#include <sys/statvfs.h>
#include <linux/sockios.h>
-#include "libudev.h"
#include "sd-daemon.h"
#include "sd-journal.h"
#include "sd-messages.h"
#include "acl-util.h"
+#include "alloc-util.h"
+#include "audit-util.h"
#include "cgroup-util.h"
#include "conf-parser.h"
+#include "dirent-util.h"
+#include "extract-word.h"
+#include "fd-util.h"
+#include "fileio.h"
#include "formats-util.h"
+#include "fs-util.h"
#include "hashmap.h"
#include "hostname-util.h"
+#include "io-util.h"
#include "journal-authenticate.h"
#include "journal-file.h"
#include "journal-internal.h"
#include "journald-kmsg.h"
#include "journald-native.h"
#include "journald-rate-limit.h"
+#include "journald-server.h"
#include "journald-stream.h"
#include "journald-syslog.h"
+#include "libudev.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"
-#include "journald-server.h"
#define USER_JOURNALS_MAX 1024
#define RECHECK_SPACE_USEC (30*USEC_PER_SEC)
+#define NOTIFY_SNDBUF_SIZE (8*1024*1024)
+
static int determine_space_for(
Server *s,
JournalMetrics *metrics,
/* 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);
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);
static int dispatch_sigusr1(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
Server *s = userdata;
+ int r;
assert(s);
- log_info("Received request to flush runtime journal from PID %"PRIu32, si->ssi_pid);
+ log_info("Received request to flush runtime journal from PID " PID_FMT, si->ssi_pid);
server_flush_to_var(s);
server_sync(s);
server_vacuum(s, false, false);
- (void) touch("/run/systemd/journal/flushed");
+ r = touch("/run/systemd/journal/flushed");
+ if (r < 0)
+ log_warning_errno(r, "Failed to touch /run/systemd/journal/flushed, ignoring: %m");
return 0;
}
static int dispatch_sigusr2(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
Server *s = userdata;
+ int r;
assert(s);
- log_info("Received request to rotate journal from PID %"PRIu32, si->ssi_pid);
+ log_info("Received request to rotate journal from PID " PID_FMT, si->ssi_pid);
server_rotate(s);
server_vacuum(s, true, true);
+ /* Let clients know when the most recent rotation happened. */
+ r = write_timestamp_file_atomic("/run/systemd/journal/rotated", now(CLOCK_MONOTONIC));
+ if (r < 0)
+ log_warning_errno(r, "Failed to write /run/systemd/journal/rotated, ignoring: %m");
+
return 0;
}
return 0;
}
+static int dispatch_sigrtmin1(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
+ Server *s = userdata;
+ int r;
+
+ assert(s);
+
+ log_debug("Received request to sync from PID " PID_FMT, si->ssi_pid);
+
+ server_sync(s);
+
+ /* Let clients know when the most recent sync happened. */
+ r = write_timestamp_file_atomic("/run/systemd/journal/synced", now(CLOCK_MONOTONIC));
+ if (r < 0)
+ log_warning_errno(r, "Failed to write /run/systemd/journal/synced, ignoring: %m");
+
+ return 0;
+}
+
static int setup_signals(Server *s) {
int r;
assert(s);
- assert(sigprocmask_many(SIG_SETMASK, NULL, SIGINT, SIGTERM, SIGUSR1, SIGUSR2, -1) >= 0);
+ assert(sigprocmask_many(SIG_SETMASK, NULL, SIGINT, SIGTERM, SIGUSR1, SIGUSR2, SIGRTMIN+1, -1) >= 0);
r = sd_event_add_signal(s->event, &s->sigusr1_event_source, SIGUSR1, dispatch_sigusr1, s);
if (r < 0)
if (r < 0)
return r;
+ /* Let's process SIGTERM late, so that we flush all queued
+ * messages to disk before we exit */
+ r = sd_event_source_set_priority(s->sigterm_event_source, SD_EVENT_PRIORITY_NORMAL+20);
+ if (r < 0)
+ return r;
+
+ /* When journald is invoked on the terminal (when debugging),
+ * it's useful if C-c is handled equivalent to SIGTERM. */
r = sd_event_add_signal(s->event, &s->sigint_event_source, SIGINT, dispatch_sigterm, s);
if (r < 0)
return r;
+ r = sd_event_source_set_priority(s->sigint_event_source, SD_EVENT_PRIORITY_NORMAL+20);
+ if (r < 0)
+ return r;
+
+ /* SIGRTMIN+1 causes an immediate sync. We process this very
+ * late, so that everything else queued at this point is
+ * really written to disk. Clients can watch
+ * /run/systemd/journal/synced with inotify until its mtime
+ * changes to see when a sync happened. */
+ r = sd_event_add_signal(s->event, &s->sigrtmin1_event_source, SIGRTMIN+1, dispatch_sigrtmin1, s);
+ if (r < 0)
+ return r;
+
+ r = sd_event_source_set_priority(s->sigrtmin1_event_source, SD_EVENT_PRIORITY_NORMAL+15);
+ if (r < 0)
+ return r;
+
return 0;
}
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);
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);
static int server_parse_config_file(Server *s) {
assert(s);
- return config_parse_many("/etc/systemd/journald.conf",
- CONF_DIRS_NULSTR("systemd/journald.conf"),
+ return config_parse_many(PKGSYSCONFDIR "/journald.conf",
+ CONF_PATHS_NULSTR("systemd/journald.conf.d"),
"Journal\0",
config_item_perf_lookup, journald_gperf_lookup,
false, 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);
+
+ /* 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/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;
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;
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");
server_cache_boot_id(s);
server_cache_machine_id(s);
+ (void) server_connect_notify(s);
+
return system_journal_open(s, false);
}
sd_event_source_unref(s->sigusr2_event_source);
sd_event_source_unref(s->sigterm_event_source);
sd_event_source_unref(s->sigint_event_source);
+ sd_event_source_unref(s->sigrtmin1_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);
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);