#define O_DIRECTORY 0
#endif
+/* Maximum length of the sun_path member in a struct sockaddr_un, excluding
+ * space for a null terminator. */
+#define MAX_UN_LEN (sizeof(((struct sockaddr_un *) 0)->sun_path) - 1)
+
static int getsockopt_int(int fd, int level, int option, const char *optname,
int *valuep);
}
}
-/* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
- * '*un_len' the size of the sockaddr_un. */
-static void
-make_sockaddr_un__(const char *name, struct sockaddr_un *un, socklen_t *un_len)
+/* Attempts to shorten 'name' by opening a file descriptor for the directory
+ * part of the name and indirecting through /proc/self/fd/<dirfd>/<basename>.
+ * On systems with Linux-like /proc, this works as long as <basename> isn't too
+ * long.
+ *
+ * On success, returns 0 and stores the short name in 'short_name' and a
+ * directory file descriptor to eventually be closed in '*dirfpd'. */
+static int
+shorten_name_via_proc(const char *name, char short_name[MAX_UN_LEN + 1],
+ int *dirfdp)
{
- un->sun_family = AF_UNIX;
- ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
- *un_len = (offsetof(struct sockaddr_un, sun_path)
- + strlen (un->sun_path) + 1);
+ char *dir, *base;
+ int dirfd;
+ int len;
+
+ if (!LINUX_DATAPATH) {
+ return ENAMETOOLONG;
+ }
+
+ dir = dir_name(name);
+ dirfd = open(dir, O_DIRECTORY | O_RDONLY);
+ if (dirfd < 0) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
+ int error = errno;
+
+ VLOG_WARN_RL(&rl, "%s: open failed (%s)", dir, ovs_strerror(error));
+ free(dir);
+
+ return error;
+ }
+ free(dir);
+
+ base = base_name(name);
+ len = snprintf(short_name, MAX_UN_LEN + 1,
+ "/proc/self/fd/%d/%s", dirfd, base);
+ free(base);
+
+ if (len >= 0 && len <= MAX_UN_LEN) {
+ *dirfdp = dirfd;
+ return 0;
+ } else {
+ close(dirfd);
+ return ENAMETOOLONG;
+ }
+}
+
+/* Attempts to shorten 'name' by creating a symlink for the directory part of
+ * the name and indirecting through <symlink>/<basename>. This works on
+ * systems that support symlinks, as long as <basename> isn't too long.
+ *
+ * On success, returns 0 and stores the short name in 'short_name' and the
+ * symbolic link to eventually delete in 'linkname'. */
+static int
+shorten_name_via_symlink(const char *name, char short_name[MAX_UN_LEN + 1],
+ char linkname[MAX_UN_LEN + 1])
+{
+ char *abs, *dir, *base;
+ const char *tmpdir;
+ int error;
+ int i;
+
+ abs = abs_file_name(NULL, name);
+ dir = dir_name(abs);
+ base = base_name(abs);
+ free(abs);
+
+ tmpdir = getenv("TMPDIR");
+ if (tmpdir == NULL) {
+ tmpdir = "/tmp";
+ }
+
+ for (i = 0; i < 1000; i++) {
+ int len;
+
+ len = snprintf(linkname, MAX_UN_LEN + 1,
+ "%s/ovs-un-c-%"PRIu32, tmpdir, random_uint32());
+ error = (len < 0 || len > MAX_UN_LEN ? ENAMETOOLONG
+ : symlink(dir, linkname) ? errno
+ : 0);
+ if (error != EEXIST) {
+ break;
+ }
+ }
+
+ if (!error) {
+ int len;
+
+ fatal_signal_add_file_to_unlink(linkname);
+
+ len = snprintf(short_name, MAX_UN_LEN + 1, "%s/%s", linkname, base);
+ if (len < 0 || len > MAX_UN_LEN) {
+ fatal_signal_unlink_file_now(linkname);
+ error = ENAMETOOLONG;
+ }
+ }
+
+ if (error) {
+ linkname[0] = '\0';
+ }
+ free(dir);
+ free(base);
+
+ return error;
}
/* Stores in '*un' a sockaddr_un that refers to file 'name'. Stores in
* '*un_len' the size of the sockaddr_un.
*
- * Returns 0 on success, otherwise a positive errno value. On success,
- * '*dirfdp' is either -1 or a nonnegative file descriptor that the caller
- * should close after using '*un' to bind or connect. On failure, '*dirfdp' is
- * -1. */
+ * Returns 0 on success, otherwise a positive errno value.
+ *
+ * Uses '*dirfdp' and 'linkname' to store references to data when the caller no
+ * longer needs to use 'un'. On success, freeing these references with
+ * free_sockaddr_un() is mandatory to avoid a leak; on failure, freeing them is
+ * unnecessary but harmless. */
static int
make_sockaddr_un(const char *name, struct sockaddr_un *un, socklen_t *un_len,
- int *dirfdp)
+ int *dirfdp, char linkname[MAX_UN_LEN + 1])
{
- enum { MAX_UN_LEN = sizeof un->sun_path - 1 };
+ char short_name[MAX_UN_LEN + 1];
*dirfdp = -1;
+ linkname[0] = '\0';
if (strlen(name) > MAX_UN_LEN) {
- static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
-
- if (LINUX_DATAPATH) {
- /* 'name' is too long to fit in a sockaddr_un, but we have a
- * workaround for that on Linux: shorten it by opening a file
- * descriptor for the directory part of the name and indirecting
- * through /proc/self/fd/<dirfd>/<basename>. */
- char *dir, *base;
- char *short_name;
- int dirfd;
-
- dir = dir_name(name);
- base = base_name(name);
-
- dirfd = open(dir, O_DIRECTORY | O_RDONLY);
- if (dirfd < 0) {
- free(base);
- free(dir);
- return errno;
- }
-
- short_name = xasprintf("/proc/self/fd/%d/%s", dirfd, base);
- free(dir);
- free(base);
-
- if (strlen(short_name) <= MAX_UN_LEN) {
- make_sockaddr_un__(short_name, un, un_len);
- free(short_name);
- *dirfdp = dirfd;
- return 0;
- }
- free(short_name);
- close(dirfd);
+ /* 'name' is too long to fit in a sockaddr_un. Try a workaround. */
+ int error = shorten_name_via_proc(name, short_name, dirfdp);
+ if (error == ENAMETOOLONG) {
+ error = shorten_name_via_symlink(name, short_name, linkname);
+ }
+ if (error) {
+ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1);
VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
- "%d bytes (even shortened)", name, MAX_UN_LEN);
- } else {
- /* 'name' is too long and we have no workaround. */
- VLOG_WARN_RL(&rl, "Unix socket name %s is longer than maximum "
- "%d bytes", name, MAX_UN_LEN);
+ "%"PRIuSIZE" bytes", name, MAX_UN_LEN);
+ return error;
}
- return ENAMETOOLONG;
- } else {
- make_sockaddr_un__(name, un, un_len);
- return 0;
+ name = short_name;
+ }
+
+ un->sun_family = AF_UNIX;
+ ovs_strzcpy(un->sun_path, name, sizeof un->sun_path);
+ *un_len = (offsetof(struct sockaddr_un, sun_path)
+ + strlen (un->sun_path) + 1);
+ return 0;
+}
+
+/* Clean up after make_sockaddr_un(). */
+static void
+free_sockaddr_un(int dirfd, const char *linkname)
+{
+ if (dirfd >= 0) {
+ close(dirfd);
+ }
+ if (linkname[0]) {
+ fatal_signal_unlink_file_now(linkname);
}
}
}
if (bind_path) {
+ char linkname[MAX_UN_LEN + 1];
struct sockaddr_un un;
socklen_t un_len;
int dirfd;
}
fatal_signal_add_file_to_unlink(bind_path);
- error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd);
+ error = make_sockaddr_un(bind_path, &un, &un_len, &dirfd, linkname);
if (!error) {
error = bind_unix_socket(fd, (struct sockaddr *) &un, un_len);
}
- if (dirfd >= 0) {
- close(dirfd);
- }
+ free_sockaddr_un(dirfd, linkname);
+
if (error) {
goto error;
}
}
if (connect_path) {
+ char linkname[MAX_UN_LEN + 1];
struct sockaddr_un un;
socklen_t un_len;
int dirfd;
- error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd);
+ error = make_sockaddr_un(connect_path, &un, &un_len, &dirfd, linkname);
if (!error
&& connect(fd, (struct sockaddr*) &un, un_len)
&& errno != EINPROGRESS) {
error = errno;
}
- if (dirfd >= 0) {
- close(dirfd);
- }
+ free_sockaddr_un(dirfd, linkname);
+
if (error) {
goto error;
}
xset_nonblocking(fds[1]);
}
-void
-xsocketpair(int domain, int type, int protocol, int fds[2])
-{
- if (socketpair(domain, type, protocol, fds)) {
- VLOG_FATAL("failed to create socketpair (%s)", ovs_strerror(errno));
- }
-}
-
static int
getsockopt_int(int fd, int level, int option, const char *optname, int *valuep)
{
VLOG_ERR_RL(&rl, "getsockopt(%s): %s", optname, ovs_strerror(error));
} else if (len != sizeof value) {
error = EINVAL;
- VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %zu)",
+ VLOG_ERR_RL(&rl, "getsockopt(%s): value is %u bytes (expected %"PRIuSIZE")",
optname, (unsigned int) len, sizeof value);
} else {
error = 0;
return ds_steal_cstr(&string);
}
-/* Returns the total of the 'iov_len' members of the 'n_iovs' in 'iovs'.
- * The caller must ensure that the total does not exceed SIZE_MAX. */
-size_t
-iovec_len(const struct iovec iovs[], size_t n_iovs)
-{
- size_t len = 0;
- size_t i;
-
- for (i = 0; i < n_iovs; i++) {
- len += iovs[i].iov_len;
- }
- return len;
-}
-
-/* Returns true if all of the 'n_iovs' iovecs in 'iovs' have length zero. */
-bool
-iovec_is_empty(const struct iovec iovs[], size_t n_iovs)
-{
- size_t i;
-
- for (i = 0; i < n_iovs; i++) {
- if (iovs[i].iov_len) {
- return false;
- }
- }
- return true;
-}
-
-/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
- * in 'fds' on Unix domain socket 'sock'. Returns the number of bytes
- * successfully sent or -1 if an error occurred. On error, sets errno
- * appropriately. */
-int
-send_iovec_and_fds(int sock,
- const struct iovec *iovs, size_t n_iovs,
- const int fds[], size_t n_fds)
-{
- ovs_assert(sock >= 0);
- if (n_fds > 0) {
- union {
- struct cmsghdr cm;
- char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
- } cmsg;
- struct msghdr msg;
-
- ovs_assert(!iovec_is_empty(iovs, n_iovs));
- ovs_assert(n_fds <= SOUTIL_MAX_FDS);
-
- memset(&cmsg, 0, sizeof cmsg);
- cmsg.cm.cmsg_len = CMSG_LEN(n_fds * sizeof *fds);
- cmsg.cm.cmsg_level = SOL_SOCKET;
- cmsg.cm.cmsg_type = SCM_RIGHTS;
- memcpy(CMSG_DATA(&cmsg.cm), fds, n_fds * sizeof *fds);
-
- msg.msg_name = NULL;
- msg.msg_namelen = 0;
- msg.msg_iov = CONST_CAST(struct iovec *, iovs);
- msg.msg_iovlen = n_iovs;
- msg.msg_control = &cmsg.cm;
- msg.msg_controllen = CMSG_SPACE(n_fds * sizeof *fds);
- msg.msg_flags = 0;
-
- return sendmsg(sock, &msg, 0);
- } else {
- return writev(sock, iovs, n_iovs);
- }
-}
-
-/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
- * in 'fds' on Unix domain socket 'sock'. If 'skip_bytes' is nonzero, then the
- * first 'skip_bytes' of data in the iovecs are not sent, and none of the file
- * descriptors are sent. The function continues to retry sending until an
- * error (other than EINTR) occurs or all the data and fds are sent.
- *
- * Returns 0 if all the data and fds were successfully sent, otherwise a
- * positive errno value. Regardless of success, stores the number of bytes
- * sent (always at least 'skip_bytes') in '*bytes_sent'. (If at least one byte
- * is sent, then all the fds have been sent.)
- *
- * 'skip_bytes' must be less than or equal to iovec_len(iovs, n_iovs). */
-int
-send_iovec_and_fds_fully(int sock,
- const struct iovec iovs[], size_t n_iovs,
- const int fds[], size_t n_fds,
- size_t skip_bytes, size_t *bytes_sent)
-{
- *bytes_sent = 0;
- while (n_iovs > 0) {
- int retval;
-
- if (skip_bytes) {
- retval = skip_bytes;
- skip_bytes = 0;
- } else if (!*bytes_sent) {
- retval = send_iovec_and_fds(sock, iovs, n_iovs, fds, n_fds);
- } else {
- retval = writev(sock, iovs, n_iovs);
- }
-
- if (retval > 0) {
- *bytes_sent += retval;
- while (retval > 0) {
- const uint8_t *base = iovs->iov_base;
- size_t len = iovs->iov_len;
-
- if (retval < len) {
- size_t sent;
- int error;
-
- error = write_fully(sock, base + retval, len - retval,
- &sent);
- *bytes_sent += sent;
- retval += sent;
- if (error) {
- return error;
- }
- }
- retval -= len;
- iovs++;
- n_iovs--;
- }
- } else if (retval == 0) {
- if (iovec_is_empty(iovs, n_iovs)) {
- break;
- }
- VLOG_WARN("send returned 0");
- return EPROTO;
- } else if (errno != EINTR) {
- return errno;
- }
- }
-
- return 0;
-}
-
-/* Sends the 'n_iovs' iovecs of data in 'iovs' and the 'n_fds' file descriptors
- * in 'fds' on Unix domain socket 'sock'. The function continues to retry
- * sending until an error (other than EAGAIN or EINTR) occurs or all the data
- * and fds are sent. Upon EAGAIN, the function blocks until the socket is
- * ready for more data.
- *
- * Returns 0 if all the data and fds were successfully sent, otherwise a
- * positive errno value. */
-int
-send_iovec_and_fds_fully_block(int sock,
- const struct iovec iovs[], size_t n_iovs,
- const int fds[], size_t n_fds)
-{
- size_t sent = 0;
-
- for (;;) {
- int error;
-
- error = send_iovec_and_fds_fully(sock, iovs, n_iovs,
- fds, n_fds, sent, &sent);
- if (error != EAGAIN) {
- return error;
- }
- poll_fd_wait(sock, POLLOUT);
- poll_block();
- }
-}
-
-/* Attempts to receive from Unix domain socket 'sock' up to 'size' bytes of
- * data into 'data' and up to SOUTIL_MAX_FDS file descriptors into 'fds'.
- *
- * - Upon success, returns the number of bytes of data copied into 'data'
- * and stores the number of received file descriptors into '*n_fdsp'.
- *
- * - On failure, returns a negative errno value and stores 0 in
- * '*n_fdsp'.
- *
- * - On EOF, returns 0 and stores 0 in '*n_fdsp'. */
-int
-recv_data_and_fds(int sock,
- void *data, size_t size,
- int fds[SOUTIL_MAX_FDS], size_t *n_fdsp)
-{
- union {
- struct cmsghdr cm;
- char control[CMSG_SPACE(SOUTIL_MAX_FDS * sizeof *fds)];
- } cmsg;
- struct msghdr msg;
- int retval;
- struct cmsghdr *p;
- size_t i;
-
- *n_fdsp = 0;
-
- do {
- struct iovec iov;
-
- iov.iov_base = data;
- iov.iov_len = size;
-
- msg.msg_name = NULL;
- msg.msg_namelen = 0;
- msg.msg_iov = &iov;
- msg.msg_iovlen = 1;
- msg.msg_control = &cmsg.cm;
- msg.msg_controllen = sizeof cmsg.control;
- msg.msg_flags = 0;
-
- retval = recvmsg(sock, &msg, 0);
- } while (retval < 0 && errno == EINTR);
- if (retval <= 0) {
- return retval < 0 ? -errno : 0;
- }
-
- for (p = CMSG_FIRSTHDR(&msg); p; p = CMSG_NXTHDR(&msg, p)) {
- if (p->cmsg_level != SOL_SOCKET || p->cmsg_type != SCM_RIGHTS) {
- VLOG_ERR("unexpected control message %d:%d",
- p->cmsg_level, p->cmsg_type);
- goto error;
- } else if (*n_fdsp) {
- VLOG_ERR("multiple SCM_RIGHTS received");
- goto error;
- } else {
- size_t n_fds = (p->cmsg_len - CMSG_LEN(0)) / sizeof *fds;
- const int *fds_data = ALIGNED_CAST(const int *, CMSG_DATA(p));
-
- ovs_assert(n_fds > 0);
- if (n_fds > SOUTIL_MAX_FDS) {
- VLOG_ERR("%zu fds received but only %d supported",
- n_fds, SOUTIL_MAX_FDS);
- for (i = 0; i < n_fds; i++) {
- close(fds_data[i]);
- }
- goto error;
- }
-
- *n_fdsp = n_fds;
- memcpy(fds, fds_data, n_fds * sizeof *fds);
- }
- }
-
- return retval;
-
-error:
- for (i = 0; i < *n_fdsp; i++) {
- close(fds[i]);
- }
- *n_fdsp = 0;
- return EPROTO;
-}
-
/* Calls ioctl() on an AF_INET sock, passing the specified 'command' and
* 'arg'. Returns 0 if successful, otherwise a positive errno value. */
int