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 void server_fix_perms(Server *s, JournalFile *f, uid_t uid) {
188 acl_permset_t permset;
193 r = fchmod(f->fd, 0640);
195 log_warning("Failed to fix access mode on %s, ignoring: %s", f->path, strerror(-r));
201 acl = acl_get_fd(f->fd);
203 log_warning("Failed to read ACL on %s, ignoring: %m", f->path);
207 r = acl_find_uid(acl, uid, &entry);
210 if (acl_create_entry(&acl, &entry) < 0 ||
211 acl_set_tag_type(entry, ACL_USER) < 0 ||
212 acl_set_qualifier(entry, &uid) < 0) {
213 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
218 /* We do not recalculate the mask unconditionally here,
219 * so that the fchmod() mask above stays intact. */
220 if (acl_get_permset(entry, &permset) < 0 ||
221 acl_add_perm(permset, ACL_READ) < 0 ||
222 calc_acl_mask_if_needed(&acl) < 0) {
223 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
227 if (acl_set_fd(f->fd, acl) < 0)
228 log_warning("Failed to set ACL on %s, ignoring: %m", f->path);
235 static JournalFile* find_journal(Server *s, uid_t uid) {
236 _cleanup_free_ char *p = NULL;
243 /* We split up user logs only on /var, not on /run. If the
244 * runtime file is open, we write to it exclusively, in order
245 * to guarantee proper order as soon as we flush /run to
246 * /var and close the runtime file. */
248 if (s->runtime_journal)
249 return s->runtime_journal;
252 return s->system_journal;
254 r = sd_id128_get_machine(&machine);
256 return s->system_journal;
258 f = hashmap_get(s->user_journals, UINT32_TO_PTR(uid));
262 if (asprintf(&p, "/var/log/journal/" SD_ID128_FORMAT_STR "/user-%lu.journal",
263 SD_ID128_FORMAT_VAL(machine), (unsigned long) uid) < 0)
264 return s->system_journal;
266 while (hashmap_size(s->user_journals) >= USER_JOURNALS_MAX) {
267 /* Too many open? Then let's close one */
268 f = hashmap_steal_first(s->user_journals);
270 journal_file_close(f);
273 r = journal_file_open_reliably(p, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, NULL, &f);
275 return s->system_journal;
277 server_fix_perms(s, f, uid);
279 r = hashmap_put(s->user_journals, UINT32_TO_PTR(uid), f);
281 journal_file_close(f);
282 return s->system_journal;
288 void server_rotate(Server *s) {
294 log_debug("Rotating...");
296 if (s->runtime_journal) {
297 r = journal_file_rotate(&s->runtime_journal, s->compress, false);
299 if (s->runtime_journal)
300 log_error("Failed to rotate %s: %s", s->runtime_journal->path, strerror(-r));
302 log_error("Failed to create new runtime journal: %s", strerror(-r));
304 server_fix_perms(s, s->runtime_journal, 0);
307 if (s->system_journal) {
308 r = journal_file_rotate(&s->system_journal, s->compress, s->seal);
310 if (s->system_journal)
311 log_error("Failed to rotate %s: %s", s->system_journal->path, strerror(-r));
313 log_error("Failed to create new system journal: %s", strerror(-r));
316 server_fix_perms(s, s->system_journal, 0);
319 HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
320 r = journal_file_rotate(&f, s->compress, s->seal);
323 log_error("Failed to rotate %s: %s", f->path, strerror(-r));
325 log_error("Failed to create user journal: %s", strerror(-r));
326 hashmap_remove(s->user_journals, k);
329 hashmap_replace(s->user_journals, k, f);
330 server_fix_perms(s, f, PTR_TO_UINT32(k));
335 void server_sync(Server *s) {
336 static const struct itimerspec sync_timer_disable = {};
342 if (s->system_journal) {
343 r = journal_file_set_offline(s->system_journal);
345 log_error("Failed to sync system journal: %s", strerror(-r));
348 HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
349 r = journal_file_set_offline(f);
351 log_error("Failed to sync user journal: %s", strerror(-r));
354 r = timerfd_settime(s->sync_timer_fd, 0, &sync_timer_disable, NULL);
356 log_error("Failed to disable max timer: %m");
358 s->sync_scheduled = false;
361 void server_vacuum(Server *s) {
366 log_debug("Vacuuming...");
368 s->oldest_file_usec = 0;
370 r = sd_id128_get_machine(&machine);
372 log_error("Failed to get machine ID: %s", strerror(-r));
376 sd_id128_to_string(machine, ids);
378 if (s->system_journal) {
379 char *p = strappenda("/var/log/journal/", ids);
381 r = journal_directory_vacuum(p, s->system_metrics.max_use, s->system_metrics.keep_free, s->max_retention_usec, &s->oldest_file_usec);
382 if (r < 0 && r != -ENOENT)
383 log_error("Failed to vacuum %s: %s", p, strerror(-r));
386 if (s->runtime_journal) {
387 char *p = strappenda("/run/log/journal/", ids);
389 r = journal_directory_vacuum(p, s->runtime_metrics.max_use, s->runtime_metrics.keep_free, s->max_retention_usec, &s->oldest_file_usec);
390 if (r < 0 && r != -ENOENT)
391 log_error("Failed to vacuum %s: %s", p, strerror(-r));
394 s->cached_available_space_timestamp = 0;
397 bool shall_try_append_again(JournalFile *f, int r) {
399 /* -E2BIG Hit configured limit
401 -EDQUOT Quota limit hit
403 -EHOSTDOWN Other machine
404 -EBUSY Unclean shutdown
405 -EPROTONOSUPPORT Unsupported feature
408 -ESHUTDOWN Already archived */
410 if (r == -E2BIG || r == -EFBIG || r == -EDQUOT || r == -ENOSPC)
411 log_debug("%s: Allocation limit reached, rotating.", f->path);
412 else if (r == -EHOSTDOWN)
413 log_info("%s: Journal file from other machine, rotating.", f->path);
414 else if (r == -EBUSY)
415 log_info("%s: Unclean shutdown, rotating.", f->path);
416 else if (r == -EPROTONOSUPPORT)
417 log_info("%s: Unsupported feature, rotating.", f->path);
418 else if (r == -EBADMSG || r == -ENODATA || r == ESHUTDOWN)
419 log_warning("%s: Journal file corrupted, rotating.", f->path);
426 static void write_to_journal(Server *s, uid_t uid, struct iovec *iovec, unsigned n, int priority) {
428 bool vacuumed = false;
435 f = find_journal(s, uid);
439 if (journal_file_rotate_suggested(f, s->max_file_usec)) {
440 log_debug("%s: Journal header limits reached or header out-of-date, rotating.", f->path);
445 f = find_journal(s, uid);
450 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
452 server_schedule_sync(s, priority);
456 if (vacuumed || !shall_try_append_again(f, r)) {
459 for (i = 0; i < n; i++)
460 size += iovec[i].iov_len;
462 log_error("Failed to write entry (%d items, %zu bytes), ignoring: %s", n, size, strerror(-r));
469 f = find_journal(s, uid);
473 log_debug("Retrying write.");
474 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
478 for (i = 0; i < n; i++)
479 size += iovec[i].iov_len;
481 log_error("Failed to write entry (%d items, %zu bytes) despite vacuuming, ignoring: %s", n, size, strerror(-r));
483 server_schedule_sync(s, priority);
486 static void dispatch_message_real(
488 struct iovec *iovec, unsigned n, unsigned m,
491 const char *label, size_t label_len,
496 char pid[sizeof("_PID=") + DECIMAL_STR_MAX(pid_t)],
497 uid[sizeof("_UID=") + DECIMAL_STR_MAX(uid_t)],
498 gid[sizeof("_GID=") + DECIMAL_STR_MAX(gid_t)],
499 owner_uid[sizeof("_SYSTEMD_OWNER_UID=") + DECIMAL_STR_MAX(uid_t)],
500 source_time[sizeof("_SOURCE_REALTIME_TIMESTAMP=") + DECIMAL_STR_MAX(usec_t)],
501 boot_id[sizeof("_BOOT_ID=") + 32] = "_BOOT_ID=",
502 machine_id[sizeof("_MACHINE_ID=") + 32] = "_MACHINE_ID=",
503 o_uid[sizeof("OBJECT_UID=") + DECIMAL_STR_MAX(uid_t)],
504 o_gid[sizeof("OBJECT_GID=") + DECIMAL_STR_MAX(gid_t)],
505 o_owner_uid[sizeof("OBJECT_SYSTEMD_OWNER_UID=") + DECIMAL_STR_MAX(uid_t)];
512 uid_t realuid = 0, owner = 0, journal_uid;
513 bool owner_valid = false;
515 char audit_session[sizeof("_AUDIT_SESSION=") + DECIMAL_STR_MAX(uint32_t)],
516 audit_loginuid[sizeof("_AUDIT_LOGINUID=") + DECIMAL_STR_MAX(uid_t)],
517 o_audit_session[sizeof("OBJECT_AUDIT_SESSION=") + DECIMAL_STR_MAX(uint32_t)],
518 o_audit_loginuid[sizeof("OBJECT_AUDIT_LOGINUID=") + DECIMAL_STR_MAX(uid_t)];
527 assert(n + N_IOVEC_META_FIELDS + (object_pid ? N_IOVEC_OBJECT_FIELDS : 0) <= m);
530 realuid = ucred->uid;
532 sprintf(pid, "_PID=%lu", (unsigned long) ucred->pid);
533 IOVEC_SET_STRING(iovec[n++], pid);
535 sprintf(uid, "_UID=%lu", (unsigned long) ucred->uid);
536 IOVEC_SET_STRING(iovec[n++], uid);
538 sprintf(gid, "_GID=%lu", (unsigned long) ucred->gid);
539 IOVEC_SET_STRING(iovec[n++], gid);
541 r = get_process_comm(ucred->pid, &t);
543 x = strappenda("_COMM=", t);
545 IOVEC_SET_STRING(iovec[n++], x);
548 r = get_process_exe(ucred->pid, &t);
550 x = strappenda("_EXE=", t);
552 IOVEC_SET_STRING(iovec[n++], x);
555 r = get_process_cmdline(ucred->pid, 0, false, &t);
557 x = strappenda("_CMDLINE=", t);
559 IOVEC_SET_STRING(iovec[n++], x);
562 r = get_process_capeff(ucred->pid, &t);
564 x = strappenda("_CAP_EFFECTIVE=", t);
566 IOVEC_SET_STRING(iovec[n++], x);
570 r = audit_session_from_pid(ucred->pid, &audit);
572 sprintf(audit_session, "_AUDIT_SESSION=%lu", (unsigned long) audit);
573 IOVEC_SET_STRING(iovec[n++], audit_session);
576 r = audit_loginuid_from_pid(ucred->pid, &loginuid);
578 sprintf(audit_loginuid, "_AUDIT_LOGINUID=%lu", (unsigned long) loginuid);
579 IOVEC_SET_STRING(iovec[n++], audit_loginuid);
583 r = cg_pid_get_path_shifted(ucred->pid, NULL, &c);
585 char *session = NULL;
587 x = strappenda("_SYSTEMD_CGROUP=", c);
588 IOVEC_SET_STRING(iovec[n++], x);
590 r = cg_path_get_session(c, &t);
592 session = strappenda("_SYSTEMD_SESSION=", t);
594 IOVEC_SET_STRING(iovec[n++], session);
597 if (cg_path_get_owner_uid(c, &owner) >= 0) {
600 sprintf(owner_uid, "_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner);
601 IOVEC_SET_STRING(iovec[n++], owner_uid);
604 if (cg_path_get_unit(c, &t) >= 0) {
605 x = strappenda("_SYSTEMD_UNIT=", t);
607 IOVEC_SET_STRING(iovec[n++], x);
608 } else if (unit_id && !session) {
609 x = strappenda("_SYSTEMD_UNIT=", unit_id);
610 IOVEC_SET_STRING(iovec[n++], x);
613 if (cg_path_get_user_unit(c, &t) >= 0) {
614 x = strappenda("_SYSTEMD_USER_UNIT=", t);
616 IOVEC_SET_STRING(iovec[n++], x);
617 } else if (unit_id && session) {
618 x = strappenda("_SYSTEMD_USER_UNIT=", unit_id);
619 IOVEC_SET_STRING(iovec[n++], x);
622 if (cg_path_get_slice(c, &t) >= 0) {
623 x = strappenda("_SYSTEMD_SLICE=", t);
625 IOVEC_SET_STRING(iovec[n++], x);
633 x = alloca(sizeof("_SELINUX_CONTEXT=") + label_len);
635 *((char*) mempcpy(stpcpy(x, "_SELINUX_CONTEXT="), label, label_len)) = 0;
636 IOVEC_SET_STRING(iovec[n++], x);
638 security_context_t con;
640 if (getpidcon(ucred->pid, &con) >= 0) {
641 x = strappenda("_SELINUX_CONTEXT=", con);
644 IOVEC_SET_STRING(iovec[n++], x);
652 r = get_process_uid(object_pid, &object_uid);
654 sprintf(o_uid, "OBJECT_UID=%lu", (unsigned long) object_uid);
655 IOVEC_SET_STRING(iovec[n++], o_uid);
658 r = get_process_gid(object_pid, &object_gid);
660 sprintf(o_gid, "OBJECT_GID=%lu", (unsigned long) object_gid);
661 IOVEC_SET_STRING(iovec[n++], o_gid);
664 r = get_process_comm(object_pid, &t);
666 x = strappenda("OBJECT_COMM=", t);
668 IOVEC_SET_STRING(iovec[n++], x);
671 r = get_process_exe(object_pid, &t);
673 x = strappenda("OBJECT_EXE=", t);
675 IOVEC_SET_STRING(iovec[n++], x);
678 r = get_process_cmdline(object_pid, 0, false, &t);
680 x = strappenda("OBJECT_CMDLINE=", t);
682 IOVEC_SET_STRING(iovec[n++], x);
686 r = audit_session_from_pid(object_pid, &audit);
688 sprintf(o_audit_session, "OBJECT_AUDIT_SESSION=%lu", (unsigned long) audit);
689 IOVEC_SET_STRING(iovec[n++], o_audit_session);
692 r = audit_loginuid_from_pid(object_pid, &loginuid);
694 sprintf(o_audit_loginuid, "OBJECT_AUDIT_LOGINUID=%lu", (unsigned long) loginuid);
695 IOVEC_SET_STRING(iovec[n++], o_audit_loginuid);
699 r = cg_pid_get_path_shifted(object_pid, NULL, &c);
701 x = strappenda("OBJECT_SYSTEMD_CGROUP=", c);
702 IOVEC_SET_STRING(iovec[n++], x);
704 r = cg_path_get_session(c, &t);
706 x = strappenda("OBJECT_SYSTEMD_SESSION=", t);
708 IOVEC_SET_STRING(iovec[n++], x);
711 if (cg_path_get_owner_uid(c, &owner) >= 0) {
712 sprintf(o_owner_uid, "OBJECT_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner);
713 IOVEC_SET_STRING(iovec[n++], o_owner_uid);
716 if (cg_path_get_unit(c, &t) >= 0) {
717 x = strappenda("OBJECT_SYSTEMD_UNIT=", t);
719 IOVEC_SET_STRING(iovec[n++], x);
722 if (cg_path_get_user_unit(c, &t) >= 0) {
723 x = strappenda("OBJECT_SYSTEMD_USER_UNIT=", t);
725 IOVEC_SET_STRING(iovec[n++], x);
734 sprintf(source_time, "_SOURCE_REALTIME_TIMESTAMP=%llu", (unsigned long long) timeval_load(tv));
735 IOVEC_SET_STRING(iovec[n++], source_time);
738 /* Note that strictly speaking storing the boot id here is
739 * redundant since the entry includes this in-line
740 * anyway. However, we need this indexed, too. */
741 r = sd_id128_get_boot(&id);
743 sd_id128_to_string(id, boot_id + strlen("_BOOT_ID="));
744 IOVEC_SET_STRING(iovec[n++], boot_id);
747 r = sd_id128_get_machine(&id);
749 sd_id128_to_string(id, machine_id + strlen("_MACHINE_ID="));
750 IOVEC_SET_STRING(iovec[n++], machine_id);
753 t = gethostname_malloc();
755 x = strappenda("_HOSTNAME=", t);
757 IOVEC_SET_STRING(iovec[n++], x);
762 if (s->split_mode == SPLIT_UID && realuid > 0)
763 /* Split up strictly by any UID */
764 journal_uid = realuid;
765 else if (s->split_mode == SPLIT_LOGIN && realuid > 0 && owner_valid && owner > 0)
766 /* Split up by login UIDs, this avoids creation of
767 * individual journals for system UIDs. We do this
768 * only if the realuid is not root, in order not to
769 * accidentally leak privileged information to the
770 * user that is logged by a privileged process that is
771 * part of an unprivileged session.*/
776 write_to_journal(s, journal_uid, iovec, n, priority);
779 void server_driver_message(Server *s, sd_id128_t message_id, const char *format, ...) {
780 char mid[11 + 32 + 1];
781 char buffer[16 + LINE_MAX + 1];
782 struct iovec iovec[N_IOVEC_META_FIELDS + 4];
785 struct ucred ucred = {};
790 IOVEC_SET_STRING(iovec[n++], "PRIORITY=6");
791 IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=driver");
793 memcpy(buffer, "MESSAGE=", 8);
794 va_start(ap, format);
795 vsnprintf(buffer + 8, sizeof(buffer) - 8, format, ap);
797 char_array_0(buffer);
798 IOVEC_SET_STRING(iovec[n++], buffer);
800 if (!sd_id128_equal(message_id, SD_ID128_NULL)) {
801 snprintf(mid, sizeof(mid), MESSAGE_ID(message_id));
803 IOVEC_SET_STRING(iovec[n++], mid);
806 ucred.pid = getpid();
807 ucred.uid = getuid();
808 ucred.gid = getgid();
810 dispatch_message_real(s, iovec, n, ELEMENTSOF(iovec), &ucred, NULL, NULL, 0, NULL, LOG_INFO, 0);
813 void server_dispatch_message(
815 struct iovec *iovec, unsigned n, unsigned m,
818 const char *label, size_t label_len,
824 _cleanup_free_ char *path = NULL;
828 assert(iovec || n == 0);
833 if (LOG_PRI(priority) > s->max_level_store)
836 /* Stop early in case the information will not be stored
838 if (s->storage == STORAGE_NONE)
844 r = cg_pid_get_path_shifted(ucred->pid, NULL, &path);
848 /* example: /user/lennart/3/foobar
849 * /system/dbus.service/foobar
851 * So let's cut of everything past the third /, since that is
852 * where user directories start */
854 c = strchr(path, '/');
856 c = strchr(c+1, '/');
858 c = strchr(c+1, '/');
864 rl = journal_rate_limit_test(s->rate_limit, path,
865 priority & LOG_PRIMASK, available_space(s, false));
870 /* Write a suppression message if we suppressed something */
872 server_driver_message(s, SD_MESSAGE_JOURNAL_DROPPED,
873 "Suppressed %u messages from %s", rl - 1, path);
876 dispatch_message_real(s, iovec, n, m, ucred, tv, label, label_len, unit_id, priority, object_pid);
880 static int system_journal_open(Server *s) {
886 r = sd_id128_get_machine(&machine);
888 log_error("Failed to get machine id: %s", strerror(-r));
892 sd_id128_to_string(machine, ids);
894 if (!s->system_journal &&
895 (s->storage == STORAGE_PERSISTENT || s->storage == STORAGE_AUTO) &&
896 access("/run/systemd/journal/flushed", F_OK) >= 0) {
898 /* If in auto mode: first try to create the machine
899 * path, but not the prefix.
901 * If in persistent mode: create /var/log/journal and
902 * the machine path */
904 if (s->storage == STORAGE_PERSISTENT)
905 (void) mkdir("/var/log/journal/", 0755);
907 fn = strappenda("/var/log/journal/", ids);
908 (void) mkdir(fn, 0755);
910 fn = strappenda(fn, "/system.journal");
911 r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, NULL, &s->system_journal);
914 server_fix_perms(s, s->system_journal, 0);
916 if (r != -ENOENT && r != -EROFS)
917 log_warning("Failed to open system journal: %s", strerror(-r));
923 if (!s->runtime_journal &&
924 (s->storage != STORAGE_NONE)) {
926 fn = strjoin("/run/log/journal/", ids, "/system.journal", NULL);
930 if (s->system_journal) {
932 /* Try to open the runtime journal, but only
933 * if it already exists, so that we can flush
934 * it into the system journal */
936 r = journal_file_open(fn, O_RDWR, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal);
941 log_warning("Failed to open runtime journal: %s", strerror(-r));
948 /* OK, we really need the runtime journal, so create
949 * it if necessary. */
951 (void) mkdir_parents(fn, 0755);
952 r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal);
956 log_error("Failed to open runtime journal: %s", strerror(-r));
961 if (s->runtime_journal)
962 server_fix_perms(s, s->runtime_journal, 0);
965 available_space(s, true);
970 int server_flush_to_var(Server *s) {
973 sd_journal *j = NULL;
977 if (s->storage != STORAGE_AUTO &&
978 s->storage != STORAGE_PERSISTENT)
981 if (!s->runtime_journal)
984 system_journal_open(s);
986 if (!s->system_journal)
989 log_debug("Flushing to /var...");
991 r = sd_id128_get_machine(&machine);
995 r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY);
997 log_error("Failed to read runtime journal: %s", strerror(-r));
1001 sd_journal_set_data_threshold(j, 0);
1003 SD_JOURNAL_FOREACH(j) {
1007 f = j->current_file;
1008 assert(f && f->current_offset > 0);
1010 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1012 log_error("Can't read entry: %s", strerror(-r));
1016 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1020 if (!shall_try_append_again(s->system_journal, r)) {
1021 log_error("Can't write entry: %s", strerror(-r));
1028 if (!s->system_journal) {
1029 log_notice("Didn't flush runtime journal since rotation of system journal wasn't successful.");
1034 log_debug("Retrying write.");
1035 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1037 log_error("Can't write entry: %s", strerror(-r));
1043 journal_file_post_change(s->system_journal);
1045 journal_file_close(s->runtime_journal);
1046 s->runtime_journal = NULL;
1049 rm_rf("/run/log/journal", false, true, false);
1051 sd_journal_close(j);
1056 int process_event(Server *s, struct epoll_event *ev) {
1060 if (ev->data.fd == s->signal_fd) {
1061 struct signalfd_siginfo sfsi;
1064 if (ev->events != EPOLLIN) {
1065 log_error("Got invalid event from epoll for %s: %"PRIx32,
1066 "signal fd", ev->events);
1070 n = read(s->signal_fd, &sfsi, sizeof(sfsi));
1071 if (n != sizeof(sfsi)) {
1076 if (errno == EINTR || errno == EAGAIN)
1082 if (sfsi.ssi_signo == SIGUSR1) {
1083 log_info("Received request to flush runtime journal from PID %"PRIu32,
1085 touch("/run/systemd/journal/flushed");
1086 server_flush_to_var(s);
1091 if (sfsi.ssi_signo == SIGUSR2) {
1092 log_info("Received request to rotate journal from PID %"PRIu32,
1099 log_info("Received SIG%s", signal_to_string(sfsi.ssi_signo));
1103 } else if (ev->data.fd == s->sync_timer_fd) {
1107 log_debug("Got sync request from epoll.");
1109 r = read(ev->data.fd, (void *)&t, sizeof(t));
1116 } else if (ev->data.fd == s->dev_kmsg_fd) {
1119 if (ev->events & EPOLLERR)
1120 log_warning("/dev/kmsg buffer overrun, some messages lost.");
1122 if (!(ev->events & EPOLLIN)) {
1123 log_error("Got invalid event from epoll for %s: %"PRIx32,
1124 "/dev/kmsg", ev->events);
1128 r = server_read_dev_kmsg(s);
1134 } else if (ev->data.fd == s->native_fd ||
1135 ev->data.fd == s->syslog_fd) {
1137 if (ev->events != EPOLLIN) {
1138 log_error("Got invalid event from epoll for %s: %"PRIx32,
1139 ev->data.fd == s->native_fd ? "native fd" : "syslog fd",
1145 struct msghdr msghdr;
1147 struct ucred *ucred = NULL;
1148 struct timeval *tv = NULL;
1149 struct cmsghdr *cmsg;
1151 size_t label_len = 0;
1153 struct cmsghdr cmsghdr;
1155 /* We use NAME_MAX space for the
1156 * SELinux label here. The kernel
1157 * currently enforces no limit, but
1158 * according to suggestions from the
1159 * SELinux people this will change and
1160 * it will probably be identical to
1161 * NAME_MAX. For now we use that, but
1162 * this should be updated one day when
1163 * the final limit is known.*/
1164 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1165 CMSG_SPACE(sizeof(struct timeval)) +
1166 CMSG_SPACE(sizeof(int)) + /* fd */
1167 CMSG_SPACE(NAME_MAX)]; /* selinux label */
1174 if (ioctl(ev->data.fd, SIOCINQ, &v) < 0) {
1175 log_error("SIOCINQ failed: %m");
1179 if (s->buffer_size < (size_t) v) {
1183 l = MAX(LINE_MAX + (size_t) v, s->buffer_size * 2);
1184 b = realloc(s->buffer, l+1);
1187 log_error("Couldn't increase buffer.");
1196 iovec.iov_base = s->buffer;
1197 iovec.iov_len = s->buffer_size;
1201 msghdr.msg_iov = &iovec;
1202 msghdr.msg_iovlen = 1;
1203 msghdr.msg_control = &control;
1204 msghdr.msg_controllen = sizeof(control);
1206 n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
1209 if (errno == EINTR || errno == EAGAIN)
1212 log_error("recvmsg() failed: %m");
1216 for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
1218 if (cmsg->cmsg_level == SOL_SOCKET &&
1219 cmsg->cmsg_type == SCM_CREDENTIALS &&
1220 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
1221 ucred = (struct ucred*) CMSG_DATA(cmsg);
1222 else if (cmsg->cmsg_level == SOL_SOCKET &&
1223 cmsg->cmsg_type == SCM_SECURITY) {
1224 label = (char*) CMSG_DATA(cmsg);
1225 label_len = cmsg->cmsg_len - CMSG_LEN(0);
1226 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1227 cmsg->cmsg_type == SO_TIMESTAMP &&
1228 cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval)))
1229 tv = (struct timeval*) CMSG_DATA(cmsg);
1230 else if (cmsg->cmsg_level == SOL_SOCKET &&
1231 cmsg->cmsg_type == SCM_RIGHTS) {
1232 fds = (int*) CMSG_DATA(cmsg);
1233 n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1237 if (ev->data.fd == s->syslog_fd) {
1238 if (n > 0 && n_fds == 0) {
1240 server_process_syslog_message(s, strstrip(s->buffer), ucred, tv, label, label_len);
1241 } else if (n_fds > 0)
1242 log_warning("Got file descriptors via syslog socket. Ignoring.");
1245 if (n > 0 && n_fds == 0)
1246 server_process_native_message(s, s->buffer, n, ucred, tv, label, label_len);
1247 else if (n == 0 && n_fds == 1)
1248 server_process_native_file(s, fds[0], ucred, tv, label, label_len);
1250 log_warning("Got too many file descriptors via native socket. Ignoring.");
1253 close_many(fds, n_fds);
1258 } else if (ev->data.fd == s->stdout_fd) {
1260 if (ev->events != EPOLLIN) {
1261 log_error("Got invalid event from epoll for %s: %"PRIx32,
1262 "stdout fd", ev->events);
1266 stdout_stream_new(s);
1270 StdoutStream *stream;
1272 if ((ev->events|EPOLLIN|EPOLLHUP) != (EPOLLIN|EPOLLHUP)) {
1273 log_error("Got invalid event from epoll for %s: %"PRIx32,
1274 "stdout stream", ev->events);
1278 /* If it is none of the well-known fds, it must be an
1279 * stdout stream fd. Note that this is a bit ugly here
1280 * (since we rely that none of the well-known fds
1281 * could be interpreted as pointer), but nonetheless
1282 * safe, since the well-known fds would never get an
1283 * fd > 4096, i.e. beyond the first memory page */
1285 stream = ev->data.ptr;
1287 if (stdout_stream_process(stream) <= 0)
1288 stdout_stream_free(stream);
1293 log_error("Unknown event.");
1297 static int open_signalfd(Server *s) {
1299 struct epoll_event ev;
1303 assert_se(sigemptyset(&mask) == 0);
1304 sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, SIGUSR2, -1);
1305 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1307 s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1308 if (s->signal_fd < 0) {
1309 log_error("signalfd(): %m");
1314 ev.events = EPOLLIN;
1315 ev.data.fd = s->signal_fd;
1317 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->signal_fd, &ev) < 0) {
1318 log_error("epoll_ctl(): %m");
1325 static int server_parse_proc_cmdline(Server *s) {
1326 _cleanup_free_ char *line = NULL;
1331 if (detect_container(NULL) > 0)
1334 r = read_one_line_file("/proc/cmdline", &line);
1336 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
1340 FOREACH_WORD_QUOTED(w, l, line, state) {
1341 _cleanup_free_ char *word;
1343 word = strndup(w, l);
1347 if (startswith(word, "systemd.journald.forward_to_syslog=")) {
1348 r = parse_boolean(word + 35);
1350 log_warning("Failed to parse forward to syslog switch %s. Ignoring.", word + 35);
1352 s->forward_to_syslog = r;
1353 } else if (startswith(word, "systemd.journald.forward_to_kmsg=")) {
1354 r = parse_boolean(word + 33);
1356 log_warning("Failed to parse forward to kmsg switch %s. Ignoring.", word + 33);
1358 s->forward_to_kmsg = r;
1359 } else if (startswith(word, "systemd.journald.forward_to_console=")) {
1360 r = parse_boolean(word + 36);
1362 log_warning("Failed to parse forward to console switch %s. Ignoring.", word + 36);
1364 s->forward_to_console = r;
1365 } else if (startswith(word, "systemd.journald"))
1366 log_warning("Invalid systemd.journald parameter. Ignoring.");
1372 static int server_parse_config_file(Server *s) {
1373 static const char fn[] = "/etc/systemd/journald.conf";
1374 _cleanup_fclose_ FILE *f = NULL;
1379 f = fopen(fn, "re");
1381 if (errno == ENOENT)
1384 log_warning("Failed to open configuration file %s: %m", fn);
1388 r = config_parse(NULL, fn, f, "Journal\0", config_item_perf_lookup,
1389 (void*) journald_gperf_lookup, false, false, s);
1391 log_warning("Failed to parse configuration file: %s", strerror(-r));
1396 static int server_open_sync_timer(Server *s) {
1398 struct epoll_event ev;
1402 s->sync_timer_fd = timerfd_create(CLOCK_MONOTONIC, TFD_CLOEXEC);
1403 if (s->sync_timer_fd < 0)
1407 ev.events = EPOLLIN;
1408 ev.data.fd = s->sync_timer_fd;
1410 r = epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->sync_timer_fd, &ev);
1412 log_error("Failed to add idle timer fd to epoll object: %m");
1419 int server_schedule_sync(Server *s, int priority) {
1424 if (priority <= LOG_CRIT) {
1425 /* Immediately sync to disk when this is of priority CRIT, ALERT, EMERG */
1430 if (s->sync_scheduled)
1433 if (s->sync_interval_usec) {
1434 struct itimerspec sync_timer_enable = {};
1436 timespec_store(&sync_timer_enable.it_value, s->sync_interval_usec);
1438 r = timerfd_settime(s->sync_timer_fd, 0, &sync_timer_enable, NULL);
1443 s->sync_scheduled = true;
1448 int server_init(Server *s) {
1454 s->sync_timer_fd = s->syslog_fd = s->native_fd = s->stdout_fd =
1455 s->signal_fd = s->epoll_fd = s->dev_kmsg_fd = -1;
1459 s->sync_interval_usec = DEFAULT_SYNC_INTERVAL_USEC;
1460 s->sync_scheduled = false;
1462 s->rate_limit_interval = DEFAULT_RATE_LIMIT_INTERVAL;
1463 s->rate_limit_burst = DEFAULT_RATE_LIMIT_BURST;
1465 s->forward_to_syslog = true;
1467 s->max_level_store = LOG_DEBUG;
1468 s->max_level_syslog = LOG_DEBUG;
1469 s->max_level_kmsg = LOG_NOTICE;
1470 s->max_level_console = LOG_INFO;
1472 memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics));
1473 memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics));
1475 server_parse_config_file(s);
1476 server_parse_proc_cmdline(s);
1477 if (!!s->rate_limit_interval ^ !!s->rate_limit_burst) {
1478 log_debug("Setting both rate limit interval and burst from %llu,%u to 0,0",
1479 (long long unsigned) s->rate_limit_interval,
1480 s->rate_limit_burst);
1481 s->rate_limit_interval = s->rate_limit_burst = 0;
1484 mkdir_p("/run/systemd/journal", 0755);
1486 s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
1487 if (!s->user_journals)
1490 s->mmap = mmap_cache_new();
1494 s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
1495 if (s->epoll_fd < 0) {
1496 log_error("Failed to create epoll object: %m");
1500 n = sd_listen_fds(true);
1502 log_error("Failed to read listening file descriptors from environment: %s", strerror(-n));
1506 for (fd = SD_LISTEN_FDS_START; fd < SD_LISTEN_FDS_START + n; fd++) {
1508 if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/run/systemd/journal/socket", 0) > 0) {
1510 if (s->native_fd >= 0) {
1511 log_error("Too many native sockets passed.");
1517 } else if (sd_is_socket_unix(fd, SOCK_STREAM, 1, "/run/systemd/journal/stdout", 0) > 0) {
1519 if (s->stdout_fd >= 0) {
1520 log_error("Too many stdout sockets passed.");
1526 } else if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/dev/log", 0) > 0) {
1528 if (s->syslog_fd >= 0) {
1529 log_error("Too many /dev/log sockets passed.");
1536 log_error("Unknown socket passed.");
1541 r = server_open_syslog_socket(s);
1545 r = server_open_native_socket(s);
1549 r = server_open_stdout_socket(s);
1553 r = server_open_dev_kmsg(s);
1557 r = server_open_kernel_seqnum(s);
1561 r = server_open_sync_timer(s);
1565 r = open_signalfd(s);
1569 s->udev = udev_new();
1573 s->rate_limit = journal_rate_limit_new(s->rate_limit_interval,
1574 s->rate_limit_burst);
1578 r = system_journal_open(s);
1585 void server_maybe_append_tags(Server *s) {
1591 n = now(CLOCK_REALTIME);
1593 if (s->system_journal)
1594 journal_file_maybe_append_tag(s->system_journal, n);
1596 HASHMAP_FOREACH(f, s->user_journals, i)
1597 journal_file_maybe_append_tag(f, n);
1601 void server_done(Server *s) {
1605 while (s->stdout_streams)
1606 stdout_stream_free(s->stdout_streams);
1608 if (s->system_journal)
1609 journal_file_close(s->system_journal);
1611 if (s->runtime_journal)
1612 journal_file_close(s->runtime_journal);
1614 while ((f = hashmap_steal_first(s->user_journals)))
1615 journal_file_close(f);
1617 hashmap_free(s->user_journals);
1619 if (s->epoll_fd >= 0)
1620 close_nointr_nofail(s->epoll_fd);
1622 if (s->signal_fd >= 0)
1623 close_nointr_nofail(s->signal_fd);
1625 if (s->syslog_fd >= 0)
1626 close_nointr_nofail(s->syslog_fd);
1628 if (s->native_fd >= 0)
1629 close_nointr_nofail(s->native_fd);
1631 if (s->stdout_fd >= 0)
1632 close_nointr_nofail(s->stdout_fd);
1634 if (s->dev_kmsg_fd >= 0)
1635 close_nointr_nofail(s->dev_kmsg_fd);
1637 if (s->sync_timer_fd >= 0)
1638 close_nointr_nofail(s->sync_timer_fd);
1641 journal_rate_limit_free(s->rate_limit);
1643 if (s->kernel_seqnum)
1644 munmap(s->kernel_seqnum, sizeof(uint64_t));
1650 mmap_cache_unref(s->mmap);
1653 udev_unref(s->udev);