chiark / gitweb /
journald: don't rotate on startup
[elogind.git] / src / journal / journald.c
index 5d3956ea460157c58ea97ab25f4fddba3cea697e..47fe6edb0cad2ac85ca42f972837551937c41595 100644 (file)
@@ -30,6 +30,7 @@
 #include <stddef.h>
 #include <sys/ioctl.h>
 #include <linux/sockios.h>
+#include <sys/statvfs.h>
 
 #include "hashmap.h"
 #include "journal-file.h"
 #include "acl-util.h"
 #include "cgroup-util.h"
 #include "list.h"
+#include "journal-rate-limit.h"
+#include "sd-journal.h"
+#include "sd-login.h"
+#include "journal-internal.h"
 
 #define USER_JOURNALS_MAX 1024
 #define STDOUT_STREAMS_MAX 4096
 
+#define DEFAULT_RATE_LIMIT_INTERVAL (10*USEC_PER_SEC)
+#define DEFAULT_RATE_LIMIT_BURST 200
+
+#define RECHECK_AVAILABLE_SPACE_USEC (30*USEC_PER_SEC)
+
+#define RECHECK_VAR_AVAILABLE_USEC (30*USEC_PER_SEC)
+
+#define SYSLOG_TIMEOUT_USEC (5*USEC_PER_SEC)
+
 typedef struct StdoutStream StdoutStream;
 
 typedef struct Server {
@@ -60,10 +74,18 @@ typedef struct Server {
         char *buffer;
         size_t buffer_size;
 
-        JournalMetrics metrics;
-        uint64_t max_use;
+        JournalRateLimit *rate_limit;
+
+        JournalMetrics runtime_metrics;
+        JournalMetrics system_metrics;
+
         bool compress;
 
+        uint64_t cached_available_space;
+        usec_t cached_available_space_timestamp;
+
+        uint64_t var_available_timestamp;
+
         LIST_HEAD(StdoutStream, stdout_streams);
         unsigned n_stdout_streams;
 } Server;
@@ -95,6 +117,92 @@ struct StdoutStream {
         LIST_FIELDS(StdoutStream, stdout_stream);
 };
 
+static int server_flush_to_var(Server *s);
+
+static uint64_t available_space(Server *s) {
+        char ids[33], *p;
+        const char *f;
+        sd_id128_t machine;
+        struct statvfs ss;
+        uint64_t sum = 0, avail = 0, ss_avail = 0;
+        int r;
+        DIR *d;
+        usec_t ts;
+        JournalMetrics *m;
+
+        ts = now(CLOCK_MONOTONIC);
+
+        if (s->cached_available_space_timestamp + RECHECK_AVAILABLE_SPACE_USEC > ts)
+                return s->cached_available_space;
+
+        r = sd_id128_get_machine(&machine);
+        if (r < 0)
+                return 0;
+
+        if (s->system_journal) {
+                f = "/var/log/journal/";
+                m = &s->system_metrics;
+        } else {
+                f = "/run/log/journal/";
+                m = &s->runtime_metrics;
+        }
+
+        assert(m);
+
+        p = strappend(f, sd_id128_to_string(machine, ids));
+        if (!p)
+                return 0;
+
+        d = opendir(p);
+        free(p);
+
+        if (!d)
+                return 0;
+
+        if (fstatvfs(dirfd(d), &ss) < 0)
+                goto finish;
+
+        for (;;) {
+                struct stat st;
+                struct dirent buf, *de;
+                int k;
+
+                k = readdir_r(d, &buf, &de);
+                if (k != 0) {
+                        r = -k;
+                        goto finish;
+                }
+
+                if (!de)
+                        break;
+
+                if (!dirent_is_file_with_suffix(de, ".journal"))
+                        continue;
+
+                if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
+                        continue;
+
+                sum += (uint64_t) st.st_blocks * (uint64_t) st.st_blksize;
+        }
+
+        avail = sum >= m->max_use ? 0 : m->max_use - sum;
+
+        ss_avail = ss.f_bsize * ss.f_bavail;
+
+        ss_avail = ss_avail < m->keep_free ? 0 : ss_avail - m->keep_free;
+
+        if (ss_avail < avail)
+                avail = ss_avail;
+
+        s->cached_available_space = avail;
+        s->cached_available_space_timestamp = ts;
+
+finish:
+        closedir(d);
+
+        return avail;
+}
+
 static void fix_perms(JournalFile *f, uid_t uid) {
         acl_t acl;
         acl_entry_t entry;
@@ -150,8 +258,12 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
 
         assert(s);
 
-        /* We split up user logs only on /var, not on /run */
-        if (!s->system_journal)
+        /* We split up user logs only on /var, not on /run. If the
+         * runtime file is open, we write to it exclusively, in order
+         * to guarantee proper order as soon as we flush /run to
+         * /var and close the runtime file. */
+
+        if (s->runtime_journal)
                 return s->runtime_journal;
 
         if (uid <= 0)
@@ -182,7 +294,7 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
                 return s->system_journal;
 
         fix_perms(f, uid);
-        f->metrics = s->metrics;
+        f->metrics = s->system_metrics;
         f->compress = s->compress;
 
         r = hashmap_put(s->user_journals, UINT32_TO_PTR(uid), f);
@@ -194,14 +306,11 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
         return f;
 }
 
-static void server_vacuum(Server *s) {
-        Iterator i;
+static void server_rotate(Server *s) {
+        JournalFile *f;
         void *k;
-        char *p;
-        char ids[33];
-        sd_id128_t machine;
+        Iterator i;
         int r;
-        JournalFile *f;
 
         log_info("Rotating...");
 
@@ -224,6 +333,13 @@ static void server_vacuum(Server *s) {
                 else
                         hashmap_replace(s->user_journals, k, f);
         }
+}
+
+static void server_vacuum(Server *s) {
+        char *p;
+        char ids[33];
+        sd_id128_t machine;
+        int r;
 
         log_info("Vacuuming...");
 
@@ -233,33 +349,89 @@ static void server_vacuum(Server *s) {
                 return;
         }
 
-        if (asprintf(&p, "/var/log/journal/%s", sd_id128_to_string(machine, ids)) < 0) {
-                log_error("Out of memory.");
-                return;
+        sd_id128_to_string(machine, ids);
+
+        if (s->system_journal) {
+                if (asprintf(&p, "/var/log/journal/%s", ids) < 0) {
+                        log_error("Out of memory.");
+                        return;
+                }
+
+                r = journal_directory_vacuum(p, s->system_metrics.max_use, s->system_metrics.keep_free);
+                if (r < 0 && r != -ENOENT)
+                        log_error("Failed to vacuum %s: %s", p, strerror(-r));
+                free(p);
         }
 
-        r = journal_directory_vacuum(p, s->max_use, s->metrics.keep_free);
-        if (r < 0 && r != -ENOENT)
-                log_error("Failed to vacuum %s: %s", p, strerror(-r));
-        free(p);
 
-        if (asprintf(&p, "/run/log/journal/%s", ids) < 0) {
-                log_error("Out of memory.");
-                return;
+        if (s->runtime_journal) {
+                if (asprintf(&p, "/run/log/journal/%s", ids) < 0) {
+                        log_error("Out of memory.");
+                        return;
+                }
+
+                r = journal_directory_vacuum(p, s->runtime_metrics.max_use, s->runtime_metrics.keep_free);
+                if (r < 0 && r != -ENOENT)
+                        log_error("Failed to vacuum %s: %s", p, strerror(-r));
+                free(p);
         }
 
-        r = journal_directory_vacuum(p, s->max_use, s->metrics.keep_free);
-        if (r < 0 && r != -ENOENT)
-                log_error("Failed to vacuum %s: %s", p, strerror(-r));
-        free(p);
+        s->cached_available_space_timestamp = 0;
+}
+
+static char *shortened_cgroup_path(pid_t pid) {
+        int r;
+        char *process_path, *init_path, *path;
+
+        assert(pid > 0);
+
+        r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, pid, &process_path);
+        if (r < 0)
+                return NULL;
+
+        r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 1, &init_path);
+        if (r < 0) {
+                free(process_path);
+                return NULL;
+        }
+
+        if (endswith(init_path, "/system"))
+                init_path[strlen(init_path) - 7] = 0;
+        else if (streq(init_path, "/"))
+                init_path[0] = 0;
+
+        if (startswith(process_path, init_path)) {
+                char *p;
+
+                p = strdup(process_path + strlen(init_path));
+                if (!p) {
+                        free(process_path);
+                        free(init_path);
+                        return NULL;
+                }
+                path = p;
+        } else {
+                path = process_path;
+                process_path = NULL;
+        }
+
+        free(process_path);
+        free(init_path);
+
+        return path;
 }
 
-static void dispatch_message(Server *s, struct iovec *iovec, unsigned n, unsigned m, struct ucred *ucred, struct timeval *tv) {
+static void dispatch_message_real(Server *s,
+                             struct iovec *iovec, unsigned n, unsigned m,
+                             struct ucred *ucred,
+                             struct timeval *tv) {
+
         char *pid = NULL, *uid = NULL, *gid = NULL,
                 *source_time = NULL, *boot_id = NULL, *machine_id = NULL,
                 *comm = NULL, *cmdline = NULL, *hostname = NULL,
                 *audit_session = NULL, *audit_loginuid = NULL,
-                *exe = NULL, *cgroup = NULL;
+                *exe = NULL, *cgroup = NULL, *session = NULL,
+                *owner_uid = NULL, *service = NULL;
 
         char idbuf[33];
         sd_id128_t id;
@@ -270,16 +442,13 @@ static void dispatch_message(Server *s, struct iovec *iovec, unsigned n, unsigne
         bool vacuumed = false;
 
         assert(s);
-        assert(iovec || n == 0);
-
-        if (n == 0)
-                return;
-
-        assert(n + 13 <= m);
+        assert(iovec);
+        assert(n > 0);
+        assert(n + 16 <= m);
 
         if (ucred) {
-                uint32_t session;
-                char *path;
+                uint32_t audit;
+                uid_t owner;
 
                 realuid = ucred->uid;
 
@@ -295,30 +464,33 @@ static void dispatch_message(Server *s, struct iovec *iovec, unsigned n, unsigne
                 r = get_process_comm(ucred->pid, &t);
                 if (r >= 0) {
                         comm = strappend("_COMM=", t);
+                        free(t);
+
                         if (comm)
                                 IOVEC_SET_STRING(iovec[n++], comm);
-                        free(t);
                 }
 
                 r = get_process_exe(ucred->pid, &t);
                 if (r >= 0) {
                         exe = strappend("_EXE=", t);
+                        free(t);
+
                         if (comm)
                                 IOVEC_SET_STRING(iovec[n++], exe);
-                        free(t);
                 }
 
                 r = get_process_cmdline(ucred->pid, LINE_MAX, false, &t);
                 if (r >= 0) {
                         cmdline = strappend("_CMDLINE=", t);
+                        free(t);
+
                         if (cmdline)
                                 IOVEC_SET_STRING(iovec[n++], cmdline);
-                        free(t);
                 }
 
-                r = audit_session_from_pid(ucred->pid, &session);
+                r = audit_session_from_pid(ucred->pid, &audit);
                 if (r >= 0)
-                        if (asprintf(&audit_session, "_AUDIT_SESSION=%lu", (unsigned long) session) >= 0)
+                        if (asprintf(&audit_session, "_AUDIT_SESSION=%lu", (unsigned long) audit) >= 0)
                                 IOVEC_SET_STRING(iovec[n++], audit_session);
 
                 r = audit_loginuid_from_pid(ucred->pid, &loginuid);
@@ -326,13 +498,34 @@ static void dispatch_message(Server *s, struct iovec *iovec, unsigned n, unsigne
                         if (asprintf(&audit_loginuid, "_AUDIT_LOGINUID=%lu", (unsigned long) loginuid) >= 0)
                                 IOVEC_SET_STRING(iovec[n++], audit_loginuid);
 
-                r = cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, ucred->pid, &path);
-                if (r >= 0) {
-                        cgroup = strappend("_SYSTEMD_CGROUP=", path);
+                t = shortened_cgroup_path(ucred->pid);
+                if (t) {
+                        cgroup = strappend("_SYSTEMD_CGROUP=", t);
+                        free(t);
+
                         if (cgroup)
                                 IOVEC_SET_STRING(iovec[n++], cgroup);
-                        free(path);
                 }
+
+                if (sd_pid_get_session(ucred->pid, &t) >= 0) {
+                        session = strappend("_SYSTEMD_SESSION=", t);
+                        free(t);
+
+                        if (session)
+                                IOVEC_SET_STRING(iovec[n++], session);
+                }
+
+                if (sd_pid_get_service(ucred->pid, &t) >= 0) {
+                        service = strappend("_SYSTEMD_SERVICE=", t);
+                        free(t);
+
+                        if (service)
+                                IOVEC_SET_STRING(iovec[n++], service);
+                }
+
+                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);
         }
 
         if (tv) {
@@ -357,13 +550,15 @@ static void dispatch_message(Server *s, struct iovec *iovec, unsigned n, unsigne
         t = gethostname_malloc();
         if (t) {
                 hostname = strappend("_HOSTNAME=", t);
+                free(t);
                 if (hostname)
                         IOVEC_SET_STRING(iovec[n++], hostname);
-                free(t);
         }
 
         assert(n <= m);
 
+        server_flush_to_var(s);
+
 retry:
         f = find_journal(s, realuid == 0 ? 0 : loginuid);
         if (!f)
@@ -374,6 +569,7 @@ retry:
                 if (r == -E2BIG && !vacuumed) {
                         log_info("Allocation limit reached.");
 
+                        server_rotate(s);
                         server_vacuum(s);
                         vacuumed = true;
 
@@ -398,11 +594,80 @@ retry:
         free(audit_session);
         free(audit_loginuid);
         free(cgroup);
+        free(session);
+        free(owner_uid);
+        free(service);
+}
+
+static void dispatch_message(Server *s,
+                             struct iovec *iovec, unsigned n, unsigned m,
+                             struct ucred *ucred,
+                             struct timeval *tv,
+                             int priority) {
+        int rl;
+        char *path = NULL, *c;
+
+        assert(s);
+        assert(iovec || n == 0);
+
+        if (n == 0)
+                return;
+
+        if (!ucred)
+                goto finish;
+
+        path = shortened_cgroup_path(ucred->pid);
+        if (!path)
+                goto finish;
+
+        /* example: /user/lennart/3/foobar
+         *          /system/dbus.service/foobar
+         *
+         * So let's cut of everything past the third /, since that is
+         * wher user directories start */
+
+        c = strchr(path, '/');
+        if (c) {
+                c = strchr(c+1, '/');
+                if (c) {
+                        c = strchr(c+1, '/');
+                        if (c)
+                                *c = 0;
+                }
+        }
+
+        rl = journal_rate_limit_test(s->rate_limit, path, priority, available_space(s));
+
+        if (rl == 0) {
+                free(path);
+                return;
+        }
+
+        if (rl > 1) {
+                int j = 0;
+                char suppress_message[LINE_MAX];
+                struct iovec suppress_iovec[18];
+
+                /* Write a suppression message if we suppressed something */
+
+                snprintf(suppress_message, sizeof(suppress_message), "MESSAGE=Suppressed %u messages from %s", rl - 1, path);
+                char_array_0(suppress_message);
+
+                IOVEC_SET_STRING(suppress_iovec[j++], "PRIORITY=5");
+                IOVEC_SET_STRING(suppress_iovec[j++], suppress_message);
+
+                dispatch_message_real(s, suppress_iovec, j, ELEMENTSOF(suppress_iovec), NULL, NULL);
+        }
+
+        free(path);
+
+finish:
+        dispatch_message_real(s, iovec, n, m, ucred, tv);
 }
 
 static void process_syslog_message(Server *s, const char *buf, struct ucred *ucred, struct timeval *tv) {
         char *message = NULL, *syslog_priority = NULL, *syslog_facility = NULL;
-        struct iovec iovec[16];
+        struct iovec iovec[19];
         unsigned n = 0;
         int priority = LOG_USER | LOG_INFO;
 
@@ -422,7 +687,7 @@ 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);
+        dispatch_message(s, iovec, n, ELEMENTSOF(iovec), ucred, tv, priority & LOG_PRIMASK);
 
         free(message);
         free(syslog_facility);
@@ -469,6 +734,7 @@ static void process_native_message(Server *s, const void *buffer, size_t buffer_
         unsigned n = 0, m = 0, j;
         const char *p;
         size_t remaining;
+        int priority = LOG_INFO;
 
         assert(s);
         assert(buffer || n == 0);
@@ -489,8 +755,9 @@ static void process_native_message(Server *s, const void *buffer, size_t buffer_
 
                 if (e == p) {
                         /* Entry separator */
-                        dispatch_message(s, iovec, n, m, ucred, tv);
+                        dispatch_message(s, iovec, n, m, ucred, tv, priority);
                         n = 0;
+                        priority = LOG_INFO;
 
                         p++;
                         remaining--;
@@ -507,11 +774,11 @@ static void process_native_message(Server *s, const void *buffer, size_t buffer_
 
                 /* A property follows */
 
-                if (n+13 >= m) {
+                if (n+16 >= m) {
                         struct iovec *c;
                         unsigned u;
 
-                        u = MAX((n+13U) * 2U, 4U);
+                        u = MAX((n+16U) * 2U, 4U);
                         c = realloc(iovec, u * sizeof(struct iovec));
                         if (!c) {
                                 log_error("Out of memory");
@@ -532,6 +799,15 @@ static void process_native_message(Server *s, const void *buffer, size_t buffer_
                                 iovec[n].iov_base = (char*) p;
                                 iovec[n].iov_len = e - p;
                                 n++;
+
+                                /* We need to determine the priority
+                                 * of this entry for the rate limiting
+                                 * logic */
+                                if (e - p == 10 &&
+                                    memcmp(p, "PRIORITY=", 10) == 0 &&
+                                    p[10] >= '0' &&
+                                    p[10] <= '9')
+                                        priority = p[10] - '0';
                         }
 
                         remaining -= (e - p) + 1;
@@ -577,7 +853,7 @@ static void process_native_message(Server *s, const void *buffer, size_t buffer_
                 }
         }
 
-        dispatch_message(s, iovec, n, m, ucred, tv);
+        dispatch_message(s, iovec, n, m, ucred, tv, priority);
 
         for (j = 0; j < n; j++)
                 if (iovec[j].iov_base < buffer ||
@@ -586,7 +862,7 @@ static void process_native_message(Server *s, const void *buffer, size_t buffer_
 }
 
 static int stdout_stream_log(StdoutStream *s, const char *p, size_t l) {
-        struct iovec iovec[15];
+        struct iovec iovec[18];
         char *message = NULL, *syslog_priority = NULL;
         unsigned n = 0;
         size_t tag_len;
@@ -630,7 +906,7 @@ static int stdout_stream_log(StdoutStream *s, const char *p, size_t l) {
                 n++;
         }
 
-        dispatch_message(s->server, iovec, n, ELEMENTSOF(iovec), &s->ucred, NULL);
+        dispatch_message(s->server, iovec, n, ELEMENTSOF(iovec), &s->ucred, NULL, priority);
 
         if (s->tee_console) {
                 int console;
@@ -889,6 +1165,243 @@ fail:
         return r;
 }
 
+static int system_journal_open(Server *s) {
+        int r;
+        char *fn;
+        sd_id128_t machine;
+        char ids[33];
+
+        r = sd_id128_get_machine(&machine);
+        if (r < 0)
+                return r;
+
+        sd_id128_to_string(machine, ids);
+
+        if (!s->system_journal) {
+
+                /* 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);
+                if (!fn)
+                        return -ENOMEM;
+
+                r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, 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;
+
+                        fix_perms(s->system_journal, 0);
+                } else if (r < 0) {
+
+                        if (r != -ENOENT && r != -EROFS)
+                                log_warning("Failed to open system journal: %s", strerror(-r));
+
+                        r = 0;
+                }
+        }
+
+        if (!s->runtime_journal) {
+
+                fn = join("/run/log/journal/", ids, "/system.journal", NULL);
+                if (!fn)
+                        return -ENOMEM;
+
+                if (s->system_journal) {
+
+                        /* Try to open the runtime journal, but only
+                         * 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);
+                        free(fn);
+
+                        if (r < 0) {
+                                if (r != -ENOENT)
+                                        log_warning("Failed to open runtime journal: %s", strerror(-r));
+
+                                r = 0;
+                        }
+
+                } else {
+
+                        /* OK, we really need the runtime journal, so create
+                         * it if necessary. */
+
+                        (void) mkdir_parents(fn, 0755);
+                        r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
+                        free(fn);
+
+                        if (r < 0) {
+                                log_error("Failed to open runtime journal: %s", strerror(-r));
+                                return r;
+                        }
+                }
+
+                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;
+
+                        fix_perms(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)
+                return 0;
+
+        ts = now(CLOCK_MONOTONIC);
+        if (s->var_available_timestamp + RECHECK_VAR_AVAILABLE_USEC > ts)
+                return 0;
+
+        s->var_available_timestamp = ts;
+
+        system_journal_open(s);
+
+        if (!s->system_journal)
+                return 0;
+
+        r = sd_id128_get_machine(&machine);
+        if (r < 0) {
+                log_error("Failed to get machine id: %s", strerror(-r));
+                return r;
+        }
+
+        r = sd_journal_open(&j, SD_JOURNAL_RUNTIME_ONLY);
+        if (r < 0) {
+                log_error("Failed to read runtime journal: %s", strerror(-r));
+                return r;
+        }
+
+        SD_JOURNAL_FOREACH(j) {
+                JournalFile *f;
+
+                f = j->current_file;
+                assert(f && f->current_offset > 0);
+
+                r = journal_file_move_to_object(f, OBJECT_ENTRY, f->current_offset, &o);
+                if (r < 0) {
+                        log_error("Can't read entry: %s", strerror(-r));
+                        goto finish;
+                }
+
+                r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
+                if (r == -E2BIG) {
+                        log_info("Allocation limit reached.");
+
+                        journal_file_post_change(s->system_journal);
+                        server_rotate(s);
+                        server_vacuum(s);
+
+                        r = journal_file_copy_entry(f, s->system_journal, o, f->current_offset, NULL, NULL, NULL);
+                }
+
+                if (r < 0) {
+                        log_error("Can't write entry: %s", strerror(-r));
+                        goto finish;
+                }
+        }
+
+finish:
+        journal_file_post_change(s->system_journal);
+
+        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);
+        }
+
+        return r;
+}
+
+static void forward_syslog(Server *s, const void *buffer, size_t length, struct ucred *ucred, struct timeval *tv) {
+        struct msghdr msghdr;
+        struct iovec iovec;
+        struct cmsghdr *cmsg;
+        union {
+                struct cmsghdr cmsghdr;
+                uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
+                            CMSG_SPACE(sizeof(struct timeval))];
+        } control;
+        union sockaddr_union sa;
+
+        assert(s);
+
+        zero(msghdr);
+
+        zero(iovec);
+        iovec.iov_base = (void*) buffer;
+        iovec.iov_len = length;
+        msghdr.msg_iov = &iovec;
+        msghdr.msg_iovlen = 1;
+
+        zero(sa);
+        sa.un.sun_family = AF_UNIX;
+        strncpy(sa.un.sun_path, "/run/systemd/syslog", sizeof(sa.un.sun_path));
+        msghdr.msg_name = &sa;
+        msghdr.msg_namelen = offsetof(union sockaddr_union, un.sun_path) + strlen(sa.un.sun_path);
+
+        zero(control);
+        msghdr.msg_control = &control;
+        msghdr.msg_controllen = sizeof(control);
+
+        cmsg = CMSG_FIRSTHDR(&msghdr);
+        cmsg->cmsg_level = SOL_SOCKET;
+        cmsg->cmsg_type = SCM_CREDENTIALS;
+        cmsg->cmsg_len = CMSG_LEN(sizeof(struct ucred));
+        memcpy(CMSG_DATA(cmsg), ucred, sizeof(struct ucred));
+        msghdr.msg_controllen = cmsg->cmsg_len;
+
+        /* Forward the syslog message we received via /dev/log to
+         * /run/systemd/syslog. Unfortunately we currently can't set
+         * the SO_TIMESTAMP auxiliary data, and hence we don't. */
+
+        if (sendmsg(s->syslog_fd, &msghdr, MSG_NOSIGNAL) >= 0)
+                return;
+
+        if (errno == ESRCH) {
+                struct ucred u;
+
+                /* Hmm, presumably the sender process vanished
+                 * by now, so let's fix it as good as we
+                 * can, and retry */
+
+                u = *ucred;
+                u.pid = getpid();
+                memcpy(CMSG_DATA(cmsg), &u, sizeof(struct ucred));
+
+                if (sendmsg(s->syslog_fd, &msghdr, MSG_NOSIGNAL) >= 0)
+                        return;
+        }
+
+        log_debug("Failed to forward syslog message: %m");
+}
+
 static int process_event(Server *s, struct epoll_event *ev) {
         assert(s);
 
@@ -913,6 +1426,11 @@ static int process_event(Server *s, struct epoll_event *ev) {
                         return -errno;
                 }
 
+                if (sfsi.ssi_signo == SIGUSR1) {
+                        server_flush_to_var(s);
+                        return 0;
+                }
+
                 log_debug("Received SIG%s", signal_to_string(sfsi.ssi_signo));
                 return 0;
 
@@ -1004,6 +1522,7 @@ static int process_event(Server *s, struct epoll_event *ev) {
                                 else
                                         s->buffer[n] = 0;
 
+                                forward_syslog(s, s->buffer, n, ucred, tv);
                                 process_syslog_message(s, strstrip(s->buffer), ucred, tv);
                         } else
                                 process_native_message(s, s->buffer, n, ucred, tv);
@@ -1048,70 +1567,11 @@ static int process_event(Server *s, struct epoll_event *ev) {
         return 0;
 }
 
-static int system_journal_open(Server *s) {
-        int r;
-        char *fn;
-        sd_id128_t machine;
-        char ids[33];
-
-        r = sd_id128_get_machine(&machine);
-        if (r < 0)
-                return r;
-
-        /* First try to create the machine path, but not the prefix */
-        fn = strappend("/var/log/journal/", sd_id128_to_string(machine, 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);
-        if (!fn)
-                return -ENOMEM;
-
-        r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->system_journal);
-        free(fn);
-
-        if (r >= 0) {
-                s->system_journal->metrics = s->metrics;
-                s->system_journal->compress = s->compress;
-
-                fix_perms(s->system_journal, 0);
-                return r;
-        }
-
-        if (r < 0 && r != -ENOENT) {
-                log_error("Failed to open system journal: %s", strerror(-r));
-                return r;
-        }
-
-        /* /var didn't work, so try /run, but this time we
-         * create the prefix too */
-        fn = join("/run/log/journal/", ids, "/system.journal", NULL);
-        if (!fn)
-                return -ENOMEM;
-
-        (void) mkdir_parents(fn, 0755);
-        r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
-        free(fn);
-
-        if (r < 0) {
-                log_error("Failed to open runtime journal: %s", strerror(-r));
-                return r;
-        }
-
-        s->runtime_journal->metrics = s->metrics;
-        s->runtime_journal->compress = s->compress;
-
-        fix_perms(s->runtime_journal, 0);
-        return r;
-}
-
 static int open_syslog_socket(Server *s) {
         union sockaddr_union sa;
         int one, r;
         struct epoll_event ev;
+        struct timeval tv;
 
         assert(s);
 
@@ -1125,7 +1585,7 @@ static int open_syslog_socket(Server *s) {
 
                 zero(sa);
                 sa.un.sun_family = AF_UNIX;
-                strncpy(sa.un.sun_path, "/run/systemd/syslog", sizeof(sa.un.sun_path));
+                strncpy(sa.un.sun_path, "/dev/log", sizeof(sa.un.sun_path));
 
                 unlink(sa.un.sun_path);
 
@@ -1152,6 +1612,15 @@ static int open_syslog_socket(Server *s) {
                 return -errno;
         }
 
+        /* Since we use the same socket for forwarding this to some
+         * other syslog implementation, make sure we don't hang
+         * forever */
+        timeval_store(&tv, SYSLOG_TIMEOUT_USEC);
+        if (setsockopt(s->syslog_fd, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv)) < 0) {
+                log_error("SO_SNDTIMEO failed: %m");
+                return -errno;
+        }
+
         zero(ev);
         ev.events = EPOLLIN;
         ev.data.fd = s->syslog_fd;
@@ -1271,7 +1740,7 @@ static int open_signalfd(Server *s) {
         assert(s);
 
         assert_se(sigemptyset(&mask) == 0);
-        sigset_add_many(&mask, SIGINT, SIGTERM, -1);
+        sigset_add_many(&mask, SIGINT, SIGTERM, SIGUSR1, -1);
         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 
         s->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
@@ -1299,12 +1768,11 @@ static int server_init(Server *s) {
 
         zero(*s);
         s->syslog_fd = s->native_fd = s->stdout_fd = s->signal_fd = s->epoll_fd = -1;
-        s->metrics.max_size = DEFAULT_MAX_SIZE;
-        s->metrics.min_size = DEFAULT_MIN_SIZE;
-        s->metrics.keep_free = DEFAULT_KEEP_FREE;
-        s->max_use = DEFAULT_MAX_USE;
         s->compress = true;
 
+        memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics));
+        memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics));
+
         s->user_journals = hashmap_new(trivial_hash_func, trivial_compare_func);
         if (!s->user_journals) {
                 log_error("Out of memory.");
@@ -1378,6 +1846,10 @@ static int server_init(Server *s) {
         if (r < 0)
                 return r;
 
+        s->rate_limit = journal_rate_limit_new(DEFAULT_RATE_LIMIT_INTERVAL, DEFAULT_RATE_LIMIT_BURST);
+        if (!s->rate_limit)
+                return -ENOMEM;
+
         return 0;
 }
 
@@ -1413,6 +1885,11 @@ static void server_done(Server *s) {
 
         if (s->stdout_fd >= 0)
                 close_nointr_nofail(s->stdout_fd);
+
+        if (s->rate_limit)
+                journal_rate_limit_free(s->rate_limit);
+
+        free(s->buffer);
 }
 
 int main(int argc, char *argv[]) {
@@ -1430,7 +1907,6 @@ int main(int argc, char *argv[]) {
         }
 
         log_set_target(LOG_TARGET_CONSOLE);
-        log_set_max_level(LOG_DEBUG);
         log_parse_environment();
         log_open();
 
@@ -1446,6 +1922,9 @@ int main(int argc, char *argv[]) {
                   "READY=1\n"
                   "STATUS=Processing requests...");
 
+        server_vacuum(&server);
+        server_flush_to_var(&server);
+
         for (;;) {
                 struct epoll_event event;