xref: /openbmc/qemu/util/vhost-user-server.c (revision 2e1cacfb)
1 /*
2  * Sharing QEMU devices via vhost-user protocol
3  *
4  * Copyright (c) Coiby Xu <coiby.xu@gmail.com>.
5  * Copyright (c) 2020 Red Hat, Inc.
6  *
7  * This work is licensed under the terms of the GNU GPL, version 2 or
8  * later.  See the COPYING file in the top-level directory.
9  */
10 #include "qemu/osdep.h"
11 #include "qemu/error-report.h"
12 #include "qemu/main-loop.h"
13 #include "qemu/vhost-user-server.h"
14 #include "block/aio-wait.h"
15 
16 /*
17  * Theory of operation:
18  *
19  * VuServer is started and stopped by vhost_user_server_start() and
20  * vhost_user_server_stop() from the main loop thread. Starting the server
21  * opens a vhost-user UNIX domain socket and listens for incoming connections.
22  * Only one connection is allowed at a time.
23  *
24  * The connection is handled by the vu_client_trip() coroutine in the
25  * VuServer->ctx AioContext. The coroutine consists of a vu_dispatch() loop
26  * where libvhost-user calls vu_message_read() to receive the next vhost-user
27  * protocol messages over the UNIX domain socket.
28  *
29  * When virtqueues are set up libvhost-user calls set_watch() to monitor kick
30  * fds. These fds are also handled in the VuServer->ctx AioContext.
31  *
32  * Both vu_client_trip() and kick fd monitoring can be stopped by shutting down
33  * the socket connection. Shutting down the socket connection causes
34  * vu_message_read() to fail since no more data can be received from the socket.
35  * After vu_dispatch() fails, vu_client_trip() calls vu_deinit() to stop
36  * libvhost-user before terminating the coroutine. vu_deinit() calls
37  * remove_watch() to stop monitoring kick fds and this stops virtqueue
38  * processing.
39  *
40  * When vu_client_trip() has finished cleaning up it schedules a BH in the main
41  * loop thread to accept the next client connection.
42  *
43  * When libvhost-user detects an error it calls panic_cb() and sets the
44  * dev->broken flag. Both vu_client_trip() and kick fd processing stop when
45  * the dev->broken flag is set.
46  *
47  * It is possible to switch AioContexts using
48  * vhost_user_server_detach_aio_context() and
49  * vhost_user_server_attach_aio_context(). They stop monitoring fds in the old
50  * AioContext and resume monitoring in the new AioContext. The vu_client_trip()
51  * coroutine remains in a yielded state during the switch. This is made
52  * possible by QIOChannel's support for spurious coroutine re-entry in
53  * qio_channel_yield(). The coroutine will restart I/O when re-entered from the
54  * new AioContext.
55  */
56 
57 static void vmsg_close_fds(VhostUserMsg *vmsg)
58 {
59     int i;
60     for (i = 0; i < vmsg->fd_num; i++) {
61         close(vmsg->fds[i]);
62     }
63 }
64 
65 static void vmsg_unblock_fds(VhostUserMsg *vmsg)
66 {
67     int i;
68 
69     /*
70      * These messages carry fd used to map memory, not to send/receive messages,
71      * so this operation is useless. In addition, in some systems this
72      * operation can fail (e.g. in macOS setting an fd returned by shm_open()
73      * non-blocking fails with errno = ENOTTY)
74      */
75     if (vmsg->request == VHOST_USER_ADD_MEM_REG ||
76         vmsg->request == VHOST_USER_SET_MEM_TABLE) {
77         return;
78     }
79 
80     for (i = 0; i < vmsg->fd_num; i++) {
81         qemu_socket_set_nonblock(vmsg->fds[i]);
82     }
83 }
84 
85 static void panic_cb(VuDev *vu_dev, const char *buf)
86 {
87     error_report("vu_panic: %s", buf);
88 }
89 
90 void vhost_user_server_inc_in_flight(VuServer *server)
91 {
92     assert(!server->wait_idle);
93     qatomic_inc(&server->in_flight);
94 }
95 
96 void vhost_user_server_dec_in_flight(VuServer *server)
97 {
98     if (qatomic_fetch_dec(&server->in_flight) == 1) {
99         if (server->wait_idle) {
100             aio_co_wake(server->co_trip);
101         }
102     }
103 }
104 
105 bool vhost_user_server_has_in_flight(VuServer *server)
106 {
107     return qatomic_load_acquire(&server->in_flight) > 0;
108 }
109 
110 static bool coroutine_fn
111 vu_message_read(VuDev *vu_dev, int conn_fd, VhostUserMsg *vmsg)
112 {
113     struct iovec iov = {
114         .iov_base = (char *)vmsg,
115         .iov_len = VHOST_USER_HDR_SIZE,
116     };
117     int rc, read_bytes = 0;
118     Error *local_err = NULL;
119     const size_t max_fds = G_N_ELEMENTS(vmsg->fds);
120     VuServer *server = container_of(vu_dev, VuServer, vu_dev);
121     QIOChannel *ioc = server->ioc;
122 
123     vmsg->fd_num = 0;
124     if (!ioc) {
125         error_report_err(local_err);
126         goto fail;
127     }
128 
129     assert(qemu_in_coroutine());
130     do {
131         size_t nfds = 0;
132         int *fds = NULL;
133 
134         /*
135          * qio_channel_readv_full may have short reads, keeping calling it
136          * until getting VHOST_USER_HDR_SIZE or 0 bytes in total
137          */
138         rc = qio_channel_readv_full(ioc, &iov, 1, &fds, &nfds, 0, &local_err);
139         if (rc < 0) {
140             if (rc == QIO_CHANNEL_ERR_BLOCK) {
141                 assert(local_err == NULL);
142                 if (server->ctx) {
143                     server->in_qio_channel_yield = true;
144                     qio_channel_yield(ioc, G_IO_IN);
145                     server->in_qio_channel_yield = false;
146                 } else {
147                     return false;
148                 }
149                 continue;
150             } else {
151                 error_report_err(local_err);
152                 goto fail;
153             }
154         }
155 
156         if (nfds > 0) {
157             if (vmsg->fd_num + nfds > max_fds) {
158                 error_report("A maximum of %zu fds are allowed, "
159                              "however got %zu fds now",
160                              max_fds, vmsg->fd_num + nfds);
161                 g_free(fds);
162                 goto fail;
163             }
164             memcpy(vmsg->fds + vmsg->fd_num, fds, nfds * sizeof(vmsg->fds[0]));
165             vmsg->fd_num += nfds;
166             g_free(fds);
167         }
168 
169         if (rc == 0) { /* socket closed */
170             goto fail;
171         }
172 
173         iov.iov_base += rc;
174         iov.iov_len -= rc;
175         read_bytes += rc;
176     } while (read_bytes != VHOST_USER_HDR_SIZE);
177 
178     /* qio_channel_readv_full will make socket fds blocking, unblock them */
179     vmsg_unblock_fds(vmsg);
180     if (vmsg->size > sizeof(vmsg->payload)) {
181         error_report("Error: too big message request: %d, "
182                      "size: vmsg->size: %u, "
183                      "while sizeof(vmsg->payload) = %zu",
184                      vmsg->request, vmsg->size, sizeof(vmsg->payload));
185         goto fail;
186     }
187 
188     struct iovec iov_payload = {
189         .iov_base = (char *)&vmsg->payload,
190         .iov_len = vmsg->size,
191     };
192     if (vmsg->size) {
193         rc = qio_channel_readv_all_eof(ioc, &iov_payload, 1, &local_err);
194         if (rc != 1) {
195             if (local_err) {
196                 error_report_err(local_err);
197             }
198             goto fail;
199         }
200     }
201 
202     return true;
203 
204 fail:
205     vmsg_close_fds(vmsg);
206 
207     return false;
208 }
209 
210 static coroutine_fn void vu_client_trip(void *opaque)
211 {
212     VuServer *server = opaque;
213     VuDev *vu_dev = &server->vu_dev;
214 
215     while (!vu_dev->broken) {
216         if (server->quiescing) {
217             server->co_trip = NULL;
218             aio_wait_kick();
219             return;
220         }
221         /* vu_dispatch() returns false if server->ctx went away */
222         if (!vu_dispatch(vu_dev) && server->ctx) {
223             break;
224         }
225     }
226 
227     if (vhost_user_server_has_in_flight(server)) {
228         /* Wait for requests to complete before we can unmap the memory */
229         server->wait_idle = true;
230         qemu_coroutine_yield();
231         server->wait_idle = false;
232     }
233     assert(!vhost_user_server_has_in_flight(server));
234 
235     vu_deinit(vu_dev);
236 
237     /* vu_deinit() should have called remove_watch() */
238     assert(QTAILQ_EMPTY(&server->vu_fd_watches));
239 
240     object_unref(OBJECT(server->sioc));
241     server->sioc = NULL;
242 
243     object_unref(OBJECT(server->ioc));
244     server->ioc = NULL;
245 
246     server->co_trip = NULL;
247     if (server->restart_listener_bh) {
248         qemu_bh_schedule(server->restart_listener_bh);
249     }
250     aio_wait_kick();
251 }
252 
253 /*
254  * a wrapper for vu_kick_cb
255  *
256  * since aio_dispatch can only pass one user data pointer to the
257  * callback function, pack VuDev and pvt into a struct. Then unpack it
258  * and pass them to vu_kick_cb
259  */
260 static void kick_handler(void *opaque)
261 {
262     VuFdWatch *vu_fd_watch = opaque;
263     VuDev *vu_dev = vu_fd_watch->vu_dev;
264 
265     vu_fd_watch->cb(vu_dev, 0, vu_fd_watch->pvt);
266 
267     /* Stop vu_client_trip() if an error occurred in vu_fd_watch->cb() */
268     if (vu_dev->broken) {
269         VuServer *server = container_of(vu_dev, VuServer, vu_dev);
270 
271         qio_channel_shutdown(server->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
272     }
273 }
274 
275 static VuFdWatch *find_vu_fd_watch(VuServer *server, int fd)
276 {
277 
278     VuFdWatch *vu_fd_watch, *next;
279     QTAILQ_FOREACH_SAFE(vu_fd_watch, &server->vu_fd_watches, next, next) {
280         if (vu_fd_watch->fd == fd) {
281             return vu_fd_watch;
282         }
283     }
284     return NULL;
285 }
286 
287 static void
288 set_watch(VuDev *vu_dev, int fd, int vu_evt,
289           vu_watch_cb cb, void *pvt)
290 {
291 
292     VuServer *server = container_of(vu_dev, VuServer, vu_dev);
293     g_assert(vu_dev);
294     g_assert(fd >= 0);
295     g_assert(cb);
296 
297     VuFdWatch *vu_fd_watch = find_vu_fd_watch(server, fd);
298 
299     if (!vu_fd_watch) {
300         vu_fd_watch = g_new0(VuFdWatch, 1);
301 
302         QTAILQ_INSERT_TAIL(&server->vu_fd_watches, vu_fd_watch, next);
303 
304         vu_fd_watch->fd = fd;
305         vu_fd_watch->cb = cb;
306         qemu_socket_set_nonblock(fd);
307         aio_set_fd_handler(server->ctx, fd, kick_handler,
308                            NULL, NULL, NULL, vu_fd_watch);
309         vu_fd_watch->vu_dev = vu_dev;
310         vu_fd_watch->pvt = pvt;
311     }
312 }
313 
314 
315 static void remove_watch(VuDev *vu_dev, int fd)
316 {
317     VuServer *server;
318     g_assert(vu_dev);
319     g_assert(fd >= 0);
320 
321     server = container_of(vu_dev, VuServer, vu_dev);
322 
323     VuFdWatch *vu_fd_watch = find_vu_fd_watch(server, fd);
324 
325     if (!vu_fd_watch) {
326         return;
327     }
328     aio_set_fd_handler(server->ctx, fd, NULL, NULL, NULL, NULL, NULL);
329 
330     QTAILQ_REMOVE(&server->vu_fd_watches, vu_fd_watch, next);
331     g_free(vu_fd_watch);
332 }
333 
334 
335 static void vu_accept(QIONetListener *listener, QIOChannelSocket *sioc,
336                       gpointer opaque)
337 {
338     VuServer *server = opaque;
339 
340     if (server->sioc) {
341         warn_report("Only one vhost-user client is allowed to "
342                     "connect the server one time");
343         return;
344     }
345 
346     if (!vu_init(&server->vu_dev, server->max_queues, sioc->fd, panic_cb,
347                  vu_message_read, set_watch, remove_watch, server->vu_iface)) {
348         error_report("Failed to initialize libvhost-user");
349         return;
350     }
351 
352     /*
353      * Unset the callback function for network listener to make another
354      * vhost-user client keeping waiting until this client disconnects
355      */
356     qio_net_listener_set_client_func(server->listener,
357                                      NULL,
358                                      NULL,
359                                      NULL);
360     server->sioc = sioc;
361     /*
362      * Increase the object reference, so sioc will not freed by
363      * qio_net_listener_channel_func which will call object_unref(OBJECT(sioc))
364      */
365     object_ref(OBJECT(server->sioc));
366     qio_channel_set_name(QIO_CHANNEL(sioc), "vhost-user client");
367     server->ioc = QIO_CHANNEL(sioc);
368     object_ref(OBJECT(server->ioc));
369 
370     /* TODO vu_message_write() spins if non-blocking! */
371     qio_channel_set_blocking(server->ioc, false, NULL);
372 
373     qio_channel_set_follow_coroutine_ctx(server->ioc, true);
374 
375     vhost_user_server_attach_aio_context(server, server->ctx);
376 }
377 
378 /* server->ctx acquired by caller */
379 void vhost_user_server_stop(VuServer *server)
380 {
381     qemu_bh_delete(server->restart_listener_bh);
382     server->restart_listener_bh = NULL;
383 
384     if (server->sioc) {
385         VuFdWatch *vu_fd_watch;
386 
387         QTAILQ_FOREACH(vu_fd_watch, &server->vu_fd_watches, next) {
388             aio_set_fd_handler(server->ctx, vu_fd_watch->fd,
389                                NULL, NULL, NULL, NULL, vu_fd_watch);
390         }
391 
392         qio_channel_shutdown(server->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
393 
394         AIO_WAIT_WHILE(server->ctx, server->co_trip);
395     }
396 
397     if (server->listener) {
398         qio_net_listener_disconnect(server->listener);
399         object_unref(OBJECT(server->listener));
400     }
401 }
402 
403 /*
404  * Allow the next client to connect to the server. Called from a BH in the main
405  * loop.
406  */
407 static void restart_listener_bh(void *opaque)
408 {
409     VuServer *server = opaque;
410 
411     qio_net_listener_set_client_func(server->listener, vu_accept, server,
412                                      NULL);
413 }
414 
415 /* Called with ctx acquired */
416 void vhost_user_server_attach_aio_context(VuServer *server, AioContext *ctx)
417 {
418     VuFdWatch *vu_fd_watch;
419 
420     server->ctx = ctx;
421 
422     if (!server->sioc) {
423         return;
424     }
425 
426     QTAILQ_FOREACH(vu_fd_watch, &server->vu_fd_watches, next) {
427         aio_set_fd_handler(ctx, vu_fd_watch->fd, kick_handler, NULL,
428                            NULL, NULL, vu_fd_watch);
429     }
430 
431     if (server->co_trip) {
432         /*
433          * The caller didn't fully shut down co_trip (this can happen on
434          * non-polling drains like in bdrv_graph_wrlock()). This is okay as long
435          * as it no longer tries to shut it down and we're guaranteed to still
436          * be in the same AioContext as before.
437          *
438          * co_ctx can still be NULL if we get multiple calls and only just
439          * scheduled a new coroutine in the else branch.
440          */
441         AioContext *co_ctx = qemu_coroutine_get_aio_context(server->co_trip);
442 
443         assert(!server->quiescing);
444         assert(!co_ctx || co_ctx == ctx);
445     } else {
446         server->co_trip = qemu_coroutine_create(vu_client_trip, server);
447         assert(!server->in_qio_channel_yield);
448         aio_co_schedule(ctx, server->co_trip);
449     }
450 }
451 
452 /* Called with server->ctx acquired */
453 void vhost_user_server_detach_aio_context(VuServer *server)
454 {
455     if (server->sioc) {
456         VuFdWatch *vu_fd_watch;
457 
458         QTAILQ_FOREACH(vu_fd_watch, &server->vu_fd_watches, next) {
459             aio_set_fd_handler(server->ctx, vu_fd_watch->fd,
460                                NULL, NULL, NULL, NULL, vu_fd_watch);
461         }
462     }
463 
464     server->ctx = NULL;
465 
466     if (server->ioc) {
467         if (server->in_qio_channel_yield) {
468             /* Stop receiving the next vhost-user message */
469             qio_channel_wake_read(server->ioc);
470         }
471     }
472 }
473 
474 bool vhost_user_server_start(VuServer *server,
475                              SocketAddress *socket_addr,
476                              AioContext *ctx,
477                              uint16_t max_queues,
478                              const VuDevIface *vu_iface,
479                              Error **errp)
480 {
481     QEMUBH *bh;
482     QIONetListener *listener;
483 
484     if (socket_addr->type != SOCKET_ADDRESS_TYPE_UNIX &&
485         socket_addr->type != SOCKET_ADDRESS_TYPE_FD) {
486         error_setg(errp, "Only socket address types 'unix' and 'fd' are supported");
487         return false;
488     }
489 
490     listener = qio_net_listener_new();
491     if (qio_net_listener_open_sync(listener, socket_addr, 1,
492                                    errp) < 0) {
493         object_unref(OBJECT(listener));
494         return false;
495     }
496 
497     bh = qemu_bh_new(restart_listener_bh, server);
498 
499     /* zero out unspecified fields */
500     *server = (VuServer) {
501         .listener              = listener,
502         .restart_listener_bh   = bh,
503         .vu_iface              = vu_iface,
504         .max_queues            = max_queues,
505         .ctx                   = ctx,
506     };
507 
508     qio_net_listener_set_name(server->listener, "vhost-user-backend-listener");
509 
510     qio_net_listener_set_client_func(server->listener,
511                                      vu_accept,
512                                      server,
513                                      NULL);
514 
515     QTAILQ_INIT(&server->vu_fd_watches);
516     return true;
517 }
518