summaryrefslogtreecommitdiffstats
path: root/src/basic/process-util.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/basic/process-util.c')
-rw-r--r--src/basic/process-util.c270
1 files changed, 219 insertions, 51 deletions
diff --git a/src/basic/process-util.c b/src/basic/process-util.c
index 4492e7d..c9d968d 100644
--- a/src/basic/process-util.c
+++ b/src/basic/process-util.c
@@ -25,6 +25,7 @@
#include "alloc-util.h"
#include "architecture.h"
#include "argv-util.h"
+#include "cgroup-util.h"
#include "dirent-util.h"
#include "env-file.h"
#include "env-util.h"
@@ -510,7 +511,7 @@ static int get_process_link_contents(pid_t pid, const char *proc_file, char **re
p = procfs_file_alloca(pid, proc_file);
r = readlink_malloc(p, ret);
- return r == -ENOENT ? -ESRCH : r;
+ return (r == -ENOENT && proc_mounted() > 0) ? -ESRCH : r;
}
int get_process_exe(pid_t pid, char **ret) {
@@ -730,6 +731,82 @@ int get_process_ppid(pid_t pid, pid_t *ret) {
return 0;
}
+int pid_get_start_time(pid_t pid, uint64_t *ret) {
+ _cleanup_free_ char *line = NULL;
+ const char *p;
+ int r;
+
+ assert(pid >= 0);
+
+ p = procfs_file_alloca(pid, "stat");
+ r = read_one_line_file(p, &line);
+ if (r == -ENOENT)
+ return -ESRCH;
+ if (r < 0)
+ return r;
+
+ /* Let's skip the pid and comm fields. The latter is enclosed in () but does not escape any () in its
+ * value, so let's skip over it manually */
+
+ p = strrchr(line, ')');
+ if (!p)
+ return -EIO;
+
+ p++;
+
+ unsigned long llu;
+
+ if (sscanf(p, " "
+ "%*c " /* state */
+ "%*u " /* ppid */
+ "%*u " /* pgrp */
+ "%*u " /* session */
+ "%*u " /* tty_nr */
+ "%*u " /* tpgid */
+ "%*u " /* flags */
+ "%*u " /* minflt */
+ "%*u " /* cminflt */
+ "%*u " /* majflt */
+ "%*u " /* cmajflt */
+ "%*u " /* utime */
+ "%*u " /* stime */
+ "%*u " /* cutime */
+ "%*u " /* cstime */
+ "%*i " /* priority */
+ "%*i " /* nice */
+ "%*u " /* num_threads */
+ "%*u " /* itrealvalue */
+ "%lu ", /* starttime */
+ &llu) != 1)
+ return -EIO;
+
+ if (ret)
+ *ret = llu;
+
+ return 0;
+}
+
+int pidref_get_start_time(const PidRef *pid, uint64_t *ret) {
+ uint64_t t;
+ int r;
+
+ if (!pidref_is_set(pid))
+ return -ESRCH;
+
+ r = pid_get_start_time(pid->pid, ret ? &t : NULL);
+ if (r < 0)
+ return r;
+
+ r = pidref_verify(pid);
+ if (r < 0)
+ return r;
+
+ if (ret)
+ *ret = t;
+
+ return 0;
+}
+
int get_process_umask(pid_t pid, mode_t *ret) {
_cleanup_free_ char *m = NULL;
const char *p;
@@ -946,31 +1023,16 @@ int kill_and_sigcont(pid_t pid, int sig) {
int getenv_for_pid(pid_t pid, const char *field, char **ret) {
_cleanup_fclose_ FILE *f = NULL;
- char *value = NULL;
const char *path;
- size_t l, sum = 0;
+ size_t sum = 0;
int r;
assert(pid >= 0);
assert(field);
assert(ret);
- if (pid == 0 || pid == getpid_cached()) {
- const char *e;
-
- e = getenv(field);
- if (!e) {
- *ret = NULL;
- return 0;
- }
-
- value = strdup(e);
- if (!value)
- return -ENOMEM;
-
- *ret = value;
- return 1;
- }
+ if (pid == 0 || pid == getpid_cached())
+ return strdup_to_full(ret, getenv(field));
if (!pid_is_valid(pid))
return -EINVAL;
@@ -983,9 +1045,9 @@ int getenv_for_pid(pid_t pid, const char *field, char **ret) {
if (r < 0)
return r;
- l = strlen(field);
for (;;) {
_cleanup_free_ char *line = NULL;
+ const char *match;
if (sum > ENVIRONMENT_BLOCK_MAX) /* Give up searching eventually */
return -ENOBUFS;
@@ -998,14 +1060,9 @@ int getenv_for_pid(pid_t pid, const char *field, char **ret) {
sum += r;
- if (strneq(line, field, l) && line[l] == '=') {
- value = strdup(line + l + 1);
- if (!value)
- return -ENOMEM;
-
- *ret = value;
- return 1;
- }
+ match = startswith(line, field);
+ if (match && *match == '=')
+ return strdup_to_full(ret, match + 1);
}
*ret = NULL;
@@ -1112,8 +1169,10 @@ int pidref_is_alive(const PidRef *pidref) {
return -ESRCH;
result = pid_is_alive(pidref->pid);
- if (result < 0)
+ if (result < 0) {
+ assert(result != -ESRCH);
return result;
+ }
r = pidref_verify(pidref);
if (r == -ESRCH)
@@ -1224,7 +1283,7 @@ int opinionated_personality(unsigned long *ret) {
if (current < 0)
return current;
- if (((unsigned long) current & 0xffff) == PER_LINUX32)
+ if (((unsigned long) current & OPINIONATED_PERSONALITY_MASK) == PER_LINUX32)
*ret = PER_LINUX32;
else
*ret = PER_LINUX;
@@ -1389,7 +1448,7 @@ static int fork_flags_to_signal(ForkFlags flags) {
int safe_fork_full(
const char *name,
const int stdio_fds[3],
- const int except_fds[],
+ int except_fds[],
size_t n_except_fds,
ForkFlags flags,
pid_t *ret_pid) {
@@ -1462,10 +1521,11 @@ int safe_fork_full(
}
}
- if ((flags & (FORK_NEW_MOUNTNS|FORK_NEW_USERNS)) != 0)
+ if ((flags & (FORK_NEW_MOUNTNS|FORK_NEW_USERNS|FORK_NEW_NETNS)) != 0)
pid = raw_clone(SIGCHLD|
(FLAGS_SET(flags, FORK_NEW_MOUNTNS) ? CLONE_NEWNS : 0) |
- (FLAGS_SET(flags, FORK_NEW_USERNS) ? CLONE_NEWUSER : 0));
+ (FLAGS_SET(flags, FORK_NEW_USERNS) ? CLONE_NEWUSER : 0) |
+ (FLAGS_SET(flags, FORK_NEW_NETNS) ? CLONE_NEWNET : 0));
else
pid = fork();
if (pid < 0)
@@ -1589,6 +1649,9 @@ int safe_fork_full(
log_full_errno(prio, r, "Failed to rearrange stdio fds: %m");
_exit(EXIT_FAILURE);
}
+
+ /* Turn off O_NONBLOCK on the fdio fds, in case it was left on */
+ stdio_disable_nonblock();
} else {
r = make_null_stdio();
if (r < 0) {
@@ -1614,6 +1677,19 @@ int safe_fork_full(
}
}
+ if (flags & FORK_PACK_FDS) {
+ /* FORK_CLOSE_ALL_FDS ensures that except_fds are the only FDs >= 3 that are
+ * open, this is including the log. This is required by pack_fds, which will
+ * get stuck in an infinite loop of any FDs other than except_fds are open. */
+ assert(FLAGS_SET(flags, FORK_CLOSE_ALL_FDS));
+
+ r = pack_fds(except_fds, n_except_fds);
+ if (r < 0) {
+ log_full_errno(prio, r, "Failed to pack file descriptors: %m");
+ _exit(EXIT_FAILURE);
+ }
+ }
+
if (flags & FORK_CLOEXEC_OFF) {
r = fd_cloexec_many(except_fds, n_except_fds, false);
if (r < 0) {
@@ -1650,10 +1726,34 @@ int safe_fork_full(
return 0;
}
+int pidref_safe_fork_full(
+ const char *name,
+ const int stdio_fds[3],
+ int except_fds[],
+ size_t n_except_fds,
+ ForkFlags flags,
+ PidRef *ret_pid) {
+
+ pid_t pid;
+ int r, q;
+
+ assert(!FLAGS_SET(flags, FORK_WAIT));
+
+ r = safe_fork_full(name, stdio_fds, except_fds, n_except_fds, flags, &pid);
+ if (r < 0)
+ return r;
+
+ q = pidref_set_pid(ret_pid, pid);
+ if (q < 0) /* Let's not fail for this, no matter what, the process exists after all, and that's key */
+ *ret_pid = PIDREF_MAKE_FROM_PID(pid);
+
+ return r;
+}
+
int namespace_fork(
const char *outer_name,
const char *inner_name,
- const int except_fds[],
+ int except_fds[],
size_t n_except_fds,
ForkFlags flags,
int pidns_fd,
@@ -1927,47 +2027,115 @@ int make_reaper_process(bool b) {
return 0;
}
-int posix_spawn_wrapper(const char *path, char *const *argv, char *const *envp, pid_t *ret_pid) {
+DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(posix_spawnattr_t*, posix_spawnattr_destroy, NULL);
+
+int posix_spawn_wrapper(
+ const char *path,
+ char * const *argv,
+ char * const *envp,
+ const char *cgroup,
+ PidRef *ret_pidref) {
+
+ short flags = POSIX_SPAWN_SETSIGMASK|POSIX_SPAWN_SETSIGDEF;
posix_spawnattr_t attr;
sigset_t mask;
- pid_t pid;
int r;
/* Forks and invokes 'path' with 'argv' and 'envp' using CLONE_VM and CLONE_VFORK, which means the
* caller will be blocked until the child either exits or exec's. The memory of the child will be
* fully shared with the memory of the parent, so that there are no copy-on-write or memory.max
- * issues. */
+ * issues.
+ *
+ * Also, move the newly-created process into 'cgroup' through POSIX_SPAWN_SETCGROUP (clone3())
+ * if available. Note that CLONE_INTO_CGROUP is only supported on cgroup v2.
+ * returns 1: We're already in the right cgroup
+ * 0: 'cgroup' not specified or POSIX_SPAWN_SETCGROUP is not supported. The caller
+ * needs to call 'cg_attach' on their own */
assert(path);
assert(argv);
- assert(ret_pid);
+ assert(ret_pidref);
assert_se(sigfillset(&mask) >= 0);
r = posix_spawnattr_init(&attr);
if (r != 0)
return -r; /* These functions return a positive errno on failure */
- /* Set all signals to SIG_DFL */
- r = posix_spawnattr_setflags(&attr, POSIX_SPAWN_SETSIGMASK|POSIX_SPAWN_SETSIGDEF);
+
+ /* Initialization needs to succeed before we can set up a destructor. */
+ _unused_ _cleanup_(posix_spawnattr_destroyp) posix_spawnattr_t *attr_destructor = &attr;
+
+#if HAVE_PIDFD_SPAWN
+ _cleanup_close_ int cgroup_fd = -EBADF;
+
+ if (cgroup) {
+ _cleanup_free_ char *resolved_cgroup = NULL;
+
+ r = cg_get_path_and_check(
+ SYSTEMD_CGROUP_CONTROLLER,
+ cgroup,
+ /* suffix= */ NULL,
+ &resolved_cgroup);
+ if (r < 0)
+ return r;
+
+ cgroup_fd = open(resolved_cgroup, O_PATH|O_DIRECTORY|O_CLOEXEC);
+ if (cgroup_fd < 0)
+ return -errno;
+
+ r = posix_spawnattr_setcgroup_np(&attr, cgroup_fd);
+ if (r != 0)
+ return -r;
+
+ flags |= POSIX_SPAWN_SETCGROUP;
+ }
+#endif
+
+ r = posix_spawnattr_setflags(&attr, flags);
if (r != 0)
- goto fail;
+ return -r;
r = posix_spawnattr_setsigmask(&attr, &mask);
if (r != 0)
- goto fail;
+ return -r;
- r = posix_spawn(&pid, path, NULL, &attr, argv, envp);
+#if HAVE_PIDFD_SPAWN
+ _cleanup_close_ int pidfd = -EBADF;
+
+ r = pidfd_spawn(&pidfd, path, NULL, &attr, argv, envp);
+ if (r == 0) {
+ r = pidref_set_pidfd_consume(ret_pidref, TAKE_FD(pidfd));
+ if (r < 0)
+ return r;
+
+ return FLAGS_SET(flags, POSIX_SPAWN_SETCGROUP);
+ }
+ if (ERRNO_IS_NOT_SUPPORTED(r)) {
+ /* clone3() could also return EOPNOTSUPP if the target cgroup is in threaded mode. */
+ if (cgroup && cg_is_threaded(cgroup) > 0)
+ return -EUCLEAN;
+
+ /* clone3() not available? */
+ } else if (!ERRNO_IS_PRIVILEGE(r))
+ return -r;
+
+ /* Compiled on a newer host, or seccomp&friends blocking clone3()? Fallback, but need to change the
+ * flags to remove the cgroup one, which is what redirects to clone3() */
+ flags &= ~POSIX_SPAWN_SETCGROUP;
+ r = posix_spawnattr_setflags(&attr, flags);
if (r != 0)
- goto fail;
+ return -r;
+#endif
- *ret_pid = pid;
+ pid_t pid;
+ r = posix_spawn(&pid, path, NULL, &attr, argv, envp);
+ if (r != 0)
+ return -r;
- posix_spawnattr_destroy(&attr);
- return 0;
+ r = pidref_set_pid(ret_pidref, pid);
+ if (r < 0)
+ return r;
-fail:
- assert(r > 0);
- posix_spawnattr_destroy(&attr);
- return -r;
+ return 0; /* We did not use CLONE_INTO_CGROUP so return 0, the caller will have to move the child */
}
int proc_dir_open(DIR **ret) {
generated by cgit v1.2.3 (git 2.39.1) at 2024-10-01 19:39:43 +0000