HALF ECHO LEARN / REFERENCE COLLECTION
v262.2src/journal/journald-manager.c
Git blob: a0fc0d8c9bae282356d0744068ab54f859896e0c
Lines 1001–2000 of 2,634
iovec, n,
&m->seqnum->seqnum,
&m->seqnum->id,
/* ret_object= */ NULL,
/* ret_offset= */ NULL);
if (r >= 0) {
manager_schedule_sync(m, priority);
return;
}
log_debug_errno(r, "Failed to write entry to %s (%zu items, %zu bytes): %m", f->path, n, iovec_total_size(iovec, n));
if (!shall_try_append_again(f, r))
return;
if (vacuumed) {
log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
"Suppressing rotation, as we already rotated immediately before write attempt. Giving up.");
return;
}
manager_rotate_journal(m, TAKE_PTR(f), uid);
manager_vacuum(m, /* verbose= */ false);
f = manager_find_journal(m, uid);
if (!f)
return;
log_debug_errno(r, "Retrying write.");
r = journal_file_append_entry(
f,
ts,
/* boot_id= */ NULL,
iovec, n,
&m->seqnum->seqnum,
&m->seqnum->id,
/* ret_object= */ NULL,
/* ret_offset= */ NULL);
if (r < 0)
log_ratelimit_error_errno(r, FAILED_TO_WRITE_ENTRY_RATELIMIT,
"Failed to write entry to %s (%zu items, %zu bytes) despite vacuuming, ignoring: %m",
f->path, n, iovec_total_size(iovec, n));
else
manager_schedule_sync(m, priority);
}
#define IOVEC_ADD_NUMERIC_FIELD(iovec, n, value, type, isset, format, field) \
if (isset(value)) { \
char *k; \
k = newa(char, STRLEN(field "=") + DECIMAL_STR_MAX(type) + 1); \
sprintf(k, field "=" format, value); \
iovec[n++] = IOVEC_MAKE_STRING(k); \
}
#define IOVEC_ADD_STRING_FIELD(iovec, n, value, field) \
if (!isempty(value)) { \
char *k; \
k = strjoina(field "=", value); \
iovec[n++] = IOVEC_MAKE_STRING(k); \
}
#define IOVEC_ADD_ID128_FIELD(iovec, n, value, field) \
if (!sd_id128_is_null(value)) { \
char *k; \
k = newa(char, STRLEN(field "=") + SD_ID128_STRING_MAX); \
sd_id128_to_string(value, stpcpy(k, field "=")); \
iovec[n++] = IOVEC_MAKE_STRING(k); \
}
static void manager_dispatch_message_real(
Manager *m,
struct iovec *iovec, size_t n, size_t mm,
const ClientContext *c,
const struct timeval *tv,
int priority,
pid_t object_pid) {
char source_time[STRLEN("_SOURCE_REALTIME_TIMESTAMP=") + DECIMAL_STR_MAX(usec_t)];
_unused_ _cleanup_free_ char *cmdline1 = NULL, *cmdline2 = NULL;
uid_t journal_uid;
ClientContext *o;
assert(m);
assert(iovec);
assert(n > 0);
assert(n +
N_IOVEC_META_FIELDS +
(pid_is_valid(object_pid) ? N_IOVEC_OBJECT_FIELDS : 0) +
client_context_extra_fields_n_iovec(c) <= mm);
if (c) {
IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->pid, pid_t, pid_is_valid, PID_FMT, "_PID");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->uid, uid_t, uid_is_valid, UID_FMT, "_UID");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->gid, gid_t, gid_is_valid, GID_FMT, "_GID");
IOVEC_ADD_STRING_FIELD(iovec, n, c->comm, "_COMM"); /* At most TASK_COMM_LENGTH (16 bytes) */
IOVEC_ADD_STRING_FIELD(iovec, n, c->exe, "_EXE"); /* A path, so at most PATH_MAX (4096 bytes) */
if (c->cmdline)
/* At most _SC_ARG_MAX (2MB usually), which is too much to put on stack.
* Let's use a heap allocation for this one. */
cmdline1 = set_iovec_string_field(iovec, &n, "_CMDLINE=", c->cmdline);
IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->capability_quintet.effective, uint64_t, capability_is_set, "%" PRIx64, "_CAP_EFFECTIVE");
IOVEC_ADD_STRING_FIELD(iovec, n, c->label, "_SELINUX_CONTEXT");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->auditid, uint32_t, audit_session_is_valid, "%" PRIu32, "_AUDIT_SESSION");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->loginuid, uid_t, uid_is_valid, UID_FMT, "_AUDIT_LOGINUID");
IOVEC_ADD_STRING_FIELD(iovec, n, c->cgroup, "_SYSTEMD_CGROUP"); /* A path */
IOVEC_ADD_STRING_FIELD(iovec, n, c->session, "_SYSTEMD_SESSION");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, c->owner_uid, uid_t, uid_is_valid, UID_FMT, "_SYSTEMD_OWNER_UID");
IOVEC_ADD_STRING_FIELD(iovec, n, c->unit, "_SYSTEMD_UNIT"); /* Unit names are bounded by UNIT_NAME_MAX */
IOVEC_ADD_STRING_FIELD(iovec, n, c->user_unit, "_SYSTEMD_USER_UNIT");
IOVEC_ADD_STRING_FIELD(iovec, n, c->slice, "_SYSTEMD_SLICE");
IOVEC_ADD_STRING_FIELD(iovec, n, c->user_slice, "_SYSTEMD_USER_SLICE");
IOVEC_ADD_ID128_FIELD(iovec, n, c->invocation_id, "_SYSTEMD_INVOCATION_ID");
if (c->extra_fields_n_iovec > 0) {
memcpy(iovec + n, c->extra_fields_iovec, c->extra_fields_n_iovec * sizeof(struct iovec));
n += c->extra_fields_n_iovec;
}
}
assert(n <= mm);
if (pid_is_valid(object_pid) && client_context_get(m, object_pid, /* ucred= */ NULL, /* label= */ NULL, /* unit_id= */ NULL, &o) >= 0) {
IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->pid, pid_t, pid_is_valid, PID_FMT, "OBJECT_PID");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->uid, uid_t, uid_is_valid, UID_FMT, "OBJECT_UID");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->gid, gid_t, gid_is_valid, GID_FMT, "OBJECT_GID");
/* See above for size limits, only ->cmdline may be large, so use a heap allocation for it. */
IOVEC_ADD_STRING_FIELD(iovec, n, o->comm, "OBJECT_COMM");
IOVEC_ADD_STRING_FIELD(iovec, n, o->exe, "OBJECT_EXE");
if (o->cmdline)
cmdline2 = set_iovec_string_field(iovec, &n, "OBJECT_CMDLINE=", o->cmdline);
IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->capability_quintet.effective, uint64_t, capability_is_set, "%" PRIx64, "OBJECT_CAP_EFFECTIVE");
IOVEC_ADD_STRING_FIELD(iovec, n, o->label, "OBJECT_SELINUX_CONTEXT");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->auditid, uint32_t, audit_session_is_valid, "%" PRIu32, "OBJECT_AUDIT_SESSION");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->loginuid, uid_t, uid_is_valid, UID_FMT, "OBJECT_AUDIT_LOGINUID");
IOVEC_ADD_STRING_FIELD(iovec, n, o->cgroup, "OBJECT_SYSTEMD_CGROUP");
IOVEC_ADD_STRING_FIELD(iovec, n, o->session, "OBJECT_SYSTEMD_SESSION");
IOVEC_ADD_NUMERIC_FIELD(iovec, n, o->owner_uid, uid_t, uid_is_valid, UID_FMT, "OBJECT_SYSTEMD_OWNER_UID");
IOVEC_ADD_STRING_FIELD(iovec, n, o->unit, "OBJECT_SYSTEMD_UNIT");
IOVEC_ADD_STRING_FIELD(iovec, n, o->user_unit, "OBJECT_SYSTEMD_USER_UNIT");
IOVEC_ADD_STRING_FIELD(iovec, n, o->slice, "OBJECT_SYSTEMD_SLICE");
IOVEC_ADD_STRING_FIELD(iovec, n, o->user_slice, "OBJECT_SYSTEMD_USER_SLICE");
IOVEC_ADD_ID128_FIELD(iovec, n, o->invocation_id, "OBJECT_SYSTEMD_INVOCATION_ID");
}
assert(n <= mm);
if (tv) {
xsprintf(source_time, "_SOURCE_REALTIME_TIMESTAMP=" USEC_FMT, timeval_load(tv));
iovec[n++] = IOVEC_MAKE_STRING(source_time);
}
/* Note that strictly speaking storing the boot id here is
* redundant since the entry includes this in-line
* anyway. However, we need this indexed, too. */
if (!isempty(m->boot_id_field))
iovec[n++] = IOVEC_MAKE_STRING(m->boot_id_field);
if (!isempty(m->machine_id_field))
iovec[n++] = IOVEC_MAKE_STRING(m->machine_id_field);
if (!isempty(m->hostname_field))
iovec[n++] = IOVEC_MAKE_STRING(m->hostname_field);
if (!isempty(m->namespace_field))
iovec[n++] = IOVEC_MAKE_STRING(m->namespace_field);
iovec[n++] = in_initrd() ? IOVEC_MAKE_STRING("_RUNTIME_SCOPE=initrd") : IOVEC_MAKE_STRING("_RUNTIME_SCOPE=system");
assert(n <= mm);
if (m->config.split_mode == SPLIT_UID && c && uid_is_valid(c->uid))
/* Split up strictly by (non-root) UID */
journal_uid = c->uid;
else if (m->config.split_mode == SPLIT_LOGIN && c && c->uid > 0 && uid_is_valid(c->owner_uid))
/* Split up by login UIDs. We do this only if the
* realuid is not root, in order not to accidentally
* leak privileged information to the user that is
* logged by a privileged process that is part of an
* unprivileged session. */
journal_uid = c->owner_uid;
else
journal_uid = 0;
/* Get the closest, linearized time we have for this log event from the event loop. (Note that we do
* not use the source time, and not even the time the event was originally seen, but instead simply
* the time we started processing it, as we want strictly linear ordering in what we write out.) */
struct dual_timestamp ts;
event_dual_timestamp_now(m->event, &ts);
(void) manager_forward_socket(m, iovec, n, &ts, priority);
manager_write_to_journal(m, journal_uid, iovec, n, &ts, priority);
}
void manager_driver_message_internal(Manager *m, pid_t object_pid, const char *format, ...) {
struct iovec *iovec;
size_t n = 0, k, mm;
va_list ap;
int r;
assert(m);
assert(format);
mm = N_IOVEC_META_FIELDS + 5 + N_IOVEC_PAYLOAD_FIELDS + client_context_extra_fields_n_iovec(m->my_context) + N_IOVEC_OBJECT_FIELDS;
iovec = newa(struct iovec, mm);
assert_cc(3 == LOG_FAC(LOG_DAEMON));
iovec[n++] = IOVEC_MAKE_STRING("SYSLOG_FACILITY=3");
iovec[n++] = IOVEC_MAKE_STRING("SYSLOG_IDENTIFIER=systemd-journald");
iovec[n++] = IOVEC_MAKE_STRING("_TRANSPORT=driver");
assert_cc(6 == LOG_INFO);
iovec[n++] = IOVEC_MAKE_STRING("PRIORITY=6");
k = n;
va_start(ap, format);
DISABLE_WARNING_FORMAT_NONLITERAL;
r = log_format_iovec(iovec, mm, &n, false, 0, format, ap);
REENABLE_WARNING;
/* Error handling below */
va_end(ap);
if (r >= 0)
manager_dispatch_message_real(m, iovec, n, mm, m->my_context, /* tv= */ NULL, LOG_INFO, object_pid);
while (k < n)
free(iovec[k++].iov_base);
if (r < 0) {
/* We failed to format the message. Emit a warning instead. */
char buf[LINE_MAX];
errno = -r;
xsprintf(buf, "MESSAGE=Entry printing failed: %m");
n = 3;
iovec[n++] = IOVEC_MAKE_STRING("PRIORITY=4");
iovec[n++] = IOVEC_MAKE_STRING(buf);
manager_dispatch_message_real(m, iovec, n, mm, m->my_context, /* tv= */ NULL, LOG_INFO, object_pid);
}
}
void manager_dispatch_message(
Manager *m,
struct iovec *iovec, size_t n, size_t mm,
ClientContext *c,
const struct timeval *tv,
int priority,
pid_t object_pid) {
uint64_t available = 0;
int rl;
assert(m);
assert(iovec || n == 0);
if (n == 0)
return;
if (LOG_PRI(priority) > m->config.max_level_store)
return;
/* Stop early in case the information will not be stored
* in a journal. */
if (m->config.storage == STORAGE_NONE)
return;
if (c && c->unit) {
(void) manager_determine_space(m, &available, /* limit= */ NULL);
rl = journal_ratelimit_test(
&m->ratelimit_groups_by_id,
c->unit,
c->log_ratelimit_interval,
c->log_ratelimit_burst,
LOG_PRI(priority),
available);
if (rl == 0)
return;
/* Write a suppression message if we suppressed something */
if (rl > 1)
manager_driver_message(m, c->pid,
LOG_MESSAGE_ID(SD_MESSAGE_JOURNAL_DROPPED_STR),
LOG_MESSAGE("Suppressed %i messages from %s", rl - 1, c->unit),
LOG_ITEM("N_DROPPED=%i", rl - 1));
}
manager_dispatch_message_real(m, iovec, n, mm, c, tv, priority, object_pid);
}
int manager_flush_to_var(Manager *m, bool require_flag_file) {
sd_journal *j = NULL;
const char *fn;
unsigned n = 0;
usec_t start;
int r, k;
assert(m);
if (!IN_SET(m->config.storage, STORAGE_AUTO, STORAGE_PERSISTENT))
return 0;
if (m->namespace) /* Flushing concept does not exist for namespace instances */
return 0;
if (!m->runtime_journal) /* Nothing to flush? */
return 0;
if (require_flag_file && !manager_flushed_flag_is_set(m))
return 0;
(void) manager_system_journal_open(m, /* flush_requested= */ true, /* relinquish_requested= */ false);
if (!m->system_journal)
return 0;
/* Offline and close the 'main' runtime journal file to allow the runtime journal to be opened with
* the SD_JOURNAL_ASSUME_IMMUTABLE flag in the below. */
m->runtime_journal = journal_file_offline_close(m->runtime_journal);
/* Reset current seqnum data to avoid unnecessary rotation when switching to system journal.
* See issue #30092. */
zero(*m->seqnum);
log_debug("Flushing to %s...", m->system_storage.path);
start = now(CLOCK_MONOTONIC);
r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY | SD_JOURNAL_ASSUME_IMMUTABLE);
if (r < 0) {
log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "Failed to read runtime journal: %m");
goto finish;
}
sd_journal_set_data_threshold(j, 0);
SD_JOURNAL_FOREACH(j) {
Object *o = NULL;
JournalFile *f;
f = j->current_file;
assert(f && f->current_offset > 0);
n++;
r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
if (r < 0) {
log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "Can't read entry: %m");
goto finish;
}
r = journal_file_copy_entry(
f,
m->system_journal,
o,
f->current_offset,
&m->seqnum->seqnum,
&m->seqnum->id);
if (r >= 0)
continue;
if (!shall_try_append_again(m->system_journal, r)) {
log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "Can't write entry: %m");
goto finish;
}
log_ratelimit_info(JOURNAL_LOG_RATELIMIT, "Rotating system journal.");
manager_rotate_journal(m, m->system_journal, /* uid= */ 0);
manager_vacuum(m, /* verbose= */ false);
if (!m->system_journal) {
log_ratelimit_notice(JOURNAL_LOG_RATELIMIT,
"Didn't flush runtime journal since rotation of system journal wasn't successful.");
r = -EIO;
goto finish;
}
log_debug("Retrying write.");
r = journal_file_copy_entry(
f,
m->system_journal,
o,
f->current_offset,
&m->seqnum->seqnum,
&m->seqnum->id);
if (r < 0) {
log_ratelimit_error_errno(r, JOURNAL_LOG_RATELIMIT, "Can't write entry: %m");
goto finish;
}
}
r = 0;
finish:
if (m->system_journal)
journal_file_post_change(m->system_journal);
/* Save parent directories of runtime journals before closing runtime journals. */
_cleanup_strv_free_ char **dirs = NULL;
(void) journal_get_directories(j, &dirs);
/* First, close all runtime journals opened in the above. */
sd_journal_close(j);
/* Remove the runtime directory if the all entries are successfully flushed to /var/. */
if (r >= 0) {
r = rm_rf(m->runtime_storage.path, REMOVE_ROOT);
if (r < 0)
log_debug_errno(r, "Failed to remove runtime journal directory %s, ignoring: %m", m->runtime_storage.path);
else
log_debug("Removed runtime journal directory %s.", m->runtime_storage.path);
/* The initrd may have a different machine ID from the host's one. Typically, that happens
* when our tests running on qemu, as the host's initrd is picked as is without updating
* the machine ID in the initrd with the one used in the image. Even in such the case, the
* runtime journals in the subdirectory named with the initrd's machine ID are flushed to
* the persistent journal. To make not the runtime journal flushed multiple times, let's
* also remove the runtime directories. */
STRV_FOREACH(p, dirs) {
r = rm_rf(*p, REMOVE_ROOT);
if (r < 0)
log_debug_errno(r, "Failed to remove additional runtime journal directory %s, ignoring: %m", *p);
else
log_debug("Removed additional runtime journal directory %s.", *p);
}
}
manager_driver_message(m, 0,
LOG_MESSAGE("Time spent on flushing to %s is %s for %u entries.",
m->system_storage.path,
FORMAT_TIMESPAN(usec_sub_unsigned(now(CLOCK_MONOTONIC), start), 0),
n));
fn = strjoina(m->runtime_directory, "/flushed");
k = touch(fn);
if (k < 0)
log_ratelimit_warning_errno(k, JOURNAL_LOG_RATELIMIT,
"Failed to touch %s, ignoring: %m", fn);
manager_refresh_idle_timer(m);
return r;
}
int manager_relinquish_var(Manager *m) {
assert(m);
if (m->config.storage == STORAGE_NONE)
return 0;
if (m->namespace) /* Concept does not exist for namespaced instances */
return -EOPNOTSUPP;
if (m->runtime_journal && !m->system_journal)
return 0;
log_debug("Relinquishing %s...", m->system_storage.path);
(void) manager_system_journal_open(m, /* flush_requested= */ false, /* relinquish_requested= */ true);
m->system_journal = journal_file_offline_close(m->system_journal);
ordered_hashmap_clear(m->user_journals);
set_clear(m->deferred_closes);
manager_refresh_idle_timer(m);
return 0;
}
int manager_process_datagram(
sd_event_source *es,
int fd,
uint32_t revents,
void *userdata) {
size_t mm;
Manager *m = ASSERT_PTR(userdata);
struct ucred *ucred = NULL;
struct timeval tv_buf, *tv = NULL;
struct cmsghdr *cmsg;
_cleanup_free_ char *label = NULL;
struct iovec iovec;
ssize_t n;
int *fds = NULL, v = 0;
size_t n_fds = 0;
/* We use NAME_MAX space for the SELinux label here. The kernel currently enforces no limit, but
* according to suggestions from the SELinux people this will change and it will probably be
* identical to NAME_MAX. For now we use that, but this should be updated one day when the final
* limit is known.
*
* Here, we need to explicitly initialize the buffer with zero, as glibc has a bug in
* __convert_scm_timestamps(), which assumes the buffer is initialized. See #20741.
* The issue is fixed on glibc-2.35 (8fba672472ae0055387e9315fc2eddfa6775ca79). */
CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) +
CMSG_SPACE_TIMEVAL +
CMSG_SPACE(sizeof(int)) + /* fd */
CMSG_SPACE(NAME_MAX) /* selinux label */) control = {};
union sockaddr_union sa = {};
struct msghdr msghdr = {
.msg_iov = &iovec,
.msg_iovlen = 1,
.msg_control = &control,
.msg_controllen = sizeof(control),
.msg_name = &sa,
.msg_namelen = sizeof(sa),
};
assert(fd == m->native_fd || fd == m->syslog_fd || fd == m->audit_fd);
if (revents != EPOLLIN)
return log_error_errno(SYNTHETIC_ERRNO(EIO),
"Got invalid event from epoll for datagram fd: %" PRIx32,
revents);
/* Try to get the right size, if we can. (Not all sockets support SIOCINQ, hence we just try, but don't rely on
* it.) */
(void) ioctl(fd, SIOCINQ, &v);
/* Fix it up, if it is too small. We use the same fixed value as auditd here. Awful! */
mm = PAGE_ALIGN(MAX3((size_t) v + 1,
(size_t) LINE_MAX,
ALIGN(sizeof(struct nlmsghdr)) + ALIGN((size_t) MAX_AUDIT_MESSAGE_LENGTH)) + 1);
if (!GREEDY_REALLOC(m->buffer, mm))
return log_oom();
iovec = IOVEC_MAKE(m->buffer, MALLOC_ELEMENTSOF(m->buffer) - 1); /* Leave room for trailing NUL we add later */
n = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
if (ERRNO_IS_NEG_TRANSIENT(n))
return 0;
if (n == -ECHRNG) {
log_ratelimit_warning_errno(n, JOURNAL_LOG_RATELIMIT,
"Got message with truncated control data (too many fds sent?), ignoring.");
return 0;
}
if (n == -EXFULL) {
log_ratelimit_warning_errno(n, JOURNAL_LOG_RATELIMIT, "Got message with truncated payload data, ignoring.");
return 0;
}
if (n < 0)
return log_ratelimit_error_errno(n, JOURNAL_LOG_RATELIMIT, "Failed to receive message: %m");
CMSG_FOREACH(cmsg, &msghdr) {
if (cmsg->cmsg_level != SOL_SOCKET)
continue;
if (cmsg->cmsg_type == SCM_CREDENTIALS &&
cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
assert(!ucred);
ucred = CMSG_TYPED_DATA(cmsg, struct ucred);
} else if (cmsg->cmsg_type == SCM_SECURITY && mac_selinux_use()) {
assert(!label);
/* Here, we ignore any errors including OOM, as the field is optional. */
(void) make_cstring(CMSG_TYPED_DATA(cmsg, char), cmsg->cmsg_len - CMSG_LEN(0),
MAKE_CSTRING_ALLOW_TRAILING_NUL, &label);
} else if (cmsg->cmsg_type == SCM_TIMESTAMP &&
cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval))) {
assert(!tv);
tv = memcpy(&tv_buf, CMSG_DATA(cmsg), sizeof(struct timeval));
} else if (cmsg->cmsg_type == SCM_RIGHTS) {
assert(!fds);
fds = CMSG_TYPED_DATA(cmsg, int);
n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
}
}
/* And a trailing NUL, just in case */
m->buffer[n] = 0;
if (fd == m->syslog_fd) {
if (n > 0 && n_fds == 0)
manager_process_syslog_message(m, m->buffer, n, ucred, tv, label);
else if (n_fds > 0)
log_ratelimit_warning(JOURNAL_LOG_RATELIMIT,
"Got file descriptors via syslog socket. Ignoring.");
if (tv)
m->syslog_timestamp = timeval_load(tv);
} else if (fd == m->native_fd) {
if (n > 0 && n_fds == 0)
manager_process_native_message(m, m->buffer, n, ucred, tv, label);
else if (n == 0 && n_fds == 1)
(void) manager_process_native_file(m, fds[0], ucred, tv, label);
else if (n_fds > 0)
log_ratelimit_warning(JOURNAL_LOG_RATELIMIT,
"Got too many file descriptors via native socket. Ignoring.");
if (tv)
m->native_timestamp = timeval_load(tv);
} else {
assert(fd == m->audit_fd);
if (n > 0 && n_fds == 0)
manager_process_audit_message(m, m->buffer, n, ucred, &sa, msghdr.msg_namelen);
else if (n_fds > 0)
log_ratelimit_warning(JOURNAL_LOG_RATELIMIT,
"Got file descriptors via audit socket. Ignoring.");
}
close_many(fds, n_fds);
if (tv)
sync_req_revalidate_by_timestamp(m);
manager_refresh_idle_timer(m);
return 0;
}
void manager_full_flush(Manager *m) {
assert(m);
(void) manager_flush_to_var(m, false);
manager_sync(m, /* wait= */ false);
manager_vacuum(m, false);
manager_space_usage_message(m, NULL);
manager_refresh_idle_timer(m);
}
static int dispatch_sigusr1(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
assert(si);
if (!si_code_from_process(si->ssi_code)) {
log_warning("Received SIGUSR1 with unexpected .si_code %i, ignoring.", si->ssi_code);
return 0;
}
if (m->namespace) {
log_warning("Received SIGUSR1 signal from PID %u, but flushing runtime journals not supported for namespaced instances, ignoring.", si->ssi_pid);
return 0;
}
log_info("Received SIGUSR1 signal from PID %u, as request to flush runtime journal.", si->ssi_pid);
manager_full_flush(m);
return 0;
}
void manager_full_rotate(Manager *m) {
const char *fn;
int r;
assert(m);
manager_rotate(m);
manager_vacuum(m, true);
if (m->system_journal)
patch_min_use(&m->system_storage);
if (m->runtime_journal)
patch_min_use(&m->runtime_storage);
/* Let clients know when the most recent rotation happened. */
fn = strjoina(m->runtime_directory, "/rotated");
r = write_timestamp_file_atomic(fn, now(CLOCK_MONOTONIC));
if (r < 0)
log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
"Failed to write %s, ignoring: %m", fn);
}
static int dispatch_sigusr2(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
assert(si);
if (!si_code_from_process(si->ssi_code)) {
log_warning("Received SIGUSR2 with unexpected .si_code %i, ignoring.", si->ssi_code);
return 0;
}
log_info("Received SIGUSR2 signal from PID %u, as request to rotate journal, rotating.", si->ssi_pid);
manager_full_rotate(m);
return 0;
}
static int dispatch_sigterm(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
_cleanup_(sd_event_source_disable_unrefp) sd_event_source *news = NULL;
Manager *m = ASSERT_PTR(userdata);
int r;
log_received_signal(LOG_INFO, si);
(void) sd_event_source_set_enabled(es, SD_EVENT_OFF); /* Make sure this handler is called at most once */
/* So on one hand we want to ensure that SIGTERMs are definitely handled in appropriate, bounded
* time. On the other hand we want that everything pending is first comprehensively processed and
* written to disk. These goals are incompatible, hence we try to find a middle ground: we'll process
* SIGTERM with high priority, but from the handler (this one right here) we'll install two new event
* sources: one low priority idle one that will issue the exit once everything else is processed (and
* which is hopefully the regular, clean codepath); and one high priority timer that acts as safety
* net: if our idle handler isn't run within 10s, we'll exit anyway.
*
* TLDR: we'll exit either when everything is processed, or after 10s max, depending on what happens
* first.
*
* Note that exiting before the idle event is hit doesn't typically mean that we lose any data, as
* messages will remain queued in the sockets they came in from, and thus can be processed when we
* start up next – unless we are going down for the final system shutdown, in which case everything
* is lost. */
r = sd_event_add_defer(m->event, &news, NULL, NULL); /* NULL handler means → exit when triggered */
if (r < 0) {
log_error_errno(r, "Failed to allocate exit idle event handler: %m");
goto fail;
}
(void) sd_event_source_set_description(news, "exit-idle");
/* Run everything relevant before this. */
r = sd_event_source_set_priority(news, SD_EVENT_PRIORITY_NORMAL+20);
if (r < 0) {
log_error_errno(r, "Failed to adjust priority of exit idle event handler: %m");
goto fail;
}
/* Give up ownership, so that this event source is freed automatically when the event loop is freed. */
r = sd_event_source_set_floating(news, true);
if (r < 0) {
log_error_errno(r, "Failed to make exit idle event handler floating: %m");
goto fail;
}
news = sd_event_source_unref(news);
r = sd_event_add_time_relative(m->event, &news, CLOCK_MONOTONIC, 10 * USEC_PER_SEC, 0, NULL, NULL);
if (r < 0) {
log_error_errno(r, "Failed to allocate exit timeout event handler: %m");
goto fail;
}
(void) sd_event_source_set_description(news, "exit-timeout");
r = sd_event_source_set_priority(news, SD_EVENT_PRIORITY_IMPORTANT-20); /* This is a safety net, with highest priority */
if (r < 0) {
log_error_errno(r, "Failed to adjust priority of exit timeout event handler: %m");
goto fail;
}
r = sd_event_source_set_floating(news, true);
if (r < 0) {
log_error_errno(r, "Failed to make exit timeout event handler floating: %m");
goto fail;
}
news = sd_event_source_unref(news);
log_debug("Exit event sources are now pending.");
return 0;
fail:
sd_event_exit(m->event, 0);
return 0;
}
void manager_full_sync(Manager *m, bool wait) {
const char *fn;
int r;
assert(m);
manager_sync(m, wait);
/* Let clients know when the most recent sync happened. */
fn = strjoina(m->runtime_directory, "/synced");
r = write_timestamp_file_atomic(fn, now(CLOCK_MONOTONIC));
if (r < 0)
log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT,
"Failed to write %s, ignoring: %m", fn);
}
static int dispatch_sigrtmin1(sd_event_source *es, const struct signalfd_siginfo *si, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
assert(si);
if (!si_code_from_process(si->ssi_code)) {
log_warning("Received SIGRTMIN1 with unexpected .si_code %i, ignoring.", si->ssi_code);
return 0;
}
log_debug("Received SIGRTMIN1 signal from PID %u, as request to sync.", si->ssi_pid);
manager_full_sync(m, /* wait= */ false);
return 0;
}
static int manager_setup_signals(Manager *m) {
int r;
assert(m);
r = sd_event_add_signal(m->event, &m->sigusr1_event_source, SIGUSR1|SD_EVENT_SIGNAL_PROCMASK, dispatch_sigusr1, m);
if (r < 0)
return r;
r = sd_event_add_signal(m->event, &m->sigusr2_event_source, SIGUSR2|SD_EVENT_SIGNAL_PROCMASK, dispatch_sigusr2, m);
if (r < 0)
return r;
r = sd_event_add_signal(m->event, &m->sigterm_event_source, SIGTERM|SD_EVENT_SIGNAL_PROCMASK, dispatch_sigterm, m);
if (r < 0)
return r;
/* Let's process SIGTERM early, so that we definitely react to it */
r = sd_event_source_set_priority(m->sigterm_event_source, SD_EVENT_PRIORITY_IMPORTANT-10);
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(m->event, &m->sigint_event_source, SIGINT|SD_EVENT_SIGNAL_PROCMASK, dispatch_sigterm, m);
if (r < 0)
return r;
r = sd_event_source_set_priority(m->sigint_event_source, SD_EVENT_PRIORITY_IMPORTANT-10);
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(m->event, &m->sigrtmin1_event_source, (SIGRTMIN+1)|SD_EVENT_SIGNAL_PROCMASK, dispatch_sigrtmin1, m);
if (r < 0)
return r;
r = sd_event_source_set_priority(m->sigrtmin1_event_source, SD_EVENT_PRIORITY_NORMAL+15);
if (r < 0)
return r;
r = sd_event_add_signal(m->event, /* ret= */ NULL, (SIGRTMIN+18)|SD_EVENT_SIGNAL_PROCMASK, sigrtmin18_handler, &m->sigrtmin18_info);
if (r < 0)
return r;
r = sd_event_add_signal(m->event, NULL, SIGHUP|SD_EVENT_SIGNAL_PROCMASK, manager_dispatch_reload_signal, m);
if (r < 0)
return r;
return 0;
}
static int manager_dispatch_sync(sd_event_source *es, usec_t t, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
manager_sync(m, /* wait= */ false);
return 0;
}
static int manager_schedule_sync(Manager *m, int priority) {
int r;
assert(m);
if (priority <= LOG_CRIT) {
/* Immediately sync to disk when this is of priority CRIT, ALERT, EMERG */
manager_sync(m, /* wait= */ false);
return 0;
}
if (!m->event || sd_event_get_state(m->event) == SD_EVENT_FINISHED) {
/* Shutting down the manager? Let's sync immediately. */
manager_sync(m, /* wait= */ false);
return 0;
}
if (m->sync_scheduled)
return 0;
if (m->config.sync_interval_usec > 0) {
if (!m->sync_event_source) {
r = sd_event_add_time_relative(
m->event,
&m->sync_event_source,
CLOCK_MONOTONIC,
m->config.sync_interval_usec, 0,
manager_dispatch_sync, m);
if (r < 0)
return r;
r = sd_event_source_set_priority(m->sync_event_source, SD_EVENT_PRIORITY_IMPORTANT);
} else {
r = sd_event_source_set_time_relative(m->sync_event_source, m->config.sync_interval_usec);
if (r < 0)
return r;
r = sd_event_source_set_enabled(m->sync_event_source, SD_EVENT_ONESHOT);
}
if (r < 0)
return r;
m->sync_scheduled = true;
}
return 0;
}
static int dispatch_hostname_change(sd_event_source *es, int fd, uint32_t revents, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
manager_cache_hostname(m);
return 0;
}
static int manager_open_hostname(Manager *m) {
int r;
assert(m);
m->hostname_fd = open("/proc/sys/kernel/hostname",
O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
if (m->hostname_fd < 0)
return log_error_errno(errno, "Failed to open %s: %m", "/proc/sys/kernel/hostname");
r = sd_event_add_io(m->event, &m->hostname_event_source, m->hostname_fd, 0, dispatch_hostname_change, m);
if (r < 0)
return log_error_errno(r, "Failed to register hostname fd in event loop: %m");
r = sd_event_source_set_priority(m->hostname_event_source, SD_EVENT_PRIORITY_IMPORTANT-10);
if (r < 0)
return log_error_errno(r, "Failed to adjust priority of hostname event source: %m");
return 0;
}
static int dispatch_notify_event(sd_event_source *es, int fd, uint32_t revents, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
int r;
assert(m->notify_event_source == es);
assert(m->notify_fd == fd);
/* The $NOTIFY_SOCKET is writable again, now send exactly one
* message on it. Either it's the watchdog 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 (!m->sent_notify_ready) {
if (send(m->notify_fd, NOTIFY_READY_MESSAGE, strlen(NOTIFY_READY_MESSAGE), MSG_DONTWAIT) < 0) {
if (errno == EAGAIN)
return 0;
return log_error_errno(errno, "Failed to send READY=1 notification message: %m");
}
m->sent_notify_ready = true;
log_debug("Sent READY=1 notification.");
} else if (m->send_watchdog) {
static const char p[] = "WATCHDOG=1";
if (send(m->notify_fd, p, strlen(p), MSG_DONTWAIT) < 0) {
if (errno == EAGAIN)
return 0;
return log_error_errno(errno, "Failed to send WATCHDOG=1 notification message: %m");
}
m->send_watchdog = false;
log_debug("Sent WATCHDOG=1 notification.");
} else if (m->stdout_streams_notify_queue)
/* Dispatch one stream notification event */
stdout_stream_send_notify(m->stdout_streams_notify_queue);
/* Leave us enabled if there's still more to do. */
if (m->send_watchdog || m->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) {
Manager *m = ASSERT_PTR(userdata);
int r;
m->send_watchdog = true;
r = sd_event_source_set_enabled(m->notify_event_source, SD_EVENT_ON);