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"
44 #include "journal-internal.h"
46 #define USER_JOURNALS_MAX 1024
47 #define STDOUT_STREAMS_MAX 4096
49 #define DEFAULT_RATE_LIMIT_INTERVAL (10*USEC_PER_SEC)
50 #define DEFAULT_RATE_LIMIT_BURST 200
52 #define RECHECK_AVAILABLE_SPACE_USEC (30*USEC_PER_SEC)
54 #define RECHECK_VAR_AVAILABLE_USEC (30*USEC_PER_SEC)
56 #define SYSLOG_TIMEOUT_USEC (5*USEC_PER_SEC)
58 typedef struct StdoutStream StdoutStream;
60 typedef struct Server {
67 JournalFile *runtime_journal;
68 JournalFile *system_journal;
69 Hashmap *user_journals;
76 JournalRateLimit *rate_limit;
78 JournalMetrics metrics;
82 uint64_t cached_available_space;
83 usec_t cached_available_space_timestamp;
85 uint64_t var_available_timestamp;
87 LIST_HEAD(StdoutStream, stdout_streams);
88 unsigned n_stdout_streams;
91 typedef enum StdoutStreamState {
93 STDOUT_STREAM_PRIORITY,
94 STDOUT_STREAM_PRIORITY_PREFIX,
95 STDOUT_STREAM_TEE_CONSOLE,
101 StdoutStreamState state;
109 bool priority_prefix:1;
112 char buffer[LINE_MAX+1];
115 LIST_FIELDS(StdoutStream, stdout_stream);
118 static int server_flush_to_var(Server *s);
120 static uint64_t available_space(Server *s) {
126 uint64_t sum = 0, avail = 0, ss_avail = 0;
129 usec_t ts = now(CLOCK_MONOTONIC);
131 if (s->cached_available_space_timestamp + RECHECK_AVAILABLE_SPACE_USEC > ts)
132 return s->cached_available_space;
134 r = sd_id128_get_machine(&machine);
138 if (s->system_journal)
139 f = "/var/log/journal/";
141 f = "/run/log/journal/";
143 p = strappend(f, sd_id128_to_string(machine, ids));
153 if (fstatvfs(dirfd(d), &ss) < 0)
158 struct dirent buf, *de;
161 k = readdir_r(d, &buf, &de);
170 if (!dirent_is_file_with_suffix(de, ".journal"))
173 if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
176 sum += (uint64_t) st.st_blocks * (uint64_t) st.st_blksize;
179 avail = sum >= s->max_use ? 0 : s->max_use - sum;
181 ss_avail = ss.f_bsize * ss.f_bavail;
183 ss_avail = ss_avail < s->metrics.keep_free ? 0 : ss_avail - s->metrics.keep_free;
185 if (ss_avail < avail)
188 s->cached_available_space = avail;
189 s->cached_available_space_timestamp = ts;
197 static void fix_perms(JournalFile *f, uid_t uid) {
200 acl_permset_t permset;
205 r = fchmod_and_fchown(f->fd, 0640, 0, 0);
207 log_warning("Failed to fix access mode/rights on %s, ignoring: %s", f->path, strerror(-r));
212 acl = acl_get_fd(f->fd);
214 log_warning("Failed to read ACL on %s, ignoring: %m", f->path);
218 r = acl_find_uid(acl, uid, &entry);
221 if (acl_create_entry(&acl, &entry) < 0 ||
222 acl_set_tag_type(entry, ACL_USER) < 0 ||
223 acl_set_qualifier(entry, &uid) < 0) {
224 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
229 if (acl_get_permset(entry, &permset) < 0 ||
230 acl_add_perm(permset, ACL_READ) < 0 ||
231 acl_calc_mask(&acl) < 0) {
232 log_warning("Failed to patch ACL on %s, ignoring: %m", f->path);
236 if (acl_set_fd(f->fd, acl) < 0)
237 log_warning("Failed to set ACL on %s, ignoring: %m", f->path);
243 static JournalFile* find_journal(Server *s, uid_t uid) {
252 /* We split up user logs only on /var, not on /run. If the
253 * runtime file is open, we write to it exclusively, in order
254 * to guarantee proper order as soon as we flush /run to
255 * /var and close the runtime file. */
257 if (s->runtime_journal)
258 return s->runtime_journal;
261 return s->system_journal;
263 r = sd_id128_get_machine(&machine);
265 return s->system_journal;
267 f = hashmap_get(s->user_journals, UINT32_TO_PTR(uid));
271 if (asprintf(&p, "/var/log/journal/%s/user-%lu.journal", sd_id128_to_string(machine, ids), (unsigned long) uid) < 0)
272 return s->system_journal;
274 while (hashmap_size(s->user_journals) >= USER_JOURNALS_MAX) {
275 /* Too many open? Then let's close one */
276 f = hashmap_steal_first(s->user_journals);
278 journal_file_close(f);
281 r = journal_file_open(p, O_RDWR|O_CREAT, 0640, s->system_journal, &f);
285 return s->system_journal;
288 f->metrics = s->metrics;
289 f->compress = s->compress;
291 r = hashmap_put(s->user_journals, UINT32_TO_PTR(uid), f);
293 journal_file_close(f);
294 return s->system_journal;
300 static void server_vacuum(Server *s) {
309 log_info("Rotating...");
311 if (s->runtime_journal) {
312 r = journal_file_rotate(&s->runtime_journal);
314 log_error("Failed to rotate %s: %s", s->runtime_journal->path, strerror(-r));
317 if (s->system_journal) {
318 r = journal_file_rotate(&s->system_journal);
320 log_error("Failed to rotate %s: %s", s->system_journal->path, strerror(-r));
323 HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
324 r = journal_file_rotate(&f);
326 log_error("Failed to rotate %s: %s", f->path, strerror(-r));
328 hashmap_replace(s->user_journals, k, f);
331 log_info("Vacuuming...");
333 r = sd_id128_get_machine(&machine);
335 log_error("Failed to get machine ID: %s", strerror(-r));
339 if (asprintf(&p, "/var/log/journal/%s", sd_id128_to_string(machine, ids)) < 0) {
340 log_error("Out of memory.");
344 r = journal_directory_vacuum(p, s->max_use, s->metrics.keep_free);
345 if (r < 0 && r != -ENOENT)
346 log_error("Failed to vacuum %s: %s", p, strerror(-r));
349 if (asprintf(&p, "/run/log/journal/%s", ids) < 0) {
350 log_error("Out of memory.");
354 r = journal_directory_vacuum(p, s->max_use, s->metrics.keep_free);
355 if (r < 0 && r != -ENOENT)
356 log_error("Failed to vacuum %s: %s", p, strerror(-r));
359 s->cached_available_space_timestamp = 0;
362 static char *shortened_cgroup_path(pid_t pid) {
364 char *process_path, *init_path, *path;
368 r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &process_path);
372 r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 1, &init_path);
378 if (streq(init_path, "/"))
381 if (startswith(process_path, init_path))
382 path = process_path + strlen(init_path);
391 static void dispatch_message_real(Server *s,
392 struct iovec *iovec, unsigned n, unsigned m,
394 struct timeval *tv) {
396 char *pid = NULL, *uid = NULL, *gid = NULL,
397 *source_time = NULL, *boot_id = NULL, *machine_id = NULL,
398 *comm = NULL, *cmdline = NULL, *hostname = NULL,
399 *audit_session = NULL, *audit_loginuid = NULL,
400 *exe = NULL, *cgroup = NULL;
406 uid_t loginuid = 0, realuid = 0;
408 bool vacuumed = false;
419 realuid = ucred->uid;
421 if (asprintf(&pid, "_PID=%lu", (unsigned long) ucred->pid) >= 0)
422 IOVEC_SET_STRING(iovec[n++], pid);
424 if (asprintf(&uid, "_UID=%lu", (unsigned long) ucred->uid) >= 0)
425 IOVEC_SET_STRING(iovec[n++], uid);
427 if (asprintf(&gid, "_GID=%lu", (unsigned long) ucred->gid) >= 0)
428 IOVEC_SET_STRING(iovec[n++], gid);
430 r = get_process_comm(ucred->pid, &t);
432 comm = strappend("_COMM=", t);
434 IOVEC_SET_STRING(iovec[n++], comm);
438 r = get_process_exe(ucred->pid, &t);
440 exe = strappend("_EXE=", t);
442 IOVEC_SET_STRING(iovec[n++], exe);
446 r = get_process_cmdline(ucred->pid, LINE_MAX, false, &t);
448 cmdline = strappend("_CMDLINE=", t);
450 IOVEC_SET_STRING(iovec[n++], cmdline);
454 r = audit_session_from_pid(ucred->pid, &session);
456 if (asprintf(&audit_session, "_AUDIT_SESSION=%lu", (unsigned long) session) >= 0)
457 IOVEC_SET_STRING(iovec[n++], audit_session);
459 r = audit_loginuid_from_pid(ucred->pid, &loginuid);
461 if (asprintf(&audit_loginuid, "_AUDIT_LOGINUID=%lu", (unsigned long) loginuid) >= 0)
462 IOVEC_SET_STRING(iovec[n++], audit_loginuid);
464 path = shortened_cgroup_path(ucred->pid);
466 cgroup = strappend("_SYSTEMD_CGROUP=", path);
468 IOVEC_SET_STRING(iovec[n++], cgroup);
475 if (asprintf(&source_time, "_SOURCE_REALTIME_TIMESTAMP=%llu",
476 (unsigned long long) timeval_load(tv)) >= 0)
477 IOVEC_SET_STRING(iovec[n++], source_time);
480 /* Note that strictly speaking storing the boot id here is
481 * redundant since the entry includes this in-line
482 * anyway. However, we need this indexed, too. */
483 r = sd_id128_get_boot(&id);
485 if (asprintf(&boot_id, "_BOOT_ID=%s", sd_id128_to_string(id, idbuf)) >= 0)
486 IOVEC_SET_STRING(iovec[n++], boot_id);
488 r = sd_id128_get_machine(&id);
490 if (asprintf(&machine_id, "_MACHINE_ID=%s", sd_id128_to_string(id, idbuf)) >= 0)
491 IOVEC_SET_STRING(iovec[n++], machine_id);
493 t = gethostname_malloc();
495 hostname = strappend("_HOSTNAME=", t);
497 IOVEC_SET_STRING(iovec[n++], hostname);
503 server_flush_to_var(s);
506 f = find_journal(s, realuid == 0 ? 0 : loginuid);
508 log_warning("Dropping message, as we can't find a place to store the data.");
510 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
512 if (r == -E2BIG && !vacuumed) {
513 log_info("Allocation limit reached.");
518 log_info("Retrying write.");
523 log_error("Failed to write entry, ignoring: %s", strerror(-r));
537 free(audit_loginuid);
541 static void dispatch_message(Server *s,
542 struct iovec *iovec, unsigned n, unsigned m,
550 assert(iovec || n == 0);
558 path = shortened_cgroup_path(ucred->pid);
562 /* example: /user/lennart/3/foobar
563 * /system/dbus.service/foobar
565 * So let's cut of everything past the third /, since that is
566 * wher user directories start */
568 c = strchr(path, '/');
570 c = strchr(c+1, '/');
572 c = strchr(c+1, '/');
578 rl = journal_rate_limit_test(s->rate_limit, path, priority, available_space(s));
587 char suppress_message[LINE_MAX];
588 struct iovec suppress_iovec[15];
590 /* Write a suppression message if we suppressed something */
592 snprintf(suppress_message, sizeof(suppress_message), "MESSAGE=Suppressed %u messages from %s", rl - 1, path);
593 char_array_0(suppress_message);
595 IOVEC_SET_STRING(suppress_iovec[j++], "PRIORITY=5");
596 IOVEC_SET_STRING(suppress_iovec[j++], suppress_message);
598 dispatch_message_real(s, suppress_iovec, j, ELEMENTSOF(suppress_iovec), NULL, NULL);
604 dispatch_message_real(s, iovec, n, m, ucred, tv);
607 static void process_syslog_message(Server *s, const char *buf, struct ucred *ucred, struct timeval *tv) {
608 char *message = NULL, *syslog_priority = NULL, *syslog_facility = NULL;
609 struct iovec iovec[16];
611 int priority = LOG_USER | LOG_INFO;
616 parse_syslog_priority((char**) &buf, &priority);
617 skip_syslog_date((char**) &buf);
619 if (asprintf(&syslog_priority, "PRIORITY=%i", priority & LOG_PRIMASK) >= 0)
620 IOVEC_SET_STRING(iovec[n++], syslog_priority);
622 if (asprintf(&syslog_facility, "SYSLOG_FACILITY=%i", LOG_FAC(priority)) >= 0)
623 IOVEC_SET_STRING(iovec[n++], syslog_facility);
625 message = strappend("MESSAGE=", buf);
627 IOVEC_SET_STRING(iovec[n++], message);
629 dispatch_message(s, iovec, n, ELEMENTSOF(iovec), ucred, tv, priority & LOG_PRIMASK);
632 free(syslog_facility);
633 free(syslog_priority);
636 static bool valid_user_field(const char *p, size_t l) {
639 /* We kinda enforce POSIX syntax recommendations for
640 environment variables here, but make a couple of additional
643 http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
645 /* No empty field names */
649 /* Don't allow names longer than 64 chars */
653 /* Variables starting with an underscore are protected */
657 /* Don't allow digits as first character */
658 if (p[0] >= '0' && p[0] <= '9')
661 /* Only allow A-Z0-9 and '_' */
662 for (a = p; a < p + l; a++)
663 if (!((*a >= 'A' && *a <= 'Z') ||
664 (*a >= '0' && *a <= '9') ||
671 static void process_native_message(Server *s, const void *buffer, size_t buffer_size, struct ucred *ucred, struct timeval *tv) {
672 struct iovec *iovec = NULL;
673 unsigned n = 0, m = 0, j;
676 int priority = LOG_INFO;
679 assert(buffer || n == 0);
682 remaining = buffer_size;
684 while (remaining > 0) {
687 e = memchr(p, '\n', remaining);
690 /* Trailing noise, let's ignore it, and flush what we collected */
691 log_debug("Received message with trailing noise, ignoring.");
696 /* Entry separator */
697 dispatch_message(s, iovec, n, m, ucred, tv, priority);
706 if (*p == '.' || *p == '#') {
707 /* Ignore control commands for now, and
709 remaining -= (e - p) + 1;
714 /* A property follows */
720 u = MAX((n+13U) * 2U, 4U);
721 c = realloc(iovec, u * sizeof(struct iovec));
723 log_error("Out of memory");
731 q = memchr(p, '=', e - p);
733 if (valid_user_field(p, q - p)) {
734 /* If the field name starts with an
735 * underscore, skip the variable,
736 * since that indidates a trusted
738 iovec[n].iov_base = (char*) p;
739 iovec[n].iov_len = e - p;
742 /* We need to determine the priority
743 * of this entry for the rate limiting
746 memcmp(p, "PRIORITY=", 10) == 0 &&
749 priority = p[10] - '0';
752 remaining -= (e - p) + 1;
759 if (remaining < e - p + 1 + sizeof(uint64_t) + 1) {
760 log_debug("Failed to parse message, ignoring.");
764 memcpy(&l, e + 1, sizeof(uint64_t));
767 if (remaining < e - p + 1 + sizeof(uint64_t) + l + 1 ||
768 e[1+sizeof(uint64_t)+l] != '\n') {
769 log_debug("Failed to parse message, ignoring.");
773 k = malloc((e - p) + 1 + l);
775 log_error("Out of memory");
781 memcpy(k + (e - p) + 1, e + 1 + sizeof(uint64_t), l);
783 if (valid_user_field(p, e - p)) {
784 iovec[n].iov_base = k;
785 iovec[n].iov_len = (e - p) + 1 + l;
790 remaining -= (e - p) + 1 + sizeof(uint64_t) + l + 1;
791 p = e + 1 + sizeof(uint64_t) + l + 1;
795 dispatch_message(s, iovec, n, m, ucred, tv, priority);
797 for (j = 0; j < n; j++)
798 if (iovec[j].iov_base < buffer ||
799 (const uint8_t*) iovec[j].iov_base >= (const uint8_t*) buffer + buffer_size)
800 free(iovec[j].iov_base);
803 static int stdout_stream_log(StdoutStream *s, const char *p, size_t l) {
804 struct iovec iovec[15];
805 char *message = NULL, *syslog_priority = NULL;
813 priority = s->priority;
815 if (s->priority_prefix &&
818 p[1] >= '0' && p[1] <= '7' &&
821 priority = p[1] - '0';
829 if (asprintf(&syslog_priority, "PRIORITY=%i", priority) >= 0)
830 IOVEC_SET_STRING(iovec[n++], syslog_priority);
832 tag_len = s->tag ? strlen(s->tag) + 2: 0;
833 message = malloc(8 + tag_len + l);
835 memcpy(message, "MESSAGE=", 8);
838 memcpy(message+8, s->tag, tag_len-2);
839 memcpy(message+8+tag_len-2, ": ", 2);
842 memcpy(message+8+tag_len, p, l);
843 iovec[n].iov_base = message;
844 iovec[n].iov_len = 8+tag_len+l;
848 dispatch_message(s->server, iovec, n, ELEMENTSOF(iovec), &s->ucred, NULL, priority);
850 if (s->tee_console) {
853 console = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
857 IOVEC_SET_STRING(iovec[n++], s->tag);
858 IOVEC_SET_STRING(iovec[n++], ": ");
861 iovec[n].iov_base = (void*) p;
862 iovec[n].iov_len = l;
865 IOVEC_SET_STRING(iovec[n++], (char*) "\n");
867 writev(console, iovec, n);
872 free(syslog_priority);
877 static int stdout_stream_line(StdoutStream *s, const char *p, size_t l) {
881 while (l > 0 && strchr(WHITESPACE, *p)) {
886 while (l > 0 && strchr(WHITESPACE, *(p+l-1)))
891 case STDOUT_STREAM_TAG:
894 s->tag = strndup(p, l);
896 log_error("Out of memory");
901 s->state = STDOUT_STREAM_PRIORITY;
904 case STDOUT_STREAM_PRIORITY:
905 if (l != 1 || *p < '0' || *p > '7') {
906 log_warning("Failed to parse log priority line.");
910 s->priority = *p - '0';
911 s->state = STDOUT_STREAM_PRIORITY_PREFIX;
914 case STDOUT_STREAM_PRIORITY_PREFIX:
915 if (l != 1 || *p < '0' || *p > '1') {
916 log_warning("Failed to parse priority prefix line.");
920 s->priority_prefix = *p - '0';
921 s->state = STDOUT_STREAM_TEE_CONSOLE;
924 case STDOUT_STREAM_TEE_CONSOLE:
925 if (l != 1 || *p < '0' || *p > '1') {
926 log_warning("Failed to parse tee to console line.");
930 s->tee_console = *p - '0';
931 s->state = STDOUT_STREAM_RUNNING;
934 case STDOUT_STREAM_RUNNING:
935 return stdout_stream_log(s, p, l);
938 assert_not_reached("Unknown stream state");
941 static int stdout_stream_scan(StdoutStream *s, bool force_flush) {
949 remaining = s->length;
954 end = memchr(p, '\n', remaining);
956 if (remaining >= LINE_MAX) {
964 r = stdout_stream_line(s, p, end - p);
972 if (force_flush && remaining > 0) {
973 r = stdout_stream_line(s, p, remaining);
982 memmove(s->buffer, p, remaining);
983 s->length = remaining;
989 static int stdout_stream_process(StdoutStream *s) {
995 l = read(s->fd, s->buffer+s->length, sizeof(s->buffer)-1-s->length);
1001 log_warning("Failed to read from stream: %m");
1006 r = stdout_stream_scan(s, true);
1014 r = stdout_stream_scan(s, false);
1022 static void stdout_stream_free(StdoutStream *s) {
1026 assert(s->server->n_stdout_streams > 0);
1027 s->server->n_stdout_streams --;
1028 LIST_REMOVE(StdoutStream, stdout_stream, s->server->stdout_streams, s);
1033 epoll_ctl(s->server->epoll_fd, EPOLL_CTL_DEL, s->fd, NULL);
1035 close_nointr_nofail(s->fd);
1042 static int stdout_stream_new(Server *s) {
1043 StdoutStream *stream;
1046 struct epoll_event ev;
1050 fd = accept4(s->stdout_fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC);
1052 if (errno == EAGAIN)
1055 log_error("Failed to accept stdout connection: %m");
1059 if (s->n_stdout_streams >= STDOUT_STREAMS_MAX) {
1060 log_warning("Too many stdout streams, refusing connection.");
1061 close_nointr_nofail(fd);
1065 stream = new0(StdoutStream, 1);
1067 log_error("Out of memory.");
1068 close_nointr_nofail(fd);
1074 len = sizeof(stream->ucred);
1075 if (getsockopt(fd, SOL_SOCKET, SO_PEERCRED, &stream->ucred, &len) < 0) {
1076 log_error("Failed to determine peer credentials: %m");
1081 if (shutdown(fd, SHUT_WR) < 0) {
1082 log_error("Failed to shutdown writing side of socket: %m");
1088 ev.data.ptr = stream;
1089 ev.events = EPOLLIN;
1090 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) {
1091 log_error("Failed to add stream to event loop: %m");
1097 LIST_PREPEND(StdoutStream, stdout_stream, s->stdout_streams, stream);
1098 s->n_stdout_streams ++;
1103 stdout_stream_free(stream);
1107 static int system_journal_open(Server *s) {
1113 r = sd_id128_get_machine(&machine);
1117 sd_id128_to_string(machine, ids);
1119 if (!s->system_journal) {
1121 /* First try to create the machine path, but not the prefix */
1122 fn = strappend("/var/log/journal/", ids);
1125 (void) mkdir(fn, 0755);
1128 /* The create the system journal file */
1129 fn = join("/var/log/journal/", ids, "/system.journal", NULL);
1133 r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->system_journal);
1137 s->system_journal->metrics = s->metrics;
1138 s->system_journal->compress = s->compress;
1140 fix_perms(s->system_journal, 0);
1146 log_error("Failed to open system journal: %s", strerror(-r));
1152 if (!s->runtime_journal) {
1154 fn = join("/run/log/journal/", ids, "/system.journal", NULL);
1158 if (s->system_journal) {
1160 /* Try to open the runtime journal, but only
1161 * if it already exists, so that we can flush
1162 * it into the system journal */
1164 r = journal_file_open(fn, O_RDWR, 0640, NULL, &s->runtime_journal);
1172 log_error("Failed to open runtime journal: %s", strerror(-r));
1179 /* OK, we really need the runtime journal, so create
1180 * it if necessary. */
1182 (void) mkdir_parents(fn, 0755);
1183 r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
1187 log_error("Failed to open runtime journal: %s", strerror(-r));
1192 if (s->runtime_journal) {
1193 s->runtime_journal->metrics = s->metrics;
1194 s->runtime_journal->compress = s->compress;
1196 fix_perms(s->runtime_journal, 0);
1203 static int server_flush_to_var(Server *s) {
1204 char path[] = "/run/log/journal/xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";
1213 if (!s->runtime_journal)
1216 ts = now(CLOCK_MONOTONIC);
1217 if (s->var_available_timestamp + RECHECK_VAR_AVAILABLE_USEC > ts)
1220 s->var_available_timestamp = ts;
1222 system_journal_open(s);
1224 if (!s->system_journal)
1227 r = sd_id128_get_machine(&machine);
1229 log_error("Failed to get machine id: %s", strerror(-r));
1233 r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY);
1235 log_error("Failed to read runtime journal: %s", strerror(-r));
1239 SD_JOURNAL_FOREACH(j) {
1242 f = j->current_file;
1243 assert(f && f->current_offset > 0);
1245 r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
1247 log_error("Can't read entry: %s", strerror(-r));
1251 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1253 log_info("Allocation limit reached.");
1255 journal_file_post_change(s->system_journal);
1258 r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
1262 log_error("Can't write entry: %s", strerror(-r));
1268 journal_file_post_change(s->system_journal);
1270 journal_file_close(s->runtime_journal);
1271 s->runtime_journal = NULL;
1274 sd_id128_to_string(machine, path + 17);
1275 rm_rf(path, false, true, false);
1281 static void forward_syslog(Server *s, const void *buffer, size_t length, struct ucred *ucred, struct timeval *tv) {
1282 struct msghdr msghdr;
1284 struct cmsghdr *cmsg;
1286 struct cmsghdr cmsghdr;
1287 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1288 CMSG_SPACE(sizeof(struct timeval))];
1290 union sockaddr_union sa;
1297 iovec.iov_base = (void*) buffer;
1298 iovec.iov_len = length;
1299 msghdr.msg_iov = &iovec;
1300 msghdr.msg_iovlen = 1;
1303 sa.un.sun_family = AF_UNIX;
1304 strncpy(sa.un.sun_path, "/run/systemd/syslog", sizeof(sa.un.sun_path));
1305 msghdr.msg_name = &sa;
1306 msghdr.msg_namelen = offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path);
1309 msghdr.msg_control = &control;
1310 msghdr.msg_controllen = sizeof(control);
1312 cmsg = CMSG_FIRSTHDR(&msghdr);
1313 cmsg->cmsg_level = SOL_SOCKET;
1314 cmsg->cmsg_type = SCM_CREDENTIALS;
1315 cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
1316 memcpy(CMSG_DATA(cmsg), ucred, sizeof(struct ucred));
1317 msghdr.msg_controllen = cmsg->cmsg_len;
1319 /* Forward the syslog message we received via /dev/log to
1320 * /run/systemd/syslog. Unfortunately we currently can't set
1321 * the SO_TIMESTAMP auxiliary data, and hence we don't. */
1323 if (sendmsg(s->syslog_fd, &msghdr, MSG_NOSIGNAL) >= 0)
1326 if (errno == ESRCH) {
1329 /* Hmm, presumably the sender process vanished
1330 * by now, so let's fix it as good as we
1335 memcpy(CMSG_DATA(cmsg), &u, sizeof(struct ucred));
1337 if (sendmsg(s->syslog_fd, &msghdr, MSG_NOSIGNAL) >= 0)
1341 log_debug("Failed to forward syslog message: %m");
1344 static int process_event(Server *s, struct epoll_event *ev) {
1347 if (ev->data.fd == s->signal_fd) {
1348 struct signalfd_siginfo sfsi;
1351 if (ev->events != EPOLLIN) {
1352 log_info("Got invalid event from epoll.");
1356 n = read(s->signal_fd, &sfsi, sizeof(sfsi));
1357 if (n != sizeof(sfsi)) {
1362 if (errno == EINTR || errno == EAGAIN)
1368 if (sfsi.ssi_signo == SIGUSR1) {
1369 server_flush_to_var(s);
1373 log_debug("Received SIG%s", signal_to_string(sfsi.ssi_signo));
1376 } else if (ev->data.fd == s->native_fd ||
1377 ev->data.fd == s->syslog_fd) {
1379 if (ev->events != EPOLLIN) {
1380 log_info("Got invalid event from epoll.");
1385 struct msghdr msghdr;
1387 struct ucred *ucred = NULL;
1388 struct timeval *tv = NULL;
1389 struct cmsghdr *cmsg;
1391 struct cmsghdr cmsghdr;
1392 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1393 CMSG_SPACE(sizeof(struct timeval))];
1398 if (ioctl(ev->data.fd, SIOCINQ, &v) < 0) {
1399 log_error("SIOCINQ failed: %m");
1406 if (s->buffer_size < (size_t) v) {
1410 l = MAX(LINE_MAX + (size_t) v, s->buffer_size * 2);
1411 b = realloc(s->buffer, l+1);
1414 log_error("Couldn't increase buffer.");
1423 iovec.iov_base = s->buffer;
1424 iovec.iov_len = s->buffer_size;
1428 msghdr.msg_iov = &iovec;
1429 msghdr.msg_iovlen = 1;
1430 msghdr.msg_control = &control;
1431 msghdr.msg_controllen = sizeof(control);
1433 n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT);
1436 if (errno == EINTR || errno == EAGAIN)
1439 log_error("recvmsg() failed: %m");
1443 for (cmsg = CMSG_FIRSTHDR(&msghdr); cmsg; cmsg = CMSG_NXTHDR(&msghdr, cmsg)) {
1445 if (cmsg->cmsg_level == SOL_SOCKET &&
1446 cmsg->cmsg_type == SCM_CREDENTIALS &&
1447 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
1448 ucred = (struct ucred*) CMSG_DATA(cmsg);
1449 else if (cmsg->cmsg_level == SOL_SOCKET &&
1450 cmsg->cmsg_type == SO_TIMESTAMP &&
1451 cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval)))
1452 tv = (struct timeval*) CMSG_DATA(cmsg);
1455 if (ev->data.fd == s->syslog_fd) {
1458 e = memchr(s->buffer, '\n', n);
1464 forward_syslog(s, s->buffer, n, ucred, tv);
1465 process_syslog_message(s, strstrip(s->buffer), ucred, tv);
1467 process_native_message(s, s->buffer, n, ucred, tv);
1472 } else if (ev->data.fd == s->stdout_fd) {
1474 if (ev->events != EPOLLIN) {
1475 log_info("Got invalid event from epoll.");
1479 stdout_stream_new(s);
1483 StdoutStream *stream;
1485 if ((ev->events|EPOLLIN|EPOLLHUP) != (EPOLLIN|EPOLLHUP)) {
1486 log_info("Got invalid event from epoll.");
1490 /* If it is none of the well-known fds, it must be an
1491 * stdout stream fd. Note that this is a bit ugly here
1492 * (since we rely that none of the well-known fds
1493 * could be interpreted as pointer), but nonetheless
1494 * safe, since the well-known fds would never get an
1495 * fd > 4096, i.e. beyond the first memory page */
1497 stream = ev->data.ptr;
1499 if (stdout_stream_process(stream) <= 0)
1500 stdout_stream_free(stream);
1505 log_error("Unknown event.");
1509 static int open_syslog_socket(Server *s) {
1510 union sockaddr_union sa;
1512 struct epoll_event ev;
1517 if (s->syslog_fd < 0) {
1519 s->syslog_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1520 if (s->syslog_fd < 0) {
1521 log_error("socket() failed: %m");
1526 sa.un.sun_family = AF_UNIX;
1527 strncpy(sa.un.sun_path, "/dev/log", sizeof(sa.un.sun_path));
1529 unlink(sa.un.sun_path);
1531 r = bind(s->syslog_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
1533 log_error("bind() failed: %m");
1537 chmod(sa.un.sun_path, 0666);
1541 r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
1543 log_error("SO_PASSCRED failed: %m");
1548 r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_TIMESTAMP, &one, sizeof(one));
1550 log_error("SO_TIMESTAMP failed: %m");
1554 /* Since we use the same socket for forwarding this to some
1555 * other syslog implementation, make sure we don't hang
1557 timeval_store(&tv, SYSLOG_TIMEOUT_USEC);
1558 if (setsockopt(s->syslog_fd, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv)) < 0) {
1559 log_error("SO_SNDTIMEO failed: %m");
1564 ev.events = EPOLLIN;
1565 ev.data.fd = s->syslog_fd;
1566 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->syslog_fd, &ev) < 0) {
1567 log_error("Failed to add syslog server fd to epoll object: %m");
1574 static int open_native_socket(Server*s) {
1575 union sockaddr_union sa;
1577 struct epoll_event ev;
1581 if (s->native_fd < 0) {
1583 s->native_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0);
1584 if (s->native_fd < 0) {
1585 log_error("socket() failed: %m");
1590 sa.un.sun_family = AF_UNIX;
1591 strncpy(sa.un.sun_path, "/run/systemd/journal", sizeof(sa.un.sun_path));
1593 unlink(sa.un.sun_path);
1595 r = bind(s->native_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
1597 log_error("bind() failed: %m");
1601 chmod(sa.un.sun_path, 0666);
1605 r = setsockopt(s->native_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
1607 log_error("SO_PASSCRED failed: %m");
1612 r = setsockopt(s->native_fd, SOL_SOCKET, SO_TIMESTAMP, &one, sizeof(one));
1614 log_error("SO_TIMESTAMP failed: %m");
1619 ev.events = EPOLLIN;
1620 ev.data.fd = s->native_fd;
1621 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->native_fd, &ev) < 0) {
1622 log_error("Failed to add native server fd to epoll object: %m");
1629 static int open_stdout_socket(Server *s) {
1630 union sockaddr_union sa;
1632 struct epoll_event ev;
1636 if (s->stdout_fd < 0) {
1638 s->stdout_fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC, 0);
1639 if (s->stdout_fd < 0) {
1640 log_error("socket() failed: %m");
1645 sa.un.sun_family = AF_UNIX;
1646 strncpy(sa.un.sun_path, "/run/systemd/stdout", sizeof(sa.un.sun_path));
1648 unlink(sa.un.sun_path);
1650 r = bind(s->stdout_fd, &sa.sa, offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path));
1652 log_error("bind() failed: %m");
1656 chmod(sa.un.sun_path, 0666);
1658 if (listen(s->stdout_fd, SOMAXCONN) < 0) {
1659 log_error("liste() failed: %m");
1665 ev.events = EPOLLIN;
1666 ev.data.fd = s->stdout_fd;
1667 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->stdout_fd, &ev) < 0) {
1668 log_error("Failed to add stdout server fd to epoll object: %m");
1675 static int open_signalfd(Server *s) {
1677 struct epoll_event ev;
1681 assert_se(sigemptyset(&mask) == 0);
1682 sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, -1);
1683 assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
1685 s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1686 if (s->signal_fd < 0) {
1687 log_error("signalfd(): %m");
1692 ev.events = EPOLLIN;
1693 ev.data.fd = s->signal_fd;
1695 if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->signal_fd, &ev) < 0) {
1696 log_error("epoll_ctl(): %m");
1703 static int server_init(Server *s) {
1709 s->syslog_fd = s->native_fd = s->stdout_fd = s->signal_fd = s->epoll_fd = -1;
1710 s->metrics.max_size = DEFAULT_MAX_SIZE;
1711 s->metrics.min_size = DEFAULT_MIN_SIZE;
1712 s->metrics.keep_free = DEFAULT_KEEP_FREE;
1713 s->max_use = DEFAULT_MAX_USE;
1716 s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
1717 if (!s->user_journals) {
1718 log_error("Out of memory.");
1722 s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
1723 if (s->epoll_fd < 0) {
1724 log_error("Failed to create epoll object: %m");
1728 n = sd_listen_fds(true);
1730 log_error("Failed to read listening file descriptors from environment: %s", strerror(-n));
1734 for (fd = SD_LISTEN_FDS_START; fd < SD_LISTEN_FDS_START + n; fd++) {
1736 if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/run/systemd/native", 0) > 0) {
1738 if (s->native_fd >= 0) {
1739 log_error("Too many native sockets passed.");
1745 } else if (sd_is_socket_unix(fd, SOCK_STREAM, 1, "/run/systemd/stdout", 0) > 0) {
1747 if (s->stdout_fd >= 0) {
1748 log_error("Too many stdout sockets passed.");
1754 } else if (sd_is_socket_unix(fd, SOCK_DGRAM, -1, "/dev/log", 0) > 0) {
1756 if (s->syslog_fd >= 0) {
1757 log_error("Too many /dev/log sockets passed.");
1764 log_error("Unknown socket passed.");
1769 r = open_syslog_socket(s);
1773 r = open_native_socket(s);
1777 r = open_stdout_socket(s);
1781 r = system_journal_open(s);
1785 r = open_signalfd(s);
1789 s->rate_limit = journal_rate_limit_new(DEFAULT_RATE_LIMIT_INTERVAL, DEFAULT_RATE_LIMIT_BURST);
1796 static void server_done(Server *s) {
1800 while (s->stdout_streams)
1801 stdout_stream_free(s->stdout_streams);
1803 if (s->system_journal)
1804 journal_file_close(s->system_journal);
1806 if (s->runtime_journal)
1807 journal_file_close(s->runtime_journal);
1809 while ((f = hashmap_steal_first(s->user_journals)))
1810 journal_file_close(f);
1812 hashmap_free(s->user_journals);
1814 if (s->epoll_fd >= 0)
1815 close_nointr_nofail(s->epoll_fd);
1817 if (s->signal_fd >= 0)
1818 close_nointr_nofail(s->signal_fd);
1820 if (s->syslog_fd >= 0)
1821 close_nointr_nofail(s->syslog_fd);
1823 if (s->native_fd >= 0)
1824 close_nointr_nofail(s->native_fd);
1826 if (s->stdout_fd >= 0)
1827 close_nointr_nofail(s->stdout_fd);
1830 journal_rate_limit_free(s->rate_limit);
1833 int main(int argc, char *argv[]) {
1837 /* if (getppid() != 1) { */
1838 /* log_error("This program should be invoked by init only."); */
1839 /* return EXIT_FAILURE; */
1843 log_error("This program does not take arguments.");
1844 return EXIT_FAILURE;
1847 log_set_target(LOG_TARGET_CONSOLE);
1848 log_set_max_level(LOG_DEBUG);
1849 log_parse_environment();
1854 r = server_init(&server);
1858 log_debug("systemd-journald running as pid %lu", (unsigned long) getpid());
1862 "STATUS=Processing requests...");
1864 server_vacuum(&server);
1865 server_flush_to_var(&server);
1868 struct epoll_event event;
1870 r = epoll_wait(server.epoll_fd, &event, 1, -1);
1876 log_error("epoll_wait() failed: %m");
1882 r = process_event(&server, &event);
1889 log_debug("systemd-journald stopped as pid %lu", (unsigned long) getpid());
1893 "STATUS=Shutting down...");
1895 server_done(&server);
1897 return r < 0 ? EXIT_FAILURE : EXIT_SUCCESS;