1 /*
2  * Copyright 6WIND S.A., 2014
3  *
4  * This work is licensed under the terms of the GNU GPL, version 2 or
5  * (at your option) any later version.  See the COPYING file in the
6  * top-level directory.
7  */
8 #include "qemu-common.h"
9 #include "qemu/sockets.h"
10 
11 #include <sys/mman.h>
12 #include <sys/types.h>
13 #include <sys/socket.h>
14 #include <sys/un.h>
15 #ifdef CONFIG_LINUX
16 #include <sys/vfs.h>
17 #endif
18 
19 #include "ivshmem-server.h"
20 
21 /* log a message on stdout if verbose=1 */
22 #define IVSHMEM_SERVER_DEBUG(server, fmt, ...) do { \
23         if ((server)->verbose) {         \
24             printf(fmt, ## __VA_ARGS__); \
25         }                                \
26     } while (0)
27 
28 /** maximum size of a huge page, used by ivshmem_server_ftruncate() */
29 #define IVSHMEM_SERVER_MAX_HUGEPAGE_SIZE (1024 * 1024 * 1024)
30 
31 /** default listen backlog (number of sockets not accepted) */
32 #define IVSHMEM_SERVER_LISTEN_BACKLOG 10
33 
34 /* send message to a client unix socket */
35 static int
36 ivshmem_server_send_one_msg(int sock_fd, int64_t peer_id, int fd)
37 {
38     int ret;
39     struct msghdr msg;
40     struct iovec iov[1];
41     union {
42         struct cmsghdr cmsg;
43         char control[CMSG_SPACE(sizeof(int))];
44     } msg_control;
45     struct cmsghdr *cmsg;
46 
47     peer_id = GINT64_TO_LE(peer_id);
48     iov[0].iov_base = &peer_id;
49     iov[0].iov_len = sizeof(peer_id);
50 
51     memset(&msg, 0, sizeof(msg));
52     msg.msg_iov = iov;
53     msg.msg_iovlen = 1;
54 
55     /* if fd is specified, add it in a cmsg */
56     if (fd >= 0) {
57         memset(&msg_control, 0, sizeof(msg_control));
58         msg.msg_control = &msg_control;
59         msg.msg_controllen = sizeof(msg_control);
60         cmsg = CMSG_FIRSTHDR(&msg);
61         cmsg->cmsg_level = SOL_SOCKET;
62         cmsg->cmsg_type = SCM_RIGHTS;
63         cmsg->cmsg_len = CMSG_LEN(sizeof(int));
64         memcpy(CMSG_DATA(cmsg), &fd, sizeof(fd));
65     }
66 
67     ret = sendmsg(sock_fd, &msg, 0);
68     if (ret <= 0) {
69         return -1;
70     }
71 
72     return 0;
73 }
74 
75 /* free a peer when the server advertises a disconnection or when the
76  * server is freed */
77 static void
78 ivshmem_server_free_peer(IvshmemServer *server, IvshmemServerPeer *peer)
79 {
80     unsigned vector;
81     IvshmemServerPeer *other_peer;
82 
83     IVSHMEM_SERVER_DEBUG(server, "free peer %" PRId64 "\n", peer->id);
84     close(peer->sock_fd);
85     QTAILQ_REMOVE(&server->peer_list, peer, next);
86 
87     /* advertise the deletion to other peers */
88     QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
89         ivshmem_server_send_one_msg(other_peer->sock_fd, peer->id, -1);
90     }
91 
92     for (vector = 0; vector < peer->vectors_count; vector++) {
93         event_notifier_cleanup(&peer->vectors[vector]);
94     }
95 
96     g_free(peer);
97 }
98 
99 /* send the peer id and the shm_fd just after a new client connection */
100 static int
101 ivshmem_server_send_initial_info(IvshmemServer *server, IvshmemServerPeer *peer)
102 {
103     int ret;
104 
105     /* send our protocol version first */
106     ret = ivshmem_server_send_one_msg(peer->sock_fd, IVSHMEM_PROTOCOL_VERSION,
107                                       -1);
108     if (ret < 0) {
109         IVSHMEM_SERVER_DEBUG(server, "cannot send version: %s\n",
110                              strerror(errno));
111         return -1;
112     }
113 
114     /* send the peer id to the client */
115     ret = ivshmem_server_send_one_msg(peer->sock_fd, peer->id, -1);
116     if (ret < 0) {
117         IVSHMEM_SERVER_DEBUG(server, "cannot send peer id: %s\n",
118                              strerror(errno));
119         return -1;
120     }
121 
122     /* send the shm_fd */
123     ret = ivshmem_server_send_one_msg(peer->sock_fd, -1, server->shm_fd);
124     if (ret < 0) {
125         IVSHMEM_SERVER_DEBUG(server, "cannot send shm fd: %s\n",
126                              strerror(errno));
127         return -1;
128     }
129 
130     return 0;
131 }
132 
133 /* handle message on listening unix socket (new client connection) */
134 static int
135 ivshmem_server_handle_new_conn(IvshmemServer *server)
136 {
137     IvshmemServerPeer *peer, *other_peer;
138     struct sockaddr_un unaddr;
139     socklen_t unaddr_len;
140     int newfd;
141     unsigned i;
142 
143     /* accept the incoming connection */
144     unaddr_len = sizeof(unaddr);
145     newfd = qemu_accept(server->sock_fd,
146                         (struct sockaddr *)&unaddr, &unaddr_len);
147 
148     if (newfd < 0) {
149         IVSHMEM_SERVER_DEBUG(server, "cannot accept() %s\n", strerror(errno));
150         return -1;
151     }
152 
153     qemu_set_nonblock(newfd);
154     IVSHMEM_SERVER_DEBUG(server, "accept()=%d\n", newfd);
155 
156     /* allocate new structure for this peer */
157     peer = g_malloc0(sizeof(*peer));
158     peer->sock_fd = newfd;
159 
160     /* get an unused peer id */
161     /* XXX: this could use id allocation such as Linux IDA, or simply
162      * a free-list */
163     for (i = 0; i < G_MAXUINT16; i++) {
164         if (ivshmem_server_search_peer(server, server->cur_id) == NULL) {
165             break;
166         }
167         server->cur_id++;
168     }
169     if (i == G_MAXUINT16) {
170         IVSHMEM_SERVER_DEBUG(server, "cannot allocate new client id\n");
171         goto fail;
172     }
173     peer->id = server->cur_id++;
174 
175     /* create eventfd, one per vector */
176     peer->vectors_count = server->n_vectors;
177     for (i = 0; i < peer->vectors_count; i++) {
178         if (event_notifier_init(&peer->vectors[i], FALSE) < 0) {
179             IVSHMEM_SERVER_DEBUG(server, "cannot create eventfd\n");
180             goto fail;
181         }
182     }
183 
184     /* send peer id and shm fd */
185     if (ivshmem_server_send_initial_info(server, peer) < 0) {
186         IVSHMEM_SERVER_DEBUG(server, "cannot send initial info\n");
187         goto fail;
188     }
189 
190     /* advertise the new peer to others */
191     QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
192         for (i = 0; i < peer->vectors_count; i++) {
193             ivshmem_server_send_one_msg(other_peer->sock_fd, peer->id,
194                                         peer->vectors[i].wfd);
195         }
196     }
197 
198     /* advertise the other peers to the new one */
199     QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
200         for (i = 0; i < peer->vectors_count; i++) {
201             ivshmem_server_send_one_msg(peer->sock_fd, other_peer->id,
202                                         other_peer->vectors[i].wfd);
203         }
204     }
205 
206     /* advertise the new peer to itself */
207     for (i = 0; i < peer->vectors_count; i++) {
208         ivshmem_server_send_one_msg(peer->sock_fd, peer->id,
209                                     event_notifier_get_fd(&peer->vectors[i]));
210     }
211 
212     QTAILQ_INSERT_TAIL(&server->peer_list, peer, next);
213     IVSHMEM_SERVER_DEBUG(server, "new peer id = %" PRId64 "\n",
214                          peer->id);
215     return 0;
216 
217 fail:
218     while (i--) {
219         event_notifier_cleanup(&peer->vectors[i]);
220     }
221     close(newfd);
222     g_free(peer);
223     return -1;
224 }
225 
226 /* Try to ftruncate a file to next power of 2 of shmsize.
227  * If it fails; all power of 2 above shmsize are tested until
228  * we reach the maximum huge page size. This is useful
229  * if the shm file is in a hugetlbfs that cannot be truncated to the
230  * shm_size value. */
231 static int
232 ivshmem_server_ftruncate(int fd, unsigned shmsize)
233 {
234     int ret;
235     struct stat mapstat;
236 
237     /* align shmsize to next power of 2 */
238     shmsize = pow2ceil(shmsize);
239 
240     if (fstat(fd, &mapstat) != -1 && mapstat.st_size == shmsize) {
241         return 0;
242     }
243 
244     while (shmsize <= IVSHMEM_SERVER_MAX_HUGEPAGE_SIZE) {
245         ret = ftruncate(fd, shmsize);
246         if (ret == 0) {
247             return ret;
248         }
249         shmsize *= 2;
250     }
251 
252     return -1;
253 }
254 
255 /* Init a new ivshmem server */
256 int
257 ivshmem_server_init(IvshmemServer *server, const char *unix_sock_path,
258                     const char *shm_path, size_t shm_size, unsigned n_vectors,
259                     bool verbose)
260 {
261     int ret;
262 
263     memset(server, 0, sizeof(*server));
264     server->verbose = verbose;
265 
266     ret = snprintf(server->unix_sock_path, sizeof(server->unix_sock_path),
267                    "%s", unix_sock_path);
268     if (ret < 0 || ret >= sizeof(server->unix_sock_path)) {
269         IVSHMEM_SERVER_DEBUG(server, "could not copy unix socket path\n");
270         return -1;
271     }
272     ret = snprintf(server->shm_path, sizeof(server->shm_path),
273                    "%s", shm_path);
274     if (ret < 0 || ret >= sizeof(server->shm_path)) {
275         IVSHMEM_SERVER_DEBUG(server, "could not copy shm path\n");
276         return -1;
277     }
278 
279     server->shm_size = shm_size;
280     server->n_vectors = n_vectors;
281 
282     QTAILQ_INIT(&server->peer_list);
283 
284     return 0;
285 }
286 
287 #ifdef CONFIG_LINUX
288 
289 #define HUGETLBFS_MAGIC       0x958458f6
290 
291 static long gethugepagesize(const char *path)
292 {
293     struct statfs fs;
294     int ret;
295 
296     do {
297         ret = statfs(path, &fs);
298     } while (ret != 0 && errno == EINTR);
299 
300     if (ret != 0) {
301         return -1;
302     }
303 
304     if (fs.f_type != HUGETLBFS_MAGIC) {
305         return -1;
306     }
307 
308     return fs.f_bsize;
309 }
310 #endif
311 
312 /* open shm, create and bind to the unix socket */
313 int
314 ivshmem_server_start(IvshmemServer *server)
315 {
316     struct sockaddr_un sun;
317     int shm_fd, sock_fd, ret;
318 
319     /* open shm file */
320 #ifdef CONFIG_LINUX
321     long hpagesize;
322 
323     hpagesize = gethugepagesize(server->shm_path);
324     if (hpagesize < 0 && errno != ENOENT) {
325         IVSHMEM_SERVER_DEBUG(server, "cannot stat shm file %s: %s\n",
326                              server->shm_path, strerror(errno));
327     }
328 
329     if (hpagesize > 0) {
330         gchar *filename = g_strdup_printf("%s/ivshmem.XXXXXX", server->shm_path);
331         IVSHMEM_SERVER_DEBUG(server, "Using hugepages: %s\n", server->shm_path);
332         shm_fd = mkstemp(filename);
333         unlink(filename);
334         g_free(filename);
335     } else
336 #endif
337     {
338         IVSHMEM_SERVER_DEBUG(server, "Using POSIX shared memory: %s\n",
339                              server->shm_path);
340         shm_fd = shm_open(server->shm_path, O_CREAT|O_RDWR, S_IRWXU);
341     }
342 
343     if (shm_fd < 0) {
344         fprintf(stderr, "cannot open shm file %s: %s\n", server->shm_path,
345                 strerror(errno));
346         return -1;
347     }
348     if (ivshmem_server_ftruncate(shm_fd, server->shm_size) < 0) {
349         fprintf(stderr, "ftruncate(%s) failed: %s\n", server->shm_path,
350                 strerror(errno));
351         goto err_close_shm;
352     }
353 
354     IVSHMEM_SERVER_DEBUG(server, "create & bind socket %s\n",
355                          server->unix_sock_path);
356 
357     /* create the unix listening socket */
358     sock_fd = socket(AF_UNIX, SOCK_STREAM, 0);
359     if (sock_fd < 0) {
360         IVSHMEM_SERVER_DEBUG(server, "cannot create socket: %s\n",
361                              strerror(errno));
362         goto err_close_shm;
363     }
364 
365     sun.sun_family = AF_UNIX;
366     ret = snprintf(sun.sun_path, sizeof(sun.sun_path), "%s",
367                    server->unix_sock_path);
368     if (ret < 0 || ret >= sizeof(sun.sun_path)) {
369         IVSHMEM_SERVER_DEBUG(server, "could not copy unix socket path\n");
370         goto err_close_sock;
371     }
372     if (bind(sock_fd, (struct sockaddr *)&sun, sizeof(sun)) < 0) {
373         IVSHMEM_SERVER_DEBUG(server, "cannot connect to %s: %s\n", sun.sun_path,
374                              strerror(errno));
375         goto err_close_sock;
376     }
377 
378     if (listen(sock_fd, IVSHMEM_SERVER_LISTEN_BACKLOG) < 0) {
379         IVSHMEM_SERVER_DEBUG(server, "listen() failed: %s\n", strerror(errno));
380         goto err_close_sock;
381     }
382 
383     server->sock_fd = sock_fd;
384     server->shm_fd = shm_fd;
385 
386     return 0;
387 
388 err_close_sock:
389     close(sock_fd);
390 err_close_shm:
391     close(shm_fd);
392     return -1;
393 }
394 
395 /* close connections to clients, the unix socket and the shm fd */
396 void
397 ivshmem_server_close(IvshmemServer *server)
398 {
399     IvshmemServerPeer *peer, *npeer;
400 
401     IVSHMEM_SERVER_DEBUG(server, "close server\n");
402 
403     QTAILQ_FOREACH_SAFE(peer, &server->peer_list, next, npeer) {
404         ivshmem_server_free_peer(server, peer);
405     }
406 
407     unlink(server->unix_sock_path);
408     close(server->sock_fd);
409     close(server->shm_fd);
410     server->sock_fd = -1;
411     server->shm_fd = -1;
412 }
413 
414 /* get the fd_set according to the unix socket and the peer list */
415 void
416 ivshmem_server_get_fds(const IvshmemServer *server, fd_set *fds, int *maxfd)
417 {
418     IvshmemServerPeer *peer;
419 
420     if (server->sock_fd == -1) {
421         return;
422     }
423 
424     FD_SET(server->sock_fd, fds);
425     if (server->sock_fd >= *maxfd) {
426         *maxfd = server->sock_fd + 1;
427     }
428 
429     QTAILQ_FOREACH(peer, &server->peer_list, next) {
430         FD_SET(peer->sock_fd, fds);
431         if (peer->sock_fd >= *maxfd) {
432             *maxfd = peer->sock_fd + 1;
433         }
434     }
435 }
436 
437 /* process incoming messages on the sockets in fd_set */
438 int
439 ivshmem_server_handle_fds(IvshmemServer *server, fd_set *fds, int maxfd)
440 {
441     IvshmemServerPeer *peer, *peer_next;
442 
443     if (server->sock_fd < maxfd && FD_ISSET(server->sock_fd, fds) &&
444         ivshmem_server_handle_new_conn(server) < 0 && errno != EINTR) {
445         IVSHMEM_SERVER_DEBUG(server, "ivshmem_server_handle_new_conn() "
446                              "failed\n");
447         return -1;
448     }
449 
450     QTAILQ_FOREACH_SAFE(peer, &server->peer_list, next, peer_next) {
451         /* any message from a peer socket result in a close() */
452         IVSHMEM_SERVER_DEBUG(server, "peer->sock_fd=%d\n", peer->sock_fd);
453         if (peer->sock_fd < maxfd && FD_ISSET(peer->sock_fd, fds)) {
454             ivshmem_server_free_peer(server, peer);
455         }
456     }
457 
458     return 0;
459 }
460 
461 /* lookup peer from its id */
462 IvshmemServerPeer *
463 ivshmem_server_search_peer(IvshmemServer *server, int64_t peer_id)
464 {
465     IvshmemServerPeer *peer;
466 
467     QTAILQ_FOREACH(peer, &server->peer_list, next) {
468         if (peer->id == peer_id) {
469             return peer;
470         }
471     }
472     return NULL;
473 }
474 
475 /* dump our info, the list of peers their vectors on stdout */
476 void
477 ivshmem_server_dump(const IvshmemServer *server)
478 {
479     const IvshmemServerPeer *peer;
480     unsigned vector;
481 
482     /* dump peers */
483     QTAILQ_FOREACH(peer, &server->peer_list, next) {
484         printf("peer_id = %" PRId64 "\n", peer->id);
485 
486         for (vector = 0; vector < peer->vectors_count; vector++) {
487             printf("  vector %d is enabled (fd=%d)\n", vector,
488                    event_notifier_get_fd(&peer->vectors[vector]));
489         }
490     }
491 }
492