1 /*
2 * QEMU Block driver for NBD
3 *
4 * Copyright (c) 2021 Virtuozzo International GmbH.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "trace.h"
27
28 #include "block/nbd.h"
29
30 #include "qapi/qapi-visit-sockets.h"
31 #include "qapi/clone-visitor.h"
32 #include "qemu/coroutine.h"
33
34 struct NBDClientConnection {
35 /* Initialization constants, never change */
36 SocketAddress *saddr; /* address to connect to */
37 QCryptoTLSCreds *tlscreds;
38 char *tlshostname;
39 NBDExportInfo initial_info;
40 bool do_negotiation;
41 bool do_retry;
42
43 QemuMutex mutex;
44
45 NBDExportInfo updated_info;
46 /*
47 * @sioc represents a successful result. While thread is running, @sioc is
48 * used only by thread and not protected by mutex. When thread is not
49 * running, @sioc is stolen by nbd_co_establish_connection() under mutex.
50 */
51 QIOChannelSocket *sioc;
52 QIOChannel *ioc;
53 /*
54 * @err represents previous attempt. It may be copied by
55 * nbd_co_establish_connection() when it reports failure.
56 */
57 Error *err;
58
59 /* All further fields are accessed only under mutex */
60 bool running; /* thread is running now */
61 bool detached; /* thread is detached and should cleanup the state */
62
63 /*
64 * wait_co: if non-NULL, which coroutine to wake in
65 * nbd_co_establish_connection() after yield()
66 */
67 Coroutine *wait_co;
68 };
69
70 /*
71 * The function isn't protected by any mutex, only call it when the client
72 * connection attempt has not yet started.
73 */
nbd_client_connection_enable_retry(NBDClientConnection * conn)74 void nbd_client_connection_enable_retry(NBDClientConnection *conn)
75 {
76 conn->do_retry = true;
77 }
78
nbd_client_connection_new(const SocketAddress * saddr,bool do_negotiation,const char * export_name,const char * x_dirty_bitmap,QCryptoTLSCreds * tlscreds,const char * tlshostname)79 NBDClientConnection *nbd_client_connection_new(const SocketAddress *saddr,
80 bool do_negotiation,
81 const char *export_name,
82 const char *x_dirty_bitmap,
83 QCryptoTLSCreds *tlscreds,
84 const char *tlshostname)
85 {
86 NBDClientConnection *conn = g_new(NBDClientConnection, 1);
87
88 object_ref(OBJECT(tlscreds));
89 *conn = (NBDClientConnection) {
90 .saddr = QAPI_CLONE(SocketAddress, saddr),
91 .tlscreds = tlscreds,
92 .tlshostname = g_strdup(tlshostname),
93 .do_negotiation = do_negotiation,
94
95 .initial_info.request_sizes = true,
96 .initial_info.mode = NBD_MODE_EXTENDED,
97 .initial_info.base_allocation = true,
98 .initial_info.x_dirty_bitmap = g_strdup(x_dirty_bitmap),
99 .initial_info.name = g_strdup(export_name ?: "")
100 };
101
102 qemu_mutex_init(&conn->mutex);
103
104 return conn;
105 }
106
nbd_client_connection_do_free(NBDClientConnection * conn)107 static void nbd_client_connection_do_free(NBDClientConnection *conn)
108 {
109 if (conn->sioc) {
110 qio_channel_close(QIO_CHANNEL(conn->sioc), NULL);
111 object_unref(OBJECT(conn->sioc));
112 }
113 error_free(conn->err);
114 qapi_free_SocketAddress(conn->saddr);
115 g_free(conn->tlshostname);
116 object_unref(OBJECT(conn->tlscreds));
117 g_free(conn->initial_info.x_dirty_bitmap);
118 g_free(conn->initial_info.name);
119 g_free(conn);
120 }
121
122 /*
123 * Connect to @addr and do NBD negotiation if @info is not null. If @tlscreds
124 * are given @outioc is returned. @outioc is provided only on success. The call
125 * may be cancelled from other thread by simply qio_channel_shutdown(sioc).
126 */
nbd_connect(QIOChannelSocket * sioc,SocketAddress * addr,NBDExportInfo * info,QCryptoTLSCreds * tlscreds,const char * tlshostname,QIOChannel ** outioc,Error ** errp)127 static int nbd_connect(QIOChannelSocket *sioc, SocketAddress *addr,
128 NBDExportInfo *info, QCryptoTLSCreds *tlscreds,
129 const char *tlshostname,
130 QIOChannel **outioc, Error **errp)
131 {
132 int ret;
133
134 if (outioc) {
135 *outioc = NULL;
136 }
137
138 ret = qio_channel_socket_connect_sync(sioc, addr, errp);
139 if (ret < 0) {
140 return ret;
141 }
142
143 qio_channel_set_delay(QIO_CHANNEL(sioc), false);
144
145 if (!info) {
146 return 0;
147 }
148
149 ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), tlscreds, tlshostname,
150 outioc, info, errp);
151 if (ret < 0) {
152 /*
153 * nbd_receive_negotiate() may setup tls ioc and return it even on
154 * failure path. In this case we should use it instead of original
155 * channel.
156 */
157 if (outioc && *outioc) {
158 qio_channel_close(*outioc, NULL);
159 object_unref(OBJECT(*outioc));
160 *outioc = NULL;
161 } else {
162 qio_channel_close(QIO_CHANNEL(sioc), NULL);
163 }
164
165 return ret;
166 }
167
168 return 0;
169 }
170
connect_thread_func(void * opaque)171 static void *connect_thread_func(void *opaque)
172 {
173 NBDClientConnection *conn = opaque;
174 int ret;
175 bool do_free;
176 uint64_t timeout = 1;
177 uint64_t max_timeout = 16;
178
179 qemu_mutex_lock(&conn->mutex);
180 while (!conn->detached) {
181 Error *local_err = NULL;
182
183 assert(!conn->sioc);
184 conn->sioc = qio_channel_socket_new();
185
186 qemu_mutex_unlock(&conn->mutex);
187
188 conn->updated_info = conn->initial_info;
189
190 ret = nbd_connect(conn->sioc, conn->saddr,
191 conn->do_negotiation ? &conn->updated_info : NULL,
192 conn->tlscreds, conn->tlshostname,
193 &conn->ioc, &local_err);
194
195 /*
196 * conn->updated_info will finally be returned to the user. Clear the
197 * pointers to our internally allocated strings, which are IN parameters
198 * of nbd_receive_negotiate() and therefore nbd_connect(). Caller
199 * shouldn't be interested in these fields.
200 */
201 conn->updated_info.x_dirty_bitmap = NULL;
202 conn->updated_info.name = NULL;
203
204 qemu_mutex_lock(&conn->mutex);
205
206 error_free(conn->err);
207 conn->err = NULL;
208 error_propagate(&conn->err, local_err);
209
210 if (ret < 0) {
211 object_unref(OBJECT(conn->sioc));
212 conn->sioc = NULL;
213 if (conn->do_retry && !conn->detached) {
214 trace_nbd_connect_thread_sleep(timeout);
215 qemu_mutex_unlock(&conn->mutex);
216
217 sleep(timeout);
218 if (timeout < max_timeout) {
219 timeout *= 2;
220 }
221
222 qemu_mutex_lock(&conn->mutex);
223 continue;
224 }
225 }
226
227 break;
228 }
229
230 /* mutex is locked */
231
232 assert(conn->running);
233 conn->running = false;
234 if (conn->wait_co) {
235 aio_co_wake(conn->wait_co);
236 conn->wait_co = NULL;
237 }
238 do_free = conn->detached;
239
240 qemu_mutex_unlock(&conn->mutex);
241
242 if (do_free) {
243 nbd_client_connection_do_free(conn);
244 }
245
246 return NULL;
247 }
248
nbd_client_connection_release(NBDClientConnection * conn)249 void nbd_client_connection_release(NBDClientConnection *conn)
250 {
251 bool do_free = false;
252
253 if (!conn) {
254 return;
255 }
256
257 WITH_QEMU_LOCK_GUARD(&conn->mutex) {
258 assert(!conn->detached);
259 if (conn->running) {
260 conn->detached = true;
261 } else {
262 do_free = true;
263 }
264 if (conn->sioc) {
265 qio_channel_shutdown(QIO_CHANNEL(conn->sioc),
266 QIO_CHANNEL_SHUTDOWN_BOTH, NULL);
267 }
268 }
269
270 if (do_free) {
271 nbd_client_connection_do_free(conn);
272 }
273 }
274
275 /*
276 * Get a new connection in context of @conn:
277 * if the thread is running, wait for completion
278 * if the thread already succeeded in the background, and user didn't get the
279 * result, just return it now
280 * otherwise the thread is not running, so start a thread and wait for
281 * completion
282 *
283 * If @blocking is false, don't wait for the thread, return immediately.
284 *
285 * If @info is not NULL, also do nbd-negotiation after successful connection.
286 * In this case info is used only as out parameter, and is fully initialized by
287 * nbd_co_establish_connection(). "IN" fields of info as well as related only to
288 * nbd_receive_export_list() would be zero (see description of NBDExportInfo in
289 * include/block/nbd.h).
290 */
291 QIOChannel *coroutine_fn
nbd_co_establish_connection(NBDClientConnection * conn,NBDExportInfo * info,bool blocking,Error ** errp)292 nbd_co_establish_connection(NBDClientConnection *conn, NBDExportInfo *info,
293 bool blocking, Error **errp)
294 {
295 QemuThread thread;
296
297 if (conn->do_negotiation) {
298 assert(info);
299 }
300
301 WITH_QEMU_LOCK_GUARD(&conn->mutex) {
302 /*
303 * Don't call nbd_co_establish_connection() in several coroutines in
304 * parallel. Only one call at once is supported.
305 */
306 assert(!conn->wait_co);
307
308 if (!conn->running) {
309 if (conn->sioc) {
310 /* Previous attempt finally succeeded in background */
311 if (conn->do_negotiation) {
312 memcpy(info, &conn->updated_info, sizeof(*info));
313 if (conn->ioc) {
314 /* TLS channel now has own reference to parent */
315 object_unref(OBJECT(conn->sioc));
316 conn->sioc = NULL;
317
318 return g_steal_pointer(&conn->ioc);
319 }
320 }
321
322 assert(!conn->ioc);
323
324 return QIO_CHANNEL(g_steal_pointer(&conn->sioc));
325 }
326
327 conn->running = true;
328 qemu_thread_create(&thread, "nbd-connect",
329 connect_thread_func, conn, QEMU_THREAD_DETACHED);
330 }
331
332 if (!blocking) {
333 if (conn->err) {
334 error_propagate(errp, error_copy(conn->err));
335 } else {
336 error_setg(errp, "No connection at the moment");
337 }
338
339 return NULL;
340 }
341
342 conn->wait_co = qemu_coroutine_self();
343 }
344
345 /*
346 * We are going to wait for connect-thread finish, but
347 * nbd_co_establish_connection_cancel() can interrupt.
348 */
349 qemu_coroutine_yield();
350
351 WITH_QEMU_LOCK_GUARD(&conn->mutex) {
352 if (conn->running) {
353 /*
354 * The connection attempt was canceled and the coroutine resumed
355 * before the connection thread finished its job. Report the
356 * attempt as failed, but leave the connection thread running,
357 * to reuse it for the next connection attempt.
358 */
359 if (conn->err) {
360 error_propagate(errp, error_copy(conn->err));
361 } else {
362 /*
363 * The only possible case here is cancelling by open_timer
364 * during nbd_open(). So, the error message is for that case.
365 * If we have more use cases, we can refactor
366 * nbd_co_establish_connection_cancel() to take an additional
367 * parameter cancel_reason, that would be passed than to the
368 * caller of cancelled nbd_co_establish_connection().
369 */
370 error_setg(errp, "Connection attempt cancelled by timeout");
371 }
372
373 return NULL;
374 } else {
375 /* Thread finished. There must be either error or sioc */
376 assert(!conn->err != !conn->sioc);
377
378 if (conn->err) {
379 error_propagate(errp, error_copy(conn->err));
380 return NULL;
381 }
382
383 if (conn->do_negotiation) {
384 memcpy(info, &conn->updated_info, sizeof(*info));
385 if (conn->ioc) {
386 /* TLS channel now has own reference to parent */
387 object_unref(OBJECT(conn->sioc));
388 conn->sioc = NULL;
389
390 return g_steal_pointer(&conn->ioc);
391 }
392 }
393
394 assert(!conn->ioc);
395
396 return QIO_CHANNEL(g_steal_pointer(&conn->sioc));
397 }
398 }
399
400 abort(); /* unreachable */
401 }
402
403 /*
404 * nbd_co_establish_connection_cancel
405 * Cancel nbd_co_establish_connection() asynchronously.
406 *
407 * Note that this function neither directly stops the thread nor closes the
408 * socket, but rather safely wakes nbd_co_establish_connection() which is
409 * sleeping in yield()
410 */
nbd_co_establish_connection_cancel(NBDClientConnection * conn)411 void nbd_co_establish_connection_cancel(NBDClientConnection *conn)
412 {
413 Coroutine *wait_co;
414
415 WITH_QEMU_LOCK_GUARD(&conn->mutex) {
416 wait_co = g_steal_pointer(&conn->wait_co);
417 }
418
419 if (wait_co) {
420 aio_co_wake(wait_co);
421 }
422 }
423