chiark / gitweb /
journald: augment journal entries from the kernel with data from udev
[elogind.git] / src / journal / journald.c
index 5ecb7f72d34a4e80742d51b78a82938bfd584022..2429dd3e278380e229b447665eed6a51587b9689 100644 (file)
 #include <sys/ioctl.h>
 #include <linux/sockios.h>
 #include <sys/statvfs.h>
+#include <sys/mman.h>
 
+#include <libudev.h>
 #include <systemd/sd-journal.h>
-#include <systemd/sd-login.h>
 #include <systemd/sd-messages.h>
 #include <systemd/sd-daemon.h>
 
+#ifdef HAVE_LOGIND
+#include <systemd/sd-login.h>
+#endif
+
 #include "mkdir.h"
 #include "hashmap.h"
 #include "journal-file.h"
@@ -43,6 +48,8 @@
 #include "list.h"
 #include "journal-rate-limit.h"
 #include "journal-internal.h"
+#include "journal-vacuum.h"
+#include "journal-authenticate.h"
 #include "conf-parser.h"
 #include "journald.h"
 #include "virt.h"
 
 #define RECHECK_AVAILABLE_SPACE_USEC (30*USEC_PER_SEC)
 
-#define RECHECK_VAR_AVAILABLE_USEC (30*USEC_PER_SEC)
-
 #define N_IOVEC_META_FIELDS 17
+#define N_IOVEC_KERNEL_FIELDS 64
+#define N_IOVEC_UDEV_FIELDS 32
 
 #define ENTRY_SIZE_MAX (1024*1024*32)
 
 typedef enum StdoutStreamState {
         STDOUT_STREAM_IDENTIFIER,
+        STDOUT_STREAM_UNIT_ID,
         STDOUT_STREAM_PRIORITY,
         STDOUT_STREAM_LEVEL_PREFIX,
         STDOUT_STREAM_FORWARD_TO_SYSLOG,
@@ -94,6 +102,7 @@ struct StdoutStream {
 #endif
 
         char *identifier;
+        char *unit_id;
         int priority;
         bool level_prefix:1;
         bool forward_to_syslog:1;
@@ -106,7 +115,15 @@ struct StdoutStream {
         LIST_FIELDS(StdoutStream, stdout_stream);
 };
 
-static int server_flush_to_var(Server *s);
+static const char* const storage_table[] = {
+        [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 uint64_t available_space(Server *s) {
         char ids[33], *p;
@@ -268,7 +285,6 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
         char *p;
         int r;
         JournalFile *f;
-        char ids[33];
         sd_id128_t machine;
 
         assert(s);
@@ -292,7 +308,8 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
         if (f)
                 return f;
 
-        if (asprintf(&p, "/var/log/journal/%s/user-%lu.journal", sd_id128_to_string(machine, ids), (unsigned long) uid) < 0)
+        if (asprintf(&p, "/var/log/journal/" SD_ID128_FORMAT_STR "/user-%lu.journal",
+                     SD_ID128_FORMAT_VAL(machine), (unsigned long) uid) < 0)
                 return s->system_journal;
 
         while (hashmap_size(s->user_journals) >= USER_JOURNALS_MAX) {
@@ -302,7 +319,7 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
                 journal_file_close(f);
         }
 
-        r = journal_file_open_reliably(p, O_RDWR|O_CREAT, 0640, s->system_journal, &f);
+        r = journal_file_open_reliably(p, O_RDWR|O_CREAT, 0640, s->compress, s->seal, &s->system_metrics, s->mmap, s->system_journal, &f);
         free(p);
 
         if (r < 0)
@@ -328,7 +345,7 @@ static void server_rotate(Server *s) {
         log_info("Rotating...");
 
         if (s->runtime_journal) {
-                r = journal_file_rotate(&s->runtime_journal);
+                r = journal_file_rotate(&s->runtime_journal, s->compress, false);
                 if (r < 0)
                         if (s->runtime_journal)
                                 log_error("Failed to rotate %s: %s", s->runtime_journal->path, strerror(-r));
@@ -339,7 +356,7 @@ static void server_rotate(Server *s) {
         }
 
         if (s->system_journal) {
-                r = journal_file_rotate(&s->system_journal);
+                r = journal_file_rotate(&s->system_journal, s->compress, s->seal);
                 if (r < 0)
                         if (s->system_journal)
                                 log_error("Failed to rotate %s: %s", s->system_journal->path, strerror(-r));
@@ -351,7 +368,7 @@ static void server_rotate(Server *s) {
         }
 
         HASHMAP_FOREACH_KEY(f, k, s->user_journals, i) {
-                r = journal_file_rotate(&f);
+                r = journal_file_rotate(&f, s->compress, s->seal);
                 if (r < 0)
                         if (f->path)
                                 log_error("Failed to rotate %s: %s", f->path, strerror(-r));
@@ -382,7 +399,7 @@ static void server_vacuum(Server *s) {
 
         if (s->system_journal) {
                 if (asprintf(&p, "/var/log/journal/%s", ids) < 0) {
-                        log_error("Out of memory.");
+                        log_oom();
                         return;
                 }
 
@@ -392,10 +409,9 @@ static void server_vacuum(Server *s) {
                 free(p);
         }
 
-
         if (s->runtime_journal) {
                 if (asprintf(&p, "/run/log/journal/%s", ids) < 0) {
-                        log_error("Out of memory.");
+                        log_oom();
                         return;
                 }
 
@@ -450,12 +466,78 @@ static char *shortened_cgroup_path(pid_t pid) {
         return path;
 }
 
+static void write_to_journal(Server *s, uid_t uid, struct iovec *iovec, unsigned n) {
+        JournalFile *f;
+        bool vacuumed = false;
+        int r;
+
+        assert(s);
+        assert(iovec);
+        assert(n > 0);
+
+        f = find_journal(s, uid);
+        if (!f)
+                return;
+
+        if (journal_file_rotate_suggested(f)) {
+                log_info("Journal header limits reached or header out-of-date, rotating.");
+                server_rotate(s);
+                server_vacuum(s);
+                vacuumed = true;
+
+                f = find_journal(s, uid);
+                if (!f)
+                        return;
+        }
+
+        for (;;) {
+                r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
+                if (r >= 0)
+                        return;
+
+                if (vacuumed ||
+                    (r != -E2BIG && /* hit limit */
+                     r != -EFBIG && /* hit fs limit */
+                     r != -EDQUOT && /* quota hit */
+                     r != -ENOSPC && /* disk full */
+                     r != -EBADMSG && /* corrupted */
+                     r != -ENODATA && /* truncated */
+                     r != -EHOSTDOWN && /* other machine */
+                     r != -EPROTONOSUPPORT && /* unsupported feature */
+                     r != -EBUSY && /* unclean shutdown */
+                     r != -ESHUTDOWN /* already archived */)) {
+                        log_error("Failed to write entry, ignoring: %s", strerror(-r));
+                        return;
+                }
+
+                if (r == -E2BIG || r == -EFBIG || r == EDQUOT || r == ENOSPC)
+                        log_info("Allocation limit reached, rotating.");
+                else if (r == -EHOSTDOWN)
+                        log_info("Journal file from other machine, rotating.");
+                else if (r == -EBUSY)
+                        log_info("Unlcean shutdown, rotating.");
+                else
+                        log_warning("Journal file corrupted, rotating.");
+
+                server_rotate(s);
+                server_vacuum(s);
+                vacuumed = true;
+
+                f = find_journal(s, uid);
+                if (!f)
+                        return;
+
+                log_info("Retrying write.");
+        }
+}
+
 static void dispatch_message_real(
                 Server *s,
                 struct iovec *iovec, unsigned n, unsigned m,
                 struct ucred *ucred,
                 struct timeval *tv,
-                const char *label, size_t label_len) {
+                const char *label, size_t label_len,
+                const char *unit_id) {
 
         char *pid = NULL, *uid = NULL, *gid = NULL,
                 *source_time = NULL, *boot_id = NULL, *machine_id = NULL,
@@ -469,8 +551,6 @@ static void dispatch_message_real(
         int r;
         char *t;
         uid_t loginuid = 0, realuid = 0;
-        JournalFile *f;
-        bool vacuumed = false;
 
         assert(s);
         assert(iovec);
@@ -479,7 +559,9 @@ static void dispatch_message_real(
 
         if (ucred) {
                 uint32_t audit;
+#ifdef HAVE_LOGIND
                 uid_t owner;
+#endif
 
                 realuid = ucred->uid;
 
@@ -538,6 +620,7 @@ static void dispatch_message_real(
                                 IOVEC_SET_STRING(iovec[n++], cgroup);
                 }
 
+#ifdef HAVE_LOGIND
                 if (sd_pid_get_session(ucred->pid, &t) >= 0) {
                         session = strappend("_SYSTEMD_SESSION=", t);
                         free(t);
@@ -546,17 +629,19 @@ static void dispatch_message_real(
                                 IOVEC_SET_STRING(iovec[n++], session);
                 }
 
-                if (sd_pid_get_unit(ucred->pid, &t) >= 0) {
-                        unit = strappend("_SYSTEMD_UNIT=", t);
-                        free(t);
-
-                        if (unit)
-                                IOVEC_SET_STRING(iovec[n++], unit);
-                }
-
                 if (sd_pid_get_owner_uid(ucred->uid, &owner) >= 0)
                         if (asprintf(&owner_uid, "_SYSTEMD_OWNER_UID=%lu", (unsigned long) owner) >= 0)
                                 IOVEC_SET_STRING(iovec[n++], owner_uid);
+#endif
+
+                if (cg_pid_get_unit(ucred->pid, &t) >= 0) {
+                        unit = strappend("_SYSTEMD_UNIT=", t);
+                        free(t);
+                } else if (unit_id)
+                        unit = strappend("_SYSTEMD_UNIT=", unit_id);
+
+                if (unit)
+                        IOVEC_SET_STRING(iovec[n++], unit);
 
 #ifdef HAVE_SELINUX
                 if (label) {
@@ -610,41 +695,7 @@ static void dispatch_message_real(
 
         assert(n <= m);
 
-        server_flush_to_var(s);
-
-retry:
-        f = find_journal(s, realuid == 0 ? 0 : loginuid);
-        if (!f)
-                log_warning("Dropping message, as we can't find a place to store the data.");
-        else {
-                r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
-
-                if ((r == -E2BIG || /* hit limit */
-                     r == -EFBIG || /* hit fs limit */
-                     r == -EDQUOT || /* quota hit */
-                     r == -ENOSPC || /* disk full */
-                     r == -EBADMSG || /* corrupted */
-                     r == -ENODATA || /* truncated */
-                     r == -EHOSTDOWN || /* other machine */
-                     r == -EPROTONOSUPPORT) && /* unsupported feature */
-                    !vacuumed) {
-
-                        if (r == -E2BIG)
-                                log_info("Allocation limit reached, rotating.");
-                        else
-                                log_warning("Journal file corrupted, rotating.");
-
-                        server_rotate(s);
-                        server_vacuum(s);
-                        vacuumed = true;
-
-                        log_info("Retrying write.");
-                        goto retry;
-                }
-
-                if (r < 0)
-                        log_error("Failed to write entry, ignoring: %s", strerror(-r));
-        }
+        write_to_journal(s, realuid == 0 ? 0 : loginuid, iovec, n);
 
         free(pid);
         free(uid);
@@ -676,7 +727,7 @@ static void driver_message(Server *s, sd_id128_t message_id, const char *format,
         assert(s);
         assert(format);
 
-        IOVEC_SET_STRING(iovec[n++], "PRIORITY=5");
+        IOVEC_SET_STRING(iovec[n++], "PRIORITY=6");
         IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=driver");
 
         memcpy(buffer, "MESSAGE=", 8);
@@ -695,7 +746,7 @@ static void driver_message(Server *s, sd_id128_t message_id, const char *format,
         ucred.uid = getuid();
         ucred.gid = getgid();
 
-        dispatch_message_real(s, iovec, n, ELEMENTSOF(iovec), &ucred, NULL, NULL, 0);
+        dispatch_message_real(s, iovec, n, ELEMENTSOF(iovec), &ucred, NULL, NULL, 0, NULL);
 }
 
 static void dispatch_message(Server *s,
@@ -703,6 +754,7 @@ static void dispatch_message(Server *s,
                              struct ucred *ucred,
                              struct timeval *tv,
                              const char *label, size_t label_len,
+                             const char *unit_id,
                              int priority) {
         int rl;
         char *path = NULL, *c;
@@ -713,6 +765,9 @@ static void dispatch_message(Server *s,
         if (n == 0)
                 return;
 
+        if (LOG_PRI(priority) > s->max_level_store)
+                return;
+
         if (!ucred)
                 goto finish;
 
@@ -750,7 +805,7 @@ static void dispatch_message(Server *s,
         free(path);
 
 finish:
-        dispatch_message_real(s, iovec, n, m, ucred, tv, label, label_len);
+        dispatch_message_real(s, iovec, n, m, ucred, tv, label, label_len, unit_id);
 }
 
 static void forward_syslog_iovec(Server *s, const struct iovec *iovec, unsigned n_iovec, struct ucred *ucred, struct timeval *tv) {
@@ -819,15 +874,19 @@ static void forward_syslog_iovec(Server *s, const struct iovec *iovec, unsigned
                         return;
         }
 
-        log_debug("Failed to forward syslog message: %m");
+        if (errno != ENOENT)
+                log_debug("Failed to forward syslog message: %m");
 }
 
-static void forward_syslog_raw(Server *s, const char *buffer, struct ucred *ucred, struct timeval *tv) {
+static void forward_syslog_raw(Server *s, int priority, const char *buffer, struct ucred *ucred, struct timeval *tv) {
         struct iovec iovec;
 
         assert(s);
         assert(buffer);
 
+        if (LOG_PRI(priority) > s->max_level_syslog)
+                return;
+
         IOVEC_SET_STRING(iovec, buffer);
         forward_syslog_iovec(s, &iovec, 1, ucred, tv);
 }
@@ -845,6 +904,9 @@ static void forward_syslog(Server *s, int priority, const char *identifier, cons
         assert(priority <= 999);
         assert(message);
 
+        if (LOG_PRI(priority) > s->max_level_syslog)
+                return;
+
         /* First: priority field */
         snprintf(header_priority, sizeof(header_priority), "<%i>", priority);
         char_array_0(header_priority);
@@ -899,13 +961,18 @@ static void forward_kmsg(Server *s, int priority, const char *identifier, const
         char header_priority[6], header_pid[16];
         int n = 0;
         char *ident_buf = NULL;
-        int fd;
 
         assert(s);
         assert(priority >= 0);
         assert(priority <= 999);
         assert(message);
 
+        if (_unlikely_(LOG_PRI(priority) > s->max_level_kmsg))
+                return;
+
+        if (_unlikely_(s->dev_kmsg_fd < 0))
+                return;
+
         /* Never allow messages with kernel facility to be written to
          * kmsg, regardless where the data comes from. */
         priority = fixup_priority(priority);
@@ -938,30 +1005,25 @@ static void forward_kmsg(Server *s, int priority, const char *identifier, const
         IOVEC_SET_STRING(iovec[n++], message);
         IOVEC_SET_STRING(iovec[n++], "\n");
 
-        fd = open("/dev/kmsg", O_WRONLY|O_NOCTTY|O_CLOEXEC);
-        if (fd < 0) {
-                log_debug("Failed to open /dev/kmsg for logging: %s", strerror(errno));
-                goto finish;
-        }
-
-        if (writev(fd, iovec, n) < 0)
+        if (writev(s->dev_kmsg_fd, iovec, n) < 0)
                 log_debug("Failed to write to /dev/kmsg for logging: %s", strerror(errno));
 
-        close_nointr_nofail(fd);
-
-finish:
         free(ident_buf);
 }
 
-static void forward_console(Server *s, const char *identifier, const char *message, struct ucred *ucred) {
+static void forward_console(Server *s, int priority, const char *identifier, const char *message, struct ucred *ucred) {
         struct iovec iovec[4];
         char header_pid[16];
         int n = 0, fd;
         char *ident_buf = NULL;
+        const char *tty;
 
         assert(s);
         assert(message);
 
+        if (LOG_PRI(priority) > s->max_level_console)
+                return;
+
         /* First: identifier and PID */
         if (ucred) {
                 if (!identifier) {
@@ -985,14 +1047,16 @@ static void forward_console(Server *s, const char *identifier, const char *messa
         IOVEC_SET_STRING(iovec[n++], message);
         IOVEC_SET_STRING(iovec[n++], "\n");
 
-        fd = open_terminal("/dev/console", O_WRONLY|O_NOCTTY|O_CLOEXEC);
+        tty = s->tty_path ? s->tty_path : "/dev/console";
+
+        fd = open_terminal(tty, O_WRONLY|O_NOCTTY|O_CLOEXEC);
         if (fd < 0) {
-                log_debug("Failed to open /dev/console for logging: %s", strerror(errno));
+                log_debug("Failed to open %s for logging: %s", tty, strerror(errno));
                 goto finish;
         }
 
         if (writev(fd, iovec, n) < 0)
-                log_debug("Failed to write to /dev/console for logging: %s", strerror(errno));
+                log_debug("Failed to write to %s for logging: %s", tty, strerror(errno));
 
         close_nointr_nofail(fd);
 
@@ -1056,14 +1120,17 @@ static void process_syslog_message(Server *s, const char *buf, struct ucred *ucr
         unsigned n = 0;
         int priority = LOG_USER | LOG_INFO;
         char *identifier = NULL, *pid = NULL;
+        const char *orig;
 
         assert(s);
         assert(buf);
 
+        orig = buf;
+        parse_syslog_priority((char**) &buf, &priority);
+
         if (s->forward_to_syslog)
-                forward_syslog_raw(s, buf, ucred, tv);
+                forward_syslog_raw(s, priority, orig, ucred, tv);
 
-        parse_syslog_priority((char**) &buf, &priority);
         skip_syslog_date((char**) &buf);
         read_identifier(&buf, &identifier, &pid);
 
@@ -1071,7 +1138,7 @@ static void process_syslog_message(Server *s, const char *buf, struct ucred *ucr
                 forward_kmsg(s, priority, identifier, buf, ucred);
 
         if (s->forward_to_console)
-                forward_console(s, identifier, buf, ucred);
+                forward_console(s, priority, identifier, buf, ucred);
 
         IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=syslog");
 
@@ -1098,7 +1165,7 @@ static void process_syslog_message(Server *s, const char *buf, struct ucred *ucr
         if (message)
                 IOVEC_SET_STRING(iovec[n++], message);
 
-        dispatch_message(s, iovec, n, ELEMENTSOF(iovec), ucred, tv, label, label_len, priority);
+        dispatch_message(s, iovec, n, ELEMENTSOF(iovec), ucred, tv, label, label_len, NULL, priority);
 
         free(message);
         free(identifier);
@@ -1106,6 +1173,7 @@ static void process_syslog_message(Server *s, const char *buf, struct ucred *ucr
         free(syslog_priority);
         free(syslog_facility);
         free(syslog_identifier);
+        free(syslog_pid);
 }
 
 static bool valid_user_field(const char *p, size_t l) {
@@ -1176,7 +1244,7 @@ static void process_native_message(
 
                 if (e == p) {
                         /* Entry separator */
-                        dispatch_message(s, iovec, n, m, ucred, tv, label, label_len, priority);
+                        dispatch_message(s, iovec, n, m, ucred, tv, label, label_len, NULL, priority);
                         n = 0;
                         priority = LOG_INFO;
 
@@ -1202,7 +1270,7 @@ static void process_native_message(
                         u = MAX((n+N_IOVEC_META_FIELDS+1) * 2U, 4U);
                         c = realloc(iovec, u * sizeof(struct iovec));
                         if (!c) {
-                                log_error("Out of memory");
+                                log_oom();
                                 break;
                         }
 
@@ -1289,7 +1357,7 @@ static void process_native_message(
 
                         k = malloc((e - p) + 1 + l);
                         if (!k) {
-                                log_error("Out of memory");
+                                log_oom();
                                 break;
                         }
 
@@ -1323,10 +1391,10 @@ static void process_native_message(
                         forward_kmsg(s, priority, identifier, message, ucred);
 
                 if (s->forward_to_console)
-                        forward_console(s, identifier, message, ucred);
+                        forward_console(s, priority, identifier, message, ucred);
         }
 
-        dispatch_message(s, iovec, n, m, ucred, tv, label, label_len, priority);
+        dispatch_message(s, iovec, n, m, ucred, tv, label, label_len, NULL, priority);
 
 finish:
         for (j = 0; j < n; j++)  {
@@ -1382,7 +1450,7 @@ static void process_native_file(
 
         p = malloc(st.st_size);
         if (!p) {
-                log_error("Out of memory");
+                log_oom();
                 return;
         }
 
@@ -1421,7 +1489,7 @@ static int stdout_stream_log(StdoutStream *s, const char *p) {
                 forward_kmsg(s->server, priority, s->identifier, p, &s->ucred);
 
         if (s->forward_to_console || s->server->forward_to_console)
-                forward_console(s->server, s->identifier, p, &s->ucred);
+                forward_console(s->server, priority, s->identifier, p, &s->ucred);
 
         IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=stdout");
 
@@ -1449,7 +1517,7 @@ static int stdout_stream_log(StdoutStream *s, const char *p) {
         }
 #endif
 
-        dispatch_message(s->server, iovec, n, ELEMENTSOF(iovec), &s->ucred, NULL, label, label_len, priority);
+        dispatch_message(s->server, iovec, n, ELEMENTSOF(iovec), &s->ucred, NULL, label, label_len, s->unit_id, priority);
 
         free(message);
         free(syslog_priority);
@@ -1474,9 +1542,21 @@ static int stdout_stream_line(StdoutStream *s, char *p) {
                         s->identifier = NULL;
                 else  {
                         s->identifier = strdup(p);
-                        if (!s->identifier) {
-                                log_error("Out of memory");
-                                return -ENOMEM;
+                        if (!s->identifier)
+                                return log_oom();
+                }
+
+                s->state = STDOUT_STREAM_UNIT_ID;
+                return 0;
+
+        case STDOUT_STREAM_UNIT_ID:
+                if (s->ucred.uid == 0) {
+                        if (isempty(p))
+                                s->unit_id = NULL;
+                        else  {
+                                s->unit_id = strdup(p);
+                                if (!s->unit_id)
+                                        return log_oom();
                         }
                 }
 
@@ -1677,9 +1757,8 @@ static int stdout_stream_new(Server *s) {
 
         stream = new0(StdoutStream, 1);
         if (!stream) {
-                log_error("Out of memory.");
                 close_nointr_nofail(fd);
-                return -ENOMEM;
+                return log_oom();
         }
 
         stream->fd = fd;
@@ -1722,110 +1801,195 @@ fail:
         return r;
 }
 
-static int parse_kernel_timestamp(char **_p, usec_t *t) {
-        usec_t r;
-        int k, i;
-        char *p;
+static bool is_us(const char *pid) {
+        pid_t t;
 
-        assert(_p);
-        assert(*_p);
-        assert(t);
+        assert(pid);
 
-        p = *_p;
+        if (parse_pid(pid, &t) < 0)
+                return false;
 
-        if (strlen(p) < 14 || p[0] != '[' || p[13] != ']' || p[6] != '.')
-                return 0;
+        return t == getpid();
+}
 
-        r = 0;
+static void dev_kmsg_record(Server *s, char *p, size_t l) {
+        struct iovec iovec[N_IOVEC_META_FIELDS + 7 + N_IOVEC_KERNEL_FIELDS + 2 + N_IOVEC_UDEV_FIELDS];
+        char *message = NULL, *syslog_priority = NULL, *syslog_pid = NULL, *syslog_facility = NULL, *syslog_identifier = NULL, *source_time = NULL;
+        int priority, r;
+        unsigned n = 0, z = 0, j;
+        usec_t usec;
+        char *identifier = NULL, *pid = NULL, *e, *f, *k;
+        uint64_t serial;
+        size_t pl;
+        char *kernel_device = NULL;
+
+        assert(s);
+        assert(p);
 
-        for (i = 1; i <= 5; i++) {
-                r *= 10;
+        if (l <= 0)
+                return;
 
-                if (p[i] == ' ')
-                        continue;
+        e = memchr(p, ',', l);
+        if (!e)
+                return;
+        *e = 0;
 
-                k = undecchar(p[i]);
-                if (k < 0)
-                        return 0;
+        r = safe_atoi(p, &priority);
+        if (r < 0 || priority < 0 || priority > 999)
+                return;
 
-                r += k;
-        }
+        if (s->forward_to_kmsg && (priority & LOG_FACMASK) != LOG_KERN)
+                return;
 
-        for (i = 7; i <= 12; i++) {
-                r *= 10;
+        l -= (e - p) + 1;
+        p = e + 1;
+        e = memchr(p, ',', l);
+        if (!e)
+                return;
+        *e = 0;
 
-                k = undecchar(p[i]);
-                if (k < 0)
-                        return 0;
+        r = safe_atou64(p, &serial);
+        if (r < 0)
+                return;
 
-                r += k;
+        if (s->kernel_seqnum) {
+                /* We already read this one? */
+                if (serial < *s->kernel_seqnum)
+                        return;
+
+                /* Did we lose any? */
+                if (serial > *s->kernel_seqnum)
+                        driver_message(s, SD_MESSAGE_JOURNAL_MISSED, "Missed %llu kernel messages", (unsigned long long) serial - *s->kernel_seqnum - 1);
+
+                /* Make sure we never read this one again. Note that
+                 * we always store the next message serial we expect
+                 * here, simply because this makes handling the first
+                 * message with serial 0 easy. */
+                *s->kernel_seqnum = serial + 1;
         }
 
-        *t = r;
-        *_p += 14;
-        *_p += strspn(*_p, WHITESPACE);
+        l -= (e - p) + 1;
+        p = e + 1;
+        f = memchr(p, ';', l);
+        if (!f)
+                return;
+        /* Kernel 3.6 has the flags field, kernel 3.5 lacks that */
+        e = memchr(p, ',', l);
+        if (!e || f < e)
+                e = f;
+        *e = 0;
 
-        return 1;
-}
+        r = parse_usec(p, &usec);
+        if (r < 0)
+                return;
 
-static bool is_us(const char *pid) {
-        pid_t t;
+        l -= (f - p) + 1;
+        p = f + 1;
+        e = memchr(p, '\n', l);
+        if (!e)
+                return;
+        *e = 0;
 
-        assert(pid);
+        pl = e - p;
+        l -= (e - p) + 1;
+        k = e + 1;
 
-        if (parse_pid(pid, &t) < 0)
-                return false;
+        for (j = 0; l > 0 && j < N_IOVEC_KERNEL_FIELDS; j++) {
+                char *m;
+                /* Meta data fields attached */
 
-        return t == getpid();
-}
+                if (*k != ' ')
+                        break;
 
-static void proc_kmsg_line(Server *s, const char *p) {
-        struct iovec iovec[N_IOVEC_META_FIELDS + 7];
-        char *message = NULL, *syslog_priority = NULL, *syslog_pid = NULL, *syslog_facility = NULL, *syslog_identifier = NULL, *source_time = NULL;
-        int priority = LOG_KERN | LOG_INFO;
-        unsigned n = 0;
-        usec_t usec;
-        char *identifier = NULL, *pid = NULL;
+                k ++, l --;
 
-        assert(s);
-        assert(p);
+                e = memchr(k, '\n', l);
+                if (!e)
+                        return;
 
-        if (isempty(p))
-                return;
+                *e = 0;
+
+                m = cunescape_length_with_prefix(k, e - k, "_KERNEL_");
+                if (!m)
+                        break;
 
-        parse_syslog_priority((char **) &p, &priority);
+                if (startswith(m, "_KERNEL_DEVICE="))
+                        kernel_device = m + 15;
 
-        if (s->forward_to_kmsg && (priority & LOG_FACMASK) != LOG_KERN)
-                return;
+                IOVEC_SET_STRING(iovec[n++], m);
+                z++;
 
-        if (parse_kernel_timestamp((char **) &p, &usec) > 0) {
-                if (asprintf(&source_time, "_SOURCE_MONOTONIC_TIMESTAMP=%llu",
-                             (unsigned long long) usec) >= 0)
-                        IOVEC_SET_STRING(iovec[n++], source_time);
+                l -= (e - k) + 1;
+                k = e + 1;
+        }
+
+        if (kernel_device) {
+                struct udev_device *ud;
+
+                ud = udev_device_new_from_device_id(s->udev, kernel_device);
+                if (ud) {
+                        const char *g;
+                        struct udev_list_entry *ll;
+                        char *b;
+
+                        g = udev_device_get_devnode(ud);
+                        if (g) {
+                                b = strappend("_UDEV_DEVNODE=", g);
+                                if (b) {
+                                        IOVEC_SET_STRING(iovec[n++], b);
+                                        z++;
+                                }
+                        }
+
+                        g = udev_device_get_sysname(ud);
+                        if (g) {
+                                b = strappend("_UDEV_SYSNAME=", g);
+                                if (b) {
+                                        IOVEC_SET_STRING(iovec[n++], b);
+                                        z++;
+                                }
+                        }
+
+                        j = 0;
+                        ll = udev_device_get_devlinks_list_entry(ud);
+                        udev_list_entry_foreach(ll, ll) {
+
+                                if (j > N_IOVEC_UDEV_FIELDS)
+                                        break;
+
+                                g = udev_list_entry_get_name(ll);
+                                b = strappend("_UDEV_DEVLINK=", g);
+                                if (g) {
+                                        IOVEC_SET_STRING(iovec[n++], b);
+                                        z++;
+                                }
+
+                                j++;
+                        }
+
+                        udev_device_unref(ud);
+                }
         }
 
+        if (asprintf(&source_time, "_SOURCE_MONOTONIC_TIMESTAMP=%llu",
+                     (unsigned long long) usec) >= 0)
+                IOVEC_SET_STRING(iovec[n++], source_time);
+
         IOVEC_SET_STRING(iovec[n++], "_TRANSPORT=kernel");
 
         if (asprintf(&syslog_priority, "PRIORITY=%i", priority & LOG_PRIMASK) >= 0)
                 IOVEC_SET_STRING(iovec[n++], syslog_priority);
 
-        if ((priority & LOG_FACMASK) == LOG_KERN) {
-
-                if (s->forward_to_syslog)
-                        forward_syslog(s, priority, "kernel", p, NULL, NULL);
-
+        if ((priority & LOG_FACMASK) == LOG_KERN)
                 IOVEC_SET_STRING(iovec[n++], "SYSLOG_IDENTIFIER=kernel");
-        else {
-                read_identifier(&p, &identifier, &pid);
+        else {
+                read_identifier((const char**) &p, &identifier, &pid);
 
                 /* Avoid any messages we generated ourselves via
                  * log_info() and friends. */
-                if (is_us(pid))
+                if (pid && is_us(pid))
                         goto finish;
 
-                if (s->forward_to_syslog)
-                        forward_syslog(s, priority, identifier, p, NULL, NULL);
-
                 if (identifier) {
                         syslog_identifier = strappend("SYSLOG_IDENTIFIER=", identifier);
                         if (syslog_identifier)
@@ -1842,13 +2006,16 @@ static void proc_kmsg_line(Server *s, const char *p) {
                         IOVEC_SET_STRING(iovec[n++], syslog_facility);
         }
 
-        message = strappend("MESSAGE=", p);
+        message = cunescape_length_with_prefix(p, pl, "MESSAGE=");
         if (message)
                 IOVEC_SET_STRING(iovec[n++], message);
 
-        dispatch_message(s, iovec, n, ELEMENTSOF(iovec), NULL, NULL, NULL, 0, priority);
+        dispatch_message(s, iovec, n, ELEMENTSOF(iovec), NULL, NULL, NULL, 0, NULL, priority);
 
 finish:
+        for (j = 0; j < z; j++)
+                free(iovec[j].iov_base);
+
         free(message);
         free(syslog_priority);
         free(syslog_identifier);
@@ -1859,41 +2026,6 @@ finish:
         free(pid);
 }
 
-static void proc_kmsg_scan(Server *s) {
-        char *p;
-        size_t remaining;
-
-        assert(s);
-
-        p = s->proc_kmsg_buffer;
-        remaining = s->proc_kmsg_length;
-        for (;;) {
-                char *end;
-                size_t skip;
-
-                end = memchr(p, '\n', remaining);
-                if (end)
-                        skip = end - p + 1;
-                else if (remaining >= sizeof(s->proc_kmsg_buffer) - 1) {
-                        end = p + sizeof(s->proc_kmsg_buffer) - 1;
-                        skip = remaining;
-                } else
-                        break;
-
-                *end = 0;
-
-                proc_kmsg_line(s, p);
-
-                remaining -= skip;
-                p += skip;
-        }
-
-        if (p > s->proc_kmsg_buffer) {
-                memmove(s->proc_kmsg_buffer, p, remaining);
-                s->proc_kmsg_length = remaining;
-        }
-}
-
 static int system_journal_open(Server *s) {
         int r;
         char *fn;
@@ -1906,31 +2038,36 @@ static int system_journal_open(Server *s) {
 
         sd_id128_to_string(machine, ids);
 
-        if (!s->system_journal) {
+        if (!s->system_journal &&
+            (s->storage == STORAGE_PERSISTENT || s->storage == STORAGE_AUTO) &&
+            access("/run/systemd/journal/flushed", F_OK) >= 0) {
+
+                /* If in auto mode: first try to create the machine
+                 * path, but not the prefix.
+                 *
+                 * If in persistent mode: create /var/log/journal and
+                 * the machine path */
+
+                if (s->storage == STORAGE_PERSISTENT)
+                        (void) mkdir("/var/log/journal/", 0755);
 
-                /* First try to create the machine path, but not the prefix */
                 fn = strappend("/var/log/journal/", ids);
                 if (!fn)
                         return -ENOMEM;
+
                 (void) mkdir(fn, 0755);
                 free(fn);
 
-                /* The create the system journal file */
-                fn = join("/var/log/journal/", ids, "/system.journal", NULL);
+                fn = strjoin("/var/log/journal/", ids, "/system.journal", NULL);
                 if (!fn)
                         return -ENOMEM;
 
-                r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, NULL, &s->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);
                 free(fn);
 
-                if (r >= 0) {
-                        journal_default_metrics(&s->system_metrics, s->system_journal->fd);
-
-                        s->system_journal->metrics = s->system_metrics;
-                        s->system_journal->compress = s->compress;
-
+                if (r >= 0)
                         server_fix_perms(s, s->system_journal, 0);
-                else if (r < 0) {
+                else if (r < 0) {
 
                         if (r != -ENOENT && r != -EROFS)
                                 log_warning("Failed to open system journal: %s", strerror(-r));
@@ -1939,9 +2076,10 @@ static int system_journal_open(Server *s) {
                 }
         }
 
-        if (!s->runtime_journal) {
+        if (!s->runtime_journal &&
+            (s->storage != STORAGE_NONE)) {
 
-                fn = join("/run/log/journal/", ids, "/system.journal", NULL);
+                fn = strjoin("/run/log/journal/", ids, "/system.journal", NULL);
                 if (!fn)
                         return -ENOMEM;
 
@@ -1951,7 +2089,7 @@ static int system_journal_open(Server *s) {
                          * if it already exists, so that we can flush
                          * it into the system journal */
 
-                        r = journal_file_open(fn, O_RDWR, 0640, NULL, &s->runtime_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) {
@@ -1967,7 +2105,7 @@ static int system_journal_open(Server *s) {
                          * it if necessary. */
 
                         (void) mkdir_parents(fn, 0755);
-                        r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
+                        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) {
@@ -1976,38 +2114,28 @@ static int system_journal_open(Server *s) {
                         }
                 }
 
-                if (s->runtime_journal) {
-                        journal_default_metrics(&s->runtime_metrics, s->runtime_journal->fd);
-
-                        s->runtime_journal->metrics = s->runtime_metrics;
-                        s->runtime_journal->compress = s->compress;
-
+                if (s->runtime_journal)
                         server_fix_perms(s, s->runtime_journal, 0);
-                }
         }
 
         return r;
 }
 
 static int server_flush_to_var(Server *s) {
-        char path[] = "/run/log/journal/xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";
         Object *o = NULL;
         int r;
         sd_id128_t machine;
         sd_journal *j;
-        usec_t ts;
 
         assert(s);
 
-        if (!s->runtime_journal)
+        if (s->storage != STORAGE_AUTO &&
+            s->storage != STORAGE_PERSISTENT)
                 return 0;
 
-        ts = now(CLOCK_MONOTONIC);
-        if (s->var_available_timestamp + RECHECK_VAR_AVAILABLE_USEC > ts)
+        if (!s->runtime_journal)
                 return 0;
 
-        s->var_available_timestamp = ts;
-
         system_journal_open(s);
 
         if (!s->system_journal)
@@ -2062,47 +2190,57 @@ finish:
         journal_file_close(s->runtime_journal);
         s->runtime_journal = NULL;
 
-        if (r >= 0) {
-                sd_id128_to_string(machine, path + 17);
-                rm_rf(path, false, true, false);
-        }
+        if (r >= 0)
+                rm_rf("/run/log/journal", false, true, false);
 
         return r;
 }
 
-static int server_read_proc_kmsg(Server *s) {
+static int server_read_dev_kmsg(Server *s) {
+        char buffer[8192+1]; /* the kernel-side limit per record is 8K currently */
         ssize_t l;
+
         assert(s);
-        assert(s->proc_kmsg_fd >= 0);
+        assert(s->dev_kmsg_fd >= 0);
 
-        l = read(s->proc_kmsg_fd, s->proc_kmsg_buffer + s->proc_kmsg_length, sizeof(s->proc_kmsg_buffer) - 1 - s->proc_kmsg_length);
+        l = read(s->dev_kmsg_fd, buffer, sizeof(buffer) - 1);
+        if (l == 0)
+                return 0;
         if (l < 0) {
+                /* Old kernels who don't allow reading from /dev/kmsg
+                 * return EINVAL when we try. So handle this cleanly,
+                 * but don' try to ever read from it again. */
+                if (errno == EINVAL) {
+                        epoll_ctl(s->epoll_fd, EPOLL_CTL_DEL, s->dev_kmsg_fd, NULL);
+                        return 0;
+                }
 
-                if (errno == EAGAIN || errno == EINTR)
+                if (errno == EAGAIN || errno == EINTR || errno == EPIPE)
                         return 0;
 
                 log_error("Failed to read from kernel: %m");
                 return -errno;
         }
 
-        s->proc_kmsg_length += l;
-
-        proc_kmsg_scan(s);
+        dev_kmsg_record(s, buffer, l);
         return 1;
 }
 
-static int server_flush_proc_kmsg(Server *s) {
+static int server_flush_dev_kmsg(Server *s) {
         int r;
 
         assert(s);
 
-        if (s->proc_kmsg_fd < 0)
+        if (s->dev_kmsg_fd < 0)
+                return 0;
+
+        if (!s->dev_kmsg_readable)
                 return 0;
 
-        log_info("Flushing /proc/kmsg...");
+        log_info("Flushing /dev/kmsg...");
 
         for (;;) {
-                r = server_read_proc_kmsg(s);
+                r = server_read_dev_kmsg(s);
                 if (r < 0)
                         return r;
 
@@ -2115,6 +2253,7 @@ static int server_flush_proc_kmsg(Server *s) {
 
 static int process_event(Server *s, struct epoll_event *ev) {
         assert(s);
+        assert(ev);
 
         if (ev->data.fd == s->signal_fd) {
                 struct signalfd_siginfo sfsi;
@@ -2137,15 +2276,23 @@ static int process_event(Server *s, struct epoll_event *ev) {
                         return -errno;
                 }
 
+                log_info("Received SIG%s", signal_to_string(sfsi.ssi_signo));
+
                 if (sfsi.ssi_signo == SIGUSR1) {
+                        touch("/run/systemd/journal/flushed");
                         server_flush_to_var(s);
-                        return 0;
+                        return 1;
+                }
+
+                if (sfsi.ssi_signo == SIGUSR2) {
+                        server_rotate(s);
+                        server_vacuum(s);
+                        return 1;
                 }
 
-                log_debug("Received SIG%s", signal_to_string(sfsi.ssi_signo));
                 return 0;
 
-        } else if (ev->data.fd == s->proc_kmsg_fd) {
+        } else if (ev->data.fd == s->dev_kmsg_fd) {
                 int r;
 
                 if (ev->events != EPOLLIN) {
@@ -2153,7 +2300,7 @@ static int process_event(Server *s, struct epoll_event *ev) {
                         return -EIO;
                 }
 
-                r = server_read_proc_kmsg(s);
+                r = server_read_dev_kmsg(s);
                 if (r < 0)
                         return r;
 
@@ -2498,28 +2645,68 @@ static int open_stdout_socket(Server *s) {
         return 0;
 }
 
-static int open_proc_kmsg(Server *s) {
+static int open_dev_kmsg(Server *s) {
         struct epoll_event ev;
 
         assert(s);
 
-        if (!s->import_proc_kmsg)
-                return 0;
-
-        s->proc_kmsg_fd = open("/proc/kmsg", O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
-        if (s->proc_kmsg_fd < 0) {
-                log_warning("Failed to open /proc/kmsg, ignoring: %m");
+        s->dev_kmsg_fd = open("/dev/kmsg", O_RDWR|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
+        if (s->dev_kmsg_fd < 0) {
+                log_warning("Failed to open /dev/kmsg, ignoring: %m");
                 return 0;
         }
 
         zero(ev);
         ev.events = EPOLLIN;
-        ev.data.fd = s->proc_kmsg_fd;
-        if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->proc_kmsg_fd, &ev) < 0) {
-                log_error("Failed to add /proc/kmsg fd to epoll object: %m");
+        ev.data.fd = s->dev_kmsg_fd;
+        if (epoll_ctl(s->epoll_fd, EPOLL_CTL_ADD, s->dev_kmsg_fd, &ev) < 0) {
+
+                /* This will fail with EPERM on older kernels where
+                 * /dev/kmsg is not readable. */
+                if (errno == EPERM)
+                        return 0;
+
+                log_error("Failed to add /dev/kmsg fd to epoll object: %m");
                 return -errno;
         }
 
+        s->dev_kmsg_readable = true;
+
+        return 0;
+}
+
+static int open_kernel_seqnum(Server *s) {
+        int fd;
+        uint64_t *p;
+
+        assert(s);
+
+        /* We store the seqnum we last read in an mmaped file. That
+         * way we can just use it like a variable, but it is
+         * persistant and automatically flushed at reboot. */
+
+        fd = open("/run/systemd/journal/kernel-seqnum", O_RDWR|O_CREAT|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, 0644);
+        if (fd < 0) {
+                log_error("Failed to open /run/systemd/journal/kernel-seqnum, ignoring: %m");
+                return 0;
+        }
+
+        if (posix_fallocate(fd, 0, sizeof(uint64_t)) < 0) {
+                log_error("Failed to allocate sequential number file, ignoring: %m");
+                close_nointr_nofail(fd);
+                return 0;
+        }
+
+        p = mmap(NULL, sizeof(uint64_t), PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
+        if (p == MAP_FAILED) {
+                log_error("Failed to map sequential number file, ignoring: %m");
+                close_nointr_nofail(fd);
+                return 0;
+        }
+
+        close_nointr_nofail(fd);
+        s->kernel_seqnum = p;
+
         return 0;
 }
 
@@ -2530,7 +2717,7 @@ static int open_signalfd(Server *s) {
         assert(s);
 
         assert_se(sigemptyset(&mask) == 0);
-        sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, -1);
+        sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, SIGUSR2, -1);
         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 
         s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
@@ -2574,25 +2761,26 @@ static int server_parse_proc_cmdline(Server *s) {
                         goto finish;
                 }
 
-                if (startswith(word, "systemd_journald.forward_to_syslog=")) {
+                if (startswith(word, "systemd.journald.forward_to_syslog=")) {
                         r = parse_boolean(word + 35);
                         if (r < 0)
                                 log_warning("Failed to parse forward to syslog switch %s. Ignoring.", word + 35);
                         else
                                 s->forward_to_syslog = r;
-                } else if (startswith(word, "systemd_journald.forward_to_kmsg=")) {
+                } else if (startswith(word, "systemd.journald.forward_to_kmsg=")) {
                         r = parse_boolean(word + 33);
                         if (r < 0)
                                 log_warning("Failed to parse forward to kmsg switch %s. Ignoring.", word + 33);
                         else
                                 s->forward_to_kmsg = r;
-                } else if (startswith(word, "systemd_journald.forward_to_console=")) {
+                } else if (startswith(word, "systemd.journald.forward_to_console=")) {
                         r = parse_boolean(word + 36);
                         if (r < 0)
                                 log_warning("Failed to parse forward to console switch %s. Ignoring.", word + 36);
                         else
                                 s->forward_to_console = r;
-                }
+                } else if (startswith(word, "systemd.journald"))
+                        log_warning("Invalid systemd.journald parameter. Ignoring.");
 
                 free(word);
         }
@@ -2636,14 +2824,19 @@ static int server_init(Server *s) {
         assert(s);
 
         zero(*s);
-        s->syslog_fd = s->native_fd = s->stdout_fd = s->signal_fd = s->epoll_fd = s->proc_kmsg_fd = -1;
+        s->syslog_fd = s->native_fd = s->stdout_fd = s->signal_fd = s->epoll_fd = s->dev_kmsg_fd = -1;
         s->compress = true;
+        s->seal = true;
 
         s->rate_limit_interval = DEFAULT_RATE_LIMIT_INTERVAL;
         s->rate_limit_burst = DEFAULT_RATE_LIMIT_BURST;
 
         s->forward_to_syslog = true;
-        s->import_proc_kmsg = true;
+
+        s->max_level_store = LOG_DEBUG;
+        s->max_level_syslog = LOG_DEBUG;
+        s->max_level_kmsg = LOG_NOTICE;
+        s->max_level_console = LOG_INFO;
 
         memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics));
         memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics));
@@ -2651,11 +2844,15 @@ static int server_init(Server *s) {
         server_parse_config_file(s);
         server_parse_proc_cmdline(s);
 
+        mkdir_p("/run/systemd/journal", 0755);
+
         s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
-        if (!s->user_journals) {
-                log_error("Out of memory.");
-                return -ENOMEM;
-        }
+        if (!s->user_journals)
+                return log_oom();
+
+        s->mmap = mmap_cache_new();
+        if (!s->mmap)
+                return log_oom();
 
         s->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
         if (s->epoll_fd < 0) {
@@ -2716,7 +2913,11 @@ static int server_init(Server *s) {
         if (r < 0)
                 return r;
 
-        r = open_proc_kmsg(s);
+        r = open_dev_kmsg(s);
+        if (r < 0)
+                return r;
+
+        r = open_kernel_seqnum(s);
         if (r < 0)
                 return r;
 
@@ -2724,6 +2925,10 @@ static int server_init(Server *s) {
         if (r < 0)
                 return r;
 
+        s->udev = udev_new();
+        if (!s->udev)
+                return -ENOMEM;
+
         s->rate_limit = journal_rate_limit_new(s->rate_limit_interval, s->rate_limit_burst);
         if (!s->rate_limit)
                 return -ENOMEM;
@@ -2768,13 +2973,23 @@ static void server_done(Server *s) {
         if (s->stdout_fd >= 0)
                 close_nointr_nofail(s->stdout_fd);
 
-        if (s->proc_kmsg_fd >= 0)
-                close_nointr_nofail(s->proc_kmsg_fd);
+        if (s->dev_kmsg_fd >= 0)
+                close_nointr_nofail(s->dev_kmsg_fd);
 
         if (s->rate_limit)
                 journal_rate_limit_free(s->rate_limit);
 
+        if (s->kernel_seqnum)
+                munmap(s->kernel_seqnum, sizeof(uint64_t));
+
         free(s->buffer);
+        free(s->tty_path);
+
+        if (s->mmap)
+                mmap_cache_unref(s->mmap);
+
+        if (s->udev)
+                udev_unref(s->udev);
 }
 
 int main(int argc, char *argv[]) {
@@ -2793,6 +3008,7 @@ int main(int argc, char *argv[]) {
 
         log_set_target(LOG_TARGET_SAFE);
         log_set_facility(LOG_SYSLOG);
+        log_set_max_level(LOG_DEBUG);
         log_parse_environment();
         log_open();
 
@@ -2804,7 +3020,7 @@ int main(int argc, char *argv[]) {
 
         server_vacuum(&server);
         server_flush_to_var(&server);
-        server_flush_proc_kmsg(&server);
+        server_flush_dev_kmsg(&server);
 
         log_debug("systemd-journald running as pid %lu", (unsigned long) getpid());
         driver_message(&server, SD_MESSAGE_JOURNAL_START, "Journal started");
@@ -2815,8 +3031,26 @@ int main(int argc, char *argv[]) {
 
         for (;;) {
                 struct epoll_event event;
+                int t;
+
+#ifdef HAVE_GCRYPT
+                usec_t u;
+
+                if (server.system_journal &&
+                    journal_file_next_evolve_usec(server.system_journal, &u)) {
+                        usec_t n;
 
-                r = epoll_wait(server.epoll_fd, &event, 1, -1);
+                        n = now(CLOCK_MONOTONIC);
+
+                        if (n >= u)
+                                t = 0;
+                        else
+                                t = (int) ((u - n + USEC_PER_MSEC - 1) / USEC_PER_MSEC);
+                } else
+#endif
+                        t = -1;
+
+                r = epoll_wait(server.epoll_fd, &event, 1, t);
                 if (r < 0) {
 
                         if (errno == EINTR)
@@ -2825,14 +3059,20 @@ int main(int argc, char *argv[]) {
                         log_error("epoll_wait() failed: %m");
                         r = -errno;
                         goto finish;
-                } else if (r == 0)
-                        break;
+                }
 
-                r = process_event(&server, &event);
-                if (r < 0)
-                        goto finish;
-                else if (r == 0)
-                        break;
+                if (r > 0) {
+                        r = process_event(&server, &event);
+                        if (r < 0)
+                                goto finish;
+                        else if (r == 0)
+                                break;
+                }
+
+#ifdef HAVE_GCRYPT
+                if (server.system_journal)
+                        journal_file_maybe_append_tag(server.system_journal, 0);
+#endif
         }
 
         log_debug("systemd-journald stopped as pid %lu", (unsigned long) getpid());