UUID in the container is already set, then that's what counts, and we use. If it isn't set, and the
+ * caller passed --uuid=, then we'll pass it in the $container_uuid env var to PID 1 of the container. The
+ * assumption is that PID 1 will then write it to /etc/machine-id to make it persistent. If --uuid= is not
+ * passed we generate a random UUID, and pass it via $container_uuid. In effect this means that /etc/machine-id
+ * in the container and our idea of the container UUID will always be in sync (at least if PID 1 in the
+ * container behaves nicely). */
+
+ etc_machine_id = prefix_roota(directory, "/etc/machine-id");
+
+ r = id128_read(etc_machine_id, ID128_FORMAT_PLAIN, &id);
+ if (r < 0) {
+ if (!IN_SET(r, -ENOENT, -ENOMEDIUM, -ENOPKG)) /* If the file is missing, empty, or uninitialized, we don't mind */
+ return log_error_errno(r, "Failed to read machine ID from container image: %m");
+
+ if (sd_id128_is_null(arg_uuid)) {
+ r = sd_id128_randomize(&arg_uuid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to acquire randomized machine UUID: %m");
+ }
+ } else {
+ if (sd_id128_is_null(id))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Machine ID in container image is zero, refusing.");
+
+ arg_uuid = id;
+ }
+
+ return 0;
+}
+
+static int recursive_chown(const char *directory, uid_t shift, uid_t range) {
+ int r;
+
+ assert(directory);
+
+ if (arg_userns_mode == USER_NAMESPACE_NO || arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_CHOWN)
+ return 0;
+
+ r = path_patch_uid(directory, arg_uid_shift, arg_uid_range);
+ if (r == -EOPNOTSUPP)
+ return log_error_errno(r, "Automatic UID/GID adjusting is only supported for UID/GID ranges starting at multiples of 2^16 with a range of 2^16.");
+ if (r == -EBADE)
+ return log_error_errno(r, "Upper 16 bits of root directory UID and GID do not match.");
+ if (r < 0)
+ return log_error_errno(r, "Failed to adjust UID/GID shift of OS tree: %m");
+ if (r == 0)
+ log_debug("Root directory of image is already owned by the right UID/GID range, skipping recursive chown operation.");
+ else
+ log_debug("Patched directory tree to match UID/GID range.");
+
+ return r;
+}
+
+/*
+ * Return values:
+ * < 0 : wait_for_terminate() failed to get the state of the
+ * container, the container was terminated by a signal, or
+ * failed for an unknown reason. No change is made to the
+ * container argument.
+ * > 0 : The program executed in the container terminated with an
+ * error. The exit code of the program executed in the
+ * container is returned. The container argument has been set
+ * to CONTAINER_TERMINATED.
+ * 0 : The container is being rebooted, has been shut down or exited
+ * successfully. The container argument has been set to either
+ * CONTAINER_TERMINATED or CONTAINER_REBOOTED.
+ *
+ * That is, success is indicated by a return value of zero, and an
+ * error is indicated by a non-zero value.
+ */
+static int wait_for_container(pid_t pid, ContainerStatus *container) {
+ siginfo_t status;
+ int r;
+
+ r = wait_for_terminate(pid, &status);
+ if (r < 0)
+ return log_warning_errno(r, "Failed to wait for container: %m");
+
+ switch (status.si_code) {
+
+ case CLD_EXITED:
+ if (status.si_status == 0)
+ log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s exited successfully.", arg_machine);
+ else
+ log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s failed with error code %i.", arg_machine, status.si_status);
+
+ *container = CONTAINER_TERMINATED;
+ return status.si_status;
+
+ case CLD_KILLED:
+ if (status.si_status == SIGINT) {
+ log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s has been shut down.", arg_machine);
+ *container = CONTAINER_TERMINATED;
+ return 0;
+
+ } else if (status.si_status == SIGHUP) {
+ log_full(arg_quiet ? LOG_DEBUG : LOG_INFO, "Container %s is being rebooted.", arg_machine);
+ *container = CONTAINER_REBOOTED;
+ return 0;
+ }
+
+ _fallthrough_;
+ case CLD_DUMPED:
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Container %s terminated by signal %s.", arg_machine, signal_to_string(status.si_status));
+
+ default:
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Container %s failed due to unknown reason.", arg_machine);
+ }
+}
+
+static int on_orderly_shutdown(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
+ pid_t pid;
+
+ pid = PTR_TO_PID(userdata);
+ if (pid > 0) {
+ if (kill(pid, arg_kill_signal) >= 0) {
+ log_info("Trying to halt container. Send SIGTERM again to trigger immediate termination.");
+ sd_event_source_set_userdata(s, NULL);
+ return 0;
+ }
+ }
+
+ sd_event_exit(sd_event_source_get_event(s), 0);
+ return 0;
+}
+
+static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *ssi, void *userdata) {
+ pid_t pid;
+
+ assert(s);
+ assert(ssi);
+
+ pid = PTR_TO_PID(userdata);
+
+ for (;;) {
+ siginfo_t si = {};
+
+ if (waitid(P_ALL, 0, &si, WNOHANG|WNOWAIT|WEXITED) < 0)
+ return log_error_errno(errno, "Failed to waitid(): %m");
+ if (si.si_pid == 0) /* No pending children. */
+ break;
+ if (si.si_pid == pid) {
+ /* The main process we care for has exited. Return from
+ * signal handler but leave the zombie. */
+ sd_event_exit(sd_event_source_get_event(s), 0);
+ break;
+ }
+
+ /* Reap all other children. */
+ (void) waitid(P_PID, si.si_pid, &si, WNOHANG|WEXITED);
+ }
+
+ return 0;
+}
+
+static int on_request_stop(sd_bus_message *m, void *userdata, sd_bus_error *error) {
+ pid_t pid;
+
+ assert(m);
+
+ pid = PTR_TO_PID(userdata);
+
+ if (arg_kill_signal > 0) {
+ log_info("Container termination requested. Attempting to halt container.");
+ (void) kill(pid, arg_kill_signal);
+ } else {
+ log_info("Container termination requested. Exiting.");
+ sd_event_exit(sd_bus_get_event(sd_bus_message_get_bus(m)), 0);
+ }
+
+ return 0;
+}
+
+static int determine_names(void) {
+ int r;
+
+ if (arg_template && !arg_directory && arg_machine) {
+
+ /* If --template= was specified then we should not
+ * search for a machine, but instead create a new one
+ * in /var/lib/machine. */
+
+ arg_directory = path_join("/var/lib/machines", arg_machine);
+ if (!arg_directory)
+ return log_oom();
+ }
+
+ if (!arg_image && !arg_directory) {
+ if (arg_machine) {
+ _cleanup_(image_unrefp) Image *i = NULL;
+
+ r = image_find(IMAGE_MACHINE, arg_machine, NULL, &i);
+ if (r == -ENOENT)
+ return log_error_errno(r, "No image for machine '%s'.", arg_machine);
+ if (r < 0)
+ return log_error_errno(r, "Failed to find image for machine '%s': %m", arg_machine);
+
+ if (IN_SET(i->type, IMAGE_RAW, IMAGE_BLOCK))
+ r = free_and_strdup(&arg_image, i->path);
+ else
+ r = free_and_strdup(&arg_directory, i->path);
+ if (r < 0)
+ return log_oom();
+
+ if (!arg_ephemeral)
+ arg_read_only = arg_read_only || i->read_only;
+ } else {
+ r = safe_getcwd(&arg_directory);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine current directory: %m");
+ }
+
+ if (!arg_directory && !arg_image)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine path, please use -D or -i.");
+ }
+
+ if (!arg_machine) {
+ if (arg_directory && path_equal(arg_directory, "/"))
+ arg_machine = gethostname_malloc();
+ else if (arg_image) {
+ char *e;
+
+ arg_machine = strdup(basename(arg_image));
+
+ /* Truncate suffix if there is one */
+ e = endswith(arg_machine, ".raw");
+ if (e)
+ *e = 0;
+ } else
+ arg_machine = strdup(basename(arg_directory));
+ if (!arg_machine)
+ return log_oom();
+
+ hostname_cleanup(arg_machine);
+ if (!hostname_is_valid(arg_machine, 0))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to determine machine name automatically, please use -M.");
+
+ /* Copy the machine name before the random suffix is added below, otherwise we won't be able
+ * to match fixed config file names. */
+ arg_settings_filename = strjoin(arg_machine, ".nspawn");
+ if (!arg_settings_filename)
+ return log_oom();
+
+ /* Add a random suffix when this is an ephemeral machine, so that we can run many
+ * instances at once without manually having to specify -M each time. */
+ if (arg_ephemeral)
+ if (strextendf(&arg_machine, "-%016" PRIx64, random_u64()) < 0)
+ return log_oom();
+ } else {
+ arg_settings_filename = strjoin(arg_machine, ".nspawn");
+ if (!arg_settings_filename)
+ return log_oom();
+ }
+
+ return 0;
+}
+
+static int chase_symlinks_and_update(char **p, unsigned flags) {
+ char *chased;
+ int r;
+
+ assert(p);
+
+ if (!*p)
+ return 0;
+
+ r = chase_symlinks(*p, NULL, flags, &chased, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to resolve path %s: %m", *p);
+
+ return free_and_replace(*p, chased);
+}
+
+static int determine_uid_shift(const char *directory) {
+
+ if (arg_userns_mode == USER_NAMESPACE_NO) {
+ arg_uid_shift = 0;
+ return 0;
+ }
+
+ if (arg_uid_shift == UID_INVALID) {
+ struct stat st;
+
+ /* Read the UID shift off the image. Maybe we can reuse this to avoid chowning. */
+
+ if (stat(directory, &st) < 0)
+ return log_error_errno(errno, "Failed to determine UID base of %s: %m", directory);
+
+ arg_uid_shift = st.st_uid & UINT32_C(0xffff0000);
+
+ if (arg_uid_shift != (st.st_gid & UINT32_C(0xffff0000)))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "UID and GID base of %s don't match.", directory);
+
+ arg_uid_range = UINT32_C(0x10000);
+
+ if (arg_uid_shift != 0) {
+ /* If the image is shifted already, then we'll fall back to classic chowning, for
+ * compatibility (and simplicity), or refuse if mapping is explicitly requested. */
+
+ if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_AUTO) {
+ log_debug("UID base of %s is non-zero, not using UID mapping.", directory);
+ arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
+ } else if (arg_userns_ownership == USER_NAMESPACE_OWNERSHIP_MAP)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "UID base of %s is not zero, UID mapping not supported.", directory);
+ }
+ }
+
+ if (!userns_shift_range_valid(arg_uid_shift, arg_uid_range))
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "UID base too high for UID range.");
+
+ return 0;
+}
+
+static unsigned long effective_clone_ns_flags(void) {
+ unsigned long flags = arg_clone_ns_flags;
+
+ if (arg_private_network)
+ flags |= CLONE_NEWNET;
+ if (arg_use_cgns)
+ flags |= CLONE_NEWCGROUP;
+ if (arg_userns_mode != USER_NAMESPACE_NO)
+ flags |= CLONE_NEWUSER;
+
+ return flags;
+}
+
+static int patch_sysctl(void) {
+
+ /* This table is inspired by runc's sysctl() function */
+ static const struct {
+ const char *key;
+ bool prefix;
+ unsigned long clone_flags;
+ } safe_sysctl[] = {
+ { "kernel.hostname", false, CLONE_NEWUTS },
+ { "kernel.domainname", false, CLONE_NEWUTS },
+ { "kernel.msgmax", false, CLONE_NEWIPC },
+ { "kernel.msgmnb", false, CLONE_NEWIPC },
+ { "kernel.msgmni", false, CLONE_NEWIPC },
+ { "kernel.sem", false, CLONE_NEWIPC },
+ { "kernel.shmall", false, CLONE_NEWIPC },
+ { "kernel.shmmax", false, CLONE_NEWIPC },
+ { "kernel.shmmni", false, CLONE_NEWIPC },
+ { "fs.mqueue.", true, CLONE_NEWIPC },
+ { "net.", true, CLONE_NEWNET },
+ };
+
+ unsigned long flags;
+ int r;
+
+ flags = effective_clone_ns_flags();
+
+ STRV_FOREACH_PAIR(k, v, arg_sysctl) {
+ bool good = false;
+ size_t i;
+
+ for (i = 0; i < ELEMENTSOF(safe_sysctl); i++) {
+
+ if (!FLAGS_SET(flags, safe_sysctl[i].clone_flags))
+ continue;
+
+ if (safe_sysctl[i].prefix)
+ good = startswith(*k, safe_sysctl[i].key);
+ else
+ good = streq(*k, safe_sysctl[i].key);
+
+ if (good)
+ break;
+ }
+
+ if (!good)
+ return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Refusing to write to sysctl '%s', as it is not safe in the selected namespaces.", *k);
+
+ r = sysctl_write(*k, *v);
+ if (r < 0)
+ return log_error_errno(r, "Failed to write sysctl '%s': %m", *k);
+ }
+
+ return 0;
+}
+
+static int inner_child(
+ Barrier *barrier,
+ const char *directory,
+ bool secondary,
+ int kmsg_socket,
+ int rtnl_socket,
+ int master_pty_socket,
+ FDSet *fds,
+ char **os_release_pairs) {
+
+ _cleanup_free_ char *home = NULL;
+ size_t n_env = 1;
+ char *envp[] = {
+ (char*) "PATH=" DEFAULT_PATH_COMPAT,
+ NULL, /* container */
+ NULL, /* TERM */
+ NULL, /* HOME */
+ NULL, /* USER */
+ NULL, /* LOGNAME */
+ NULL, /* container_uuid */
+ NULL, /* LISTEN_FDS */
+ NULL, /* LISTEN_PID */
+ NULL, /* NOTIFY_SOCKET */
+ NULL, /* CREDENTIALS_DIRECTORY */
+ NULL, /* LANG */
+ NULL
+ };
+ const char *exec_target;
+ _cleanup_strv_free_ char **env_use = NULL;
+ int r, which_failed;
+
+ /* This is the "inner" child process, i.e. the one forked off by the "outer" child process, which is the one
+ * the container manager itself forked off. At the time of clone() it gained its own CLONE_NEWNS, CLONE_NEWPID,
+ * CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER namespaces. Note that it has its own CLONE_NEWNS namespace,
+ * separate from the CLONE_NEWNS created for the "outer" child, and also separate from the host's CLONE_NEWNS
+ * namespace. The reason for having two levels of CLONE_NEWNS namespaces is that the "inner" one is owned by
+ * the CLONE_NEWUSER namespace of the container, while the "outer" one is owned by the host's CLONE_NEWUSER
+ * namespace.
+ *
+ * Note at this point we have no CLONE_NEWNET namespace yet. We'll acquire that one later through
+ * unshare(). See below. */
+
+ assert(barrier);
+ assert(directory);
+ assert(kmsg_socket >= 0);
+
+ log_debug("Inner child is initializing.");
+
+ if (arg_userns_mode != USER_NAMESPACE_NO) {
+ /* Tell the parent, that it now can write the UID map. */
+ (void) barrier_place(barrier); /* #1 */
+
+ /* Wait until the parent wrote the UID map */
+ if (!barrier_place_and_sync(barrier)) /* #2 */
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
+
+ /* Become the new root user inside our namespace */
+ r = reset_uid_gid();
+ if (r < 0)
+ return log_error_errno(r, "Couldn't become new root: %m");
+
+ /* Creating a new user namespace means all MS_SHARED mounts become MS_SLAVE. Let's put them
+ * back to MS_SHARED here, since that's what we want as defaults. (This will not reconnect
+ * propagation, but simply create new peer groups for all our mounts). */
+ r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SHARED|MS_REC, NULL);
+ if (r < 0)
+ return r;
+ }
+
+ r = mount_all(NULL,
+ arg_mount_settings | MOUNT_IN_USERNS,
+ arg_uid_shift,
+ arg_selinux_apifs_context);
+ if (r < 0)
+ return r;
+
+ if (!arg_network_namespace_path && arg_private_network) {
+ r = unshare(CLONE_NEWNET);
+ if (r < 0)
+ return log_error_errno(errno, "Failed to unshare network namespace: %m");
+
+ /* Tell the parent that it can setup network interfaces. */
+ (void) barrier_place(barrier); /* #3 */
+ }
+
+ r = mount_sysfs(NULL, arg_mount_settings);
+ if (r < 0)
+ return r;
+
+ /* Wait until we are cgroup-ified, so that we
+ * can mount the right cgroup path writable */
+ if (!barrier_place_and_sync(barrier)) /* #4 */
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
+ "Parent died too early");
+
+ if (arg_use_cgns) {
+ r = unshare(CLONE_NEWCGROUP);
+ if (r < 0)
+ return log_error_errno(errno, "Failed to unshare cgroup namespace: %m");
+ r = mount_cgroups(
+ "",
+ arg_unified_cgroup_hierarchy,
+ arg_userns_mode != USER_NAMESPACE_NO,
+ arg_uid_shift,
+ arg_uid_range,
+ arg_selinux_apifs_context,
+ true);
+ } else
+ r = mount_systemd_cgroup_writable("", arg_unified_cgroup_hierarchy);
+ if (r < 0)
+ return r;
+
+ r = setup_boot_id();
+ if (r < 0)
+ return r;
+
+ r = setup_kmsg(kmsg_socket);
+ if (r < 0)
+ return r;
+ kmsg_socket = safe_close(kmsg_socket);
+
+ r = mount_custom(
+ "/",
+ arg_custom_mounts,
+ arg_n_custom_mounts,
+ 0,
+ 0,
+ arg_selinux_apifs_context,
+ MOUNT_NON_ROOT_ONLY | MOUNT_IN_USERNS);
+ if (r < 0)
+ return r;
+
+ if (setsid() < 0)
+ return log_error_errno(errno, "setsid() failed: %m");
+
+ if (arg_private_network)
+ (void) loopback_setup();
+
+ if (arg_expose_ports) {
+ r = expose_port_send_rtnl(rtnl_socket);
+ if (r < 0)
+ return r;
+ rtnl_socket = safe_close(rtnl_socket);
+ }
+
+ if (arg_console_mode != CONSOLE_PIPE) {
+ _cleanup_close_ int master = -1;
+ _cleanup_free_ char *console = NULL;
+
+ /* Allocate a pty and make it available as /dev/console. */
+ master = openpt_allocate(O_RDWR|O_NONBLOCK, &console);
+ if (master < 0)
+ return log_error_errno(master, "Failed to allocate a pty: %m");
+
+ r = setup_dev_console(console);
+ if (r < 0)
+ return log_error_errno(r, "Failed to set up /dev/console: %m");
+
+ r = send_one_fd(master_pty_socket, master, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to send master fd: %m");
+ master_pty_socket = safe_close(master_pty_socket);
+
+ r = setup_stdio_as_dev_console();
+ if (r < 0)
+ return r;
+ }
+
+ r = patch_sysctl();
+ if (r < 0)
+ return r;
+
+ if (arg_oom_score_adjust_set) {
+ r = set_oom_score_adjust(arg_oom_score_adjust);
+ if (r < 0)
+ return log_error_errno(r, "Failed to adjust OOM score: %m");
+ }
+
+ if (arg_cpu_set.set)
+ if (sched_setaffinity(0, arg_cpu_set.allocated, arg_cpu_set.set) < 0)
+ return log_error_errno(errno, "Failed to set CPU affinity: %m");
+
+ (void) setup_hostname();
+
+ if (arg_personality != PERSONALITY_INVALID) {
+ r = safe_personality(arg_personality);
+ if (r < 0)
+ return log_error_errno(r, "personality() failed: %m");
+ } else if (secondary) {
+ r = safe_personality(PER_LINUX32);
+ if (r < 0)
+ return log_error_errno(r, "personality() failed: %m");
+ }
+
+ r = setrlimit_closest_all((const struct rlimit *const*) arg_rlimit, &which_failed);
+ if (r < 0)
+ return log_error_errno(r, "Failed to apply resource limit RLIMIT_%s: %m", rlimit_to_string(which_failed));
+
+#if HAVE_SECCOMP
+ if (arg_seccomp) {
+
+ if (is_seccomp_available()) {
+
+ r = seccomp_load(arg_seccomp);
+ if (ERRNO_IS_SECCOMP_FATAL(r))
+ return log_error_errno(r, "Failed to install seccomp filter: %m");
+ if (r < 0)
+ log_debug_errno(r, "Failed to install seccomp filter: %m");
+ }
+ } else
+#endif
+ {
+ r = setup_seccomp(arg_caps_retain, arg_syscall_allow_list, arg_syscall_deny_list);
+ if (r < 0)
+ return r;
+ }
+
+ if (arg_suppress_sync) {
+#if HAVE_SECCOMP
+ r = seccomp_suppress_sync();
+ if (r < 0)
+ log_debug_errno(r, "Failed to install sync() suppression seccomp filter, ignoring: %m");
+#else
+ log_debug("systemd is built without SECCOMP support. Ignoring --suppress-sync= command line option and SuppressSync= setting.");
+#endif
+ }
+
+#if HAVE_SELINUX
+ if (arg_selinux_context)
+ if (setexeccon(arg_selinux_context) < 0)
+ return log_error_errno(errno, "setexeccon(\"%s\") failed: %m", arg_selinux_context);
+#endif
+
+ /* Make sure we keep the caps across the uid/gid dropping, so that we can retain some selected caps
+ * if we need to later on. */
+ if (prctl(PR_SET_KEEPCAPS, 1) < 0)
+ return log_error_errno(errno, "Failed to set PR_SET_KEEPCAPS: %m");
+
+ if (uid_is_valid(arg_uid) || gid_is_valid(arg_gid))
+ r = change_uid_gid_raw(arg_uid, arg_gid, arg_supplementary_gids, arg_n_supplementary_gids, arg_console_mode != CONSOLE_PIPE);
+ else
+ r = change_uid_gid(arg_user, arg_console_mode != CONSOLE_PIPE, &home);
+ if (r < 0)
+ return r;
+
+ r = drop_capabilities(getuid());
+ if (r < 0)
+ return log_error_errno(r, "Dropping capabilities failed: %m");
+
+ if (arg_no_new_privileges)
+ if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0)
+ return log_error_errno(errno, "Failed to disable new privileges: %m");
+
+ /* LXC sets container=lxc, so follow the scheme here */
+ envp[n_env++] = strjoina("container=", arg_container_service_name);
+
+ envp[n_env] = strv_find_prefix(environ, "TERM=");
+ if (envp[n_env])
+ n_env++;
+
+ if (home || !uid_is_valid(arg_uid) || arg_uid == 0)
+ if (asprintf(envp + n_env++, "HOME=%s", home ?: "/root") < 0)
+ return log_oom();
+
+ if (arg_user || !uid_is_valid(arg_uid) || arg_uid == 0)
+ if (asprintf(envp + n_env++, "USER=%s", arg_user ?: "root") < 0 ||
+ asprintf(envp + n_env++, "LOGNAME=%s", arg_user ? arg_user : "root") < 0)
+ return log_oom();
+
+ assert(!sd_id128_is_null(arg_uuid));
+
+ if (asprintf(envp + n_env++, "container_uuid=%s", SD_ID128_TO_UUID_STRING(arg_uuid)) < 0)
+ return log_oom();
+
+ if (fdset_size(fds) > 0) {
+ r = fdset_cloexec(fds, false);
+ if (r < 0)
+ return log_error_errno(r, "Failed to unset O_CLOEXEC for file descriptors.");
+
+ if ((asprintf(envp + n_env++, "LISTEN_FDS=%u", fdset_size(fds)) < 0) ||
+ (asprintf(envp + n_env++, "LISTEN_PID=1") < 0))
+ return log_oom();
+ }
+ if (asprintf(envp + n_env++, "NOTIFY_SOCKET=%s", NSPAWN_NOTIFY_SOCKET_PATH) < 0)
+ return log_oom();
+
+ if (arg_n_credentials > 0) {
+ envp[n_env] = strdup("CREDENTIALS_DIRECTORY=/run/host/credentials");
+ if (!envp[n_env])
+ return log_oom();
+ n_env++;
+ }
+
+ if (arg_start_mode != START_BOOT) {
+ envp[n_env] = strdup("LANG=" SYSTEMD_NSPAWN_LOCALE);
+ if (!envp[n_env])
+ return log_oom();
+ n_env++;
+ }
+
+ env_use = strv_env_merge(envp, os_release_pairs, arg_setenv);
+ if (!env_use)
+ return log_oom();
+
+ /* Let the parent know that we are ready and
+ * wait until the parent is ready with the
+ * setup, too... */
+ if (!barrier_place_and_sync(barrier)) /* #5 */
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Parent died too early");
+
+ if (arg_chdir)
+ if (chdir(arg_chdir) < 0)
+ return log_error_errno(errno, "Failed to change to specified working directory %s: %m", arg_chdir);
+
+ if (arg_start_mode == START_PID2) {
+ r = stub_pid1(arg_uuid);
+ if (r < 0)
+ return r;
+ }
+
+ if (arg_console_mode != CONSOLE_PIPE) {
+ /* So far our pty wasn't controlled by any process. Finally, it's time to change that, if we
+ * are configured for that. Acquire it as controlling tty. */
+ if (ioctl(STDIN_FILENO, TIOCSCTTY) < 0)
+ return log_error_errno(errno, "Failed to acquire controlling TTY: %m");
+ }
+
+ log_debug("Inner child completed, invoking payload.");
+
+ /* Now, explicitly close the log, so that we then can close all remaining fds. Closing the log explicitly first
+ * has the benefit that the logging subsystem knows about it, and is thus ready to be reopened should we need
+ * it again. Note that the other fds closed here are at least the locking and barrier fds. */
+ log_close();
+ log_set_open_when_needed(true);
+ log_settle_target();
+
+ (void) fdset_close_others(fds);
+
+ if (arg_start_mode == START_BOOT) {
+ char **a;
+ size_t m;
+
+ /* Automatically search for the init system */
+
+ m = strv_length(arg_parameters);
+ a = newa(char*, m + 2);
+ memcpy_safe(a + 1, arg_parameters, m * sizeof(char*));
+ a[1 + m] = NULL;
+
+ FOREACH_STRING(init,
+ "/usr/lib/systemd/systemd",
+ "/lib/systemd/systemd",
+ "/sbin/init") {
+ a[0] = (char*) init;
+ execve(a[0], a, env_use);
+ }
+
+ exec_target = "/usr/lib/systemd/systemd, /lib/systemd/systemd, /sbin/init";
+ } else if (!strv_isempty(arg_parameters)) {
+ const char *dollar_path;
+
+ exec_target = arg_parameters[0];
+
+ /* Use the user supplied search $PATH if there is one, or DEFAULT_PATH_COMPAT if not to search the
+ * binary. */
+ dollar_path = strv_env_get(env_use, "PATH");
+ if (dollar_path) {
+ if (setenv("PATH", dollar_path, 1) < 0)
+ return log_error_errno(errno, "Failed to update $PATH: %m");
+ }
+
+ execvpe(arg_parameters[0], arg_parameters, env_use);
+ } else {
+ if (!arg_chdir)
+ /* If we cannot change the directory, we'll end up in /, that is expected. */
+ (void) chdir(home ?: "/root");
+
+ execle(DEFAULT_USER_SHELL, "-" DEFAULT_USER_SHELL_NAME, NULL, env_use);
+ if (!streq(DEFAULT_USER_SHELL, "/bin/bash"))
+ execle("/bin/bash", "-bash", NULL, env_use);
+ if (!streq(DEFAULT_USER_SHELL, "/bin/sh"))
+ execle("/bin/sh", "-sh", NULL, env_use);
+
+ exec_target = DEFAULT_USER_SHELL ", /bin/bash, /bin/sh";
+ }
+
+ return log_error_errno(errno, "execv(%s) failed: %m", exec_target);
+}
+
+static int setup_notify_child(void) {
+ _cleanup_close_ int fd = -1;
+ static const union sockaddr_union sa = {
+ .un.sun_family = AF_UNIX,
+ .un.sun_path = NSPAWN_NOTIFY_SOCKET_PATH,
+ };
+ int r;
+
+ fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
+ if (fd < 0)
+ return log_error_errno(errno, "Failed to allocate notification socket: %m");
+
+ (void) mkdir_parents(NSPAWN_NOTIFY_SOCKET_PATH, 0755);
+ (void) sockaddr_un_unlink(&sa.un);
+
+ r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
+ if (r < 0)
+ return log_error_errno(errno, "bind(" NSPAWN_NOTIFY_SOCKET_PATH ") failed: %m");
+
+ r = userns_lchown(NSPAWN_NOTIFY_SOCKET_PATH, 0, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to chown " NSPAWN_NOTIFY_SOCKET_PATH ": %m");
+
+ r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
+ if (r < 0)
+ return log_error_errno(r, "SO_PASSCRED failed: %m");
+
+ return TAKE_FD(fd);
+}
+
+static int outer_child(
+ Barrier *barrier,
+ const char *directory,
+ DissectedImage *dissected_image,
+ bool secondary,
+ int pid_socket,
+ int uuid_socket,
+ int notify_socket,
+ int kmsg_socket,
+ int rtnl_socket,
+ int uid_shift_socket,
+ int master_pty_socket,
+ int unified_cgroup_hierarchy_socket,
+ FDSet *fds,
+ int netns_fd) {
+
+ _cleanup_(bind_user_context_freep) BindUserContext *bind_user_context = NULL;
+ _cleanup_strv_free_ char **os_release_pairs = NULL;
+ _cleanup_close_ int fd = -1;
+ bool idmap = false;
+ const char *p;
+ pid_t pid;
+ ssize_t l;
+ int r;
+
+ /* This is the "outer" child process, i.e the one forked off by the container manager itself. It
+ * already has its own CLONE_NEWNS namespace (which was created by the clone()). It still lives in
+ * the host's CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, CLONE_NEWUSER and CLONE_NEWNET
+ * namespaces. After it completed a number of initializations a second child (the "inner" one) is
+ * forked off it, and it exits. */
+
+ assert(barrier);
+ assert(directory);
+ assert(pid_socket >= 0);
+ assert(uuid_socket >= 0);
+ assert(notify_socket >= 0);
+ assert(master_pty_socket >= 0);
+ assert(kmsg_socket >= 0);
+
+ log_debug("Outer child is initializing.");
+
+ r = load_os_release_pairs_with_prefix("/", "container_host_", &os_release_pairs);
+ if (r < 0)
+ log_debug_errno(r, "Failed to read os-release from host for container, ignoring: %m");
+
+ if (prctl(PR_SET_PDEATHSIG, SIGKILL) < 0)
+ return log_error_errno(errno, "PR_SET_PDEATHSIG failed: %m");
+
+ r = reset_audit_loginuid();
+ if (r < 0)
+ return r;
+
+ /* Mark everything as slave, so that we still receive mounts from the real root, but don't propagate
+ * mounts to the real root. */
+ r = mount_follow_verbose(LOG_ERR, NULL, "/", NULL, MS_SLAVE|MS_REC, NULL);
+ if (r < 0)
+ return r;
+
+ if (dissected_image) {
+ /* If we are operating on a disk image, then mount its root directory now, but leave out the
+ * rest. We can read the UID shift from it if we need to. Further down we'll mount the rest,
+ * but then with the uid shift known. That way we can mount VFAT file systems shifted to the
+ * right place right away. This makes sure ESP partitions and userns are compatible. */
+
+ r = dissected_image_mount_and_warn(
+ dissected_image,
+ directory,
+ arg_uid_shift,
+ arg_uid_range,
+ DISSECT_IMAGE_MOUNT_ROOT_ONLY|
+ DISSECT_IMAGE_DISCARD_ON_LOOP|
+ DISSECT_IMAGE_USR_NO_ROOT|
+ (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK|DISSECT_IMAGE_GROWFS)|
+ (arg_start_mode == START_BOOT ? DISSECT_IMAGE_VALIDATE_OS : 0));
+ if (r < 0)
+ return r;
+ }
+
+ r = determine_uid_shift(directory);
+ if (r < 0)
+ return r;
+
+ if (arg_userns_mode != USER_NAMESPACE_NO) {
+ /* Let the parent know which UID shift we read from the image */
+ l = send(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), MSG_NOSIGNAL);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to send UID shift: %m");
+ if (l != sizeof(arg_uid_shift))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Short write while sending UID shift.");
+
+ if (arg_userns_mode == USER_NAMESPACE_PICK) {
+ /* When we are supposed to pick the UID shift, the parent will check now whether the
+ * UID shift we just read from the image is available. If yes, it will send the UID
+ * shift back to us, if not it will pick a different one, and send it back to us. */
+
+ l = recv(uid_shift_socket, &arg_uid_shift, sizeof(arg_uid_shift), 0);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to recv UID shift: %m");
+ if (l != sizeof(arg_uid_shift))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Short read while receiving UID shift.");
+ }
+
+ log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
+ "Selected user namespace base " UID_FMT " and range " UID_FMT ".", arg_uid_shift, arg_uid_range);
+ }
+
+ if (path_equal(directory, "/")) {
+ /* If the directory we shall boot is the host, let's operate on a bind mount at a different
+ * place, so that we can make changes to its mount structure (for example, to implement
+ * --volatile=) without this interfering with our ability to access files such as
+ * /etc/localtime to copy into the container. Note that we use a fixed place for this
+ * (instead of a temporary directory, since we are living in our own mount namspace here
+ * already, and thus don't need to be afraid of colliding with anyone else's mounts). */
+ (void) mkdir_p("/run/systemd/nspawn-root", 0755);
+
+ r = mount_nofollow_verbose(LOG_ERR, "/", "/run/systemd/nspawn-root", NULL, MS_BIND|MS_REC, NULL);
+ if (r < 0)
+ return r;
+
+ directory = "/run/systemd/nspawn-root";
+ }
+
+ r = setup_pivot_root(
+ directory,
+ arg_pivot_root_new,
+ arg_pivot_root_old);
+ if (r < 0)
+ return r;
+
+ r = setup_volatile_mode(
+ directory,
+ arg_volatile_mode,
+ arg_uid_shift,
+ arg_selinux_apifs_context);
+ if (r < 0)
+ return r;
+
+ r = bind_user_prepare(
+ directory,
+ arg_bind_user,
+ arg_uid_shift,
+ arg_uid_range,
+ &arg_custom_mounts, &arg_n_custom_mounts,
+ &bind_user_context);
+ if (r < 0)
+ return r;
+
+ if (arg_userns_mode != USER_NAMESPACE_NO && bind_user_context) {
+ /* Send the user maps we determined to the parent, so that it installs it in our user
+ * namespace UID map table */
+
+ for (size_t i = 0; i < bind_user_context->n_data; i++) {
+ uid_t map[] = {
+ bind_user_context->data[i].payload_user->uid,
+ bind_user_context->data[i].host_user->uid,
+ (uid_t) bind_user_context->data[i].payload_group->gid,
+ (uid_t) bind_user_context->data[i].host_group->gid,
+ };
+
+ l = send(uid_shift_socket, map, sizeof(map), MSG_NOSIGNAL);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to send user UID map: %m");
+ if (l != sizeof(map))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Short write while sending user UID map.");
+ }
+ }
+
+ r = mount_custom(
+ directory,
+ arg_custom_mounts,
+ arg_n_custom_mounts,
+ arg_uid_shift,
+ arg_uid_range,
+ arg_selinux_apifs_context,
+ MOUNT_ROOT_ONLY);
+ if (r < 0)
+ return r;
+
+ /* Make sure we always have a mount that we can move to root later on. */
+ r = make_mount_point(directory);
+ if (r < 0)
+ return r;
+
+ if (arg_userns_mode != USER_NAMESPACE_NO &&
+ IN_SET(arg_userns_ownership, USER_NAMESPACE_OWNERSHIP_MAP, USER_NAMESPACE_OWNERSHIP_AUTO) &&
+ arg_uid_shift != 0) {
+
+ r = remount_idmap(directory, arg_uid_shift, arg_uid_range, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT);
+ if (r == -EINVAL || ERRNO_IS_NOT_SUPPORTED(r)) {
+ /* This might fail because the kernel or file system doesn't support idmapping. We
+ * can't really distinguish this nicely, nor do we have any guarantees about the
+ * error codes we see, could be EOPNOTSUPP or EINVAL. */
+ if (arg_userns_ownership != USER_NAMESPACE_OWNERSHIP_AUTO)
+ return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "ID mapped mounts are apparently not available, sorry.");
+
+ log_debug("ID mapped mounts are apparently not available on this kernel or for the selected file system, reverting to recursive chown()ing.");
+ arg_userns_ownership = USER_NAMESPACE_OWNERSHIP_CHOWN;
+ } else if (r < 0)
+ return log_error_errno(r, "Failed to set up ID mapped mounts: %m");
+ else {
+ log_debug("ID mapped mounts available, making use of them.");
+ idmap = true;
+ }
+ }
+
+ if (dissected_image) {
+ /* Now we know the uid shift, let's now mount everything else that might be in the image. */
+ r = dissected_image_mount(
+ dissected_image,
+ directory,
+ arg_uid_shift,
+ arg_uid_range,
+ DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY|
+ DISSECT_IMAGE_DISCARD_ON_LOOP|
+ DISSECT_IMAGE_USR_NO_ROOT|
+ (arg_read_only ? DISSECT_IMAGE_READ_ONLY : DISSECT_IMAGE_FSCK|DISSECT_IMAGE_GROWFS)|
+ (idmap ? DISSECT_IMAGE_MOUNT_IDMAPPED : 0));
+ if (r == -EUCLEAN)
+ return log_error_errno(r, "File system check for image failed: %m");
+ if (r < 0)
+ return log_error_errno(r, "Failed to mount image file system: %m");
+ }
+
+ if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
+ /* OK, we don't know yet which cgroup mode to use yet. Let's figure it out, and tell the parent. */
+
+ r = detect_unified_cgroup_hierarchy_from_image(directory);
+ if (r < 0)
+ return r;
+
+ l = send(unified_cgroup_hierarchy_socket, &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), MSG_NOSIGNAL);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to send cgroup mode: %m");
+ if (l != sizeof(arg_unified_cgroup_hierarchy))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Short write while sending cgroup mode.");
+
+ unified_cgroup_hierarchy_socket = safe_close(unified_cgroup_hierarchy_socket);
+ }
+
+ /* Mark everything as shared so our mounts get propagated down. This is required to make new bind
+ * mounts available in systemd services inside the container that create a new mount namespace. See
+ * https://github.com/systemd/systemd/issues/3860 Further submounts (such as /dev) done after this
+ * will inherit the shared propagation mode.
+ *
+ * IMPORTANT: Do not overmount the root directory anymore from now on to enable moving the root
+ * directory mount to root later on.
+ * https://github.com/systemd/systemd/issues/3847#issuecomment-562735251
+ */
+ r = mount_nofollow_verbose(LOG_ERR, NULL, directory, NULL, MS_SHARED|MS_REC, NULL);
+ if (r < 0)
+ return r;
+
+ r = recursive_chown(directory, arg_uid_shift, arg_uid_range);
+ if (r < 0)
+ return r;
+
+ r = base_filesystem_create(directory, arg_uid_shift, (gid_t) arg_uid_shift);
+ if (r < 0)
+ return r;
+
+ if (arg_read_only && arg_volatile_mode == VOLATILE_NO &&
+ !has_custom_root_mount(arg_custom_mounts, arg_n_custom_mounts)) {
+ r = bind_remount_recursive(directory, MS_RDONLY, MS_RDONLY, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to make tree read-only: %m");
+ }
+
+ r = mount_all(directory,
+ arg_mount_settings,
+ arg_uid_shift,
+ arg_selinux_apifs_context);
+ if (r < 0)
+ return r;
+
+ r = copy_devnodes(directory);
+ if (r < 0)
+ return r;
+
+ r = make_extra_nodes(directory);
+ if (r < 0)
+ return r;
+
+ (void) dev_setup(directory, arg_uid_shift, arg_uid_shift);
+
+ p = prefix_roota(directory, "/run/host");
+ (void) make_inaccessible_nodes(p, arg_uid_shift, arg_uid_shift);
+
+ r = setup_pts(directory);
+ if (r < 0)
+ return r;
+
+ r = setup_propagate(directory);
+ if (r < 0)
+ return r;
+
+ r = setup_keyring();
+ if (r < 0)
+ return r;
+
+ r = setup_credentials(directory);
+ if (r < 0)
+ return r;
+
+ r = bind_user_setup(bind_user_context, directory);
+ if (r < 0)
+ return r;
+
+ r = mount_custom(
+ directory,
+ arg_custom_mounts,
+ arg_n_custom_mounts,
+ arg_uid_shift,
+ arg_uid_range,
+ arg_selinux_apifs_context,
+ MOUNT_NON_ROOT_ONLY);
+ if (r < 0)
+ return r;
+
+ r = setup_timezone(directory);
+ if (r < 0)
+ return r;
+
+ r = setup_resolv_conf(directory);
+ if (r < 0)
+ return r;
+
+ r = setup_machine_id(directory);
+ if (r < 0)
+ return r;
+
+ r = setup_journal(directory);
+ if (r < 0)
+ return r;
+
+ /* The same stuff as the $container env var, but nicely readable for the entire payload */
+ p = prefix_roota(directory, "/run/host/container-manager");
+ (void) write_string_file(p, arg_container_service_name, WRITE_STRING_FILE_CREATE);
+
+ /* The same stuff as the $container_uuid env var */
+ p = prefix_roota(directory, "/run/host/container-uuid");
+ (void) write_string_filef(p, WRITE_STRING_FILE_CREATE, SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(arg_uuid));
+
+ if (!arg_use_cgns) {
+ r = mount_cgroups(
+ directory,
+ arg_unified_cgroup_hierarchy,
+ arg_userns_mode != USER_NAMESPACE_NO,
+ arg_uid_shift,
+ arg_uid_range,
+ arg_selinux_apifs_context,
+ false);
+ if (r < 0)
+ return r;
+ }
+
+ r = mount_move_root(directory);
+ if (r < 0)
+ return log_error_errno(r, "Failed to move root directory: %m");
+
+ fd = setup_notify_child();
+ if (fd < 0)
+ return fd;
+
+ pid = raw_clone(SIGCHLD|CLONE_NEWNS|
+ arg_clone_ns_flags |
+ (arg_userns_mode != USER_NAMESPACE_NO ? CLONE_NEWUSER : 0));
+ if (pid < 0)
+ return log_error_errno(errno, "Failed to fork inner child: %m");
+ if (pid == 0) {
+ pid_socket = safe_close(pid_socket);
+ uuid_socket = safe_close(uuid_socket);
+ notify_socket = safe_close(notify_socket);
+ uid_shift_socket = safe_close(uid_shift_socket);
+
+ /* The inner child has all namespaces that are requested, so that we all are owned by the
+ * user if user namespaces are turned on. */
+
+ if (arg_network_namespace_path) {
+ r = namespace_enter(-1, -1, netns_fd, -1, -1);
+ if (r < 0)
+ return log_error_errno(r, "Failed to join network namespace: %m");
+ }
+
+ r = inner_child(barrier, directory, secondary, kmsg_socket, rtnl_socket, master_pty_socket, fds, os_release_pairs);
+ if (r < 0)
+ _exit(EXIT_FAILURE);
+
+ _exit(EXIT_SUCCESS);
+ }
+
+ l = send(pid_socket, &pid, sizeof(pid), MSG_NOSIGNAL);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to send PID: %m");
+ if (l != sizeof(pid))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Short write while sending PID.");
+
+ l = send(uuid_socket, &arg_uuid, sizeof(arg_uuid), MSG_NOSIGNAL);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to send machine ID: %m");
+ if (l != sizeof(arg_uuid))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO),
+ "Short write while sending machine ID.");
+
+ l = send_one_fd(notify_socket, fd, 0);
+ if (l < 0)
+ return log_error_errno(l, "Failed to send notify fd: %m");
+
+ pid_socket = safe_close(pid_socket);
+ uuid_socket = safe_close(uuid_socket);
+ notify_socket = safe_close(notify_socket);
+ master_pty_socket = safe_close(master_pty_socket);
+ kmsg_socket = safe_close(kmsg_socket);
+ rtnl_socket = safe_close(rtnl_socket);
+ netns_fd = safe_close(netns_fd);
+
+ return 0;
+}
+
+static int uid_shift_pick(uid_t *shift, LockFile *ret_lock_file) {
+ bool tried_hashed = false;
+ unsigned n_tries = 100;
+ uid_t candidate;
+ int r;
+
+ assert(shift);
+ assert(ret_lock_file);
+ assert(arg_userns_mode == USER_NAMESPACE_PICK);
+ assert(arg_uid_range == 0x10000U);
+
+ candidate = *shift;
+
+ (void) mkdir("/run/systemd/nspawn-uid", 0755);
+
+ for (;;) {
+ char lock_path[STRLEN("/run/systemd/nspawn-uid/") + DECIMAL_STR_MAX(uid_t) + 1];
+ _cleanup_(release_lock_file) LockFile lf = LOCK_FILE_INIT;
+
+ if (--n_tries <= 0)
+ return -EBUSY;
+
+ if (candidate < CONTAINER_UID_BASE_MIN || candidate > CONTAINER_UID_BASE_MAX)
+ goto next;
+ if ((candidate & UINT32_C(0xFFFF)) != 0)
+ goto next;
+
+ xsprintf(lock_path, "/run/systemd/nspawn-uid/" UID_FMT, candidate);
+ r = make_lock_file(lock_path, LOCK_EX|LOCK_NB, &lf);
+ if (r == -EBUSY) /* Range already taken by another nspawn instance */
+ goto next;
+ if (r < 0)
+ return r;
+
+ /* Make some superficial checks whether the range is currently known in the user database */
+ if (getpwuid(candidate))
+ goto next;
+ if (getpwuid(candidate + UINT32_C(0xFFFE)))
+ goto next;
+ if (getgrgid(candidate))
+ goto next;
+ if (getgrgid(candidate + UINT32_C(0xFFFE)))
+ goto next;
+
+ *ret_lock_file = lf;
+ lf = (struct LockFile) LOCK_FILE_INIT;
+ *shift = candidate;
+ return 0;
+
+ next:
+ if (arg_machine && !tried_hashed) {
+ /* Try to hash the base from the container name */
+
+ static const uint8_t hash_key[] = {
+ 0xe1, 0x56, 0xe0, 0xf0, 0x4a, 0xf0, 0x41, 0xaf,
+ 0x96, 0x41, 0xcf, 0x41, 0x33, 0x94, 0xff, 0x72
+ };
+
+ candidate = (uid_t) siphash24(arg_machine, strlen(arg_machine), hash_key);
+
+ tried_hashed = true;
+ } else
+ random_bytes(&candidate, sizeof(candidate));
+
+ candidate = (candidate % (CONTAINER_UID_BASE_MAX - CONTAINER_UID_BASE_MIN)) + CONTAINER_UID_BASE_MIN;
+ candidate &= (uid_t) UINT32_C(0xFFFF0000);
+ }
+}
+
+static int add_one_uid_map(
+ char **p,
+ uid_t container_uid,
+ uid_t host_uid,
+ uid_t range) {
+
+ return strextendf(p,
+ UID_FMT " " UID_FMT " " UID_FMT "\n",
+ container_uid, host_uid, range);
+}
+
+static int make_uid_map_string(
+ const uid_t bind_user_uid[],
+ size_t n_bind_user_uid,
+ size_t offset,
+ char **ret) {
+
+ _cleanup_free_ char *s = NULL;
+ uid_t previous_uid = 0;
+ int r;
+
+ assert(n_bind_user_uid == 0 || bind_user_uid);
+ assert(IN_SET(offset, 0, 2)); /* used to switch between UID and GID map */
+ assert(ret);
+
+ /* The bind_user_uid[] array is a series of 4 uid_t values, for each --bind-user= entry one
+ * quadruplet, consisting of host and container UID + GID. */
+
+ for (size_t i = 0; i < n_bind_user_uid; i++) {
+ uid_t payload_uid = bind_user_uid[i*4+offset],
+ host_uid = bind_user_uid[i*4+offset+1];
+
+ assert(previous_uid <= payload_uid);
+ assert(payload_uid < arg_uid_range);
+
+ /* Add a range to close the gap to previous entry */
+ if (payload_uid > previous_uid) {
+ r = add_one_uid_map(&s, previous_uid, arg_uid_shift + previous_uid, payload_uid - previous_uid);
+ if (r < 0)
+ return r;
+ }
+
+ /* Map this specific user */
+ r = add_one_uid_map(&s, payload_uid, host_uid, 1);
+ if (r < 0)
+ return r;
+
+ previous_uid = payload_uid + 1;
+ }
+
+ /* And add a range to close the gap to finish the range */
+ if (arg_uid_range > previous_uid) {
+ r = add_one_uid_map(&s, previous_uid, arg_uid_shift + previous_uid, arg_uid_range - previous_uid);
+ if (r < 0)
+ return r;
+ }
+
+ assert(s);
+
+ *ret = TAKE_PTR(s);
+ return 0;
+}
+
+static int setup_uid_map(
+ pid_t pid,
+ const uid_t bind_user_uid[],
+ size_t n_bind_user_uid) {
+
+ char uid_map[STRLEN("/proc//uid_map") + DECIMAL_STR_MAX(uid_t) + 1];
+ _cleanup_free_ char *s = NULL;
+ int r;
+
+ assert(pid > 1);
+
+ /* Build the UID map string */
+ if (make_uid_map_string(bind_user_uid, n_bind_user_uid, 0, &s) < 0) /* offset=0 contains the UID pair */
+ return log_oom();
+
+ xsprintf(uid_map, "/proc/" PID_FMT "/uid_map", pid);
+ r = write_string_file(uid_map, s, WRITE_STRING_FILE_DISABLE_BUFFER);
+ if (r < 0)
+ return log_error_errno(r, "Failed to write UID map: %m");
+
+ /* And now build the GID map string */
+ s = mfree(s);
+ if (make_uid_map_string(bind_user_uid, n_bind_user_uid, 2, &s) < 0) /* offset=2 contains the GID pair */
+ return log_oom();
+
+ xsprintf(uid_map, "/proc/" PID_FMT "/gid_map", pid);
+ r = write_string_file(uid_map, s, WRITE_STRING_FILE_DISABLE_BUFFER);
+ if (r < 0)
+ return log_error_errno(r, "Failed to write GID map: %m");
+
+ return 0;
+}
+
+static int nspawn_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
+ char buf[NOTIFY_BUFFER_MAX+1];
+ char *p = NULL;
+ struct iovec iovec = {
+ .iov_base = buf,
+ .iov_len = sizeof(buf)-1,
+ };
+ CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
+ CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)) control;
+ struct msghdr msghdr = {
+ .msg_iov = &iovec,
+ .msg_iovlen = 1,
+ .msg_control = &control,
+ .msg_controllen = sizeof(control),
+ };
+ struct ucred *ucred;
+ ssize_t n;
+ pid_t inner_child_pid;
+ _cleanup_strv_free_ char **tags = NULL;
+ int r;
+
+ assert(userdata);
+
+ inner_child_pid = PTR_TO_PID(userdata);
+
+ if (revents != EPOLLIN) {
+ log_warning("Got unexpected poll event for notify fd.");
+ return 0;
+ }
+
+ n = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
+ if (n < 0) {
+ if (ERRNO_IS_TRANSIENT(n))
+ return 0;
+ if (n == -EXFULL) {
+ log_warning("Got message with truncated control data (too many fds sent?), ignoring.");
+ return 0;
+ }
+ return log_warning_errno(n, "Couldn't read notification socket: %m");
+ }
+
+ cmsg_close_all(&msghdr);
+
+ ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
+ if (!ucred || ucred->pid != inner_child_pid) {
+ log_debug("Received notify message without valid credentials. Ignoring.");
+ return 0;
+ }
+
+ if ((size_t) n >= sizeof(buf)) {
+ log_warning("Received notify message exceeded maximum size. Ignoring.");
+ return 0;
+ }
+
+ buf[n] = 0;
+ tags = strv_split(buf, "\n\r");
+ if (!tags)
+ return log_oom();
+
+ if (strv_contains(tags, "READY=1")) {
+ r = sd_notify(false, "READY=1\n");
+ if (r < 0)
+ log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
+ }
+
+ p = strv_find_startswith(tags, "STATUS=");
+ if (p)
+ (void) sd_notifyf(false, "STATUS=Container running: %s", p);
+
+ return 0;
+}
+
+static int setup_notify_parent(sd_event *event, int fd, pid_t *inner_child_pid, sd_event_source **notify_event_source) {
+ int r;
+
+ r = sd_event_add_io(event, notify_event_source, fd, EPOLLIN, nspawn_dispatch_notify_fd, inner_child_pid);
+ if (r < 0)
+ return log_error_errno(r, "Failed to allocate notify event source: %m");
+
+ (void) sd_event_source_set_description(*notify_event_source, "nspawn-notify");
+
+ return 0;
+}
+
+static int merge_settings(Settings *settings, const char *path) {
+ int rl;
+
+ assert(settings);
+ assert(path);
+
+ /* Copy over bits from the settings, unless they have been explicitly masked by command line switches. Note
+ * that this steals the fields of the Settings* structure, and hence modifies it. */
+
+ if ((arg_settings_mask & SETTING_START_MODE) == 0 &&
+ settings->start_mode >= 0) {
+ arg_start_mode = settings->start_mode;
+ strv_free_and_replace(arg_parameters, settings->parameters);
+ }
+
+ if ((arg_settings_mask & SETTING_EPHEMERAL) == 0 &&
+ settings->ephemeral >= 0)
+ arg_ephemeral = settings->ephemeral;
+
+ if ((arg_settings_mask & SETTING_DIRECTORY) == 0 &&
+ settings->root) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring root directory setting, file %s is not trusted.", path);
+ else
+ free_and_replace(arg_directory, settings->root);
+ }
+
+ if ((arg_settings_mask & SETTING_PIVOT_ROOT) == 0 &&
+ settings->pivot_root_new) {
+ free_and_replace(arg_pivot_root_new, settings->pivot_root_new);
+ free_and_replace(arg_pivot_root_old, settings->pivot_root_old);
+ }
+
+ if ((arg_settings_mask & SETTING_WORKING_DIRECTORY) == 0 &&
+ settings->working_directory)
+ free_and_replace(arg_chdir, settings->working_directory);
+
+ if ((arg_settings_mask & SETTING_ENVIRONMENT) == 0 &&
+ settings->environment)
+ strv_free_and_replace(arg_setenv, settings->environment);
+
+ if ((arg_settings_mask & SETTING_USER) == 0) {
+
+ if (settings->user)
+ free_and_replace(arg_user, settings->user);
+
+ if (uid_is_valid(settings->uid))
+ arg_uid = settings->uid;
+ if (gid_is_valid(settings->gid))
+ arg_gid = settings->gid;
+ if (settings->n_supplementary_gids > 0) {
+ free_and_replace(arg_supplementary_gids, settings->supplementary_gids);
+ arg_n_supplementary_gids = settings->n_supplementary_gids;
+ }
+ }
+
+ if ((arg_settings_mask & SETTING_CAPABILITY) == 0) {
+ uint64_t plus, minus;
+ uint64_t network_minus = 0;
+ uint64_t ambient;
+
+ /* Note that we copy both the simple plus/minus caps here, and the full quintet from the
+ * Settings structure */
+
+ plus = settings->capability;
+ minus = settings->drop_capability;
+
+ if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
+ settings_network_configured(settings)) {
+ if (settings_private_network(settings))
+ plus |= UINT64_C(1) << CAP_NET_ADMIN;
+ else
+ network_minus |= UINT64_C(1) << CAP_NET_ADMIN;
+ }
+
+ if (!arg_settings_trusted && plus != 0) {
+ if (settings->capability != 0)
+ log_warning("Ignoring Capability= setting, file %s is not trusted.", path);
+ } else {
+ arg_caps_retain &= ~network_minus;
+ arg_caps_retain |= plus;
+ }
+
+ arg_caps_retain &= ~minus;
+
+ /* Copy the full capabilities over too */
+ if (capability_quintet_is_set(&settings->full_capabilities)) {
+ if (!arg_settings_trusted)
+ log_warning("Ignoring capability settings, file %s is not trusted.", path);
+ else
+ arg_full_capabilities = settings->full_capabilities;
+ }
+
+ ambient = settings->ambient_capability;
+ if (!arg_settings_trusted && ambient != 0)
+ log_warning("Ignoring AmbientCapability= setting, file %s is not trusted.", path);
+ else
+ arg_caps_ambient |= ambient;
+ }
+
+ if ((arg_settings_mask & SETTING_KILL_SIGNAL) == 0 &&
+ settings->kill_signal > 0)
+ arg_kill_signal = settings->kill_signal;
+
+ if ((arg_settings_mask & SETTING_PERSONALITY) == 0 &&
+ settings->personality != PERSONALITY_INVALID)
+ arg_personality = settings->personality;
+
+ if ((arg_settings_mask & SETTING_MACHINE_ID) == 0 &&
+ !sd_id128_is_null(settings->machine_id)) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring MachineID= setting, file %s is not trusted.", path);
+ else
+ arg_uuid = settings->machine_id;
+ }
+
+ if ((arg_settings_mask & SETTING_READ_ONLY) == 0 &&
+ settings->read_only >= 0)
+ arg_read_only = settings->read_only;
+
+ if ((arg_settings_mask & SETTING_VOLATILE_MODE) == 0 &&
+ settings->volatile_mode != _VOLATILE_MODE_INVALID)
+ arg_volatile_mode = settings->volatile_mode;
+
+ if ((arg_settings_mask & SETTING_CUSTOM_MOUNTS) == 0 &&
+ settings->n_custom_mounts > 0) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring TemporaryFileSystem=, Bind= and BindReadOnly= settings, file %s is not trusted.", path);
+ else {
+ custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
+ arg_custom_mounts = TAKE_PTR(settings->custom_mounts);
+ arg_n_custom_mounts = settings->n_custom_mounts;
+ settings->n_custom_mounts = 0;
+ }
+ }
+
+ if ((arg_settings_mask & SETTING_NETWORK) == 0 &&
+ settings_network_configured(settings)) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring network settings, file %s is not trusted.", path);
+ else {
+ arg_network_veth = settings_network_veth(settings);
+ arg_private_network = settings_private_network(settings);
+
+ strv_free_and_replace(arg_network_interfaces, settings->network_interfaces);
+ strv_free_and_replace(arg_network_macvlan, settings->network_macvlan);
+ strv_free_and_replace(arg_network_ipvlan, settings->network_ipvlan);
+ strv_free_and_replace(arg_network_veth_extra, settings->network_veth_extra);
+
+ free_and_replace(arg_network_bridge, settings->network_bridge);
+ free_and_replace(arg_network_zone, settings->network_zone);
+
+ free_and_replace(arg_network_namespace_path, settings->network_namespace_path);
+ }
+ }
+
+ if ((arg_settings_mask & SETTING_EXPOSE_PORTS) == 0 &&
+ settings->expose_ports) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring Port= setting, file %s is not trusted.", path);
+ else {
+ expose_port_free_all(arg_expose_ports);
+ arg_expose_ports = TAKE_PTR(settings->expose_ports);
+ }
+ }
+
+ if ((arg_settings_mask & SETTING_USERNS) == 0 &&
+ settings->userns_mode != _USER_NAMESPACE_MODE_INVALID) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring PrivateUsers= and PrivateUsersChown= settings, file %s is not trusted.", path);
+ else {
+ arg_userns_mode = settings->userns_mode;
+ arg_uid_shift = settings->uid_shift;
+ arg_uid_range = settings->uid_range;
+ arg_userns_ownership = settings->userns_ownership;
+ }
+ }
+
+ if ((arg_settings_mask & SETTING_BIND_USER) == 0 &&
+ !strv_isempty(settings->bind_user))
+ strv_free_and_replace(arg_bind_user, settings->bind_user);
+
+ if ((arg_settings_mask & SETTING_NOTIFY_READY) == 0 &&
+ settings->notify_ready >= 0)
+ arg_notify_ready = settings->notify_ready;
+
+ if ((arg_settings_mask & SETTING_SYSCALL_FILTER) == 0) {
+
+ if (!strv_isempty(settings->syscall_allow_list) || !strv_isempty(settings->syscall_deny_list)) {
+ if (!arg_settings_trusted && !strv_isempty(settings->syscall_allow_list))
+ log_warning("Ignoring SystemCallFilter= settings, file %s is not trusted.", path);
+ else {
+ strv_free_and_replace(arg_syscall_allow_list, settings->syscall_allow_list);
+ strv_free_and_replace(arg_syscall_deny_list, settings->syscall_deny_list);
+ }
+ }
+
+#if HAVE_SECCOMP
+ if (settings->seccomp) {
+ if (!arg_settings_trusted)
+ log_warning("Ignoring SECCOMP filter, file %s is not trusted.", path);
+ else {
+ seccomp_release(arg_seccomp);
+ arg_seccomp = TAKE_PTR(settings->seccomp);
+ }
+ }
+#endif
+ }
+
+ for (rl = 0; rl < _RLIMIT_MAX; rl ++) {
+ if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)))
+ continue;
+
+ if (!settings->rlimit[rl])
+ continue;
+
+ if (!arg_settings_trusted) {
+ log_warning("Ignoring Limit%s= setting, file '%s' is not trusted.", rlimit_to_string(rl), path);
+ continue;
+ }
+
+ free_and_replace(arg_rlimit[rl], settings->rlimit[rl]);
+ }
+
+ if ((arg_settings_mask & SETTING_HOSTNAME) == 0 &&
+ settings->hostname)
+ free_and_replace(arg_hostname, settings->hostname);
+
+ if ((arg_settings_mask & SETTING_NO_NEW_PRIVILEGES) == 0 &&
+ settings->no_new_privileges >= 0)
+ arg_no_new_privileges = settings->no_new_privileges;
+
+ if ((arg_settings_mask & SETTING_OOM_SCORE_ADJUST) == 0 &&
+ settings->oom_score_adjust_set) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring OOMScoreAdjust= setting, file '%s' is not trusted.", path);
+ else {
+ arg_oom_score_adjust = settings->oom_score_adjust;
+ arg_oom_score_adjust_set = true;
+ }
+ }
+
+ if ((arg_settings_mask & SETTING_CPU_AFFINITY) == 0 &&
+ settings->cpu_set.set) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring CPUAffinity= setting, file '%s' is not trusted.", path);
+ else {
+ cpu_set_reset(&arg_cpu_set);
+ arg_cpu_set = settings->cpu_set;
+ settings->cpu_set = (CPUSet) {};
+ }
+ }
+
+ if ((arg_settings_mask & SETTING_RESOLV_CONF) == 0 &&
+ settings->resolv_conf != _RESOLV_CONF_MODE_INVALID)
+ arg_resolv_conf = settings->resolv_conf;
+
+ if ((arg_settings_mask & SETTING_LINK_JOURNAL) == 0 &&
+ settings->link_journal != _LINK_JOURNAL_INVALID) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring journal link setting, file '%s' is not trusted.", path);
+ else {
+ arg_link_journal = settings->link_journal;
+ arg_link_journal_try = settings->link_journal_try;
+ }
+ }
+
+ if ((arg_settings_mask & SETTING_TIMEZONE) == 0 &&
+ settings->timezone != _TIMEZONE_MODE_INVALID)
+ arg_timezone = settings->timezone;
+
+ if ((arg_settings_mask & SETTING_SLICE) == 0 &&
+ settings->slice) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring slice setting, file '%s' is not trusted.", path);
+ else
+ free_and_replace(arg_slice, settings->slice);
+ }
+
+ if ((arg_settings_mask & SETTING_USE_CGNS) == 0 &&
+ settings->use_cgns >= 0) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring cgroup namespace setting, file '%s' is not trusted.", path);
+ else
+ arg_use_cgns = settings->use_cgns;
+ }
+
+ if ((arg_settings_mask & SETTING_CLONE_NS_FLAGS) == 0 &&
+ settings->clone_ns_flags != ULONG_MAX) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring namespace setting, file '%s' is not trusted.", path);
+ else
+ arg_clone_ns_flags = settings->clone_ns_flags;
+ }
+
+ if ((arg_settings_mask & SETTING_CONSOLE_MODE) == 0 &&
+ settings->console_mode >= 0) {
+
+ if (!arg_settings_trusted)
+ log_warning("Ignoring console mode setting, file '%s' is not trusted.", path);
+ else
+ arg_console_mode = settings->console_mode;
+ }
+
+ if ((arg_settings_mask & SETTING_SUPPRESS_SYNC) == 0 &&
+ settings->suppress_sync >= 0)
+ arg_suppress_sync = settings->suppress_sync;
+
+ /* The following properties can only be set through the OCI settings logic, not from the command line, hence we
+ * don't consult arg_settings_mask for them. */
+
+ sd_bus_message_unref(arg_property_message);
+ arg_property_message = TAKE_PTR(settings->properties);
+
+ arg_console_width = settings->console_width;
+ arg_console_height = settings->console_height;
+
+ device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
+ arg_extra_nodes = TAKE_PTR(settings->extra_nodes);
+ arg_n_extra_nodes = settings->n_extra_nodes;
+ settings->n_extra_nodes = 0;
+
+ return 0;
+}
+
+static int load_settings(void) {
+ _cleanup_(settings_freep) Settings *settings = NULL;
+ _cleanup_fclose_ FILE *f = NULL;
+ _cleanup_free_ char *p = NULL;
+ int r;
+
+ if (arg_oci_bundle)
+ return 0;
+
+ /* If all settings are masked, there's no point in looking for
+ * the settings file */
+ if (FLAGS_SET(arg_settings_mask, _SETTINGS_MASK_ALL))
+ return 0;
+
+ /* We first look in the admin's directories in /etc and /run */
+ FOREACH_STRING(i, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
+ _cleanup_free_ char *j = NULL;
+
+ j = path_join(i, arg_settings_filename);
+ if (!j)
+ return log_oom();
+
+ f = fopen(j, "re");
+ if (f) {
+ p = TAKE_PTR(j);
+
+ /* By default, we trust configuration from /etc and /run */
+ if (arg_settings_trusted < 0)
+ arg_settings_trusted = true;
+
+ break;
+ }
+
+ if (errno != ENOENT)
+ return log_error_errno(errno, "Failed to open %s: %m", j);
+ }
+
+ if (!f) {
+ /* After that, let's look for a file next to the
+ * actual image we shall boot. */
+
+ if (arg_image) {
+ p = file_in_same_dir(arg_image, arg_settings_filename);
+ if (!p)
+ return log_oom();
+ } else if (arg_directory && !path_equal(arg_directory, "/")) {
+ p = file_in_same_dir(arg_directory, arg_settings_filename);
+ if (!p)
+ return log_oom();
+ }
+
+ if (p) {
+ f = fopen(p, "re");
+ if (!f && errno != ENOENT)
+ return log_error_errno(errno, "Failed to open %s: %m", p);
+
+ /* By default, we do not trust configuration from /var/lib/machines */
+ if (arg_settings_trusted < 0)
+ arg_settings_trusted = false;
+ }
+ }
+
+ if (!f)
+ return 0;
+
+ log_debug("Settings are trusted: %s", yes_no(arg_settings_trusted));
+
+ r = settings_load(f, p, &settings);
+ if (r < 0)
+ return r;
+
+ return merge_settings(settings, p);
+}
+
+static int load_oci_bundle(void) {
+ _cleanup_(settings_freep) Settings *settings = NULL;
+ int r;
+
+ if (!arg_oci_bundle)
+ return 0;
+
+ /* By default let's trust OCI bundles */
+ if (arg_settings_trusted < 0)
+ arg_settings_trusted = true;
+
+ r = oci_load(NULL, arg_oci_bundle, &settings);
+ if (r < 0)
+ return r;
+
+ return merge_settings(settings, arg_oci_bundle);
+}
+
+static int run_container(
+ DissectedImage *dissected_image,
+ bool secondary,
+ FDSet *fds,
+ char veth_name[IFNAMSIZ], bool *veth_created,
+ struct ExposeArgs *expose_args,
+ int *master, pid_t *pid, int *ret) {
+
+ static const struct sigaction sa = {
+ .sa_handler = nop_signal_handler,
+ .sa_flags = SA_NOCLDSTOP|SA_RESTART,
+ };
+
+ _cleanup_(release_lock_file) LockFile uid_shift_lock = LOCK_FILE_INIT;
+ _cleanup_close_ int etc_passwd_lock = -1;
+ _cleanup_close_pair_ int
+ kmsg_socket_pair[2] = { -1, -1 },
+ rtnl_socket_pair[2] = { -1, -1 },
+ pid_socket_pair[2] = { -1, -1 },
+ uuid_socket_pair[2] = { -1, -1 },
+ notify_socket_pair[2] = { -1, -1 },
+ uid_shift_socket_pair[2] = { -1, -1 },
+ master_pty_socket_pair[2] = { -1, -1 },
+ unified_cgroup_hierarchy_socket_pair[2] = { -1, -1};
+
+ _cleanup_close_ int notify_socket = -1;
+ _cleanup_(barrier_destroy) Barrier barrier = BARRIER_NULL;
+ _cleanup_(sd_event_source_unrefp) sd_event_source *notify_event_source = NULL;
+ _cleanup_(sd_event_unrefp) sd_event *event = NULL;
+ _cleanup_(pty_forward_freep) PTYForward *forward = NULL;
+ _cleanup_(sd_netlink_unrefp) sd_netlink *rtnl = NULL;
+ _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
+ _cleanup_free_ uid_t *bind_user_uid = NULL;
+ size_t n_bind_user_uid = 0;
+ ContainerStatus container_status = 0;
+ int ifi = 0, r;
+ ssize_t l;
+ sigset_t mask_chld;
+ _cleanup_close_ int child_netns_fd = -1;
+
+ assert_se(sigemptyset(&mask_chld) == 0);
+ assert_se(sigaddset(&mask_chld, SIGCHLD) == 0);
+
+ if (arg_userns_mode == USER_NAMESPACE_PICK) {
+ /* When we shall pick the UID/GID range, let's first lock /etc/passwd, so that we can safely
+ * check with getpwuid() if the specific user already exists. Note that /etc might be
+ * read-only, in which case this will fail with EROFS. But that's really OK, as in that case we
+ * can be reasonably sure that no users are going to be added. Note that getpwuid() checks are
+ * really just an extra safety net. We kinda assume that the UID range we allocate from is
+ * really ours. */
+
+ etc_passwd_lock = take_etc_passwd_lock(NULL);
+ if (etc_passwd_lock < 0 && etc_passwd_lock != -EROFS)
+ return log_error_errno(etc_passwd_lock, "Failed to take /etc/passwd lock: %m");
+ }
+
+ r = barrier_create(&barrier);
+ if (r < 0)
+ return log_error_errno(r, "Cannot initialize IPC barrier: %m");
+
+ if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, kmsg_socket_pair) < 0)
+ return log_error_errno(errno, "Failed to create kmsg socket pair: %m");
+
+ if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, rtnl_socket_pair) < 0)
+ return log_error_errno(errno, "Failed to create rtnl socket pair: %m");
+
+ if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, pid_socket_pair) < 0)
+ return log_error_errno(errno, "Failed to create pid socket pair: %m");
+
+ if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uuid_socket_pair) < 0)
+ return log_error_errno(errno, "Failed to create id socket pair: %m");
+
+ if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, notify_socket_pair) < 0)
+ return log_error_errno(errno, "Failed to create notify socket pair: %m");
+
+ if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, master_pty_socket_pair) < 0)
+ return log_error_errno(errno, "Failed to create console socket pair: %m");
+
+ if (arg_userns_mode != USER_NAMESPACE_NO)
+ if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, uid_shift_socket_pair) < 0)
+ return log_error_errno(errno, "Failed to create uid shift socket pair: %m");
+
+ if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN)
+ if (socketpair(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0, unified_cgroup_hierarchy_socket_pair) < 0)
+ return log_error_errno(errno, "Failed to create unified cgroup socket pair: %m");
+
+ /* Child can be killed before execv(), so handle SIGCHLD in order to interrupt
+ * parent's blocking calls and give it a chance to call wait() and terminate. */
+ r = sigprocmask(SIG_UNBLOCK, &mask_chld, NULL);
+ if (r < 0)
+ return log_error_errno(errno, "Failed to change the signal mask: %m");
+
+ r = sigaction(SIGCHLD, &sa, NULL);
+ if (r < 0)
+ return log_error_errno(errno, "Failed to install SIGCHLD handler: %m");
+
+ if (arg_network_namespace_path) {
+ child_netns_fd = open(arg_network_namespace_path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
+ if (child_netns_fd < 0)
+ return log_error_errno(errno, "Cannot open file %s: %m", arg_network_namespace_path);
+
+ r = fd_is_ns(child_netns_fd, CLONE_NEWNET);
+ if (r == -EUCLEAN)
+ log_debug_errno(r, "Cannot determine if passed network namespace path '%s' really refers to a network namespace, assuming it does.", arg_network_namespace_path);
+ else if (r < 0)
+ return log_error_errno(r, "Failed to check %s fs type: %m", arg_network_namespace_path);
+ else if (r == 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Path %s doesn't refer to a network namespace, refusing.", arg_network_namespace_path);
+ }
+
+ *pid = raw_clone(SIGCHLD|CLONE_NEWNS);
+ if (*pid < 0)
+ return log_error_errno(errno, "clone() failed%s: %m",
+ errno == EINVAL ?
+ ", do you have namespace support enabled in your kernel? (You need UTS, IPC, PID and NET namespacing built in)" : "");
+
+ if (*pid == 0) {
+ /* The outer child only has a file system namespace. */
+ barrier_set_role(&barrier, BARRIER_CHILD);
+
+ kmsg_socket_pair[0] = safe_close(kmsg_socket_pair[0]);
+ rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
+ pid_socket_pair[0] = safe_close(pid_socket_pair[0]);
+ uuid_socket_pair[0] = safe_close(uuid_socket_pair[0]);
+ notify_socket_pair[0] = safe_close(notify_socket_pair[0]);
+ master_pty_socket_pair[0] = safe_close(master_pty_socket_pair[0]);
+ uid_shift_socket_pair[0] = safe_close(uid_shift_socket_pair[0]);
+ unified_cgroup_hierarchy_socket_pair[0] = safe_close(unified_cgroup_hierarchy_socket_pair[0]);
+
+ (void) reset_all_signal_handlers();
+ (void) reset_signal_mask();
+
+ r = outer_child(&barrier,
+ arg_directory,
+ dissected_image,
+ secondary,
+ pid_socket_pair[1],
+ uuid_socket_pair[1],
+ notify_socket_pair[1],
+ kmsg_socket_pair[1],
+ rtnl_socket_pair[1],
+ uid_shift_socket_pair[1],
+ master_pty_socket_pair[1],
+ unified_cgroup_hierarchy_socket_pair[1],
+ fds,
+ child_netns_fd);
+ if (r < 0)
+ _exit(EXIT_FAILURE);
+
+ _exit(EXIT_SUCCESS);
+ }
+
+ barrier_set_role(&barrier, BARRIER_PARENT);
+
+ fdset_close(fds);
+
+ kmsg_socket_pair[1] = safe_close(kmsg_socket_pair[1]);
+ rtnl_socket_pair[1] = safe_close(rtnl_socket_pair[1]);
+ pid_socket_pair[1] = safe_close(pid_socket_pair[1]);
+ uuid_socket_pair[1] = safe_close(uuid_socket_pair[1]);
+ notify_socket_pair[1] = safe_close(notify_socket_pair[1]);
+ master_pty_socket_pair[1] = safe_close(master_pty_socket_pair[1]);
+ uid_shift_socket_pair[1] = safe_close(uid_shift_socket_pair[1]);
+ unified_cgroup_hierarchy_socket_pair[1] = safe_close(unified_cgroup_hierarchy_socket_pair[1]);
+
+ if (arg_userns_mode != USER_NAMESPACE_NO) {
+ /* The child just let us know the UID shift it might have read from the image. */
+ l = recv(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, 0);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to read UID shift: %m");
+ if (l != sizeof arg_uid_shift)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading UID shift.");
+
+ if (arg_userns_mode == USER_NAMESPACE_PICK) {
+ /* If we are supposed to pick the UID shift, let's try to use the shift read from the
+ * image, but if that's already in use, pick a new one, and report back to the child,
+ * which one we now picked. */
+
+ r = uid_shift_pick(&arg_uid_shift, &uid_shift_lock);
+ if (r < 0)
+ return log_error_errno(r, "Failed to pick suitable UID/GID range: %m");
+
+ l = send(uid_shift_socket_pair[0], &arg_uid_shift, sizeof arg_uid_shift, MSG_NOSIGNAL);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to send UID shift: %m");
+ if (l != sizeof arg_uid_shift)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short write while writing UID shift.");
+ }
+
+ n_bind_user_uid = strv_length(arg_bind_user);
+ if (n_bind_user_uid > 0) {
+ /* Right after the UID shift, we'll receive the list of UID mappings for the
+ * --bind-user= logic. Always a quadruplet of payload and host UID + GID. */
+
+ bind_user_uid = new(uid_t, n_bind_user_uid*4);
+ if (!bind_user_uid)
+ return log_oom();
+
+ for (size_t i = 0; i < n_bind_user_uid; i++) {
+ l = recv(uid_shift_socket_pair[0], bind_user_uid + i*4, sizeof(uid_t)*4, 0);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to read user UID map pair: %m");
+ if (l != sizeof(uid_t)*4)
+ return log_full_errno(l == 0 ? LOG_DEBUG : LOG_WARNING,
+ SYNTHETIC_ERRNO(EIO),
+ "Short read while reading bind user UID pairs.");
+ }
+ }
+ }
+
+ if (arg_unified_cgroup_hierarchy == CGROUP_UNIFIED_UNKNOWN) {
+ /* The child let us know the support cgroup mode it might have read from the image. */
+ l = recv(unified_cgroup_hierarchy_socket_pair[0], &arg_unified_cgroup_hierarchy, sizeof(arg_unified_cgroup_hierarchy), 0);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to read cgroup mode: %m");
+ if (l != sizeof(arg_unified_cgroup_hierarchy))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading cgroup mode (%zi bytes).%s",
+ l, l == 0 ? " The child is most likely dead." : "");
+ }
+
+ /* Wait for the outer child. */
+ r = wait_for_terminate_and_check("(sd-namespace)", *pid, WAIT_LOG_ABNORMAL);
+ if (r < 0)
+ return r;
+ if (r != EXIT_SUCCESS)
+ return -EIO;
+
+ /* And now retrieve the PID of the inner child. */
+ l = recv(pid_socket_pair[0], pid, sizeof *pid, 0);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to read inner child PID: %m");
+ if (l != sizeof *pid)
+ return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading inner child PID.");
+
+ /* We also retrieve container UUID in case it was generated by outer child */
+ l = recv(uuid_socket_pair[0], &arg_uuid, sizeof arg_uuid, 0);
+ if (l < 0)
+ return log_error_errno(errno, "Failed to read container machine ID: %m");
+ if (l != sizeof(arg_uuid))
+ return log_error_errno(SYNTHETIC_ERRNO(EIO), "Short read while reading container machined ID.");
+
+ /* We also retrieve the socket used for notifications generated by outer child */
+ notify_socket = receive_one_fd(notify_socket_pair[0], 0);
+ if (notify_socket < 0)
+ return log_error_errno(notify_socket,
+ "Failed to receive notification socket from the outer child: %m");
+
+ log_debug("Init process invoked as PID "PID_FMT, *pid);
+
+ if (arg_userns_mode != USER_NAMESPACE_NO) {
+ if (!barrier_place_and_sync(&barrier)) /* #1 */
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
+
+ r = setup_uid_map(*pid, bind_user_uid, n_bind_user_uid);
+ if (r < 0)
+ return r;
+
+ (void) barrier_place(&barrier); /* #2 */
+ }
+
+ if (arg_private_network) {
+ if (!arg_network_namespace_path) {
+ /* Wait until the child has unshared its network namespace. */
+ if (!barrier_place_and_sync(&barrier)) /* #3 */
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early");
+ }
+
+ if (child_netns_fd < 0) {
+ /* Make sure we have an open file descriptor to the child's network
+ * namespace so it stays alive even if the child exits. */
+ r = namespace_open(*pid, NULL, NULL, &child_netns_fd, NULL, NULL);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open child network namespace: %m");
+ }
+
+ r = move_network_interfaces(child_netns_fd, arg_network_interfaces);
+ if (r < 0)
+ return r;
+
+ if (arg_network_veth) {
+ r = setup_veth(arg_machine, *pid, veth_name,
+ arg_network_bridge || arg_network_zone);
+ if (r < 0)
+ return r;
+ else if (r > 0)
+ ifi = r;
+
+ if (arg_network_bridge) {
+ /* Add the interface to a bridge */
+ r = setup_bridge(veth_name, arg_network_bridge, false);
+ if (r < 0)
+ return r;
+ if (r > 0)
+ ifi = r;
+ } else if (arg_network_zone) {
+ /* Add the interface to a bridge, possibly creating it */
+ r = setup_bridge(veth_name, arg_network_zone, true);
+ if (r < 0)
+ return r;
+ if (r > 0)
+ ifi = r;
+ }
+ }
+
+ r = setup_veth_extra(arg_machine, *pid, arg_network_veth_extra);
+ if (r < 0)
+ return r;
+
+ /* We created the primary and extra veth links now; let's remember this, so that we know to
+ remove them later on. Note that we don't bother with removing veth links that were created
+ here when their setup failed half-way, because in that case the kernel should be able to
+ remove them on its own, since they cannot be referenced by anything yet. */
+ *veth_created = true;
+
+ r = setup_macvlan(arg_machine, *pid, arg_network_macvlan);
+ if (r < 0)
+ return r;
+
+ r = setup_ipvlan(arg_machine, *pid, arg_network_ipvlan);
+ if (r < 0)
+ return r;
+ }
+
+ if (arg_register || !arg_keep_unit) {
+ r = sd_bus_default_system(&bus);
+ if (r < 0)
+ return log_error_errno(r, "Failed to open system bus: %m");
+
+ r = sd_bus_set_close_on_exit(bus, false);
+ if (r < 0)
+ return log_error_errno(r, "Failed to disable close-on-exit behaviour: %m");
+ }
+
+ if (!arg_keep_unit) {
+ /* When a new scope is created for this container, then we'll be registered as its controller, in which
+ * case PID 1 will send us a friendly RequestStop signal, when it is asked to terminate the
+ * scope. Let's hook into that, and cleanly shut down the container, and print a friendly message. */
+
+ r = sd_bus_match_signal_async(
+ bus,
+ NULL,
+ "org.freedesktop.systemd1",
+ NULL,
+ "org.freedesktop.systemd1.Scope",
+ "RequestStop",
+ on_request_stop, NULL, PID_TO_PTR(*pid));
+ if (r < 0)
+ return log_error_errno(r, "Failed to request RequestStop match: %m");
+ }
+
+ if (arg_register) {
+ r = register_machine(
+ bus,
+ arg_machine,
+ *pid,
+ arg_directory,
+ arg_uuid,
+ ifi,
+ arg_slice,
+ arg_custom_mounts, arg_n_custom_mounts,
+ arg_kill_signal,
+ arg_property,
+ arg_property_message,
+ arg_keep_unit,
+ arg_container_service_name);
+ if (r < 0)
+ return r;
+
+ } else if (!arg_keep_unit) {
+ r = allocate_scope(
+ bus,
+ arg_machine,
+ *pid,
+ arg_slice,
+ arg_custom_mounts, arg_n_custom_mounts,
+ arg_kill_signal,
+ arg_property,
+ arg_property_message);
+ if (r < 0)
+ return r;
+
+ } else if (arg_slice || arg_property)
+ log_notice("Machine and scope registration turned off, --slice= and --property= settings will have no effect.");
+
+ r = create_subcgroup(*pid, arg_keep_unit, arg_unified_cgroup_hierarchy);
+ if (r < 0)
+ return r;
+
+ r = sync_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
+ if (r < 0)
+ return r;
+
+ r = chown_cgroup(*pid, arg_unified_cgroup_hierarchy, arg_uid_shift);
+ if (r < 0)
+ return r;
+
+ /* Notify the child that the parent is ready with all
+ * its setup (including cgroup-ification), and that
+ * the child can now hand over control to the code to
+ * run inside the container. */
+ (void) barrier_place(&barrier); /* #4 */
+
+ /* Block SIGCHLD here, before notifying child.
+ * process_pty() will handle it with the other signals. */
+ assert_se(sigprocmask(SIG_BLOCK, &mask_chld, NULL) >= 0);
+
+ /* Reset signal to default */
+ r = default_signals(SIGCHLD);
+ if (r < 0)
+ return log_error_errno(r, "Failed to reset SIGCHLD: %m");
+
+ r = sd_event_new(&event);
+ if (r < 0)
+ return log_error_errno(r, "Failed to get default event source: %m");
+
+ (void) sd_event_set_watchdog(event, true);
+
+ if (bus) {
+ r = sd_bus_attach_event(bus, event, 0);
+ if (r < 0)
+ return log_error_errno(r, "Failed to attach bus to event loop: %m");
+ }
+
+ r = setup_notify_parent(event, notify_socket, PID_TO_PTR(*pid), &notify_event_source);
+ if (r < 0)
+ return r;
+
+ /* Let the child know that we are ready and wait that the child is completely ready now. */
+ if (!barrier_place_and_sync(&barrier)) /* #5 */
+ return log_error_errno(SYNTHETIC_ERRNO(ESRCH), "Child died too early.");
+
+ /* At this point we have made use of the UID we picked, and thus nss-systemd/systemd-machined.service
+ * will make them appear in getpwuid(), thus we can release the /etc/passwd lock. */
+ etc_passwd_lock = safe_close(etc_passwd_lock);
+
+ (void) sd_notifyf(false,
+ "STATUS=Container running.\n"
+ "X_NSPAWN_LEADER_PID=" PID_FMT, *pid);
+ if (!arg_notify_ready) {
+ r = sd_notify(false, "READY=1\n");
+ if (r < 0)
+ log_warning_errno(r, "Failed to send readiness notification, ignoring: %m");
+ }
+
+ if (arg_kill_signal > 0) {
+ /* Try to kill the init system on SIGINT or SIGTERM */
+ (void) sd_event_add_signal(event, NULL, SIGINT, on_orderly_shutdown, PID_TO_PTR(*pid));
+ (void) sd_event_add_signal(event, NULL, SIGTERM, on_orderly_shutdown, PID_TO_PTR(*pid));
+ } else {
+ /* Immediately exit */
+ (void) sd_event_add_signal(event, NULL, SIGINT, NULL, NULL);
+ (void) sd_event_add_signal(event, NULL, SIGTERM, NULL, NULL);
+ }
+
+ /* Exit when the child exits */
+ (void) sd_event_add_signal(event, NULL, SIGCHLD, on_sigchld, PID_TO_PTR(*pid));
+
+ if (arg_expose_ports) {
+ r = expose_port_watch_rtnl(event, rtnl_socket_pair[0], on_address_change, expose_args, &rtnl);
+ if (r < 0)
+ return r;
+
+ (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, AF_INET, &expose_args->address4);
+ (void) expose_port_execute(rtnl, &expose_args->fw_ctx, arg_expose_ports, AF_INET6, &expose_args->address6);
+ }
+
+ rtnl_socket_pair[0] = safe_close(rtnl_socket_pair[0]);
+
+ if (arg_console_mode != CONSOLE_PIPE) {
+ _cleanup_close_ int fd = -1;
+ PTYForwardFlags flags = 0;
+
+ /* Retrieve the master pty allocated by inner child */
+ fd = receive_one_fd(master_pty_socket_pair[0], 0);
+ if (fd < 0)
+ return log_error_errno(fd, "Failed to receive master pty from the inner child: %m");
+
+ switch (arg_console_mode) {
+
+ case CONSOLE_READ_ONLY:
+ flags |= PTY_FORWARD_READ_ONLY;
+
+ _fallthrough_;
+
+ case CONSOLE_INTERACTIVE:
+ flags |= PTY_FORWARD_IGNORE_VHANGUP;
+
+ r = pty_forward_new(event, fd, flags, &forward);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create PTY forwarder: %m");
+
+ if (arg_console_width != UINT_MAX || arg_console_height != UINT_MAX)
+ (void) pty_forward_set_width_height(forward,
+ arg_console_width,
+ arg_console_height);
+ break;
+
+ default:
+ assert(arg_console_mode == CONSOLE_PASSIVE);
+ }
+
+ *master = TAKE_FD(fd);
+ }
+
+ r = sd_event_loop(event);
+ if (r < 0)
+ return log_error_errno(r, "Failed to run event loop: %m");
+
+ if (forward) {
+ char last_char = 0;
+
+ (void) pty_forward_get_last_char(forward, &last_char);
+ forward = pty_forward_free(forward);
+
+ if (!arg_quiet && last_char != '\n')
+ putc('\n', stdout);
+ }
+
+ /* Kill if it is not dead yet anyway */
+ if (!arg_register && !arg_keep_unit && bus)
+ terminate_scope(bus, arg_machine);
+
+ /* Normally redundant, but better safe than sorry */
+ (void) kill(*pid, SIGKILL);
+
+ if (arg_private_network) {
+ /* Move network interfaces back to the parent network namespace. We use `safe_fork`
+ * to avoid having to move the parent to the child network namespace. */
+ r = safe_fork(NULL, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_WAIT|FORK_LOG, NULL);
+ if (r < 0)
+ return r;
+
+ if (r == 0) {
+ _cleanup_close_ int parent_netns_fd = -1;
+
+ r = namespace_open(getpid(), NULL, NULL, &parent_netns_fd, NULL, NULL);
+ if (r < 0) {
+ log_error_errno(r, "Failed to open parent network namespace: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ r = namespace_enter(-1, -1, child_netns_fd, -1, -1);
+ if (r < 0) {
+ log_error_errno(r, "Failed to enter child network namespace: %m");
+ _exit(EXIT_FAILURE);
+ }
+
+ r = move_network_interfaces(parent_netns_fd, arg_network_interfaces);
+ if (r < 0)
+ log_error_errno(r, "Failed to move network interfaces back to parent network namespace: %m");
+
+ _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
+ }
+ }
+
+ r = wait_for_container(TAKE_PID(*pid), &container_status);
+
+ /* Tell machined that we are gone. */
+ if (bus)
+ (void) unregister_machine(bus, arg_machine);
+
+ if (r < 0)
+ /* We failed to wait for the container, or the container exited abnormally. */
+ return r;
+ if (r > 0 || container_status == CONTAINER_TERMINATED) {
+ /* r > 0 → The container exited with a non-zero status.
+ * As a special case, we need to replace 133 with a different value,
+ * because 133 is special-cased in the service file to reboot the container.
+ * otherwise → The container exited with zero status and a reboot was not requested.
+ */
+ if (r == EXIT_FORCE_RESTART)
+ r = EXIT_FAILURE; /* replace 133 with the general failure code */
+ *ret = r;
+ return 0; /* finito */
+ }
+
+ /* CONTAINER_REBOOTED, loop again */
+
+ if (arg_keep_unit) {
+ /* Special handling if we are running as a service: instead of simply
+ * restarting the machine we want to restart the entire service, so let's
+ * inform systemd about this with the special exit code 133. The service
+ * file uses RestartForceExitStatus=133 so that this results in a full
+ * nspawn restart. This is necessary since we might have cgroup parameters
+ * set we want to have flushed out. */
+ *ret = EXIT_FORCE_RESTART;
+ return 0; /* finito */
+ }
+
+ expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, AF_INET, &expose_args->address4);
+ expose_port_flush(&expose_args->fw_ctx, arg_expose_ports, AF_INET6, &expose_args->address6);
+
+ (void) remove_veth_links(veth_name, arg_network_veth_extra);
+ *veth_created = false;
+ return 1; /* loop again */
+}
+
+static int initialize_rlimits(void) {
+ /* The default resource limits the kernel passes to PID 1, as per kernel 5.16. Let's pass our container payload
+ * the same values as the kernel originally passed to PID 1, in order to minimize differences between host and
+ * container execution environments. */
+
+ static const struct rlimit kernel_defaults[_RLIMIT_MAX] = {
+ [RLIMIT_AS] = { RLIM_INFINITY, RLIM_INFINITY },
+ [RLIMIT_CORE] = { 0, RLIM_INFINITY },
+ [RLIMIT_CPU] = { RLIM_INFINITY, RLIM_INFINITY },
+ [RLIMIT_DATA] = { RLIM_INFINITY, RLIM_INFINITY },
+ [RLIMIT_FSIZE] = { RLIM_INFINITY, RLIM_INFINITY },
+ [RLIMIT_LOCKS] = { RLIM_INFINITY, RLIM_INFINITY },
+ [RLIMIT_MEMLOCK] = { DEFAULT_RLIMIT_MEMLOCK, DEFAULT_RLIMIT_MEMLOCK },
+ [RLIMIT_MSGQUEUE] = { 819200, 819200 },
+ [RLIMIT_NICE] = { 0, 0 },
+ [RLIMIT_NOFILE] = { 1024, 4096 },
+ [RLIMIT_RSS] = { RLIM_INFINITY, RLIM_INFINITY },
+ [RLIMIT_RTPRIO] = { 0, 0 },
+ [RLIMIT_RTTIME] = { RLIM_INFINITY, RLIM_INFINITY },
+ [RLIMIT_STACK] = { 8388608, RLIM_INFINITY },
+
+ /* The kernel scales the default for RLIMIT_NPROC and RLIMIT_SIGPENDING based on the system's amount of
+ * RAM. To provide best compatibility we'll read these limits off PID 1 instead of hardcoding them
+ * here. This is safe as we know that PID 1 doesn't change these two limits and thus the original
+ * kernel's initialization should still be valid during runtime — at least if PID 1 is systemd. Note
+ * that PID 1 changes a number of other resource limits during early initialization which is why we
+ * don't read the other limits from PID 1 but prefer the static table above. */
+ };
+
+ int rl;
+
+ for (rl = 0; rl < _RLIMIT_MAX; rl++) {
+ /* Let's only fill in what the user hasn't explicitly configured anyway */
+ if ((arg_settings_mask & (SETTING_RLIMIT_FIRST << rl)) == 0) {
+ const struct rlimit *v;
+ struct rlimit buffer;
+
+ if (IN_SET(rl, RLIMIT_NPROC, RLIMIT_SIGPENDING)) {
+ /* For these two let's read the limits off PID 1. See above for an explanation. */
+
+ if (prlimit(1, rl, NULL, &buffer) < 0)
+ return log_error_errno(errno, "Failed to read resource limit RLIMIT_%s of PID 1: %m", rlimit_to_string(rl));
+
+ v = &buffer;
+ } else if (rl == RLIMIT_NOFILE) {
+ /* We nowadays bump RLIMIT_NOFILE's hard limit early in PID 1 for all
+ * userspace. Given that nspawn containers are often run without our PID 1,
+ * let's grant the containers a raised RLIMIT_NOFILE hard limit by default,
+ * so that container userspace gets similar resources as host userspace
+ * gets. */
+ buffer = kernel_defaults[rl];
+ buffer.rlim_max = MIN((rlim_t) read_nr_open(), (rlim_t) HIGH_RLIMIT_NOFILE);
+ v = &buffer;
+ } else
+ v = kernel_defaults + rl;
+
+ arg_rlimit[rl] = newdup(struct rlimit, v, 1);
+ if (!arg_rlimit[rl])
+ return log_oom();
+ }
+
+ if (DEBUG_LOGGING) {
+ _cleanup_free_ char *k = NULL;
+
+ (void) rlimit_format(arg_rlimit[rl], &k);
+ log_debug("Setting RLIMIT_%s to %s.", rlimit_to_string(rl), k);
+ }
+ }
+
+ return 0;
+}
+
+static int cant_be_in_netns(void) {
+ _cleanup_close_ int fd = -1;
+ struct ucred ucred;
+ int r;
+
+ /* Check if we are in the same netns as udev. If we aren't, then device monitoring (and thus waiting
+ * for loopback block devices) won't work, and we will hang. Detect this case and exit early with a
+ * nice message. */
+
+ if (!arg_image) /* only matters if --image= us used, i.e. we actually need to use loopback devices */
+ return 0;
+
+ fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_NONBLOCK|SOCK_CLOEXEC, 0);
+ if (fd < 0)
+ return log_error_errno(errno, "Failed to allocate udev control socket: %m");
+
+ r = connect_unix_path(fd, AT_FDCWD, "/run/udev/control");
+ if (r < 0) {
+ if (r == -ENOENT || ERRNO_IS_DISCONNECT(r))
+ return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Sorry, but --image= requires access to the host's /run/ hierarchy, since we need access to udev.");
+
+ return log_error_errno(r, "Failed to connect socket to udev control socket: %m");
+ }
+
+ r = getpeercred(fd, &ucred);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine peer of udev control socket: %m");
+
+ r = in_same_namespace(ucred.pid, 0, NAMESPACE_NET);
+ if (r < 0)
+ return log_error_errno(r, "Failed to determine network namespace of udev: %m");
+ if (r == 0)
+ return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
+ "Sorry, but --image= is only supported in the main network namespace, since we need access to udev/AF_NETLINK.");
+ return 0;
+}
+
+static int run(int argc, char *argv[]) {
+ bool secondary = false, remove_directory = false, remove_image = false,
+ veth_created = false, remove_tmprootdir = false;
+ _cleanup_close_ int master = -1;
+ _cleanup_fdset_free_ FDSet *fds = NULL;
+ int r, n_fd_passed, ret = EXIT_SUCCESS;
+ char veth_name[IFNAMSIZ] = "";
+ struct ExposeArgs expose_args = {};
+ _cleanup_(release_lock_file) LockFile tree_global_lock = LOCK_FILE_INIT, tree_local_lock = LOCK_FILE_INIT;
+ char tmprootdir[] = "/tmp/nspawn-root-XXXXXX";
+ _cleanup_(loop_device_unrefp) LoopDevice *loop = NULL;
+ _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
+ _cleanup_(fw_ctx_freep) FirewallContext *fw_ctx = NULL;
+ pid_t pid = 0;
+
+ log_parse_environment();
+ log_open();
+
+ r = parse_argv(argc, argv);
+ if (r <= 0)
+ goto finish;
+
+ if (geteuid() != 0) {
+ r = log_warning_errno(SYNTHETIC_ERRNO(EPERM),
+ argc >= 2 ? "Need to be root." :
+ "Need to be root (and some arguments are usually required).\nHint: try --help");
+ goto finish;
+ }
+
+ r = cant_be_in_netns();
+ if (r < 0)
+ goto finish;
+
+ r = initialize_rlimits();
+ if (r < 0)
+ goto finish;
+
+ r = load_oci_bundle();
+ if (r < 0)
+ goto finish;
+
+ r = determine_names();
+ if (r < 0)
+ goto finish;
+
+ r = load_settings();
+ if (r < 0)
+ goto finish;
+
+ /* If we're not unsharing the network namespace and are unsharing the user namespace, we won't have
+ * permissions to bind ports in the container, so let's drop the CAP_NET_BIND_SERVICE capability to
+ * indicate that. */
+ if (!arg_private_network && arg_userns_mode != USER_NAMESPACE_NO && arg_uid_shift > 0)
+ arg_caps_retain &= ~(UINT64_C(1) << CAP_NET_BIND_SERVICE);
+
+ r = cg_unified();
+ if (r < 0) {
+ log_error_errno(r, "Failed to determine whether the unified cgroups hierarchy is used: %m");
+ goto finish;
+ }
+
+ r = verify_arguments();
+ if (r < 0)
+ goto finish;
+
+ /* Reapply environment settings. */
+ (void) detect_unified_cgroup_hierarchy_from_environment();
+
+ /* Ignore SIGPIPE here, because we use splice() on the ptyfwd stuff and that will generate SIGPIPE if
+ * the result is closed. Note that the container payload child will reset signal mask+handler anyway,
+ * so just turning this off here means we only turn it off in nspawn itself, not any children. */
+ (void) ignore_signals(SIGPIPE);
+
+ n_fd_passed = sd_listen_fds(false);
+ if (n_fd_passed > 0) {
+ r = fdset_new_listen_fds(&fds, false);
+ if (r < 0) {
+ log_error_errno(r, "Failed to collect file descriptors: %m");
+ goto finish;
+ }
+ }
+
+ /* The "default" umask. This is appropriate for most file and directory
+ * operations performed by nspawn, and is the umask that will be used for
+ * the child. Functions like copy_devnodes() change the umask temporarily. */
+ umask(0022);
+
+ if (arg_directory) {
+ assert(!arg_image);
+
+ /* Safety precaution: let's not allow running images from the live host OS image, as long as
+ * /var from the host will propagate into container dynamically (because bad things happen if
+ * two systems write to the same /var). Let's allow it for the special cases where /var is
+ * either copied (i.e. --ephemeral) or replaced (i.e. --volatile=yes|state). */
+ if (path_equal(arg_directory, "/") && !(arg_ephemeral || IN_SET(arg_volatile_mode, VOLATILE_YES, VOLATILE_STATE))) {
+ log_error("Spawning container on root directory is not supported. Consider using --ephemeral, --volatile=yes or --volatile=state.");
+ r = -EINVAL;
+ goto finish;
+ }
+
+ if (arg_ephemeral) {
+ _cleanup_free_ char *np = NULL;
+
+ r = chase_symlinks_and_update(&arg_directory, 0);
+ if (r < 0)
+ goto finish;
+
+ /* If the specified path is a mount point we generate the new snapshot immediately
+ * inside it under a random name. However if the specified is not a mount point we
+ * create the new snapshot in the parent directory, just next to it. */
+ r = path_is_mount_point(arg_directory, NULL, 0);
+ if (r < 0) {
+ log_error_errno(r, "Failed to determine whether directory %s is mount point: %m", arg_directory);
+ goto finish;
+ }
+ if (r > 0)
+ r = tempfn_random_child(arg_directory, "machine.", &np);
+ else
+ r = tempfn_random(arg_directory, "machine.", &np);
+ if (r < 0) {
+ log_error_errno(r, "Failed to generate name for directory snapshot: %m");
+ goto finish;
+ }
+
+ /* We take an exclusive lock on this image, since it's our private, ephemeral copy
+ * only owned by us and no one else. */
+ r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
+ if (r < 0) {
+ log_error_errno(r, "Failed to lock %s: %m", np);
+ goto finish;
+ }
+
+ {
+ BLOCK_SIGNALS(SIGINT);
+ r = btrfs_subvol_snapshot(arg_directory, np,
+ (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
+ BTRFS_SNAPSHOT_FALLBACK_COPY |
+ BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
+ BTRFS_SNAPSHOT_RECURSIVE |
+ BTRFS_SNAPSHOT_QUOTA |
+ BTRFS_SNAPSHOT_SIGINT);
+ }
+ if (r == -EINTR) {
+ log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", np);
+ goto finish;
+ }
+ if (r < 0) {
+ log_error_errno(r, "Failed to create snapshot %s from %s: %m", np, arg_directory);
+ goto finish;
+ }
+
+ free_and_replace(arg_directory, np);
+ remove_directory = true;
+ } else {
+ r = chase_symlinks_and_update(&arg_directory, arg_template ? CHASE_NONEXISTENT : 0);
+ if (r < 0)
+ goto finish;
+
+ r = image_path_lock(arg_directory, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
+ if (r == -EBUSY) {
+ log_error_errno(r, "Directory tree %s is currently busy.", arg_directory);
+ goto finish;
+ }
+ if (r < 0) {
+ log_error_errno(r, "Failed to lock %s: %m", arg_directory);
+ goto finish;
+ }
+
+ if (arg_template) {
+ r = chase_symlinks_and_update(&arg_template, 0);
+ if (r < 0)
+ goto finish;
+
+ {
+ BLOCK_SIGNALS(SIGINT);
+ r = btrfs_subvol_snapshot(arg_template, arg_directory,
+ (arg_read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) |
+ BTRFS_SNAPSHOT_FALLBACK_COPY |
+ BTRFS_SNAPSHOT_FALLBACK_DIRECTORY |
+ BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE |
+ BTRFS_SNAPSHOT_RECURSIVE |
+ BTRFS_SNAPSHOT_QUOTA |
+ BTRFS_SNAPSHOT_SIGINT);
+ }
+ if (r == -EEXIST)
+ log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
+ "Directory %s already exists, not populating from template %s.", arg_directory, arg_template);
+ else if (r == -EINTR) {
+ log_error_errno(r, "Interrupted while copying file system tree to %s, removed again.", arg_directory);
+ goto finish;
+ } else if (r < 0) {
+ log_error_errno(r, "Couldn't create snapshot %s from %s: %m", arg_directory, arg_template);
+ goto finish;
+ } else
+ log_full(arg_quiet ? LOG_DEBUG : LOG_INFO,
+ "Populated %s from template %s.", arg_directory, arg_template);
+ }
+ }
+
+ if (arg_start_mode == START_BOOT) {
+ _cleanup_free_ char *b = NULL;
+ const char *p;
+
+ if (arg_pivot_root_new) {
+ b = path_join(arg_directory, arg_pivot_root_new);
+ if (!b) {
+ r = log_oom();
+ goto finish;
+ }
+
+ p = b;
+ } else
+ p = arg_directory;
+
+ if (path_is_os_tree(p) <= 0) {
+ r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Directory %s doesn't look like an OS root directory (os-release file is missing). Refusing.", p);
+ goto finish;
+ }
+ } else {
+ _cleanup_free_ char *p = NULL;
+
+ if (arg_pivot_root_new)
+ p = path_join(arg_directory, arg_pivot_root_new, "/usr/");
+ else
+ p = path_join(arg_directory, "/usr/");
+ if (!p) {
+ r = log_oom();
+ goto finish;
+ }
+
+ if (laccess(p, F_OK) < 0) {
+ r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+ "Directory %s doesn't look like it has an OS tree (/usr/ directory is missing). Refusing.", arg_directory);
+ goto finish;
+ }
+ }
+
+ } else {
+ DissectImageFlags dissect_image_flags =
+ DISSECT_IMAGE_GENERIC_ROOT |
+ DISSECT_IMAGE_REQUIRE_ROOT |
+ DISSECT_IMAGE_RELAX_VAR_CHECK |
+ DISSECT_IMAGE_USR_NO_ROOT |
+ DISSECT_IMAGE_ADD_PARTITION_DEVICES |
+ DISSECT_IMAGE_PIN_PARTITION_DEVICES;
+ assert(arg_image);
+ assert(!arg_template);
+
+ r = chase_symlinks_and_update(&arg_image, 0);
+ if (r < 0)
+ goto finish;
+
+ if (arg_ephemeral) {
+ _cleanup_free_ char *np = NULL;
+
+ r = tempfn_random(arg_image, "machine.", &np);
+ if (r < 0) {
+ log_error_errno(r, "Failed to generate name for image snapshot: %m");
+ goto finish;
+ }
+
+ /* Always take an exclusive lock on our own ephemeral copy. */
+ r = image_path_lock(np, LOCK_EX|LOCK_NB, &tree_global_lock, &tree_local_lock);
+ if (r < 0) {
+ r = log_error_errno(r, "Failed to create image lock: %m");
+ goto finish;
+ }
+
+ {
+ BLOCK_SIGNALS(SIGINT);
+ r = copy_file(arg_image, np, O_EXCL, arg_read_only ? 0400 : 0600, FS_NOCOW_FL, FS_NOCOW_FL, COPY_REFLINK|COPY_CRTIME|COPY_SIGINT);
+ }
+ if (r == -EINTR) {
+ log_error_errno(r, "Interrupted while copying image file to %s, removed again.", np);
+ goto finish;
+ }
+ if (r < 0) {
+ r = log_error_errno(r, "Failed to copy image file: %m");
+ goto finish;
+ }
+
+ free_and_replace(arg_image, np);
+ remove_image = true;
+ } else {
+ r = image_path_lock(arg_image, (arg_read_only ? LOCK_SH : LOCK_EX) | LOCK_NB, &tree_global_lock, &tree_local_lock);
+ if (r == -EBUSY) {
+ r = log_error_errno(r, "Disk image %s is currently busy.", arg_image);
+ goto finish;
+ }
+ if (r < 0) {
+ r = log_error_errno(r, "Failed to create image lock: %m");
+ goto finish;
+ }
+
+ r = verity_settings_load(
+ &arg_verity_settings,
+ arg_image, NULL, NULL);
+ if (r < 0) {
+ log_error_errno(r, "Failed to read verity artefacts for %s: %m", arg_image);
+ goto finish;
+ }
+
+ if (arg_verity_settings.data_path)
+ dissect_image_flags |= DISSECT_IMAGE_NO_PARTITION_TABLE;
+ }
+
+ if (!mkdtemp(tmprootdir)) {
+ r = log_error_errno(errno, "Failed to create temporary directory: %m");
+ goto finish;
+ }
+
+ remove_tmprootdir = true;
+
+ arg_directory = strdup(tmprootdir);
+ if (!arg_directory) {
+ r = log_oom();
+ goto finish;
+ }
+
+ r = loop_device_make_by_path(
+ arg_image,
+ arg_read_only ? O_RDONLY : O_RDWR,
+ FLAGS_SET(dissect_image_flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
+ LOCK_SH,
+ &loop);
+ if (r < 0) {
+ log_error_errno(r, "Failed to set up loopback block device: %m");
+ goto finish;
+ }
+
+ r = dissect_loop_device_and_warn(
+ loop,
+ &arg_verity_settings,
+ NULL,
+ dissect_image_flags,
+ &dissected_image);
+ if (r == -ENOPKG) {
+ /* dissected_image_and_warn() already printed a brief error message. Extend on that with more details */
+ log_notice("Note that the disk image needs to\n"
+ " a) either contain only a single MBR partition of type 0x83 that is marked bootable\n"
+ " b) or contain a single GPT partition of type 0FC63DAF-8483-4772-8E79-3D69D8477DE4\n"
+ " c) or follow https://systemd.io/DISCOVERABLE_PARTITIONS\n"
+ " d) or contain a file system without a partition table\n"
+ "in order to be bootable with systemd-nspawn.");
+ goto finish;
+ }
+ if (r < 0)
+ goto finish;
+
+ r = dissected_image_load_verity_sig_partition(
+ dissected_image,
+ loop->fd,
+ &arg_verity_settings);
+ if (r < 0)
+ goto finish;
+
+ if (dissected_image->has_verity && !arg_verity_settings.root_hash && !dissected_image->has_verity_sig)
+ log_notice("Note: image %s contains verity information, but no root hash specified and no embedded "
+ "root hash signature found! Proceeding without integrity checking.", arg_image);
+
+ r = dissected_image_decrypt_interactively(
+ dissected_image,
+ NULL,
+ &arg_verity_settings,
+ 0);
+ if (r < 0)
+ goto finish;
+
+ /* Now that we mounted the image, let's try to remove it again, if it is ephemeral */
+ if (remove_image && unlink(arg_image) >= 0)
+ remove_image = false;
+ }
+
+ r = custom_mount_prepare_all(arg_directory, arg_custom_mounts, arg_n_custom_mounts);
+ if (r < 0)
+ goto finish;
+
+ if (arg_console_mode < 0)
+ arg_console_mode =
+ isatty(STDIN_FILENO) > 0 &&
+ isatty(STDOUT_FILENO) > 0 ? CONSOLE_INTERACTIVE : CONSOLE_READ_ONLY;
+
+ if (arg_console_mode == CONSOLE_PIPE) /* if we pass STDERR on to the container, don't add our own logs into it too */
+ arg_quiet = true;
+
+ if (!arg_quiet)
+ log_info("Spawning container %s on %s.\nPress ^] three times within 1s to kill container.",
+ arg_machine, arg_image ?: arg_directory);
+
+ assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGCHLD, SIGWINCH, SIGTERM, SIGINT, -1) >= 0);
+
+ if (prctl(PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) < 0) {
+ r = log_error_errno(errno, "Failed to become subreaper: %m");
+ goto finish;
+ }
+
+ if (arg_expose_ports) {
+ r = fw_ctx_new(&fw_ctx);
+ if (r < 0) {
+ log_error_errno(r, "Cannot expose configured ports, firewall initialization failed: %m");
+ goto finish;
+ }
+ expose_args.fw_ctx = fw_ctx;
+ }
+ for (;;) {
+ r = run_container(dissected_image,
+ secondary,
+ fds,
+ veth_name, &veth_created,
+ &expose_args, &master,
+ &pid, &ret);
+ if (r <= 0)
+ break;
+ }
+
+finish:
+ (void) sd_notify(false,
+ r == 0 && ret == EXIT_FORCE_RESTART ? "STOPPING=1\nSTATUS=Restarting..." :
+ "STOPPING=1\nSTATUS=Terminating...");
+
+ if (pid > 0)
+ (void) kill(pid, SIGKILL);
+
+ /* Try to flush whatever is still queued in the pty */
+ if (master >= 0) {
+ (void) copy_bytes(master, STDOUT_FILENO, UINT64_MAX, 0);
+ master = safe_close(master);
+ }
+
+ if (pid > 0)
+ (void) wait_for_terminate(pid, NULL);
+
+ pager_close();
+
+ if (remove_directory && arg_directory) {
+ int k;
+
+ k = rm_rf(arg_directory, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
+ if (k < 0)
+ log_warning_errno(k, "Cannot remove '%s', ignoring: %m", arg_directory);
+ }
+
+ if (remove_image && arg_image) {
+ if (unlink(arg_image) < 0)
+ log_warning_errno(errno, "Can't remove image file '%s', ignoring: %m", arg_image);
+ }
+
+ if (remove_tmprootdir) {
+ if (rmdir(tmprootdir) < 0)
+ log_debug_errno(errno, "Can't remove temporary root directory '%s', ignoring: %m", tmprootdir);
+ }
+
+ if (arg_machine) {
+ const char *p;
+
+ p = strjoina("/run/systemd/nspawn/propagate/", arg_machine);
+ (void) rm_rf(p, REMOVE_ROOT);
+ }
+
+ expose_port_flush(&fw_ctx, arg_expose_ports, AF_INET, &expose_args.address4);
+ expose_port_flush(&fw_ctx, arg_expose_ports, AF_INET6, &expose_args.address6);
+
+ if (veth_created)
+ (void) remove_veth_links(veth_name, arg_network_veth_extra);
+ (void) remove_bridge(arg_network_zone);
+
+ custom_mount_free_all(arg_custom_mounts, arg_n_custom_mounts);
+ expose_port_free_all(arg_expose_ports);
+ rlimit_free_all(arg_rlimit);
+ device_node_array_free(arg_extra_nodes, arg_n_extra_nodes);
+ credential_free_all(arg_credentials, arg_n_credentials);
+
+ if (r < 0)
+ return r;
+
+ return ret;
+}
+
+DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
diff --git a/src/nspawn/nspawn.h b/src/nspawn/nspawn.h
new file mode 100644
index 0000000..27fb0b4
--- /dev/null
+++ b/src/nspawn/nspawn.h
@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <sys/types.h>
+
+int userns_lchown(const char *p, uid_t uid, gid_t gid);
+int userns_mkdir(const char *root, const char *path, mode_t mode, uid_t uid, gid_t gid);
diff --git a/src/nspawn/test-nspawn-tables.c b/src/nspawn/test-nspawn-tables.c
new file mode 100644
index 0000000..3baf5d3
--- /dev/null
+++ b/src/nspawn/test-nspawn-tables.c
@@ -0,0 +1,11 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "nspawn-settings.h"
+#include "test-tables.h"
+
+int main(int argc, char **argv) {
+ test_table(resolv_conf_mode, RESOLV_CONF_MODE);
+ test_table(timezone_mode, TIMEZONE_MODE);
+
+ return 0;
+}
diff --git a/src/nspawn/test-nspawn-util.c b/src/nspawn/test-nspawn-util.c
new file mode 100644
index 0000000..08c8050
--- /dev/null
+++ b/src/nspawn/test-nspawn-util.c
@@ -0,0 +1,22 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include "nspawn-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "tests.h"
+
+TEST(systemd_installation_has_version) {
+ int r;
+
+ FOREACH_STRING(version, "0", "231", STRINGIFY(PROJECT_VERSION), "999") {
+ r = systemd_installation_has_version(saved_argv[1], version);
+ assert_se(r >= 0);
+ log_info("%s has systemd >= %s: %s",
+ saved_argv[1] ?: "Current installation", version, yes_no(r));
+ }
+}
+
+/* This program can be called with a path to an installation root.
+ * For example: build/test-nspawn-util /var/lib/machines/rawhide
+ */
+DEFINE_TEST_MAIN(LOG_DEBUG);
diff --git a/src/nspawn/test-patch-uid.c b/src/nspawn/test-patch-uid.c
new file mode 100644
index 0000000..f8f44b0
--- /dev/null
+++ b/src/nspawn/test-patch-uid.c
@@ -0,0 +1,43 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <stdlib.h>
+
+#include "log.h"
+#include "nspawn-patch-uid.h"
+#include "user-util.h"
+#include "string-util.h"
+#include "tests.h"
+
+int main(int argc, char *argv[]) {
+ uid_t shift, range;
+ int r;
+
+ test_setup_logging(LOG_DEBUG);
+
+ if (argc != 4) {
+ log_error("Expected PATH SHIFT RANGE parameters.");
+ return EXIT_FAILURE;
+ }
+
+ r = parse_uid(argv[2], &shift);
+ if (r < 0) {
+ log_error_errno(r, "Failed to parse UID shift %s.", argv[2]);
+ return EXIT_FAILURE;
+ }
+
+ r = parse_gid(argv[3], &range);
+ if (r < 0) {
+ log_error_errno(r, "Failed to parse UID range %s.", argv[3]);
+ return EXIT_FAILURE;
+ }
+
+ r = path_patch_uid(argv[1], shift, range);
+ if (r < 0) {
+ log_error_errno(r, "Failed to patch directory tree: %m");
+ return EXIT_FAILURE;
+ }
+
+ log_info("Changed: %s", yes_no(r));
+
+ return EXIT_SUCCESS;
+}