diff options
Diffstat (limited to 'src/journal/journald-server.c')
| -rw-r--r-- | src/journal/journald-server.c | 609 |
1 files changed, 414 insertions, 195 deletions
diff --git a/src/journal/journald-server.c b/src/journal/journald-server.c index fa2e9b9825..7a70dcbc57 100644 --- a/src/journal/journald-server.c +++ b/src/journal/journald-server.c @@ -19,45 +19,55 @@ along with systemd; If not, see <http://www.gnu.org/licenses/>. ***/ -#include <sys/signalfd.h> -#include <sys/ioctl.h> -#include <linux/sockios.h> -#include <sys/statvfs.h> -#include <sys/mman.h> - #ifdef HAVE_SELINUX #include <selinux/selinux.h> #endif +#include <sys/ioctl.h> +#include <sys/mman.h> +#include <sys/signalfd.h> +#include <sys/statvfs.h> +#include <linux/sockios.h> -#include <libudev.h> - +#include "libudev.h" +#include "sd-daemon.h" #include "sd-journal.h" #include "sd-messages.h" -#include "sd-daemon.h" -#include "mkdir.h" -#include "rm-rf.h" -#include "hashmap.h" -#include "journal-file.h" -#include "socket-util.h" + +#include "acl-util.h" +#include "alloc-util.h" +#include "audit-util.h" #include "cgroup-util.h" -#include "missing.h" #include "conf-parser.h" -#include "selinux-util.h" -#include "acl-util.h" +#include "dirent-util.h" +#include "extract-word.h" +#include "fd-util.h" #include "formats-util.h" -#include "process-util.h" +#include "fs-util.h" +#include "hashmap.h" #include "hostname-util.h" -#include "signal-util.h" +#include "io-util.h" +#include "journal-authenticate.h" +#include "journal-file.h" #include "journal-internal.h" #include "journal-vacuum.h" -#include "journal-authenticate.h" -#include "journald-rate-limit.h" +#include "journald-audit.h" #include "journald-kmsg.h" -#include "journald-syslog.h" -#include "journald-stream.h" #include "journald-native.h" -#include "journald-audit.h" +#include "journald-rate-limit.h" #include "journald-server.h" +#include "journald-stream.h" +#include "journald-syslog.h" +#include "missing.h" +#include "mkdir.h" +#include "parse-util.h" +#include "proc-cmdline.h" +#include "process-util.h" +#include "rm-rf.h" +#include "selinux-util.h" +#include "signal-util.h" +#include "socket-util.h" +#include "string-table.h" +#include "string-util.h" #define USER_JOURNALS_MAX 1024 @@ -66,88 +76,63 @@ #define DEFAULT_RATE_LIMIT_BURST 1000 #define DEFAULT_MAX_FILE_USEC USEC_PER_MONTH -#define RECHECK_AVAILABLE_SPACE_USEC (30*USEC_PER_SEC) +#define RECHECK_SPACE_USEC (30*USEC_PER_SEC) -static const char* const storage_table[_STORAGE_MAX] = { - [STORAGE_AUTO] = "auto", - [STORAGE_VOLATILE] = "volatile", - [STORAGE_PERSISTENT] = "persistent", - [STORAGE_NONE] = "none" -}; - -DEFINE_STRING_TABLE_LOOKUP(storage, Storage); -DEFINE_CONFIG_PARSE_ENUM(config_parse_storage, storage, Storage, "Failed to parse storage setting"); - -static const char* const split_mode_table[_SPLIT_MAX] = { - [SPLIT_LOGIN] = "login", - [SPLIT_UID] = "uid", - [SPLIT_NONE] = "none", -}; - -DEFINE_STRING_TABLE_LOOKUP(split_mode, SplitMode); -DEFINE_CONFIG_PARSE_ENUM(config_parse_split_mode, split_mode, SplitMode, "Failed to parse split mode setting"); +#define NOTIFY_SNDBUF_SIZE (8*1024*1024) -static uint64_t available_space(Server *s, bool verbose) { - char ids[33]; - _cleanup_free_ char *p = NULL; - sd_id128_t machine; - struct statvfs ss; - uint64_t sum = 0, ss_avail = 0, avail = 0; - int r; +static int determine_space_for( + Server *s, + JournalMetrics *metrics, + const char *path, + const char *name, + bool verbose, + bool patch_min_use, + uint64_t *available, + uint64_t *limit) { + + uint64_t sum = 0, ss_avail, avail; _cleanup_closedir_ DIR *d = NULL; + struct dirent *de; + struct statvfs ss; + const char *p; usec_t ts; - const char *f; - JournalMetrics *m; - - ts = now(CLOCK_MONOTONIC); - if (s->cached_available_space_timestamp + RECHECK_AVAILABLE_SPACE_USEC > ts - && !verbose) - return s->cached_available_space; + assert(s); + assert(metrics); + assert(path); + assert(name); - r = sd_id128_get_machine(&machine); - if (r < 0) - return 0; + ts = now(CLOCK_MONOTONIC); - if (s->system_journal) { - f = "/var/log/journal/"; - m = &s->system_metrics; - } else { - f = "/run/log/journal/"; - m = &s->runtime_metrics; - } + if (!verbose && s->cached_space_timestamp + RECHECK_SPACE_USEC > ts) { - assert(m); + if (available) + *available = s->cached_space_available; + if (limit) + *limit = s->cached_space_limit; - p = strappend(f, sd_id128_to_string(machine, ids)); - if (!p) return 0; + } + p = strjoina(path, SERVER_MACHINE_ID(s)); d = opendir(p); if (!d) - return 0; + return log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_ERR, errno, "Failed to open %s: %m", p); if (fstatvfs(dirfd(d), &ss) < 0) - return 0; + return log_error_errno(errno, "Failed to fstatvfs(%s): %m", p); - for (;;) { + FOREACH_DIRENT_ALL(de, d, break) { struct stat st; - struct dirent *de; - - errno = 0; - de = readdir(d); - if (!de && errno != 0) - return 0; - - if (!de) - break; if (!endswith(de->d_name, ".journal") && !endswith(de->d_name, ".journal~")) continue; - if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0) + if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0) { + log_debug_errno(errno, "Failed to stat %s/%s, ignoring: %m", p, de->d_name); continue; + } if (!S_ISREG(st.st_mode)) continue; @@ -155,45 +140,72 @@ static uint64_t available_space(Server *s, bool verbose) { sum += (uint64_t) st.st_blocks * 512UL; } - ss_avail = ss.f_bsize * ss.f_bavail; - - /* If we reached a high mark, we will always allow this much - * again, unless usage goes above max_use. This watermark - * value is cached so that we don't give up space on pressure, - * but hover below the maximum usage. */ + /* If request, then let's bump the min_use limit to the + * current usage on disk. We do this when starting up and + * first opening the journal files. This way sudden spikes in + * disk usage will not cause journald to vacuum files without + * bounds. Note that this means that only a restart of + * journald will make it reset this value. */ - if (m->use < sum) - m->use = sum; + if (patch_min_use) + metrics->min_use = MAX(metrics->min_use, sum); - avail = LESS_BY(ss_avail, m->keep_free); + ss_avail = ss.f_bsize * ss.f_bavail; + avail = LESS_BY(ss_avail, metrics->keep_free); - s->cached_available_space = LESS_BY(MIN(m->max_use, avail), sum); - s->cached_available_space_timestamp = ts; + s->cached_space_limit = MIN(MAX(sum + avail, metrics->min_use), metrics->max_use); + s->cached_space_available = LESS_BY(s->cached_space_limit, sum); + s->cached_space_timestamp = ts; if (verbose) { char fb1[FORMAT_BYTES_MAX], fb2[FORMAT_BYTES_MAX], fb3[FORMAT_BYTES_MAX], - fb4[FORMAT_BYTES_MAX], fb5[FORMAT_BYTES_MAX]; + fb4[FORMAT_BYTES_MAX], fb5[FORMAT_BYTES_MAX], fb6[FORMAT_BYTES_MAX]; server_driver_message(s, SD_MESSAGE_JOURNAL_USAGE, - "%s is currently using %s.\n" + "%s (%s) is currently using %s.\n" "Maximum allowed usage is set to %s.\n" "Leaving at least %s free (of currently available %s of space).\n" - "Enforced usage limit is thus %s.", - s->system_journal ? "Permanent journal (/var/log/journal/)" : "Runtime journal (/run/log/journal/)", + "Enforced usage limit is thus %s, of which %s are still available.", + name, path, format_bytes(fb1, sizeof(fb1), sum), - format_bytes(fb2, sizeof(fb2), m->max_use), - format_bytes(fb3, sizeof(fb3), m->keep_free), + format_bytes(fb2, sizeof(fb2), metrics->max_use), + format_bytes(fb3, sizeof(fb3), metrics->keep_free), format_bytes(fb4, sizeof(fb4), ss_avail), - format_bytes(fb5, sizeof(fb5), s->cached_available_space + sum)); + format_bytes(fb5, sizeof(fb5), s->cached_space_limit), + format_bytes(fb6, sizeof(fb6), s->cached_space_available)); } - return s->cached_available_space; + if (available) + *available = s->cached_space_available; + if (limit) + *limit = s->cached_space_limit; + + return 1; +} + +static int determine_space(Server *s, bool verbose, bool patch_min_use, uint64_t *available, uint64_t *limit) { + JournalMetrics *metrics; + const char *path, *name; + + assert(s); + + if (s->system_journal) { + path = "/var/log/journal/"; + metrics = &s->system_metrics; + name = "System journal"; + } else { + path = "/run/log/journal/"; + metrics = &s->runtime_metrics; + name = "Runtime journal"; + } + + return determine_space_for(s, metrics, path, name, verbose, patch_min_use, available, limit); } void server_fix_perms(Server *s, JournalFile *f, uid_t uid) { int r; #ifdef HAVE_ACL - acl_t acl; + _cleanup_(acl_freep) acl_t acl = NULL; acl_entry_t entry; acl_permset_t permset; #endif @@ -202,7 +214,7 @@ void server_fix_perms(Server *s, JournalFile *f, uid_t uid) { r = fchmod(f->fd, 0640); if (r < 0) - log_warning_errno(r, "Failed to fix access mode on %s, ignoring: %m", f->path); + log_warning_errno(errno, "Failed to fix access mode on %s, ignoring: %m", f->path); #ifdef HAVE_ACL if (uid <= SYSTEM_UID_MAX) @@ -221,24 +233,27 @@ void server_fix_perms(Server *s, JournalFile *f, uid_t uid) { acl_set_tag_type(entry, ACL_USER) < 0 || acl_set_qualifier(entry, &uid) < 0) { log_warning_errno(errno, "Failed to patch ACL on %s, ignoring: %m", f->path); - goto finish; + return; } } /* We do not recalculate the mask unconditionally here, * so that the fchmod() mask above stays intact. */ if (acl_get_permset(entry, &permset) < 0 || - acl_add_perm(permset, ACL_READ) < 0 || - calc_acl_mask_if_needed(&acl) < 0) { + acl_add_perm(permset, ACL_READ) < 0) { log_warning_errno(errno, "Failed to patch ACL on %s, ignoring: %m", f->path); - goto finish; + return; + } + + r = calc_acl_mask_if_needed(&acl); + if (r < 0) { + log_warning_errno(r, "Failed to patch ACL on %s, ignoring: %m", f->path); + return; } if (acl_set_fd(f->fd, acl) < 0) log_warning_errno(errno, "Failed to set ACL on %s, ignoring: %m", f->path); -finish: - acl_free(acl); #endif } @@ -328,8 +343,8 @@ void server_rotate(Server *s) { log_debug("Rotating..."); - do_rotate(s, &s->runtime_journal, "runtime", false, 0); - do_rotate(s, &s->system_journal, "system", s->seal, 0); + (void) do_rotate(s, &s->runtime_journal, "runtime", false, 0); + (void) do_rotate(s, &s->system_journal, "system", s->seal, 0); ORDERED_HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) { r = do_rotate(s, &f, "user", s->seal, PTR_TO_UINT32(k)); @@ -350,13 +365,13 @@ void server_sync(Server *s) { if (s->system_journal) { r = journal_file_set_offline(s->system_journal); if (r < 0) - log_error_errno(r, "Failed to sync system journal: %m"); + log_warning_errno(r, "Failed to sync system journal, ignoring: %m"); } ORDERED_HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) { r = journal_file_set_offline(f); if (r < 0) - log_error_errno(r, "Failed to sync user journal: %m"); + log_warning_errno(r, "Failed to sync user journal, ignoring: %m"); } if (s->sync_event_source) { @@ -370,43 +385,50 @@ void server_sync(Server *s) { static void do_vacuum( Server *s, - const char *id, JournalFile *f, - const char* path, - JournalMetrics *metrics) { + JournalMetrics *metrics, + const char *path, + const char *name, + bool verbose, + bool patch_min_use) { const char *p; + uint64_t limit; int r; + assert(s); + assert(metrics); + assert(path); + assert(name); + if (!f) return; - p = strjoina(path, id); - r = journal_directory_vacuum(p, metrics->max_use, s->max_retention_usec, &s->oldest_file_usec, false); + p = strjoina(path, SERVER_MACHINE_ID(s)); + + limit = metrics->max_use; + (void) determine_space_for(s, metrics, path, name, verbose, patch_min_use, NULL, &limit); + + r = journal_directory_vacuum(p, limit, metrics->n_max_files, s->max_retention_usec, &s->oldest_file_usec, verbose); if (r < 0 && r != -ENOENT) - log_error_errno(r, "Failed to vacuum %s: %m", p); + log_warning_errno(r, "Failed to vacuum %s, ignoring: %m", p); } -void server_vacuum(Server *s) { - char ids[33]; - sd_id128_t machine; - int r; +int server_vacuum(Server *s, bool verbose, bool patch_min_use) { + assert(s); log_debug("Vacuuming..."); s->oldest_file_usec = 0; - r = sd_id128_get_machine(&machine); - if (r < 0) { - log_error_errno(r, "Failed to get machine ID: %m"); - return; - } - sd_id128_to_string(machine, ids); + do_vacuum(s, s->system_journal, &s->system_metrics, "/var/log/journal/", "System journal", verbose, patch_min_use); + do_vacuum(s, s->runtime_journal, &s->runtime_metrics, "/run/log/journal/", "Runtime journal", verbose, patch_min_use); - do_vacuum(s, ids, s->system_journal, "/var/log/journal/", &s->system_metrics); - do_vacuum(s, ids, s->runtime_journal, "/run/log/journal/", &s->runtime_metrics); + s->cached_space_limit = 0; + s->cached_space_available = 0; + s->cached_space_timestamp = 0; - s->cached_available_space_timestamp = 0; + return 0; } static void server_cache_machine_id(Server *s) { @@ -504,7 +526,7 @@ static void write_to_journal(Server *s, uid_t uid, struct iovec *iovec, unsigned if (journal_file_rotate_suggested(f, s->max_file_usec)) { log_debug("%s: Journal header limits reached or header out-of-date, rotating.", f->path); server_rotate(s); - server_vacuum(s); + server_vacuum(s, false, false); vacuumed = true; f = find_journal(s, uid); @@ -524,7 +546,7 @@ static void write_to_journal(Server *s, uid_t uid, struct iovec *iovec, unsigned } server_rotate(s); - server_vacuum(s); + server_vacuum(s, false, false); f = find_journal(s, uid); if (!f) @@ -825,7 +847,7 @@ static void dispatch_message_real( void server_driver_message(Server *s, sd_id128_t message_id, const char *format, ...) { char mid[11 + 32 + 1]; char buffer[16 + LINE_MAX + 1]; - struct iovec iovec[N_IOVEC_META_FIELDS + 4]; + struct iovec iovec[N_IOVEC_META_FIELDS + 6]; int n = 0; va_list ap; struct ucred ucred = {}; @@ -833,6 +855,9 @@ void server_driver_message(Server *s, sd_id128_t message_id, const char *format, assert(s); assert(format); + IOVEC_SET_STRING(iovec[n++], "SYSLOG_FACILITY=3"); + IOVEC_SET_STRING(iovec[n++], "SYSLOG_IDENTIFIER=systemd-journald"); + IOVEC_SET_STRING(iovec[n++], "PRIORITY=6"); IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=driver"); @@ -866,6 +891,7 @@ void server_dispatch_message( int rl, r; _cleanup_free_ char *path = NULL; + uint64_t available = 0; char *c; assert(s); @@ -905,9 +931,8 @@ void server_dispatch_message( } } - rl = journal_rate_limit_test(s->rate_limit, path, - priority & LOG_PRIMASK, available_space(s, false)); - + (void) determine_space(s, false, false, &available, NULL); + rl = journal_rate_limit_test(s->rate_limit, path, priority & LOG_PRIMASK, available); if (rl == 0) return; @@ -922,16 +947,8 @@ finish: static int system_journal_open(Server *s, bool flush_requested) { - int r; - char *fn; - sd_id128_t machine; - char ids[33]; - - r = sd_id128_get_machine(&machine); - if (r < 0) - return log_error_errno(r, "Failed to get machine id: %m"); - - sd_id128_to_string(machine, ids); + const char *fn; + int r = 0; if (!s->system_journal && (s->storage == STORAGE_PERSISTENT || s->storage == STORAGE_AUTO) && @@ -947,15 +964,15 @@ static int system_journal_open(Server *s, bool flush_requested) { if (s->storage == STORAGE_PERSISTENT) (void) mkdir_p("/var/log/journal/", 0755); - fn = strjoina("/var/log/journal/", ids); + fn = strjoina("/var/log/journal/", SERVER_MACHINE_ID(s)); (void) mkdir(fn, 0755); fn = strjoina(fn, "/system.journal"); r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, NULL, &s->system_journal); - - if (r >= 0) + if (r >= 0) { server_fix_perms(s, s->system_journal, 0); - else if (r < 0) { + (void) determine_space_for(s, &s->system_metrics, "/var/log/journal/", "System journal", true, true, NULL, NULL); + } else if (r < 0) { if (r != -ENOENT && r != -EROFS) log_warning_errno(r, "Failed to open system journal: %m"); @@ -966,9 +983,7 @@ static int system_journal_open(Server *s, bool flush_requested) { if (!s->runtime_journal && (s->storage != STORAGE_NONE)) { - fn = strjoin("/run/log/journal/", ids, "/system.journal", NULL); - if (!fn) - return -ENOMEM; + fn = strjoina("/run/log/journal/", SERVER_MACHINE_ID(s), "/system.journal"); if (s->system_journal) { @@ -977,8 +992,6 @@ static int system_journal_open(Server *s, bool flush_requested) { * it into the system journal */ r = journal_file_open(fn, O_RDWR, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal); - free(fn); - if (r < 0) { if (r != -ENOENT) log_warning_errno(r, "Failed to open runtime journal: %m"); @@ -996,18 +1009,16 @@ static int system_journal_open(Server *s, bool flush_requested) { (void) mkdir_parents(fn, 0750); r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, s->compress, false, &s->runtime_metrics, s->mmap, NULL, &s->runtime_journal); - free(fn); - if (r < 0) return log_error_errno(r, "Failed to open runtime journal: %m"); } - if (s->runtime_journal) + if (s->runtime_journal) { server_fix_perms(s, s->runtime_journal, 0); + (void) determine_space_for(s, &s->runtime_metrics, "/run/log/journal/", "Runtime journal", true, true, NULL, NULL); + } } - available_space(s, true); - return r; } @@ -1028,7 +1039,7 @@ int server_flush_to_var(Server *s) { if (!s->runtime_journal) return 0; - system_journal_open(s, true); + (void) system_journal_open(s, true); if (!s->system_journal) return 0; @@ -1072,7 +1083,7 @@ int server_flush_to_var(Server *s) { } server_rotate(s); - server_vacuum(s); + server_vacuum(s, false, false); if (!s->system_journal) { log_notice("Didn't flush runtime journal since rotation of system journal wasn't successful."); @@ -1088,11 +1099,12 @@ int server_flush_to_var(Server *s) { } } + r = 0; + finish: journal_file_post_change(s->system_journal); - journal_file_close(s->runtime_journal); - s->runtime_journal = NULL; + s->runtime_journal = journal_file_close(s->runtime_journal); if (r >= 0) (void) rm_rf("/run/log/journal", REMOVE_ROOT); @@ -1235,9 +1247,9 @@ static int dispatch_sigusr1(sd_event_source *es, const struct signalfd_siginfo * server_flush_to_var(s); server_sync(s); - server_vacuum(s); + server_vacuum(s, false, false); - touch("/run/systemd/journal/flushed"); + (void) touch("/run/systemd/journal/flushed"); return 0; } @@ -1249,7 +1261,7 @@ static int dispatch_sigusr2(sd_event_source *es, const struct signalfd_siginfo * log_info("Received request to rotate journal from PID %"PRIu32, si->ssi_pid); server_rotate(s); - server_vacuum(s); + server_vacuum(s, true, true); return 0; } @@ -1293,8 +1305,7 @@ static int setup_signals(Server *s) { static int server_parse_proc_cmdline(Server *s) { _cleanup_free_ char *line = NULL; - const char *w, *state; - size_t l; + const char *p; int r; r = proc_cmdline(&line); @@ -1303,12 +1314,16 @@ static int server_parse_proc_cmdline(Server *s) { return 0; } - FOREACH_WORD_QUOTED(w, l, line, state) { + p = line; + for(;;) { _cleanup_free_ char *word; - word = strndup(w, l); - if (!word) - return -ENOMEM; + r = extract_first_word(&p, &word, NULL, 0); + if (r < 0) + return log_error_errno(r, "Failed to parse journald syntax \"%s\": %m", line); + + if (r == 0) + break; if (startswith(word, "systemd.journald.forward_to_syslog=")) { r = parse_boolean(word + 35); @@ -1337,8 +1352,8 @@ static int server_parse_proc_cmdline(Server *s) { } else if (startswith(word, "systemd.journald")) log_warning("Invalid systemd.journald parameter. Ignoring."); } - /* do not warn about state here, since probably systemd already did */ + /* do not warn about state here, since probably systemd already did */ return 0; } @@ -1434,8 +1449,7 @@ static int server_open_hostname(Server *s) { /* kernels prior to 3.2 don't support polling this file. Ignore * the failure. */ if (r == -EPERM) { - log_warning("Failed to register hostname fd in event loop: %s. Ignoring.", - strerror(-r)); + log_warning_errno(r, "Failed to register hostname fd in event loop, ignoring: %m"); s->hostname_fd = safe_close(s->hostname_fd); return 0; } @@ -1450,17 +1464,189 @@ static int server_open_hostname(Server *s) { return 0; } +static int dispatch_notify_event(sd_event_source *es, int fd, uint32_t revents, void *userdata) { + Server *s = userdata; + int r; + + assert(s); + assert(s->notify_event_source == es); + assert(s->notify_fd == fd); + + if (revents != EPOLLOUT) { + log_error("Invalid events on notify file descriptor."); + return -EINVAL; + } + + /* The $NOTIFY_SOCKET is writable again, now send exactly one + * message on it. Either it's the wtachdog event, the initial + * READY=1 event or an stdout stream event. If there's nothing + * to write anymore, turn our event source off. The next time + * there's something to send it will be turned on again. */ + + if (!s->sent_notify_ready) { + static const char p[] = + "READY=1\n" + "STATUS=Processing requests..."; + ssize_t l; + + l = send(s->notify_fd, p, strlen(p), MSG_DONTWAIT); + if (l < 0) { + if (errno == EAGAIN) + return 0; + + return log_error_errno(errno, "Failed to send READY=1 notification message: %m"); + } + + s->sent_notify_ready = true; + log_debug("Sent READY=1 notification."); + + } else if (s->send_watchdog) { + + static const char p[] = + "WATCHDOG=1"; + + ssize_t l; + + l = send(s->notify_fd, p, strlen(p), MSG_DONTWAIT); + if (l < 0) { + if (errno == EAGAIN) + return 0; + + return log_error_errno(errno, "Failed to send WATCHDOG=1 notification message: %m"); + } + + s->send_watchdog = false; + log_debug("Sent WATCHDOG=1 notification."); + + } else if (s->stdout_streams_notify_queue) + /* Dispatch one stream notification event */ + stdout_stream_send_notify(s->stdout_streams_notify_queue); + + /* Leave us enabled if there's still more to to do. */ + if (s->send_watchdog || s->stdout_streams_notify_queue) + return 0; + + /* There was nothing to do anymore, let's turn ourselves off. */ + r = sd_event_source_set_enabled(es, SD_EVENT_OFF); + if (r < 0) + return log_error_errno(r, "Failed to turn off notify event source: %m"); + + return 0; +} + +static int dispatch_watchdog(sd_event_source *es, uint64_t usec, void *userdata) { + Server *s = userdata; + int r; + + assert(s); + + s->send_watchdog = true; + + r = sd_event_source_set_enabled(s->notify_event_source, SD_EVENT_ON); + if (r < 0) + log_warning_errno(r, "Failed to turn on notify event source: %m"); + + r = sd_event_source_set_time(s->watchdog_event_source, usec + s->watchdog_usec / 2); + if (r < 0) + return log_error_errno(r, "Failed to restart watchdog event source: %m"); + + r = sd_event_source_set_enabled(s->watchdog_event_source, SD_EVENT_ON); + if (r < 0) + return log_error_errno(r, "Failed to enable watchdog event source: %m"); + + return 0; +} + +static int server_connect_notify(Server *s) { + union sockaddr_union sa = { + .un.sun_family = AF_UNIX, + }; + const char *e; + int r; + + assert(s); + assert(s->notify_fd < 0); + assert(!s->notify_event_source); + + /* + So here's the problem: we'd like to send notification + messages to PID 1, but we cannot do that via sd_notify(), + since that's synchronous, and we might end up blocking on + it. Specifically: given that PID 1 might block on + dbus-daemon during IPC, and dbus-daemon is logging to us, + and might hence block on us, we might end up in a deadlock + if we block on sending PID 1 notification messages -- by + generating a full blocking circle. To avoid this, let's + create a non-blocking socket, and connect it to the + notification socket, and then wait for POLLOUT before we + send anything. This should efficiently avoid any deadlocks, + as we'll never block on PID 1, hence PID 1 can safely block + on dbus-daemon which can safely block on us again. + + Don't think that this issue is real? It is, see: + https://github.com/systemd/systemd/issues/1505 + */ + + e = getenv("NOTIFY_SOCKET"); + if (!e) + return 0; + + if ((e[0] != '@' && e[0] != '/') || e[1] == 0) { + log_error("NOTIFY_SOCKET set to an invalid value: %s", e); + return -EINVAL; + } + + if (strlen(e) > sizeof(sa.un.sun_path)) { + log_error("NOTIFY_SOCKET path too long: %s", e); + return -EINVAL; + } + + s->notify_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); + if (s->notify_fd < 0) + return log_error_errno(errno, "Failed to create notify socket: %m"); + + (void) fd_inc_sndbuf(s->notify_fd, NOTIFY_SNDBUF_SIZE); + + strncpy(sa.un.sun_path, e, sizeof(sa.un.sun_path)); + if (sa.un.sun_path[0] == '@') + sa.un.sun_path[0] = 0; + + r = connect(s->notify_fd, &sa.sa, offsetof(struct sockaddr_un, sun_path) + strlen(e)); + if (r < 0) + return log_error_errno(errno, "Failed to connect to notify socket: %m"); + + r = sd_event_add_io(s->event, &s->notify_event_source, s->notify_fd, EPOLLOUT, dispatch_notify_event, s); + if (r < 0) + return log_error_errno(r, "Failed to watch notification socket: %m"); + + if (sd_watchdog_enabled(false, &s->watchdog_usec) > 0) { + s->send_watchdog = true; + + r = sd_event_add_time(s->event, &s->watchdog_event_source, CLOCK_MONOTONIC, now(CLOCK_MONOTONIC) + s->watchdog_usec/2, s->watchdog_usec*3/4, dispatch_watchdog, s); + if (r < 0) + return log_error_errno(r, "Failed to add watchdog time event: %m"); + } + + /* This should fire pretty soon, which we'll use to send the + * READY=1 event. */ + + return 0; +} + int server_init(Server *s) { _cleanup_fdset_free_ FDSet *fds = NULL; int n, r, fd; + bool no_sockets; assert(s); zero(*s); - s->syslog_fd = s->native_fd = s->stdout_fd = s->dev_kmsg_fd = s->audit_fd = s->hostname_fd = -1; + s->syslog_fd = s->native_fd = s->stdout_fd = s->dev_kmsg_fd = s->audit_fd = s->hostname_fd = s->notify_fd = -1; s->compress = true; s->seal = true; + s->watchdog_usec = USEC_INFINITY; + s->sync_interval_usec = DEFAULT_SYNC_INTERVAL_USEC; s->sync_scheduled = false; @@ -1477,18 +1663,19 @@ int server_init(Server *s) { s->max_level_console = LOG_INFO; s->max_level_wall = LOG_EMERG; - memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics)); - memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics)); + journal_reset_metrics(&s->system_metrics); + journal_reset_metrics(&s->runtime_metrics); server_parse_config_file(s); server_parse_proc_cmdline(s); + if (!!s->rate_limit_interval ^ !!s->rate_limit_burst) { log_debug("Setting both rate limit interval and burst from "USEC_FMT",%u to 0,0", s->rate_limit_interval, s->rate_limit_burst); s->rate_limit_interval = s->rate_limit_burst = 0; } - mkdir_p("/run/systemd/journal", 0755); + (void) mkdir_p("/run/systemd/journal", 0755); s->user_journals = ordered_hashmap_new(NULL); if (!s->user_journals) @@ -1502,8 +1689,6 @@ int server_init(Server *s) { if (r < 0) return log_error_errno(r, "Failed to create event loop: %m"); - sd_event_set_watchdog(s->event, true); - n = sd_listen_fds(true); if (n < 0) return log_error_errno(n, "Failed to read listening file descriptors from environment: %m"); @@ -1561,30 +1746,44 @@ int server_init(Server *s) { } } - r = server_open_stdout_socket(s, fds); - if (r < 0) - return r; + /* Try to restore streams, but don't bother if this fails */ + (void) server_restore_streams(s, fds); if (fdset_size(fds) > 0) { log_warning("%u unknown file descriptors passed, closing.", fdset_size(fds)); fds = fdset_free(fds); } + no_sockets = s->native_fd < 0 && s->stdout_fd < 0 && s->syslog_fd < 0 && s->audit_fd < 0; + + /* always open stdout, syslog, native, and kmsg sockets */ + + /* systemd-journald.socket: /run/systemd/journal/stdout */ + r = server_open_stdout_socket(s); + if (r < 0) + return r; + + /* systemd-journald-dev-log.socket: /run/systemd/journal/dev-log */ r = server_open_syslog_socket(s); if (r < 0) return r; + /* systemd-journald.socket: /run/systemd/journal/socket */ r = server_open_native_socket(s); if (r < 0) return r; + /* /dev/ksmg */ r = server_open_dev_kmsg(s); if (r < 0) return r; - r = server_open_audit(s); - if (r < 0) - return r; + /* Unless we got *some* sockets and not audit, open audit socket */ + if (s->audit_fd >= 0 || no_sockets) { + r = server_open_audit(s); + if (r < 0) + return r; + } r = server_open_kernel_seqnum(s); if (r < 0) @@ -1614,11 +1813,9 @@ int server_init(Server *s) { server_cache_boot_id(s); server_cache_machine_id(s); - r = system_journal_open(s, false); - if (r < 0) - return r; + (void) server_connect_notify(s); - return 0; + return system_journal_open(s, false); } void server_maybe_append_tags(Server *s) { @@ -1666,6 +1863,8 @@ void server_done(Server *s) { sd_event_source_unref(s->sigterm_event_source); sd_event_source_unref(s->sigint_event_source); sd_event_source_unref(s->hostname_event_source); + sd_event_source_unref(s->notify_event_source); + sd_event_source_unref(s->watchdog_event_source); sd_event_unref(s->event); safe_close(s->syslog_fd); @@ -1674,6 +1873,7 @@ void server_done(Server *s) { safe_close(s->dev_kmsg_fd); safe_close(s->audit_fd); safe_close(s->hostname_fd); + safe_close(s->notify_fd); if (s->rate_limit) journal_rate_limit_free(s->rate_limit); @@ -1691,3 +1891,22 @@ void server_done(Server *s) { udev_unref(s->udev); } + +static const char* const storage_table[_STORAGE_MAX] = { + [STORAGE_AUTO] = "auto", + [STORAGE_VOLATILE] = "volatile", + [STORAGE_PERSISTENT] = "persistent", + [STORAGE_NONE] = "none" +}; + +DEFINE_STRING_TABLE_LOOKUP(storage, Storage); +DEFINE_CONFIG_PARSE_ENUM(config_parse_storage, storage, Storage, "Failed to parse storage setting"); + +static const char* const split_mode_table[_SPLIT_MAX] = { + [SPLIT_LOGIN] = "login", + [SPLIT_UID] = "uid", + [SPLIT_NONE] = "none", +}; + +DEFINE_STRING_TABLE_LOOKUP(split_mode, SplitMode); +DEFINE_CONFIG_PARSE_ENUM(config_parse_split_mode, split_mode, SplitMode, "Failed to parse split mode setting"); |