Adding upstream version 247.3.upstream/247.3upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

9 files changed, 2963 insertions, 0 deletions

diff --git a/src/coredump/coredump-vacuum.c b/src/coredump/coredump-vacuum.c
new file mode 100644
index 0000000..30c67ff
--- /dev/null
+++ b/src/coredump/coredump-vacuum.c
@@ -0,0 +1,248 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <sys/statvfs.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "alloc-util.h"
+#include "coredump-vacuum.h"
+#include "dirent-util.h"
+#include "fd-util.h"
+#include "fs-util.h"
+#include "hashmap.h"
+#include "macro.h"
+#include "memory-util.h"
+#include "string-util.h"
+#include "time-util.h"
+#include "user-util.h"
+
+#define DEFAULT_MAX_USE_LOWER (uint64_t) (1ULL*1024ULL*1024ULL)           /* 1 MiB */
+#define DEFAULT_MAX_USE_UPPER (uint64_t) (4ULL*1024ULL*1024ULL*1024ULL)   /* 4 GiB */
+#define DEFAULT_KEEP_FREE_UPPER (uint64_t) (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
+#define DEFAULT_KEEP_FREE (uint64_t) (1024ULL*1024ULL)                    /* 1 MB */
+
+struct vacuum_candidate {
+        unsigned n_files;
+        char *oldest_file;
+        usec_t oldest_mtime;
+};
+
+static void vacuum_candidate_free(struct vacuum_candidate *c) {
+        if (!c)
+                return;
+
+        free(c->oldest_file);
+        free(c);
+}
+
+DEFINE_TRIVIAL_CLEANUP_FUNC(struct vacuum_candidate*, vacuum_candidate_free);
+
+static void vacuum_candidate_hashmap_free(Hashmap *h) {
+        hashmap_free_with_destructor(h, vacuum_candidate_free);
+}
+
+DEFINE_TRIVIAL_CLEANUP_FUNC(Hashmap*, vacuum_candidate_hashmap_free);
+
+static int uid_from_file_name(const char *filename, uid_t *uid) {
+        const char *p, *e, *u;
+
+        p = startswith(filename, "core.");
+        if (!p)
+                return -EINVAL;
+
+        /* Skip the comm field */
+        p = strchr(p, '.');
+        if (!p)
+                return -EINVAL;
+        p++;
+
+        /* Find end up UID */
+        e = strchr(p, '.');
+        if (!e)
+                return -EINVAL;
+
+        u = strndupa(p, e-p);
+        return parse_uid(u, uid);
+}
+
+static bool vacuum_necessary(int fd, uint64_t sum, uint64_t keep_free, uint64_t max_use) {
+        uint64_t fs_size = 0, fs_free = (uint64_t) -1;
+        struct statvfs sv;
+
+        assert(fd >= 0);
+
+        if (fstatvfs(fd, &sv) >= 0) {
+                fs_size = sv.f_frsize * sv.f_blocks;
+                fs_free = sv.f_frsize * sv.f_bfree;
+        }
+
+        if (max_use == (uint64_t) -1) {
+
+                if (fs_size > 0) {
+                        max_use = PAGE_ALIGN(fs_size / 10); /* 10% */
+
+                        if (max_use > DEFAULT_MAX_USE_UPPER)
+                                max_use = DEFAULT_MAX_USE_UPPER;
+
+                        if (max_use < DEFAULT_MAX_USE_LOWER)
+                                max_use = DEFAULT_MAX_USE_LOWER;
+                } else
+                        max_use = DEFAULT_MAX_USE_LOWER;
+        } else
+                max_use = PAGE_ALIGN(max_use);
+
+        if (max_use > 0 && sum > max_use)
+                return true;
+
+        if (keep_free == (uint64_t) -1) {
+
+                if (fs_size > 0) {
+                        keep_free = PAGE_ALIGN((fs_size * 3) / 20); /* 15% */
+
+                        if (keep_free > DEFAULT_KEEP_FREE_UPPER)
+                                keep_free = DEFAULT_KEEP_FREE_UPPER;
+                } else
+                        keep_free = DEFAULT_KEEP_FREE;
+        } else
+                keep_free = PAGE_ALIGN(keep_free);
+
+        if (keep_free > 0 && fs_free < keep_free)
+                return true;
+
+        return false;
+}
+
+int coredump_vacuum(int exclude_fd, uint64_t keep_free, uint64_t max_use) {
+        _cleanup_closedir_ DIR *d = NULL;
+        struct stat exclude_st;
+        int r;
+
+        if (keep_free == 0 && max_use == 0)
+                return 0;
+
+        if (exclude_fd >= 0) {
+                if (fstat(exclude_fd, &exclude_st) < 0)
+                        return log_error_errno(errno, "Failed to fstat(): %m");
+        }
+
+        /* This algorithm will keep deleting the oldest file of the
+         * user with the most coredumps until we are back in the size
+         * limits. Note that vacuuming for journal files is different,
+         * because we rely on rate-limiting of the messages there,
+         * to avoid being flooded. */
+
+        d = opendir("/var/lib/systemd/coredump");
+        if (!d) {
+                if (errno == ENOENT)
+                        return 0;
+
+                return log_error_errno(errno, "Can't open coredump directory: %m");
+        }
+
+        for (;;) {
+                _cleanup_(vacuum_candidate_hashmap_freep) Hashmap *h = NULL;
+                struct vacuum_candidate *worst = NULL;
+                struct dirent *de;
+                uint64_t sum = 0;
+
+                rewinddir(d);
+
+                FOREACH_DIRENT(de, d, goto fail) {
+                        struct vacuum_candidate *c;
+                        struct stat st;
+                        uid_t uid;
+                        usec_t t;
+
+                        r = uid_from_file_name(de->d_name, &uid);
+                        if (r < 0)
+                                continue;
+
+                        if (fstatat(dirfd(d), de->d_name, &st, AT_NO_AUTOMOUNT|AT_SYMLINK_NOFOLLOW) < 0) {
+                                if (errno == ENOENT)
+                                        continue;
+
+                                log_warning_errno(errno, "Failed to stat /var/lib/systemd/coredump/%s: %m", de->d_name);
+                                continue;
+                        }
+
+                        if (!S_ISREG(st.st_mode))
+                                continue;
+
+                        if (exclude_fd >= 0 &&
+                            exclude_st.st_dev == st.st_dev &&
+                            exclude_st.st_ino == st.st_ino)
+                                continue;
+
+                        r = hashmap_ensure_allocated(&h, NULL);
+                        if (r < 0)
+                                return log_oom();
+
+                        t = timespec_load(&st.st_mtim);
+
+                        c = hashmap_get(h, UID_TO_PTR(uid));
+                        if (c) {
+
+                                if (t < c->oldest_mtime) {
+                                        char *n;
+
+                                        n = strdup(de->d_name);
+                                        if (!n)
+                                                return log_oom();
+
+                                        free(c->oldest_file);
+                                        c->oldest_file = n;
+                                        c->oldest_mtime = t;
+                                }
+
+                        } else {
+                                _cleanup_(vacuum_candidate_freep) struct vacuum_candidate *n = NULL;
+
+                                n = new0(struct vacuum_candidate, 1);
+                                if (!n)
+                                        return log_oom();
+
+                                n->oldest_file = strdup(de->d_name);
+                                if (!n->oldest_file)
+                                        return log_oom();
+
+                                n->oldest_mtime = t;
+
+                                r = hashmap_put(h, UID_TO_PTR(uid), n);
+                                if (r < 0)
+                                        return log_oom();
+
+                                c = TAKE_PTR(n);
+                        }
+
+                        c->n_files++;
+
+                        if (!worst ||
+                            worst->n_files < c->n_files ||
+                            (worst->n_files == c->n_files && c->oldest_mtime < worst->oldest_mtime))
+                                worst = c;
+
+                        sum += st.st_blocks * 512;
+                }
+
+                if (!worst)
+                        break;
+
+                r = vacuum_necessary(dirfd(d), sum, keep_free, max_use);
+                if (r <= 0)
+                        return r;
+
+                r = unlinkat_deallocate(dirfd(d), worst->oldest_file, 0);
+                if (r == -ENOENT)
+                        continue;
+                if (r < 0)
+                        return log_error_errno(r, "Failed to remove file %s: %m", worst->oldest_file);
+
+                log_info("Removed old coredump %s.", worst->oldest_file);
+        }
+
+        return 0;
+
+fail:
+        return log_error_errno(errno, "Failed to read directory: %m");
+}
diff --git a/src/coredump/coredump-vacuum.h b/src/coredump/coredump-vacuum.h
new file mode 100644
index 0000000..8ad5baf
--- /dev/null
+++ b/src/coredump/coredump-vacuum.h
@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+#include <inttypes.h>
+#include <sys/types.h>
+
+int coredump_vacuum(int exclude_fd, uint64_t keep_free, uint64_t max_use);
diff --git a/src/coredump/coredump.c b/src/coredump/coredump.c
new file mode 100644
index 0000000..0a1cb91
--- /dev/null
+++ b/src/coredump/coredump.c
@@ -0,0 +1,1330 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <errno.h>
+#include <stdio.h>
+#include <sys/prctl.h>
+#include <sys/xattr.h>
+#include <unistd.h>
+
+#if HAVE_ELFUTILS
+#include <dwarf.h>
+#include <elfutils/libdwfl.h>
+#endif
+
+#include "sd-daemon.h"
+#include "sd-journal.h"
+#include "sd-login.h"
+#include "sd-messages.h"
+
+#include "acl-util.h"
+#include "alloc-util.h"
+#include "capability-util.h"
+#include "cgroup-util.h"
+#include "compress.h"
+#include "conf-parser.h"
+#include "copy.h"
+#include "coredump-vacuum.h"
+#include "dirent-util.h"
+#include "escape.h"
+#include "fd-util.h"
+#include "fileio.h"
+#include "fs-util.h"
+#include "io-util.h"
+#include "journal-importer.h"
+#include "log.h"
+#include "macro.h"
+#include "main-func.h"
+#include "memory-util.h"
+#include "mkdir.h"
+#include "parse-util.h"
+#include "process-util.h"
+#include "signal-util.h"
+#include "socket-util.h"
+#include "special.h"
+#include "stacktrace.h"
+#include "string-table.h"
+#include "string-util.h"
+#include "strv.h"
+#include "tmpfile-util.h"
+#include "user-record.h"
+#include "user-util.h"
+
+/* The maximum size up to which we process coredumps */
+#define PROCESS_SIZE_MAX ((uint64_t) (2LLU*1024LLU*1024LLU*1024LLU))
+
+/* The maximum size up to which we leave the coredump around on disk */
+#define EXTERNAL_SIZE_MAX PROCESS_SIZE_MAX
+
+/* The maximum size up to which we store the coredump in the journal */
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+#define JOURNAL_SIZE_MAX ((size_t) (767LU*1024LU*1024LU))
+#else
+/* oss-fuzz limits memory usage. */
+#define JOURNAL_SIZE_MAX ((size_t) (10LU*1024LU*1024LU))
+#endif
+
+/* Make sure to not make this larger than the maximum journal entry
+ * size. See DATA_SIZE_MAX in journal-importer.h. */
+assert_cc(JOURNAL_SIZE_MAX <= DATA_SIZE_MAX);
+
+enum {
+        /* We use these as array indexes for our process metadata cache.
+         *
+         * The first indices of the cache stores the same metadata as the ones passed by
+         * the kernel via argv[], ie the strings array passed by the kernel according to
+         * our pattern defined in /proc/sys/kernel/core_pattern (see man:core(5)). */
+
+        META_ARGV_PID,          /* %P: as seen in the initial pid namespace */
+        META_ARGV_UID,          /* %u: as seen in the initial user namespace */
+        META_ARGV_GID,          /* %g: as seen in the initial user namespace */
+        META_ARGV_SIGNAL,       /* %s: number of signal causing dump */
+        META_ARGV_TIMESTAMP,    /* %t: time of dump, expressed as seconds since the Epoch (we expand this to µs granularity) */
+        META_ARGV_RLIMIT,       /* %c: core file size soft resource limit */
+        META_ARGV_HOSTNAME,     /* %h: hostname */
+        _META_ARGV_MAX,
+
+        /* The following indexes are cached for a couple of special fields we use (and
+         * thereby need to be retrieved quickly) for naming coredump files, and attaching
+         * xattrs. Unlike the previous ones they are retrieved from the runtime
+         * environment. */
+
+        META_COMM = _META_ARGV_MAX,
+        _META_MANDATORY_MAX,
+
+        /* The rest are similar to the previous ones except that we won't fail if one of
+         * them is missing. */
+
+        META_EXE = _META_MANDATORY_MAX,
+        META_UNIT,
+        _META_MAX
+};
+
+static const char * const meta_field_names[_META_MAX] = {
+        [META_ARGV_PID]          = "COREDUMP_PID=",
+        [META_ARGV_UID]          = "COREDUMP_UID=",
+        [META_ARGV_GID]          = "COREDUMP_GID=",
+        [META_ARGV_SIGNAL]       = "COREDUMP_SIGNAL=",
+        [META_ARGV_TIMESTAMP]    = "COREDUMP_TIMESTAMP=",
+        [META_ARGV_RLIMIT]       = "COREDUMP_RLIMIT=",
+        [META_ARGV_HOSTNAME]     = "COREDUMP_HOSTNAME=",
+        [META_COMM]              = "COREDUMP_COMM=",
+        [META_EXE]               = "COREDUMP_EXE=",
+        [META_UNIT]              = "COREDUMP_UNIT=",
+};
+
+typedef struct Context {
+        const char *meta[_META_MAX];
+        pid_t pid;
+        bool is_pid1;
+        bool is_journald;
+} Context;
+
+typedef enum CoredumpStorage {
+        COREDUMP_STORAGE_NONE,
+        COREDUMP_STORAGE_EXTERNAL,
+        COREDUMP_STORAGE_JOURNAL,
+        _COREDUMP_STORAGE_MAX,
+        _COREDUMP_STORAGE_INVALID = -1
+} CoredumpStorage;
+
+static const char* const coredump_storage_table[_COREDUMP_STORAGE_MAX] = {
+        [COREDUMP_STORAGE_NONE] = "none",
+        [COREDUMP_STORAGE_EXTERNAL] = "external",
+        [COREDUMP_STORAGE_JOURNAL] = "journal",
+};
+
+DEFINE_PRIVATE_STRING_TABLE_LOOKUP(coredump_storage, CoredumpStorage);
+static DEFINE_CONFIG_PARSE_ENUM(config_parse_coredump_storage, coredump_storage, CoredumpStorage, "Failed to parse storage setting");
+
+static CoredumpStorage arg_storage = COREDUMP_STORAGE_EXTERNAL;
+static bool arg_compress = true;
+static uint64_t arg_process_size_max = PROCESS_SIZE_MAX;
+static uint64_t arg_external_size_max = EXTERNAL_SIZE_MAX;
+static uint64_t arg_journal_size_max = JOURNAL_SIZE_MAX;
+static uint64_t arg_keep_free = (uint64_t) -1;
+static uint64_t arg_max_use = (uint64_t) -1;
+
+static int parse_config(void) {
+        static const ConfigTableItem items[] = {
+                { "Coredump", "Storage",          config_parse_coredump_storage,  0, &arg_storage           },
+                { "Coredump", "Compress",         config_parse_bool,              0, &arg_compress          },
+                { "Coredump", "ProcessSizeMax",   config_parse_iec_uint64,        0, &arg_process_size_max  },
+                { "Coredump", "ExternalSizeMax",  config_parse_iec_uint64,        0, &arg_external_size_max },
+                { "Coredump", "JournalSizeMax",   config_parse_iec_size,          0, &arg_journal_size_max  },
+                { "Coredump", "KeepFree",         config_parse_iec_uint64,        0, &arg_keep_free         },
+                { "Coredump", "MaxUse",           config_parse_iec_uint64,        0, &arg_max_use           },
+                {}
+        };
+
+        return config_parse_many_nulstr(
+                        PKGSYSCONFDIR "/coredump.conf",
+                        CONF_PATHS_NULSTR("systemd/coredump.conf.d"),
+                        "Coredump\0",
+                        config_item_table_lookup, items,
+                        CONFIG_PARSE_WARN,
+                        NULL,
+                        NULL);
+}
+
+static uint64_t storage_size_max(void) {
+        if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
+                return arg_external_size_max;
+        if (arg_storage == COREDUMP_STORAGE_JOURNAL)
+                return arg_journal_size_max;
+        assert(arg_storage == COREDUMP_STORAGE_NONE);
+        return 0;
+}
+
+static int fix_acl(int fd, uid_t uid) {
+
+#if HAVE_ACL
+        int r;
+
+        assert(fd >= 0);
+        assert(uid_is_valid(uid));
+
+        if (uid_is_system(uid) || uid_is_dynamic(uid) || uid == UID_NOBODY)
+                return 0;
+
+        /* Make sure normal users can read (but not write or delete) their own coredumps */
+        r = fd_add_uid_acl_permission(fd, uid, ACL_READ);
+        if (r < 0)
+                return log_error_errno(r, "Failed to adjust ACL of the coredump: %m");
+#endif
+
+        return 0;
+}
+
+static int fix_xattr(int fd, const Context *context) {
+
+        static const char * const xattrs[_META_MAX] = {
+                [META_ARGV_PID]          = "user.coredump.pid",
+                [META_ARGV_UID]          = "user.coredump.uid",
+                [META_ARGV_GID]          = "user.coredump.gid",
+                [META_ARGV_SIGNAL]       = "user.coredump.signal",
+                [META_ARGV_TIMESTAMP]    = "user.coredump.timestamp",
+                [META_ARGV_RLIMIT]       = "user.coredump.rlimit",
+                [META_ARGV_HOSTNAME]     = "user.coredump.hostname",
+                [META_COMM]              = "user.coredump.comm",
+                [META_EXE]               = "user.coredump.exe",
+        };
+
+        int r = 0;
+        unsigned i;
+
+        assert(fd >= 0);
+
+        /* Attach some metadata to coredumps via extended
+         * attributes. Just because we can. */
+
+        for (i = 0; i < _META_MAX; i++) {
+                int k;
+
+                if (isempty(context->meta[i]) || !xattrs[i])
+                        continue;
+
+                k = fsetxattr(fd, xattrs[i], context->meta[i], strlen(context->meta[i]), XATTR_CREATE);
+                if (k < 0 && r == 0)
+                        r = -errno;
+        }
+
+        return r;
+}
+
+#define filename_escape(s) xescape((s), "./ ")
+
+static const char *coredump_tmpfile_name(const char *s) {
+        return s ? s : "(unnamed temporary file)";
+}
+
+static int fix_permissions(
+                int fd,
+                const char *filename,
+                const char *target,
+                const Context *context,
+                uid_t uid) {
+
+        int r;
+
+        assert(fd >= 0);
+        assert(target);
+        assert(context);
+
+        /* Ignore errors on these */
+        (void) fchmod(fd, 0640);
+        (void) fix_acl(fd, uid);
+        (void) fix_xattr(fd, context);
+
+        if (fsync(fd) < 0)
+                return log_error_errno(errno, "Failed to sync coredump %s: %m", coredump_tmpfile_name(filename));
+
+        (void) fsync_directory_of_file(fd);
+
+        r = link_tmpfile(fd, filename, target);
+        if (r < 0)
+                return log_error_errno(r, "Failed to move coredump %s into place: %m", target);
+
+        return 0;
+}
+
+static int maybe_remove_external_coredump(const char *filename, uint64_t size) {
+
+        /* Returns 1 if might remove, 0 if will not remove, < 0 on error. */
+
+        if (arg_storage == COREDUMP_STORAGE_EXTERNAL &&
+            size <= arg_external_size_max)
+                return 0;
+
+        if (!filename)
+                return 1;
+
+        if (unlink(filename) < 0 && errno != ENOENT)
+                return log_error_errno(errno, "Failed to unlink %s: %m", filename);
+
+        return 1;
+}
+
+static int make_filename(const Context *context, char **ret) {
+        _cleanup_free_ char *c = NULL, *u = NULL, *p = NULL, *t = NULL;
+        sd_id128_t boot = {};
+        int r;
+
+        assert(context);
+
+        c = filename_escape(context->meta[META_COMM]);
+        if (!c)
+                return -ENOMEM;
+
+        u = filename_escape(context->meta[META_ARGV_UID]);
+        if (!u)
+                return -ENOMEM;
+
+        r = sd_id128_get_boot(&boot);
+        if (r < 0)
+                return r;
+
+        p = filename_escape(context->meta[META_ARGV_PID]);
+        if (!p)
+                return -ENOMEM;
+
+        t = filename_escape(context->meta[META_ARGV_TIMESTAMP]);
+        if (!t)
+                return -ENOMEM;
+
+        if (asprintf(ret,
+                     "/var/lib/systemd/coredump/core.%s.%s." SD_ID128_FORMAT_STR ".%s.%s",
+                     c,
+                     u,
+                     SD_ID128_FORMAT_VAL(boot),
+                     p,
+                     t) < 0)
+                return -ENOMEM;
+
+        return 0;
+}
+
+static int save_external_coredump(
+                const Context *context,
+                int input_fd,
+                char **ret_filename,
+                int *ret_node_fd,
+                int *ret_data_fd,
+                uint64_t *ret_size,
+                bool *ret_truncated) {
+
+        _cleanup_free_ char *fn = NULL, *tmp = NULL;
+        _cleanup_close_ int fd = -1;
+        uint64_t rlimit, process_limit, max_size;
+        struct stat st;
+        uid_t uid;
+        int r;
+
+        assert(context);
+        assert(ret_filename);
+        assert(ret_node_fd);
+        assert(ret_data_fd);
+        assert(ret_size);
+
+        r = parse_uid(context->meta[META_ARGV_UID], &uid);
+        if (r < 0)
+                return log_error_errno(r, "Failed to parse UID: %m");
+
+        r = safe_atou64(context->meta[META_ARGV_RLIMIT], &rlimit);
+        if (r < 0)
+                return log_error_errno(r, "Failed to parse resource limit '%s': %m",
+                                       context->meta[META_ARGV_RLIMIT]);
+        if (rlimit < page_size())
+                /* Is coredumping disabled? Then don't bother saving/processing the
+                 * coredump.  Anything below PAGE_SIZE cannot give a readable coredump
+                 * (the kernel uses ELF_EXEC_PAGESIZE which is not easily accessible, but
+                 * is usually the same as PAGE_SIZE. */
+                return log_info_errno(SYNTHETIC_ERRNO(EBADSLT),
+                                      "Resource limits disable core dumping for process %s (%s).",
+                                      context->meta[META_ARGV_PID], context->meta[META_COMM]);
+
+        process_limit = MAX(arg_process_size_max, storage_size_max());
+        if (process_limit == 0)
+                return log_debug_errno(SYNTHETIC_ERRNO(EBADSLT),
+                                       "Limits for coredump processing and storage are both 0, not dumping core.");
+
+        /* Never store more than the process configured, or than we actually shall keep or process */
+        max_size = MIN(rlimit, process_limit);
+
+        r = make_filename(context, &fn);
+        if (r < 0)
+                return log_error_errno(r, "Failed to determine coredump file name: %m");
+
+        (void) mkdir_p_label("/var/lib/systemd/coredump", 0755);
+
+        fd = open_tmpfile_linkable(fn, O_RDWR|O_CLOEXEC, &tmp);
+        if (fd < 0)
+                return log_error_errno(fd, "Failed to create temporary file for coredump %s: %m", fn);
+
+        r = copy_bytes(input_fd, fd, max_size, 0);
+        if (r < 0) {
+                log_error_errno(r, "Cannot store coredump of %s (%s): %m",
+                                context->meta[META_ARGV_PID], context->meta[META_COMM]);
+                goto fail;
+        }
+        *ret_truncated = r == 1;
+        if (*ret_truncated)
+                log_struct(LOG_INFO,
+                           LOG_MESSAGE("Core file was truncated to %zu bytes.", max_size),
+                           "SIZE_LIMIT=%zu", max_size,
+                           "MESSAGE_ID=" SD_MESSAGE_TRUNCATED_CORE_STR);
+
+        if (fstat(fd, &st) < 0) {
+                log_error_errno(errno, "Failed to fstat core file %s: %m", coredump_tmpfile_name(tmp));
+                goto fail;
+        }
+
+        if (lseek(fd, 0, SEEK_SET) == (off_t) -1) {
+                log_error_errno(errno, "Failed to seek on %s: %m", coredump_tmpfile_name(tmp));
+                goto fail;
+        }
+
+#if HAVE_COMPRESSION
+        /* If we will remove the coredump anyway, do not compress. */
+        if (arg_compress && !maybe_remove_external_coredump(NULL, st.st_size)) {
+
+                _cleanup_free_ char *fn_compressed = NULL, *tmp_compressed = NULL;
+                _cleanup_close_ int fd_compressed = -1;
+
+                fn_compressed = strjoin(fn, COMPRESSED_EXT);
+                if (!fn_compressed) {
+                        log_oom();
+                        goto uncompressed;
+                }
+
+                fd_compressed = open_tmpfile_linkable(fn_compressed, O_RDWR|O_CLOEXEC, &tmp_compressed);
+                if (fd_compressed < 0) {
+                        log_error_errno(fd_compressed, "Failed to create temporary file for coredump %s: %m", fn_compressed);
+                        goto uncompressed;
+                }
+
+                r = compress_stream(fd, fd_compressed, -1);
+                if (r < 0) {
+                        log_error_errno(r, "Failed to compress %s: %m", coredump_tmpfile_name(tmp_compressed));
+                        goto fail_compressed;
+                }
+
+                r = fix_permissions(fd_compressed, tmp_compressed, fn_compressed, context, uid);
+                if (r < 0)
+                        goto fail_compressed;
+
+                /* OK, this worked, we can get rid of the uncompressed version now */
+                if (tmp)
+                        unlink_noerrno(tmp);
+
+                *ret_filename = TAKE_PTR(fn_compressed);     /* compressed */
+                *ret_node_fd = TAKE_FD(fd_compressed);      /* compressed */
+                *ret_data_fd = TAKE_FD(fd);                 /* uncompressed */
+                *ret_size = (uint64_t) st.st_size; /* uncompressed */
+
+                return 0;
+
+        fail_compressed:
+                if (tmp_compressed)
+                        (void) unlink(tmp_compressed);
+        }
+
+uncompressed:
+#endif
+
+        r = fix_permissions(fd, tmp, fn, context, uid);
+        if (r < 0)
+                goto fail;
+
+        *ret_filename = TAKE_PTR(fn);
+        *ret_data_fd = TAKE_FD(fd);
+        *ret_node_fd = -1;
+        *ret_size = (uint64_t) st.st_size;
+
+        return 0;
+
+fail:
+        if (tmp)
+                (void) unlink(tmp);
+        return r;
+}
+
+static int allocate_journal_field(int fd, size_t size, char **ret, size_t *ret_size) {
+        _cleanup_free_ char *field = NULL;
+        ssize_t n;
+
+        assert(fd >= 0);
+        assert(ret);
+        assert(ret_size);
+
+        if (lseek(fd, 0, SEEK_SET) == (off_t) -1)
+                return log_warning_errno(errno, "Failed to seek: %m");
+
+        field = malloc(9 + size);
+        if (!field) {
+                log_warning("Failed to allocate memory for coredump, coredump will not be stored.");
+                return -ENOMEM;
+        }
+
+        memcpy(field, "COREDUMP=", 9);
+
+        n = read(fd, field + 9, size);
+        if (n < 0)
+                return log_error_errno((int) n, "Failed to read core data: %m");
+        if ((size_t) n < size)
+                return log_error_errno(SYNTHETIC_ERRNO(EIO),
+                                       "Core data too short.");
+
+        *ret = TAKE_PTR(field);
+        *ret_size = size + 9;
+
+        return 0;
+}
+
+/* Joins /proc/[pid]/fd/ and /proc/[pid]/fdinfo/ into the following lines:
+ * 0:/dev/pts/23
+ * pos:    0
+ * flags:  0100002
+ *
+ * 1:/dev/pts/23
+ * pos:    0
+ * flags:  0100002
+ *
+ * 2:/dev/pts/23
+ * pos:    0
+ * flags:  0100002
+ * EOF
+ */
+static int compose_open_fds(pid_t pid, char **open_fds) {
+        _cleanup_closedir_ DIR *proc_fd_dir = NULL;
+        _cleanup_close_ int proc_fdinfo_fd = -1;
+        _cleanup_free_ char *buffer = NULL;
+        _cleanup_fclose_ FILE *stream = NULL;
+        const char *fddelim = "", *path;
+        struct dirent *dent = NULL;
+        size_t size = 0;
+        int r;
+
+        assert(pid >= 0);
+        assert(open_fds != NULL);
+
+        path = procfs_file_alloca(pid, "fd");
+        proc_fd_dir = opendir(path);
+        if (!proc_fd_dir)
+                return -errno;
+
+        proc_fdinfo_fd = openat(dirfd(proc_fd_dir), "../fdinfo", O_DIRECTORY|O_NOFOLLOW|O_CLOEXEC|O_PATH);
+        if (proc_fdinfo_fd < 0)
+                return -errno;
+
+        stream = open_memstream_unlocked(&buffer, &size);
+        if (!stream)
+                return -ENOMEM;
+
+        FOREACH_DIRENT(dent, proc_fd_dir, return -errno) {
+                _cleanup_fclose_ FILE *fdinfo = NULL;
+                _cleanup_free_ char *fdname = NULL;
+                _cleanup_close_ int fd = -1;
+
+                r = readlinkat_malloc(dirfd(proc_fd_dir), dent->d_name, &fdname);
+                if (r < 0)
+                        return r;
+
+                fprintf(stream, "%s%s:%s\n", fddelim, dent->d_name, fdname);
+                fddelim = "\n";
+
+                /* Use the directory entry from /proc/[pid]/fd with /proc/[pid]/fdinfo */
+                fd = openat(proc_fdinfo_fd, dent->d_name, O_NOFOLLOW|O_CLOEXEC|O_RDONLY);
+                if (fd < 0)
+                        continue;
+
+                fdinfo = take_fdopen(&fd, "r");
+                if (!fdinfo)
+                        continue;
+
+                for (;;) {
+                        _cleanup_free_ char *line = NULL;
+
+                        r = read_line(fdinfo, LONG_LINE_MAX, &line);
+                        if (r < 0)
+                                return r;
+                        if (r == 0)
+                                break;
+
+                        fputs(line, stream);
+                        fputc('\n', stream);
+                }
+        }
+
+        errno = 0;
+        stream = safe_fclose(stream);
+
+        if (errno > 0)
+                return -errno;
+
+        *open_fds = TAKE_PTR(buffer);
+
+        return 0;
+}
+
+static int get_process_ns(pid_t pid, const char *namespace, ino_t *ns) {
+        const char *p;
+        struct stat stbuf;
+        _cleanup_close_ int proc_ns_dir_fd;
+
+        p = procfs_file_alloca(pid, "ns");
+
+        proc_ns_dir_fd = open(p, O_DIRECTORY | O_CLOEXEC | O_RDONLY);
+        if (proc_ns_dir_fd < 0)
+                return -errno;
+
+        if (fstatat(proc_ns_dir_fd, namespace, &stbuf, /* flags */0) < 0)
+                return -errno;
+
+        *ns = stbuf.st_ino;
+        return 0;
+}
+
+static int get_mount_namespace_leader(pid_t pid, pid_t *container_pid) {
+        pid_t cpid = pid, ppid = 0;
+        ino_t proc_mntns;
+        int r;
+
+        r = get_process_ns(pid, "mnt", &proc_mntns);
+        if (r < 0)
+                return r;
+
+        for (;;) {
+                ino_t parent_mntns;
+
+                r = get_process_ppid(cpid, &ppid);
+                if (r < 0)
+                        return r;
+
+                r = get_process_ns(ppid, "mnt", &parent_mntns);
+                if (r < 0)
+                        return r;
+
+                if (proc_mntns != parent_mntns)
+                        break;
+
+                if (ppid == 1)
+                        return -ENOENT;
+
+                cpid = ppid;
+        }
+
+        *container_pid = ppid;
+        return 0;
+}
+
+/* Returns 1 if the parent was found.
+ * Returns 0 if there is not a process we can call the pid's
+ * container parent (the pid's process isn't 'containerized').
+ * Returns a negative number on errors.
+ */
+static int get_process_container_parent_cmdline(pid_t pid, char** cmdline) {
+        int r = 0;
+        pid_t container_pid;
+        const char *proc_root_path;
+        struct stat root_stat, proc_root_stat;
+
+        /* To compare inodes of / and /proc/[pid]/root */
+        if (stat("/", &root_stat) < 0)
+                return -errno;
+
+        proc_root_path = procfs_file_alloca(pid, "root");
+        if (stat(proc_root_path, &proc_root_stat) < 0)
+                return -errno;
+
+        /* The process uses system root. */
+        if (proc_root_stat.st_ino == root_stat.st_ino) {
+                *cmdline = NULL;
+                return 0;
+        }
+
+        r = get_mount_namespace_leader(pid, &container_pid);
+        if (r < 0)
+                return r;
+
+        r = get_process_cmdline(container_pid, SIZE_MAX, 0, cmdline);
+        if (r < 0)
+                return r;
+
+        return 1;
+}
+
+static int change_uid_gid(const Context *context) {
+        uid_t uid;
+        gid_t gid;
+        int r;
+
+        r = parse_uid(context->meta[META_ARGV_UID], &uid);
+        if (r < 0)
+                return r;
+
+        if (uid_is_system(uid)) {
+                const char *user = "systemd-coredump";
+
+                r = get_user_creds(&user, &uid, &gid, NULL, NULL, 0);
+                if (r < 0) {
+                        log_warning_errno(r, "Cannot resolve %s user. Proceeding to dump core as root: %m", user);
+                        uid = gid = 0;
+                }
+        } else {
+                r = parse_gid(context->meta[META_ARGV_GID], &gid);
+                if (r < 0)
+                        return r;
+        }
+
+        return drop_privileges(uid, gid, 0);
+}
+
+static int submit_coredump(
+                Context *context,
+                struct iovec_wrapper *iovw,
+                int input_fd) {
+
+        _cleanup_close_ int coredump_fd = -1, coredump_node_fd = -1;
+        _cleanup_free_ char *filename = NULL, *coredump_data = NULL;
+        _cleanup_free_ char *stacktrace = NULL;
+        char *core_message;
+        uint64_t coredump_size = UINT64_MAX;
+        bool truncated = false;
+        int r;
+
+        assert(context);
+        assert(iovw);
+        assert(input_fd >= 0);
+
+        /* Vacuum before we write anything again */
+        (void) coredump_vacuum(-1, arg_keep_free, arg_max_use);
+
+        /* Always stream the coredump to disk, if that's possible */
+        r = save_external_coredump(context, input_fd,
+                                   &filename, &coredump_node_fd, &coredump_fd, &coredump_size, &truncated);
+        if (r < 0)
+                /* Skip whole core dumping part */
+                goto log;
+
+        /* If we don't want to keep the coredump on disk, remove it now, as later on we
+         * will lack the privileges for it. However, we keep the fd to it, so that we can
+         * still process it and log it. */
+        r = maybe_remove_external_coredump(filename, coredump_size);
+        if (r < 0)
+                return r;
+        if (r == 0) {
+                (void) iovw_put_string_field(iovw, "COREDUMP_FILENAME=", filename);
+
+        } else if (arg_storage == COREDUMP_STORAGE_EXTERNAL)
+                log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
+                         coredump_size, arg_external_size_max);
+
+        /* Vacuum again, but exclude the coredump we just created */
+        (void) coredump_vacuum(coredump_node_fd >= 0 ? coredump_node_fd : coredump_fd, arg_keep_free, arg_max_use);
+
+        /* Now, let's drop privileges to become the user who owns the segfaulted process
+         * and allocate the coredump memory under the user's uid. This also ensures that
+         * the credentials journald will see are the ones of the coredumping user, thus
+         * making sure the user gets access to the core dump. Let's also get rid of all
+         * capabilities, if we run as root, we won't need them anymore. */
+        r = change_uid_gid(context);
+        if (r < 0)
+                return log_error_errno(r, "Failed to drop privileges: %m");
+
+#if HAVE_ELFUTILS
+        /* Try to get a stack trace if we can */
+        if (coredump_size > arg_process_size_max) {
+                log_debug("Not generating stack trace: core size %"PRIu64" is greater "
+                          "than %"PRIu64" (the configured maximum)",
+                          coredump_size, arg_process_size_max);
+        } else
+                coredump_make_stack_trace(coredump_fd, context->meta[META_EXE], &stacktrace);
+#endif
+
+log:
+        core_message = strjoina("Process ", context->meta[META_ARGV_PID],
+                                " (", context->meta[META_COMM], ") of user ",
+                                context->meta[META_ARGV_UID], " dumped core.",
+                                context->is_journald && filename ? "\nCoredump diverted to " : NULL,
+                                context->is_journald && filename ? filename : NULL);
+
+        core_message = strjoina(core_message, stacktrace ? "\n\n" : NULL, stacktrace);
+
+        if (context->is_journald) {
+                /* We cannot log to the journal, so just print the message.
+                 * The target was set previously to something safe. */
+                log_dispatch(LOG_ERR, 0, core_message);
+                return 0;
+        }
+
+        (void) iovw_put_string_field(iovw, "MESSAGE=", core_message);
+
+        if (truncated)
+                (void) iovw_put_string_field(iovw, "COREDUMP_TRUNCATED=", "1");
+
+        /* Optionally store the entire coredump in the journal */
+        if (arg_storage == COREDUMP_STORAGE_JOURNAL) {
+                if (coredump_size <= arg_journal_size_max) {
+                        size_t sz = 0;
+
+                        /* Store the coredump itself in the journal */
+
+                        r = allocate_journal_field(coredump_fd, (size_t) coredump_size, &coredump_data, &sz);
+                        if (r >= 0) {
+                                if (iovw_put(iovw, coredump_data, sz) >= 0)
+                                        TAKE_PTR(coredump_data);
+                        } else
+                                log_warning_errno(r, "Failed to attach the core to the journal entry: %m");
+                } else
+                        log_info("The core will not be stored: size %"PRIu64" is greater than %"PRIu64" (the configured maximum)",
+                                 coredump_size, arg_journal_size_max);
+        }
+
+        r = sd_journal_sendv(iovw->iovec, iovw->count);
+        if (r < 0)
+                return log_error_errno(r, "Failed to log coredump: %m");
+
+        return 0;
+}
+
+static int save_context(Context *context, const struct iovec_wrapper *iovw) {
+        unsigned n, i, count = 0;
+        const char *unit;
+        int r;
+
+        assert(context);
+        assert(iovw);
+        assert(iovw->count >= _META_ARGV_MAX);
+
+        /* The context does not allocate any memory on its own */
+
+        for (n = 0; n < iovw->count; n++) {
+                struct iovec *iovec = iovw->iovec + n;
+
+                for (i = 0; i < ELEMENTSOF(meta_field_names); i++) {
+                        char *p;
+
+                        /* Note that these strings are NUL terminated, because we made sure that a
+                         * trailing NUL byte is in the buffer, though not included in the iov_len
+                         * count (see process_socket() and gather_pid_metadata_*()) */
+                        assert(((char*) iovec->iov_base)[iovec->iov_len] == 0);
+
+                        p = startswith(iovec->iov_base, meta_field_names[i]);
+                        if (p) {
+                                context->meta[i] = p;
+                                count++;
+                                break;
+                        }
+                }
+        }
+
+        if (!context->meta[META_ARGV_PID])
+                return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                                       "Failed to find the PID of crashing process");
+
+        r = parse_pid(context->meta[META_ARGV_PID], &context->pid);
+        if (r < 0)
+                return log_error_errno(r, "Failed to parse PID \"%s\": %m", context->meta[META_ARGV_PID]);
+
+        unit = context->meta[META_UNIT];
+        context->is_pid1 = streq(context->meta[META_ARGV_PID], "1") || streq_ptr(unit, SPECIAL_INIT_SCOPE);
+        context->is_journald = streq_ptr(unit, SPECIAL_JOURNALD_SERVICE);
+
+        return 0;
+}
+
+static int process_socket(int fd) {
+        _cleanup_close_ int input_fd = -1;
+        Context context = {};
+        struct iovec_wrapper iovw = {};
+        struct iovec iovec;
+        int i, r;
+
+        assert(fd >= 0);
+
+        log_setup_service();
+
+        log_debug("Processing coredump received on stdin...");
+
+        for (;;) {
+                CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(int))) control;
+                struct msghdr mh = {
+                        .msg_control = &control,
+                        .msg_controllen = sizeof(control),
+                        .msg_iovlen = 1,
+                };
+                ssize_t n;
+                ssize_t l;
+
+                l = next_datagram_size_fd(fd);
+                if (l < 0) {
+                        r = log_error_errno(l, "Failed to determine datagram size to read: %m");
+                        goto finish;
+                }
+
+                iovec.iov_len = l;
+                iovec.iov_base = malloc(l + 1);
+                if (!iovec.iov_base) {
+                        r = log_oom();
+                        goto finish;
+                }
+
+                mh.msg_iov = &iovec;
+
+                n = recvmsg_safe(fd, &mh, MSG_CMSG_CLOEXEC);
+                if (n < 0)  {
+                        free(iovec.iov_base);
+                        r = log_error_errno(n, "Failed to receive datagram: %m");
+                        goto finish;
+                }
+
+                /* The final zero-length datagram carries the file descriptor and tells us
+                 * that we're done. */
+                if (n == 0) {
+                        struct cmsghdr *found;
+
+                        free(iovec.iov_base);
+
+                        found = cmsg_find(&mh, SOL_SOCKET, SCM_RIGHTS, CMSG_LEN(sizeof(int)));
+                        if (!found) {
+                                cmsg_close_all(&mh);
+                                r = log_error_errno(SYNTHETIC_ERRNO(EBADMSG),
+                                                    "Coredump file descriptor missing.");
+                                goto finish;
+                        }
+
+                        assert(input_fd < 0);
+                        input_fd = *(int*) CMSG_DATA(found);
+                        break;
+                } else
+                        cmsg_close_all(&mh);
+
+                /* Add trailing NUL byte, in case these are strings */
+                ((char*) iovec.iov_base)[n] = 0;
+                iovec.iov_len = (size_t) n;
+
+                r = iovw_put(&iovw, iovec.iov_base, iovec.iov_len);
+                if (r < 0)
+                        goto finish;
+        }
+
+        /* Make sure we got all data we really need */
+        assert(input_fd >= 0);
+
+        r = save_context(&context, &iovw);
+        if (r < 0)
+                goto finish;
+
+        /* Make sure we received at least all fields we need. */
+        for (i = 0; i < _META_MANDATORY_MAX; i++)
+                if (!context.meta[i]) {
+                        r = log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                                            "A mandatory argument (%i) has not been sent, aborting.",
+                                            i);
+                        goto finish;
+                }
+
+        r = submit_coredump(&context, &iovw, input_fd);
+
+finish:
+        iovw_free_contents(&iovw, true);
+        return r;
+}
+
+static int send_iovec(const struct iovec_wrapper *iovw, int input_fd) {
+
+        static const union sockaddr_union sa = {
+                .un.sun_family = AF_UNIX,
+                .un.sun_path = "/run/systemd/coredump",
+        };
+        _cleanup_close_ int fd = -1;
+        size_t i;
+        int r;
+
+        assert(iovw);
+        assert(input_fd >= 0);
+
+        fd = socket(AF_UNIX, SOCK_SEQPACKET|SOCK_CLOEXEC, 0);
+        if (fd < 0)
+                return log_error_errno(errno, "Failed to create coredump socket: %m");
+
+        if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
+                return log_error_errno(errno, "Failed to connect to coredump service: %m");
+
+        for (i = 0; i < iovw->count; i++) {
+                struct msghdr mh = {
+                        .msg_iov = iovw->iovec + i,
+                        .msg_iovlen = 1,
+                };
+                struct iovec copy[2];
+
+                for (;;) {
+                        if (sendmsg(fd, &mh, MSG_NOSIGNAL) >= 0)
+                                break;
+
+                        if (errno == EMSGSIZE && mh.msg_iov[0].iov_len > 0) {
+                                /* This field didn't fit? That's a pity. Given that this is
+                                 * just metadata, let's truncate the field at half, and try
+                                 * again. We append three dots, in order to show that this is
+                                 * truncated. */
+
+                                if (mh.msg_iov != copy) {
+                                        /* We don't want to modify the caller's iovec, hence
+                                         * let's create our own array, consisting of two new
+                                         * iovecs, where the first is a (truncated) copy of
+                                         * what we want to send, and the second one contains
+                                         * the trailing dots. */
+                                        copy[0] = iovw->iovec[i];
+                                        copy[1] = IOVEC_MAKE(((char[]){'.', '.', '.'}), 3);
+
+                                        mh.msg_iov = copy;
+                                        mh.msg_iovlen = 2;
+                                }
+
+                                copy[0].iov_len /= 2; /* halve it, and try again */
+                                continue;
+                        }
+
+                        return log_error_errno(errno, "Failed to send coredump datagram: %m");
+                }
+        }
+
+        r = send_one_fd(fd, input_fd, 0);
+        if (r < 0)
+                return log_error_errno(r, "Failed to send coredump fd: %m");
+
+        return 0;
+}
+
+static int gather_pid_metadata_from_argv(
+                struct iovec_wrapper *iovw,
+                Context *context,
+                int argc, char **argv) {
+
+        _cleanup_free_ char *free_timestamp = NULL;
+        int i, r, signo;
+        char *t;
+
+        /* We gather all metadata that were passed via argv[] into an array of iovecs that
+         * we'll forward to the socket unit */
+
+        if (argc < _META_ARGV_MAX)
+                return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                                       "Not enough arguments passed by the kernel (%i, expected %i).",
+                                       argc, _META_ARGV_MAX);
+
+        for (i = 0; i < _META_ARGV_MAX; i++) {
+
+                t = argv[i];
+
+                switch (i) {
+
+                case META_ARGV_TIMESTAMP:
+                        /* The journal fields contain the timestamp padded with six
+                         * zeroes, so that the kernel-supplied 1s granularity timestamps
+                         * becomes 1µs granularity, i.e. the granularity systemd usually
+                         * operates in. */
+                        t = free_timestamp = strjoin(argv[i], "000000");
+                        if (!t)
+                                return log_oom();
+                        break;
+
+                case META_ARGV_SIGNAL:
+                        /* For signal, record its pretty name too */
+                        if (safe_atoi(argv[i], &signo) >= 0 && SIGNAL_VALID(signo))
+                                (void) iovw_put_string_field(iovw, "COREDUMP_SIGNAL_NAME=SIG",
+                                                             signal_to_string(signo));
+                        break;
+
+                default:
+                        break;
+                }
+
+                r = iovw_put_string_field(iovw, meta_field_names[i], t);
+                if (r < 0)
+                        return r;
+        }
+
+        /* Cache some of the process metadata we collected so far and that we'll need to
+         * access soon */
+        return save_context(context, iovw);
+}
+
+static int gather_pid_metadata(struct iovec_wrapper *iovw, Context *context) {
+        uid_t owner_uid;
+        pid_t pid;
+        char *t;
+        const char *p;
+        int r;
+
+        /* Note that if we fail on oom later on, we do not roll-back changes to the iovec
+         * structure. (It remains valid, with the first iovec fields initialized.) */
+
+        pid = context->pid;
+
+        /* The following is mandatory */
+        r = get_process_comm(pid, &t);
+        if (r < 0)
+                return log_error_errno(r, "Failed to get COMM: %m");
+
+        r = iovw_put_string_field_free(iovw, "COREDUMP_COMM=", t);
+        if (r < 0)
+                return r;
+
+        /* The following are optional but we used them if present */
+        r = get_process_exe(pid, &t);
+        if (r >= 0)
+                r = iovw_put_string_field_free(iovw, "COREDUMP_EXE=", t);
+        if (r < 0)
+                log_warning_errno(r, "Failed to get EXE, ignoring: %m");
+
+        if (cg_pid_get_unit(pid, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_UNIT=", t);
+
+        /* The next are optional */
+        if (cg_pid_get_user_unit(pid, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_USER_UNIT=", t);
+
+        if (sd_pid_get_session(pid, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_SESSION=", t);
+
+        if (sd_pid_get_owner_uid(pid, &owner_uid) >= 0) {
+                r = asprintf(&t, UID_FMT, owner_uid);
+                if (r > 0)
+                        (void) iovw_put_string_field_free(iovw, "COREDUMP_OWNER_UID=", t);
+        }
+
+        if (sd_pid_get_slice(pid, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_SLICE=", t);
+
+        if (get_process_cmdline(pid, SIZE_MAX, 0, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_CMDLINE=", t);
+
+        if (cg_pid_get_path_shifted(pid, NULL, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_CGROUP=", t);
+
+        if (compose_open_fds(pid, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_OPEN_FDS=", t);
+
+        p = procfs_file_alloca(pid, "status");
+        if (read_full_file(p, &t, NULL) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_STATUS=", t);
+
+        p = procfs_file_alloca(pid, "maps");
+        if (read_full_file(p, &t, NULL) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_MAPS=", t);
+
+        p = procfs_file_alloca(pid, "limits");
+        if (read_full_file(p, &t, NULL) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_LIMITS=", t);
+
+        p = procfs_file_alloca(pid, "cgroup");
+        if (read_full_file(p, &t, NULL) >=0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_CGROUP=", t);
+
+        p = procfs_file_alloca(pid, "mountinfo");
+        if (read_full_file(p, &t, NULL) >=0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_PROC_MOUNTINFO=", t);
+
+        if (get_process_cwd(pid, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_CWD=", t);
+
+        if (get_process_root(pid, &t) >= 0) {
+                bool proc_self_root_is_slash;
+
+                proc_self_root_is_slash = strcmp(t, "/") == 0;
+
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_ROOT=", t);
+
+                /* If the process' root is "/", then there is a chance it has
+                 * mounted own root and hence being containerized. */
+                if (proc_self_root_is_slash && get_process_container_parent_cmdline(pid, &t) > 0)
+                        (void) iovw_put_string_field_free(iovw, "COREDUMP_CONTAINER_CMDLINE=", t);
+        }
+
+        if (get_process_environ(pid, &t) >= 0)
+                (void) iovw_put_string_field_free(iovw, "COREDUMP_ENVIRON=", t);
+
+        /* we successfully acquired all metadata */
+        return save_context(context, iovw);
+}
+
+static int process_kernel(int argc, char* argv[]) {
+        Context context = {};
+        struct iovec_wrapper *iovw;
+        int r;
+
+        log_debug("Processing coredump received from the kernel...");
+
+        iovw = iovw_new();
+        if (!iovw)
+                return log_oom();
+
+        (void) iovw_put_string_field(iovw, "MESSAGE_ID=", SD_MESSAGE_COREDUMP_STR);
+        (void) iovw_put_string_field(iovw, "PRIORITY=", STRINGIFY(LOG_CRIT));
+
+        /* Collect all process metadata passed by the kernel through argv[] */
+        r = gather_pid_metadata_from_argv(iovw, &context, argc - 1, argv + 1);
+        if (r < 0)
+                goto finish;
+
+        /* Collect the rest of the process metadata retrieved from the runtime */
+        r = gather_pid_metadata(iovw, &context);
+        if (r < 0)
+                goto finish;
+
+        if (!context.is_journald) {
+                /* OK, now we know it's not the journal, hence we can make use of it now. */
+                log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
+                log_open();
+        }
+
+        /* If this is PID 1 disable coredump collection, we'll unlikely be able to process
+         * it later on.
+         *
+         * FIXME: maybe we should disable coredumps generation from the beginning and
+         * re-enable it only when we know it's either safe (ie we're not running OOM) or
+         * it's not pid1 ? */
+        if (context.is_pid1) {
+                log_notice("Due to PID 1 having crashed coredump collection will now be turned off.");
+                disable_coredumps();
+        }
+
+        if (context.is_journald || context.is_pid1)
+                r = submit_coredump(&context, iovw, STDIN_FILENO);
+        else
+                r = send_iovec(iovw, STDIN_FILENO);
+
+ finish:
+        iovw = iovw_free_free(iovw);
+        return r;
+}
+
+static int process_backtrace(int argc, char *argv[]) {
+        Context context = {};
+        struct iovec_wrapper *iovw;
+        char *message;
+        size_t i;
+        int r;
+         _cleanup_(journal_importer_cleanup) JournalImporter importer = JOURNAL_IMPORTER_INIT(STDIN_FILENO);
+
+        log_debug("Processing backtrace on stdin...");
+
+        iovw = iovw_new();
+        if (!iovw)
+                return log_oom();
+
+        (void) iovw_put_string_field(iovw, "MESSAGE_ID=", SD_MESSAGE_BACKTRACE_STR);
+        (void) iovw_put_string_field(iovw, "PRIORITY=", STRINGIFY(LOG_CRIT));
+
+        /* Collect all process metadata from argv[] by making sure to skip the
+         * '--backtrace' option */
+        r = gather_pid_metadata_from_argv(iovw, &context, argc - 2, argv + 2);
+        if (r < 0)
+                goto finish;
+
+        /* Collect the rest of the process metadata retrieved from the runtime */
+        r = gather_pid_metadata(iovw, &context);
+        if (r < 0)
+                goto finish;
+
+        for (;;) {
+                r = journal_importer_process_data(&importer);
+                if (r < 0) {
+                        log_error_errno(r, "Failed to parse journal entry on stdin: %m");
+                        goto finish;
+                }
+                if (r == 1 ||                        /* complete entry */
+                    journal_importer_eof(&importer)) /* end of data */
+                        break;
+        }
+
+        if (journal_importer_eof(&importer)) {
+                log_warning("Did not receive a full journal entry on stdin, ignoring message sent by reporter");
+
+                message = strjoina("Process ", context.meta[META_ARGV_PID],
+                                  " (", context.meta[META_COMM], ")"
+                                  " of user ", context.meta[META_ARGV_UID],
+                                  " failed with ", context.meta[META_ARGV_SIGNAL]);
+
+                r = iovw_put_string_field(iovw, "MESSAGE=", message);
+                if (r < 0)
+                        return r;
+        } else {
+                /* The imported iovecs are not supposed to be freed by us so let's store
+                 * them at the end of the array so we can skip them while freeing the
+                 * rest. */
+                for (i = 0; i < importer.iovw.count; i++) {
+                        struct iovec *iovec = importer.iovw.iovec + i;
+
+                        iovw_put(iovw, iovec->iov_base, iovec->iov_len);
+                }
+        }
+
+        r = sd_journal_sendv(iovw->iovec, iovw->count);
+        if (r < 0)
+                log_error_errno(r, "Failed to log backtrace: %m");
+
+ finish:
+        iovw->count -= importer.iovw.count;
+        iovw = iovw_free_free(iovw);
+        return r;
+}
+
+static int run(int argc, char *argv[]) {
+        int r;
+
+        /* First, log to a safe place, since we don't know what crashed and it might
+         * be journald which we'd rather not log to then. */
+
+        log_set_target(LOG_TARGET_KMSG);
+        log_open();
+
+        /* Make sure we never enter a loop */
+        (void) prctl(PR_SET_DUMPABLE, 0);
+
+        /* Ignore all parse errors */
+        (void) parse_config();
+
+        log_debug("Selected storage '%s'.", coredump_storage_to_string(arg_storage));
+        log_debug("Selected compression %s.", yes_no(arg_compress));
+
+        r = sd_listen_fds(false);
+        if (r < 0)
+                return log_error_errno(r, "Failed to determine the number of file descriptors: %m");
+
+        /* If we got an fd passed, we are running in coredumpd mode. Otherwise we
+         * are invoked from the kernel as coredump handler. */
+        if (r == 0) {
+                if (streq_ptr(argv[1], "--backtrace"))
+                        return process_backtrace(argc, argv);
+                else
+                        return process_kernel(argc, argv);
+        } else if (r == 1)
+                return process_socket(SD_LISTEN_FDS_START);
+
+        return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                               "Received unexpected number of file descriptors.");
+}
+
+DEFINE_MAIN_FUNCTION(run);
diff --git a/src/coredump/coredump.conf b/src/coredump/coredump.conf
new file mode 100644
index 0000000..c2f0643
--- /dev/null
+++ b/src/coredump/coredump.conf
@@ -0,0 +1,21 @@
+#  This file is part of systemd.
+#
+#  systemd is free software; you can redistribute it and/or modify it
+#  under the terms of the GNU Lesser General Public License as published by
+#  the Free Software Foundation; either version 2.1 of the License, or
+#  (at your option) any later version.
+#
+# Entries in this file show the compile time defaults.
+# You can change settings by editing this file.
+# Defaults can be restored by simply deleting this file.
+#
+# See coredump.conf(5) for details.
+
+[Coredump]
+#Storage=external
+#Compress=yes
+#ProcessSizeMax=2G
+#ExternalSizeMax=2G
+#JournalSizeMax=767M
+#MaxUse=
+#KeepFree=
diff --git a/src/coredump/coredumpctl.c b/src/coredump/coredumpctl.c
new file mode 100644
index 0000000..91356ad
--- /dev/null
+++ b/src/coredump/coredumpctl.c
@@ -0,0 +1,1114 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <fcntl.h>
+#include <getopt.h>
+#include <locale.h>
+#include <stdio.h>
+#include <unistd.h>
+
+#include "sd-bus.h"
+#include "sd-journal.h"
+#include "sd-messages.h"
+
+#include "alloc-util.h"
+#include "bus-error.h"
+#include "bus-util.h"
+#include "compress.h"
+#include "def.h"
+#include "fd-util.h"
+#include "fs-util.h"
+#include "glob-util.h"
+#include "journal-internal.h"
+#include "journal-util.h"
+#include "log.h"
+#include "macro.h"
+#include "main-func.h"
+#include "pager.h"
+#include "parse-util.h"
+#include "path-util.h"
+#include "pretty-print.h"
+#include "process-util.h"
+#include "rlimit-util.h"
+#include "sigbus.h"
+#include "signal-util.h"
+#include "string-util.h"
+#include "strv.h"
+#include "terminal-util.h"
+#include "tmpfile-util.h"
+#include "user-util.h"
+#include "util.h"
+#include "verbs.h"
+
+#define SHORT_BUS_CALL_TIMEOUT_USEC (3 * USEC_PER_SEC)
+
+static usec_t arg_since = USEC_INFINITY, arg_until = USEC_INFINITY;
+static const char* arg_field = NULL;
+static const char *arg_debugger = NULL;
+static const char *arg_directory = NULL;
+static char **arg_file = NULL;
+static PagerFlags arg_pager_flags = 0;
+static int arg_no_legend = false;
+static int arg_one = false;
+static const char* arg_output = NULL;
+static bool arg_reverse = false;
+static bool arg_quiet = false;
+
+STATIC_DESTRUCTOR_REGISTER(arg_file, strv_freep);
+
+static int add_match(sd_journal *j, const char *match) {
+        _cleanup_free_ char *p = NULL;
+        const char* prefix, *pattern;
+        pid_t pid;
+        int r;
+
+        if (strchr(match, '='))
+                prefix = "";
+        else if (strchr(match, '/')) {
+                r = path_make_absolute_cwd(match, &p);
+                if (r < 0)
+                        return log_error_errno(r, "path_make_absolute_cwd(\"%s\"): %m", match);
+
+                match = p;
+                prefix = "COREDUMP_EXE=";
+        } else if (parse_pid(match, &pid) >= 0)
+                prefix = "COREDUMP_PID=";
+        else
+                prefix = "COREDUMP_COMM=";
+
+        pattern = strjoina(prefix, match);
+        log_debug("Adding match: %s", pattern);
+        r = sd_journal_add_match(j, pattern, 0);
+        if (r < 0)
+                return log_error_errno(r, "Failed to add match \"%s\": %m", match);
+
+        return 0;
+}
+
+static int add_matches(sd_journal *j, char **matches) {
+        char **match;
+        int r;
+
+        r = sd_journal_add_match(j, "MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR, 0);
+        if (r < 0)
+                return log_error_errno(r, "Failed to add match \"%s\": %m", "MESSAGE_ID=" SD_MESSAGE_COREDUMP_STR);
+
+        r = sd_journal_add_match(j, "MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR, 0);
+        if (r < 0)
+                return log_error_errno(r, "Failed to add match \"%s\": %m", "MESSAGE_ID=" SD_MESSAGE_BACKTRACE_STR);
+
+        STRV_FOREACH(match, matches) {
+                r = add_match(j, *match);
+                if (r < 0)
+                        return r;
+        }
+
+        return 0;
+}
+
+static int acquire_journal(sd_journal **ret, char **matches) {
+        _cleanup_(sd_journal_closep) sd_journal *j = NULL;
+        int r;
+
+        assert(ret);
+
+        if (arg_directory) {
+                r = sd_journal_open_directory(&j, arg_directory, 0);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to open journals in directory: %s: %m", arg_directory);
+        } else if (arg_file) {
+                r = sd_journal_open_files(&j, (const char**)arg_file, 0);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to open journal files: %m");
+        } else {
+                r = sd_journal_open(&j, SD_JOURNAL_LOCAL_ONLY);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to open journal: %m");
+        }
+
+        r = journal_access_check_and_warn(j, arg_quiet, true);
+        if (r < 0)
+                return r;
+
+        r = add_matches(j, matches);
+        if (r < 0)
+                return r;
+
+        if (DEBUG_LOGGING) {
+                _cleanup_free_ char *filter;
+
+                filter = journal_make_match_string(j);
+                log_debug("Journal filter: %s", filter);
+        }
+
+        *ret = TAKE_PTR(j);
+
+        return 0;
+}
+
+static int help(void) {
+        _cleanup_free_ char *link = NULL;
+        int r;
+
+        r = terminal_urlify_man("coredumpctl", "1", &link);
+        if (r < 0)
+                return log_oom();
+
+        printf("%s [OPTIONS...] COMMAND ...\n\n"
+               "%sList or retrieve coredumps from the journal.%s\n"
+               "\nCommands:\n"
+               "  list [MATCHES...]  List available coredumps (default)\n"
+               "  info [MATCHES...]  Show detailed information about one or more coredumps\n"
+               "  dump [MATCHES...]  Print first matching coredump to stdout\n"
+               "  debug [MATCHES...] Start a debugger for the first matching coredump\n"
+               "\nOptions:\n"
+               "  -h --help              Show this help\n"
+               "     --version           Print version string\n"
+               "     --no-pager          Do not pipe output into a pager\n"
+               "     --no-legend         Do not print the column headers\n"
+               "     --debugger=DEBUGGER Use the given debugger\n"
+               "  -1                     Show information about most recent entry only\n"
+               "  -S --since=DATE        Only print coredumps since the date\n"
+               "  -U --until=DATE        Only print coredumps until the date\n"
+               "  -r --reverse           Show the newest entries first\n"
+               "  -F --field=FIELD       List all values a certain field takes\n"
+               "  -o --output=FILE       Write output to FILE\n"
+               "     --file=PATH         Use journal file\n"
+               "  -D --directory=DIR     Use journal files from directory\n\n"
+               "  -q --quiet             Do not show info messages and privilege warning\n"
+               "\nSee the %s for details.\n"
+               , program_invocation_short_name
+               , ansi_highlight()
+               , ansi_normal()
+               , link
+        );
+
+        return 0;
+}
+
+static int parse_argv(int argc, char *argv[]) {
+        enum {
+                ARG_VERSION = 0x100,
+                ARG_NO_PAGER,
+                ARG_NO_LEGEND,
+                ARG_DEBUGGER,
+                ARG_FILE,
+        };
+
+        int c, r;
+
+        static const struct option options[] = {
+                { "help",         no_argument,       NULL, 'h'           },
+                { "version" ,     no_argument,       NULL, ARG_VERSION   },
+                { "no-pager",     no_argument,       NULL, ARG_NO_PAGER  },
+                { "no-legend",    no_argument,       NULL, ARG_NO_LEGEND },
+                { "debugger",     required_argument, NULL, ARG_DEBUGGER  },
+                { "output",       required_argument, NULL, 'o'           },
+                { "field",        required_argument, NULL, 'F'           },
+                { "file",         required_argument, NULL, ARG_FILE      },
+                { "directory",    required_argument, NULL, 'D'           },
+                { "reverse",      no_argument,       NULL, 'r'           },
+                { "since",        required_argument, NULL, 'S'           },
+                { "until",        required_argument, NULL, 'U'           },
+                { "quiet",        no_argument,       NULL, 'q'           },
+                {}
+        };
+
+        assert(argc >= 0);
+        assert(argv);
+
+        while ((c = getopt_long(argc, argv, "ho:F:1D:rS:U:q", options, NULL)) >= 0)
+                switch(c) {
+                case 'h':
+                        return help();
+
+                case ARG_VERSION:
+                        return version();
+
+                case ARG_NO_PAGER:
+                        arg_pager_flags |= PAGER_DISABLE;
+                        break;
+
+                case ARG_NO_LEGEND:
+                        arg_no_legend = true;
+                        break;
+
+                case ARG_DEBUGGER:
+                        arg_debugger = optarg;
+                        break;
+
+                case ARG_FILE:
+                        r = glob_extend(&arg_file, optarg, GLOB_NOCHECK);
+                        if (r < 0)
+                                return log_error_errno(r, "Failed to add paths: %m");
+                        break;
+
+                case 'o':
+                        if (arg_output)
+                                return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                                                       "Cannot set output more than once.");
+
+                        arg_output = optarg;
+                        break;
+
+                case 'S':
+                        r = parse_timestamp(optarg, &arg_since);
+                        if (r < 0)
+                                return log_error_errno(r, "Failed to parse timestamp '%s': %m", optarg);
+                        break;
+
+                case 'U':
+                        r = parse_timestamp(optarg, &arg_until);
+                        if (r < 0)
+                                return log_error_errno(r, "Failed to parse timestamp '%s': %m", optarg);
+                        break;
+
+                case 'F':
+                        if (arg_field)
+                                return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                                                       "Cannot use --field/-F more than once.");
+                        arg_field = optarg;
+                        break;
+
+                case '1':
+                        arg_one = true;
+                        break;
+
+                case 'D':
+                        arg_directory = optarg;
+                        break;
+
+                case 'r':
+                        arg_reverse = true;
+                        break;
+
+                case 'q':
+                        arg_quiet = true;
+                        break;
+
+                case '?':
+                        return -EINVAL;
+
+                default:
+                        assert_not_reached("Unhandled option");
+                }
+
+        if (arg_since != USEC_INFINITY && arg_until != USEC_INFINITY &&
+            arg_since > arg_until)
+                return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                                       "--since= must be before --until=.");
+
+        return 1;
+}
+
+static int retrieve(const void *data,
+                    size_t len,
+                    const char *name,
+                    char **var) {
+
+        size_t ident;
+        char *v;
+
+        ident = strlen(name) + 1; /* name + "=" */
+
+        if (len < ident)
+                return 0;
+
+        if (memcmp(data, name, ident - 1) != 0)
+                return 0;
+
+        if (((const char*) data)[ident - 1] != '=')
+                return 0;
+
+        v = strndup((const char*)data + ident, len - ident);
+        if (!v)
+                return log_oom();
+
+        free_and_replace(*var, v);
+        return 1;
+}
+
+static int print_field(FILE* file, sd_journal *j) {
+        const void *d;
+        size_t l;
+
+        assert(file);
+        assert(j);
+
+        assert(arg_field);
+
+        /* A (user-specified) field may appear more than once for a given entry.
+         * We will print all of the occurrences.
+         * This is different below for fields that systemd-coredump uses,
+         * because they cannot meaningfully appear more than once.
+         */
+        SD_JOURNAL_FOREACH_DATA(j, d, l) {
+                _cleanup_free_ char *value = NULL;
+                int r;
+
+                r = retrieve(d, l, arg_field, &value);
+                if (r < 0)
+                        return r;
+                if (r > 0)
+                        fprintf(file, "%s\n", value);
+        }
+
+        return 0;
+}
+
+#define RETRIEVE(d, l, name, arg)                    \
+        {                                            \
+                int _r = retrieve(d, l, name, &arg); \
+                if (_r < 0)                          \
+                        return _r;                   \
+                if (_r > 0)                          \
+                        continue;                    \
+        }
+
+static int print_list(FILE* file, sd_journal *j, int had_legend) {
+        _cleanup_free_ char
+                *mid = NULL, *pid = NULL, *uid = NULL, *gid = NULL,
+                *sgnl = NULL, *exe = NULL, *comm = NULL, *cmdline = NULL,
+                *filename = NULL, *truncated = NULL, *coredump = NULL;
+        const void *d;
+        size_t l;
+        usec_t t;
+        char buf[FORMAT_TIMESTAMP_MAX];
+        int r;
+        const char *present;
+        bool normal_coredump;
+
+        assert(file);
+        assert(j);
+
+        SD_JOURNAL_FOREACH_DATA(j, d, l) {
+                RETRIEVE(d, l, "MESSAGE_ID", mid);
+                RETRIEVE(d, l, "COREDUMP_PID", pid);
+                RETRIEVE(d, l, "COREDUMP_UID", uid);
+                RETRIEVE(d, l, "COREDUMP_GID", gid);
+                RETRIEVE(d, l, "COREDUMP_SIGNAL", sgnl);
+                RETRIEVE(d, l, "COREDUMP_EXE", exe);
+                RETRIEVE(d, l, "COREDUMP_COMM", comm);
+                RETRIEVE(d, l, "COREDUMP_CMDLINE", cmdline);
+                RETRIEVE(d, l, "COREDUMP_FILENAME", filename);
+                RETRIEVE(d, l, "COREDUMP_TRUNCATED", truncated);
+                RETRIEVE(d, l, "COREDUMP", coredump);
+        }
+
+        if (!pid && !uid && !gid && !sgnl && !exe && !comm && !cmdline && !filename) {
+                log_warning("Empty coredump log entry");
+                return -EINVAL;
+        }
+
+        r = sd_journal_get_realtime_usec(j, &t);
+        if (r < 0)
+                return log_error_errno(r, "Failed to get realtime timestamp: %m");
+
+        format_timestamp(buf, sizeof(buf), t);
+
+        if (!had_legend && !arg_no_legend)
+                fprintf(file, "%-*s %*s %*s %*s %*s %-*s %s\n",
+                        FORMAT_TIMESTAMP_WIDTH, "TIME",
+                        6, "PID",
+                        5, "UID",
+                        5, "GID",
+                        3, "SIG",
+                        9, "COREFILE",
+                           "EXE");
+
+        normal_coredump = streq_ptr(mid, SD_MESSAGE_COREDUMP_STR);
+
+        if (filename)
+                if (access(filename, R_OK) == 0)
+                        present = "present";
+                else if (errno == ENOENT)
+                        present = "missing";
+                else
+                        present = "error";
+        else if (coredump)
+                present = "journal";
+        else if (normal_coredump)
+                present = "none";
+        else
+                present = "-";
+
+        if (STR_IN_SET(present, "present", "journal") && truncated && parse_boolean(truncated) > 0)
+                present = "truncated";
+
+        fprintf(file, "%-*s %*s %*s %*s %*s %-*s %s\n",
+                FORMAT_TIMESTAMP_WIDTH, buf,
+                6, strna(pid),
+                5, strna(uid),
+                5, strna(gid),
+                3, normal_coredump ? strna(sgnl) : "-",
+                9, present,
+                strna(exe ?: (comm ?: cmdline)));
+
+        return 0;
+}
+
+static int print_info(FILE *file, sd_journal *j, bool need_space) {
+        _cleanup_free_ char
+                *mid = NULL, *pid = NULL, *uid = NULL, *gid = NULL,
+                *sgnl = NULL, *exe = NULL, *comm = NULL, *cmdline = NULL,
+                *unit = NULL, *user_unit = NULL, *session = NULL,
+                *boot_id = NULL, *machine_id = NULL, *hostname = NULL,
+                *slice = NULL, *cgroup = NULL, *owner_uid = NULL,
+                *message = NULL, *timestamp = NULL, *filename = NULL,
+                *truncated = NULL, *coredump = NULL;
+        const void *d;
+        size_t l;
+        bool normal_coredump;
+        int r;
+
+        assert(file);
+        assert(j);
+
+        SD_JOURNAL_FOREACH_DATA(j, d, l) {
+                RETRIEVE(d, l, "MESSAGE_ID", mid);
+                RETRIEVE(d, l, "COREDUMP_PID", pid);
+                RETRIEVE(d, l, "COREDUMP_UID", uid);
+                RETRIEVE(d, l, "COREDUMP_GID", gid);
+                RETRIEVE(d, l, "COREDUMP_SIGNAL", sgnl);
+                RETRIEVE(d, l, "COREDUMP_EXE", exe);
+                RETRIEVE(d, l, "COREDUMP_COMM", comm);
+                RETRIEVE(d, l, "COREDUMP_CMDLINE", cmdline);
+                RETRIEVE(d, l, "COREDUMP_UNIT", unit);
+                RETRIEVE(d, l, "COREDUMP_USER_UNIT", user_unit);
+                RETRIEVE(d, l, "COREDUMP_SESSION", session);
+                RETRIEVE(d, l, "COREDUMP_OWNER_UID", owner_uid);
+                RETRIEVE(d, l, "COREDUMP_SLICE", slice);
+                RETRIEVE(d, l, "COREDUMP_CGROUP", cgroup);
+                RETRIEVE(d, l, "COREDUMP_TIMESTAMP", timestamp);
+                RETRIEVE(d, l, "COREDUMP_FILENAME", filename);
+                RETRIEVE(d, l, "COREDUMP_TRUNCATED", truncated);
+                RETRIEVE(d, l, "COREDUMP", coredump);
+                RETRIEVE(d, l, "_BOOT_ID", boot_id);
+                RETRIEVE(d, l, "_MACHINE_ID", machine_id);
+                RETRIEVE(d, l, "_HOSTNAME", hostname);
+                RETRIEVE(d, l, "MESSAGE", message);
+        }
+
+        if (need_space)
+                fputs("\n", file);
+
+        normal_coredump = streq_ptr(mid, SD_MESSAGE_COREDUMP_STR);
+
+        if (comm)
+                fprintf(file,
+                        "           PID: %s%s%s (%s)\n",
+                        ansi_highlight(), strna(pid), ansi_normal(), comm);
+        else
+                fprintf(file,
+                        "           PID: %s%s%s\n",
+                        ansi_highlight(), strna(pid), ansi_normal());
+
+        if (uid) {
+                uid_t n;
+
+                if (parse_uid(uid, &n) >= 0) {
+                        _cleanup_free_ char *u = NULL;
+
+                        u = uid_to_name(n);
+                        fprintf(file,
+                                "           UID: %s (%s)\n",
+                                uid, u);
+                } else {
+                        fprintf(file,
+                                "           UID: %s\n",
+                                uid);
+                }
+        }
+
+        if (gid) {
+                gid_t n;
+
+                if (parse_gid(gid, &n) >= 0) {
+                        _cleanup_free_ char *g = NULL;
+
+                        g = gid_to_name(n);
+                        fprintf(file,
+                                "           GID: %s (%s)\n",
+                                gid, g);
+                } else {
+                        fprintf(file,
+                                "           GID: %s\n",
+                                gid);
+                }
+        }
+
+        if (sgnl) {
+                int sig;
+                const char *name = normal_coredump ? "Signal" : "Reason";
+
+                if (normal_coredump && safe_atoi(sgnl, &sig) >= 0)
+                        fprintf(file, "        %s: %s (%s)\n", name, sgnl, signal_to_string(sig));
+                else
+                        fprintf(file, "        %s: %s\n", name, sgnl);
+        }
+
+        if (timestamp) {
+                usec_t u;
+
+                r = safe_atou64(timestamp, &u);
+                if (r >= 0) {
+                        char absolute[FORMAT_TIMESTAMP_MAX], relative[FORMAT_TIMESPAN_MAX];
+
+                        fprintf(file,
+                                "     Timestamp: %s (%s)\n",
+                                format_timestamp(absolute, sizeof(absolute), u),
+                                format_timestamp_relative(relative, sizeof(relative), u));
+
+                } else
+                        fprintf(file, "     Timestamp: %s\n", timestamp);
+        }
+
+        if (cmdline)
+                fprintf(file, "  Command Line: %s\n", cmdline);
+        if (exe)
+                fprintf(file, "    Executable: %s%s%s\n", ansi_highlight(), exe, ansi_normal());
+        if (cgroup)
+                fprintf(file, " Control Group: %s\n", cgroup);
+        if (unit)
+                fprintf(file, "          Unit: %s\n", unit);
+        if (user_unit)
+                fprintf(file, "     User Unit: %s\n", user_unit);
+        if (slice)
+                fprintf(file, "         Slice: %s\n", slice);
+        if (session)
+                fprintf(file, "       Session: %s\n", session);
+        if (owner_uid) {
+                uid_t n;
+
+                if (parse_uid(owner_uid, &n) >= 0) {
+                        _cleanup_free_ char *u = NULL;
+
+                        u = uid_to_name(n);
+                        fprintf(file,
+                                "     Owner UID: %s (%s)\n",
+                                owner_uid, u);
+                } else {
+                        fprintf(file,
+                                "     Owner UID: %s\n",
+                                owner_uid);
+                }
+        }
+        if (boot_id)
+                fprintf(file, "       Boot ID: %s\n", boot_id);
+        if (machine_id)
+                fprintf(file, "    Machine ID: %s\n", machine_id);
+        if (hostname)
+                fprintf(file, "      Hostname: %s\n", hostname);
+
+        if (filename) {
+                bool inacc, trunc;
+
+                inacc = access(filename, R_OK) < 0;
+                trunc = truncated && parse_boolean(truncated) > 0;
+
+                if (inacc || trunc)
+                        fprintf(file, "       Storage: %s%s (%s%s%s)%s\n",
+                                ansi_highlight_red(),
+                                filename,
+                                inacc ? "inaccessible" : "",
+                                inacc && trunc ? ", " : "",
+                                trunc ? "truncated" : "",
+                                ansi_normal());
+                else
+                        fprintf(file, "       Storage: %s\n", filename);
+        }
+
+        else if (coredump)
+                fprintf(file, "       Storage: journal\n");
+        else
+                fprintf(file, "       Storage: none\n");
+
+        if (message) {
+                _cleanup_free_ char *m = NULL;
+
+                m = strreplace(message, "\n", "\n                ");
+
+                fprintf(file, "       Message: %s\n", strstrip(m ?: message));
+        }
+
+        return 0;
+}
+
+static int focus(sd_journal *j) {
+        int r;
+
+        r = sd_journal_seek_tail(j);
+        if (r == 0)
+                r = sd_journal_previous(j);
+        if (r < 0)
+                return log_error_errno(r, "Failed to search journal: %m");
+        if (r == 0)
+                return log_error_errno(SYNTHETIC_ERRNO(ESRCH),
+                                       "No match found.");
+        return r;
+}
+
+static int print_entry(sd_journal *j, unsigned n_found, bool verb_is_info) {
+        assert(j);
+
+        if (verb_is_info)
+                return print_info(stdout, j, n_found);
+        else if (arg_field)
+                return print_field(stdout, j);
+        else
+                return print_list(stdout, j, n_found);
+}
+
+static int dump_list(int argc, char **argv, void *userdata) {
+        _cleanup_(sd_journal_closep) sd_journal *j = NULL;
+        unsigned n_found = 0;
+        bool verb_is_info;
+        int r;
+
+        verb_is_info = (argc >= 1 && streq(argv[0], "info"));
+
+        r = acquire_journal(&j, argv + 1);
+        if (r < 0)
+                return r;
+
+        (void) pager_open(arg_pager_flags);
+
+        /* The coredumps are likely to compressed, and for just
+         * listing them we don't need to decompress them, so let's
+         * pick a fairly low data threshold here */
+        sd_journal_set_data_threshold(j, 4096);
+
+        /* "info" without pattern implies "-1" */
+        if (arg_one || (verb_is_info && argc == 1)) {
+                r = focus(j);
+                if (r < 0)
+                        return r;
+
+                return print_entry(j, 0, verb_is_info);
+        } else {
+                if (arg_since != USEC_INFINITY && !arg_reverse)
+                        r = sd_journal_seek_realtime_usec(j, arg_since);
+                else if (arg_until != USEC_INFINITY && arg_reverse)
+                        r = sd_journal_seek_realtime_usec(j, arg_until);
+                else if (arg_reverse)
+                        r = sd_journal_seek_tail(j);
+                else
+                        r = sd_journal_seek_head(j);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to seek to date: %m");
+
+                for (;;) {
+                        if (!arg_reverse)
+                                r = sd_journal_next(j);
+                        else
+                                r = sd_journal_previous(j);
+
+                        if (r < 0)
+                                return log_error_errno(r, "Failed to iterate through journal: %m");
+
+                        if (r == 0)
+                                break;
+
+                        if (arg_until != USEC_INFINITY && !arg_reverse) {
+                                usec_t usec;
+
+                                r = sd_journal_get_realtime_usec(j, &usec);
+                                if (r < 0)
+                                        return log_error_errno(r, "Failed to determine timestamp: %m");
+                                if (usec > arg_until)
+                                        continue;
+                        }
+
+                        if (arg_since != USEC_INFINITY && arg_reverse) {
+                                usec_t usec;
+
+                                r = sd_journal_get_realtime_usec(j, &usec);
+                                if (r < 0)
+                                        return log_error_errno(r, "Failed to determine timestamp: %m");
+                                if (usec < arg_since)
+                                        continue;
+                        }
+
+                        r = print_entry(j, n_found++, verb_is_info);
+                        if (r < 0)
+                                return r;
+                }
+
+                if (!arg_field && n_found <= 0) {
+                        if (!arg_quiet)
+                                log_notice("No coredumps found.");
+                        return -ESRCH;
+                }
+        }
+
+        return 0;
+}
+
+static int save_core(sd_journal *j, FILE *file, char **path, bool *unlink_temp) {
+        const char *data;
+        _cleanup_free_ char *filename = NULL;
+        size_t len;
+        int r, fd;
+        _cleanup_close_ int fdt = -1;
+        char *temp = NULL;
+
+        assert(!(file && path));         /* At most one can be specified */
+        assert(!!path == !!unlink_temp); /* Those must be specified together */
+
+        /* Look for a coredump on disk first. */
+        r = sd_journal_get_data(j, "COREDUMP_FILENAME", (const void**) &data, &len);
+        if (r == 0) {
+                r = retrieve(data, len, "COREDUMP_FILENAME", &filename);
+                if (r < 0)
+                        return r;
+                assert(r > 0);
+
+                if (access(filename, R_OK) < 0)
+                        return log_error_errno(errno, "File \"%s\" is not readable: %m", filename);
+
+                if (path && !ENDSWITH_SET(filename, ".xz", ".lz4", ".zst")) {
+                        *path = TAKE_PTR(filename);
+
+                        return 0;
+                }
+
+        } else {
+                if (r != -ENOENT)
+                        return log_error_errno(r, "Failed to retrieve COREDUMP_FILENAME field: %m");
+                /* Check that we can have a COREDUMP field. We still haven't set a high
+                 * data threshold, so we'll get a few kilobytes at most.
+                 */
+
+                r = sd_journal_get_data(j, "COREDUMP", (const void**) &data, &len);
+                if (r == -ENOENT)
+                        return log_error_errno(r, "Coredump entry has no core attached (neither internally in the journal nor externally on disk).");
+                if (r < 0)
+                        return log_error_errno(r, "Failed to retrieve COREDUMP field: %m");
+        }
+
+        if (path) {
+                const char *vt;
+
+                /* Create a temporary file to write the uncompressed core to. */
+
+                r = var_tmp_dir(&vt);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to acquire temporary directory path: %m");
+
+                temp = path_join(vt, "coredump-XXXXXX");
+                if (!temp)
+                        return log_oom();
+
+                fdt = mkostemp_safe(temp);
+                if (fdt < 0)
+                        return log_error_errno(fdt, "Failed to create temporary file: %m");
+                log_debug("Created temporary file %s", temp);
+
+                fd = fdt;
+        } else {
+                /* If neither path or file are specified, we will write to stdout. Let's now check
+                 * if stdout is connected to a tty. We checked that the file exists, or that the
+                 * core might be stored in the journal. In this second case, if we found the entry,
+                 * in all likelihood we will be able to access the COREDUMP= field.  In either case,
+                 * we stop before doing any "real" work, i.e. before starting decompression or
+                 * reading from the file or creating temporary files.
+                 */
+                if (!file) {
+                        if (on_tty())
+                                return log_error_errno(SYNTHETIC_ERRNO(ENOTTY),
+                                                       "Refusing to dump core to tty"
+                                                       " (use shell redirection or specify --output).");
+                        file = stdout;
+                }
+
+                fd = fileno(file);
+        }
+
+        if (filename) {
+#if HAVE_COMPRESSION
+                _cleanup_close_ int fdf;
+
+                fdf = open(filename, O_RDONLY | O_CLOEXEC);
+                if (fdf < 0) {
+                        r = log_error_errno(errno, "Failed to open %s: %m", filename);
+                        goto error;
+                }
+
+                r = decompress_stream(filename, fdf, fd, -1);
+                if (r < 0) {
+                        log_error_errno(r, "Failed to decompress %s: %m", filename);
+                        goto error;
+                }
+#else
+                r = log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
+                                    "Cannot decompress file. Compiled without compression support.");
+                goto error;
+#endif
+        } else {
+                ssize_t sz;
+
+                /* We want full data, nothing truncated. */
+                sd_journal_set_data_threshold(j, 0);
+
+                r = sd_journal_get_data(j, "COREDUMP", (const void**) &data, &len);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to retrieve COREDUMP field: %m");
+
+                assert(len >= 9);
+                data += 9;
+                len -= 9;
+
+                sz = write(fd, data, len);
+                if (sz < 0) {
+                        r = log_error_errno(errno, "Failed to write output: %m");
+                        goto error;
+                }
+                if (sz != (ssize_t) len) {
+                        log_error("Short write to output.");
+                        r = -EIO;
+                        goto error;
+                }
+        }
+
+        if (temp) {
+                *path = temp;
+                *unlink_temp = true;
+        }
+        return 0;
+
+error:
+        if (temp) {
+                (void) unlink(temp);
+                log_debug("Removed temporary file %s", temp);
+        }
+        return r;
+}
+
+static int dump_core(int argc, char **argv, void *userdata) {
+        _cleanup_(sd_journal_closep) sd_journal *j = NULL;
+        _cleanup_fclose_ FILE *f = NULL;
+        int r;
+
+        if (arg_field)
+                return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                                       "Option --field/-F only makes sense with list");
+
+        r = acquire_journal(&j, argv + 1);
+        if (r < 0)
+                return r;
+
+        r = focus(j);
+        if (r < 0)
+                return r;
+
+        if (arg_output) {
+                f = fopen(arg_output, "we");
+                if (!f)
+                        return log_error_errno(errno, "Failed to open \"%s\" for writing: %m", arg_output);
+        }
+
+        print_info(f ? stdout : stderr, j, false);
+
+        r = save_core(j, f, NULL, NULL);
+        if (r < 0)
+                return r;
+
+        r = sd_journal_previous(j);
+        if (r > 0 && !arg_quiet)
+                log_notice("More than one entry matches, ignoring rest.");
+
+        return 0;
+}
+
+static int run_debug(int argc, char **argv, void *userdata) {
+        _cleanup_(sd_journal_closep) sd_journal *j = NULL;
+        _cleanup_free_ char *exe = NULL, *path = NULL, *debugger = NULL;
+        bool unlink_path = false;
+        const char *data, *fork_name;
+        size_t len;
+        pid_t pid;
+        int r;
+
+        if (!arg_debugger) {
+                char *env_debugger;
+
+                env_debugger = getenv("SYSTEMD_DEBUGGER");
+                if (env_debugger)
+                        arg_debugger = env_debugger;
+                else
+                        arg_debugger = "gdb";
+        }
+
+        debugger = strdup(arg_debugger);
+        if (!debugger)
+                return -ENOMEM;
+
+        if (arg_field)
+                return log_error_errno(SYNTHETIC_ERRNO(EINVAL),
+                                       "Option --field/-F only makes sense with list");
+
+        r = acquire_journal(&j, argv + 1);
+        if (r < 0)
+                return r;
+
+        r = focus(j);
+        if (r < 0)
+                return r;
+
+        print_info(stdout, j, false);
+        fputs("\n", stdout);
+
+        r = sd_journal_get_data(j, "COREDUMP_EXE", (const void**) &data, &len);
+        if (r < 0)
+                return log_error_errno(r, "Failed to retrieve COREDUMP_EXE field: %m");
+
+        assert(len > STRLEN("COREDUMP_EXE="));
+        data += STRLEN("COREDUMP_EXE=");
+        len -= STRLEN("COREDUMP_EXE=");
+
+        exe = strndup(data, len);
+        if (!exe)
+                return log_oom();
+
+        if (endswith(exe, " (deleted)"))
+                return log_error_errno(SYNTHETIC_ERRNO(ENOENT),
+                                       "Binary already deleted.");
+
+        if (!path_is_absolute(exe))
+                return log_error_errno(SYNTHETIC_ERRNO(ENOENT),
+                                       "Binary is not an absolute path.");
+
+        r = save_core(j, NULL, &path, &unlink_path);
+        if (r < 0)
+                return r;
+
+        /* Don't interfere with gdb and its handling of SIGINT. */
+        (void) ignore_signals(SIGINT, -1);
+
+        fork_name = strjoina("(", debugger, ")");
+
+        r = safe_fork(fork_name, FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_LOG, &pid);
+        if (r < 0)
+                goto finish;
+        if (r == 0) {
+                execlp(debugger, debugger, exe, "-c", path, NULL);
+                log_open();
+                log_error_errno(errno, "Failed to invoke %s: %m", debugger);
+                _exit(EXIT_FAILURE);
+        }
+
+        r = wait_for_terminate_and_check(debugger, pid, WAIT_LOG_ABNORMAL);
+
+finish:
+        (void) default_signals(SIGINT, -1);
+
+        if (unlink_path) {
+                log_debug("Removed temporary file %s", path);
+                (void) unlink(path);
+        }
+
+        return r;
+}
+
+static int check_units_active(void) {
+        _cleanup_(sd_bus_flush_close_unrefp) sd_bus *bus = NULL;
+        _cleanup_(sd_bus_message_unrefp) sd_bus_message *m = NULL;
+        _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
+        _cleanup_(sd_bus_message_unrefp) sd_bus_message *reply = NULL;
+        int c = 0, r;
+        const char *id, *state, *substate;
+
+        if (arg_quiet)
+                return false;
+
+        r = sd_bus_default_system(&bus);
+        if (r < 0)
+                return log_error_errno(r, "Failed to acquire bus: %m");
+
+        r = sd_bus_message_new_method_call(
+                        bus,
+                        &m,
+                        "org.freedesktop.systemd1",
+                        "/org/freedesktop/systemd1",
+                        "org.freedesktop.systemd1.Manager",
+                        "ListUnitsByPatterns");
+        if (r < 0)
+                return bus_log_create_error(r);
+
+        r = sd_bus_message_append_strv(m, NULL);
+        if (r < 0)
+                return bus_log_create_error(r);
+
+        r = sd_bus_message_append_strv(m, STRV_MAKE("systemd-coredump@*.service"));
+        if (r < 0)
+                return bus_log_create_error(r);
+
+        r = sd_bus_call(bus, m, SHORT_BUS_CALL_TIMEOUT_USEC, &error, &reply);
+        if (r < 0)
+                return log_error_errno(r, "Failed to check if any systemd-coredump@.service units are running: %s",
+                                       bus_error_message(&error, r));
+
+        r = sd_bus_message_enter_container(reply, SD_BUS_TYPE_ARRAY, "(ssssssouso)");
+        if (r < 0)
+                return bus_log_parse_error(r);
+
+        while ((r = sd_bus_message_read(
+                                reply, "(ssssssouso)",
+                                &id,  NULL,  NULL,  &state,  &substate,
+                                NULL,  NULL,  NULL,  NULL,  NULL)) > 0) {
+                bool found = !STR_IN_SET(state, "inactive", "dead", "failed");
+                log_debug("Unit %s is %s/%s, %scounting it.", id, state, substate, found ? "" : "not ");
+                c += found;
+        }
+        if (r < 0)
+                return bus_log_parse_error(r);
+
+        r = sd_bus_message_exit_container(reply);
+        if (r < 0)
+                return bus_log_parse_error(r);
+
+        return c;
+}
+
+static int coredumpctl_main(int argc, char *argv[]) {
+
+        static const Verb verbs[] = {
+                { "list",  VERB_ANY, VERB_ANY, VERB_DEFAULT, dump_list },
+                { "info",  VERB_ANY, VERB_ANY, 0,            dump_list },
+                { "dump",  VERB_ANY, VERB_ANY, 0,            dump_core },
+                { "debug", VERB_ANY, VERB_ANY, 0,            run_debug },
+                { "gdb",   VERB_ANY, VERB_ANY, 0,            run_debug },
+                {}
+        };
+
+        return dispatch_verb(argc, argv, verbs, NULL);
+}
+
+static int run(int argc, char *argv[]) {
+        int r, units_active;
+
+        setlocale(LC_ALL, "");
+        log_setup_cli();
+
+        /* The journal merging logic potentially needs a lot of fds. */
+        (void) rlimit_nofile_bump(HIGH_RLIMIT_NOFILE);
+
+        r = parse_argv(argc, argv);
+        if (r <= 0)
+                return r;
+
+        sigbus_install();
+
+        units_active = check_units_active(); /* error is treated the same as 0 */
+
+        r = coredumpctl_main(argc, argv);
+
+        if (units_active > 0)
+                printf("%s-- Notice: %d systemd-coredump@.service %s, output may be incomplete.%s\n",
+                       ansi_highlight_red(),
+                       units_active, units_active == 1 ? "unit is running" : "units are running",
+                       ansi_normal());
+
+        return r;
+}
+
+DEFINE_MAIN_FUNCTION_WITH_POSITIVE_FAILURE(run);
diff --git a/src/coredump/meson.build b/src/coredump/meson.build
new file mode 100644
index 0000000..ebd99bd
--- /dev/null
+++ b/src/coredump/meson.build
@@ -0,0 +1,28 @@
+# SPDX-License-Identifier: LGPL-2.1-or-later
+
+systemd_coredump_sources = files('''
+        coredump.c
+        coredump-vacuum.c
+        coredump-vacuum.h
+'''.split())
+
+if conf.get('HAVE_ELFUTILS') == 1
+        systemd_coredump_sources += files(['stacktrace.c',
+                                           'stacktrace.h'])
+endif
+
+coredumpctl_sources = files('coredumpctl.c')
+
+if conf.get('ENABLE_COREDUMP') == 1 and install_sysconfdir
+        install_data('coredump.conf',
+                     install_dir : pkgsysconfdir)
+endif
+
+tests += [
+        [['src/coredump/test-coredump-vacuum.c',
+          'src/coredump/coredump-vacuum.c',
+          'src/coredump/coredump-vacuum.h'],
+         [],
+         [],
+         'ENABLE_COREDUMP', 'manual'],
+]
diff --git a/src/coredump/stacktrace.c b/src/coredump/stacktrace.c
new file mode 100644
index 0000000..a29ab12
--- /dev/null
+++ b/src/coredump/stacktrace.c
@@ -0,0 +1,198 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <dwarf.h>
+#include <elfutils/libdwfl.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "alloc-util.h"
+#include "fileio.h"
+#include "fd-util.h"
+#include "format-util.h"
+#include "macro.h"
+#include "stacktrace.h"
+#include "string-util.h"
+#include "util.h"
+
+#define FRAMES_MAX 64
+#define THREADS_MAX 64
+
+struct stack_context {
+        FILE *f;
+        Dwfl *dwfl;
+        Elf *elf;
+        unsigned n_thread;
+        unsigned n_frame;
+};
+
+static int frame_callback(Dwfl_Frame *frame, void *userdata) {
+        struct stack_context *c = userdata;
+        Dwarf_Addr pc, pc_adjusted, bias = 0;
+        _cleanup_free_ Dwarf_Die *scopes = NULL;
+        const char *fname = NULL, *symbol = NULL;
+        Dwfl_Module *module;
+        bool is_activation;
+        uint64_t module_offset = 0;
+
+        assert(frame);
+        assert(c);
+
+        if (c->n_frame >= FRAMES_MAX)
+                return DWARF_CB_ABORT;
+
+        if (!dwfl_frame_pc(frame, &pc, &is_activation))
+                return DWARF_CB_ABORT;
+
+        pc_adjusted = pc - (is_activation ? 0 : 1);
+
+        module = dwfl_addrmodule(c->dwfl, pc_adjusted);
+        if (module) {
+                Dwarf_Die *s, *cudie;
+                int n;
+                Dwarf_Addr start;
+
+                cudie = dwfl_module_addrdie(module, pc_adjusted, &bias);
+                if (cudie) {
+                        n = dwarf_getscopes(cudie, pc_adjusted - bias, &scopes);
+                        for (s = scopes; s < scopes + n; s++) {
+                                if (IN_SET(dwarf_tag(s), DW_TAG_subprogram, DW_TAG_inlined_subroutine, DW_TAG_entry_point)) {
+                                        Dwarf_Attribute *a, space;
+
+                                        a = dwarf_attr_integrate(s, DW_AT_MIPS_linkage_name, &space);
+                                        if (!a)
+                                                a = dwarf_attr_integrate(s, DW_AT_linkage_name, &space);
+                                        if (a)
+                                                symbol = dwarf_formstring(a);
+                                        if (!symbol)
+                                                symbol = dwarf_diename(s);
+
+                                        if (symbol)
+                                                break;
+                                }
+                        }
+                }
+
+                if (!symbol)
+                        symbol = dwfl_module_addrname(module, pc_adjusted);
+
+                fname = dwfl_module_info(module, NULL, &start, NULL, NULL, NULL, NULL, NULL);
+                module_offset = pc - start;
+        }
+
+        fprintf(c->f, "#%-2u 0x%016" PRIx64 " %s (%s + 0x%" PRIx64 ")\n", c->n_frame, (uint64_t) pc, strna(symbol), strna(fname), module_offset);
+        c->n_frame++;
+
+        return DWARF_CB_OK;
+}
+
+static int thread_callback(Dwfl_Thread *thread, void *userdata) {
+        struct stack_context *c = userdata;
+        pid_t tid;
+
+        assert(thread);
+        assert(c);
+
+        if (c->n_thread >= THREADS_MAX)
+                return DWARF_CB_ABORT;
+
+        if (c->n_thread != 0)
+                fputc('\n', c->f);
+
+        c->n_frame = 0;
+
+        tid = dwfl_thread_tid(thread);
+        fprintf(c->f, "Stack trace of thread " PID_FMT ":\n", tid);
+
+        if (dwfl_thread_getframes(thread, frame_callback, c) < 0)
+                return DWARF_CB_ABORT;
+
+        c->n_thread++;
+
+        return DWARF_CB_OK;
+}
+
+static int make_stack_trace(int fd, const char *executable, char **ret) {
+
+        static const Dwfl_Callbacks callbacks = {
+                .find_elf = dwfl_build_id_find_elf,
+                .find_debuginfo = dwfl_standard_find_debuginfo,
+        };
+
+        struct stack_context c = {};
+        char *buf = NULL;
+        size_t sz = 0;
+        int r;
+
+        assert(fd >= 0);
+        assert(ret);
+
+        if (lseek(fd, 0, SEEK_SET) == (off_t) -1)
+                return -errno;
+
+        c.f = open_memstream_unlocked(&buf, &sz);
+        if (!c.f)
+                return -ENOMEM;
+
+        elf_version(EV_CURRENT);
+
+        c.elf = elf_begin(fd, ELF_C_READ_MMAP, NULL);
+        if (!c.elf) {
+                r = -EINVAL;
+                goto finish;
+        }
+
+        c.dwfl = dwfl_begin(&callbacks);
+        if (!c.dwfl) {
+                r = -EINVAL;
+                goto finish;
+        }
+
+        if (dwfl_core_file_report(c.dwfl, c.elf, executable) < 0) {
+                r = -EINVAL;
+                goto finish;
+        }
+
+        if (dwfl_report_end(c.dwfl, NULL, NULL) != 0) {
+                r = -EINVAL;
+                goto finish;
+        }
+
+        if (dwfl_core_file_attach(c.dwfl, c.elf) < 0) {
+                r = -EINVAL;
+                goto finish;
+        }
+
+        if (dwfl_getthreads(c.dwfl, thread_callback, &c) < 0) {
+                r = -EINVAL;
+                goto finish;
+        }
+
+        c.f = safe_fclose(c.f);
+
+        *ret = TAKE_PTR(buf);
+
+        r = 0;
+
+finish:
+        if (c.dwfl)
+                dwfl_end(c.dwfl);
+
+        if (c.elf)
+                elf_end(c.elf);
+
+        safe_fclose(c.f);
+
+        free(buf);
+
+        return r;
+}
+
+void coredump_make_stack_trace(int fd, const char *executable, char **ret) {
+        int r;
+
+        r = make_stack_trace(fd, executable, ret);
+        if (r == -EINVAL)
+                log_warning("Failed to generate stack trace: %s", dwfl_errmsg(dwfl_errno()));
+        else if (r < 0)
+                log_warning_errno(r, "Failed to generate stack trace: %m");
+}
diff --git a/src/coredump/stacktrace.h b/src/coredump/stacktrace.h
new file mode 100644
index 0000000..b935748
--- /dev/null
+++ b/src/coredump/stacktrace.h
@@ -0,0 +1,4 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+#pragma once
+
+void coredump_make_stack_trace(int fd, const char *executable, char **ret);
diff --git a/src/coredump/test-coredump-vacuum.c b/src/coredump/test-coredump-vacuum.c
new file mode 100644
index 0000000..ac212ea
--- /dev/null
+++ b/src/coredump/test-coredump-vacuum.c
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: LGPL-2.1-or-later */
+
+#include <stdlib.h>
+
+#include "coredump-vacuum.h"
+
+int main(int argc, char *argv[]) {
+
+        if (coredump_vacuum(-1, (uint64_t) -1, 70 * 1024) < 0)
+                return EXIT_FAILURE;
+
+        return EXIT_SUCCESS;
+}