chiark / gitweb /
journal: if we encounter a corrupted file, rotate and go on
[elogind.git] / src / journal / journald.c
index 1ec1542a14d7e92268503b721d2baa815e53a599..e9ac897de7fd37793f5fa654274ebf5ed65e063b 100644 (file)
 #include <sys/signalfd.h>
 #include <unistd.h>
 #include <fcntl.h>
-#include <sys/acl.h>
-#include <acl/libacl.h>
 #include <stddef.h>
 #include <sys/ioctl.h>
 #include <linux/sockios.h>
 #include <sys/statvfs.h>
+#include <sys/user.h>
 
 #include <systemd/sd-journal.h>
 #include <systemd/sd-login.h>
@@ -40,7 +39,6 @@
 #include "hashmap.h"
 #include "journal-file.h"
 #include "socket-util.h"
-#include "acl-util.h"
 #include "cgroup-util.h"
 #include "list.h"
 #include "journal-rate-limit.h"
 #include "conf-parser.h"
 #include "journald.h"
 #include "virt.h"
+#include "missing.h"
+
+#ifdef HAVE_ACL
+#include <sys/acl.h>
+#include <acl/libacl.h>
+#include "acl-util.h"
+#endif
+
+#ifdef HAVE_SELINUX
+#include <selinux/selinux.h>
+#endif
 
 #define USER_JOURNALS_MAX 1024
 #define STDOUT_STREAMS_MAX 4096
@@ -59,9 +68,9 @@
 
 #define RECHECK_VAR_AVAILABLE_USEC (30*USEC_PER_SEC)
 
-#define SYSLOG_TIMEOUT_USEC (250*USEC_PER_MSEC)
+#define N_IOVEC_META_FIELDS 17
 
-#define N_IOVEC_META_FIELDS 16
+#define ENTRY_SIZE_MAX (1024*1024*32)
 
 typedef enum StdoutStreamState {
         STDOUT_STREAM_IDENTIFIER,
@@ -80,6 +89,9 @@ struct StdoutStream {
         int fd;
 
         struct ucred ucred;
+#ifdef HAVE_SELINUX
+        security_context_t security_context;
+#endif
 
         char *identifier;
         int priority;
@@ -180,18 +192,42 @@ finish:
         return avail;
 }
 
-static void fix_perms(JournalFile *f, uid_t uid) {
+static void server_read_file_gid(Server *s) {
+        const char *adm = "adm";
+        int r;
+
+        assert(s);
+
+        if (s->file_gid_valid)
+                return;
+
+        r = get_group_creds(&adm, &s->file_gid);
+        if (r < 0)
+                log_warning("Failed to resolve 'adm' group: %s", strerror(-r));
+
+        /* if we couldn't read the gid, then it will be 0, but that's
+         * fine and we shouldn't try to resolve the group again, so
+         * let's just pretend it worked right-away. */
+        s->file_gid_valid = true;
+}
+
+static void server_fix_perms(Server *s, JournalFile *f, uid_t uid) {
+        int r;
+#ifdef HAVE_ACL
         acl_t acl;
         acl_entry_t entry;
         acl_permset_t permset;
-        int r;
+#endif
 
         assert(f);
 
-        r = fchmod_and_fchown(f->fd, 0640, 0, 0);
+        server_read_file_gid(s);
+
+        r = fchmod_and_fchown(f->fd, 0640, 0, s->file_gid);
         if (r < 0)
                 log_warning("Failed to fix access mode/rights on %s, ignoring: %s", f->path, strerror(-r));
 
+#ifdef HAVE_ACL
         if (uid <= 0)
                 return;
 
@@ -224,6 +260,7 @@ static void fix_perms(JournalFile *f, uid_t uid) {
 
 finish:
         acl_free(acl);
+#endif
 }
 
 static JournalFile* find_journal(Server *s, uid_t uid) {
@@ -264,13 +301,13 @@ static JournalFile* find_journal(Server *s, uid_t uid) {
                 journal_file_close(f);
         }
 
-        r = journal_file_open(p, O_RDWR|O_CREAT, 0640, s->system_journal, &f);
+        r = journal_file_open_reliably(p, O_RDWR|O_CREAT, 0640, s->system_journal, &f);
         free(p);
 
         if (r < 0)
                 return s->system_journal;
 
-        fix_perms(f, uid);
+        server_fix_perms(s, f, uid);
         f->metrics = s->system_metrics;
         f->compress = s->compress;
 
@@ -398,17 +435,19 @@ static char *shortened_cgroup_path(pid_t pid) {
         return path;
 }
 
-static void dispatch_message_real(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,
+                const char *label, size_t label_len) {
 
         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, *session = NULL,
-                *owner_uid = NULL, *unit = NULL;
+                *owner_uid = NULL, *unit = NULL, *selinux_context = NULL;
 
         char idbuf[33];
         sd_id128_t id;
@@ -452,7 +491,7 @@ static void dispatch_message_real(Server *s,
                         exe = strappend("_EXE=", t);
                         free(t);
 
-                        if (comm)
+                        if (exe)
                                 IOVEC_SET_STRING(iovec[n++], exe);
                 }
 
@@ -503,6 +542,28 @@ static void dispatch_message_real(Server *s,
                 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);
+
+#ifdef HAVE_SELINUX
+                if (label) {
+                        selinux_context = malloc(sizeof("_SELINUX_CONTEXT=") + label_len);
+                        if (selinux_context) {
+                                memcpy(selinux_context, "_SELINUX_CONTEXT=", sizeof("_SELINUX_CONTEXT=")-1);
+                                memcpy(selinux_context+sizeof("_SELINUX_CONTEXT=")-1, label, label_len);
+                                selinux_context[sizeof("_SELINUX_CONTEXT=")-1+label_len] = 0;
+                                IOVEC_SET_STRING(iovec[n++], selinux_context);
+                        }
+                } else {
+                        security_context_t con;
+
+                        if (getpidcon(ucred->pid, &con) >= 0) {
+                                selinux_context = strappend("_SELINUX_CONTEXT=", con);
+                                if (selinux_context)
+                                        IOVEC_SET_STRING(iovec[n++], selinux_context);
+
+                                freecon(con);
+                        }
+                }
+#endif
         }
 
         if (tv) {
@@ -543,8 +604,12 @@ retry:
         else {
                 r = journal_file_append_entry(f, NULL, iovec, n, &s->seqnum, NULL, NULL);
 
-                if (r == -E2BIG && !vacuumed) {
-                        log_info("Allocation limit reached.");
+                if ((r == -EBADMSG || r == -E2BIG) && !vacuumed) {
+
+                        if (r == -E2BIG)
+                                log_info("Allocation limit reached, rotating.");
+                        else
+                                log_warning("Journal file corrupted, rotating.");
 
                         server_rotate(s);
                         server_vacuum(s);
@@ -574,6 +639,7 @@ retry:
         free(session);
         free(owner_uid);
         free(unit);
+        free(selinux_context);
 }
 
 static void driver_message(Server *s, sd_id128_t message_id, const char *format, ...) {
@@ -606,13 +672,14 @@ 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);
+        dispatch_message_real(s, iovec, n, ELEMENTSOF(iovec), &ucred, NULL, NULL, 0);
 }
 
 static void dispatch_message(Server *s,
                              struct iovec *iovec, unsigned n, unsigned m,
                              struct ucred *ucred,
                              struct timeval *tv,
+                             const char *label, size_t label_len,
                              int priority) {
         int rl;
         char *path = NULL, *c;
@@ -660,7 +727,7 @@ static void dispatch_message(Server *s,
         free(path);
 
 finish:
-        dispatch_message_real(s, iovec, n, m, ucred, tv);
+        dispatch_message_real(s, iovec, n, m, ucred, tv, label, label_len);
 }
 
 static void forward_syslog_iovec(Server *s, const struct iovec *iovec, unsigned n_iovec, struct ucred *ucred, struct timeval *tv) {
@@ -706,6 +773,11 @@ static void forward_syslog_iovec(Server *s, const struct iovec *iovec, unsigned
         if (sendmsg(s->syslog_fd, &msghdr, MSG_NOSIGNAL) >= 0)
                 return;
 
+        /* The socket is full? I guess the syslog implementation is
+         * too slow, and we shouldn't wait for that... */
+        if (errno == EAGAIN)
+                return;
+
         if (ucred && errno == ESRCH) {
                 struct ucred u;
 
@@ -719,6 +791,9 @@ static void forward_syslog_iovec(Server *s, const struct iovec *iovec, unsigned
 
                 if (sendmsg(s->syslog_fd, &msghdr, MSG_NOSIGNAL) >= 0)
                         return;
+
+                if (errno == EAGAIN)
+                        return;
         }
 
         log_debug("Failed to forward syslog message: %m");
@@ -902,13 +977,14 @@ finish:
         free(ident_buf);
 }
 
-static void read_identifier(const char **buf, char **identifier) {
+static void read_identifier(const char **buf, char **identifier, char **pid) {
         const char *p;
         char *t;
         size_t l, e;
 
         assert(buf);
         assert(identifier);
+        assert(pid);
 
         p = *buf;
 
@@ -928,6 +1004,10 @@ static void read_identifier(const char **buf, char **identifier) {
                 for (;;) {
 
                         if (p[k] == '[') {
+                                t = strndup(p+k+1, l-k-2);
+                                if (t)
+                                        *pid = t;
+
                                 l = k;
                                 break;
                         }
@@ -947,12 +1027,12 @@ static void read_identifier(const char **buf, char **identifier) {
         *buf += strspn(*buf, WHITESPACE);
 }
 
-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, *syslog_identifier = NULL;
-        struct iovec iovec[N_IOVEC_META_FIELDS + 5];
+static void process_syslog_message(Server *s, const char *buf, struct ucred *ucred, struct timeval *tv, const char *label, size_t label_len) {
+        char *message = NULL, *syslog_priority = NULL, *syslog_facility = NULL, *syslog_identifier = NULL, *syslog_pid = NULL;
+        struct iovec iovec[N_IOVEC_META_FIELDS + 6];
         unsigned n = 0;
         int priority = LOG_USER | LOG_INFO;
-        char *identifier = NULL;
+        char *identifier = NULL, *pid = NULL;
 
         assert(s);
         assert(buf);
@@ -962,7 +1042,7 @@ static void process_syslog_message(Server *s, const char *buf, struct ucred *ucr
 
         parse_syslog_priority((char**) &buf, &priority);
         skip_syslog_date((char**) &buf);
-        read_identifier(&buf, &identifier);
+        read_identifier(&buf, &identifier, &pid);
 
         if (s->forward_to_kmsg)
                 forward_kmsg(s, priority, identifier, buf, ucred);
@@ -985,14 +1065,21 @@ static void process_syslog_message(Server *s, const char *buf, struct ucred *ucr
                         IOVEC_SET_STRING(iovec[n++], syslog_identifier);
         }
 
+        if (pid) {
+                syslog_pid = strappend("SYSLOG_PID=", pid);
+                if (syslog_pid)
+                        IOVEC_SET_STRING(iovec[n++], syslog_pid);
+        }
+
         message = strappend("MESSAGE=", buf);
         if (message)
                 IOVEC_SET_STRING(iovec[n++], message);
 
-        dispatch_message(s, iovec, n, ELEMENTSOF(iovec), ucred, tv, priority);
+        dispatch_message(s, iovec, n, ELEMENTSOF(iovec), ucred, tv, label, label_len, priority);
 
         free(message);
         free(identifier);
+        free(pid);
         free(syslog_priority);
         free(syslog_facility);
         free(syslog_identifier);
@@ -1033,7 +1120,13 @@ static bool valid_user_field(const char *p, size_t l) {
         return true;
 }
 
-static void process_native_message(Server *s, const void *buffer, size_t buffer_size, struct ucred *ucred, struct timeval *tv) {
+static void process_native_message(
+                Server *s,
+                const void *buffer, size_t buffer_size,
+                struct ucred *ucred,
+                struct timeval *tv,
+                const char *label, size_t label_len) {
+
         struct iovec *iovec = NULL;
         unsigned n = 0, m = 0, j, tn = (unsigned) -1;
         const char *p;
@@ -1060,7 +1153,7 @@ 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, priority);
+                        dispatch_message(s, iovec, n, m, ucred, tv, label, label_len, priority);
                         n = 0;
                         priority = LOG_INFO;
 
@@ -1209,7 +1302,7 @@ static void process_native_message(Server *s, const void *buffer, size_t buffer_
                         forward_console(s, identifier, message, ucred);
         }
 
-        dispatch_message(s, iovec, n, m, ucred, tv, priority);
+        dispatch_message(s, iovec, n, m, ucred, tv, label, label_len, priority);
 
 finish:
         for (j = 0; j < n; j++)  {
@@ -1221,19 +1314,77 @@ finish:
                         free(iovec[j].iov_base);
         }
 
+        free(iovec);
         free(identifier);
         free(message);
 }
 
+static void process_native_file(
+                Server *s,
+                int fd,
+                struct ucred *ucred,
+                struct timeval *tv,
+                const char *label, size_t label_len) {
+
+        struct stat st;
+        void *p;
+        ssize_t n;
+
+        assert(s);
+        assert(fd >= 0);
+
+        /* Data is in the passed file, since it didn't fit in a
+         * datagram. We can't map the file here, since clients might
+         * then truncate it and trigger a SIGBUS for us. So let's
+         * stupidly read it */
+
+        if (fstat(fd, &st) < 0) {
+                log_error("Failed to stat passed file, ignoring: %m");
+                return;
+        }
+
+        if (!S_ISREG(st.st_mode)) {
+                log_error("File passed is not regular. Ignoring.");
+                return;
+        }
+
+        if (st.st_size <= 0)
+                return;
+
+        if (st.st_size > ENTRY_SIZE_MAX) {
+                log_error("File passed too large. Ignoring.");
+                return;
+        }
+
+        p = malloc(st.st_size);
+        if (!p) {
+                log_error("Out of memory");
+                return;
+        }
+
+        n = pread(fd, p, st.st_size, 0);
+        if (n < 0)
+                log_error("Failed to read file, ignoring: %s", strerror(-n));
+        else if (n > 0)
+                process_native_message(s, p, n, ucred, tv, label, label_len);
+
+        free(p);
+}
+
 static int stdout_stream_log(StdoutStream *s, const char *p) {
         struct iovec iovec[N_IOVEC_META_FIELDS + 5];
         char *message = NULL, *syslog_priority = NULL, *syslog_facility = NULL, *syslog_identifier = NULL;
         unsigned n = 0;
         int priority;
+        char *label = NULL;
+        size_t label_len = 0;
 
         assert(s);
         assert(p);
 
+        if (isempty(p))
+                return 0;
+
         priority = s->priority;
 
         if (s->level_prefix)
@@ -1267,7 +1418,14 @@ static int stdout_stream_log(StdoutStream *s, const char *p) {
         if (message)
                 IOVEC_SET_STRING(iovec[n++], message);
 
-        dispatch_message(s->server, iovec, n, ELEMENTSOF(iovec), &s->ucred, NULL, priority);
+#ifdef HAVE_SELINUX
+        if (s->security_context) {
+                label = (char*) s->security_context;
+                label_len = strlen((char*) s->security_context);
+        }
+#endif
+
+        dispatch_message(s->server, iovec, n, ELEMENTSOF(iovec), &s->ucred, NULL, label, label_len, priority);
 
         free(message);
         free(syslog_priority);
@@ -1288,10 +1446,14 @@ static int stdout_stream_line(StdoutStream *s, char *p) {
         switch (s->state) {
 
         case STDOUT_STREAM_IDENTIFIER:
-                s->identifier = strdup(p);
-                if (!s->identifier) {
-                        log_error("Out of memory");
-                        return -ENOMEM;
+                if (isempty(p))
+                        s->identifier = NULL;
+                else  {
+                        s->identifier = strdup(p);
+                        if (!s->identifier) {
+                                log_error("Out of memory");
+                                return -ENOMEM;
+                        }
                 }
 
                 s->state = STDOUT_STREAM_PRIORITY;
@@ -1376,7 +1538,7 @@ static int stdout_stream_scan(StdoutStream *s, bool force_flush) {
                         skip = end - p + 1;
                 else if (remaining >= sizeof(s->buffer) - 1) {
                         end = p + sizeof(s->buffer) - 1;
-                        skip = sizeof(s->buffer) - 1;
+                        skip = remaining;
                 } else
                         break;
 
@@ -1457,6 +1619,11 @@ static void stdout_stream_free(StdoutStream *s) {
                 close_nointr_nofail(s->fd);
         }
 
+#ifdef HAVE_SELINUX
+        if (s->security_context)
+                freecon(s->security_context);
+#endif
+
         free(s->identifier);
         free(s);
 }
@@ -1500,6 +1667,11 @@ static int stdout_stream_new(Server *s) {
                 goto fail;
         }
 
+#ifdef HAVE_SELINUX
+        if (getpeercon(fd, &stream->security_context) < 0)
+                log_error("Failed to determine peer security context.");
+#endif
+
         if (shutdown(fd, SHUT_WR) < 0) {
                 log_error("Failed to shutdown writing side of socket: %m");
                 r = -errno;
@@ -1526,6 +1698,161 @@ fail:
         return r;
 }
 
+static int parse_kernel_timestamp(char **_p, usec_t *t) {
+        usec_t r;
+        int k, i;
+        char *p;
+
+        assert(_p);
+        assert(*_p);
+        assert(t);
+
+        p = *_p;
+
+        if (strlen(p) < 14 || p[0] != '[' || p[13] != ']' || p[6] != '.')
+                return 0;
+
+        r = 0;
+
+        for (i = 1; i <= 5; i++) {
+                r *= 10;
+
+                if (p[i] == ' ')
+                        continue;
+
+                k = undecchar(p[i]);
+                if (k < 0)
+                        return 0;
+
+                r += k;
+        }
+
+        for (i = 7; i <= 12; i++) {
+                r *= 10;
+
+                k = undecchar(p[i]);
+                if (k < 0)
+                        return 0;
+
+                r += k;
+        }
+
+        *t = r;
+        *_p += 14;
+        *_p += strspn(*_p, WHITESPACE);
+
+        return 1;
+}
+
+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;
+
+        assert(s);
+        assert(p);
+
+        if (isempty(p))
+                return;
+
+        parse_syslog_priority((char **) &p, &priority);
+
+        if (s->forward_to_kmsg && (priority & LOG_FACMASK) != LOG_KERN)
+                return;
+
+        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);
+        }
+
+        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);
+
+                IOVEC_SET_STRING(iovec[n++], "SYSLOG_IDENTIFIER=kernel");
+        } else {
+                read_identifier(&p, &identifier, &pid);
+
+                if (s->forward_to_syslog)
+                        forward_syslog(s, priority, identifier, p, NULL, NULL);
+
+                if (identifier) {
+                        syslog_identifier = strappend("SYSLOG_IDENTIFIER=", identifier);
+                        if (syslog_identifier)
+                                IOVEC_SET_STRING(iovec[n++], syslog_identifier);
+                }
+
+                if (pid) {
+                        syslog_pid = strappend("SYSLOG_PID=", pid);
+                        if (syslog_pid)
+                                IOVEC_SET_STRING(iovec[n++], syslog_pid);
+                }
+
+                if (asprintf(&syslog_facility, "SYSLOG_FACILITY=%i", LOG_FAC(priority)) >= 0)
+                        IOVEC_SET_STRING(iovec[n++], syslog_facility);
+        }
+
+        message = strappend("MESSAGE=", p);
+        if (message)
+                IOVEC_SET_STRING(iovec[n++], message);
+
+        dispatch_message(s, iovec, n, ELEMENTSOF(iovec), NULL, NULL, NULL, 0, priority);
+
+        free(message);
+        free(syslog_priority);
+        free(syslog_identifier);
+        free(syslog_pid);
+        free(syslog_facility);
+        free(source_time);
+        free(identifier);
+        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;
@@ -1552,7 +1879,7 @@ static int system_journal_open(Server *s) {
                 if (!fn)
                         return -ENOMEM;
 
-                r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->system_journal);
+                r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, NULL, &s->system_journal);
                 free(fn);
 
                 if (r >= 0) {
@@ -1561,7 +1888,7 @@ static int system_journal_open(Server *s) {
                         s->system_journal->metrics = s->system_metrics;
                         s->system_journal->compress = s->compress;
 
-                        fix_perms(s->system_journal, 0);
+                        server_fix_perms(s, s->system_journal, 0);
                 } else if (r < 0) {
 
                         if (r != -ENOENT && r != -EROFS)
@@ -1583,7 +1910,7 @@ 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_reliably(fn, O_RDWR, 0640, NULL, &s->runtime_journal);
                         free(fn);
 
                         if (r < 0) {
@@ -1599,7 +1926,7 @@ static int system_journal_open(Server *s) {
                          * it if necessary. */
 
                         (void) mkdir_parents(fn, 0755);
-                        r = journal_file_open(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
+                        r = journal_file_open_reliably(fn, O_RDWR|O_CREAT, 0640, NULL, &s->runtime_journal);
                         free(fn);
 
                         if (r < 0) {
@@ -1614,7 +1941,7 @@ static int system_journal_open(Server *s) {
                         s->runtime_journal->metrics = s->runtime_metrics;
                         s->runtime_journal->compress = s->compress;
 
-                        fix_perms(s->runtime_journal, 0);
+                        server_fix_perms(s, s->runtime_journal, 0);
                 }
         }
 
@@ -1645,6 +1972,8 @@ static int server_flush_to_var(Server *s) {
         if (!s->system_journal)
                 return 0;
 
+        log_info("Flushing to /var...");
+
         r = sd_id128_get_machine(&machine);
         if (r < 0) {
                 log_error("Failed to get machine id: %s", strerror(-r));
@@ -1700,6 +2029,49 @@ finish:
         return r;
 }
 
+static int server_read_proc_kmsg(Server *s) {
+        ssize_t l;
+        assert(s);
+        assert(s->proc_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);
+        if (l < 0) {
+
+                if (errno == EAGAIN || errno == EINTR)
+                        return 0;
+
+                log_error("Failed to read from kernel: %m");
+                return -errno;
+        }
+
+        s->proc_kmsg_length += l;
+
+        proc_kmsg_scan(s);
+        return 1;
+}
+
+static int server_flush_proc_kmsg(Server *s) {
+        int r;
+
+        assert(s);
+
+        if (s->proc_kmsg_fd < 0)
+                return 0;
+
+        log_info("Flushing /proc/kmsg...");
+
+        for (;;) {
+                r = server_read_proc_kmsg(s);
+                if (r < 0)
+                        return r;
+
+                if (r == 0)
+                        break;
+        }
+
+        return 0;
+}
+
 static int process_event(Server *s, struct epoll_event *ev) {
         assert(s);
 
@@ -1719,7 +2091,7 @@ static int process_event(Server *s, struct epoll_event *ev) {
                                 return -EIO;
 
                         if (errno == EINTR || errno == EAGAIN)
-                                return 0;
+                                return 1;
 
                         return -errno;
                 }
@@ -1732,6 +2104,20 @@ static int process_event(Server *s, struct epoll_event *ev) {
                 log_debug("Received SIG%s", signal_to_string(sfsi.ssi_signo));
                 return 0;
 
+        } else if (ev->data.fd == s->proc_kmsg_fd) {
+                int r;
+
+                if (ev->events != EPOLLIN) {
+                        log_info("Got invalid event from epoll.");
+                        return -EIO;
+                }
+
+                r = server_read_proc_kmsg(s);
+                if (r < 0)
+                        return r;
+
+                return 1;
+
         } else if (ev->data.fd == s->native_fd ||
                    ev->data.fd == s->syslog_fd) {
 
@@ -1746,22 +2132,25 @@ static int process_event(Server *s, struct epoll_event *ev) {
                         struct ucred *ucred = NULL;
                         struct timeval *tv = NULL;
                         struct cmsghdr *cmsg;
+                        char *label = NULL;
+                        size_t label_len = 0;
                         union {
                                 struct cmsghdr cmsghdr;
                                 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
-                                            CMSG_SPACE(sizeof(struct timeval))];
+                                            CMSG_SPACE(sizeof(struct timeval)) +
+                                            CMSG_SPACE(sizeof(int)) +
+                                            CMSG_SPACE(PAGE_SIZE)]; /* selinux label */
                         } control;
                         ssize_t n;
                         int v;
+                        int *fds = NULL;
+                        unsigned n_fds = 0;
 
                         if (ioctl(ev->data.fd, SIOCINQ, &v) < 0) {
                                 log_error("SIOCINQ failed: %m");
                                 return -errno;
                         }
 
-                        if (v <= 0)
-                                return 1;
-
                         if (s->buffer_size < (size_t) v) {
                                 void *b;
                                 size_t l;
@@ -1789,7 +2178,7 @@ static int process_event(Server *s, struct epoll_event *ev) {
                         msghdr.msg_control = &control;
                         msghdr.msg_controllen = sizeof(control);
 
-                        n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT);
+                        n = recvmsg(ev->data.fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
                         if (n < 0) {
 
                                 if (errno == EINTR || errno == EAGAIN)
@@ -1806,23 +2195,44 @@ static int process_event(Server *s, struct epoll_event *ev) {
                                     cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred)))
                                         ucred = (struct ucred*) CMSG_DATA(cmsg);
                                 else if (cmsg->cmsg_level == SOL_SOCKET &&
+                                         cmsg->cmsg_type == SCM_SECURITY) {
+                                        label = (char*) CMSG_DATA(cmsg);
+                                        label_len = cmsg->cmsg_len - CMSG_LEN(0);
+                                } else if (cmsg->cmsg_level == SOL_SOCKET &&
                                          cmsg->cmsg_type == SO_TIMESTAMP &&
                                          cmsg->cmsg_len == CMSG_LEN(sizeof(struct timeval)))
                                         tv = (struct timeval*) CMSG_DATA(cmsg);
+                                else if (cmsg->cmsg_level == SOL_SOCKET &&
+                                         cmsg->cmsg_type == SCM_RIGHTS) {
+                                        fds = (int*) CMSG_DATA(cmsg);
+                                        n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
+                                }
                         }
 
                         if (ev->data.fd == s->syslog_fd) {
                                 char *e;
 
-                                e = memchr(s->buffer, '\n', n);
-                                if (e)
-                                        *e = 0;
-                                else
-                                        s->buffer[n] = 0;
+                                if (n > 0 && n_fds == 0) {
+                                        e = memchr(s->buffer, '\n', n);
+                                        if (e)
+                                                *e = 0;
+                                        else
+                                                s->buffer[n] = 0;
+
+                                        process_syslog_message(s, strstrip(s->buffer), ucred, tv, label, label_len);
+                                } else if (n_fds > 0)
+                                        log_warning("Got file descriptors via syslog socket. Ignoring.");
+
+                        } else {
+                                if (n > 0 && n_fds == 0)
+                                        process_native_message(s, s->buffer, n, ucred, tv, label, label_len);
+                                else if (n == 0 && n_fds == 1)
+                                        process_native_file(s, fds[0], ucred, tv, label, label_len);
+                                else if (n_fds > 0)
+                                        log_warning("Got too many file descriptors via native socket. Ignoring.");
+                        }
 
-                                process_syslog_message(s, strstrip(s->buffer), ucred, tv);
-                        } else
-                                process_native_message(s, s->buffer, n, ucred, tv);
+                        close_many(fds, n_fds);
                 }
 
                 return 1;
@@ -1868,13 +2278,12 @@ static int open_syslog_socket(Server *s) {
         union sockaddr_union sa;
         int one, r;
         struct epoll_event ev;
-        struct timeval tv;
 
         assert(s);
 
         if (s->syslog_fd < 0) {
 
-                s->syslog_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0);
+                s->syslog_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
                 if (s->syslog_fd < 0) {
                         log_error("socket() failed: %m");
                         return -errno;
@@ -1893,7 +2302,8 @@ static int open_syslog_socket(Server *s) {
                 }
 
                 chmod(sa.un.sun_path, 0666);
-        }
+        } else
+                fd_nonblock(s->syslog_fd, 1);
 
         one = 1;
         r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
@@ -1902,6 +2312,13 @@ static int open_syslog_socket(Server *s) {
                 return -errno;
         }
 
+#ifdef HAVE_SELINUX
+        one = 1;
+        r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_PASSSEC, &one, sizeof(one));
+        if (r < 0)
+                log_warning("SO_PASSSEC failed: %m");
+#endif
+
         one = 1;
         r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_TIMESTAMP, &one, sizeof(one));
         if (r < 0) {
@@ -1909,15 +2326,6 @@ 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;
@@ -1938,7 +2346,7 @@ static int open_native_socket(Server*s) {
 
         if (s->native_fd < 0) {
 
-                s->native_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0);
+                s->native_fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
                 if (s->native_fd < 0) {
                         log_error("socket() failed: %m");
                         return -errno;
@@ -1957,7 +2365,8 @@ static int open_native_socket(Server*s) {
                 }
 
                 chmod(sa.un.sun_path, 0666);
-        }
+        } else
+                fd_nonblock(s->native_fd, 1);
 
         one = 1;
         r = setsockopt(s->native_fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
@@ -1966,6 +2375,13 @@ static int open_native_socket(Server*s) {
                 return -errno;
         }
 
+#ifdef HAVE_SELINUX
+        one = 1;
+        r = setsockopt(s->syslog_fd, SOL_SOCKET, SO_PASSSEC, &one, sizeof(one));
+        if (r < 0)
+                log_warning("SO_PASSSEC failed: %m");
+#endif
+
         one = 1;
         r = setsockopt(s->native_fd, SOL_SOCKET, SO_TIMESTAMP, &one, sizeof(one));
         if (r < 0) {
@@ -1993,7 +2409,7 @@ static int open_stdout_socket(Server *s) {
 
         if (s->stdout_fd < 0) {
 
-                s->stdout_fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC, 0);
+                s->stdout_fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
                 if (s->stdout_fd < 0) {
                         log_error("socket() failed: %m");
                         return -errno;
@@ -2017,7 +2433,8 @@ static int open_stdout_socket(Server *s) {
                         log_error("liste() failed: %m");
                         return -errno;
                 }
-        }
+        } else
+                fd_nonblock(s->stdout_fd, 1);
 
         zero(ev);
         ev.events = EPOLLIN;
@@ -2030,6 +2447,32 @@ static int open_stdout_socket(Server *s) {
         return 0;
 }
 
+static int open_proc_kmsg(Server *s) {
+        struct epoll_event ev;
+
+        assert(s);
+
+        if (!s->import_proc_kmsg)
+                return 0;
+
+
+        s->proc_kmsg_fd = open("/proc/kmsg", O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
+        if (s->proc_kmsg_fd < 0) {
+                log_warning("Failed to open /proc/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");
+                return -errno;
+        }
+
+        return 0;
+}
+
 static int open_signalfd(Server *s) {
         sigset_t mask;
         struct epoll_event ev;
@@ -2143,13 +2586,14 @@ 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 = -1;
+        s->syslog_fd = s->native_fd = s->stdout_fd = s->signal_fd = s->epoll_fd = s->proc_kmsg_fd = -1;
         s->compress = 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;
 
         memset(&s->system_metrics, 0xFF, sizeof(s->system_metrics));
         memset(&s->runtime_metrics, 0xFF, sizeof(s->runtime_metrics));
@@ -2222,7 +2666,7 @@ static int server_init(Server *s) {
         if (r < 0)
                 return r;
 
-        r = system_journal_open(s);
+        r = open_proc_kmsg(s);
         if (r < 0)
                 return r;
 
@@ -2234,6 +2678,10 @@ static int server_init(Server *s) {
         if (!s->rate_limit)
                 return -ENOMEM;
 
+        r = system_journal_open(s);
+        if (r < 0)
+                return r;
+
         return 0;
 }
 
@@ -2270,6 +2718,9 @@ 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->rate_limit)
                 journal_rate_limit_free(s->rate_limit);
 
@@ -2302,6 +2753,7 @@ int main(int argc, char *argv[]) {
 
         server_vacuum(&server);
         server_flush_to_var(&server);
+        server_flush_proc_kmsg(&server);
 
         log_debug("systemd-journald running as pid %lu", (unsigned long) getpid());
         driver_message(&server, SD_MESSAGE_JOURNAL_START, "Journal started");