journal: if the data to be sent is larger than the maximum datagram size resort to passing a temporary fd over native protocol

src/journal/journal-send.c

@@ -23,6 +23,8 @@
 #include <sys/un.h>
 #include <errno.h>
 #include <stddef.h>
+#include <unistd.h>
+#include <fcntl.h>
 
 #include "sd-journal.h"
 #include "util.h"
@@ -132,12 +134,19 @@ fail:
 }
 
 _public_ int sd_journal_sendv(const struct iovec *iov, int n) {
-        int fd;
+        int fd, buffer_fd;
         struct iovec *w;
         uint64_t *l;
         int i, j = 0;
         struct msghdr mh;
         struct sockaddr_un sa;
+        char path[] = "/tmp/journal.XXXXXX";
+        ssize_t k;
+        union {
+                struct cmsghdr cmsghdr;
+                uint8_t buf[CMSG_SPACE(sizeof(int))];
+        } control;
+        struct cmsghdr *cmsg;
 
         if (!iov || n <= 0)
                 return -EINVAL;
@@ -203,7 +212,51 @@ _public_ int sd_journal_sendv(const struct iovec *iov, int n) {
         mh.msg_iov = w;
         mh.msg_iovlen = j;
 
-        if (sendmsg(fd, &mh, MSG_NOSIGNAL) < 0)
+        k = sendmsg(fd, &mh, MSG_NOSIGNAL);
+        if (k >= 0)
+                return 0;
+
+        if (errno != EMSGSIZE)
+                return -errno;
+
+        /* Message doesn't fit... Let's dump the data in a temporary
+         * file and just pass a file descriptor of it to the other
+         * side */
+
+        buffer_fd = mkostemp(path, O_CLOEXEC|O_RDWR);
+        if (buffer_fd < 0)
+                return -errno;
+
+        if (unlink(path) < 0) {
+                close_nointr_nofail(buffer_fd);
+                return -errno;
+        }
+
+        n = writev(buffer_fd, w, j);
+        if (n < 0)  {
+                close_nointr_nofail(buffer_fd);
+                return -errno;
+        }
+
+        mh.msg_iov = NULL;
+        mh.msg_iovlen = 0;
+
+        zero(control);
+        mh.msg_control = &control;
+        mh.msg_controllen = sizeof(control);
+
+        cmsg = CMSG_FIRSTHDR(&mh);
+        cmsg->cmsg_level = SOL_SOCKET;
+        cmsg->cmsg_type = SCM_RIGHTS;
+        cmsg->cmsg_len = CMSG_LEN(sizeof(int));
+        memcpy(CMSG_DATA(cmsg), &buffer_fd, sizeof(int));
+
+        mh.msg_controllen = cmsg->cmsg_len;
+
+        k = sendmsg(fd, &mh, MSG_NOSIGNAL);
+        close_nointr_nofail(buffer_fd);
+
+        if (k < 0)
                 return -errno;
 
         return 0;

src/journal/journald.c

@@ -70,6 +70,8 @@
 
 #define N_IOVEC_META_FIELDS 17
 
+#define ENTRY_SIZE_MAX (1024*1024*32)
+
 typedef enum StdoutStreamState {
         STDOUT_STREAM_IDENTIFIER,
         STDOUT_STREAM_PRIORITY,
@@ -1291,6 +1293,52 @@ finish:
         free(message);
 }
 
+static void process_native_file(Server *s, int fd, struct ucred *ucred, struct timeval *tv) {
+        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);
+
+        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;
@@ -1300,6 +1348,9 @@ static int stdout_stream_log(StdoutStream *s, const char *p) {
         assert(s);
         assert(p);
 
+        if (isempty(p))
+                return 0;
+
         priority = s->priority;
 
         if (s->level_prefix)
@@ -1649,6 +1700,9 @@ static void proc_kmsg_line(Server *s, const char *p) {
         assert(s);
         assert(p);
 
+        if (isempty(p))
+                return;
+
         parse_syslog_priority((char **) &p, &priority);
 
         if (s->forward_to_kmsg && (priority & LOG_FACMASK) != LOG_KERN)
@@ -1697,7 +1751,6 @@ static void proc_kmsg_line(Server *s, const char *p) {
         if (message)
                 IOVEC_SET_STRING(iovec[n++], message);
 
-
         dispatch_message(s, iovec, n, ELEMENTSOF(iovec), NULL, NULL, priority);
 
         free(message);
@@ -2027,19 +2080,19 @@ static int process_event(Server *s, struct epoll_event *ev) {
                         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))];
                         } 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;
@@ -2067,7 +2120,7 @@ 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)
@@ -2087,11 +2140,17 @@ static int process_event(Server *s, struct epoll_event *ev) {
                                          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;
 
+                                if (n > 0 && n_fds == 0) {
                                         e = memchr(s->buffer, '\n', n);
                                         if (e)
                                                 *e = 0;
@@ -2099,8 +2158,19 @@ static int process_event(Server *s, struct epoll_event *ev) {
                                                 s->buffer[n] = 0;
 
                                         process_syslog_message(s, strstrip(s->buffer), ucred, tv);
-                        } else
+                                } 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);
+                                else if (n == 0 && n_fds == 1)
+                                        process_native_file(s, fds[0], ucred, tv);
+                                else if (n_fds > 0)
+                                        log_warning("Got too many file descriptors via native socket. Ignoring.");
+                        }
+
+                        close_many(fds, n_fds);
                 }
 
                 return 1;