]>
Commit | Line | Data |
---|---|---|
75818250 TS |
1 | /* |
2 | * QEMU Block driver for NBD | |
3 | * | |
f7651539 | 4 | * Copyright (c) 2019 Virtuozzo International GmbH. |
86f8cdf3 | 5 | * Copyright (C) 2016 Red Hat, Inc. |
75818250 | 6 | * Copyright (C) 2008 Bull S.A.S. |
bd5921b4 | 7 | * Author: Laurent Vivier <Laurent.Vivier@bull.net> |
75818250 TS |
8 | * |
9 | * Some parts: | |
10 | * Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws> | |
11 | * | |
12 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
13 | * of this software and associated documentation files (the "Software"), to deal | |
14 | * in the Software without restriction, including without limitation the rights | |
15 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
16 | * copies of the Software, and to permit persons to whom the Software is | |
17 | * furnished to do so, subject to the following conditions: | |
18 | * | |
19 | * The above copyright notice and this permission notice shall be included in | |
20 | * all copies or substantial portions of the Software. | |
21 | * | |
22 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
23 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
24 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
25 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
26 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
27 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
28 | * THE SOFTWARE. | |
29 | */ | |
30 | ||
80c71a24 | 31 | #include "qemu/osdep.h" |
86f8cdf3 VSO |
32 | |
33 | #include "trace.h" | |
1de7afc9 | 34 | #include "qemu/uri.h" |
922a01a0 | 35 | #include "qemu/option.h" |
86f8cdf3 | 36 | #include "qemu/cutils.h" |
db725815 | 37 | #include "qemu/main-loop.h" |
fee091cd | 38 | #include "qemu/atomic.h" |
86f8cdf3 | 39 | |
9af23989 | 40 | #include "qapi/qapi-visit-sockets.h" |
2019d68b | 41 | #include "qapi/qmp/qstring.h" |
1dc4718d | 42 | #include "qapi/clone-visitor.h" |
86f8cdf3 VSO |
43 | |
44 | #include "block/qdict.h" | |
45 | #include "block/nbd.h" | |
46 | #include "block/block_int.h" | |
75818250 | 47 | |
fee091cd LS |
48 | #include "qemu/yank.h" |
49 | ||
1d45f8b5 | 50 | #define EN_OPTSTR ":exportname=" |
86f8cdf3 VSO |
51 | #define MAX_NBD_REQUESTS 16 |
52 | ||
53 | #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs)) | |
54 | #define INDEX_TO_HANDLE(bs, index) ((index) ^ (uint64_t)(intptr_t)(bs)) | |
55 | ||
56 | typedef struct { | |
57 | Coroutine *coroutine; | |
58 | uint64_t offset; /* original offset of the request */ | |
59 | bool receiving; /* waiting for connection_co? */ | |
60 | } NBDClientRequest; | |
61 | ||
a34b1e5e | 62 | typedef enum NBDClientState { |
f7651539 VSO |
63 | NBD_CLIENT_CONNECTING_WAIT, |
64 | NBD_CLIENT_CONNECTING_NOWAIT, | |
a34b1e5e VSO |
65 | NBD_CLIENT_CONNECTED, |
66 | NBD_CLIENT_QUIT | |
67 | } NBDClientState; | |
68 | ||
1dc4718d VSO |
69 | typedef enum NBDConnectThreadState { |
70 | /* No thread, no pending results */ | |
71 | CONNECT_THREAD_NONE, | |
72 | ||
73 | /* Thread is running, no results for now */ | |
74 | CONNECT_THREAD_RUNNING, | |
75 | ||
76 | /* | |
77 | * Thread is running, but requestor exited. Thread should close | |
78 | * the new socket and free the connect state on exit. | |
79 | */ | |
80 | CONNECT_THREAD_RUNNING_DETACHED, | |
81 | ||
82 | /* Thread finished, results are stored in a state */ | |
83 | CONNECT_THREAD_FAIL, | |
84 | CONNECT_THREAD_SUCCESS | |
85 | } NBDConnectThreadState; | |
86 | ||
87 | typedef struct NBDConnectThread { | |
88 | /* Initialization constants */ | |
89 | SocketAddress *saddr; /* address to connect to */ | |
90 | /* | |
91 | * Bottom half to schedule on completion. Scheduled only if bh_ctx is not | |
92 | * NULL | |
93 | */ | |
94 | QEMUBHFunc *bh_func; | |
95 | void *bh_opaque; | |
96 | ||
97 | /* | |
98 | * Result of last attempt. Valid in FAIL and SUCCESS states. | |
99 | * If you want to steal error, don't forget to set pointer to NULL. | |
100 | */ | |
101 | QIOChannelSocket *sioc; | |
102 | Error *err; | |
103 | ||
104 | /* state and bh_ctx are protected by mutex */ | |
105 | QemuMutex mutex; | |
106 | NBDConnectThreadState state; /* current state of the thread */ | |
107 | AioContext *bh_ctx; /* where to schedule bh (NULL means don't schedule) */ | |
108 | } NBDConnectThread; | |
109 | ||
611ae1d7 | 110 | typedef struct BDRVNBDState { |
86f8cdf3 VSO |
111 | QIOChannelSocket *sioc; /* The master data channel */ |
112 | QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */ | |
113 | NBDExportInfo info; | |
114 | ||
115 | CoMutex send_mutex; | |
116 | CoQueue free_sema; | |
117 | Coroutine *connection_co; | |
78c81a3f | 118 | Coroutine *teardown_co; |
f7651539 VSO |
119 | QemuCoSleepState *connection_co_sleep_ns_state; |
120 | bool drained; | |
121 | bool wait_drained_end; | |
86f8cdf3 | 122 | int in_flight; |
a34b1e5e | 123 | NBDClientState state; |
f7651539 VSO |
124 | int connect_status; |
125 | Error *connect_err; | |
126 | bool wait_in_flight; | |
86f8cdf3 | 127 | |
46f56631 VSO |
128 | QEMUTimer *reconnect_delay_timer; |
129 | ||
86f8cdf3 VSO |
130 | NBDClientRequest requests[MAX_NBD_REQUESTS]; |
131 | NBDReply reply; | |
132 | BlockDriverState *bs; | |
03504d05 | 133 | |
8f071c9d VSO |
134 | /* Connection parameters */ |
135 | uint32_t reconnect_delay; | |
62cf396b | 136 | SocketAddress *saddr; |
491d6c7c | 137 | char *export, *tlscredsid; |
8f071c9d VSO |
138 | QCryptoTLSCreds *tlscreds; |
139 | const char *hostname; | |
140 | char *x_dirty_bitmap; | |
dbc7b014 | 141 | bool alloc_depth; |
1dc4718d VSO |
142 | |
143 | bool wait_connect; | |
144 | NBDConnectThread *connect_thread; | |
75818250 TS |
145 | } BDRVNBDState; |
146 | ||
fee091cd LS |
147 | static int nbd_establish_connection(BlockDriverState *bs, SocketAddress *saddr, |
148 | Error **errp); | |
149 | static int nbd_co_establish_connection(BlockDriverState *bs, Error **errp); | |
1dc4718d VSO |
150 | static void nbd_co_establish_connection_cancel(BlockDriverState *bs, |
151 | bool detach); | |
fee091cd LS |
152 | static int nbd_client_handshake(BlockDriverState *bs, Error **errp); |
153 | static void nbd_yank(void *opaque); | |
f7651539 | 154 | |
7f493662 PN |
155 | static void nbd_clear_bdrvstate(BDRVNBDState *s) |
156 | { | |
157 | object_unref(OBJECT(s->tlscreds)); | |
158 | qapi_free_SocketAddress(s->saddr); | |
159 | s->saddr = NULL; | |
160 | g_free(s->export); | |
161 | s->export = NULL; | |
162 | g_free(s->tlscredsid); | |
163 | s->tlscredsid = NULL; | |
164 | g_free(s->x_dirty_bitmap); | |
165 | s->x_dirty_bitmap = NULL; | |
166 | } | |
167 | ||
a34b1e5e VSO |
168 | static void nbd_channel_error(BDRVNBDState *s, int ret) |
169 | { | |
f7651539 | 170 | if (ret == -EIO) { |
fee091cd | 171 | if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTED) { |
f7651539 VSO |
172 | s->state = s->reconnect_delay ? NBD_CLIENT_CONNECTING_WAIT : |
173 | NBD_CLIENT_CONNECTING_NOWAIT; | |
174 | } | |
175 | } else { | |
fee091cd | 176 | if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTED) { |
f7651539 VSO |
177 | qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); |
178 | } | |
179 | s->state = NBD_CLIENT_QUIT; | |
180 | } | |
a34b1e5e VSO |
181 | } |
182 | ||
611ae1d7 | 183 | static void nbd_recv_coroutines_wake_all(BDRVNBDState *s) |
86f8cdf3 VSO |
184 | { |
185 | int i; | |
186 | ||
187 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { | |
188 | NBDClientRequest *req = &s->requests[i]; | |
189 | ||
190 | if (req->coroutine && req->receiving) { | |
191 | aio_co_wake(req->coroutine); | |
192 | } | |
193 | } | |
194 | } | |
195 | ||
46f56631 VSO |
196 | static void reconnect_delay_timer_del(BDRVNBDState *s) |
197 | { | |
198 | if (s->reconnect_delay_timer) { | |
46f56631 VSO |
199 | timer_free(s->reconnect_delay_timer); |
200 | s->reconnect_delay_timer = NULL; | |
201 | } | |
202 | } | |
203 | ||
204 | static void reconnect_delay_timer_cb(void *opaque) | |
205 | { | |
206 | BDRVNBDState *s = opaque; | |
207 | ||
fee091cd | 208 | if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTING_WAIT) { |
46f56631 VSO |
209 | s->state = NBD_CLIENT_CONNECTING_NOWAIT; |
210 | while (qemu_co_enter_next(&s->free_sema, NULL)) { | |
211 | /* Resume all queued requests */ | |
212 | } | |
213 | } | |
214 | ||
215 | reconnect_delay_timer_del(s); | |
216 | } | |
217 | ||
218 | static void reconnect_delay_timer_init(BDRVNBDState *s, uint64_t expire_time_ns) | |
219 | { | |
fee091cd | 220 | if (qatomic_load_acquire(&s->state) != NBD_CLIENT_CONNECTING_WAIT) { |
46f56631 VSO |
221 | return; |
222 | } | |
223 | ||
224 | assert(!s->reconnect_delay_timer); | |
225 | s->reconnect_delay_timer = aio_timer_new(bdrv_get_aio_context(s->bs), | |
226 | QEMU_CLOCK_REALTIME, | |
227 | SCALE_NS, | |
228 | reconnect_delay_timer_cb, s); | |
229 | timer_mod(s->reconnect_delay_timer, expire_time_ns); | |
230 | } | |
231 | ||
86f8cdf3 VSO |
232 | static void nbd_client_detach_aio_context(BlockDriverState *bs) |
233 | { | |
611ae1d7 VSO |
234 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
235 | ||
46f56631 VSO |
236 | /* Timer is deleted in nbd_client_co_drain_begin() */ |
237 | assert(!s->reconnect_delay_timer); | |
3b5e4db6 RK |
238 | /* |
239 | * If reconnect is in progress we may have no ->ioc. It will be | |
240 | * re-instantiated in the proper aio context once the connection is | |
241 | * reestablished. | |
242 | */ | |
243 | if (s->ioc) { | |
244 | qio_channel_detach_aio_context(QIO_CHANNEL(s->ioc)); | |
245 | } | |
86f8cdf3 VSO |
246 | } |
247 | ||
248 | static void nbd_client_attach_aio_context_bh(void *opaque) | |
249 | { | |
250 | BlockDriverState *bs = opaque; | |
611ae1d7 | 251 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 | 252 | |
ddde5ee7 RK |
253 | if (s->connection_co) { |
254 | /* | |
255 | * The node is still drained, so we know the coroutine has yielded in | |
256 | * nbd_read_eof(), the only place where bs->in_flight can reach 0, or | |
257 | * it is entered for the first time. Both places are safe for entering | |
258 | * the coroutine. | |
259 | */ | |
260 | qemu_aio_coroutine_enter(bs->aio_context, s->connection_co); | |
261 | } | |
86f8cdf3 VSO |
262 | bdrv_dec_in_flight(bs); |
263 | } | |
264 | ||
265 | static void nbd_client_attach_aio_context(BlockDriverState *bs, | |
266 | AioContext *new_context) | |
267 | { | |
611ae1d7 VSO |
268 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
269 | ||
f7651539 VSO |
270 | /* |
271 | * s->connection_co is either yielded from nbd_receive_reply or from | |
272 | * nbd_co_reconnect_loop() | |
273 | */ | |
fee091cd | 274 | if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTED) { |
f7651539 VSO |
275 | qio_channel_attach_aio_context(QIO_CHANNEL(s->ioc), new_context); |
276 | } | |
86f8cdf3 VSO |
277 | |
278 | bdrv_inc_in_flight(bs); | |
279 | ||
280 | /* | |
281 | * Need to wait here for the BH to run because the BH must run while the | |
282 | * node is still drained. | |
283 | */ | |
284 | aio_wait_bh_oneshot(new_context, nbd_client_attach_aio_context_bh, bs); | |
285 | } | |
286 | ||
f7651539 VSO |
287 | static void coroutine_fn nbd_client_co_drain_begin(BlockDriverState *bs) |
288 | { | |
289 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
86f8cdf3 | 290 | |
f7651539 VSO |
291 | s->drained = true; |
292 | if (s->connection_co_sleep_ns_state) { | |
293 | qemu_co_sleep_wake(s->connection_co_sleep_ns_state); | |
294 | } | |
1dc4718d VSO |
295 | |
296 | nbd_co_establish_connection_cancel(bs, false); | |
8c517de2 | 297 | |
46f56631 VSO |
298 | reconnect_delay_timer_del(s); |
299 | ||
fee091cd | 300 | if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTING_WAIT) { |
8c517de2 VSO |
301 | s->state = NBD_CLIENT_CONNECTING_NOWAIT; |
302 | qemu_co_queue_restart_all(&s->free_sema); | |
303 | } | |
f7651539 VSO |
304 | } |
305 | ||
306 | static void coroutine_fn nbd_client_co_drain_end(BlockDriverState *bs) | |
86f8cdf3 | 307 | { |
611ae1d7 | 308 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 | 309 | |
f7651539 VSO |
310 | s->drained = false; |
311 | if (s->wait_drained_end) { | |
312 | s->wait_drained_end = false; | |
313 | aio_co_wake(s->connection_co); | |
314 | } | |
315 | } | |
316 | ||
86f8cdf3 | 317 | |
f7651539 VSO |
318 | static void nbd_teardown_connection(BlockDriverState *bs) |
319 | { | |
320 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
321 | ||
fbeb3e63 | 322 | if (s->ioc) { |
f7651539 | 323 | /* finish any pending coroutines */ |
f7651539 | 324 | qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); |
fbeb3e63 VSO |
325 | } else if (s->sioc) { |
326 | /* abort negotiation */ | |
327 | qio_channel_shutdown(QIO_CHANNEL(s->sioc), QIO_CHANNEL_SHUTDOWN_BOTH, | |
328 | NULL); | |
f7651539 | 329 | } |
fbeb3e63 | 330 | |
f7651539 VSO |
331 | s->state = NBD_CLIENT_QUIT; |
332 | if (s->connection_co) { | |
333 | if (s->connection_co_sleep_ns_state) { | |
334 | qemu_co_sleep_wake(s->connection_co_sleep_ns_state); | |
335 | } | |
1dc4718d | 336 | nbd_co_establish_connection_cancel(bs, true); |
f7651539 | 337 | } |
78c81a3f HR |
338 | if (qemu_in_coroutine()) { |
339 | s->teardown_co = qemu_coroutine_self(); | |
340 | /* connection_co resumes us when it terminates */ | |
341 | qemu_coroutine_yield(); | |
342 | s->teardown_co = NULL; | |
343 | } else { | |
344 | BDRV_POLL_WHILE(bs, s->connection_co); | |
345 | } | |
346 | assert(!s->connection_co); | |
f7651539 | 347 | } |
86f8cdf3 | 348 | |
f7651539 VSO |
349 | static bool nbd_client_connecting(BDRVNBDState *s) |
350 | { | |
fee091cd LS |
351 | NBDClientState state = qatomic_load_acquire(&s->state); |
352 | return state == NBD_CLIENT_CONNECTING_WAIT || | |
353 | state == NBD_CLIENT_CONNECTING_NOWAIT; | |
f7651539 VSO |
354 | } |
355 | ||
356 | static bool nbd_client_connecting_wait(BDRVNBDState *s) | |
357 | { | |
fee091cd | 358 | return qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTING_WAIT; |
f7651539 VSO |
359 | } |
360 | ||
1dc4718d VSO |
361 | static void connect_bh(void *opaque) |
362 | { | |
363 | BDRVNBDState *state = opaque; | |
364 | ||
365 | assert(state->wait_connect); | |
366 | state->wait_connect = false; | |
367 | aio_co_wake(state->connection_co); | |
368 | } | |
369 | ||
370 | static void nbd_init_connect_thread(BDRVNBDState *s) | |
371 | { | |
372 | s->connect_thread = g_new(NBDConnectThread, 1); | |
373 | ||
374 | *s->connect_thread = (NBDConnectThread) { | |
375 | .saddr = QAPI_CLONE(SocketAddress, s->saddr), | |
376 | .state = CONNECT_THREAD_NONE, | |
377 | .bh_func = connect_bh, | |
378 | .bh_opaque = s, | |
379 | }; | |
380 | ||
381 | qemu_mutex_init(&s->connect_thread->mutex); | |
382 | } | |
383 | ||
384 | static void nbd_free_connect_thread(NBDConnectThread *thr) | |
385 | { | |
386 | if (thr->sioc) { | |
387 | qio_channel_close(QIO_CHANNEL(thr->sioc), NULL); | |
388 | } | |
389 | error_free(thr->err); | |
390 | qapi_free_SocketAddress(thr->saddr); | |
391 | g_free(thr); | |
392 | } | |
393 | ||
394 | static void *connect_thread_func(void *opaque) | |
395 | { | |
396 | NBDConnectThread *thr = opaque; | |
397 | int ret; | |
398 | bool do_free = false; | |
399 | ||
400 | thr->sioc = qio_channel_socket_new(); | |
401 | ||
402 | error_free(thr->err); | |
403 | thr->err = NULL; | |
404 | ret = qio_channel_socket_connect_sync(thr->sioc, thr->saddr, &thr->err); | |
405 | if (ret < 0) { | |
406 | object_unref(OBJECT(thr->sioc)); | |
407 | thr->sioc = NULL; | |
408 | } | |
409 | ||
410 | qemu_mutex_lock(&thr->mutex); | |
411 | ||
412 | switch (thr->state) { | |
413 | case CONNECT_THREAD_RUNNING: | |
414 | thr->state = ret < 0 ? CONNECT_THREAD_FAIL : CONNECT_THREAD_SUCCESS; | |
415 | if (thr->bh_ctx) { | |
416 | aio_bh_schedule_oneshot(thr->bh_ctx, thr->bh_func, thr->bh_opaque); | |
417 | ||
418 | /* play safe, don't reuse bh_ctx on further connection attempts */ | |
419 | thr->bh_ctx = NULL; | |
420 | } | |
421 | break; | |
422 | case CONNECT_THREAD_RUNNING_DETACHED: | |
423 | do_free = true; | |
424 | break; | |
425 | default: | |
426 | abort(); | |
427 | } | |
428 | ||
429 | qemu_mutex_unlock(&thr->mutex); | |
430 | ||
431 | if (do_free) { | |
432 | nbd_free_connect_thread(thr); | |
433 | } | |
434 | ||
435 | return NULL; | |
436 | } | |
437 | ||
fee091cd | 438 | static int coroutine_fn |
1dc4718d VSO |
439 | nbd_co_establish_connection(BlockDriverState *bs, Error **errp) |
440 | { | |
fee091cd | 441 | int ret; |
1dc4718d VSO |
442 | QemuThread thread; |
443 | BDRVNBDState *s = bs->opaque; | |
1dc4718d VSO |
444 | NBDConnectThread *thr = s->connect_thread; |
445 | ||
446 | qemu_mutex_lock(&thr->mutex); | |
447 | ||
448 | switch (thr->state) { | |
449 | case CONNECT_THREAD_FAIL: | |
450 | case CONNECT_THREAD_NONE: | |
451 | error_free(thr->err); | |
452 | thr->err = NULL; | |
453 | thr->state = CONNECT_THREAD_RUNNING; | |
454 | qemu_thread_create(&thread, "nbd-connect", | |
455 | connect_thread_func, thr, QEMU_THREAD_DETACHED); | |
456 | break; | |
457 | case CONNECT_THREAD_SUCCESS: | |
458 | /* Previous attempt finally succeeded in background */ | |
459 | thr->state = CONNECT_THREAD_NONE; | |
fee091cd | 460 | s->sioc = thr->sioc; |
1dc4718d | 461 | thr->sioc = NULL; |
fee091cd LS |
462 | yank_register_function(BLOCKDEV_YANK_INSTANCE(bs->node_name), |
463 | nbd_yank, bs); | |
1dc4718d | 464 | qemu_mutex_unlock(&thr->mutex); |
fee091cd | 465 | return 0; |
1dc4718d VSO |
466 | case CONNECT_THREAD_RUNNING: |
467 | /* Already running, will wait */ | |
468 | break; | |
469 | default: | |
470 | abort(); | |
471 | } | |
472 | ||
473 | thr->bh_ctx = qemu_get_current_aio_context(); | |
474 | ||
475 | qemu_mutex_unlock(&thr->mutex); | |
476 | ||
477 | ||
478 | /* | |
479 | * We are going to wait for connect-thread finish, but | |
480 | * nbd_client_co_drain_begin() can interrupt. | |
481 | * | |
482 | * Note that wait_connect variable is not visible for connect-thread. It | |
483 | * doesn't need mutex protection, it used only inside home aio context of | |
484 | * bs. | |
485 | */ | |
486 | s->wait_connect = true; | |
487 | qemu_coroutine_yield(); | |
488 | ||
489 | qemu_mutex_lock(&thr->mutex); | |
490 | ||
491 | switch (thr->state) { | |
492 | case CONNECT_THREAD_SUCCESS: | |
493 | case CONNECT_THREAD_FAIL: | |
494 | thr->state = CONNECT_THREAD_NONE; | |
495 | error_propagate(errp, thr->err); | |
496 | thr->err = NULL; | |
fee091cd | 497 | s->sioc = thr->sioc; |
1dc4718d | 498 | thr->sioc = NULL; |
fee091cd LS |
499 | if (s->sioc) { |
500 | yank_register_function(BLOCKDEV_YANK_INSTANCE(bs->node_name), | |
501 | nbd_yank, bs); | |
502 | } | |
503 | ret = (s->sioc ? 0 : -1); | |
1dc4718d VSO |
504 | break; |
505 | case CONNECT_THREAD_RUNNING: | |
506 | case CONNECT_THREAD_RUNNING_DETACHED: | |
507 | /* | |
508 | * Obviously, drained section wants to start. Report the attempt as | |
509 | * failed. Still connect thread is executing in background, and its | |
510 | * result may be used for next connection attempt. | |
511 | */ | |
fee091cd | 512 | ret = -1; |
1dc4718d VSO |
513 | error_setg(errp, "Connection attempt cancelled by other operation"); |
514 | break; | |
515 | ||
516 | case CONNECT_THREAD_NONE: | |
517 | /* | |
518 | * Impossible. We've seen this thread running. So it should be | |
519 | * running or at least give some results. | |
520 | */ | |
521 | abort(); | |
522 | ||
523 | default: | |
524 | abort(); | |
525 | } | |
526 | ||
527 | qemu_mutex_unlock(&thr->mutex); | |
528 | ||
fee091cd | 529 | return ret; |
1dc4718d VSO |
530 | } |
531 | ||
532 | /* | |
533 | * nbd_co_establish_connection_cancel | |
534 | * Cancel nbd_co_establish_connection asynchronously: it will finish soon, to | |
535 | * allow drained section to begin. | |
536 | * | |
537 | * If detach is true, also cleanup the state (or if thread is running, move it | |
538 | * to CONNECT_THREAD_RUNNING_DETACHED state). s->connect_thread becomes NULL if | |
539 | * detach is true. | |
540 | */ | |
541 | static void nbd_co_establish_connection_cancel(BlockDriverState *bs, | |
542 | bool detach) | |
543 | { | |
544 | BDRVNBDState *s = bs->opaque; | |
545 | NBDConnectThread *thr = s->connect_thread; | |
546 | bool wake = false; | |
547 | bool do_free = false; | |
548 | ||
549 | qemu_mutex_lock(&thr->mutex); | |
550 | ||
551 | if (thr->state == CONNECT_THREAD_RUNNING) { | |
552 | /* We can cancel only in running state, when bh is not yet scheduled */ | |
553 | thr->bh_ctx = NULL; | |
554 | if (s->wait_connect) { | |
555 | s->wait_connect = false; | |
556 | wake = true; | |
557 | } | |
558 | if (detach) { | |
559 | thr->state = CONNECT_THREAD_RUNNING_DETACHED; | |
560 | s->connect_thread = NULL; | |
561 | } | |
562 | } else if (detach) { | |
563 | do_free = true; | |
564 | } | |
565 | ||
566 | qemu_mutex_unlock(&thr->mutex); | |
567 | ||
568 | if (do_free) { | |
569 | nbd_free_connect_thread(thr); | |
570 | s->connect_thread = NULL; | |
571 | } | |
572 | ||
573 | if (wake) { | |
574 | aio_co_wake(s->connection_co); | |
575 | } | |
576 | } | |
577 | ||
f7651539 VSO |
578 | static coroutine_fn void nbd_reconnect_attempt(BDRVNBDState *s) |
579 | { | |
fa35591b | 580 | int ret; |
f7651539 VSO |
581 | Error *local_err = NULL; |
582 | ||
583 | if (!nbd_client_connecting(s)) { | |
584 | return; | |
585 | } | |
586 | ||
587 | /* Wait for completion of all in-flight requests */ | |
588 | ||
589 | qemu_co_mutex_lock(&s->send_mutex); | |
590 | ||
591 | while (s->in_flight > 0) { | |
592 | qemu_co_mutex_unlock(&s->send_mutex); | |
593 | nbd_recv_coroutines_wake_all(s); | |
594 | s->wait_in_flight = true; | |
595 | qemu_coroutine_yield(); | |
596 | s->wait_in_flight = false; | |
597 | qemu_co_mutex_lock(&s->send_mutex); | |
598 | } | |
599 | ||
600 | qemu_co_mutex_unlock(&s->send_mutex); | |
601 | ||
602 | if (!nbd_client_connecting(s)) { | |
603 | return; | |
604 | } | |
605 | ||
606 | /* | |
607 | * Now we are sure that nobody is accessing the channel, and no one will | |
608 | * try until we set the state to CONNECTED. | |
609 | */ | |
610 | ||
611 | /* Finalize previous connection if any */ | |
612 | if (s->ioc) { | |
8a509afd | 613 | qio_channel_detach_aio_context(QIO_CHANNEL(s->ioc)); |
fee091cd LS |
614 | yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), |
615 | nbd_yank, s->bs); | |
f7651539 VSO |
616 | object_unref(OBJECT(s->sioc)); |
617 | s->sioc = NULL; | |
618 | object_unref(OBJECT(s->ioc)); | |
619 | s->ioc = NULL; | |
620 | } | |
621 | ||
fee091cd | 622 | if (nbd_co_establish_connection(s->bs, &local_err) < 0) { |
fa35591b VSO |
623 | ret = -ECONNREFUSED; |
624 | goto out; | |
625 | } | |
626 | ||
dd1ec1a4 VSO |
627 | bdrv_dec_in_flight(s->bs); |
628 | ||
fee091cd | 629 | ret = nbd_client_handshake(s->bs, &local_err); |
fa35591b | 630 | |
dd1ec1a4 VSO |
631 | if (s->drained) { |
632 | s->wait_drained_end = true; | |
633 | while (s->drained) { | |
634 | /* | |
635 | * We may be entered once from nbd_client_attach_aio_context_bh | |
636 | * and then from nbd_client_co_drain_end. So here is a loop. | |
637 | */ | |
638 | qemu_coroutine_yield(); | |
639 | } | |
640 | } | |
641 | bdrv_inc_in_flight(s->bs); | |
642 | ||
fa35591b VSO |
643 | out: |
644 | s->connect_status = ret; | |
f7651539 VSO |
645 | error_free(s->connect_err); |
646 | s->connect_err = NULL; | |
647 | error_propagate(&s->connect_err, local_err); | |
648 | ||
fa35591b VSO |
649 | if (ret >= 0) { |
650 | /* successfully connected */ | |
651 | s->state = NBD_CLIENT_CONNECTED; | |
652 | qemu_co_queue_restart_all(&s->free_sema); | |
f7651539 | 653 | } |
f7651539 VSO |
654 | } |
655 | ||
656 | static coroutine_fn void nbd_co_reconnect_loop(BDRVNBDState *s) | |
657 | { | |
f7651539 VSO |
658 | uint64_t timeout = 1 * NANOSECONDS_PER_SECOND; |
659 | uint64_t max_timeout = 16 * NANOSECONDS_PER_SECOND; | |
660 | ||
fee091cd | 661 | if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTING_WAIT) { |
46f56631 VSO |
662 | reconnect_delay_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + |
663 | s->reconnect_delay * NANOSECONDS_PER_SECOND); | |
664 | } | |
665 | ||
f7651539 VSO |
666 | nbd_reconnect_attempt(s); |
667 | ||
668 | while (nbd_client_connecting(s)) { | |
f7651539 VSO |
669 | if (s->drained) { |
670 | bdrv_dec_in_flight(s->bs); | |
671 | s->wait_drained_end = true; | |
672 | while (s->drained) { | |
673 | /* | |
674 | * We may be entered once from nbd_client_attach_aio_context_bh | |
675 | * and then from nbd_client_co_drain_end. So here is a loop. | |
676 | */ | |
677 | qemu_coroutine_yield(); | |
678 | } | |
679 | bdrv_inc_in_flight(s->bs); | |
12c75e20 VSO |
680 | } else { |
681 | qemu_co_sleep_ns_wakeable(QEMU_CLOCK_REALTIME, timeout, | |
682 | &s->connection_co_sleep_ns_state); | |
99d72dba VSO |
683 | if (s->drained) { |
684 | continue; | |
685 | } | |
12c75e20 VSO |
686 | if (timeout < max_timeout) { |
687 | timeout *= 2; | |
688 | } | |
f7651539 VSO |
689 | } |
690 | ||
691 | nbd_reconnect_attempt(s); | |
692 | } | |
46f56631 VSO |
693 | |
694 | reconnect_delay_timer_del(s); | |
86f8cdf3 VSO |
695 | } |
696 | ||
697 | static coroutine_fn void nbd_connection_entry(void *opaque) | |
698 | { | |
611ae1d7 | 699 | BDRVNBDState *s = opaque; |
86f8cdf3 VSO |
700 | uint64_t i; |
701 | int ret = 0; | |
702 | Error *local_err = NULL; | |
703 | ||
fee091cd | 704 | while (qatomic_load_acquire(&s->state) != NBD_CLIENT_QUIT) { |
86f8cdf3 VSO |
705 | /* |
706 | * The NBD client can only really be considered idle when it has | |
707 | * yielded from qio_channel_readv_all_eof(), waiting for data. This is | |
708 | * the point where the additional scheduled coroutine entry happens | |
709 | * after nbd_client_attach_aio_context(). | |
710 | * | |
711 | * Therefore we keep an additional in_flight reference all the time and | |
712 | * only drop it temporarily here. | |
713 | */ | |
f7651539 VSO |
714 | |
715 | if (nbd_client_connecting(s)) { | |
716 | nbd_co_reconnect_loop(s); | |
717 | } | |
718 | ||
fee091cd | 719 | if (qatomic_load_acquire(&s->state) != NBD_CLIENT_CONNECTED) { |
f7651539 VSO |
720 | continue; |
721 | } | |
722 | ||
86f8cdf3 VSO |
723 | assert(s->reply.handle == 0); |
724 | ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, &local_err); | |
725 | ||
726 | if (local_err) { | |
727 | trace_nbd_read_reply_entry_fail(ret, error_get_pretty(local_err)); | |
728 | error_free(local_err); | |
f7651539 | 729 | local_err = NULL; |
86f8cdf3 VSO |
730 | } |
731 | if (ret <= 0) { | |
a34b1e5e | 732 | nbd_channel_error(s, ret ? ret : -EIO); |
f7651539 | 733 | continue; |
86f8cdf3 VSO |
734 | } |
735 | ||
736 | /* | |
737 | * There's no need for a mutex on the receive side, because the | |
738 | * handler acts as a synchronization point and ensures that only | |
739 | * one coroutine is called until the reply finishes. | |
740 | */ | |
741 | i = HANDLE_TO_INDEX(s, s->reply.handle); | |
742 | if (i >= MAX_NBD_REQUESTS || | |
743 | !s->requests[i].coroutine || | |
744 | !s->requests[i].receiving || | |
745 | (nbd_reply_is_structured(&s->reply) && !s->info.structured_reply)) | |
746 | { | |
a34b1e5e | 747 | nbd_channel_error(s, -EINVAL); |
f7651539 | 748 | continue; |
86f8cdf3 VSO |
749 | } |
750 | ||
751 | /* | |
752 | * We're woken up again by the request itself. Note that there | |
753 | * is no race between yielding and reentering connection_co. This | |
754 | * is because: | |
755 | * | |
756 | * - if the request runs on the same AioContext, it is only | |
757 | * entered after we yield | |
758 | * | |
759 | * - if the request runs on a different AioContext, reentering | |
760 | * connection_co happens through a bottom half, which can only | |
761 | * run after we yield. | |
762 | */ | |
763 | aio_co_wake(s->requests[i].coroutine); | |
764 | qemu_coroutine_yield(); | |
765 | } | |
766 | ||
f7651539 | 767 | qemu_co_queue_restart_all(&s->free_sema); |
86f8cdf3 VSO |
768 | nbd_recv_coroutines_wake_all(s); |
769 | bdrv_dec_in_flight(s->bs); | |
770 | ||
771 | s->connection_co = NULL; | |
f7651539 | 772 | if (s->ioc) { |
8a509afd | 773 | qio_channel_detach_aio_context(QIO_CHANNEL(s->ioc)); |
fee091cd LS |
774 | yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), |
775 | nbd_yank, s->bs); | |
f7651539 VSO |
776 | object_unref(OBJECT(s->sioc)); |
777 | s->sioc = NULL; | |
778 | object_unref(OBJECT(s->ioc)); | |
779 | s->ioc = NULL; | |
780 | } | |
781 | ||
78c81a3f HR |
782 | if (s->teardown_co) { |
783 | aio_co_wake(s->teardown_co); | |
784 | } | |
86f8cdf3 VSO |
785 | aio_wait_kick(); |
786 | } | |
787 | ||
788 | static int nbd_co_send_request(BlockDriverState *bs, | |
789 | NBDRequest *request, | |
790 | QEMUIOVector *qiov) | |
791 | { | |
611ae1d7 | 792 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
a34b1e5e | 793 | int rc, i = -1; |
86f8cdf3 VSO |
794 | |
795 | qemu_co_mutex_lock(&s->send_mutex); | |
f7651539 | 796 | while (s->in_flight == MAX_NBD_REQUESTS || nbd_client_connecting_wait(s)) { |
86f8cdf3 VSO |
797 | qemu_co_queue_wait(&s->free_sema, &s->send_mutex); |
798 | } | |
a34b1e5e | 799 | |
fee091cd | 800 | if (qatomic_load_acquire(&s->state) != NBD_CLIENT_CONNECTED) { |
a34b1e5e VSO |
801 | rc = -EIO; |
802 | goto err; | |
803 | } | |
804 | ||
86f8cdf3 VSO |
805 | s->in_flight++; |
806 | ||
807 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { | |
808 | if (s->requests[i].coroutine == NULL) { | |
809 | break; | |
810 | } | |
811 | } | |
812 | ||
813 | g_assert(qemu_in_coroutine()); | |
814 | assert(i < MAX_NBD_REQUESTS); | |
815 | ||
816 | s->requests[i].coroutine = qemu_coroutine_self(); | |
817 | s->requests[i].offset = request->from; | |
818 | s->requests[i].receiving = false; | |
819 | ||
820 | request->handle = INDEX_TO_HANDLE(s, i); | |
821 | ||
86f8cdf3 VSO |
822 | assert(s->ioc); |
823 | ||
824 | if (qiov) { | |
825 | qio_channel_set_cork(s->ioc, true); | |
826 | rc = nbd_send_request(s->ioc, request); | |
fee091cd LS |
827 | if (qatomic_load_acquire(&s->state) == NBD_CLIENT_CONNECTED && |
828 | rc >= 0) { | |
86f8cdf3 VSO |
829 | if (qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov, |
830 | NULL) < 0) { | |
831 | rc = -EIO; | |
832 | } | |
833 | } else if (rc >= 0) { | |
834 | rc = -EIO; | |
835 | } | |
836 | qio_channel_set_cork(s->ioc, false); | |
837 | } else { | |
838 | rc = nbd_send_request(s->ioc, request); | |
839 | } | |
840 | ||
841 | err: | |
842 | if (rc < 0) { | |
a34b1e5e VSO |
843 | nbd_channel_error(s, rc); |
844 | if (i != -1) { | |
845 | s->requests[i].coroutine = NULL; | |
846 | s->in_flight--; | |
847 | } | |
f7651539 VSO |
848 | if (s->in_flight == 0 && s->wait_in_flight) { |
849 | aio_co_wake(s->connection_co); | |
850 | } else { | |
851 | qemu_co_queue_next(&s->free_sema); | |
852 | } | |
86f8cdf3 VSO |
853 | } |
854 | qemu_co_mutex_unlock(&s->send_mutex); | |
855 | return rc; | |
856 | } | |
857 | ||
858 | static inline uint16_t payload_advance16(uint8_t **payload) | |
859 | { | |
860 | *payload += 2; | |
861 | return lduw_be_p(*payload - 2); | |
862 | } | |
863 | ||
864 | static inline uint32_t payload_advance32(uint8_t **payload) | |
865 | { | |
866 | *payload += 4; | |
867 | return ldl_be_p(*payload - 4); | |
868 | } | |
869 | ||
870 | static inline uint64_t payload_advance64(uint8_t **payload) | |
871 | { | |
872 | *payload += 8; | |
873 | return ldq_be_p(*payload - 8); | |
874 | } | |
875 | ||
611ae1d7 | 876 | static int nbd_parse_offset_hole_payload(BDRVNBDState *s, |
86f8cdf3 VSO |
877 | NBDStructuredReplyChunk *chunk, |
878 | uint8_t *payload, uint64_t orig_offset, | |
879 | QEMUIOVector *qiov, Error **errp) | |
880 | { | |
881 | uint64_t offset; | |
882 | uint32_t hole_size; | |
883 | ||
884 | if (chunk->length != sizeof(offset) + sizeof(hole_size)) { | |
885 | error_setg(errp, "Protocol error: invalid payload for " | |
886 | "NBD_REPLY_TYPE_OFFSET_HOLE"); | |
887 | return -EINVAL; | |
888 | } | |
889 | ||
890 | offset = payload_advance64(&payload); | |
891 | hole_size = payload_advance32(&payload); | |
892 | ||
893 | if (!hole_size || offset < orig_offset || hole_size > qiov->size || | |
894 | offset > orig_offset + qiov->size - hole_size) { | |
895 | error_setg(errp, "Protocol error: server sent chunk exceeding requested" | |
896 | " region"); | |
897 | return -EINVAL; | |
898 | } | |
611ae1d7 VSO |
899 | if (s->info.min_block && |
900 | !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) { | |
86f8cdf3 VSO |
901 | trace_nbd_structured_read_compliance("hole"); |
902 | } | |
903 | ||
904 | qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size); | |
905 | ||
906 | return 0; | |
907 | } | |
908 | ||
909 | /* | |
910 | * nbd_parse_blockstatus_payload | |
911 | * Based on our request, we expect only one extent in reply, for the | |
912 | * base:allocation context. | |
913 | */ | |
611ae1d7 | 914 | static int nbd_parse_blockstatus_payload(BDRVNBDState *s, |
86f8cdf3 VSO |
915 | NBDStructuredReplyChunk *chunk, |
916 | uint8_t *payload, uint64_t orig_length, | |
917 | NBDExtent *extent, Error **errp) | |
918 | { | |
919 | uint32_t context_id; | |
920 | ||
921 | /* The server succeeded, so it must have sent [at least] one extent */ | |
922 | if (chunk->length < sizeof(context_id) + sizeof(*extent)) { | |
923 | error_setg(errp, "Protocol error: invalid payload for " | |
924 | "NBD_REPLY_TYPE_BLOCK_STATUS"); | |
925 | return -EINVAL; | |
926 | } | |
927 | ||
928 | context_id = payload_advance32(&payload); | |
611ae1d7 | 929 | if (s->info.context_id != context_id) { |
86f8cdf3 VSO |
930 | error_setg(errp, "Protocol error: unexpected context id %d for " |
931 | "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context " | |
932 | "id is %d", context_id, | |
611ae1d7 | 933 | s->info.context_id); |
86f8cdf3 VSO |
934 | return -EINVAL; |
935 | } | |
936 | ||
937 | extent->length = payload_advance32(&payload); | |
938 | extent->flags = payload_advance32(&payload); | |
939 | ||
940 | if (extent->length == 0) { | |
941 | error_setg(errp, "Protocol error: server sent status chunk with " | |
942 | "zero length"); | |
943 | return -EINVAL; | |
944 | } | |
945 | ||
946 | /* | |
947 | * A server sending unaligned block status is in violation of the | |
948 | * protocol, but as qemu-nbd 3.1 is such a server (at least for | |
949 | * POSIX files that are not a multiple of 512 bytes, since qemu | |
950 | * rounds files up to 512-byte multiples but lseek(SEEK_HOLE) | |
951 | * still sees an implicit hole beyond the real EOF), it's nicer to | |
952 | * work around the misbehaving server. If the request included | |
953 | * more than the final unaligned block, truncate it back to an | |
954 | * aligned result; if the request was only the final block, round | |
955 | * up to the full block and change the status to fully-allocated | |
956 | * (always a safe status, even if it loses information). | |
957 | */ | |
611ae1d7 VSO |
958 | if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length, |
959 | s->info.min_block)) { | |
86f8cdf3 | 960 | trace_nbd_parse_blockstatus_compliance("extent length is unaligned"); |
611ae1d7 | 961 | if (extent->length > s->info.min_block) { |
86f8cdf3 | 962 | extent->length = QEMU_ALIGN_DOWN(extent->length, |
611ae1d7 | 963 | s->info.min_block); |
86f8cdf3 | 964 | } else { |
611ae1d7 | 965 | extent->length = s->info.min_block; |
86f8cdf3 VSO |
966 | extent->flags = 0; |
967 | } | |
968 | } | |
969 | ||
970 | /* | |
971 | * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have | |
972 | * sent us any more than one extent, nor should it have included | |
973 | * status beyond our request in that extent. However, it's easy | |
974 | * enough to ignore the server's noncompliance without killing the | |
975 | * connection; just ignore trailing extents, and clamp things to | |
976 | * the length of our request. | |
977 | */ | |
978 | if (chunk->length > sizeof(context_id) + sizeof(*extent)) { | |
979 | trace_nbd_parse_blockstatus_compliance("more than one extent"); | |
980 | } | |
981 | if (extent->length > orig_length) { | |
982 | extent->length = orig_length; | |
983 | trace_nbd_parse_blockstatus_compliance("extent length too large"); | |
984 | } | |
985 | ||
dbc7b014 EB |
986 | /* |
987 | * HACK: if we are using x-dirty-bitmaps to access | |
988 | * qemu:allocation-depth, treat all depths > 2 the same as 2, | |
989 | * since nbd_client_co_block_status is only expecting the low two | |
990 | * bits to be set. | |
991 | */ | |
992 | if (s->alloc_depth && extent->flags > 2) { | |
993 | extent->flags = 2; | |
994 | } | |
995 | ||
86f8cdf3 VSO |
996 | return 0; |
997 | } | |
998 | ||
999 | /* | |
1000 | * nbd_parse_error_payload | |
1001 | * on success @errp contains message describing nbd error reply | |
1002 | */ | |
1003 | static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk, | |
1004 | uint8_t *payload, int *request_ret, | |
1005 | Error **errp) | |
1006 | { | |
1007 | uint32_t error; | |
1008 | uint16_t message_size; | |
1009 | ||
1010 | assert(chunk->type & (1 << 15)); | |
1011 | ||
1012 | if (chunk->length < sizeof(error) + sizeof(message_size)) { | |
1013 | error_setg(errp, | |
1014 | "Protocol error: invalid payload for structured error"); | |
1015 | return -EINVAL; | |
1016 | } | |
1017 | ||
1018 | error = nbd_errno_to_system_errno(payload_advance32(&payload)); | |
1019 | if (error == 0) { | |
1020 | error_setg(errp, "Protocol error: server sent structured error chunk " | |
1021 | "with error = 0"); | |
1022 | return -EINVAL; | |
1023 | } | |
1024 | ||
1025 | *request_ret = -error; | |
1026 | message_size = payload_advance16(&payload); | |
1027 | ||
1028 | if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) { | |
1029 | error_setg(errp, "Protocol error: server sent structured error chunk " | |
1030 | "with incorrect message size"); | |
1031 | return -EINVAL; | |
1032 | } | |
1033 | ||
1034 | /* TODO: Add a trace point to mention the server complaint */ | |
1035 | ||
1036 | /* TODO handle ERROR_OFFSET */ | |
1037 | ||
1038 | return 0; | |
1039 | } | |
1040 | ||
611ae1d7 | 1041 | static int nbd_co_receive_offset_data_payload(BDRVNBDState *s, |
86f8cdf3 VSO |
1042 | uint64_t orig_offset, |
1043 | QEMUIOVector *qiov, Error **errp) | |
1044 | { | |
1045 | QEMUIOVector sub_qiov; | |
1046 | uint64_t offset; | |
1047 | size_t data_size; | |
1048 | int ret; | |
1049 | NBDStructuredReplyChunk *chunk = &s->reply.structured; | |
1050 | ||
1051 | assert(nbd_reply_is_structured(&s->reply)); | |
1052 | ||
1053 | /* The NBD spec requires at least one byte of payload */ | |
1054 | if (chunk->length <= sizeof(offset)) { | |
1055 | error_setg(errp, "Protocol error: invalid payload for " | |
1056 | "NBD_REPLY_TYPE_OFFSET_DATA"); | |
1057 | return -EINVAL; | |
1058 | } | |
1059 | ||
1060 | if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) { | |
1061 | return -EIO; | |
1062 | } | |
1063 | ||
1064 | data_size = chunk->length - sizeof(offset); | |
1065 | assert(data_size); | |
1066 | if (offset < orig_offset || data_size > qiov->size || | |
1067 | offset > orig_offset + qiov->size - data_size) { | |
1068 | error_setg(errp, "Protocol error: server sent chunk exceeding requested" | |
1069 | " region"); | |
1070 | return -EINVAL; | |
1071 | } | |
1072 | if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) { | |
1073 | trace_nbd_structured_read_compliance("data"); | |
1074 | } | |
1075 | ||
1076 | qemu_iovec_init(&sub_qiov, qiov->niov); | |
1077 | qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size); | |
1078 | ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp); | |
1079 | qemu_iovec_destroy(&sub_qiov); | |
1080 | ||
1081 | return ret < 0 ? -EIO : 0; | |
1082 | } | |
1083 | ||
1084 | #define NBD_MAX_MALLOC_PAYLOAD 1000 | |
1085 | static coroutine_fn int nbd_co_receive_structured_payload( | |
611ae1d7 | 1086 | BDRVNBDState *s, void **payload, Error **errp) |
86f8cdf3 VSO |
1087 | { |
1088 | int ret; | |
1089 | uint32_t len; | |
1090 | ||
1091 | assert(nbd_reply_is_structured(&s->reply)); | |
1092 | ||
1093 | len = s->reply.structured.length; | |
1094 | ||
1095 | if (len == 0) { | |
1096 | return 0; | |
1097 | } | |
1098 | ||
1099 | if (payload == NULL) { | |
1100 | error_setg(errp, "Unexpected structured payload"); | |
1101 | return -EINVAL; | |
1102 | } | |
1103 | ||
1104 | if (len > NBD_MAX_MALLOC_PAYLOAD) { | |
1105 | error_setg(errp, "Payload too large"); | |
1106 | return -EINVAL; | |
1107 | } | |
1108 | ||
1109 | *payload = g_new(char, len); | |
1110 | ret = nbd_read(s->ioc, *payload, len, "structured payload", errp); | |
1111 | if (ret < 0) { | |
1112 | g_free(*payload); | |
1113 | *payload = NULL; | |
1114 | return ret; | |
1115 | } | |
1116 | ||
1117 | return 0; | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | * nbd_co_do_receive_one_chunk | |
1122 | * for simple reply: | |
1123 | * set request_ret to received reply error | |
1124 | * if qiov is not NULL: read payload to @qiov | |
1125 | * for structured reply chunk: | |
1126 | * if error chunk: read payload, set @request_ret, do not set @payload | |
1127 | * else if offset_data chunk: read payload data to @qiov, do not set @payload | |
1128 | * else: read payload to @payload | |
1129 | * | |
1130 | * If function fails, @errp contains corresponding error message, and the | |
1131 | * connection with the server is suspect. If it returns 0, then the | |
1132 | * transaction succeeded (although @request_ret may be a negative errno | |
1133 | * corresponding to the server's error reply), and errp is unchanged. | |
1134 | */ | |
1135 | static coroutine_fn int nbd_co_do_receive_one_chunk( | |
611ae1d7 | 1136 | BDRVNBDState *s, uint64_t handle, bool only_structured, |
86f8cdf3 VSO |
1137 | int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp) |
1138 | { | |
1139 | int ret; | |
1140 | int i = HANDLE_TO_INDEX(s, handle); | |
1141 | void *local_payload = NULL; | |
1142 | NBDStructuredReplyChunk *chunk; | |
1143 | ||
1144 | if (payload) { | |
1145 | *payload = NULL; | |
1146 | } | |
1147 | *request_ret = 0; | |
1148 | ||
1149 | /* Wait until we're woken up by nbd_connection_entry. */ | |
1150 | s->requests[i].receiving = true; | |
1151 | qemu_coroutine_yield(); | |
1152 | s->requests[i].receiving = false; | |
fee091cd | 1153 | if (qatomic_load_acquire(&s->state) != NBD_CLIENT_CONNECTED) { |
86f8cdf3 VSO |
1154 | error_setg(errp, "Connection closed"); |
1155 | return -EIO; | |
1156 | } | |
1157 | assert(s->ioc); | |
1158 | ||
1159 | assert(s->reply.handle == handle); | |
1160 | ||
1161 | if (nbd_reply_is_simple(&s->reply)) { | |
1162 | if (only_structured) { | |
1163 | error_setg(errp, "Protocol error: simple reply when structured " | |
1164 | "reply chunk was expected"); | |
1165 | return -EINVAL; | |
1166 | } | |
1167 | ||
1168 | *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error); | |
1169 | if (*request_ret < 0 || !qiov) { | |
1170 | return 0; | |
1171 | } | |
1172 | ||
1173 | return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov, | |
1174 | errp) < 0 ? -EIO : 0; | |
1175 | } | |
1176 | ||
1177 | /* handle structured reply chunk */ | |
1178 | assert(s->info.structured_reply); | |
1179 | chunk = &s->reply.structured; | |
1180 | ||
1181 | if (chunk->type == NBD_REPLY_TYPE_NONE) { | |
1182 | if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) { | |
1183 | error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without" | |
1184 | " NBD_REPLY_FLAG_DONE flag set"); | |
1185 | return -EINVAL; | |
1186 | } | |
1187 | if (chunk->length) { | |
1188 | error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with" | |
1189 | " nonzero length"); | |
1190 | return -EINVAL; | |
1191 | } | |
1192 | return 0; | |
1193 | } | |
1194 | ||
1195 | if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) { | |
1196 | if (!qiov) { | |
1197 | error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk"); | |
1198 | return -EINVAL; | |
1199 | } | |
1200 | ||
1201 | return nbd_co_receive_offset_data_payload(s, s->requests[i].offset, | |
1202 | qiov, errp); | |
1203 | } | |
1204 | ||
1205 | if (nbd_reply_type_is_error(chunk->type)) { | |
1206 | payload = &local_payload; | |
1207 | } | |
1208 | ||
1209 | ret = nbd_co_receive_structured_payload(s, payload, errp); | |
1210 | if (ret < 0) { | |
1211 | return ret; | |
1212 | } | |
1213 | ||
1214 | if (nbd_reply_type_is_error(chunk->type)) { | |
1215 | ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp); | |
1216 | g_free(local_payload); | |
1217 | return ret; | |
1218 | } | |
1219 | ||
1220 | return 0; | |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * nbd_co_receive_one_chunk | |
1225 | * Read reply, wake up connection_co and set s->quit if needed. | |
1226 | * Return value is a fatal error code or normal nbd reply error code | |
1227 | */ | |
1228 | static coroutine_fn int nbd_co_receive_one_chunk( | |
611ae1d7 | 1229 | BDRVNBDState *s, uint64_t handle, bool only_structured, |
86f8cdf3 VSO |
1230 | int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload, |
1231 | Error **errp) | |
1232 | { | |
1233 | int ret = nbd_co_do_receive_one_chunk(s, handle, only_structured, | |
1234 | request_ret, qiov, payload, errp); | |
1235 | ||
1236 | if (ret < 0) { | |
5cf42b1c | 1237 | memset(reply, 0, sizeof(*reply)); |
a34b1e5e | 1238 | nbd_channel_error(s, ret); |
86f8cdf3 VSO |
1239 | } else { |
1240 | /* For assert at loop start in nbd_connection_entry */ | |
5cf42b1c | 1241 | *reply = s->reply; |
86f8cdf3 | 1242 | } |
f7651539 | 1243 | s->reply.handle = 0; |
86f8cdf3 | 1244 | |
f7651539 VSO |
1245 | if (s->connection_co && !s->wait_in_flight) { |
1246 | /* | |
1247 | * We must check s->wait_in_flight, because we may entered by | |
1248 | * nbd_recv_coroutines_wake_all(), in this case we should not | |
1249 | * wake connection_co here, it will woken by last request. | |
1250 | */ | |
86f8cdf3 VSO |
1251 | aio_co_wake(s->connection_co); |
1252 | } | |
1253 | ||
1254 | return ret; | |
1255 | } | |
1256 | ||
1257 | typedef struct NBDReplyChunkIter { | |
1258 | int ret; | |
1259 | int request_ret; | |
1260 | Error *err; | |
1261 | bool done, only_structured; | |
1262 | } NBDReplyChunkIter; | |
1263 | ||
1264 | static void nbd_iter_channel_error(NBDReplyChunkIter *iter, | |
1265 | int ret, Error **local_err) | |
1266 | { | |
d9366135 | 1267 | assert(local_err && *local_err); |
86f8cdf3 VSO |
1268 | assert(ret < 0); |
1269 | ||
1270 | if (!iter->ret) { | |
1271 | iter->ret = ret; | |
1272 | error_propagate(&iter->err, *local_err); | |
1273 | } else { | |
1274 | error_free(*local_err); | |
1275 | } | |
1276 | ||
1277 | *local_err = NULL; | |
1278 | } | |
1279 | ||
1280 | static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret) | |
1281 | { | |
1282 | assert(ret < 0); | |
1283 | ||
1284 | if (!iter->request_ret) { | |
1285 | iter->request_ret = ret; | |
1286 | } | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * NBD_FOREACH_REPLY_CHUNK | |
1291 | * The pointer stored in @payload requires g_free() to free it. | |
1292 | */ | |
1293 | #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \ | |
1294 | qiov, reply, payload) \ | |
1295 | for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \ | |
1296 | nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);) | |
1297 | ||
1298 | /* | |
1299 | * nbd_reply_chunk_iter_receive | |
1300 | * The pointer stored in @payload requires g_free() to free it. | |
1301 | */ | |
611ae1d7 | 1302 | static bool nbd_reply_chunk_iter_receive(BDRVNBDState *s, |
86f8cdf3 VSO |
1303 | NBDReplyChunkIter *iter, |
1304 | uint64_t handle, | |
1305 | QEMUIOVector *qiov, NBDReply *reply, | |
1306 | void **payload) | |
1307 | { | |
1308 | int ret, request_ret; | |
1309 | NBDReply local_reply; | |
1310 | NBDStructuredReplyChunk *chunk; | |
1311 | Error *local_err = NULL; | |
fee091cd | 1312 | if (qatomic_load_acquire(&s->state) != NBD_CLIENT_CONNECTED) { |
86f8cdf3 VSO |
1313 | error_setg(&local_err, "Connection closed"); |
1314 | nbd_iter_channel_error(iter, -EIO, &local_err); | |
1315 | goto break_loop; | |
1316 | } | |
1317 | ||
1318 | if (iter->done) { | |
1319 | /* Previous iteration was last. */ | |
1320 | goto break_loop; | |
1321 | } | |
1322 | ||
1323 | if (reply == NULL) { | |
1324 | reply = &local_reply; | |
1325 | } | |
1326 | ||
1327 | ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured, | |
1328 | &request_ret, qiov, reply, payload, | |
1329 | &local_err); | |
1330 | if (ret < 0) { | |
1331 | nbd_iter_channel_error(iter, ret, &local_err); | |
1332 | } else if (request_ret < 0) { | |
1333 | nbd_iter_request_error(iter, request_ret); | |
1334 | } | |
1335 | ||
1336 | /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */ | |
fee091cd LS |
1337 | if (nbd_reply_is_simple(reply) || |
1338 | qatomic_load_acquire(&s->state) != NBD_CLIENT_CONNECTED) { | |
86f8cdf3 VSO |
1339 | goto break_loop; |
1340 | } | |
1341 | ||
1342 | chunk = &reply->structured; | |
1343 | iter->only_structured = true; | |
1344 | ||
1345 | if (chunk->type == NBD_REPLY_TYPE_NONE) { | |
1346 | /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */ | |
1347 | assert(chunk->flags & NBD_REPLY_FLAG_DONE); | |
1348 | goto break_loop; | |
1349 | } | |
1350 | ||
1351 | if (chunk->flags & NBD_REPLY_FLAG_DONE) { | |
1352 | /* This iteration is last. */ | |
1353 | iter->done = true; | |
1354 | } | |
1355 | ||
1356 | /* Execute the loop body */ | |
1357 | return true; | |
1358 | ||
1359 | break_loop: | |
1360 | s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL; | |
1361 | ||
1362 | qemu_co_mutex_lock(&s->send_mutex); | |
1363 | s->in_flight--; | |
f7651539 VSO |
1364 | if (s->in_flight == 0 && s->wait_in_flight) { |
1365 | aio_co_wake(s->connection_co); | |
1366 | } else { | |
1367 | qemu_co_queue_next(&s->free_sema); | |
1368 | } | |
86f8cdf3 VSO |
1369 | qemu_co_mutex_unlock(&s->send_mutex); |
1370 | ||
1371 | return false; | |
1372 | } | |
1373 | ||
611ae1d7 | 1374 | static int nbd_co_receive_return_code(BDRVNBDState *s, uint64_t handle, |
86f8cdf3 VSO |
1375 | int *request_ret, Error **errp) |
1376 | { | |
1377 | NBDReplyChunkIter iter; | |
1378 | ||
1379 | NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) { | |
1380 | /* nbd_reply_chunk_iter_receive does all the work */ | |
1381 | } | |
1382 | ||
1383 | error_propagate(errp, iter.err); | |
1384 | *request_ret = iter.request_ret; | |
1385 | return iter.ret; | |
1386 | } | |
1387 | ||
611ae1d7 | 1388 | static int nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t handle, |
86f8cdf3 VSO |
1389 | uint64_t offset, QEMUIOVector *qiov, |
1390 | int *request_ret, Error **errp) | |
1391 | { | |
1392 | NBDReplyChunkIter iter; | |
1393 | NBDReply reply; | |
1394 | void *payload = NULL; | |
1395 | Error *local_err = NULL; | |
1396 | ||
1397 | NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply, | |
1398 | qiov, &reply, &payload) | |
1399 | { | |
1400 | int ret; | |
1401 | NBDStructuredReplyChunk *chunk = &reply.structured; | |
1402 | ||
1403 | assert(nbd_reply_is_structured(&reply)); | |
1404 | ||
1405 | switch (chunk->type) { | |
1406 | case NBD_REPLY_TYPE_OFFSET_DATA: | |
1407 | /* | |
1408 | * special cased in nbd_co_receive_one_chunk, data is already | |
1409 | * in qiov | |
1410 | */ | |
1411 | break; | |
1412 | case NBD_REPLY_TYPE_OFFSET_HOLE: | |
1413 | ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload, | |
1414 | offset, qiov, &local_err); | |
1415 | if (ret < 0) { | |
a34b1e5e | 1416 | nbd_channel_error(s, ret); |
86f8cdf3 VSO |
1417 | nbd_iter_channel_error(&iter, ret, &local_err); |
1418 | } | |
1419 | break; | |
1420 | default: | |
1421 | if (!nbd_reply_type_is_error(chunk->type)) { | |
1422 | /* not allowed reply type */ | |
a34b1e5e | 1423 | nbd_channel_error(s, -EINVAL); |
86f8cdf3 VSO |
1424 | error_setg(&local_err, |
1425 | "Unexpected reply type: %d (%s) for CMD_READ", | |
1426 | chunk->type, nbd_reply_type_lookup(chunk->type)); | |
1427 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); | |
1428 | } | |
1429 | } | |
1430 | ||
1431 | g_free(payload); | |
1432 | payload = NULL; | |
1433 | } | |
1434 | ||
1435 | error_propagate(errp, iter.err); | |
1436 | *request_ret = iter.request_ret; | |
1437 | return iter.ret; | |
1438 | } | |
1439 | ||
611ae1d7 | 1440 | static int nbd_co_receive_blockstatus_reply(BDRVNBDState *s, |
86f8cdf3 VSO |
1441 | uint64_t handle, uint64_t length, |
1442 | NBDExtent *extent, | |
1443 | int *request_ret, Error **errp) | |
1444 | { | |
1445 | NBDReplyChunkIter iter; | |
1446 | NBDReply reply; | |
1447 | void *payload = NULL; | |
1448 | Error *local_err = NULL; | |
1449 | bool received = false; | |
1450 | ||
1451 | assert(!extent->length); | |
1452 | NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, &reply, &payload) { | |
1453 | int ret; | |
1454 | NBDStructuredReplyChunk *chunk = &reply.structured; | |
1455 | ||
1456 | assert(nbd_reply_is_structured(&reply)); | |
1457 | ||
1458 | switch (chunk->type) { | |
1459 | case NBD_REPLY_TYPE_BLOCK_STATUS: | |
1460 | if (received) { | |
a34b1e5e | 1461 | nbd_channel_error(s, -EINVAL); |
86f8cdf3 VSO |
1462 | error_setg(&local_err, "Several BLOCK_STATUS chunks in reply"); |
1463 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); | |
1464 | } | |
1465 | received = true; | |
1466 | ||
1467 | ret = nbd_parse_blockstatus_payload(s, &reply.structured, | |
1468 | payload, length, extent, | |
1469 | &local_err); | |
1470 | if (ret < 0) { | |
a34b1e5e | 1471 | nbd_channel_error(s, ret); |
86f8cdf3 VSO |
1472 | nbd_iter_channel_error(&iter, ret, &local_err); |
1473 | } | |
1474 | break; | |
1475 | default: | |
1476 | if (!nbd_reply_type_is_error(chunk->type)) { | |
a34b1e5e | 1477 | nbd_channel_error(s, -EINVAL); |
86f8cdf3 VSO |
1478 | error_setg(&local_err, |
1479 | "Unexpected reply type: %d (%s) " | |
1480 | "for CMD_BLOCK_STATUS", | |
1481 | chunk->type, nbd_reply_type_lookup(chunk->type)); | |
1482 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); | |
1483 | } | |
1484 | } | |
1485 | ||
1486 | g_free(payload); | |
1487 | payload = NULL; | |
1488 | } | |
1489 | ||
1490 | if (!extent->length && !iter.request_ret) { | |
1491 | error_setg(&local_err, "Server did not reply with any status extents"); | |
1492 | nbd_iter_channel_error(&iter, -EIO, &local_err); | |
1493 | } | |
1494 | ||
1495 | error_propagate(errp, iter.err); | |
1496 | *request_ret = iter.request_ret; | |
1497 | return iter.ret; | |
1498 | } | |
1499 | ||
1500 | static int nbd_co_request(BlockDriverState *bs, NBDRequest *request, | |
1501 | QEMUIOVector *write_qiov) | |
1502 | { | |
1503 | int ret, request_ret; | |
1504 | Error *local_err = NULL; | |
611ae1d7 | 1505 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1506 | |
1507 | assert(request->type != NBD_CMD_READ); | |
1508 | if (write_qiov) { | |
1509 | assert(request->type == NBD_CMD_WRITE); | |
1510 | assert(request->len == iov_size(write_qiov->iov, write_qiov->niov)); | |
1511 | } else { | |
1512 | assert(request->type != NBD_CMD_WRITE); | |
1513 | } | |
86f8cdf3 | 1514 | |
f7651539 VSO |
1515 | do { |
1516 | ret = nbd_co_send_request(bs, request, write_qiov); | |
1517 | if (ret < 0) { | |
1518 | continue; | |
1519 | } | |
1520 | ||
1521 | ret = nbd_co_receive_return_code(s, request->handle, | |
1522 | &request_ret, &local_err); | |
1523 | if (local_err) { | |
1524 | trace_nbd_co_request_fail(request->from, request->len, | |
1525 | request->handle, request->flags, | |
1526 | request->type, | |
1527 | nbd_cmd_lookup(request->type), | |
1528 | ret, error_get_pretty(local_err)); | |
1529 | error_free(local_err); | |
1530 | local_err = NULL; | |
1531 | } | |
1532 | } while (ret < 0 && nbd_client_connecting_wait(s)); | |
1533 | ||
86f8cdf3 VSO |
1534 | return ret ? ret : request_ret; |
1535 | } | |
1536 | ||
1537 | static int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset, | |
1538 | uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1539 | { | |
1540 | int ret, request_ret; | |
1541 | Error *local_err = NULL; | |
611ae1d7 | 1542 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1543 | NBDRequest request = { |
1544 | .type = NBD_CMD_READ, | |
1545 | .from = offset, | |
1546 | .len = bytes, | |
1547 | }; | |
1548 | ||
1549 | assert(bytes <= NBD_MAX_BUFFER_SIZE); | |
1550 | assert(!flags); | |
1551 | ||
1552 | if (!bytes) { | |
1553 | return 0; | |
1554 | } | |
1555 | /* | |
1556 | * Work around the fact that the block layer doesn't do | |
1557 | * byte-accurate sizing yet - if the read exceeds the server's | |
1558 | * advertised size because the block layer rounded size up, then | |
1559 | * truncate the request to the server and tail-pad with zero. | |
1560 | */ | |
611ae1d7 | 1561 | if (offset >= s->info.size) { |
86f8cdf3 VSO |
1562 | assert(bytes < BDRV_SECTOR_SIZE); |
1563 | qemu_iovec_memset(qiov, 0, 0, bytes); | |
1564 | return 0; | |
1565 | } | |
611ae1d7 VSO |
1566 | if (offset + bytes > s->info.size) { |
1567 | uint64_t slop = offset + bytes - s->info.size; | |
86f8cdf3 VSO |
1568 | |
1569 | assert(slop < BDRV_SECTOR_SIZE); | |
1570 | qemu_iovec_memset(qiov, bytes - slop, 0, slop); | |
1571 | request.len -= slop; | |
1572 | } | |
1573 | ||
f7651539 VSO |
1574 | do { |
1575 | ret = nbd_co_send_request(bs, &request, NULL); | |
1576 | if (ret < 0) { | |
1577 | continue; | |
1578 | } | |
1579 | ||
1580 | ret = nbd_co_receive_cmdread_reply(s, request.handle, offset, qiov, | |
1581 | &request_ret, &local_err); | |
1582 | if (local_err) { | |
1583 | trace_nbd_co_request_fail(request.from, request.len, request.handle, | |
1584 | request.flags, request.type, | |
1585 | nbd_cmd_lookup(request.type), | |
1586 | ret, error_get_pretty(local_err)); | |
1587 | error_free(local_err); | |
1588 | local_err = NULL; | |
1589 | } | |
1590 | } while (ret < 0 && nbd_client_connecting_wait(s)); | |
86f8cdf3 | 1591 | |
86f8cdf3 VSO |
1592 | return ret ? ret : request_ret; |
1593 | } | |
1594 | ||
1595 | static int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset, | |
1596 | uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1597 | { | |
611ae1d7 | 1598 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1599 | NBDRequest request = { |
1600 | .type = NBD_CMD_WRITE, | |
1601 | .from = offset, | |
1602 | .len = bytes, | |
1603 | }; | |
1604 | ||
611ae1d7 | 1605 | assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); |
86f8cdf3 | 1606 | if (flags & BDRV_REQ_FUA) { |
611ae1d7 | 1607 | assert(s->info.flags & NBD_FLAG_SEND_FUA); |
86f8cdf3 VSO |
1608 | request.flags |= NBD_CMD_FLAG_FUA; |
1609 | } | |
1610 | ||
1611 | assert(bytes <= NBD_MAX_BUFFER_SIZE); | |
1612 | ||
1613 | if (!bytes) { | |
1614 | return 0; | |
1615 | } | |
1616 | return nbd_co_request(bs, &request, qiov); | |
1617 | } | |
1618 | ||
1619 | static int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, | |
1620 | int bytes, BdrvRequestFlags flags) | |
1621 | { | |
611ae1d7 | 1622 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1623 | NBDRequest request = { |
1624 | .type = NBD_CMD_WRITE_ZEROES, | |
1625 | .from = offset, | |
1626 | .len = bytes, | |
1627 | }; | |
1628 | ||
611ae1d7 VSO |
1629 | assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); |
1630 | if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) { | |
86f8cdf3 VSO |
1631 | return -ENOTSUP; |
1632 | } | |
1633 | ||
1634 | if (flags & BDRV_REQ_FUA) { | |
611ae1d7 | 1635 | assert(s->info.flags & NBD_FLAG_SEND_FUA); |
86f8cdf3 VSO |
1636 | request.flags |= NBD_CMD_FLAG_FUA; |
1637 | } | |
1638 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
1639 | request.flags |= NBD_CMD_FLAG_NO_HOLE; | |
1640 | } | |
f061656c EB |
1641 | if (flags & BDRV_REQ_NO_FALLBACK) { |
1642 | assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO); | |
1643 | request.flags |= NBD_CMD_FLAG_FAST_ZERO; | |
1644 | } | |
86f8cdf3 VSO |
1645 | |
1646 | if (!bytes) { | |
1647 | return 0; | |
1648 | } | |
1649 | return nbd_co_request(bs, &request, NULL); | |
1650 | } | |
1651 | ||
1652 | static int nbd_client_co_flush(BlockDriverState *bs) | |
1653 | { | |
611ae1d7 | 1654 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1655 | NBDRequest request = { .type = NBD_CMD_FLUSH }; |
1656 | ||
611ae1d7 | 1657 | if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) { |
86f8cdf3 VSO |
1658 | return 0; |
1659 | } | |
1660 | ||
1661 | request.from = 0; | |
1662 | request.len = 0; | |
1663 | ||
1664 | return nbd_co_request(bs, &request, NULL); | |
1665 | } | |
1666 | ||
1667 | static int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, | |
1668 | int bytes) | |
1669 | { | |
611ae1d7 | 1670 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1671 | NBDRequest request = { |
1672 | .type = NBD_CMD_TRIM, | |
1673 | .from = offset, | |
1674 | .len = bytes, | |
1675 | }; | |
1676 | ||
611ae1d7 VSO |
1677 | assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); |
1678 | if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) { | |
86f8cdf3 VSO |
1679 | return 0; |
1680 | } | |
1681 | ||
1682 | return nbd_co_request(bs, &request, NULL); | |
1683 | } | |
1684 | ||
1685 | static int coroutine_fn nbd_client_co_block_status( | |
1686 | BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes, | |
1687 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
1688 | { | |
1689 | int ret, request_ret; | |
1690 | NBDExtent extent = { 0 }; | |
611ae1d7 | 1691 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1692 | Error *local_err = NULL; |
1693 | ||
1694 | NBDRequest request = { | |
1695 | .type = NBD_CMD_BLOCK_STATUS, | |
1696 | .from = offset, | |
6bf792b4 | 1697 | .len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment), |
611ae1d7 | 1698 | MIN(bytes, s->info.size - offset)), |
86f8cdf3 VSO |
1699 | .flags = NBD_CMD_FLAG_REQ_ONE, |
1700 | }; | |
1701 | ||
611ae1d7 | 1702 | if (!s->info.base_allocation) { |
86f8cdf3 VSO |
1703 | *pnum = bytes; |
1704 | *map = offset; | |
1705 | *file = bs; | |
1706 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * Work around the fact that the block layer doesn't do | |
1711 | * byte-accurate sizing yet - if the status request exceeds the | |
1712 | * server's advertised size because the block layer rounded size | |
1713 | * up, we truncated the request to the server (above), or are | |
1714 | * called on just the hole. | |
1715 | */ | |
611ae1d7 | 1716 | if (offset >= s->info.size) { |
86f8cdf3 VSO |
1717 | *pnum = bytes; |
1718 | assert(bytes < BDRV_SECTOR_SIZE); | |
1719 | /* Intentionally don't report offset_valid for the hole */ | |
1720 | return BDRV_BLOCK_ZERO; | |
1721 | } | |
1722 | ||
611ae1d7 VSO |
1723 | if (s->info.min_block) { |
1724 | assert(QEMU_IS_ALIGNED(request.len, s->info.min_block)); | |
86f8cdf3 | 1725 | } |
f7651539 VSO |
1726 | do { |
1727 | ret = nbd_co_send_request(bs, &request, NULL); | |
1728 | if (ret < 0) { | |
1729 | continue; | |
1730 | } | |
1731 | ||
1732 | ret = nbd_co_receive_blockstatus_reply(s, request.handle, bytes, | |
1733 | &extent, &request_ret, | |
1734 | &local_err); | |
1735 | if (local_err) { | |
1736 | trace_nbd_co_request_fail(request.from, request.len, request.handle, | |
1737 | request.flags, request.type, | |
1738 | nbd_cmd_lookup(request.type), | |
1739 | ret, error_get_pretty(local_err)); | |
1740 | error_free(local_err); | |
1741 | local_err = NULL; | |
1742 | } | |
1743 | } while (ret < 0 && nbd_client_connecting_wait(s)); | |
86f8cdf3 | 1744 | |
86f8cdf3 VSO |
1745 | if (ret < 0 || request_ret < 0) { |
1746 | return ret ? ret : request_ret; | |
1747 | } | |
1748 | ||
1749 | assert(extent.length); | |
1750 | *pnum = extent.length; | |
1751 | *map = offset; | |
1752 | *file = bs; | |
1753 | return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) | | |
1754 | (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) | | |
1755 | BDRV_BLOCK_OFFSET_VALID; | |
1756 | } | |
1757 | ||
e99754b4 ML |
1758 | static int nbd_client_reopen_prepare(BDRVReopenState *state, |
1759 | BlockReopenQueue *queue, Error **errp) | |
1760 | { | |
1761 | BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque; | |
1762 | ||
1763 | if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) { | |
1764 | error_setg(errp, "Can't reopen read-only NBD mount as read/write"); | |
1765 | return -EACCES; | |
1766 | } | |
1767 | return 0; | |
1768 | } | |
1769 | ||
fee091cd LS |
1770 | static void nbd_yank(void *opaque) |
1771 | { | |
1772 | BlockDriverState *bs = opaque; | |
1773 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
1774 | ||
1775 | qatomic_store_release(&s->state, NBD_CLIENT_QUIT); | |
1776 | qio_channel_shutdown(QIO_CHANNEL(s->sioc), QIO_CHANNEL_SHUTDOWN_BOTH, NULL); | |
1777 | } | |
1778 | ||
86f8cdf3 VSO |
1779 | static void nbd_client_close(BlockDriverState *bs) |
1780 | { | |
611ae1d7 | 1781 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1782 | NBDRequest request = { .type = NBD_CMD_DISC }; |
1783 | ||
f7651539 VSO |
1784 | if (s->ioc) { |
1785 | nbd_send_request(s->ioc, &request); | |
1786 | } | |
86f8cdf3 VSO |
1787 | |
1788 | nbd_teardown_connection(bs); | |
1789 | } | |
1790 | ||
fee091cd LS |
1791 | static int nbd_establish_connection(BlockDriverState *bs, |
1792 | SocketAddress *saddr, | |
1793 | Error **errp) | |
86f8cdf3 | 1794 | { |
795d946d | 1795 | ERRP_GUARD(); |
fee091cd | 1796 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 | 1797 | |
fee091cd LS |
1798 | s->sioc = qio_channel_socket_new(); |
1799 | qio_channel_set_name(QIO_CHANNEL(s->sioc), "nbd-client"); | |
86f8cdf3 | 1800 | |
fee091cd | 1801 | qio_channel_socket_connect_sync(s->sioc, saddr, errp); |
795d946d | 1802 | if (*errp) { |
fee091cd LS |
1803 | object_unref(OBJECT(s->sioc)); |
1804 | s->sioc = NULL; | |
1805 | return -1; | |
86f8cdf3 VSO |
1806 | } |
1807 | ||
fee091cd LS |
1808 | yank_register_function(BLOCKDEV_YANK_INSTANCE(bs->node_name), nbd_yank, bs); |
1809 | qio_channel_set_delay(QIO_CHANNEL(s->sioc), false); | |
86f8cdf3 | 1810 | |
fee091cd | 1811 | return 0; |
86f8cdf3 VSO |
1812 | } |
1813 | ||
fee091cd LS |
1814 | /* nbd_client_handshake takes ownership on s->sioc. On failure it's unref'ed. */ |
1815 | static int nbd_client_handshake(BlockDriverState *bs, Error **errp) | |
86f8cdf3 | 1816 | { |
611ae1d7 | 1817 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
a8e2bb6a | 1818 | AioContext *aio_context = bdrv_get_aio_context(bs); |
86f8cdf3 VSO |
1819 | int ret; |
1820 | ||
fa35591b | 1821 | trace_nbd_client_handshake(s->export); |
fee091cd LS |
1822 | qio_channel_set_blocking(QIO_CHANNEL(s->sioc), false, NULL); |
1823 | qio_channel_attach_aio_context(QIO_CHANNEL(s->sioc), aio_context); | |
86f8cdf3 | 1824 | |
611ae1d7 VSO |
1825 | s->info.request_sizes = true; |
1826 | s->info.structured_reply = true; | |
1827 | s->info.base_allocation = true; | |
8f071c9d VSO |
1828 | s->info.x_dirty_bitmap = g_strdup(s->x_dirty_bitmap); |
1829 | s->info.name = g_strdup(s->export ?: ""); | |
fee091cd | 1830 | ret = nbd_receive_negotiate(aio_context, QIO_CHANNEL(s->sioc), s->tlscreds, |
8f071c9d | 1831 | s->hostname, &s->ioc, &s->info, errp); |
611ae1d7 VSO |
1832 | g_free(s->info.x_dirty_bitmap); |
1833 | g_free(s->info.name); | |
86f8cdf3 | 1834 | if (ret < 0) { |
fee091cd LS |
1835 | yank_unregister_function(BLOCKDEV_YANK_INSTANCE(bs->node_name), |
1836 | nbd_yank, bs); | |
1837 | object_unref(OBJECT(s->sioc)); | |
fbeb3e63 | 1838 | s->sioc = NULL; |
86f8cdf3 VSO |
1839 | return ret; |
1840 | } | |
dbc7b014 EB |
1841 | if (s->x_dirty_bitmap) { |
1842 | if (!s->info.base_allocation) { | |
1843 | error_setg(errp, "requested x-dirty-bitmap %s not found", | |
1844 | s->x_dirty_bitmap); | |
1845 | ret = -EINVAL; | |
1846 | goto fail; | |
1847 | } | |
1848 | if (strcmp(s->x_dirty_bitmap, "qemu:allocation-depth") == 0) { | |
1849 | s->alloc_depth = true; | |
1850 | } | |
86f8cdf3 | 1851 | } |
611ae1d7 | 1852 | if (s->info.flags & NBD_FLAG_READ_ONLY) { |
86f8cdf3 VSO |
1853 | ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp); |
1854 | if (ret < 0) { | |
1855 | goto fail; | |
1856 | } | |
1857 | } | |
611ae1d7 | 1858 | if (s->info.flags & NBD_FLAG_SEND_FUA) { |
86f8cdf3 VSO |
1859 | bs->supported_write_flags = BDRV_REQ_FUA; |
1860 | bs->supported_zero_flags |= BDRV_REQ_FUA; | |
1861 | } | |
611ae1d7 | 1862 | if (s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) { |
86f8cdf3 | 1863 | bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP; |
f061656c EB |
1864 | if (s->info.flags & NBD_FLAG_SEND_FAST_ZERO) { |
1865 | bs->supported_zero_flags |= BDRV_REQ_NO_FALLBACK; | |
1866 | } | |
86f8cdf3 VSO |
1867 | } |
1868 | ||
611ae1d7 | 1869 | if (!s->ioc) { |
fee091cd | 1870 | s->ioc = QIO_CHANNEL(s->sioc); |
611ae1d7 | 1871 | object_ref(OBJECT(s->ioc)); |
86f8cdf3 VSO |
1872 | } |
1873 | ||
fa35591b | 1874 | trace_nbd_client_handshake_success(s->export); |
86f8cdf3 VSO |
1875 | |
1876 | return 0; | |
1877 | ||
1878 | fail: | |
1879 | /* | |
a8e2bb6a VSO |
1880 | * We have connected, but must fail for other reasons. |
1881 | * Send NBD_CMD_DISC as a courtesy to the server. | |
86f8cdf3 VSO |
1882 | */ |
1883 | { | |
1884 | NBDRequest request = { .type = NBD_CMD_DISC }; | |
1885 | ||
fee091cd | 1886 | nbd_send_request(s->ioc ?: QIO_CHANNEL(s->sioc), &request); |
86f8cdf3 | 1887 | |
fee091cd LS |
1888 | yank_unregister_function(BLOCKDEV_YANK_INSTANCE(bs->node_name), |
1889 | nbd_yank, bs); | |
1890 | object_unref(OBJECT(s->sioc)); | |
fbeb3e63 | 1891 | s->sioc = NULL; |
86f8cdf3 VSO |
1892 | |
1893 | return ret; | |
1894 | } | |
1895 | } | |
1896 | ||
8f071c9d VSO |
1897 | /* |
1898 | * Parse nbd_open options | |
1899 | */ | |
86f8cdf3 | 1900 | |
f53a1feb | 1901 | static int nbd_parse_uri(const char *filename, QDict *options) |
1d7d2a9d PB |
1902 | { |
1903 | URI *uri; | |
1904 | const char *p; | |
1905 | QueryParams *qp = NULL; | |
1906 | int ret = 0; | |
f53a1feb | 1907 | bool is_unix; |
1d7d2a9d PB |
1908 | |
1909 | uri = uri_parse(filename); | |
1910 | if (!uri) { | |
1911 | return -EINVAL; | |
1912 | } | |
1913 | ||
1914 | /* transport */ | |
f69165a8 | 1915 | if (!g_strcmp0(uri->scheme, "nbd")) { |
f53a1feb | 1916 | is_unix = false; |
f69165a8 | 1917 | } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) { |
f53a1feb | 1918 | is_unix = false; |
f69165a8 | 1919 | } else if (!g_strcmp0(uri->scheme, "nbd+unix")) { |
f53a1feb | 1920 | is_unix = true; |
1d7d2a9d PB |
1921 | } else { |
1922 | ret = -EINVAL; | |
1923 | goto out; | |
1924 | } | |
1925 | ||
2485f22f EB |
1926 | p = uri->path ? uri->path : ""; |
1927 | if (p[0] == '/') { | |
1928 | p++; | |
1929 | } | |
1d7d2a9d | 1930 | if (p[0]) { |
46f5ac20 | 1931 | qdict_put_str(options, "export", p); |
1d7d2a9d PB |
1932 | } |
1933 | ||
1934 | qp = query_params_parse(uri->query); | |
f53a1feb | 1935 | if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) { |
1d7d2a9d PB |
1936 | ret = -EINVAL; |
1937 | goto out; | |
1938 | } | |
1939 | ||
f53a1feb | 1940 | if (is_unix) { |
1d7d2a9d PB |
1941 | /* nbd+unix:///export?socket=path */ |
1942 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
1943 | ret = -EINVAL; | |
1944 | goto out; | |
1945 | } | |
46f5ac20 EB |
1946 | qdict_put_str(options, "server.type", "unix"); |
1947 | qdict_put_str(options, "server.path", qp->p[0].value); | |
1d7d2a9d | 1948 | } else { |
23307908 | 1949 | QString *host; |
f84d431b HR |
1950 | char *port_str; |
1951 | ||
bebbf7fa | 1952 | /* nbd[+tcp]://host[:port]/export */ |
1d7d2a9d PB |
1953 | if (!uri->server) { |
1954 | ret = -EINVAL; | |
1955 | goto out; | |
1956 | } | |
f17c90be | 1957 | |
23307908 JT |
1958 | /* strip braces from literal IPv6 address */ |
1959 | if (uri->server[0] == '[') { | |
1960 | host = qstring_from_substr(uri->server, 1, | |
ba891d68 | 1961 | strlen(uri->server) - 1); |
23307908 JT |
1962 | } else { |
1963 | host = qstring_from_str(uri->server); | |
1964 | } | |
1965 | ||
46f5ac20 | 1966 | qdict_put_str(options, "server.type", "inet"); |
9445673e | 1967 | qdict_put(options, "server.host", host); |
f84d431b HR |
1968 | |
1969 | port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT); | |
46f5ac20 | 1970 | qdict_put_str(options, "server.port", port_str); |
f84d431b | 1971 | g_free(port_str); |
1d7d2a9d PB |
1972 | } |
1973 | ||
1974 | out: | |
1975 | if (qp) { | |
1976 | query_params_free(qp); | |
1977 | } | |
1978 | uri_free(uri); | |
1979 | return ret; | |
1980 | } | |
1981 | ||
48c38e0b HR |
1982 | static bool nbd_has_filename_options_conflict(QDict *options, Error **errp) |
1983 | { | |
1984 | const QDictEntry *e; | |
1985 | ||
1986 | for (e = qdict_first(options); e; e = qdict_next(options, e)) { | |
1987 | if (!strcmp(e->key, "host") || | |
1988 | !strcmp(e->key, "port") || | |
1989 | !strcmp(e->key, "path") || | |
491d6c7c HR |
1990 | !strcmp(e->key, "export") || |
1991 | strstart(e->key, "server.", NULL)) | |
48c38e0b HR |
1992 | { |
1993 | error_setg(errp, "Option '%s' cannot be used with a file name", | |
1994 | e->key); | |
1995 | return true; | |
1996 | } | |
1997 | } | |
1998 | ||
1999 | return false; | |
2000 | } | |
2001 | ||
6963a30d KW |
2002 | static void nbd_parse_filename(const char *filename, QDict *options, |
2003 | Error **errp) | |
75818250 | 2004 | { |
df18c04e | 2005 | g_autofree char *file = NULL; |
33897dc7 NT |
2006 | char *export_name; |
2007 | const char *host_spec; | |
75818250 | 2008 | const char *unixpath; |
75818250 | 2009 | |
48c38e0b | 2010 | if (nbd_has_filename_options_conflict(options, errp)) { |
681e7ad0 KW |
2011 | return; |
2012 | } | |
2013 | ||
1d7d2a9d | 2014 | if (strstr(filename, "://")) { |
6963a30d KW |
2015 | int ret = nbd_parse_uri(filename, options); |
2016 | if (ret < 0) { | |
2017 | error_setg(errp, "No valid URL specified"); | |
2018 | } | |
2019 | return; | |
1d7d2a9d PB |
2020 | } |
2021 | ||
7267c094 | 2022 | file = g_strdup(filename); |
1d45f8b5 | 2023 | |
33897dc7 NT |
2024 | export_name = strstr(file, EN_OPTSTR); |
2025 | if (export_name) { | |
2026 | if (export_name[strlen(EN_OPTSTR)] == 0) { | |
df18c04e | 2027 | return; |
1d45f8b5 | 2028 | } |
33897dc7 NT |
2029 | export_name[0] = 0; /* truncate 'file' */ |
2030 | export_name += strlen(EN_OPTSTR); | |
f53a1feb | 2031 | |
46f5ac20 | 2032 | qdict_put_str(options, "export", export_name); |
1d45f8b5 LV |
2033 | } |
2034 | ||
33897dc7 NT |
2035 | /* extract the host_spec - fail if it's not nbd:... */ |
2036 | if (!strstart(file, "nbd:", &host_spec)) { | |
6963a30d | 2037 | error_setg(errp, "File name string for NBD must start with 'nbd:'"); |
df18c04e | 2038 | return; |
1d45f8b5 | 2039 | } |
75818250 | 2040 | |
f53a1feb | 2041 | if (!*host_spec) { |
df18c04e | 2042 | return; |
f53a1feb KW |
2043 | } |
2044 | ||
33897dc7 NT |
2045 | /* are we a UNIX or TCP socket? */ |
2046 | if (strstart(host_spec, "unix:", &unixpath)) { | |
46f5ac20 EB |
2047 | qdict_put_str(options, "server.type", "unix"); |
2048 | qdict_put_str(options, "server.path", unixpath); | |
75818250 | 2049 | } else { |
0785bd7a | 2050 | InetSocketAddress *addr = g_new(InetSocketAddress, 1); |
f53a1feb | 2051 | |
0785bd7a MA |
2052 | if (inet_parse(addr, host_spec, errp)) { |
2053 | goto out_inet; | |
f17c90be | 2054 | } |
75818250 | 2055 | |
46f5ac20 EB |
2056 | qdict_put_str(options, "server.type", "inet"); |
2057 | qdict_put_str(options, "server.host", addr->host); | |
2058 | qdict_put_str(options, "server.port", addr->port); | |
0785bd7a | 2059 | out_inet: |
f53a1feb KW |
2060 | qapi_free_InetSocketAddress(addr); |
2061 | } | |
f53a1feb KW |
2062 | } |
2063 | ||
491d6c7c HR |
2064 | static bool nbd_process_legacy_socket_options(QDict *output_options, |
2065 | QemuOpts *legacy_opts, | |
2066 | Error **errp) | |
f53a1feb | 2067 | { |
491d6c7c HR |
2068 | const char *path = qemu_opt_get(legacy_opts, "path"); |
2069 | const char *host = qemu_opt_get(legacy_opts, "host"); | |
2070 | const char *port = qemu_opt_get(legacy_opts, "port"); | |
2071 | const QDictEntry *e; | |
f53a1feb | 2072 | |
491d6c7c HR |
2073 | if (!path && !host && !port) { |
2074 | return true; | |
2075 | } | |
03504d05 | 2076 | |
491d6c7c HR |
2077 | for (e = qdict_first(output_options); e; e = qdict_next(output_options, e)) |
2078 | { | |
2079 | if (strstart(e->key, "server.", NULL)) { | |
2080 | error_setg(errp, "Cannot use 'server' and path/host/port at the " | |
2081 | "same time"); | |
2082 | return false; | |
681e7ad0 | 2083 | } |
33897dc7 | 2084 | } |
491d6c7c HR |
2085 | |
2086 | if (path && host) { | |
2087 | error_setg(errp, "path and host may not be used at the same time"); | |
2088 | return false; | |
2089 | } else if (path) { | |
2090 | if (port) { | |
2091 | error_setg(errp, "port may not be used without host"); | |
2092 | return false; | |
2093 | } | |
2094 | ||
46f5ac20 EB |
2095 | qdict_put_str(output_options, "server.type", "unix"); |
2096 | qdict_put_str(output_options, "server.path", path); | |
491d6c7c | 2097 | } else if (host) { |
46f5ac20 EB |
2098 | qdict_put_str(output_options, "server.type", "inet"); |
2099 | qdict_put_str(output_options, "server.host", host); | |
2100 | qdict_put_str(output_options, "server.port", | |
2101 | port ?: stringify(NBD_DEFAULT_PORT)); | |
442045cb | 2102 | } |
f53a1feb | 2103 | |
491d6c7c HR |
2104 | return true; |
2105 | } | |
f53a1feb | 2106 | |
62cf396b MA |
2107 | static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options, |
2108 | Error **errp) | |
491d6c7c | 2109 | { |
62cf396b | 2110 | SocketAddress *saddr = NULL; |
491d6c7c | 2111 | QDict *addr = NULL; |
491d6c7c | 2112 | Visitor *iv = NULL; |
491d6c7c HR |
2113 | |
2114 | qdict_extract_subqdict(options, &addr, "server."); | |
2115 | if (!qdict_size(addr)) { | |
2116 | error_setg(errp, "NBD server address missing"); | |
2117 | goto done; | |
f53a1feb KW |
2118 | } |
2119 | ||
af91062e MA |
2120 | iv = qobject_input_visitor_new_flat_confused(addr, errp); |
2121 | if (!iv) { | |
491d6c7c HR |
2122 | goto done; |
2123 | } | |
bebbf7fa | 2124 | |
af175e85 | 2125 | if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) { |
491d6c7c HR |
2126 | goto done; |
2127 | } | |
7a5ed437 | 2128 | |
491d6c7c | 2129 | done: |
cb3e7f08 | 2130 | qobject_unref(addr); |
491d6c7c | 2131 | visit_free(iv); |
7a5ed437 | 2132 | return saddr; |
33897dc7 | 2133 | } |
1d45f8b5 | 2134 | |
75822a12 DB |
2135 | static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp) |
2136 | { | |
2137 | Object *obj; | |
2138 | QCryptoTLSCreds *creds; | |
2139 | ||
2140 | obj = object_resolve_path_component( | |
2141 | object_get_objects_root(), id); | |
2142 | if (!obj) { | |
2143 | error_setg(errp, "No TLS credentials with id '%s'", | |
2144 | id); | |
2145 | return NULL; | |
2146 | } | |
2147 | creds = (QCryptoTLSCreds *) | |
2148 | object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS); | |
2149 | if (!creds) { | |
2150 | error_setg(errp, "Object with id '%s' is not TLS credentials", | |
2151 | id); | |
2152 | return NULL; | |
2153 | } | |
2154 | ||
2155 | if (creds->endpoint != QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT) { | |
2156 | error_setg(errp, | |
2157 | "Expecting TLS credentials with a client endpoint"); | |
2158 | return NULL; | |
2159 | } | |
2160 | object_ref(obj); | |
2161 | return creds; | |
2162 | } | |
2163 | ||
2164 | ||
7ccc44fd HR |
2165 | static QemuOptsList nbd_runtime_opts = { |
2166 | .name = "nbd", | |
2167 | .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head), | |
2168 | .desc = { | |
2169 | { | |
2170 | .name = "host", | |
2171 | .type = QEMU_OPT_STRING, | |
2172 | .help = "TCP host to connect to", | |
2173 | }, | |
2174 | { | |
2175 | .name = "port", | |
2176 | .type = QEMU_OPT_STRING, | |
2177 | .help = "TCP port to connect to", | |
2178 | }, | |
2179 | { | |
2180 | .name = "path", | |
2181 | .type = QEMU_OPT_STRING, | |
2182 | .help = "Unix socket path to connect to", | |
2183 | }, | |
2184 | { | |
2185 | .name = "export", | |
2186 | .type = QEMU_OPT_STRING, | |
2187 | .help = "Name of the NBD export to open", | |
2188 | }, | |
2189 | { | |
2190 | .name = "tls-creds", | |
2191 | .type = QEMU_OPT_STRING, | |
2192 | .help = "ID of the TLS credentials to use", | |
2193 | }, | |
216ee365 EB |
2194 | { |
2195 | .name = "x-dirty-bitmap", | |
2196 | .type = QEMU_OPT_STRING, | |
2197 | .help = "experimental: expose named dirty bitmap in place of " | |
2198 | "block status", | |
2199 | }, | |
b172ae2e VSO |
2200 | { |
2201 | .name = "reconnect-delay", | |
2202 | .type = QEMU_OPT_NUMBER, | |
2203 | .help = "On an unexpected disconnect, the nbd client tries to " | |
2204 | "connect again until succeeding or encountering a serious " | |
2205 | "error. During the first @reconnect-delay seconds, all " | |
2206 | "requests are paused and will be rerun on a successful " | |
2207 | "reconnect. After that time, any delayed requests and all " | |
2208 | "future requests before a successful reconnect will " | |
2209 | "immediately fail. Default 0", | |
2210 | }, | |
c4365735 | 2211 | { /* end of list */ } |
7ccc44fd HR |
2212 | }, |
2213 | }; | |
2214 | ||
8f071c9d VSO |
2215 | static int nbd_process_options(BlockDriverState *bs, QDict *options, |
2216 | Error **errp) | |
33897dc7 NT |
2217 | { |
2218 | BDRVNBDState *s = bs->opaque; | |
8f071c9d | 2219 | QemuOpts *opts; |
75822a12 | 2220 | int ret = -EINVAL; |
ae255e52 | 2221 | |
7ccc44fd | 2222 | opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 2223 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
7ccc44fd HR |
2224 | goto error; |
2225 | } | |
2226 | ||
62cf396b | 2227 | /* Translate @host, @port, and @path to a SocketAddress */ |
491d6c7c HR |
2228 | if (!nbd_process_legacy_socket_options(options, opts, errp)) { |
2229 | goto error; | |
2230 | } | |
2231 | ||
33897dc7 | 2232 | /* Pop the config into our state object. Exit if invalid. */ |
491d6c7c HR |
2233 | s->saddr = nbd_config(s, options, errp); |
2234 | if (!s->saddr) { | |
75822a12 DB |
2235 | goto error; |
2236 | } | |
2237 | ||
491d6c7c | 2238 | s->export = g_strdup(qemu_opt_get(opts, "export")); |
93676c88 EB |
2239 | if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) { |
2240 | error_setg(errp, "export name too long to send to server"); | |
2241 | goto error; | |
2242 | } | |
491d6c7c | 2243 | |
03504d05 HR |
2244 | s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds")); |
2245 | if (s->tlscredsid) { | |
8f071c9d VSO |
2246 | s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp); |
2247 | if (!s->tlscreds) { | |
75822a12 DB |
2248 | goto error; |
2249 | } | |
2250 | ||
ca0b64e5 | 2251 | /* TODO SOCKET_ADDRESS_KIND_FD where fd has AF_INET or AF_INET6 */ |
62cf396b | 2252 | if (s->saddr->type != SOCKET_ADDRESS_TYPE_INET) { |
75822a12 DB |
2253 | error_setg(errp, "TLS only supported over IP sockets"); |
2254 | goto error; | |
2255 | } | |
8f071c9d | 2256 | s->hostname = s->saddr->u.inet.host; |
33897dc7 NT |
2257 | } |
2258 | ||
8f071c9d | 2259 | s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap")); |
93676c88 EB |
2260 | if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) { |
2261 | error_setg(errp, "x-dirty-bitmap query too long to send to server"); | |
2262 | goto error; | |
2263 | } | |
2264 | ||
8f071c9d VSO |
2265 | s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0); |
2266 | ||
2267 | ret = 0; | |
d42f78e9 | 2268 | |
75822a12 | 2269 | error: |
03504d05 | 2270 | if (ret < 0) { |
7f493662 | 2271 | nbd_clear_bdrvstate(s); |
03504d05 | 2272 | } |
7ccc44fd | 2273 | qemu_opts_del(opts); |
75822a12 | 2274 | return ret; |
e183ef75 PB |
2275 | } |
2276 | ||
8f071c9d VSO |
2277 | static int nbd_open(BlockDriverState *bs, QDict *options, int flags, |
2278 | Error **errp) | |
2279 | { | |
2280 | int ret; | |
2281 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
2282 | ||
2283 | ret = nbd_process_options(bs, options, errp); | |
2284 | if (ret < 0) { | |
2285 | return ret; | |
2286 | } | |
2287 | ||
2288 | s->bs = bs; | |
2289 | qemu_co_mutex_init(&s->send_mutex); | |
2290 | qemu_co_queue_init(&s->free_sema); | |
2291 | ||
fee091cd LS |
2292 | if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) { |
2293 | return -EEXIST; | |
2294 | } | |
2295 | ||
fa35591b VSO |
2296 | /* |
2297 | * establish TCP connection, return error if it fails | |
2298 | * TODO: Configurable retry-until-timeout behaviour. | |
2299 | */ | |
fee091cd LS |
2300 | if (nbd_establish_connection(bs, s->saddr, errp) < 0) { |
2301 | yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name)); | |
fa35591b VSO |
2302 | return -ECONNREFUSED; |
2303 | } | |
2304 | ||
fee091cd | 2305 | ret = nbd_client_handshake(bs, errp); |
8f071c9d | 2306 | if (ret < 0) { |
fee091cd | 2307 | yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name)); |
8198cf5e | 2308 | nbd_clear_bdrvstate(s); |
8f071c9d VSO |
2309 | return ret; |
2310 | } | |
2311 | /* successfully connected */ | |
2312 | s->state = NBD_CLIENT_CONNECTED; | |
2313 | ||
1dc4718d VSO |
2314 | nbd_init_connect_thread(s); |
2315 | ||
8f071c9d VSO |
2316 | s->connection_co = qemu_coroutine_create(nbd_connection_entry, s); |
2317 | bdrv_inc_in_flight(bs); | |
2318 | aio_co_schedule(bdrv_get_aio_context(bs), s->connection_co); | |
2319 | ||
2320 | return 0; | |
2321 | } | |
2322 | ||
1486d04a PB |
2323 | static int nbd_co_flush(BlockDriverState *bs) |
2324 | { | |
f53a829b | 2325 | return nbd_client_co_flush(bs); |
1486d04a PB |
2326 | } |
2327 | ||
fa21e6fa DL |
2328 | static void nbd_refresh_limits(BlockDriverState *bs, Error **errp) |
2329 | { | |
611ae1d7 | 2330 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
fd8d372d | 2331 | uint32_t min = s->info.min_block; |
081dd1fe EB |
2332 | uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block); |
2333 | ||
7da537f7 EB |
2334 | /* |
2335 | * If the server did not advertise an alignment: | |
2336 | * - a size that is not sector-aligned implies that an alignment | |
2337 | * of 1 can be used to access those tail bytes | |
2338 | * - advertisement of block status requires an alignment of 1, so | |
2339 | * that we don't violate block layer constraints that block | |
2340 | * status is always aligned (as we can't control whether the | |
2341 | * server will report sub-sector extents, such as a hole at EOF | |
2342 | * on an unaligned POSIX file) | |
2343 | * - otherwise, assume the server is so old that we are safer avoiding | |
2344 | * sub-sector requests | |
2345 | */ | |
2346 | if (!min) { | |
2347 | min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) || | |
2348 | s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE; | |
2349 | } | |
2350 | ||
2351 | bs->bl.request_alignment = min; | |
714eb0db | 2352 | bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min); |
081dd1fe EB |
2353 | bs->bl.max_pwrite_zeroes = max; |
2354 | bs->bl.max_transfer = max; | |
2355 | ||
2356 | if (s->info.opt_block && | |
2357 | s->info.opt_block > bs->bl.opt_transfer) { | |
2358 | bs->bl.opt_transfer = s->info.opt_block; | |
2359 | } | |
fa21e6fa DL |
2360 | } |
2361 | ||
75818250 TS |
2362 | static void nbd_close(BlockDriverState *bs) |
2363 | { | |
03504d05 HR |
2364 | BDRVNBDState *s = bs->opaque; |
2365 | ||
f53a829b | 2366 | nbd_client_close(bs); |
fee091cd | 2367 | yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name)); |
7f493662 | 2368 | nbd_clear_bdrvstate(s); |
75818250 TS |
2369 | } |
2370 | ||
a2b333c0 NS |
2371 | /* |
2372 | * NBD cannot truncate, but if the caller asks to truncate to the same size, or | |
2373 | * to a smaller size with exact=false, there is no reason to fail the | |
2374 | * operation. | |
2375 | * | |
2376 | * Preallocation mode is ignored since it does not seems useful to fail when | |
2377 | * we never change anything. | |
2378 | */ | |
2379 | static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset, | |
2380 | bool exact, PreallocMode prealloc, | |
2381 | BdrvRequestFlags flags, Error **errp) | |
2382 | { | |
2383 | BDRVNBDState *s = bs->opaque; | |
2384 | ||
2385 | if (offset != s->info.size && exact) { | |
2386 | error_setg(errp, "Cannot resize NBD nodes"); | |
2387 | return -ENOTSUP; | |
2388 | } | |
2389 | ||
2390 | if (offset > s->info.size) { | |
2391 | error_setg(errp, "Cannot grow NBD nodes"); | |
2392 | return -EINVAL; | |
2393 | } | |
2394 | ||
2395 | return 0; | |
2396 | } | |
2397 | ||
75818250 TS |
2398 | static int64_t nbd_getlength(BlockDriverState *bs) |
2399 | { | |
2400 | BDRVNBDState *s = bs->opaque; | |
2401 | ||
611ae1d7 | 2402 | return s->info.size; |
75818250 TS |
2403 | } |
2404 | ||
998b3a1e | 2405 | static void nbd_refresh_filename(BlockDriverState *bs) |
2019d68b | 2406 | { |
03504d05 | 2407 | BDRVNBDState *s = bs->opaque; |
491d6c7c | 2408 | const char *host = NULL, *port = NULL, *path = NULL; |
5c86bdf1 | 2409 | size_t len = 0; |
491d6c7c | 2410 | |
62cf396b | 2411 | if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) { |
9445673e | 2412 | const InetSocketAddress *inet = &s->saddr->u.inet; |
491d6c7c HR |
2413 | if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) { |
2414 | host = inet->host; | |
2415 | port = inet->port; | |
2416 | } | |
62cf396b | 2417 | } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) { |
9445673e MA |
2418 | path = s->saddr->u.q_unix.path; |
2419 | } /* else can't represent as pseudo-filename */ | |
2019d68b | 2420 | |
491d6c7c | 2421 | if (path && s->export) { |
5c86bdf1 EB |
2422 | len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), |
2423 | "nbd+unix:///%s?socket=%s", s->export, path); | |
491d6c7c | 2424 | } else if (path && !s->export) { |
5c86bdf1 EB |
2425 | len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), |
2426 | "nbd+unix://?socket=%s", path); | |
491d6c7c | 2427 | } else if (host && s->export) { |
5c86bdf1 EB |
2428 | len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), |
2429 | "nbd://%s:%s/%s", host, port, s->export); | |
491d6c7c | 2430 | } else if (host && !s->export) { |
5c86bdf1 EB |
2431 | len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), |
2432 | "nbd://%s:%s", host, port); | |
2433 | } | |
00d69986 | 2434 | if (len >= sizeof(bs->exact_filename)) { |
5c86bdf1 EB |
2435 | /* Name is too long to represent exactly, so leave it empty. */ |
2436 | bs->exact_filename[0] = '\0'; | |
ec0de768 | 2437 | } |
2019d68b HR |
2438 | } |
2439 | ||
8a6239c0 HR |
2440 | static char *nbd_dirname(BlockDriverState *bs, Error **errp) |
2441 | { | |
2442 | /* The generic bdrv_dirname() implementation is able to work out some | |
2443 | * directory name for NBD nodes, but that would be wrong. So far there is no | |
2444 | * specification for how "export paths" would work, so NBD does not have | |
2445 | * directory names. */ | |
2446 | error_setg(errp, "Cannot generate a base directory for NBD nodes"); | |
2447 | return NULL; | |
2448 | } | |
2449 | ||
2654267c HR |
2450 | static const char *const nbd_strong_runtime_opts[] = { |
2451 | "path", | |
2452 | "host", | |
2453 | "port", | |
2454 | "export", | |
2455 | "tls-creds", | |
2456 | "server.", | |
2457 | ||
2458 | NULL | |
2459 | }; | |
2460 | ||
5efa9d5a | 2461 | static BlockDriver bdrv_nbd = { |
69447cd8 SH |
2462 | .format_name = "nbd", |
2463 | .protocol_name = "nbd", | |
2464 | .instance_size = sizeof(BDRVNBDState), | |
2465 | .bdrv_parse_filename = nbd_parse_filename, | |
5a5e7f8c ML |
2466 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
2467 | .create_opts = &bdrv_create_opts_simple, | |
69447cd8 | 2468 | .bdrv_file_open = nbd_open, |
e99754b4 | 2469 | .bdrv_reopen_prepare = nbd_client_reopen_prepare, |
70c4fb26 EB |
2470 | .bdrv_co_preadv = nbd_client_co_preadv, |
2471 | .bdrv_co_pwritev = nbd_client_co_pwritev, | |
fa778fff | 2472 | .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, |
69447cd8 SH |
2473 | .bdrv_close = nbd_close, |
2474 | .bdrv_co_flush_to_os = nbd_co_flush, | |
447e57c3 | 2475 | .bdrv_co_pdiscard = nbd_client_co_pdiscard, |
fa21e6fa | 2476 | .bdrv_refresh_limits = nbd_refresh_limits, |
a2b333c0 | 2477 | .bdrv_co_truncate = nbd_co_truncate, |
69447cd8 | 2478 | .bdrv_getlength = nbd_getlength, |
86f8cdf3 VSO |
2479 | .bdrv_detach_aio_context = nbd_client_detach_aio_context, |
2480 | .bdrv_attach_aio_context = nbd_client_attach_aio_context, | |
f7651539 VSO |
2481 | .bdrv_co_drain_begin = nbd_client_co_drain_begin, |
2482 | .bdrv_co_drain_end = nbd_client_co_drain_end, | |
2019d68b | 2483 | .bdrv_refresh_filename = nbd_refresh_filename, |
78a33ab5 | 2484 | .bdrv_co_block_status = nbd_client_co_block_status, |
8a6239c0 | 2485 | .bdrv_dirname = nbd_dirname, |
2654267c | 2486 | .strong_runtime_opts = nbd_strong_runtime_opts, |
1d7d2a9d PB |
2487 | }; |
2488 | ||
2489 | static BlockDriver bdrv_nbd_tcp = { | |
69447cd8 SH |
2490 | .format_name = "nbd", |
2491 | .protocol_name = "nbd+tcp", | |
2492 | .instance_size = sizeof(BDRVNBDState), | |
2493 | .bdrv_parse_filename = nbd_parse_filename, | |
5a5e7f8c ML |
2494 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
2495 | .create_opts = &bdrv_create_opts_simple, | |
69447cd8 | 2496 | .bdrv_file_open = nbd_open, |
e99754b4 | 2497 | .bdrv_reopen_prepare = nbd_client_reopen_prepare, |
70c4fb26 EB |
2498 | .bdrv_co_preadv = nbd_client_co_preadv, |
2499 | .bdrv_co_pwritev = nbd_client_co_pwritev, | |
fa778fff | 2500 | .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, |
69447cd8 SH |
2501 | .bdrv_close = nbd_close, |
2502 | .bdrv_co_flush_to_os = nbd_co_flush, | |
447e57c3 | 2503 | .bdrv_co_pdiscard = nbd_client_co_pdiscard, |
fa21e6fa | 2504 | .bdrv_refresh_limits = nbd_refresh_limits, |
a2b333c0 | 2505 | .bdrv_co_truncate = nbd_co_truncate, |
69447cd8 | 2506 | .bdrv_getlength = nbd_getlength, |
86f8cdf3 VSO |
2507 | .bdrv_detach_aio_context = nbd_client_detach_aio_context, |
2508 | .bdrv_attach_aio_context = nbd_client_attach_aio_context, | |
f7651539 VSO |
2509 | .bdrv_co_drain_begin = nbd_client_co_drain_begin, |
2510 | .bdrv_co_drain_end = nbd_client_co_drain_end, | |
2019d68b | 2511 | .bdrv_refresh_filename = nbd_refresh_filename, |
78a33ab5 | 2512 | .bdrv_co_block_status = nbd_client_co_block_status, |
8a6239c0 | 2513 | .bdrv_dirname = nbd_dirname, |
2654267c | 2514 | .strong_runtime_opts = nbd_strong_runtime_opts, |
1d7d2a9d PB |
2515 | }; |
2516 | ||
2517 | static BlockDriver bdrv_nbd_unix = { | |
69447cd8 SH |
2518 | .format_name = "nbd", |
2519 | .protocol_name = "nbd+unix", | |
2520 | .instance_size = sizeof(BDRVNBDState), | |
2521 | .bdrv_parse_filename = nbd_parse_filename, | |
5a5e7f8c ML |
2522 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
2523 | .create_opts = &bdrv_create_opts_simple, | |
69447cd8 | 2524 | .bdrv_file_open = nbd_open, |
e99754b4 | 2525 | .bdrv_reopen_prepare = nbd_client_reopen_prepare, |
70c4fb26 EB |
2526 | .bdrv_co_preadv = nbd_client_co_preadv, |
2527 | .bdrv_co_pwritev = nbd_client_co_pwritev, | |
fa778fff | 2528 | .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, |
69447cd8 SH |
2529 | .bdrv_close = nbd_close, |
2530 | .bdrv_co_flush_to_os = nbd_co_flush, | |
447e57c3 | 2531 | .bdrv_co_pdiscard = nbd_client_co_pdiscard, |
fa21e6fa | 2532 | .bdrv_refresh_limits = nbd_refresh_limits, |
a2b333c0 | 2533 | .bdrv_co_truncate = nbd_co_truncate, |
69447cd8 | 2534 | .bdrv_getlength = nbd_getlength, |
86f8cdf3 VSO |
2535 | .bdrv_detach_aio_context = nbd_client_detach_aio_context, |
2536 | .bdrv_attach_aio_context = nbd_client_attach_aio_context, | |
f7651539 VSO |
2537 | .bdrv_co_drain_begin = nbd_client_co_drain_begin, |
2538 | .bdrv_co_drain_end = nbd_client_co_drain_end, | |
2019d68b | 2539 | .bdrv_refresh_filename = nbd_refresh_filename, |
78a33ab5 | 2540 | .bdrv_co_block_status = nbd_client_co_block_status, |
8a6239c0 | 2541 | .bdrv_dirname = nbd_dirname, |
2654267c | 2542 | .strong_runtime_opts = nbd_strong_runtime_opts, |
75818250 | 2543 | }; |
5efa9d5a AL |
2544 | |
2545 | static void bdrv_nbd_init(void) | |
2546 | { | |
2547 | bdrv_register(&bdrv_nbd); | |
1d7d2a9d PB |
2548 | bdrv_register(&bdrv_nbd_tcp); |
2549 | bdrv_register(&bdrv_nbd_unix); | |
5efa9d5a AL |
2550 | } |
2551 | ||
2552 | block_init(bdrv_nbd_init); |