master
c 469 lines 13.3 KB
Raw
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/osdep.h"
9 #include "qapi/error.h"
10 #include "qemu/host-utils.h"
11 #include "qemu/sockets.h"
12
13 #include <sys/socket.h>
14 #include <sys/un.h>
15
16 #include "ivshmem-server.h"
17
18 /* log a message on stdout if verbose=1 */
19 #define IVSHMEM_SERVER_DEBUG(server, fmt, ...) do { \
20 if ((server)->verbose) { \
21 printf(fmt, ## __VA_ARGS__); \
22 } \
23 } while (0)
24
25 /** maximum size of a huge page, used by ivshmem_server_ftruncate() */
26 #define IVSHMEM_SERVER_MAX_HUGEPAGE_SIZE (1024 * 1024 * 1024)
27
28 /** default listen backlog (number of sockets not accepted) */
29 #define IVSHMEM_SERVER_LISTEN_BACKLOG 10
30
31 /* send message to a client unix socket */
32 static int
33 ivshmem_server_send_one_msg(int sock_fd, int64_t peer_id, int fd)
34 {
35 int ret;
36 struct msghdr msg;
37 struct iovec iov[1];
38 union {
39 struct cmsghdr cmsg;
40 char control[CMSG_SPACE(sizeof(int))];
41 } msg_control;
42 struct cmsghdr *cmsg;
43
44 peer_id = GINT64_TO_LE(peer_id);
45 iov[0].iov_base = &peer_id;
46 iov[0].iov_len = sizeof(peer_id);
47
48 memset(&msg, 0, sizeof(msg));
49 msg.msg_iov = iov;
50 msg.msg_iovlen = 1;
51
52 /* if fd is specified, add it in a cmsg */
53 if (fd >= 0) {
54 memset(&msg_control, 0, sizeof(msg_control));
55 msg.msg_control = &msg_control;
56 msg.msg_controllen = sizeof(msg_control);
57 cmsg = CMSG_FIRSTHDR(&msg);
58 cmsg->cmsg_level = SOL_SOCKET;
59 cmsg->cmsg_type = SCM_RIGHTS;
60 cmsg->cmsg_len = CMSG_LEN(sizeof(int));
61 memcpy(CMSG_DATA(cmsg), &fd, sizeof(fd));
62 }
63
64 ret = sendmsg(sock_fd, &msg, 0);
65 if (ret <= 0) {
66 return -1;
67 }
68
69 return 0;
70 }
71
72 /* free a peer when the server advertises a disconnection or when the
73 * server is freed */
74 static void
75 ivshmem_server_free_peer(IvshmemServer *server, IvshmemServerPeer *peer)
76 {
77 unsigned vector;
78 IvshmemServerPeer *other_peer;
79
80 IVSHMEM_SERVER_DEBUG(server, "free peer %" PRId64 "\n", peer->id);
81 close(peer->sock_fd);
82 QTAILQ_REMOVE(&server->peer_list, peer, next);
83
84 /* advertise the deletion to other peers */
85 QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
86 ivshmem_server_send_one_msg(other_peer->sock_fd, peer->id, -1);
87 }
88
89 for (vector = 0; vector < peer->vectors_count; vector++) {
90 event_notifier_cleanup(&peer->vectors[vector]);
91 }
92
93 g_free(peer);
94 }
95
96 /* send the peer id and the shm_fd just after a new client connection */
97 static int
98 ivshmem_server_send_initial_info(IvshmemServer *server, IvshmemServerPeer *peer)
99 {
100 int ret;
101
102 /* send our protocol version first */
103 ret = ivshmem_server_send_one_msg(peer->sock_fd, IVSHMEM_PROTOCOL_VERSION,
104 -1);
105 if (ret < 0) {
106 IVSHMEM_SERVER_DEBUG(server, "cannot send version: %s\n",
107 strerror(errno));
108 return -1;
109 }
110
111 /* send the peer id to the client */
112 ret = ivshmem_server_send_one_msg(peer->sock_fd, peer->id, -1);
113 if (ret < 0) {
114 IVSHMEM_SERVER_DEBUG(server, "cannot send peer id: %s\n",
115 strerror(errno));
116 return -1;
117 }
118
119 /* send the shm_fd */
120 ret = ivshmem_server_send_one_msg(peer->sock_fd, -1, server->shm_fd);
121 if (ret < 0) {
122 IVSHMEM_SERVER_DEBUG(server, "cannot send shm fd: %s\n",
123 strerror(errno));
124 return -1;
125 }
126
127 return 0;
128 }
129
130 /* handle message on listening unix socket (new client connection) */
131 static int
132 ivshmem_server_handle_new_conn(IvshmemServer *server)
133 {
134 IvshmemServerPeer *peer, *other_peer;
135 struct sockaddr_un unaddr;
136 socklen_t unaddr_len;
137 int newfd;
138 unsigned i;
139 Error *local_err = NULL;
140
141 /* accept the incoming connection */
142 unaddr_len = sizeof(unaddr);
143 newfd = qemu_accept(server->sock_fd,
144 (struct sockaddr *)&unaddr, &unaddr_len);
145
146 if (newfd < 0) {
147 IVSHMEM_SERVER_DEBUG(server, "cannot accept() %s\n", strerror(errno));
148 return -1;
149 }
150
151 IVSHMEM_SERVER_DEBUG(server, "accept()=%d\n", newfd);
152
153 if (!qemu_set_blocking(newfd, false, &local_err)) {
154 error_report_err(local_err);
155 close(newfd);
156 return -1;
157 }
158
159 /* allocate new structure for this peer */
160 peer = g_malloc0(sizeof(*peer));
161 peer->sock_fd = newfd;
162
163 /* get an unused peer id */
164 /* XXX: this could use id allocation such as Linux IDA, or simply
165 * a free-list */
166 for (i = 0; i < G_MAXUINT16; i++) {
167 if (ivshmem_server_search_peer(server, server->cur_id) == NULL) {
168 break;
169 }
170 server->cur_id++;
171 }
172 if (i == G_MAXUINT16) {
173 IVSHMEM_SERVER_DEBUG(server, "cannot allocate new client id\n");
174 close(newfd);
175 g_free(peer);
176 return -1;
177 }
178 peer->id = server->cur_id++;
179
180 /* create eventfd, one per vector */
181 peer->vectors_count = server->n_vectors;
182 for (i = 0; i < peer->vectors_count; i++) {
183 if (event_notifier_init(&peer->vectors[i], FALSE) < 0) {
184 IVSHMEM_SERVER_DEBUG(server, "cannot create eventfd\n");
185 goto fail;
186 }
187 }
188
189 /* send peer id and shm fd */
190 if (ivshmem_server_send_initial_info(server, peer) < 0) {
191 IVSHMEM_SERVER_DEBUG(server, "cannot send initial info\n");
192 goto fail;
193 }
194
195 /* advertise the new peer to others */
196 QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
197 for (i = 0; i < peer->vectors_count; i++) {
198 ivshmem_server_send_one_msg(other_peer->sock_fd, peer->id,
199 peer->vectors[i].wfd);
200 }
201 }
202
203 /* advertise the other peers to the new one */
204 QTAILQ_FOREACH(other_peer, &server->peer_list, next) {
205 for (i = 0; i < peer->vectors_count; i++) {
206 ivshmem_server_send_one_msg(peer->sock_fd, other_peer->id,
207 other_peer->vectors[i].wfd);
208 }
209 }
210
211 /* advertise the new peer to itself */
212 for (i = 0; i < peer->vectors_count; i++) {
213 ivshmem_server_send_one_msg(peer->sock_fd, peer->id,
214 event_notifier_get_fd(&peer->vectors[i]));
215 }
216
217 QTAILQ_INSERT_TAIL(&server->peer_list, peer, next);
218 IVSHMEM_SERVER_DEBUG(server, "new peer id = %" PRId64 "\n",
219 peer->id);
220 return 0;
221
222 fail:
223 while (i--) {
224 event_notifier_cleanup(&peer->vectors[i]);
225 }
226 close(newfd);
227 g_free(peer);
228 return -1;
229 }
230
231 /* Try to ftruncate a file to next power of 2 of shmsize.
232 * If it fails; all power of 2 above shmsize are tested until
233 * we reach the maximum huge page size. This is useful
234 * if the shm file is in a hugetlbfs that cannot be truncated to the
235 * shm_size value. */
236 static int
237 ivshmem_server_ftruncate(int fd, unsigned shmsize)
238 {
239 int ret;
240 struct stat mapstat;
241
242 /* align shmsize to next power of 2 */
243 shmsize = pow2ceil(shmsize);
244
245 if (fstat(fd, &mapstat) != -1 && mapstat.st_size == shmsize) {
246 return 0;
247 }
248
249 while (shmsize <= IVSHMEM_SERVER_MAX_HUGEPAGE_SIZE) {
250 ret = ftruncate(fd, shmsize);
251 if (ret == 0) {
252 return ret;
253 }
254 shmsize *= 2;
255 }
256
257 return -1;
258 }
259
260 /* Init a new ivshmem server */
261 int
262 ivshmem_server_init(IvshmemServer *server, const char *unix_sock_path,
263 const char *shm_path, bool use_shm_open,
264 size_t shm_size, unsigned n_vectors,
265 bool verbose)
266 {
267 int ret;
268
269 memset(server, 0, sizeof(*server));
270 server->verbose = verbose;
271
272 ret = snprintf(server->unix_sock_path, sizeof(server->unix_sock_path),
273 "%s", unix_sock_path);
274 if (ret < 0 || ret >= sizeof(server->unix_sock_path)) {
275 IVSHMEM_SERVER_DEBUG(server, "could not copy unix socket path\n");
276 return -1;
277 }
278 ret = snprintf(server->shm_path, sizeof(server->shm_path),
279 "%s", shm_path);
280 if (ret < 0 || ret >= sizeof(server->shm_path)) {
281 IVSHMEM_SERVER_DEBUG(server, "could not copy shm path\n");
282 return -1;
283 }
284
285 server->use_shm_open = use_shm_open;
286 server->shm_size = shm_size;
287 server->n_vectors = n_vectors;
288
289 QTAILQ_INIT(&server->peer_list);
290
291 return 0;
292 }
293
294 /* open shm, create and bind to the unix socket */
295 int
296 ivshmem_server_start(IvshmemServer *server)
297 {
298 struct sockaddr_un s_un;
299 int shm_fd, sock_fd, ret;
300
301 /* open shm file */
302 if (server->use_shm_open) {
303 IVSHMEM_SERVER_DEBUG(server, "Using POSIX shared memory: %s\n",
304 server->shm_path);
305 shm_fd = shm_open(server->shm_path, O_CREAT | O_RDWR, S_IRWXU);
306 } else {
307 gchar *filename = g_strdup_printf("%s/ivshmem.XXXXXX", server->shm_path);
308 IVSHMEM_SERVER_DEBUG(server, "Using file-backed shared memory: %s\n",
309 server->shm_path);
310 shm_fd = mkstemp(filename);
311 unlink(filename);
312 g_free(filename);
313 }
314
315 if (shm_fd < 0) {
316 fprintf(stderr, "cannot open shm file %s: %s\n", server->shm_path,
317 strerror(errno));
318 return -1;
319 }
320 if (ivshmem_server_ftruncate(shm_fd, server->shm_size) < 0) {
321 fprintf(stderr, "ftruncate(%s) failed: %s\n", server->shm_path,
322 strerror(errno));
323 goto err_close_shm;
324 }
325
326 IVSHMEM_SERVER_DEBUG(server, "create & bind socket %s\n",
327 server->unix_sock_path);
328
329 /* create the unix listening socket */
330 sock_fd = socket(AF_UNIX, SOCK_STREAM, 0);
331 if (sock_fd < 0) {
332 IVSHMEM_SERVER_DEBUG(server, "cannot create socket: %s\n",
333 strerror(errno));
334 goto err_close_shm;
335 }
336
337 s_un.sun_family = AF_UNIX;
338 ret = snprintf(s_un.sun_path, sizeof(s_un.sun_path), "%s",
339 server->unix_sock_path);
340 if (ret < 0 || ret >= sizeof(s_un.sun_path)) {
341 IVSHMEM_SERVER_DEBUG(server, "could not copy unix socket path\n");
342 goto err_close_sock;
343 }
344 if (bind(sock_fd, (struct sockaddr *)&s_un, sizeof(s_un)) < 0) {
345 IVSHMEM_SERVER_DEBUG(server, "cannot connect to %s: %s\n", s_un.sun_path,
346 strerror(errno));
347 goto err_close_sock;
348 }
349
350 if (listen(sock_fd, IVSHMEM_SERVER_LISTEN_BACKLOG) < 0) {
351 IVSHMEM_SERVER_DEBUG(server, "listen() failed: %s\n", strerror(errno));
352 goto err_close_sock;
353 }
354
355 server->sock_fd = sock_fd;
356 server->shm_fd = shm_fd;
357
358 return 0;
359
360 err_close_sock:
361 close(sock_fd);
362 err_close_shm:
363 if (server->use_shm_open) {
364 shm_unlink(server->shm_path);
365 }
366 close(shm_fd);
367 return -1;
368 }
369
370 /* close connections to clients, the unix socket and the shm fd */
371 void
372 ivshmem_server_close(IvshmemServer *server)
373 {
374 IvshmemServerPeer *peer, *npeer;
375
376 IVSHMEM_SERVER_DEBUG(server, "close server\n");
377
378 QTAILQ_FOREACH_SAFE(peer, &server->peer_list, next, npeer) {
379 ivshmem_server_free_peer(server, peer);
380 }
381
382 unlink(server->unix_sock_path);
383 if (server->use_shm_open) {
384 shm_unlink(server->shm_path);
385 }
386 close(server->sock_fd);
387 close(server->shm_fd);
388 server->sock_fd = -1;
389 server->shm_fd = -1;
390 }
391
392 /* get the fd_set according to the unix socket and the peer list */
393 void
394 ivshmem_server_get_fds(const IvshmemServer *server, fd_set *fds, int *maxfd)
395 {
396 IvshmemServerPeer *peer;
397
398 if (server->sock_fd == -1) {
399 return;
400 }
401
402 FD_SET(server->sock_fd, fds);
403 if (server->sock_fd >= *maxfd) {
404 *maxfd = server->sock_fd + 1;
405 }
406
407 QTAILQ_FOREACH(peer, &server->peer_list, next) {
408 FD_SET(peer->sock_fd, fds);
409 if (peer->sock_fd >= *maxfd) {
410 *maxfd = peer->sock_fd + 1;
411 }
412 }
413 }
414
415 /* process incoming messages on the sockets in fd_set */
416 int
417 ivshmem_server_handle_fds(IvshmemServer *server, fd_set *fds, int maxfd)
418 {
419 IvshmemServerPeer *peer, *peer_next;
420
421 if (server->sock_fd < maxfd && FD_ISSET(server->sock_fd, fds) &&
422 ivshmem_server_handle_new_conn(server) < 0 && errno != EINTR) {
423 IVSHMEM_SERVER_DEBUG(server, "ivshmem_server_handle_new_conn() "
424 "failed\n");
425 return -1;
426 }
427
428 QTAILQ_FOREACH_SAFE(peer, &server->peer_list, next, peer_next) {
429 /* any message from a peer socket result in a close() */
430 IVSHMEM_SERVER_DEBUG(server, "peer->sock_fd=%d\n", peer->sock_fd);
431 if (peer->sock_fd < maxfd && FD_ISSET(peer->sock_fd, fds)) {
432 ivshmem_server_free_peer(server, peer);
433 }
434 }
435
436 return 0;
437 }
438
439 /* lookup peer from its id */
440 IvshmemServerPeer *
441 ivshmem_server_search_peer(IvshmemServer *server, int64_t peer_id)
442 {
443 IvshmemServerPeer *peer;
444
445 QTAILQ_FOREACH(peer, &server->peer_list, next) {
446 if (peer->id == peer_id) {
447 return peer;
448 }
449 }
450 return NULL;
451 }
452
453 /* dump our info, the list of peers their vectors on stdout */
454 void
455 ivshmem_server_dump(const IvshmemServer *server)
456 {
457 const IvshmemServerPeer *peer;
458 unsigned vector;
459
460 /* dump peers */
461 QTAILQ_FOREACH(peer, &server->peer_list, next) {
462 printf("peer_id = %" PRId64 "\n", peer->id);
463
464 for (vector = 0; vector < peer->vectors_count; vector++) {
465 printf(" vector %d is enabled (fd=%d)\n", vector,
466 event_notifier_get_fd(&peer->vectors[vector]));
467 }
468 }
469 }