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 General Public License as published by
10 the Free Software Foundation; either version 2 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 General Public License for more details.
18 You should have received a copy of the GNU 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 <acl/libacl.h>
31 #include <sys/ioctl.h>
32 #include <linux/sockios.h>
33 #include <sys/statvfs.h>
36 #include "journal-file.h"
37 #include "sd-daemon.h"
38 #include "socket-util.h"
40 #include "cgroup-util.h"
42 #include "journal-rate-limit.h"
43 #include "sd-journal.h"
45 #include "journal-internal.h"
47 #define USER_JOURNALS_MAX 1024
48 #define STDOUT_STREAMS_MAX 4096
50 #define DEFAULT_RATE_LIMIT_INTERVAL (10*USEC_PER_SEC)
51 #define DEFAULT_RATE_LIMIT_BURST 200
53 #define RECHECK_AVAILABLE_SPACE_USEC (30*USEC_PER_SEC)
55 #define RECHECK_VAR_AVAILABLE_USEC (30*USEC_PER_SEC)
57 #define SYSLOG_TIMEOUT_USEC (5*USEC_PER_SEC)
59 typedef struct StdoutStream StdoutStream;
61 typedef struct Server {
68 JournalFile *runtime_journal;
69 JournalFile *system_journal;
70 Hashmap *user_journals;
77 JournalRateLimit *rate_limit;
79 JournalMetrics runtime_metrics;
80 JournalMetrics system_metrics;
84 uint64_t cached_available_space;
85 usec_t cached_available_space_timestamp;
87 uint64_t var_available_timestamp;
89 LIST_HEAD(StdoutStream, stdout_streams);
90 unsigned n_stdout_streams;
93 typedef enum StdoutStreamState {
95 STDOUT_STREAM_PRIORITY,
96 STDOUT_STREAM_PRIORITY_PREFIX,
97 STDOUT_STREAM_TEE_CONSOLE,
101 struct StdoutStream {
103 StdoutStreamState state;
111 bool priority_prefix:1;
114 char buffer[LINE_MAX+1];
117 LIST_FIELDS(StdoutStream, stdout_stream);
120 static int server_flush_to_var(Server *s);
122 static uint64_t available_space(Server *s) {
127 uint64_t sum = 0, avail = 0, ss_avail = 0;
133 ts = now(CLOCK_MONOTONIC);
135 if (s->cached_available_space_timestamp + RECHECK_AVAILABLE_SPACE_USEC > ts)
136 return s->cached_available_space;
138 r = sd_id128_get_machine(&machine);
142 if (s->system_journal) {
143 f = "/var/log/journal/";
144 m = &s->system_metrics;
146 f = "/run/log/journal/";
147 m = &s->runtime_metrics;
152 p = strappend(f, sd_id128_to_string(machine, ids));
162 if (fstatvfs(dirfd(d), &ss) < 0)
167 struct dirent buf, *de;
170 k = readdir_r(d, &buf, &de);
179 if (!dirent_is_file_with_suffix(de, ".journal"))
182 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
185 sum += (uint64_t) st.st_blocks * (uint64_t) st.st_blksize;
188 avail = sum >= m->max_use ? 0 : m->max_use - sum;
190 ss_avail = ss.f_bsize * ss.f_bavail;
192 ss_avail = ss_avail < m->keep_free ? 0 : ss_avail - m->keep_free;
194 if (ss_avail < avail)
197 s->cached_available_space = avail;
198 s->cached_available_space_timestamp = ts;
206 static void fix_perms(JournalFile *f, uid_t uid) {
209 acl_permset_t permset;
214 r = fchmod_and_fchown(f->fd, 0640, 0, 0);
216 log_warning("Failed to fix access mode/rights on %s, ignoring: %s", f->path, strerror(-r));
221 acl = acl_get_fd(f->fd);
223 log_warning("Failed to read ACL on %s, ignoring: %m", f->path);
227 r = acl_find_uid(acl, uid, &entry);
230 if (acl_create_entry(&acl, &entry) < 0 ||
231 acl_set_tag_type(entry, ACL_USER) < 0 ||
232 acl_set_qualifier(entry, &uid) < 0) {
233 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
238 if (acl_get_permset(entry, &permset) < 0 ||
239 acl_add_perm(permset, ACL_READ) < 0 ||
240 acl_calc_mask(&acl) < 0) {
241 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
245 if (acl_set_fd(f->fd, acl) < 0)
246 log_warning("Failed to set ACL on %s, ignoring: %m", f->path);
252 static JournalFile* find_journal(Server *s, uid_t uid) {
261 /* We split up user logs only on /var, not on /run. If the
262 * runtime file is open, we write to it exclusively, in order
263 * to guarantee proper order as soon as we flush /run to
264 * /var and close the runtime file. */
266 if (s->runtime_journal)
267 return s->runtime_journal;
270 return s->system_journal;
272 r = sd_id128_get_machine(&machine);
274 return s->system_journal;
276 f = hashmap_get(s->user_journals, UINT32_TO_PTR(uid));
280 if (asprintf(&p, "/var/log/journal/%s/user-%lu.journal", sd_id128_to_string(machine, ids), (unsigned long) uid) < 0)
281 return s->system_journal;
283 while (hashmap_size(s->user_journals) >= USER_JOURNALS_MAX) {
284 /* Too many open? Then let's close one */
285 f = hashmap_steal_first(s->user_journals);
287 journal_file_close(f);
290 r = journal_file_open(p, O_RDWR|O_CREAT, 0640, s->system_journal, &f);
294 return s->system_journal;
297 f->metrics = s->system_metrics;
298 f->compress = s->compress;
300 r = hashmap_put(s->user_journals, UINT32_TO_PTR(uid), f);
302 journal_file_close(f);
303 return s->system_journal;
309 static void server_vacuum(Server *s) {
318 log_info("Rotating...");
320 if (s->runtime_journal) {
321 r = journal_file_rotate(&s->runtime_journal);
323 log_error("Failed to rotate %s: %s", s->runtime_journal->path, strerror(-r));
326 if (s->system_journal) {
327 r = journal_file_rotate(&s->system_journal);
329 log_error("Failed to rotate %s: %s", s->system_journal->path, strerror(-r));
332 HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
333 r = journal_file_rotate(&f);
335 log_error("Failed to rotate %s: %s", f->path, strerror(-r));
337 hashmap_replace(s->user_journals, k, f);
340 log_info("Vacuuming...");
342 r = sd_id128_get_machine(&machine);
344 log_error("Failed to get machine ID: %s", strerror(-r));
348 sd_id128_to_string(machine, ids);
350 if (s->system_journal) {
351 if (asprintf(&p, "/var/log/journal/%s", ids) < 0) {
352 log_error("Out of memory.");
356 r = journal_directory_vacuum(p, s->system_metrics.max_use, s->system_metrics.keep_free);
357 if (r < 0 && r != -ENOENT)
358 log_error("Failed to vacuum %s: %s", p, strerror(-r));
363 if (s->runtime_journal) {
364 if (asprintf(&p, "/run/log/journal/%s", ids) < 0) {
365 log_error("Out of memory.");
369 r = journal_directory_vacuum(p, s->runtime_metrics.max_use, s->runtime_metrics.keep_free);
370 if (r < 0 && r != -ENOENT)
371 log_error("Failed to vacuum %s: %s", p, strerror(-r));
375 s->cached_available_space_timestamp = 0;
378 static char *shortened_cgroup_path(pid_t pid) {
380 char *process_path, *init_path, *path;
384 r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &process_path);
388 r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 1, &init_path);
394 if (streq(init_path, "/"))
397 if (startswith(process_path, init_path)) {
400 p = strdup(process_path + strlen(init_path));
418 static void dispatch_message_real(Server *s,
419 struct iovec *iovec, unsigned n, unsigned m,
421 struct timeval *tv) {
423 char *pid = NULL, *uid = NULL, *gid = NULL,
424 *source_time = NULL, *boot_id = NULL, *machine_id = NULL,
425 *comm = NULL, *cmdline = NULL, *hostname = NULL,
426 *audit_session = NULL, *audit_loginuid = NULL,
427 *exe = NULL, *cgroup = NULL, *session = NULL,
428 *owner_uid = NULL, *service = NULL;
434 uid_t loginuid = 0, realuid = 0;
436 bool vacuumed = false;
447 realuid = ucred->uid;
449 if (asprintf(&pid, "_PID=%lu", (unsigned long) ucred->pid) >= 0)
450 IOVEC_SET_STRING(iovec[n++], pid);
452 if (asprintf(&uid, "_UID=%lu", (unsigned long) ucred->uid) >= 0)
453 IOVEC_SET_STRING(iovec[n++], uid);
455 if (asprintf(&gid, "_GID=%lu", (unsigned long) ucred->gid) >= 0)
456 IOVEC_SET_STRING(iovec[n++], gid);
458 r = get_process_comm(ucred->pid, &t);
460 comm = strappend("_COMM=", t);
464 IOVEC_SET_STRING(iovec[n++], comm);
467 r = get_process_exe(ucred->pid, &t);
469 exe = strappend("_EXE=", t);
473 IOVEC_SET_STRING(iovec[n++], exe);
476 r = get_process_cmdline(ucred->pid, LINE_MAX, false, &t);
478 cmdline = strappend("_CMDLINE=", t);
482 IOVEC_SET_STRING(iovec[n++], cmdline);
485 r = audit_session_from_pid(ucred->pid, &audit);
487 if (asprintf(&audit_session, "_AUDIT_SESSION=%lu", (unsigned long) audit) >= 0)
488 IOVEC_SET_STRING(iovec[n++], audit_session);
490 r = audit_loginuid_from_pid(ucred->pid, &loginuid);
492 if (asprintf(&audit_loginuid, "_AUDIT_LOGINUID=%lu", (unsigned long) loginuid) >= 0)
493 IOVEC_SET_STRING(iovec[n++], audit_loginuid);
495 t = shortened_cgroup_path(ucred->pid);
497 cgroup = strappend("_SYSTEMD_CGROUP=", t);
501 IOVEC_SET_STRING(iovec[n++], cgroup);
504 if (sd_pid_get_session(ucred->pid, &t) >= 0) {
505 session = strappend("_SYSTEMD_SESSION=", t);
509 IOVEC_SET_STRING(iovec[n++], session);
512 if (sd_pid_get_service(ucred->pid, &t) >= 0) {
513 service = strappend("_SYSTEMD_SERVICE=", t);
517 IOVEC_SET_STRING(iovec[n++], service);
520 if (sd_pid_get_owner_uid(ucred->uid, &owner) >= 0)
521 if (asprintf(&owner_uid, "_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner) >= 0)
522 IOVEC_SET_STRING(iovec[n++], owner_uid);
526 if (asprintf(&source_time, "_SOURCE_REALTIME_TIMESTAMP=%llu",
527 (unsigned long long) timeval_load(tv)) >= 0)
528 IOVEC_SET_STRING(iovec[n++], source_time);
531 /* Note that strictly speaking storing the boot id here is
532 * redundant since the entry includes this in-line
533 * anyway. However, we need this indexed, too. */
534 r = sd_id128_get_boot(&id);
536 if (asprintf(&boot_id, "_BOOT_ID=%s", sd_id128_to_string(id, idbuf)) >= 0)
537 IOVEC_SET_STRING(iovec[n++], boot_id);
539 r = sd_id128_get_machine(&id);
541 if (asprintf(&machine_id, "_MACHINE_ID=%s", sd_id128_to_string(id, idbuf)) >= 0)
542 IOVEC_SET_STRING(iovec[n++], machine_id);
544 t = gethostname_malloc();
546 hostname = strappend("_HOSTNAME=", t);
549 IOVEC_SET_STRING(iovec[n++], hostname);
554 server_flush_to_var(s);
557 f = find_journal(s, realuid == 0 ? 0 : loginuid);
559 log_warning("Dropping message, as we can't find a place to store the data.");
561 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
563 if (r == -E2BIG && !vacuumed) {
564 log_info("Allocation limit reached.");
569 log_info("Retrying write.");
574 log_error("Failed to write entry, ignoring: %s", strerror(-r));
588 free(audit_loginuid);
595 static void dispatch_message(Server *s,
596 struct iovec *iovec, unsigned n, unsigned m,
601 char *path = NULL, *c;
604 assert(iovec || n == 0);
612 path = shortened_cgroup_path(ucred->pid);
616 /* example: /user/lennart/3/foobar
617 * /system/dbus.service/foobar
619 * So let's cut of everything past the third /, since that is
620 * wher user directories start */
622 c = strchr(path, '/');
624 c = strchr(c+1, '/');
626 c = strchr(c+1, '/');
632 rl = journal_rate_limit_test(s->rate_limit, path, priority, available_space(s));
641 char suppress_message[LINE_MAX];
642 struct iovec suppress_iovec[18];
644 /* Write a suppression message if we suppressed something */
646 snprintf(suppress_message, sizeof(suppress_message), "MESSAGE=Suppressed %u messages from %s", rl - 1, path);
647 char_array_0(suppress_message);
649 IOVEC_SET_STRING(suppress_iovec[j++], "PRIORITY=5");
650 IOVEC_SET_STRING(suppress_iovec[j++], suppress_message);
652 dispatch_message_real(s, suppress_iovec, j, ELEMENTSOF(suppress_iovec), NULL, NULL);
658 dispatch_message_real(s, iovec, n, m, ucred, tv);
661 static void process_syslog_message(Server *s, const char *buf, struct ucred *ucred, struct timeval *tv) {
662 char *message = NULL, *syslog_priority = NULL, *syslog_facility = NULL;
663 struct iovec iovec[19];
665 int priority = LOG_USER | LOG_INFO;
670 parse_syslog_priority((char**) &buf, &priority);
671 skip_syslog_date((char**) &buf);
673 if (asprintf(&syslog_priority, "PRIORITY=%i", priority & LOG_PRIMASK) >= 0)
674 IOVEC_SET_STRING(iovec[n++], syslog_priority);
676 if (asprintf(&syslog_facility, "SYSLOG_FACILITY=%i", LOG_FAC(priority)) >= 0)
677 IOVEC_SET_STRING(iovec[n++], syslog_facility);
679 message = strappend("MESSAGE=", buf);
681 IOVEC_SET_STRING(iovec[n++], message);
683 dispatch_message(s, iovec, n, ELEMENTSOF(iovec), ucred, tv, priority & LOG_PRIMASK);
686 free(syslog_facility);
687 free(syslog_priority);
690 static bool valid_user_field(const char *p, size_t l) {
693 /* We kinda enforce POSIX syntax recommendations for
694 environment variables here, but make a couple of additional
697 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
699 /* No empty field names */
703 /* Don't allow names longer than 64 chars */
707 /* Variables starting with an underscore are protected */
711 /* Don't allow digits as first character */
712 if (p[0] >= '0' && p[0] <= '9')
715 /* Only allow A-Z0-9 and '_' */
716 for (a = p; a < p + l; a++)
717 if (!((*a >= 'A' && *a <= 'Z') ||
718 (*a >= '0' && *a <= '9') ||
725 static void process_native_message(Server *s, const void *buffer, size_t buffer_size, struct ucred *ucred, struct timeval *tv) {
726 struct iovec *iovec = NULL;
727 unsigned n = 0, m = 0, j;
730 int priority = LOG_INFO;
733 assert(buffer || n == 0);
736 remaining = buffer_size;
738 while (remaining > 0) {
741 e = memchr(p, '\n', remaining);
744 /* Trailing noise, let's ignore it, and flush what we collected */
745 log_debug("Received message with trailing noise, ignoring.");
750 /* Entry separator */
751 dispatch_message(s, iovec, n, m, ucred, tv, priority);
760 if (*p == '.' || *p == '#') {
761 /* Ignore control commands for now, and
763 remaining -= (e - p) + 1;
768 /* A property follows */
774 u = MAX((n+16U) * 2U, 4U);
775 c = realloc(iovec, u * sizeof(struct iovec));
777 log_error("Out of memory");
785 q = memchr(p, '=', e - p);
787 if (valid_user_field(p, q - p)) {
788 /* If the field name starts with an
789 * underscore, skip the variable,
790 * since that indidates a trusted
792 iovec[n].iov_base = (char*) p;
793 iovec[n].iov_len = e - p;
796 /* We need to determine the priority
797 * of this entry for the rate limiting
800 memcmp(p, "PRIORITY=", 10) == 0 &&
803 priority = p[10] - '0';
806 remaining -= (e - p) + 1;
813 if (remaining < e - p + 1 + sizeof(uint64_t) + 1) {
814 log_debug("Failed to parse message, ignoring.");
818 memcpy(&l, e + 1, sizeof(uint64_t));
821 if (remaining < e - p + 1 + sizeof(uint64_t) + l + 1 ||
822 e[1+sizeof(uint64_t)+l] != '\n') {
823 log_debug("Failed to parse message, ignoring.");
827 k = malloc((e - p) + 1 + l);
829 log_error("Out of memory");
835 memcpy(k + (e - p) + 1, e + 1 + sizeof(uint64_t), l);
837 if (valid_user_field(p, e - p)) {
838 iovec[n].iov_base = k;
839 iovec[n].iov_len = (e - p) + 1 + l;
844 remaining -= (e - p) + 1 + sizeof(uint64_t) + l + 1;
845 p = e + 1 + sizeof(uint64_t) + l + 1;
849 dispatch_message(s, iovec, n, m, ucred, tv, priority);
851 for (j = 0; j < n; j++)
852 if (iovec[j].iov_base < buffer ||
853 (const uint8_t*) iovec[j].iov_base >= (const uint8_t*) buffer + buffer_size)
854 free(iovec[j].iov_base);
857 static int stdout_stream_log(StdoutStream *s, const char *p, size_t l) {
858 struct iovec iovec[18];
859 char *message = NULL, *syslog_priority = NULL;
867 priority = s->priority;
869 if (s->priority_prefix &&
872 p[1] >= '0' && p[1] <= '7' &&
875 priority = p[1] - '0';
883 if (asprintf(&syslog_priority, "PRIORITY=%i", priority) >= 0)
884 IOVEC_SET_STRING(iovec[n++], syslog_priority);
886 tag_len = s->tag ? strlen(s->tag) + 2: 0;
887 message = malloc(8 + tag_len + l);
889 memcpy(message, "MESSAGE=", 8);
892 memcpy(message+8, s->tag, tag_len-2);
893 memcpy(message+8+tag_len-2, ": ", 2);
896 memcpy(message+8+tag_len, p, l);
897 iovec[n].iov_base = message;
898 iovec[n].iov_len = 8+tag_len+l;
902 dispatch_message(s->server, iovec, n, ELEMENTSOF(iovec), &s->ucred, NULL, priority);
904 if (s->tee_console) {
907 console = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
911 IOVEC_SET_STRING(iovec[n++], s->tag);
912 IOVEC_SET_STRING(iovec[n++], ": ");
915 iovec[n].iov_base = (void*) p;
916 iovec[n].iov_len = l;
919 IOVEC_SET_STRING(iovec[n++], (char*) "\n");
921 writev(console, iovec, n);
926 free(syslog_priority);
931 static int stdout_stream_line(StdoutStream *s, const char *p, size_t l) {
935 while (l > 0 && strchr(WHITESPACE, *p)) {
940 while (l > 0 && strchr(WHITESPACE, *(p+l-1)))
945 case STDOUT_STREAM_TAG:
948 s->tag = strndup(p, l);
950 log_error("Out of memory");
955 s->state = STDOUT_STREAM_PRIORITY;
958 case STDOUT_STREAM_PRIORITY:
959 if (l != 1 || *p < '0' || *p > '7') {
960 log_warning("Failed to parse log priority line.");
964 s->priority = *p - '0';
965 s->state = STDOUT_STREAM_PRIORITY_PREFIX;
968 case STDOUT_STREAM_PRIORITY_PREFIX:
969 if (l != 1 || *p < '0' || *p > '1') {
970 log_warning("Failed to parse priority prefix line.");
974 s->priority_prefix = *p - '0';
975 s->state = STDOUT_STREAM_TEE_CONSOLE;
978 case STDOUT_STREAM_TEE_CONSOLE:
979 if (l != 1 || *p < '0' || *p > '1') {
980 log_warning("Failed to parse tee to console line.");
984 s->tee_console = *p - '0';
985 s->state = STDOUT_STREAM_RUNNING;
988 case STDOUT_STREAM_RUNNING:
989 return stdout_stream_log(s, p, l);
992 assert_not_reached("Unknown stream state");
995 static int stdout_stream_scan(StdoutStream *s, bool force_flush) {
1003 remaining = s->length;
1008 end = memchr(p, '\n', remaining);
1010 if (remaining >= LINE_MAX) {
1018 r = stdout_stream_line(s, p, end - p);
1026 if (force_flush && remaining > 0) {
1027 r = stdout_stream_line(s, p, remaining);
1035 if (p > s->buffer) {
1036 memmove(s->buffer, p, remaining);
1037 s->length = remaining;
1043 static int stdout_stream_process(StdoutStream *s) {
1049 l = read(s->fd, s->buffer+s->length, sizeof(s->buffer)-1-s->length);
1052 if (errno == EAGAIN)
1055 log_warning("Failed to read from stream: %m");
1060 r = stdout_stream_scan(s, true);
1068 r = stdout_stream_scan(s, false);
1076 static void stdout_stream_free(StdoutStream *s) {
1080 assert(s->server->n_stdout_streams > 0);
1081 s->server->n_stdout_streams --;
1082 LIST_REMOVE(StdoutStream, stdout_stream, s->server->stdout_streams, s);
1087 epoll_ctl(s->server->epoll_fd, EPOLL_CTL_DEL, s->fd, NULL);
1089 close_nointr_nofail(s->fd);
1096 static int stdout_stream_new(Server *s) {
1097 StdoutStream *stream;
1100 struct epoll_event ev;
1104 fd = accept4(s->stdout_fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
1106 if (errno == EAGAIN)
1109 log_error("Failed to accept stdout connection: %m");
1113 if (s->n_stdout_streams >= STDOUT_STREAMS_MAX) {
1114 log_warning("Too many stdout streams, refusing connection.");
1115 close_nointr_nofail(fd);
1119 stream = new0(StdoutStream, 1);
1121 log_error("Out of memory.");
1122 close_nointr_nofail(fd);
1128 len = sizeof(stream->ucred);
1129 if (getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &stream->ucred, &len) < 0) {
1130 log_error("Failed to determine peer credentials: %m");
1135 if (shutdown(fd, SHUT_WR) < 0) {
1136 log_error("Failed to shutdown writing side of socket: %m");
1142 ev.data.ptr = stream;
1143 ev.events = EPOLLIN;
1144 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) {
1145 log_error("Failed to add stream to event loop: %m");
1151 LIST_PREPEND(StdoutStream, stdout_stream, s->stdout_streams, stream);
1152 s->n_stdout_streams ++;
1157 stdout_stream_free(stream);
1161 static int system_journal_open(Server *s) {
1167 r = sd_id128_get_machine(&machine);
1171 sd_id128_to_string(machine, ids);
1173 if (!s->system_journal) {
1175 /* First try to create the machine path, but not the prefix */
1176 fn = strappend("/var/log/journal/", ids);
1179 (void) mkdir(fn, 0755);
1182 /* The create the system journal file */
1183 fn = join("/var/log/journal/", ids, "/system.journal", NULL);
1187 r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->system_journal);
1191 journal_default_metrics(&s->system_metrics, s->system_journal->fd);
1193 s->system_journal->metrics = s->system_metrics;
1194 s->system_journal->compress = s->compress;
1196 fix_perms(s->system_journal, 0);
1202 log_error("Failed to open system journal: %s", strerror(-r));
1208 if (!s->runtime_journal) {
1210 fn = join("/run/log/journal/", ids, "/system.journal", NULL);
1214 if (s->system_journal) {
1216 /* Try to open the runtime journal, but only
1217 * if it already exists, so that we can flush
1218 * it into the system journal */
1220 r = journal_file_open(fn, O_RDWR, 0640, NULL, &s->runtime_journal);
1228 log_error("Failed to open runtime journal: %s", strerror(-r));
1235 /* OK, we really need the runtime journal, so create
1236 * it if necessary. */
1238 (void) mkdir_parents(fn, 0755);
1239 r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
1243 log_error("Failed to open runtime journal: %s", strerror(-r));
1248 if (s->runtime_journal) {
1249 journal_default_metrics(&s->runtime_metrics, s->runtime_journal->fd);
1251 s->runtime_journal->metrics = s->runtime_metrics;
1252 s->runtime_journal->compress = s->compress;
1254 fix_perms(s->runtime_journal, 0);
1261 static int server_flush_to_var(Server *s) {
1262 char path[] = "/run/log/journal/xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";
1271 if (!s->runtime_journal)
1274 ts = now(CLOCK_MONOTONIC);
1275 if (s->var_available_timestamp + RECHECK_VAR_AVAILABLE_USEC > ts)
1278 s->var_available_timestamp = ts;
1280 system_journal_open(s);
1282 if (!s->system_journal)
1285 r = sd_id128_get_machine(&machine);
1287 log_error("Failed to get machine id: %s", strerror(-r));
1291 r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY);
1293 log_error("Failed to read runtime journal: %s", strerror(-r));
1297 SD_JOURNAL_FOREACH(j) {
1300 f = j->current_file;
1301 assert(f && f->current_offset > 0);
1303 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1305 log_error("Can't read entry: %s", strerror(-r));
1309 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1311 log_info("Allocation limit reached.");
1313 journal_file_post_change(s->system_journal);
1316 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1320 log_error("Can't write entry: %s", strerror(-r));
1326 journal_file_post_change(s->system_journal);
1328 journal_file_close(s->runtime_journal);
1329 s->runtime_journal = NULL;
1332 sd_id128_to_string(machine, path + 17);
1333 rm_rf(path, false, true, false);
1339 static void forward_syslog(Server *s, const void *buffer, size_t length, struct ucred *ucred, struct timeval *tv) {
1340 struct msghdr msghdr;
1342 struct cmsghdr *cmsg;
1344 struct cmsghdr cmsghdr;
1345 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1346 CMSG_SPACE(sizeof(struct timeval))];
1348 union sockaddr_union sa;
1355 iovec.iov_base = (void*) buffer;
1356 iovec.iov_len = length;
1357 msghdr.msg_iov = &iovec;
1358 msghdr.msg_iovlen = 1;
1361 sa.un.sun_family = AF_UNIX;
1362 strncpy(sa.un.sun_path, "/run/systemd/syslog", sizeof(sa.un.sun_path));
1363 msghdr.msg_name = &sa;
1364 msghdr.msg_namelen = offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path);
1367 msghdr.msg_control = &control;
1368 msghdr.msg_controllen = sizeof(control);
1370 cmsg = CMSG_FIRSTHDR(&msghdr);
1371 cmsg->cmsg_level = SOL_SOCKET;
1372 cmsg->cmsg_type = SCM_CREDENTIALS;
1373 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
1374 memcpy(CMSG_DATA(cmsg), ucred, sizeof(struct ucred));
1375 msghdr.msg_controllen = cmsg->cmsg_len;
1377 /* Forward the syslog message we received via /dev/log to
1378 * /run/systemd/syslog. Unfortunately we currently can't set
1379 * the SO_TIMESTAMP auxiliary data, and hence we don't. */
1381 if (sendmsg(s->syslog_fd, &msghdr, MSG_NOSIGNAL) >= 0)
1384 if (errno == ESRCH) {
1387 /* Hmm, presumably the sender process vanished
1388 * by now, so let's fix it as good as we
1393 memcpy(CMSG_DATA(cmsg), &u, sizeof(struct ucred));
1395 if (sendmsg(s->syslog_fd, &msghdr, MSG_NOSIGNAL) >= 0)
1399 log_debug("Failed to forward syslog message: %m");
1402 static int process_event(Server *s, struct epoll_event *ev) {
1405 if (ev->data.fd == s->signal_fd) {
1406 struct signalfd_siginfo sfsi;
1409 if (ev->events != EPOLLIN) {
1410 log_info("Got invalid event from epoll.");
1414 n = read(s->signal_fd, &sfsi, sizeof(sfsi));
1415 if (n != sizeof(sfsi)) {
1420 if (errno == EINTR || errno == EAGAIN)
1426 if (sfsi.ssi_signo == SIGUSR1) {
1427 server_flush_to_var(s);
1431 log_debug("Received SIG%s", signal_to_string(sfsi.ssi_signo));
1434 } else if (ev->data.fd == s->native_fd ||
1435 ev->data.fd == s->syslog_fd) {
1437 if (ev->events != EPOLLIN) {
1438 log_info("Got invalid event from epoll.");
1443 struct msghdr msghdr;
1445 struct ucred *ucred = NULL;
1446 struct timeval *tv = NULL;
1447 struct cmsghdr *cmsg;
1449 struct cmsghdr cmsghdr;
1450 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1451 CMSG_SPACE(sizeof(struct timeval))];
1456 if (ioctl(ev->data.fd, SIOCINQ, &v) < 0) {
1457 log_error("SIOCINQ failed: %m");
1464 if (s->buffer_size < (size_t) v) {
1468 l = MAX(LINE_MAX + (size_t) v, s->buffer_size * 2);
1469 b = realloc(s->buffer, l+1);
1472 log_error("Couldn't increase buffer.");
1481 iovec.iov_base = s->buffer;
1482 iovec.iov_len = s->buffer_size;
1486 msghdr.msg_iov = &iovec;
1487 msghdr.msg_iovlen = 1;
1488 msghdr.msg_control = &control;
1489 msghdr.msg_controllen = sizeof(control);
1491 n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT);
1494 if (errno == EINTR || errno == EAGAIN)
1497 log_error("recvmsg() failed: %m");
1501 for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
1503 if (cmsg->cmsg_level == SOL_SOCKET &&
1504 cmsg->cmsg_type == SCM_CREDENTIALS &&
1505 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
1506 ucred = (struct ucred*) CMSG_DATA(cmsg);
1507 else if (cmsg->cmsg_level == SOL_SOCKET &&
1508 cmsg->cmsg_type == SO_TIMESTAMP &&
1509 cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval)))
1510 tv = (struct timeval*) CMSG_DATA(cmsg);
1513 if (ev->data.fd == s->syslog_fd) {
1516 e = memchr(s->buffer, '\n', n);
1522 forward_syslog(s, s->buffer, n, ucred, tv);
1523 process_syslog_message(s, strstrip(s->buffer), ucred, tv);
1525 process_native_message(s, s->buffer, n, ucred, tv);
1530 } else if (ev->data.fd == s->stdout_fd) {
1532 if (ev->events != EPOLLIN) {
1533 log_info("Got invalid event from epoll.");
1537 stdout_stream_new(s);
1541 StdoutStream *stream;
1543 if ((ev->events|EPOLLIN|EPOLLHUP) != (EPOLLIN|EPOLLHUP)) {
1544 log_info("Got invalid event from epoll.");
1548 /* If it is none of the well-known fds, it must be an
1549 * stdout stream fd. Note that this is a bit ugly here
1550 * (since we rely that none of the well-known fds
1551 * could be interpreted as pointer), but nonetheless
1552 * safe, since the well-known fds would never get an
1553 * fd > 4096, i.e. beyond the first memory page */
1555 stream = ev->data.ptr;
1557 if (stdout_stream_process(stream) <= 0)
1558 stdout_stream_free(stream);
1563 log_error("Unknown event.");
1567 static int open_syslog_socket(Server *s) {
1568 union sockaddr_union sa;
1570 struct epoll_event ev;
1575 if (s->syslog_fd < 0) {
1577 s->syslog_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1578 if (s->syslog_fd < 0) {
1579 log_error("socket() failed: %m");
1584 sa.un.sun_family = AF_UNIX;
1585 strncpy(sa.un.sun_path, "/dev/log", sizeof(sa.un.sun_path));
1587 unlink(sa.un.sun_path);
1589 r = bind(s->syslog_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
1591 log_error("bind() failed: %m");
1595 chmod(sa.un.sun_path, 0666);
1599 r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
1601 log_error("SO_PASSCRED failed: %m");
1606 r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_TIMESTAMP, &one, sizeof(one));
1608 log_error("SO_TIMESTAMP failed: %m");
1612 /* Since we use the same socket for forwarding this to some
1613 * other syslog implementation, make sure we don't hang
1615 timeval_store(&tv, SYSLOG_TIMEOUT_USEC);
1616 if (setsockopt(s->syslog_fd, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv)) < 0) {
1617 log_error("SO_SNDTIMEO failed: %m");
1622 ev.events = EPOLLIN;
1623 ev.data.fd = s->syslog_fd;
1624 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->syslog_fd, &ev) < 0) {
1625 log_error("Failed to add syslog server fd to epoll object: %m");
1632 static int open_native_socket(Server*s) {
1633 union sockaddr_union sa;
1635 struct epoll_event ev;
1639 if (s->native_fd < 0) {
1641 s->native_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1642 if (s->native_fd < 0) {
1643 log_error("socket() failed: %m");
1648 sa.un.sun_family = AF_UNIX;
1649 strncpy(sa.un.sun_path, "/run/systemd/journal", sizeof(sa.un.sun_path));
1651 unlink(sa.un.sun_path);
1653 r = bind(s->native_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
1655 log_error("bind() failed: %m");
1659 chmod(sa.un.sun_path, 0666);
1663 r = setsockopt(s->native_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
1665 log_error("SO_PASSCRED failed: %m");
1670 r = setsockopt(s->native_fd, SOL_SOCKET, SO_TIMESTAMP, &one, sizeof(one));
1672 log_error("SO_TIMESTAMP failed: %m");
1677 ev.events = EPOLLIN;
1678 ev.data.fd = s->native_fd;
1679 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->native_fd, &ev) < 0) {
1680 log_error("Failed to add native server fd to epoll object: %m");
1687 static int open_stdout_socket(Server *s) {
1688 union sockaddr_union sa;
1690 struct epoll_event ev;
1694 if (s->stdout_fd < 0) {
1696 s->stdout_fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC, 0);
1697 if (s->stdout_fd < 0) {
1698 log_error("socket() failed: %m");
1703 sa.un.sun_family = AF_UNIX;
1704 strncpy(sa.un.sun_path, "/run/systemd/stdout", sizeof(sa.un.sun_path));
1706 unlink(sa.un.sun_path);
1708 r = bind(s->stdout_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
1710 log_error("bind() failed: %m");
1714 chmod(sa.un.sun_path, 0666);
1716 if (listen(s->stdout_fd, SOMAXCONN) < 0) {
1717 log_error("liste() failed: %m");
1723 ev.events = EPOLLIN;
1724 ev.data.fd = s->stdout_fd;
1725 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->stdout_fd, &ev) < 0) {
1726 log_error("Failed to add stdout server fd to epoll object: %m");
1733 static int open_signalfd(Server *s) {
1735 struct epoll_event ev;
1739 assert_se(sigemptyset(&mask) == 0);
1740 sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, -1);
1741 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1743 s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1744 if (s->signal_fd < 0) {
1745 log_error("signalfd(): %m");
1750 ev.events = EPOLLIN;
1751 ev.data.fd = s->signal_fd;
1753 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->signal_fd, &ev) < 0) {
1754 log_error("epoll_ctl(): %m");
1761 static int server_init(Server *s) {
1767 s->syslog_fd = s->native_fd = s->stdout_fd = s->signal_fd = s->epoll_fd = -1;
1770 memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics));
1771 memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics));
1773 s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
1774 if (!s->user_journals) {
1775 log_error("Out of memory.");
1779 s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
1780 if (s->epoll_fd < 0) {
1781 log_error("Failed to create epoll object: %m");
1785 n = sd_listen_fds(true);
1787 log_error("Failed to read listening file descriptors from environment: %s", strerror(-n));
1791 for (fd = SD_LISTEN_FDS_START; fd < SD_LISTEN_FDS_START + n; fd++) {
1793 if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/run/systemd/native", 0) > 0) {
1795 if (s->native_fd >= 0) {
1796 log_error("Too many native sockets passed.");
1802 } else if (sd_is_socket_unix(fd, SOCK_STREAM, 1, "/run/systemd/stdout", 0) > 0) {
1804 if (s->stdout_fd >= 0) {
1805 log_error("Too many stdout sockets passed.");
1811 } else if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/dev/log", 0) > 0) {
1813 if (s->syslog_fd >= 0) {
1814 log_error("Too many /dev/log sockets passed.");
1821 log_error("Unknown socket passed.");
1826 r = open_syslog_socket(s);
1830 r = open_native_socket(s);
1834 r = open_stdout_socket(s);
1838 r = system_journal_open(s);
1842 r = open_signalfd(s);
1846 s->rate_limit = journal_rate_limit_new(DEFAULT_RATE_LIMIT_INTERVAL, DEFAULT_RATE_LIMIT_BURST);
1853 static void server_done(Server *s) {
1857 while (s->stdout_streams)
1858 stdout_stream_free(s->stdout_streams);
1860 if (s->system_journal)
1861 journal_file_close(s->system_journal);
1863 if (s->runtime_journal)
1864 journal_file_close(s->runtime_journal);
1866 while ((f = hashmap_steal_first(s->user_journals)))
1867 journal_file_close(f);
1869 hashmap_free(s->user_journals);
1871 if (s->epoll_fd >= 0)
1872 close_nointr_nofail(s->epoll_fd);
1874 if (s->signal_fd >= 0)
1875 close_nointr_nofail(s->signal_fd);
1877 if (s->syslog_fd >= 0)
1878 close_nointr_nofail(s->syslog_fd);
1880 if (s->native_fd >= 0)
1881 close_nointr_nofail(s->native_fd);
1883 if (s->stdout_fd >= 0)
1884 close_nointr_nofail(s->stdout_fd);
1887 journal_rate_limit_free(s->rate_limit);
1892 int main(int argc, char *argv[]) {
1896 /* if (getppid() != 1) { */
1897 /* log_error("This program should be invoked by init only."); */
1898 /* return EXIT_FAILURE; */
1902 log_error("This program does not take arguments.");
1903 return EXIT_FAILURE;
1906 log_set_target(LOG_TARGET_CONSOLE);
1907 log_parse_environment();
1912 r = server_init(&server);
1916 log_debug("systemd-journald running as pid %lu", (unsigned long) getpid());
1920 "STATUS=Processing requests...");
1922 server_vacuum(&server);
1923 server_flush_to_var(&server);
1926 struct epoll_event event;
1928 r = epoll_wait(server.epoll_fd, &event, 1, -1);
1934 log_error("epoll_wait() failed: %m");
1940 r = process_event(&server, &event);
1947 log_debug("systemd-journald stopped as pid %lu", (unsigned long) getpid());
1951 "STATUS=Shutting down...");
1953 server_done(&server);
1955 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;