1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2011 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
22 #include <sys/epoll.h>
23 #include <sys/socket.h>
25 #include <sys/signalfd.h>
29 #include <sys/ioctl.h>
30 #include <linux/sockios.h>
31 #include <sys/statvfs.h>
35 #include <systemd/sd-journal.h>
36 #include <systemd/sd-messages.h>
37 #include <systemd/sd-daemon.h>
40 #include <systemd/sd-login.h>
45 #include "journal-file.h"
46 #include "socket-util.h"
47 #include "cgroup-util.h"
51 #include "conf-parser.h"
52 #include "journal-internal.h"
53 #include "journal-vacuum.h"
54 #include "journal-authenticate.h"
56 #include "journald-rate-limit.h"
57 #include "journald-kmsg.h"
58 #include "journald-syslog.h"
59 #include "journald-stream.h"
60 #include "journald-console.h"
61 #include "journald-native.h"
65 #include <acl/libacl.h>
70 #include <selinux/selinux.h>
73 #define USER_JOURNALS_MAX 1024
75 #define DEFAULT_RATE_LIMIT_INTERVAL (10*USEC_PER_SEC)
76 #define DEFAULT_RATE_LIMIT_BURST 200
78 #define RECHECK_AVAILABLE_SPACE_USEC (30*USEC_PER_SEC)
80 static const char* const storage_table[] = {
81 [STORAGE_AUTO] = "auto",
82 [STORAGE_VOLATILE] = "volatile",
83 [STORAGE_PERSISTENT] = "persistent",
84 [STORAGE_NONE] = "none"
87 DEFINE_STRING_TABLE_LOOKUP(storage, Storage);
88 DEFINE_CONFIG_PARSE_ENUM(config_parse_storage, storage, Storage, "Failed to parse storage setting");
90 static const char* const split_mode_table[] = {
91 [SPLIT_NONE] = "none",
93 [SPLIT_LOGIN] = "login"
96 DEFINE_STRING_TABLE_LOOKUP(split_mode, SplitMode);
97 DEFINE_CONFIG_PARSE_ENUM(config_parse_split_mode, split_mode, SplitMode, "Failed to parse split mode setting");
99 static uint64_t available_space(Server *s) {
104 uint64_t sum = 0, avail = 0, ss_avail = 0;
110 ts = now(CLOCK_MONOTONIC);
112 if (s->cached_available_space_timestamp + RECHECK_AVAILABLE_SPACE_USEC > ts)
113 return s->cached_available_space;
115 r = sd_id128_get_machine(&machine);
119 if (s->system_journal) {
120 f = "/var/log/journal/";
121 m = &s->system_metrics;
123 f = "/run/log/journal/";
124 m = &s->runtime_metrics;
129 p = strappend(f, sd_id128_to_string(machine, ids));
139 if (fstatvfs(dirfd(d), &ss) < 0)
145 union dirent_storage buf;
147 r = readdir_r(d, &buf.de, &de);
154 if (!endswith(de->d_name, ".journal") &&
155 !endswith(de->d_name, ".journal~"))
158 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
161 if (!S_ISREG(st.st_mode))
164 sum += (uint64_t) st.st_blocks * 512UL;
167 avail = sum >= m->max_use ? 0 : m->max_use - sum;
169 ss_avail = ss.f_bsize * ss.f_bavail;
171 ss_avail = ss_avail < m->keep_free ? 0 : ss_avail - m->keep_free;
173 if (ss_avail < avail)
176 s->cached_available_space = avail;
177 s->cached_available_space_timestamp = ts;
185 static void server_read_file_gid(Server *s) {
186 const char *adm = "adm";
191 if (s->file_gid_valid)
194 r = get_group_creds(&adm, &s->file_gid);
196 log_warning("Failed to resolve 'adm' group: %s", strerror(-r));
198 /* if we couldn't read the gid, then it will be 0, but that's
199 * fine and we shouldn't try to resolve the group again, so
200 * let's just pretend it worked right-away. */
201 s->file_gid_valid = true;
204 static void server_fix_perms(Server *s, JournalFile *f, uid_t uid) {
209 acl_permset_t permset;
214 server_read_file_gid(s);
216 r = fchmod_and_fchown(f->fd, 0640, 0, s->file_gid);
218 log_warning("Failed to fix access mode/rights on %s, ignoring: %s", f->path, strerror(-r));
224 acl = acl_get_fd(f->fd);
226 log_warning("Failed to read ACL on %s, ignoring: %m", f->path);
230 r = acl_find_uid(acl, uid, &entry);
233 if (acl_create_entry(&acl, &entry) < 0 ||
234 acl_set_tag_type(entry, ACL_USER) < 0 ||
235 acl_set_qualifier(entry, &uid) < 0) {
236 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
241 if (acl_get_permset(entry, &permset) < 0 ||
242 acl_add_perm(permset, ACL_READ) < 0 ||
243 acl_calc_mask(&acl) < 0) {
244 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
248 if (acl_set_fd(f->fd, acl) < 0)
249 log_warning("Failed to set ACL on %s, ignoring: %m", f->path);
256 static JournalFile* find_journal(Server *s, uid_t uid) {
264 /* We split up user logs only on /var, not on /run. If the
265 * runtime file is open, we write to it exclusively, in order
266 * to guarantee proper order as soon as we flush /run to
267 * /var and close the runtime file. */
269 if (s->runtime_journal)
270 return s->runtime_journal;
273 return s->system_journal;
275 r = sd_id128_get_machine(&machine);
277 return s->system_journal;
279 f = hashmap_get(s->user_journals, UINT32_TO_PTR(uid));
283 if (asprintf(&p, "/var/log/journal/" SD_ID128_FORMAT_STR "/user-%lu.journal",
284 SD_ID128_FORMAT_VAL(machine), (unsigned long) uid) < 0)
285 return s->system_journal;
287 while (hashmap_size(s->user_journals) >= USER_JOURNALS_MAX) {
288 /* Too many open? Then let's close one */
289 f = hashmap_steal_first(s->user_journals);
291 journal_file_close(f);
294 r = journal_file_open_reliably(p, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, s->system_journal, &f);
298 return s->system_journal;
300 server_fix_perms(s, f, uid);
302 r = hashmap_put(s->user_journals, UINT32_TO_PTR(uid), f);
304 journal_file_close(f);
305 return s->system_journal;
311 static void server_rotate(Server *s) {
317 log_debug("Rotating...");
319 if (s->runtime_journal) {
320 r = journal_file_rotate(&s->runtime_journal, s->compress, false);
322 if (s->runtime_journal)
323 log_error("Failed to rotate %s: %s", s->runtime_journal->path, strerror(-r));
325 log_error("Failed to create new runtime journal: %s", strerror(-r));
327 server_fix_perms(s, s->runtime_journal, 0);
330 if (s->system_journal) {
331 r = journal_file_rotate(&s->system_journal, s->compress, s->seal);
333 if (s->system_journal)
334 log_error("Failed to rotate %s: %s", s->system_journal->path, strerror(-r));
336 log_error("Failed to create new system journal: %s", strerror(-r));
339 server_fix_perms(s, s->system_journal, 0);
342 HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
343 r = journal_file_rotate(&f, s->compress, s->seal);
346 log_error("Failed to rotate %s: %s", f->path, strerror(-r));
348 log_error("Failed to create user journal: %s", strerror(-r));
350 hashmap_replace(s->user_journals, k, f);
351 server_fix_perms(s, f, PTR_TO_UINT32(k));
356 static void server_vacuum(Server *s) {
362 log_debug("Vacuuming...");
364 r = sd_id128_get_machine(&machine);
366 log_error("Failed to get machine ID: %s", strerror(-r));
370 sd_id128_to_string(machine, ids);
372 if (s->system_journal) {
373 p = strappend("/var/log/journal/", ids);
379 r = journal_directory_vacuum(p, s->system_metrics.max_use, s->system_metrics.keep_free);
380 if (r < 0 && r != -ENOENT)
381 log_error("Failed to vacuum %s: %s", p, strerror(-r));
385 if (s->runtime_journal) {
386 p = strappend("/run/log/journal/", ids);
392 r = journal_directory_vacuum(p, s->runtime_metrics.max_use, s->runtime_metrics.keep_free);
393 if (r < 0 && r != -ENOENT)
394 log_error("Failed to vacuum %s: %s", p, strerror(-r));
398 s->cached_available_space_timestamp = 0;
401 static char *shortened_cgroup_path(pid_t pid) {
403 char *process_path, *init_path, *path;
407 r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &process_path);
411 r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 1, &init_path);
417 if (endswith(init_path, "/system"))
418 init_path[strlen(init_path) - 7] = 0;
419 else if (streq(init_path, "/"))
422 if (startswith(process_path, init_path)) {
425 p = strdup(process_path + strlen(init_path));
443 static void write_to_journal(Server *s, uid_t uid, struct iovec *iovec, unsigned n) {
445 bool vacuumed = false;
452 f = find_journal(s, uid);
456 if (journal_file_rotate_suggested(f)) {
457 log_debug("Journal header limits reached or header out-of-date, rotating.");
462 f = find_journal(s, uid);
468 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
473 (r != -E2BIG && /* hit limit */
474 r != -EFBIG && /* hit fs limit */
475 r != -EDQUOT && /* quota hit */
476 r != -ENOSPC && /* disk full */
477 r != -EBADMSG && /* corrupted */
478 r != -ENODATA && /* truncated */
479 r != -EHOSTDOWN && /* other machine */
480 r != -EPROTONOSUPPORT && /* unsupported feature */
481 r != -EBUSY && /* unclean shutdown */
482 r != -ESHUTDOWN /* already archived */)) {
483 log_error("Failed to write entry, ignoring: %s", strerror(-r));
487 if (r == -E2BIG || r == -EFBIG || r == EDQUOT || r == ENOSPC)
488 log_debug("Allocation limit reached, rotating.");
489 else if (r == -EHOSTDOWN)
490 log_info("Journal file from other machine, rotating.");
491 else if (r == -EBUSY)
492 log_info("Unclean shutdown, rotating.");
494 log_warning("Journal file corrupted, rotating.");
500 f = find_journal(s, uid);
504 log_debug("Retrying write.");
508 static void dispatch_message_real(
510 struct iovec *iovec, unsigned n, unsigned m,
513 const char *label, size_t label_len,
514 const char *unit_id) {
516 char *pid = NULL, *uid = NULL, *gid = NULL,
517 *source_time = NULL, *boot_id = NULL, *machine_id = NULL,
518 *comm = NULL, *cmdline = NULL, *hostname = NULL,
519 *audit_session = NULL, *audit_loginuid = NULL,
520 *exe = NULL, *cgroup = NULL, *session = NULL,
521 *owner_uid = NULL, *unit = NULL, *selinux_context = NULL;
527 uid_t loginuid = 0, realuid = 0;
532 assert(n + N_IOVEC_META_FIELDS <= m);
540 realuid = ucred->uid;
542 if (asprintf(&pid, "_PID=%lu", (unsigned long) ucred->pid) >= 0)
543 IOVEC_SET_STRING(iovec[n++], pid);
545 if (asprintf(&uid, "_UID=%lu", (unsigned long) ucred->uid) >= 0)
546 IOVEC_SET_STRING(iovec[n++], uid);
548 if (asprintf(&gid, "_GID=%lu", (unsigned long) ucred->gid) >= 0)
549 IOVEC_SET_STRING(iovec[n++], gid);
551 r = get_process_comm(ucred->pid, &t);
553 comm = strappend("_COMM=", t);
557 IOVEC_SET_STRING(iovec[n++], comm);
560 r = get_process_exe(ucred->pid, &t);
562 exe = strappend("_EXE=", t);
566 IOVEC_SET_STRING(iovec[n++], exe);
569 r = get_process_cmdline(ucred->pid, LINE_MAX, false, &t);
571 cmdline = strappend("_CMDLINE=", t);
575 IOVEC_SET_STRING(iovec[n++], cmdline);
578 r = audit_session_from_pid(ucred->pid, &audit);
580 if (asprintf(&audit_session, "_AUDIT_SESSION=%lu", (unsigned long) audit) >= 0)
581 IOVEC_SET_STRING(iovec[n++], audit_session);
583 r = audit_loginuid_from_pid(ucred->pid, &loginuid);
585 if (asprintf(&audit_loginuid, "_AUDIT_LOGINUID=%lu", (unsigned long) loginuid) >= 0)
586 IOVEC_SET_STRING(iovec[n++], audit_loginuid);
588 t = shortened_cgroup_path(ucred->pid);
590 cgroup = strappend("_SYSTEMD_CGROUP=", t);
594 IOVEC_SET_STRING(iovec[n++], cgroup);
598 if (sd_pid_get_session(ucred->pid, &t) >= 0) {
599 session = strappend("_SYSTEMD_SESSION=", t);
603 IOVEC_SET_STRING(iovec[n++], session);
606 if (sd_pid_get_owner_uid(ucred->uid, &owner) >= 0)
607 if (asprintf(&owner_uid, "_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner) >= 0)
608 IOVEC_SET_STRING(iovec[n++], owner_uid);
611 if (cg_pid_get_unit(ucred->pid, &t) >= 0) {
612 unit = strappend("_SYSTEMD_UNIT=", t);
615 unit = strappend("_SYSTEMD_UNIT=", unit_id);
618 IOVEC_SET_STRING(iovec[n++], unit);
622 selinux_context = malloc(sizeof("_SELINUX_CONTEXT=") + label_len);
623 if (selinux_context) {
624 memcpy(selinux_context, "_SELINUX_CONTEXT=", sizeof("_SELINUX_CONTEXT=")-1);
625 memcpy(selinux_context+sizeof("_SELINUX_CONTEXT=")-1, label, label_len);
626 selinux_context[sizeof("_SELINUX_CONTEXT=")-1+label_len] = 0;
627 IOVEC_SET_STRING(iovec[n++], selinux_context);
630 security_context_t con;
632 if (getpidcon(ucred->pid, &con) >= 0) {
633 selinux_context = strappend("_SELINUX_CONTEXT=", con);
635 IOVEC_SET_STRING(iovec[n++], selinux_context);
644 if (asprintf(&source_time, "_SOURCE_REALTIME_TIMESTAMP=%llu",
645 (unsigned long long) timeval_load(tv)) >= 0)
646 IOVEC_SET_STRING(iovec[n++], source_time);
649 /* Note that strictly speaking storing the boot id here is
650 * redundant since the entry includes this in-line
651 * anyway. However, we need this indexed, too. */
652 r = sd_id128_get_boot(&id);
654 if (asprintf(&boot_id, "_BOOT_ID=%s", sd_id128_to_string(id, idbuf)) >= 0)
655 IOVEC_SET_STRING(iovec[n++], boot_id);
657 r = sd_id128_get_machine(&id);
659 if (asprintf(&machine_id, "_MACHINE_ID=%s", sd_id128_to_string(id, idbuf)) >= 0)
660 IOVEC_SET_STRING(iovec[n++], machine_id);
662 t = gethostname_malloc();
664 hostname = strappend("_HOSTNAME=", t);
667 IOVEC_SET_STRING(iovec[n++], hostname);
673 s->split_mode == SPLIT_NONE ? 0 :
674 (s->split_mode == SPLIT_UID ? realuid :
675 (realuid == 0 ? 0 : loginuid)), iovec, n);
688 free(audit_loginuid);
693 free(selinux_context);
696 void server_driver_message(Server *s, sd_id128_t message_id, const char *format, ...) {
697 char mid[11 + 32 + 1];
698 char buffer[16 + LINE_MAX + 1];
699 struct iovec iovec[N_IOVEC_META_FIELDS + 4];
707 IOVEC_SET_STRING(iovec[n++], "PRIORITY=6");
708 IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=driver");
710 memcpy(buffer, "MESSAGE=", 8);
711 va_start(ap, format);
712 vsnprintf(buffer + 8, sizeof(buffer) - 8, format, ap);
714 char_array_0(buffer);
715 IOVEC_SET_STRING(iovec[n++], buffer);
717 if (!sd_id128_equal(message_id, SD_ID128_NULL)) {
718 snprintf(mid, sizeof(mid), "MESSAGE_ID=" SD_ID128_FORMAT_STR, SD_ID128_FORMAT_VAL(message_id));
720 IOVEC_SET_STRING(iovec[n++], mid);
724 ucred.pid = getpid();
725 ucred.uid = getuid();
726 ucred.gid = getgid();
728 dispatch_message_real(s, iovec, n, ELEMENTSOF(iovec), &ucred, NULL, NULL, 0, NULL);
731 void server_dispatch_message(
733 struct iovec *iovec, unsigned n, unsigned m,
736 const char *label, size_t label_len,
741 char *path = NULL, *c;
744 assert(iovec || n == 0);
749 if (LOG_PRI(priority) > s->max_level_store)
755 path = shortened_cgroup_path(ucred->pid);
759 /* example: /user/lennart/3/foobar
760 * /system/dbus.service/foobar
762 * So let's cut of everything past the third /, since that is
763 * wher user directories start */
765 c = strchr(path, '/');
767 c = strchr(c+1, '/');
769 c = strchr(c+1, '/');
775 rl = journal_rate_limit_test(s->rate_limit, path, priority & LOG_PRIMASK, available_space(s));
782 /* Write a suppression message if we suppressed something */
784 server_driver_message(s, SD_MESSAGE_JOURNAL_DROPPED, "Suppressed %u messages from %s", rl - 1, path);
789 dispatch_message_real(s, iovec, n, m, ucred, tv, label, label_len, unit_id);
793 static int system_journal_open(Server *s) {
799 r = sd_id128_get_machine(&machine);
803 sd_id128_to_string(machine, ids);
805 if (!s->system_journal &&
806 (s->storage == STORAGE_PERSISTENT || s->storage == STORAGE_AUTO) &&
807 access("/run/systemd/journal/flushed", F_OK) >= 0) {
809 /* If in auto mode: first try to create the machine
810 * path, but not the prefix.
812 * If in persistent mode: create /var/log/journal and
813 * the machine path */
815 if (s->storage == STORAGE_PERSISTENT)
816 (void) mkdir("/var/log/journal/", 0755);
818 fn = strappend("/var/log/journal/", ids);
822 (void) mkdir(fn, 0755);
825 fn = strjoin("/var/log/journal/", ids, "/system.journal", NULL);
829 r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, NULL, &s->system_journal);
833 char fb[FORMAT_BYTES_MAX];
835 server_fix_perms(s, s->system_journal, 0);
836 server_driver_message(s, SD_ID128_NULL, "Allowing system journal files to grow to %s.",
837 format_bytes(fb, sizeof(fb), s->system_metrics.max_use));
841 if (r != -ENOENT && r != -EROFS)
842 log_warning("Failed to open system journal: %s", strerror(-r));
848 if (!s->runtime_journal &&
849 (s->storage != STORAGE_NONE)) {
851 fn = strjoin("/run/log/journal/", ids, "/system.journal", NULL);
855 if (s->system_journal) {
857 /* Try to open the runtime journal, but only
858 * if it already exists, so that we can flush
859 * it into the system journal */
861 r = journal_file_open(fn, O_RDWR, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal);
866 log_warning("Failed to open runtime journal: %s", strerror(-r));
873 /* OK, we really need the runtime journal, so create
874 * it if necessary. */
876 (void) mkdir_parents(fn, 0755);
877 r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal);
881 log_error("Failed to open runtime journal: %s", strerror(-r));
886 if (s->runtime_journal) {
887 char fb[FORMAT_BYTES_MAX];
889 server_fix_perms(s, s->runtime_journal, 0);
890 server_driver_message(s, SD_ID128_NULL, "Allowing runtime journal files to grow to %s.",
891 format_bytes(fb, sizeof(fb), s->runtime_metrics.max_use));
898 static int server_flush_to_var(Server *s) {
906 if (s->storage != STORAGE_AUTO &&
907 s->storage != STORAGE_PERSISTENT)
910 if (!s->runtime_journal)
913 system_journal_open(s);
915 if (!s->system_journal)
918 log_debug("Flushing to /var...");
920 r = sd_id128_get_machine(&machine);
922 log_error("Failed to get machine id: %s", strerror(-r));
926 r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY);
928 log_error("Failed to read runtime journal: %s", strerror(-r));
932 SD_JOURNAL_FOREACH(j) {
936 assert(f && f->current_offset > 0);
938 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
940 log_error("Can't read entry: %s", strerror(-r));
944 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
946 log_debug("Allocation limit reached.");
948 journal_file_post_change(s->system_journal);
952 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
956 log_error("Can't write entry: %s", strerror(-r));
962 journal_file_post_change(s->system_journal);
964 journal_file_close(s->runtime_journal);
965 s->runtime_journal = NULL;
968 rm_rf("/run/log/journal", false, true, false);
973 static int process_event(Server *s, struct epoll_event *ev) {
977 if (ev->data.fd == s->signal_fd) {
978 struct signalfd_siginfo sfsi;
981 if (ev->events != EPOLLIN) {
982 log_error("Got invalid event from epoll.");
986 n = read(s->signal_fd, &sfsi, sizeof(sfsi));
987 if (n != sizeof(sfsi)) {
992 if (errno == EINTR || errno == EAGAIN)
998 log_info("Received SIG%s", signal_to_string(sfsi.ssi_signo));
1000 if (sfsi.ssi_signo == SIGUSR1) {
1001 touch("/run/systemd/journal/flushed");
1002 server_flush_to_var(s);
1006 if (sfsi.ssi_signo == SIGUSR2) {
1014 } else if (ev->data.fd == s->dev_kmsg_fd) {
1017 if (ev->events != EPOLLIN) {
1018 log_error("Got invalid event from epoll.");
1022 r = server_read_dev_kmsg(s);
1028 } else if (ev->data.fd == s->native_fd ||
1029 ev->data.fd == s->syslog_fd) {
1031 if (ev->events != EPOLLIN) {
1032 log_error("Got invalid event from epoll.");
1037 struct msghdr msghdr;
1039 struct ucred *ucred = NULL;
1040 struct timeval *tv = NULL;
1041 struct cmsghdr *cmsg;
1043 size_t label_len = 0;
1045 struct cmsghdr cmsghdr;
1047 /* We use NAME_MAX space for the
1048 * SELinux label here. The kernel
1049 * currently enforces no limit, but
1050 * according to suggestions from the
1051 * SELinux people this will change and
1052 * it will probably be identical to
1053 * NAME_MAX. For now we use that, but
1054 * this should be updated one day when
1055 * the final limit is known.*/
1056 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1057 CMSG_SPACE(sizeof(struct timeval)) +
1058 CMSG_SPACE(sizeof(int)) + /* fd */
1059 CMSG_SPACE(NAME_MAX)]; /* selinux label */
1066 if (ioctl(ev->data.fd, SIOCINQ, &v) < 0) {
1067 log_error("SIOCINQ failed: %m");
1071 if (s->buffer_size < (size_t) v) {
1075 l = MAX(LINE_MAX + (size_t) v, s->buffer_size * 2);
1076 b = realloc(s->buffer, l+1);
1079 log_error("Couldn't increase buffer.");
1088 iovec.iov_base = s->buffer;
1089 iovec.iov_len = s->buffer_size;
1093 msghdr.msg_iov = &iovec;
1094 msghdr.msg_iovlen = 1;
1095 msghdr.msg_control = &control;
1096 msghdr.msg_controllen = sizeof(control);
1098 n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
1101 if (errno == EINTR || errno == EAGAIN)
1104 log_error("recvmsg() failed: %m");
1108 for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
1110 if (cmsg->cmsg_level == SOL_SOCKET &&
1111 cmsg->cmsg_type == SCM_CREDENTIALS &&
1112 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
1113 ucred = (struct ucred*) CMSG_DATA(cmsg);
1114 else if (cmsg->cmsg_level == SOL_SOCKET &&
1115 cmsg->cmsg_type == SCM_SECURITY) {
1116 label = (char*) CMSG_DATA(cmsg);
1117 label_len = cmsg->cmsg_len - CMSG_LEN(0);
1118 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1119 cmsg->cmsg_type == SO_TIMESTAMP &&
1120 cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval)))
1121 tv = (struct timeval*) CMSG_DATA(cmsg);
1122 else if (cmsg->cmsg_level == SOL_SOCKET &&
1123 cmsg->cmsg_type == SCM_RIGHTS) {
1124 fds = (int*) CMSG_DATA(cmsg);
1125 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1129 if (ev->data.fd == s->syslog_fd) {
1132 if (n > 0 && n_fds == 0) {
1133 e = memchr(s->buffer, '\n', n);
1139 server_process_syslog_message(s, strstrip(s->buffer), ucred, tv, label, label_len);
1140 } else if (n_fds > 0)
1141 log_warning("Got file descriptors via syslog socket. Ignoring.");
1144 if (n > 0 && n_fds == 0)
1145 server_process_native_message(s, s->buffer, n, ucred, tv, label, label_len);
1146 else if (n == 0 && n_fds == 1)
1147 server_process_native_file(s, fds[0], ucred, tv, label, label_len);
1149 log_warning("Got too many file descriptors via native socket. Ignoring.");
1152 close_many(fds, n_fds);
1157 } else if (ev->data.fd == s->stdout_fd) {
1159 if (ev->events != EPOLLIN) {
1160 log_error("Got invalid event from epoll.");
1164 stdout_stream_new(s);
1168 StdoutStream *stream;
1170 if ((ev->events|EPOLLIN|EPOLLHUP) != (EPOLLIN|EPOLLHUP)) {
1171 log_error("Got invalid event from epoll.");
1175 /* If it is none of the well-known fds, it must be an
1176 * stdout stream fd. Note that this is a bit ugly here
1177 * (since we rely that none of the well-known fds
1178 * could be interpreted as pointer), but nonetheless
1179 * safe, since the well-known fds would never get an
1180 * fd > 4096, i.e. beyond the first memory page */
1182 stream = ev->data.ptr;
1184 if (stdout_stream_process(stream) <= 0)
1185 stdout_stream_free(stream);
1190 log_error("Unknown event.");
1194 static int open_signalfd(Server *s) {
1196 struct epoll_event ev;
1200 assert_se(sigemptyset(&mask) == 0);
1201 sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, SIGUSR2, -1);
1202 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1204 s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1205 if (s->signal_fd < 0) {
1206 log_error("signalfd(): %m");
1211 ev.events = EPOLLIN;
1212 ev.data.fd = s->signal_fd;
1214 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->signal_fd, &ev) < 0) {
1215 log_error("epoll_ctl(): %m");
1222 static int server_parse_proc_cmdline(Server *s) {
1223 char *line, *w, *state;
1227 if (detect_container(NULL) > 0)
1230 r = read_one_line_file("/proc/cmdline", &line);
1232 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
1236 FOREACH_WORD_QUOTED(w, l, line, state) {
1239 word = strndup(w, l);
1245 if (startswith(word, "systemd.journald.forward_to_syslog=")) {
1246 r = parse_boolean(word + 35);
1248 log_warning("Failed to parse forward to syslog switch %s. Ignoring.", word + 35);
1250 s->forward_to_syslog = r;
1251 } else if (startswith(word, "systemd.journald.forward_to_kmsg=")) {
1252 r = parse_boolean(word + 33);
1254 log_warning("Failed to parse forward to kmsg switch %s. Ignoring.", word + 33);
1256 s->forward_to_kmsg = r;
1257 } else if (startswith(word, "systemd.journald.forward_to_console=")) {
1258 r = parse_boolean(word + 36);
1260 log_warning("Failed to parse forward to console switch %s. Ignoring.", word + 36);
1262 s->forward_to_console = r;
1263 } else if (startswith(word, "systemd.journald"))
1264 log_warning("Invalid systemd.journald parameter. Ignoring.");
1276 static int server_parse_config_file(Server *s) {
1283 fn = "/etc/systemd/journald.conf";
1284 f = fopen(fn, "re");
1286 if (errno == ENOENT)
1289 log_warning("Failed to open configuration file %s: %m", fn);
1293 r = config_parse(fn, f, "Journal\0", config_item_perf_lookup, (void*) journald_gperf_lookup, false, s);
1295 log_warning("Failed to parse configuration file: %s", strerror(-r));
1302 static int server_init(Server *s) {
1308 s->syslog_fd = s->native_fd = s->stdout_fd = s->signal_fd = s->epoll_fd = s->dev_kmsg_fd = -1;
1312 s->rate_limit_interval = DEFAULT_RATE_LIMIT_INTERVAL;
1313 s->rate_limit_burst = DEFAULT_RATE_LIMIT_BURST;
1315 s->forward_to_syslog = true;
1317 s->max_level_store = LOG_DEBUG;
1318 s->max_level_syslog = LOG_DEBUG;
1319 s->max_level_kmsg = LOG_NOTICE;
1320 s->max_level_console = LOG_INFO;
1322 memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics));
1323 memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics));
1325 server_parse_config_file(s);
1326 server_parse_proc_cmdline(s);
1328 mkdir_p("/run/systemd/journal", 0755);
1330 s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
1331 if (!s->user_journals)
1334 s->mmap = mmap_cache_new();
1338 s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
1339 if (s->epoll_fd < 0) {
1340 log_error("Failed to create epoll object: %m");
1344 n = sd_listen_fds(true);
1346 log_error("Failed to read listening file descriptors from environment: %s", strerror(-n));
1350 for (fd = SD_LISTEN_FDS_START; fd < SD_LISTEN_FDS_START + n; fd++) {
1352 if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/run/systemd/journal/socket", 0) > 0) {
1354 if (s->native_fd >= 0) {
1355 log_error("Too many native sockets passed.");
1361 } else if (sd_is_socket_unix(fd, SOCK_STREAM, 1, "/run/systemd/journal/stdout", 0) > 0) {
1363 if (s->stdout_fd >= 0) {
1364 log_error("Too many stdout sockets passed.");
1370 } else if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/dev/log", 0) > 0) {
1372 if (s->syslog_fd >= 0) {
1373 log_error("Too many /dev/log sockets passed.");
1380 log_error("Unknown socket passed.");
1385 r = server_open_syslog_socket(s);
1389 r = server_open_native_socket(s);
1393 r = server_open_stdout_socket(s);
1397 r = server_open_dev_kmsg(s);
1401 r = server_open_kernel_seqnum(s);
1405 r = open_signalfd(s);
1409 s->udev = udev_new();
1413 s->rate_limit = journal_rate_limit_new(s->rate_limit_interval, s->rate_limit_burst);
1417 r = system_journal_open(s);
1424 static void server_maybe_append_tags(Server *s) {
1430 n = now(CLOCK_REALTIME);
1432 if (s->system_journal)
1433 journal_file_maybe_append_tag(s->system_journal, n);
1435 HASHMAP_FOREACH(f, s->user_journals, i)
1436 journal_file_maybe_append_tag(f, n);
1440 static void server_done(Server *s) {
1444 while (s->stdout_streams)
1445 stdout_stream_free(s->stdout_streams);
1447 if (s->system_journal)
1448 journal_file_close(s->system_journal);
1450 if (s->runtime_journal)
1451 journal_file_close(s->runtime_journal);
1453 while ((f = hashmap_steal_first(s->user_journals)))
1454 journal_file_close(f);
1456 hashmap_free(s->user_journals);
1458 if (s->epoll_fd >= 0)
1459 close_nointr_nofail(s->epoll_fd);
1461 if (s->signal_fd >= 0)
1462 close_nointr_nofail(s->signal_fd);
1464 if (s->syslog_fd >= 0)
1465 close_nointr_nofail(s->syslog_fd);
1467 if (s->native_fd >= 0)
1468 close_nointr_nofail(s->native_fd);
1470 if (s->stdout_fd >= 0)
1471 close_nointr_nofail(s->stdout_fd);
1473 if (s->dev_kmsg_fd >= 0)
1474 close_nointr_nofail(s->dev_kmsg_fd);
1477 journal_rate_limit_free(s->rate_limit);
1479 if (s->kernel_seqnum)
1480 munmap(s->kernel_seqnum, sizeof(uint64_t));
1486 mmap_cache_unref(s->mmap);
1489 udev_unref(s->udev);
1492 int main(int argc, char *argv[]) {
1496 /* if (getppid() != 1) { */
1497 /* log_error("This program should be invoked by init only."); */
1498 /* return EXIT_FAILURE; */
1502 log_error("This program does not take arguments.");
1503 return EXIT_FAILURE;
1506 log_set_target(LOG_TARGET_SAFE);
1507 log_set_facility(LOG_SYSLOG);
1508 log_parse_environment();
1513 r = server_init(&server);
1517 server_vacuum(&server);
1518 server_flush_to_var(&server);
1519 server_flush_dev_kmsg(&server);
1521 log_debug("systemd-journald running as pid %lu", (unsigned long) getpid());
1522 server_driver_message(&server, SD_MESSAGE_JOURNAL_START, "Journal started");
1526 "STATUS=Processing requests...");
1529 struct epoll_event event;
1535 if (server.system_journal &&
1536 journal_file_next_evolve_usec(server.system_journal, &u)) {
1539 n = now(CLOCK_REALTIME);
1544 t = (int) ((u - n + USEC_PER_MSEC - 1) / USEC_PER_MSEC);
1549 r = epoll_wait(server.epoll_fd, &event, 1, t);
1555 log_error("epoll_wait() failed: %m");
1561 r = process_event(&server, &event);
1568 server_maybe_append_tags(&server);
1569 server_maybe_warn_forward_syslog_missed(&server);
1572 log_debug("systemd-journald stopped as pid %lu", (unsigned long) getpid());
1573 server_driver_message(&server, SD_MESSAGE_JOURNAL_STOP, "Journal stopped");
1577 "STATUS=Shutting down...");
1579 server_done(&server);
1581 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;