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/signalfd.h>
23 #include <sys/ioctl.h>
24 #include <linux/sockios.h>
25 #include <sys/statvfs.h>
27 #include <sys/timerfd.h>
30 #include <systemd/sd-journal.h>
31 #include <systemd/sd-messages.h>
32 #include <systemd/sd-daemon.h>
37 #include "journal-file.h"
38 #include "socket-util.h"
39 #include "cgroup-util.h"
43 #include "conf-parser.h"
44 #include "journal-internal.h"
45 #include "journal-vacuum.h"
46 #include "journal-authenticate.h"
47 #include "journald-server.h"
48 #include "journald-rate-limit.h"
49 #include "journald-kmsg.h"
50 #include "journald-syslog.h"
51 #include "journald-stream.h"
52 #include "journald-console.h"
53 #include "journald-native.h"
57 #include <acl/libacl.h>
62 #include <selinux/selinux.h>
65 #define USER_JOURNALS_MAX 1024
67 #define DEFAULT_SYNC_INTERVAL_USEC (5*USEC_PER_MINUTE)
68 #define DEFAULT_RATE_LIMIT_INTERVAL (30*USEC_PER_SEC)
69 #define DEFAULT_RATE_LIMIT_BURST 1000
71 #define RECHECK_AVAILABLE_SPACE_USEC (30*USEC_PER_SEC)
73 static const char* const storage_table[] = {
74 [STORAGE_AUTO] = "auto",
75 [STORAGE_VOLATILE] = "volatile",
76 [STORAGE_PERSISTENT] = "persistent",
77 [STORAGE_NONE] = "none"
80 DEFINE_STRING_TABLE_LOOKUP(storage, Storage);
81 DEFINE_CONFIG_PARSE_ENUM(config_parse_storage, storage, Storage, "Failed to parse storage setting");
83 static const char* const split_mode_table[] = {
84 [SPLIT_NONE] = "none",
86 [SPLIT_LOGIN] = "login"
89 DEFINE_STRING_TABLE_LOOKUP(split_mode, SplitMode);
90 DEFINE_CONFIG_PARSE_ENUM(config_parse_split_mode, split_mode, SplitMode, "Failed to parse split mode setting");
92 static uint64_t available_space(Server *s, bool verbose) {
94 _cleanup_free_ char *p = NULL;
97 uint64_t sum = 0, ss_avail = 0, avail = 0;
99 _cleanup_closedir_ DIR *d = NULL;
104 ts = now(CLOCK_MONOTONIC);
106 if (s->cached_available_space_timestamp + RECHECK_AVAILABLE_SPACE_USEC > ts
108 return s->cached_available_space;
110 r = sd_id128_get_machine(&machine);
114 if (s->system_journal) {
115 f = "/var/log/journal/";
116 m = &s->system_metrics;
118 f = "/run/log/journal/";
119 m = &s->runtime_metrics;
124 p = strappend(f, sd_id128_to_string(machine, ids));
132 if (fstatvfs(dirfd(d), &ss) < 0)
138 union dirent_storage buf;
140 r = readdir_r(d, &buf.de, &de);
147 if (!endswith(de->d_name, ".journal") &&
148 !endswith(de->d_name, ".journal~"))
151 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
154 if (!S_ISREG(st.st_mode))
157 sum += (uint64_t) st.st_blocks * 512UL;
160 ss_avail = ss.f_bsize * ss.f_bavail;
161 avail = ss_avail > m->keep_free ? ss_avail - m->keep_free : 0;
163 s->cached_available_space = MIN(m->max_use, avail) > sum ? MIN(m->max_use, avail) - sum : 0;
164 s->cached_available_space_timestamp = ts;
167 char fb1[FORMAT_BYTES_MAX], fb2[FORMAT_BYTES_MAX], fb3[FORMAT_BYTES_MAX],
168 fb4[FORMAT_BYTES_MAX], fb5[FORMAT_BYTES_MAX];
170 server_driver_message(s, SD_MESSAGE_JOURNAL_USAGE,
171 "%s journal is using %s (max %s, leaving %s of free %s, current limit %s).",
172 s->system_journal ? "Permanent" : "Runtime",
173 format_bytes(fb1, sizeof(fb1), sum),
174 format_bytes(fb2, sizeof(fb2), m->max_use),
175 format_bytes(fb3, sizeof(fb3), m->keep_free),
176 format_bytes(fb4, sizeof(fb4), ss_avail),
177 format_bytes(fb5, sizeof(fb5), MIN(m->max_use, avail)));
180 return s->cached_available_space;
183 static void server_read_file_gid(Server *s) {
184 const char *g = "systemd-journal";
189 if (s->file_gid_valid)
192 r = get_group_creds(&g, &s->file_gid);
194 log_warning("Failed to resolve '%s' group: %s", g, strerror(-r));
196 /* if we couldn't read the gid, then it will be 0, but that's
197 * fine and we shouldn't try to resolve the group again, so
198 * let's just pretend it worked right-away. */
199 s->file_gid_valid = true;
202 void server_fix_perms(Server *s, JournalFile *f, uid_t uid) {
207 acl_permset_t permset;
212 server_read_file_gid(s);
214 r = fchmod_and_fchown(f->fd, 0640, 0, s->file_gid);
216 log_warning("Failed to fix access mode/rights on %s, ignoring: %s", f->path, strerror(-r));
222 acl = acl_get_fd(f->fd);
224 log_warning("Failed to read ACL on %s, ignoring: %m", f->path);
228 r = acl_find_uid(acl, uid, &entry);
231 if (acl_create_entry(&acl, &entry) < 0 ||
232 acl_set_tag_type(entry, ACL_USER) < 0 ||
233 acl_set_qualifier(entry, &uid) < 0) {
234 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
239 /* We do not recalculate the mask unconditionally here,
240 * so that the fchmod() mask above stays intact. */
241 if (acl_get_permset(entry, &permset) < 0 ||
242 acl_add_perm(permset, ACL_READ) < 0 ||
243 calc_acl_mask_if_needed(&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) {
257 _cleanup_free_ char *p = NULL;
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, NULL, &f);
296 return s->system_journal;
298 server_fix_perms(s, f, uid);
300 r = hashmap_put(s->user_journals, UINT32_TO_PTR(uid), f);
302 journal_file_close(f);
303 return s->system_journal;
309 void server_rotate(Server *s) {
315 log_debug("Rotating...");
317 if (s->runtime_journal) {
318 r = journal_file_rotate(&s->runtime_journal, s->compress, false);
320 if (s->runtime_journal)
321 log_error("Failed to rotate %s: %s", s->runtime_journal->path, strerror(-r));
323 log_error("Failed to create new runtime journal: %s", strerror(-r));
325 server_fix_perms(s, s->runtime_journal, 0);
328 if (s->system_journal) {
329 r = journal_file_rotate(&s->system_journal, s->compress, s->seal);
331 if (s->system_journal)
332 log_error("Failed to rotate %s: %s", s->system_journal->path, strerror(-r));
334 log_error("Failed to create new system journal: %s", strerror(-r));
337 server_fix_perms(s, s->system_journal, 0);
340 HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
341 r = journal_file_rotate(&f, s->compress, s->seal);
344 log_error("Failed to rotate %s: %s", f->path, strerror(-r));
346 log_error("Failed to create user journal: %s", strerror(-r));
348 hashmap_replace(s->user_journals, k, f);
349 server_fix_perms(s, f, PTR_TO_UINT32(k));
354 void server_sync(Server *s) {
355 static const struct itimerspec sync_timer_disable = {};
361 if (s->system_journal) {
362 r = journal_file_set_offline(s->system_journal);
364 log_error("Failed to sync system journal: %s", strerror(-r));
367 HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
368 r = journal_file_set_offline(f);
370 log_error("Failed to sync user journal: %s", strerror(-r));
373 r = timerfd_settime(s->sync_timer_fd, 0, &sync_timer_disable, NULL);
375 log_error("Failed to disable max timer: %m");
377 s->sync_scheduled = false;
380 void server_vacuum(Server *s) {
385 log_debug("Vacuuming...");
387 s->oldest_file_usec = 0;
389 r = sd_id128_get_machine(&machine);
391 log_error("Failed to get machine ID: %s", strerror(-r));
395 sd_id128_to_string(machine, ids);
397 if (s->system_journal) {
398 char *p = strappenda("/var/log/journal/", ids);
400 r = journal_directory_vacuum(p, s->system_metrics.max_use, s->system_metrics.keep_free, s->max_retention_usec, &s->oldest_file_usec);
401 if (r < 0 && r != -ENOENT)
402 log_error("Failed to vacuum %s: %s", p, strerror(-r));
405 if (s->runtime_journal) {
406 char *p = strappenda("/run/log/journal/", ids);
408 r = journal_directory_vacuum(p, s->runtime_metrics.max_use, s->runtime_metrics.keep_free, s->max_retention_usec, &s->oldest_file_usec);
409 if (r < 0 && r != -ENOENT)
410 log_error("Failed to vacuum %s: %s", p, strerror(-r));
413 s->cached_available_space_timestamp = 0;
416 bool shall_try_append_again(JournalFile *f, int r) {
418 /* -E2BIG Hit configured limit
420 -EDQUOT Quota limit hit
422 -EHOSTDOWN Other machine
423 -EBUSY Unclean shutdown
424 -EPROTONOSUPPORT Unsupported feature
427 -ESHUTDOWN Already archived */
429 if (r == -E2BIG || r == -EFBIG || r == -EDQUOT || r == -ENOSPC)
430 log_debug("%s: Allocation limit reached, rotating.", f->path);
431 else if (r == -EHOSTDOWN)
432 log_info("%s: Journal file from other machine, rotating.", f->path);
433 else if (r == -EBUSY)
434 log_info("%s: Unclean shutdown, rotating.", f->path);
435 else if (r == -EPROTONOSUPPORT)
436 log_info("%s: Unsupported feature, rotating.", f->path);
437 else if (r == -EBADMSG || r == -ENODATA || r == ESHUTDOWN)
438 log_warning("%s: Journal file corrupted, rotating.", f->path);
445 static void write_to_journal(Server *s, uid_t uid, struct iovec *iovec, unsigned n, int priority) {
447 bool vacuumed = false;
454 f = find_journal(s, uid);
458 if (journal_file_rotate_suggested(f, s->max_file_usec)) {
459 log_debug("%s: Journal header limits reached or header out-of-date, rotating.", f->path);
464 f = find_journal(s, uid);
469 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
471 server_schedule_sync(s, priority);
475 if (vacuumed || !shall_try_append_again(f, r)) {
478 for (i = 0; i < n; i++)
479 size += iovec[i].iov_len;
481 log_error("Failed to write entry (%d items, %zu bytes), ignoring: %s", n, size, strerror(-r));
488 f = find_journal(s, uid);
492 log_debug("Retrying write.");
493 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
497 for (i = 0; i < n; i++)
498 size += iovec[i].iov_len;
500 log_error("Failed to write entry (%d items, %zu bytes) despite vacuuming, ignoring: %s", n, size, strerror(-r));
502 server_schedule_sync(s, priority);
505 static void dispatch_message_real(
507 struct iovec *iovec, unsigned n, unsigned m,
510 const char *label, size_t label_len,
515 char pid[sizeof("_PID=") + DECIMAL_STR_MAX(pid_t)],
516 uid[sizeof("_UID=") + DECIMAL_STR_MAX(uid_t)],
517 gid[sizeof("_GID=") + DECIMAL_STR_MAX(gid_t)],
518 owner_uid[sizeof("_SYSTEMD_OWNER_UID=") + DECIMAL_STR_MAX(uid_t)],
519 source_time[sizeof("_SOURCE_REALTIME_TIMESTAMP=") + DECIMAL_STR_MAX(usec_t)],
520 boot_id[sizeof("_BOOT_ID=") + 32] = "_BOOT_ID=",
521 machine_id[sizeof("_MACHINE_ID=") + 32] = "_MACHINE_ID=",
522 o_uid[sizeof("OBJECT_UID=") + DECIMAL_STR_MAX(uid_t)],
523 o_gid[sizeof("OBJECT_GID=") + DECIMAL_STR_MAX(gid_t)],
524 o_owner_uid[sizeof("OBJECT_SYSTEMD_OWNER_UID=") + DECIMAL_STR_MAX(uid_t)];
531 uid_t realuid = 0, owner = 0, journal_uid;
532 bool owner_valid = false;
534 char audit_session[sizeof("_AUDIT_SESSION=") + DECIMAL_STR_MAX(uint32_t)],
535 audit_loginuid[sizeof("_AUDIT_LOGINUID=") + DECIMAL_STR_MAX(uid_t)],
536 o_audit_session[sizeof("OBJECT_AUDIT_SESSION=") + DECIMAL_STR_MAX(uint32_t)],
537 o_audit_loginuid[sizeof("OBJECT_AUDIT_LOGINUID=") + DECIMAL_STR_MAX(uid_t)];
546 assert(n + N_IOVEC_META_FIELDS + (object_pid ? N_IOVEC_OBJECT_FIELDS : 0) <= m);
549 realuid = ucred->uid;
551 sprintf(pid, "_PID=%lu", (unsigned long) ucred->pid);
552 IOVEC_SET_STRING(iovec[n++], pid);
554 sprintf(uid, "_UID=%lu", (unsigned long) ucred->uid);
555 IOVEC_SET_STRING(iovec[n++], uid);
557 sprintf(gid, "_GID=%lu", (unsigned long) ucred->gid);
558 IOVEC_SET_STRING(iovec[n++], gid);
560 r = get_process_comm(ucred->pid, &t);
562 x = strappenda("_COMM=", t);
564 IOVEC_SET_STRING(iovec[n++], x);
567 r = get_process_exe(ucred->pid, &t);
569 x = strappenda("_EXE=", t);
571 IOVEC_SET_STRING(iovec[n++], x);
574 r = get_process_cmdline(ucred->pid, 0, false, &t);
576 x = strappenda("_CMDLINE=", t);
578 IOVEC_SET_STRING(iovec[n++], x);
581 r = get_process_capeff(ucred->pid, &t);
583 x = strappenda("_CAP_EFFECTIVE=", t);
585 IOVEC_SET_STRING(iovec[n++], x);
589 r = audit_session_from_pid(ucred->pid, &audit);
591 sprintf(audit_session, "_AUDIT_SESSION=%lu", (unsigned long) audit);
592 IOVEC_SET_STRING(iovec[n++], audit_session);
595 r = audit_loginuid_from_pid(ucred->pid, &loginuid);
597 sprintf(audit_loginuid, "_AUDIT_LOGINUID=%lu", (unsigned long) loginuid);
598 IOVEC_SET_STRING(iovec[n++], audit_loginuid);
602 r = cg_pid_get_path_shifted(ucred->pid, NULL, &c);
604 char *session = NULL;
606 x = strappenda("_SYSTEMD_CGROUP=", c);
607 IOVEC_SET_STRING(iovec[n++], x);
609 r = cg_path_get_session(c, &t);
611 session = strappenda("_SYSTEMD_SESSION=", t);
613 IOVEC_SET_STRING(iovec[n++], session);
616 if (cg_path_get_owner_uid(c, &owner) >= 0) {
619 sprintf(owner_uid, "_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner);
620 IOVEC_SET_STRING(iovec[n++], owner_uid);
623 if (cg_path_get_unit(c, &t) >= 0) {
624 x = strappenda("_SYSTEMD_UNIT=", t);
626 IOVEC_SET_STRING(iovec[n++], x);
627 } else if (unit_id && !session) {
628 x = strappenda("_SYSTEMD_UNIT=", unit_id);
629 IOVEC_SET_STRING(iovec[n++], x);
632 if (cg_path_get_user_unit(c, &t) >= 0) {
633 x = strappenda("_SYSTEMD_USER_UNIT=", t);
635 IOVEC_SET_STRING(iovec[n++], x);
636 } else if (unit_id && session) {
637 x = strappenda("_SYSTEMD_USER_UNIT=", unit_id);
638 IOVEC_SET_STRING(iovec[n++], x);
646 x = alloca(sizeof("_SELINUX_CONTEXT=") + label_len);
648 *((char*) mempcpy(stpcpy(x, "_SELINUX_CONTEXT="), label, label_len)) = 0;
649 IOVEC_SET_STRING(iovec[n++], x);
651 security_context_t con;
653 if (getpidcon(ucred->pid, &con) >= 0) {
654 x = strappenda("_SELINUX_CONTEXT=", con);
657 IOVEC_SET_STRING(iovec[n++], x);
665 r = get_process_uid(object_pid, &object_uid);
667 sprintf(o_uid, "OBJECT_UID=%lu", (unsigned long) object_uid);
668 IOVEC_SET_STRING(iovec[n++], o_uid);
671 r = get_process_gid(object_pid, &object_gid);
673 sprintf(o_gid, "OBJECT_GID=%lu", (unsigned long) object_gid);
674 IOVEC_SET_STRING(iovec[n++], o_gid);
677 r = get_process_comm(object_pid, &t);
679 x = strappenda("OBJECT_COMM=", t);
681 IOVEC_SET_STRING(iovec[n++], x);
684 r = get_process_exe(object_pid, &t);
686 x = strappenda("OBJECT_EXE=", t);
688 IOVEC_SET_STRING(iovec[n++], x);
691 r = get_process_cmdline(object_pid, 0, false, &t);
693 x = strappenda("OBJECT_CMDLINE=", t);
695 IOVEC_SET_STRING(iovec[n++], x);
699 r = audit_session_from_pid(object_pid, &audit);
701 sprintf(o_audit_session, "OBJECT_AUDIT_SESSION=%lu", (unsigned long) audit);
702 IOVEC_SET_STRING(iovec[n++], o_audit_session);
705 r = audit_loginuid_from_pid(object_pid, &loginuid);
707 sprintf(o_audit_loginuid, "OBJECT_AUDIT_LOGINUID=%lu", (unsigned long) loginuid);
708 IOVEC_SET_STRING(iovec[n++], o_audit_loginuid);
712 r = cg_pid_get_path_shifted(object_pid, NULL, &c);
714 x = strappenda("OBJECT_SYSTEMD_CGROUP=", c);
715 IOVEC_SET_STRING(iovec[n++], x);
717 r = cg_path_get_session(c, &t);
719 x = strappenda("OBJECT_SYSTEMD_SESSION=", t);
721 IOVEC_SET_STRING(iovec[n++], x);
724 if (cg_path_get_owner_uid(c, &owner) >= 0) {
725 sprintf(o_owner_uid, "OBJECT_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner);
726 IOVEC_SET_STRING(iovec[n++], o_owner_uid);
729 if (cg_path_get_unit(c, &t) >= 0) {
730 x = strappenda("OBJECT_SYSTEMD_UNIT=", t);
732 IOVEC_SET_STRING(iovec[n++], x);
735 if (cg_path_get_user_unit(c, &t) >= 0) {
736 x = strappenda("OBJECT_SYSTEMD_USER_UNIT=", t);
738 IOVEC_SET_STRING(iovec[n++], x);
747 sprintf(source_time, "_SOURCE_REALTIME_TIMESTAMP=%llu", (unsigned long long) timeval_load(tv));
748 IOVEC_SET_STRING(iovec[n++], source_time);
751 /* Note that strictly speaking storing the boot id here is
752 * redundant since the entry includes this in-line
753 * anyway. However, we need this indexed, too. */
754 r = sd_id128_get_boot(&id);
756 sd_id128_to_string(id, boot_id + strlen("_BOOT_ID="));
757 IOVEC_SET_STRING(iovec[n++], boot_id);
760 r = sd_id128_get_machine(&id);
762 sd_id128_to_string(id, machine_id + strlen("_MACHINE_ID="));
763 IOVEC_SET_STRING(iovec[n++], machine_id);
766 t = gethostname_malloc();
768 x = strappenda("_HOSTNAME=", t);
770 IOVEC_SET_STRING(iovec[n++], x);
775 if (s->split_mode == SPLIT_UID && realuid > 0)
776 /* Split up strictly by any UID */
777 journal_uid = realuid;
778 else if (s->split_mode == SPLIT_LOGIN && realuid > 0 && owner_valid && owner > 0)
779 /* Split up by login UIDs, this avoids creation of
780 * individual journals for system UIDs. We do this
781 * only if the realuid is not root, in order not to
782 * accidentally leak privileged information to the
783 * user that is logged by a privileged process that is
784 * part of an unprivileged session.*/
789 write_to_journal(s, journal_uid, iovec, n, priority);
792 void server_driver_message(Server *s, sd_id128_t message_id, const char *format, ...) {
793 char mid[11 + 32 + 1];
794 char buffer[16 + LINE_MAX + 1];
795 struct iovec iovec[N_IOVEC_META_FIELDS + 4];
798 struct ucred ucred = {};
803 IOVEC_SET_STRING(iovec[n++], "PRIORITY=6");
804 IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=driver");
806 memcpy(buffer, "MESSAGE=", 8);
807 va_start(ap, format);
808 vsnprintf(buffer + 8, sizeof(buffer) - 8, format, ap);
810 char_array_0(buffer);
811 IOVEC_SET_STRING(iovec[n++], buffer);
813 if (!sd_id128_equal(message_id, SD_ID128_NULL)) {
814 snprintf(mid, sizeof(mid), MESSAGE_ID(message_id));
816 IOVEC_SET_STRING(iovec[n++], mid);
819 ucred.pid = getpid();
820 ucred.uid = getuid();
821 ucred.gid = getgid();
823 dispatch_message_real(s, iovec, n, ELEMENTSOF(iovec), &ucred, NULL, NULL, 0, NULL, LOG_INFO, 0);
826 void server_dispatch_message(
828 struct iovec *iovec, unsigned n, unsigned m,
831 const char *label, size_t label_len,
837 _cleanup_free_ char *path = NULL;
841 assert(iovec || n == 0);
846 if (LOG_PRI(priority) > s->max_level_store)
849 /* Stop early in case the information will not be stored
851 if (s->storage == STORAGE_NONE)
857 r = cg_pid_get_path_shifted(ucred->pid, NULL, &path);
861 /* example: /user/lennart/3/foobar
862 * /system/dbus.service/foobar
864 * So let's cut of everything past the third /, since that is
865 * where user directories start */
867 c = strchr(path, '/');
869 c = strchr(c+1, '/');
871 c = strchr(c+1, '/');
877 rl = journal_rate_limit_test(s->rate_limit, path,
878 priority & LOG_PRIMASK, available_space(s, false));
883 /* Write a suppression message if we suppressed something */
885 server_driver_message(s, SD_MESSAGE_JOURNAL_DROPPED,
886 "Suppressed %u messages from %s", rl - 1, path);
889 dispatch_message_real(s, iovec, n, m, ucred, tv, label, label_len, unit_id, priority, object_pid);
893 static int system_journal_open(Server *s) {
899 r = sd_id128_get_machine(&machine);
903 sd_id128_to_string(machine, ids);
905 if (!s->system_journal &&
906 (s->storage == STORAGE_PERSISTENT || s->storage == STORAGE_AUTO) &&
907 access("/run/systemd/journal/flushed", F_OK) >= 0) {
909 /* If in auto mode: first try to create the machine
910 * path, but not the prefix.
912 * If in persistent mode: create /var/log/journal and
913 * the machine path */
915 if (s->storage == STORAGE_PERSISTENT)
916 (void) mkdir("/var/log/journal/", 0755);
918 fn = strappenda("/var/log/journal/", ids);
919 (void) mkdir(fn, 0755);
921 fn = strappenda(fn, "/system.journal");
922 r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, NULL, &s->system_journal);
925 server_fix_perms(s, s->system_journal, 0);
927 if (r != -ENOENT && r != -EROFS)
928 log_warning("Failed to open system journal: %s", strerror(-r));
934 if (!s->runtime_journal &&
935 (s->storage != STORAGE_NONE)) {
937 fn = strjoin("/run/log/journal/", ids, "/system.journal", NULL);
941 if (s->system_journal) {
943 /* Try to open the runtime journal, but only
944 * if it already exists, so that we can flush
945 * it into the system journal */
947 r = journal_file_open(fn, O_RDWR, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal);
952 log_warning("Failed to open runtime journal: %s", strerror(-r));
959 /* OK, we really need the runtime journal, so create
960 * it if necessary. */
962 (void) mkdir_parents(fn, 0755);
963 r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal);
967 log_error("Failed to open runtime journal: %s", strerror(-r));
972 if (s->runtime_journal)
973 server_fix_perms(s, s->runtime_journal, 0);
976 available_space(s, true);
981 int server_flush_to_var(Server *s) {
984 sd_journal *j = NULL;
988 if (s->storage != STORAGE_AUTO &&
989 s->storage != STORAGE_PERSISTENT)
992 if (!s->runtime_journal)
995 system_journal_open(s);
997 if (!s->system_journal)
1000 log_debug("Flushing to /var...");
1002 r = sd_id128_get_machine(&machine);
1004 log_error("Failed to get machine id: %s", strerror(-r));
1008 r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY);
1010 log_error("Failed to read runtime journal: %s", strerror(-r));
1014 sd_journal_set_data_threshold(j, 0);
1016 SD_JOURNAL_FOREACH(j) {
1020 f = j->current_file;
1021 assert(f && f->current_offset > 0);
1023 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1025 log_error("Can't read entry: %s", strerror(-r));
1029 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1033 if (!shall_try_append_again(s->system_journal, r)) {
1034 log_error("Can't write entry: %s", strerror(-r));
1041 if (!s->system_journal) {
1042 log_notice("Didn't flush runtime journal since rotation of system journal wasn't successful.");
1047 log_debug("Retrying write.");
1048 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1050 log_error("Can't write entry: %s", strerror(-r));
1056 journal_file_post_change(s->system_journal);
1058 journal_file_close(s->runtime_journal);
1059 s->runtime_journal = NULL;
1062 rm_rf("/run/log/journal", false, true, false);
1064 sd_journal_close(j);
1069 int process_event(Server *s, struct epoll_event *ev) {
1073 if (ev->data.fd == s->signal_fd) {
1074 struct signalfd_siginfo sfsi;
1077 if (ev->events != EPOLLIN) {
1078 log_error("Got invalid event from epoll.");
1082 n = read(s->signal_fd, &sfsi, sizeof(sfsi));
1083 if (n != sizeof(sfsi)) {
1088 if (errno == EINTR || errno == EAGAIN)
1094 if (sfsi.ssi_signo == SIGUSR1) {
1095 log_info("Received request to flush runtime journal from PID %"PRIu32,
1097 touch("/run/systemd/journal/flushed");
1098 server_flush_to_var(s);
1103 if (sfsi.ssi_signo == SIGUSR2) {
1104 log_info("Received request to rotate journal from PID %"PRIu32,
1111 log_info("Received SIG%s", signal_to_string(sfsi.ssi_signo));
1115 } else if (ev->data.fd == s->sync_timer_fd) {
1119 log_debug("Got sync request from epoll.");
1121 r = read(ev->data.fd, (void *)&t, sizeof(t));
1128 } else if (ev->data.fd == s->dev_kmsg_fd) {
1131 if (ev->events != EPOLLIN) {
1132 log_error("Got invalid event from epoll.");
1136 r = server_read_dev_kmsg(s);
1142 } else if (ev->data.fd == s->native_fd ||
1143 ev->data.fd == s->syslog_fd) {
1145 if (ev->events != EPOLLIN) {
1146 log_error("Got invalid event from epoll.");
1151 struct msghdr msghdr;
1153 struct ucred *ucred = NULL;
1154 struct timeval *tv = NULL;
1155 struct cmsghdr *cmsg;
1157 size_t label_len = 0;
1159 struct cmsghdr cmsghdr;
1161 /* We use NAME_MAX space for the
1162 * SELinux label here. The kernel
1163 * currently enforces no limit, but
1164 * according to suggestions from the
1165 * SELinux people this will change and
1166 * it will probably be identical to
1167 * NAME_MAX. For now we use that, but
1168 * this should be updated one day when
1169 * the final limit is known.*/
1170 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1171 CMSG_SPACE(sizeof(struct timeval)) +
1172 CMSG_SPACE(sizeof(int)) + /* fd */
1173 CMSG_SPACE(NAME_MAX)]; /* selinux label */
1180 if (ioctl(ev->data.fd, SIOCINQ, &v) < 0) {
1181 log_error("SIOCINQ failed: %m");
1185 if (s->buffer_size < (size_t) v) {
1189 l = MAX(LINE_MAX + (size_t) v, s->buffer_size * 2);
1190 b = realloc(s->buffer, l+1);
1193 log_error("Couldn't increase buffer.");
1202 iovec.iov_base = s->buffer;
1203 iovec.iov_len = s->buffer_size;
1207 msghdr.msg_iov = &iovec;
1208 msghdr.msg_iovlen = 1;
1209 msghdr.msg_control = &control;
1210 msghdr.msg_controllen = sizeof(control);
1212 n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
1215 if (errno == EINTR || errno == EAGAIN)
1218 log_error("recvmsg() failed: %m");
1222 for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
1224 if (cmsg->cmsg_level == SOL_SOCKET &&
1225 cmsg->cmsg_type == SCM_CREDENTIALS &&
1226 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
1227 ucred = (struct ucred*) CMSG_DATA(cmsg);
1228 else if (cmsg->cmsg_level == SOL_SOCKET &&
1229 cmsg->cmsg_type == SCM_SECURITY) {
1230 label = (char*) CMSG_DATA(cmsg);
1231 label_len = cmsg->cmsg_len - CMSG_LEN(0);
1232 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1233 cmsg->cmsg_type == SO_TIMESTAMP &&
1234 cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval)))
1235 tv = (struct timeval*) CMSG_DATA(cmsg);
1236 else if (cmsg->cmsg_level == SOL_SOCKET &&
1237 cmsg->cmsg_type == SCM_RIGHTS) {
1238 fds = (int*) CMSG_DATA(cmsg);
1239 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1243 if (ev->data.fd == s->syslog_fd) {
1244 if (n > 0 && n_fds == 0) {
1246 server_process_syslog_message(s, strstrip(s->buffer), ucred, tv, label, label_len);
1247 } else if (n_fds > 0)
1248 log_warning("Got file descriptors via syslog socket. Ignoring.");
1251 if (n > 0 && n_fds == 0)
1252 server_process_native_message(s, s->buffer, n, ucred, tv, label, label_len);
1253 else if (n == 0 && n_fds == 1)
1254 server_process_native_file(s, fds[0], ucred, tv, label, label_len);
1256 log_warning("Got too many file descriptors via native socket. Ignoring.");
1259 close_many(fds, n_fds);
1264 } else if (ev->data.fd == s->stdout_fd) {
1266 if (ev->events != EPOLLIN) {
1267 log_error("Got invalid event from epoll.");
1271 stdout_stream_new(s);
1275 StdoutStream *stream;
1277 if ((ev->events|EPOLLIN|EPOLLHUP) != (EPOLLIN|EPOLLHUP)) {
1278 log_error("Got invalid event from epoll.");
1282 /* If it is none of the well-known fds, it must be an
1283 * stdout stream fd. Note that this is a bit ugly here
1284 * (since we rely that none of the well-known fds
1285 * could be interpreted as pointer), but nonetheless
1286 * safe, since the well-known fds would never get an
1287 * fd > 4096, i.e. beyond the first memory page */
1289 stream = ev->data.ptr;
1291 if (stdout_stream_process(stream) <= 0)
1292 stdout_stream_free(stream);
1297 log_error("Unknown event.");
1301 static int open_signalfd(Server *s) {
1303 struct epoll_event ev;
1307 assert_se(sigemptyset(&mask) == 0);
1308 sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, SIGUSR2, -1);
1309 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1311 s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1312 if (s->signal_fd < 0) {
1313 log_error("signalfd(): %m");
1318 ev.events = EPOLLIN;
1319 ev.data.fd = s->signal_fd;
1321 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->signal_fd, &ev) < 0) {
1322 log_error("epoll_ctl(): %m");
1329 static int server_parse_proc_cmdline(Server *s) {
1330 _cleanup_free_ char *line = NULL;
1335 if (detect_container(NULL) > 0)
1338 r = read_one_line_file("/proc/cmdline", &line);
1340 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
1344 FOREACH_WORD_QUOTED(w, l, line, state) {
1345 _cleanup_free_ char *word;
1347 word = strndup(w, l);
1351 if (startswith(word, "systemd.journald.forward_to_syslog=")) {
1352 r = parse_boolean(word + 35);
1354 log_warning("Failed to parse forward to syslog switch %s. Ignoring.", word + 35);
1356 s->forward_to_syslog = r;
1357 } else if (startswith(word, "systemd.journald.forward_to_kmsg=")) {
1358 r = parse_boolean(word + 33);
1360 log_warning("Failed to parse forward to kmsg switch %s. Ignoring.", word + 33);
1362 s->forward_to_kmsg = r;
1363 } else if (startswith(word, "systemd.journald.forward_to_console=")) {
1364 r = parse_boolean(word + 36);
1366 log_warning("Failed to parse forward to console switch %s. Ignoring.", word + 36);
1368 s->forward_to_console = r;
1369 } else if (startswith(word, "systemd.journald"))
1370 log_warning("Invalid systemd.journald parameter. Ignoring.");
1376 static int server_parse_config_file(Server *s) {
1377 static const char fn[] = "/etc/systemd/journald.conf";
1378 _cleanup_fclose_ FILE *f = NULL;
1383 f = fopen(fn, "re");
1385 if (errno == ENOENT)
1388 log_warning("Failed to open configuration file %s: %m", fn);
1392 r = config_parse(NULL, fn, f, "Journal\0", config_item_perf_lookup,
1393 (void*) journald_gperf_lookup, false, false, s);
1395 log_warning("Failed to parse configuration file: %s", strerror(-r));
1400 static int server_open_sync_timer(Server *s) {
1402 struct epoll_event ev;
1406 s->sync_timer_fd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC);
1407 if (s->sync_timer_fd < 0)
1411 ev.events = EPOLLIN;
1412 ev.data.fd = s->sync_timer_fd;
1414 r = epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->sync_timer_fd, &ev);
1416 log_error("Failed to add idle timer fd to epoll object: %m");
1423 int server_schedule_sync(Server *s, int priority) {
1428 if (priority <= LOG_CRIT) {
1429 /* Immediately sync to disk when this is of priority CRIT, ALERT, EMERG */
1434 if (s->sync_scheduled)
1437 if (s->sync_interval_usec) {
1438 struct itimerspec sync_timer_enable = {};
1440 timespec_store(&sync_timer_enable.it_value, s->sync_interval_usec);
1442 r = timerfd_settime(s->sync_timer_fd, 0, &sync_timer_enable, NULL);
1447 s->sync_scheduled = true;
1452 int server_init(Server *s) {
1458 s->sync_timer_fd = s->syslog_fd = s->native_fd = s->stdout_fd =
1459 s->signal_fd = s->epoll_fd = s->dev_kmsg_fd = -1;
1463 s->sync_interval_usec = DEFAULT_SYNC_INTERVAL_USEC;
1464 s->sync_scheduled = false;
1466 s->rate_limit_interval = DEFAULT_RATE_LIMIT_INTERVAL;
1467 s->rate_limit_burst = DEFAULT_RATE_LIMIT_BURST;
1469 s->forward_to_syslog = true;
1471 s->max_level_store = LOG_DEBUG;
1472 s->max_level_syslog = LOG_DEBUG;
1473 s->max_level_kmsg = LOG_NOTICE;
1474 s->max_level_console = LOG_INFO;
1476 memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics));
1477 memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics));
1479 server_parse_config_file(s);
1480 server_parse_proc_cmdline(s);
1481 if (!!s->rate_limit_interval ^ !!s->rate_limit_burst) {
1482 log_debug("Setting both rate limit interval and burst from %llu,%u to 0,0",
1483 (long long unsigned) s->rate_limit_interval,
1484 s->rate_limit_burst);
1485 s->rate_limit_interval = s->rate_limit_burst = 0;
1488 mkdir_p("/run/systemd/journal", 0755);
1490 s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
1491 if (!s->user_journals)
1494 s->mmap = mmap_cache_new();
1498 s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
1499 if (s->epoll_fd < 0) {
1500 log_error("Failed to create epoll object: %m");
1504 n = sd_listen_fds(true);
1506 log_error("Failed to read listening file descriptors from environment: %s", strerror(-n));
1510 for (fd = SD_LISTEN_FDS_START; fd < SD_LISTEN_FDS_START + n; fd++) {
1512 if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/run/systemd/journal/socket", 0) > 0) {
1514 if (s->native_fd >= 0) {
1515 log_error("Too many native sockets passed.");
1521 } else if (sd_is_socket_unix(fd, SOCK_STREAM, 1, "/run/systemd/journal/stdout", 0) > 0) {
1523 if (s->stdout_fd >= 0) {
1524 log_error("Too many stdout sockets passed.");
1530 } else if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/dev/log", 0) > 0) {
1532 if (s->syslog_fd >= 0) {
1533 log_error("Too many /dev/log sockets passed.");
1540 log_error("Unknown socket passed.");
1545 r = server_open_syslog_socket(s);
1549 r = server_open_native_socket(s);
1553 r = server_open_stdout_socket(s);
1557 r = server_open_dev_kmsg(s);
1561 r = server_open_kernel_seqnum(s);
1565 r = server_open_sync_timer(s);
1569 r = open_signalfd(s);
1573 s->udev = udev_new();
1577 s->rate_limit = journal_rate_limit_new(s->rate_limit_interval,
1578 s->rate_limit_burst);
1582 r = system_journal_open(s);
1589 void server_maybe_append_tags(Server *s) {
1595 n = now(CLOCK_REALTIME);
1597 if (s->system_journal)
1598 journal_file_maybe_append_tag(s->system_journal, n);
1600 HASHMAP_FOREACH(f, s->user_journals, i)
1601 journal_file_maybe_append_tag(f, n);
1605 void server_done(Server *s) {
1609 while (s->stdout_streams)
1610 stdout_stream_free(s->stdout_streams);
1612 if (s->system_journal)
1613 journal_file_close(s->system_journal);
1615 if (s->runtime_journal)
1616 journal_file_close(s->runtime_journal);
1618 while ((f = hashmap_steal_first(s->user_journals)))
1619 journal_file_close(f);
1621 hashmap_free(s->user_journals);
1623 if (s->epoll_fd >= 0)
1624 close_nointr_nofail(s->epoll_fd);
1626 if (s->signal_fd >= 0)
1627 close_nointr_nofail(s->signal_fd);
1629 if (s->syslog_fd >= 0)
1630 close_nointr_nofail(s->syslog_fd);
1632 if (s->native_fd >= 0)
1633 close_nointr_nofail(s->native_fd);
1635 if (s->stdout_fd >= 0)
1636 close_nointr_nofail(s->stdout_fd);
1638 if (s->dev_kmsg_fd >= 0)
1639 close_nointr_nofail(s->dev_kmsg_fd);
1641 if (s->sync_timer_fd >= 0)
1642 close_nointr_nofail(s->sync_timer_fd);
1645 journal_rate_limit_free(s->rate_limit);
1647 if (s->kernel_seqnum)
1648 munmap(s->kernel_seqnum, sizeof(uint64_t));
1654 mmap_cache_unref(s->mmap);
1657 udev_unref(s->udev);