]>
Commit | Line | Data |
---|---|---|
75818250 | 1 | /* |
22efd811 | 2 | * QEMU Block driver for NBD |
75818250 | 3 | * |
f7651539 | 4 | * Copyright (c) 2019 Virtuozzo International GmbH. |
22efd811 | 5 | * Copyright Red Hat |
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" |
86f8cdf3 | 38 | |
9af23989 | 39 | #include "qapi/qapi-visit-sockets.h" |
2019d68b | 40 | #include "qapi/qmp/qstring.h" |
1dc4718d | 41 | #include "qapi/clone-visitor.h" |
86f8cdf3 VSO |
42 | |
43 | #include "block/qdict.h" | |
44 | #include "block/nbd.h" | |
45 | #include "block/block_int.h" | |
a71d597b | 46 | #include "block/coroutines.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 | ||
8cb98a72 EB |
53 | #define COOKIE_TO_INDEX(cookie) ((cookie) - 1) |
54 | #define INDEX_TO_COOKIE(index) ((index) + 1) | |
86f8cdf3 VSO |
55 | |
56 | typedef struct { | |
57 | Coroutine *coroutine; | |
58 | uint64_t offset; /* original offset of the request */ | |
4ddb5d2f | 59 | bool receiving; /* sleeping in the yield in nbd_receive_replies */ |
86f8cdf3 VSO |
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 | ||
611ae1d7 | 69 | typedef struct BDRVNBDState { |
95a078ea | 70 | QIOChannel *ioc; /* The current I/O channel */ |
86f8cdf3 VSO |
71 | NBDExportInfo info; |
72 | ||
ee19d953 | 73 | /* |
dba5156c | 74 | * Protects state, free_sema, in_flight, requests[].coroutine, |
ee19d953 PB |
75 | * reconnect_delay_timer. |
76 | */ | |
77 | QemuMutex requests_lock; | |
dba5156c | 78 | NBDClientState state; |
86f8cdf3 | 79 | CoQueue free_sema; |
1b8f7776 | 80 | unsigned in_flight; |
ee19d953 PB |
81 | NBDClientRequest requests[MAX_NBD_REQUESTS]; |
82 | QEMUTimer *reconnect_delay_timer; | |
4ddb5d2f | 83 | |
620c5cb5 | 84 | /* Protects sending data on the socket. */ |
ee19d953 | 85 | CoMutex send_mutex; |
620c5cb5 PB |
86 | |
87 | /* | |
88 | * Protects receiving reply headers from the socket, as well as the | |
89 | * fields reply and requests[].receiving | |
90 | */ | |
4ddb5d2f | 91 | CoMutex receive_mutex; |
620c5cb5 | 92 | NBDReply reply; |
86f8cdf3 | 93 | |
be16b8bf | 94 | QEMUTimer *open_timer; |
46f56631 | 95 | |
86f8cdf3 | 96 | BlockDriverState *bs; |
03504d05 | 97 | |
8f071c9d VSO |
98 | /* Connection parameters */ |
99 | uint32_t reconnect_delay; | |
be16b8bf | 100 | uint32_t open_timeout; |
62cf396b | 101 | SocketAddress *saddr; |
a0cd6d29 DB |
102 | char *export; |
103 | char *tlscredsid; | |
8f071c9d | 104 | QCryptoTLSCreds *tlscreds; |
a0cd6d29 | 105 | char *tlshostname; |
8f071c9d | 106 | char *x_dirty_bitmap; |
dbc7b014 | 107 | bool alloc_depth; |
1dc4718d | 108 | |
90ddc64f | 109 | NBDClientConnection *conn; |
75818250 TS |
110 | } BDRVNBDState; |
111 | ||
fee091cd | 112 | static void nbd_yank(void *opaque); |
f7651539 | 113 | |
bbba1c37 | 114 | static void nbd_clear_bdrvstate(BlockDriverState *bs) |
7f493662 | 115 | { |
bbba1c37 | 116 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
e8b35bf5 | 117 | |
248d4701 VSO |
118 | nbd_client_connection_release(s->conn); |
119 | s->conn = NULL; | |
bbba1c37 VSO |
120 | |
121 | yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name)); | |
122 | ||
8a39c381 HR |
123 | /* Must not leave timers behind that would access freed data */ |
124 | assert(!s->reconnect_delay_timer); | |
125 | assert(!s->open_timer); | |
126 | ||
7f493662 PN |
127 | object_unref(OBJECT(s->tlscreds)); |
128 | qapi_free_SocketAddress(s->saddr); | |
129 | s->saddr = NULL; | |
130 | g_free(s->export); | |
131 | s->export = NULL; | |
132 | g_free(s->tlscredsid); | |
133 | s->tlscredsid = NULL; | |
a0cd6d29 DB |
134 | g_free(s->tlshostname); |
135 | s->tlshostname = NULL; | |
7f493662 PN |
136 | g_free(s->x_dirty_bitmap); |
137 | s->x_dirty_bitmap = NULL; | |
138 | } | |
139 | ||
a80a9a1c | 140 | /* Called with s->receive_mutex taken. */ |
0c43c6fc | 141 | static bool coroutine_fn nbd_recv_coroutine_wake_one(NBDClientRequest *req) |
04a953b2 VSO |
142 | { |
143 | if (req->receiving) { | |
144 | req->receiving = false; | |
145 | aio_co_wake(req->coroutine); | |
146 | return true; | |
147 | } | |
148 | ||
149 | return false; | |
150 | } | |
151 | ||
a80a9a1c | 152 | static void coroutine_fn nbd_recv_coroutines_wake(BDRVNBDState *s) |
3bc0bd1f VSO |
153 | { |
154 | int i; | |
155 | ||
a80a9a1c | 156 | QEMU_LOCK_GUARD(&s->receive_mutex); |
3bc0bd1f | 157 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { |
a80a9a1c | 158 | if (nbd_recv_coroutine_wake_one(&s->requests[i])) { |
04a953b2 | 159 | return; |
3bc0bd1f VSO |
160 | } |
161 | } | |
162 | } | |
163 | ||
dba5156c PB |
164 | /* Called with s->requests_lock held. */ |
165 | static void coroutine_fn nbd_channel_error_locked(BDRVNBDState *s, int ret) | |
a34b1e5e | 166 | { |
dba5156c | 167 | if (s->state == NBD_CLIENT_CONNECTED) { |
cb116da7 VSO |
168 | qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); |
169 | } | |
170 | ||
f7651539 | 171 | if (ret == -EIO) { |
dba5156c | 172 | if (s->state == NBD_CLIENT_CONNECTED) { |
f7651539 VSO |
173 | s->state = s->reconnect_delay ? NBD_CLIENT_CONNECTING_WAIT : |
174 | NBD_CLIENT_CONNECTING_NOWAIT; | |
175 | } | |
176 | } else { | |
f7651539 VSO |
177 | s->state = NBD_CLIENT_QUIT; |
178 | } | |
a34b1e5e VSO |
179 | } |
180 | ||
dba5156c PB |
181 | static void coroutine_fn nbd_channel_error(BDRVNBDState *s, int ret) |
182 | { | |
183 | QEMU_LOCK_GUARD(&s->requests_lock); | |
184 | nbd_channel_error_locked(s, ret); | |
185 | } | |
186 | ||
46f56631 VSO |
187 | static void reconnect_delay_timer_del(BDRVNBDState *s) |
188 | { | |
189 | if (s->reconnect_delay_timer) { | |
46f56631 VSO |
190 | timer_free(s->reconnect_delay_timer); |
191 | s->reconnect_delay_timer = NULL; | |
192 | } | |
193 | } | |
194 | ||
195 | static void reconnect_delay_timer_cb(void *opaque) | |
196 | { | |
197 | BDRVNBDState *s = opaque; | |
198 | ||
dba5156c PB |
199 | reconnect_delay_timer_del(s); |
200 | WITH_QEMU_LOCK_GUARD(&s->requests_lock) { | |
201 | if (s->state != NBD_CLIENT_CONNECTING_WAIT) { | |
202 | return; | |
203 | } | |
46f56631 | 204 | s->state = NBD_CLIENT_CONNECTING_NOWAIT; |
46f56631 | 205 | } |
dba5156c | 206 | nbd_co_establish_connection_cancel(s->conn); |
46f56631 VSO |
207 | } |
208 | ||
209 | static void reconnect_delay_timer_init(BDRVNBDState *s, uint64_t expire_time_ns) | |
210 | { | |
46f56631 VSO |
211 | assert(!s->reconnect_delay_timer); |
212 | s->reconnect_delay_timer = aio_timer_new(bdrv_get_aio_context(s->bs), | |
213 | QEMU_CLOCK_REALTIME, | |
214 | SCALE_NS, | |
215 | reconnect_delay_timer_cb, s); | |
216 | timer_mod(s->reconnect_delay_timer, expire_time_ns); | |
217 | } | |
218 | ||
f7651539 VSO |
219 | static void nbd_teardown_connection(BlockDriverState *bs) |
220 | { | |
221 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
222 | ||
4ddb5d2f VSO |
223 | assert(!s->in_flight); |
224 | ||
fbeb3e63 | 225 | if (s->ioc) { |
f7651539 | 226 | qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); |
4ddb5d2f VSO |
227 | yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), |
228 | nbd_yank, s->bs); | |
229 | object_unref(OBJECT(s->ioc)); | |
230 | s->ioc = NULL; | |
f7651539 | 231 | } |
fbeb3e63 | 232 | |
dba5156c PB |
233 | WITH_QEMU_LOCK_GUARD(&s->requests_lock) { |
234 | s->state = NBD_CLIENT_QUIT; | |
235 | } | |
f7651539 | 236 | } |
86f8cdf3 | 237 | |
be16b8bf VSO |
238 | static void open_timer_del(BDRVNBDState *s) |
239 | { | |
240 | if (s->open_timer) { | |
241 | timer_free(s->open_timer); | |
242 | s->open_timer = NULL; | |
243 | } | |
244 | } | |
245 | ||
246 | static void open_timer_cb(void *opaque) | |
247 | { | |
248 | BDRVNBDState *s = opaque; | |
249 | ||
250 | nbd_co_establish_connection_cancel(s->conn); | |
251 | open_timer_del(s); | |
252 | } | |
253 | ||
254 | static void open_timer_init(BDRVNBDState *s, uint64_t expire_time_ns) | |
255 | { | |
256 | assert(!s->open_timer); | |
257 | s->open_timer = aio_timer_new(bdrv_get_aio_context(s->bs), | |
258 | QEMU_CLOCK_REALTIME, | |
259 | SCALE_NS, | |
260 | open_timer_cb, s); | |
261 | timer_mod(s->open_timer, expire_time_ns); | |
262 | } | |
263 | ||
8d45185c | 264 | static bool nbd_client_will_reconnect(BDRVNBDState *s) |
f7651539 | 265 | { |
dba5156c PB |
266 | /* |
267 | * Called only after a socket error, so this is not performance sensitive. | |
268 | */ | |
269 | QEMU_LOCK_GUARD(&s->requests_lock); | |
270 | return s->state == NBD_CLIENT_CONNECTING_WAIT; | |
f7651539 | 271 | } |
dba5156c | 272 | |
e9ba7788 VSO |
273 | /* |
274 | * Update @bs with information learned during a completed negotiation process. | |
275 | * Return failure if the server's advertised options are incompatible with the | |
276 | * client's needs. | |
277 | */ | |
278 | static int nbd_handle_updated_info(BlockDriverState *bs, Error **errp) | |
279 | { | |
280 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
281 | int ret; | |
282 | ||
283 | if (s->x_dirty_bitmap) { | |
284 | if (!s->info.base_allocation) { | |
285 | error_setg(errp, "requested x-dirty-bitmap %s not found", | |
286 | s->x_dirty_bitmap); | |
287 | return -EINVAL; | |
288 | } | |
289 | if (strcmp(s->x_dirty_bitmap, "qemu:allocation-depth") == 0) { | |
290 | s->alloc_depth = true; | |
291 | } | |
292 | } | |
293 | ||
294 | if (s->info.flags & NBD_FLAG_READ_ONLY) { | |
295 | ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp); | |
296 | if (ret < 0) { | |
297 | return ret; | |
298 | } | |
299 | } | |
300 | ||
301 | if (s->info.flags & NBD_FLAG_SEND_FUA) { | |
302 | bs->supported_write_flags = BDRV_REQ_FUA; | |
303 | bs->supported_zero_flags |= BDRV_REQ_FUA; | |
304 | } | |
305 | ||
306 | if (s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) { | |
307 | bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP; | |
308 | if (s->info.flags & NBD_FLAG_SEND_FAST_ZERO) { | |
309 | bs->supported_zero_flags |= BDRV_REQ_NO_FALLBACK; | |
310 | } | |
311 | } | |
312 | ||
313 | trace_nbd_client_handshake_success(s->export); | |
314 | ||
315 | return 0; | |
316 | } | |
317 | ||
a71d597b | 318 | int coroutine_fn nbd_co_do_establish_connection(BlockDriverState *bs, |
8610b449 | 319 | bool blocking, Error **errp) |
f7651539 | 320 | { |
51edbf53 | 321 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
fa35591b | 322 | int ret; |
1581a70d | 323 | IO_CODE(); |
f7651539 | 324 | |
69aa0d37 | 325 | assert_bdrv_graph_readable(); |
51edbf53 VSO |
326 | assert(!s->ioc); |
327 | ||
4ddb5d2f | 328 | s->ioc = nbd_co_establish_connection(s->conn, &s->info, blocking, errp); |
51edbf53 VSO |
329 | if (!s->ioc) { |
330 | return -ECONNREFUSED; | |
331 | } | |
332 | ||
0b9cd6b9 LS |
333 | yank_register_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), nbd_yank, |
334 | bs); | |
335 | ||
51edbf53 VSO |
336 | ret = nbd_handle_updated_info(s->bs, NULL); |
337 | if (ret < 0) { | |
338 | /* | |
339 | * We have connected, but must fail for other reasons. | |
340 | * Send NBD_CMD_DISC as a courtesy to the server. | |
341 | */ | |
297365b4 | 342 | NBDRequest request = { .type = NBD_CMD_DISC, .mode = s->info.mode }; |
51edbf53 VSO |
343 | |
344 | nbd_send_request(s->ioc, &request); | |
345 | ||
0b9cd6b9 LS |
346 | yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), |
347 | nbd_yank, bs); | |
51edbf53 VSO |
348 | object_unref(OBJECT(s->ioc)); |
349 | s->ioc = NULL; | |
350 | ||
351 | return ret; | |
352 | } | |
353 | ||
354 | qio_channel_set_blocking(s->ioc, false, NULL); | |
06e0f098 | 355 | qio_channel_set_follow_coroutine_ctx(s->ioc, true); |
51edbf53 | 356 | |
51edbf53 | 357 | /* successfully connected */ |
dba5156c PB |
358 | WITH_QEMU_LOCK_GUARD(&s->requests_lock) { |
359 | s->state = NBD_CLIENT_CONNECTED; | |
360 | } | |
51edbf53 VSO |
361 | |
362 | return 0; | |
363 | } | |
364 | ||
dba5156c | 365 | /* Called with s->requests_lock held. */ |
8d45185c PB |
366 | static bool nbd_client_connecting(BDRVNBDState *s) |
367 | { | |
dba5156c PB |
368 | return s->state == NBD_CLIENT_CONNECTING_WAIT || |
369 | s->state == NBD_CLIENT_CONNECTING_NOWAIT; | |
8d45185c PB |
370 | } |
371 | ||
ee19d953 | 372 | /* Called with s->requests_lock taken. */ |
69aa0d37 | 373 | static void coroutine_fn GRAPH_RDLOCK nbd_reconnect_attempt(BDRVNBDState *s) |
51edbf53 | 374 | { |
8bb100c9 | 375 | int ret; |
dba5156c | 376 | bool blocking = s->state == NBD_CLIENT_CONNECTING_WAIT; |
8610b449 PB |
377 | |
378 | /* | |
379 | * Now we are sure that nobody is accessing the channel, and no one will | |
380 | * try until we set the state to CONNECTED. | |
381 | */ | |
4ddb5d2f | 382 | assert(nbd_client_connecting(s)); |
8610b449 | 383 | assert(s->in_flight == 1); |
f7651539 | 384 | |
8bb100c9 DL |
385 | trace_nbd_reconnect_attempt(s->bs->in_flight); |
386 | ||
8610b449 | 387 | if (blocking && !s->reconnect_delay_timer) { |
4ddb5d2f | 388 | /* |
8610b449 | 389 | * It's the first reconnect attempt after switching to |
4ddb5d2f VSO |
390 | * NBD_CLIENT_CONNECTING_WAIT |
391 | */ | |
8610b449 | 392 | g_assert(s->reconnect_delay); |
4ddb5d2f VSO |
393 | reconnect_delay_timer_init(s, |
394 | qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + | |
395 | s->reconnect_delay * NANOSECONDS_PER_SECOND); | |
f7651539 VSO |
396 | } |
397 | ||
f7651539 VSO |
398 | /* Finalize previous connection if any */ |
399 | if (s->ioc) { | |
fee091cd LS |
400 | yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), |
401 | nbd_yank, s->bs); | |
f7651539 VSO |
402 | object_unref(OBJECT(s->ioc)); |
403 | s->ioc = NULL; | |
404 | } | |
405 | ||
ee19d953 | 406 | qemu_mutex_unlock(&s->requests_lock); |
8bb100c9 DL |
407 | ret = nbd_co_do_establish_connection(s->bs, blocking, NULL); |
408 | trace_nbd_reconnect_attempt_result(ret, s->bs->in_flight); | |
ee19d953 | 409 | qemu_mutex_lock(&s->requests_lock); |
3ce1fc16 HR |
410 | |
411 | /* | |
412 | * The reconnect attempt is done (maybe successfully, maybe not), so | |
413 | * we no longer need this timer. Delete it so it will not outlive | |
414 | * this I/O request (so draining removes all timers). | |
415 | */ | |
416 | reconnect_delay_timer_del(s); | |
f7651539 VSO |
417 | } |
418 | ||
22efd811 | 419 | static coroutine_fn int nbd_receive_replies(BDRVNBDState *s, uint64_t cookie) |
f7651539 | 420 | { |
4ddb5d2f | 421 | int ret; |
8cb98a72 | 422 | uint64_t ind = COOKIE_TO_INDEX(cookie), ind2; |
4ddb5d2f | 423 | QEMU_LOCK_GUARD(&s->receive_mutex); |
f7651539 | 424 | |
4ddb5d2f | 425 | while (true) { |
22efd811 | 426 | if (s->reply.cookie == cookie) { |
4ddb5d2f VSO |
427 | /* We are done */ |
428 | return 0; | |
f7651539 VSO |
429 | } |
430 | ||
22efd811 | 431 | if (s->reply.cookie != 0) { |
4ddb5d2f VSO |
432 | /* |
433 | * Some other request is being handled now. It should already be | |
22efd811 | 434 | * woken by whoever set s->reply.cookie (or never wait in this |
4ddb5d2f VSO |
435 | * yield). So, we should not wake it here. |
436 | */ | |
8cb98a72 | 437 | ind2 = COOKIE_TO_INDEX(s->reply.cookie); |
4ddb5d2f | 438 | assert(!s->requests[ind2].receiving); |
86f8cdf3 | 439 | |
4ddb5d2f VSO |
440 | s->requests[ind].receiving = true; |
441 | qemu_co_mutex_unlock(&s->receive_mutex); | |
f7651539 | 442 | |
4ddb5d2f VSO |
443 | qemu_coroutine_yield(); |
444 | /* | |
a80a9a1c | 445 | * We may be woken for 2 reasons: |
4ddb5d2f | 446 | * 1. From this function, executing in parallel coroutine, when our |
22efd811 | 447 | * cookie is received. |
a80a9a1c | 448 | * 2. From nbd_co_receive_one_chunk(), when previous request is |
22efd811 | 449 | * finished and s->reply.cookie set to 0. |
4ddb5d2f VSO |
450 | * Anyway, it's OK to lock the mutex and go to the next iteration. |
451 | */ | |
f7651539 | 452 | |
4ddb5d2f VSO |
453 | qemu_co_mutex_lock(&s->receive_mutex); |
454 | assert(!s->requests[ind].receiving); | |
f7651539 VSO |
455 | continue; |
456 | } | |
457 | ||
22efd811 EB |
458 | /* We are under mutex and cookie is 0. We have to do the dirty work. */ |
459 | assert(s->reply.cookie == 0); | |
4ddb5d2f | 460 | ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, NULL); |
86f8cdf3 | 461 | if (ret <= 0) { |
4ddb5d2f VSO |
462 | ret = ret ? ret : -EIO; |
463 | nbd_channel_error(s, ret); | |
464 | return ret; | |
86f8cdf3 | 465 | } |
ac132d05 EB |
466 | if (nbd_reply_is_structured(&s->reply) && |
467 | s->info.mode < NBD_MODE_STRUCTURED) { | |
a34b1e5e | 468 | nbd_channel_error(s, -EINVAL); |
4ddb5d2f | 469 | return -EINVAL; |
86f8cdf3 | 470 | } |
8cb98a72 | 471 | ind2 = COOKIE_TO_INDEX(s->reply.cookie); |
8846b7d1 | 472 | if (ind2 >= MAX_NBD_REQUESTS || !s->requests[ind2].coroutine) { |
4ddb5d2f VSO |
473 | nbd_channel_error(s, -EINVAL); |
474 | return -EINVAL; | |
475 | } | |
22efd811 | 476 | if (s->reply.cookie == cookie) { |
8846b7d1 PB |
477 | /* We are done */ |
478 | return 0; | |
479 | } | |
4ddb5d2f | 480 | nbd_recv_coroutine_wake_one(&s->requests[ind2]); |
78c81a3f | 481 | } |
86f8cdf3 VSO |
482 | } |
483 | ||
69aa0d37 EGE |
484 | static int coroutine_fn GRAPH_RDLOCK |
485 | nbd_co_send_request(BlockDriverState *bs, NBDRequest *request, | |
486 | QEMUIOVector *qiov) | |
86f8cdf3 | 487 | { |
611ae1d7 | 488 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
a34b1e5e | 489 | int rc, i = -1; |
86f8cdf3 | 490 | |
ee19d953 | 491 | qemu_mutex_lock(&s->requests_lock); |
4ddb5d2f | 492 | while (s->in_flight == MAX_NBD_REQUESTS || |
dba5156c | 493 | (s->state != NBD_CLIENT_CONNECTED && s->in_flight > 0)) { |
ee19d953 | 494 | qemu_co_queue_wait(&s->free_sema, &s->requests_lock); |
86f8cdf3 | 495 | } |
a34b1e5e | 496 | |
8610b449 | 497 | s->in_flight++; |
dba5156c | 498 | if (s->state != NBD_CLIENT_CONNECTED) { |
8610b449 PB |
499 | if (nbd_client_connecting(s)) { |
500 | nbd_reconnect_attempt(s); | |
501 | qemu_co_queue_restart_all(&s->free_sema); | |
502 | } | |
dba5156c | 503 | if (s->state != NBD_CLIENT_CONNECTED) { |
8610b449 PB |
504 | rc = -EIO; |
505 | goto err; | |
506 | } | |
a34b1e5e VSO |
507 | } |
508 | ||
86f8cdf3 VSO |
509 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { |
510 | if (s->requests[i].coroutine == NULL) { | |
511 | break; | |
512 | } | |
513 | } | |
514 | ||
86f8cdf3 | 515 | assert(i < MAX_NBD_REQUESTS); |
86f8cdf3 VSO |
516 | s->requests[i].coroutine = qemu_coroutine_self(); |
517 | s->requests[i].offset = request->from; | |
518 | s->requests[i].receiving = false; | |
ee19d953 | 519 | qemu_mutex_unlock(&s->requests_lock); |
86f8cdf3 | 520 | |
ee19d953 | 521 | qemu_co_mutex_lock(&s->send_mutex); |
8cb98a72 | 522 | request->cookie = INDEX_TO_COOKIE(i); |
297365b4 | 523 | request->mode = s->info.mode; |
86f8cdf3 | 524 | |
86f8cdf3 VSO |
525 | assert(s->ioc); |
526 | ||
527 | if (qiov) { | |
528 | qio_channel_set_cork(s->ioc, true); | |
529 | rc = nbd_send_request(s->ioc, request); | |
2866ddd1 EB |
530 | if (rc >= 0 && qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov, |
531 | NULL) < 0) { | |
86f8cdf3 VSO |
532 | rc = -EIO; |
533 | } | |
534 | qio_channel_set_cork(s->ioc, false); | |
535 | } else { | |
536 | rc = nbd_send_request(s->ioc, request); | |
537 | } | |
ee19d953 | 538 | qemu_co_mutex_unlock(&s->send_mutex); |
86f8cdf3 | 539 | |
86f8cdf3 | 540 | if (rc < 0) { |
ee19d953 PB |
541 | qemu_mutex_lock(&s->requests_lock); |
542 | err: | |
dba5156c | 543 | nbd_channel_error_locked(s, rc); |
a34b1e5e VSO |
544 | if (i != -1) { |
545 | s->requests[i].coroutine = NULL; | |
f7651539 | 546 | } |
8610b449 | 547 | s->in_flight--; |
6690302b | 548 | qemu_co_queue_next(&s->free_sema); |
ee19d953 | 549 | qemu_mutex_unlock(&s->requests_lock); |
86f8cdf3 | 550 | } |
86f8cdf3 VSO |
551 | return rc; |
552 | } | |
553 | ||
554 | static inline uint16_t payload_advance16(uint8_t **payload) | |
555 | { | |
556 | *payload += 2; | |
557 | return lduw_be_p(*payload - 2); | |
558 | } | |
559 | ||
560 | static inline uint32_t payload_advance32(uint8_t **payload) | |
561 | { | |
562 | *payload += 4; | |
563 | return ldl_be_p(*payload - 4); | |
564 | } | |
565 | ||
566 | static inline uint64_t payload_advance64(uint8_t **payload) | |
567 | { | |
568 | *payload += 8; | |
569 | return ldq_be_p(*payload - 8); | |
570 | } | |
571 | ||
611ae1d7 | 572 | static int nbd_parse_offset_hole_payload(BDRVNBDState *s, |
86f8cdf3 VSO |
573 | NBDStructuredReplyChunk *chunk, |
574 | uint8_t *payload, uint64_t orig_offset, | |
575 | QEMUIOVector *qiov, Error **errp) | |
576 | { | |
577 | uint64_t offset; | |
578 | uint32_t hole_size; | |
579 | ||
580 | if (chunk->length != sizeof(offset) + sizeof(hole_size)) { | |
581 | error_setg(errp, "Protocol error: invalid payload for " | |
582 | "NBD_REPLY_TYPE_OFFSET_HOLE"); | |
583 | return -EINVAL; | |
584 | } | |
585 | ||
586 | offset = payload_advance64(&payload); | |
587 | hole_size = payload_advance32(&payload); | |
588 | ||
589 | if (!hole_size || offset < orig_offset || hole_size > qiov->size || | |
590 | offset > orig_offset + qiov->size - hole_size) { | |
591 | error_setg(errp, "Protocol error: server sent chunk exceeding requested" | |
592 | " region"); | |
593 | return -EINVAL; | |
594 | } | |
611ae1d7 VSO |
595 | if (s->info.min_block && |
596 | !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) { | |
86f8cdf3 VSO |
597 | trace_nbd_structured_read_compliance("hole"); |
598 | } | |
599 | ||
600 | qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size); | |
601 | ||
602 | return 0; | |
603 | } | |
604 | ||
605 | /* | |
606 | * nbd_parse_blockstatus_payload | |
607 | * Based on our request, we expect only one extent in reply, for the | |
608 | * base:allocation context. | |
609 | */ | |
611ae1d7 | 610 | static int nbd_parse_blockstatus_payload(BDRVNBDState *s, |
86f8cdf3 VSO |
611 | NBDStructuredReplyChunk *chunk, |
612 | uint8_t *payload, uint64_t orig_length, | |
d95ffb6f | 613 | NBDExtent32 *extent, Error **errp) |
86f8cdf3 VSO |
614 | { |
615 | uint32_t context_id; | |
616 | ||
617 | /* The server succeeded, so it must have sent [at least] one extent */ | |
618 | if (chunk->length < sizeof(context_id) + sizeof(*extent)) { | |
619 | error_setg(errp, "Protocol error: invalid payload for " | |
620 | "NBD_REPLY_TYPE_BLOCK_STATUS"); | |
621 | return -EINVAL; | |
622 | } | |
623 | ||
624 | context_id = payload_advance32(&payload); | |
611ae1d7 | 625 | if (s->info.context_id != context_id) { |
86f8cdf3 VSO |
626 | error_setg(errp, "Protocol error: unexpected context id %d for " |
627 | "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context " | |
628 | "id is %d", context_id, | |
611ae1d7 | 629 | s->info.context_id); |
86f8cdf3 VSO |
630 | return -EINVAL; |
631 | } | |
632 | ||
633 | extent->length = payload_advance32(&payload); | |
634 | extent->flags = payload_advance32(&payload); | |
635 | ||
636 | if (extent->length == 0) { | |
637 | error_setg(errp, "Protocol error: server sent status chunk with " | |
638 | "zero length"); | |
639 | return -EINVAL; | |
640 | } | |
641 | ||
642 | /* | |
643 | * A server sending unaligned block status is in violation of the | |
644 | * protocol, but as qemu-nbd 3.1 is such a server (at least for | |
645 | * POSIX files that are not a multiple of 512 bytes, since qemu | |
646 | * rounds files up to 512-byte multiples but lseek(SEEK_HOLE) | |
647 | * still sees an implicit hole beyond the real EOF), it's nicer to | |
648 | * work around the misbehaving server. If the request included | |
649 | * more than the final unaligned block, truncate it back to an | |
650 | * aligned result; if the request was only the final block, round | |
651 | * up to the full block and change the status to fully-allocated | |
652 | * (always a safe status, even if it loses information). | |
653 | */ | |
611ae1d7 VSO |
654 | if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length, |
655 | s->info.min_block)) { | |
86f8cdf3 | 656 | trace_nbd_parse_blockstatus_compliance("extent length is unaligned"); |
611ae1d7 | 657 | if (extent->length > s->info.min_block) { |
86f8cdf3 | 658 | extent->length = QEMU_ALIGN_DOWN(extent->length, |
611ae1d7 | 659 | s->info.min_block); |
86f8cdf3 | 660 | } else { |
611ae1d7 | 661 | extent->length = s->info.min_block; |
86f8cdf3 VSO |
662 | extent->flags = 0; |
663 | } | |
664 | } | |
665 | ||
666 | /* | |
667 | * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have | |
668 | * sent us any more than one extent, nor should it have included | |
669 | * status beyond our request in that extent. However, it's easy | |
670 | * enough to ignore the server's noncompliance without killing the | |
671 | * connection; just ignore trailing extents, and clamp things to | |
672 | * the length of our request. | |
673 | */ | |
674 | if (chunk->length > sizeof(context_id) + sizeof(*extent)) { | |
675 | trace_nbd_parse_blockstatus_compliance("more than one extent"); | |
676 | } | |
677 | if (extent->length > orig_length) { | |
678 | extent->length = orig_length; | |
679 | trace_nbd_parse_blockstatus_compliance("extent length too large"); | |
680 | } | |
681 | ||
dbc7b014 EB |
682 | /* |
683 | * HACK: if we are using x-dirty-bitmaps to access | |
684 | * qemu:allocation-depth, treat all depths > 2 the same as 2, | |
685 | * since nbd_client_co_block_status is only expecting the low two | |
686 | * bits to be set. | |
687 | */ | |
688 | if (s->alloc_depth && extent->flags > 2) { | |
689 | extent->flags = 2; | |
690 | } | |
691 | ||
86f8cdf3 VSO |
692 | return 0; |
693 | } | |
694 | ||
695 | /* | |
696 | * nbd_parse_error_payload | |
697 | * on success @errp contains message describing nbd error reply | |
698 | */ | |
699 | static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk, | |
700 | uint8_t *payload, int *request_ret, | |
701 | Error **errp) | |
702 | { | |
703 | uint32_t error; | |
704 | uint16_t message_size; | |
705 | ||
706 | assert(chunk->type & (1 << 15)); | |
707 | ||
708 | if (chunk->length < sizeof(error) + sizeof(message_size)) { | |
709 | error_setg(errp, | |
710 | "Protocol error: invalid payload for structured error"); | |
711 | return -EINVAL; | |
712 | } | |
713 | ||
714 | error = nbd_errno_to_system_errno(payload_advance32(&payload)); | |
715 | if (error == 0) { | |
716 | error_setg(errp, "Protocol error: server sent structured error chunk " | |
717 | "with error = 0"); | |
718 | return -EINVAL; | |
719 | } | |
720 | ||
721 | *request_ret = -error; | |
722 | message_size = payload_advance16(&payload); | |
723 | ||
724 | if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) { | |
725 | error_setg(errp, "Protocol error: server sent structured error chunk " | |
726 | "with incorrect message size"); | |
727 | return -EINVAL; | |
728 | } | |
729 | ||
730 | /* TODO: Add a trace point to mention the server complaint */ | |
731 | ||
732 | /* TODO handle ERROR_OFFSET */ | |
733 | ||
734 | return 0; | |
735 | } | |
736 | ||
0c43c6fc PB |
737 | static int coroutine_fn |
738 | nbd_co_receive_offset_data_payload(BDRVNBDState *s, uint64_t orig_offset, | |
739 | QEMUIOVector *qiov, Error **errp) | |
86f8cdf3 VSO |
740 | { |
741 | QEMUIOVector sub_qiov; | |
742 | uint64_t offset; | |
743 | size_t data_size; | |
744 | int ret; | |
745 | NBDStructuredReplyChunk *chunk = &s->reply.structured; | |
746 | ||
747 | assert(nbd_reply_is_structured(&s->reply)); | |
748 | ||
749 | /* The NBD spec requires at least one byte of payload */ | |
750 | if (chunk->length <= sizeof(offset)) { | |
751 | error_setg(errp, "Protocol error: invalid payload for " | |
752 | "NBD_REPLY_TYPE_OFFSET_DATA"); | |
753 | return -EINVAL; | |
754 | } | |
755 | ||
756 | if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) { | |
757 | return -EIO; | |
758 | } | |
759 | ||
760 | data_size = chunk->length - sizeof(offset); | |
761 | assert(data_size); | |
762 | if (offset < orig_offset || data_size > qiov->size || | |
763 | offset > orig_offset + qiov->size - data_size) { | |
764 | error_setg(errp, "Protocol error: server sent chunk exceeding requested" | |
765 | " region"); | |
766 | return -EINVAL; | |
767 | } | |
768 | if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) { | |
769 | trace_nbd_structured_read_compliance("data"); | |
770 | } | |
771 | ||
772 | qemu_iovec_init(&sub_qiov, qiov->niov); | |
773 | qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size); | |
774 | ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp); | |
775 | qemu_iovec_destroy(&sub_qiov); | |
776 | ||
777 | return ret < 0 ? -EIO : 0; | |
778 | } | |
779 | ||
780 | #define NBD_MAX_MALLOC_PAYLOAD 1000 | |
781 | static coroutine_fn int nbd_co_receive_structured_payload( | |
611ae1d7 | 782 | BDRVNBDState *s, void **payload, Error **errp) |
86f8cdf3 VSO |
783 | { |
784 | int ret; | |
785 | uint32_t len; | |
786 | ||
787 | assert(nbd_reply_is_structured(&s->reply)); | |
788 | ||
789 | len = s->reply.structured.length; | |
790 | ||
791 | if (len == 0) { | |
792 | return 0; | |
793 | } | |
794 | ||
795 | if (payload == NULL) { | |
796 | error_setg(errp, "Unexpected structured payload"); | |
797 | return -EINVAL; | |
798 | } | |
799 | ||
800 | if (len > NBD_MAX_MALLOC_PAYLOAD) { | |
801 | error_setg(errp, "Payload too large"); | |
802 | return -EINVAL; | |
803 | } | |
804 | ||
805 | *payload = g_new(char, len); | |
806 | ret = nbd_read(s->ioc, *payload, len, "structured payload", errp); | |
807 | if (ret < 0) { | |
808 | g_free(*payload); | |
809 | *payload = NULL; | |
810 | return ret; | |
811 | } | |
812 | ||
813 | return 0; | |
814 | } | |
815 | ||
816 | /* | |
817 | * nbd_co_do_receive_one_chunk | |
818 | * for simple reply: | |
819 | * set request_ret to received reply error | |
820 | * if qiov is not NULL: read payload to @qiov | |
821 | * for structured reply chunk: | |
822 | * if error chunk: read payload, set @request_ret, do not set @payload | |
823 | * else if offset_data chunk: read payload data to @qiov, do not set @payload | |
824 | * else: read payload to @payload | |
825 | * | |
826 | * If function fails, @errp contains corresponding error message, and the | |
827 | * connection with the server is suspect. If it returns 0, then the | |
828 | * transaction succeeded (although @request_ret may be a negative errno | |
829 | * corresponding to the server's error reply), and errp is unchanged. | |
830 | */ | |
831 | static coroutine_fn int nbd_co_do_receive_one_chunk( | |
22efd811 | 832 | BDRVNBDState *s, uint64_t cookie, bool only_structured, |
86f8cdf3 VSO |
833 | int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp) |
834 | { | |
835 | int ret; | |
8cb98a72 | 836 | int i = COOKIE_TO_INDEX(cookie); |
86f8cdf3 VSO |
837 | void *local_payload = NULL; |
838 | NBDStructuredReplyChunk *chunk; | |
839 | ||
840 | if (payload) { | |
841 | *payload = NULL; | |
842 | } | |
843 | *request_ret = 0; | |
844 | ||
22efd811 | 845 | ret = nbd_receive_replies(s, cookie); |
172f5f1a | 846 | if (ret < 0) { |
86f8cdf3 VSO |
847 | error_setg(errp, "Connection closed"); |
848 | return -EIO; | |
849 | } | |
850 | assert(s->ioc); | |
851 | ||
22efd811 | 852 | assert(s->reply.cookie == cookie); |
86f8cdf3 VSO |
853 | |
854 | if (nbd_reply_is_simple(&s->reply)) { | |
855 | if (only_structured) { | |
856 | error_setg(errp, "Protocol error: simple reply when structured " | |
857 | "reply chunk was expected"); | |
858 | return -EINVAL; | |
859 | } | |
860 | ||
861 | *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error); | |
862 | if (*request_ret < 0 || !qiov) { | |
863 | return 0; | |
864 | } | |
865 | ||
866 | return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov, | |
867 | errp) < 0 ? -EIO : 0; | |
868 | } | |
869 | ||
870 | /* handle structured reply chunk */ | |
ac132d05 | 871 | assert(s->info.mode >= NBD_MODE_STRUCTURED); |
86f8cdf3 VSO |
872 | chunk = &s->reply.structured; |
873 | ||
874 | if (chunk->type == NBD_REPLY_TYPE_NONE) { | |
875 | if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) { | |
876 | error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without" | |
877 | " NBD_REPLY_FLAG_DONE flag set"); | |
878 | return -EINVAL; | |
879 | } | |
880 | if (chunk->length) { | |
881 | error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with" | |
882 | " nonzero length"); | |
883 | return -EINVAL; | |
884 | } | |
885 | return 0; | |
886 | } | |
887 | ||
888 | if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) { | |
889 | if (!qiov) { | |
890 | error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk"); | |
891 | return -EINVAL; | |
892 | } | |
893 | ||
894 | return nbd_co_receive_offset_data_payload(s, s->requests[i].offset, | |
895 | qiov, errp); | |
896 | } | |
897 | ||
898 | if (nbd_reply_type_is_error(chunk->type)) { | |
899 | payload = &local_payload; | |
900 | } | |
901 | ||
902 | ret = nbd_co_receive_structured_payload(s, payload, errp); | |
903 | if (ret < 0) { | |
904 | return ret; | |
905 | } | |
906 | ||
907 | if (nbd_reply_type_is_error(chunk->type)) { | |
908 | ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp); | |
909 | g_free(local_payload); | |
910 | return ret; | |
911 | } | |
912 | ||
913 | return 0; | |
914 | } | |
915 | ||
916 | /* | |
917 | * nbd_co_receive_one_chunk | |
918 | * Read reply, wake up connection_co and set s->quit if needed. | |
919 | * Return value is a fatal error code or normal nbd reply error code | |
920 | */ | |
921 | static coroutine_fn int nbd_co_receive_one_chunk( | |
22efd811 | 922 | BDRVNBDState *s, uint64_t cookie, bool only_structured, |
86f8cdf3 VSO |
923 | int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload, |
924 | Error **errp) | |
925 | { | |
22efd811 | 926 | int ret = nbd_co_do_receive_one_chunk(s, cookie, only_structured, |
86f8cdf3 VSO |
927 | request_ret, qiov, payload, errp); |
928 | ||
929 | if (ret < 0) { | |
5cf42b1c | 930 | memset(reply, 0, sizeof(*reply)); |
a34b1e5e | 931 | nbd_channel_error(s, ret); |
86f8cdf3 VSO |
932 | } else { |
933 | /* For assert at loop start in nbd_connection_entry */ | |
5cf42b1c | 934 | *reply = s->reply; |
86f8cdf3 | 935 | } |
22efd811 | 936 | s->reply.cookie = 0; |
86f8cdf3 | 937 | |
a80a9a1c | 938 | nbd_recv_coroutines_wake(s); |
86f8cdf3 VSO |
939 | |
940 | return ret; | |
941 | } | |
942 | ||
943 | typedef struct NBDReplyChunkIter { | |
944 | int ret; | |
945 | int request_ret; | |
946 | Error *err; | |
947 | bool done, only_structured; | |
948 | } NBDReplyChunkIter; | |
949 | ||
950 | static void nbd_iter_channel_error(NBDReplyChunkIter *iter, | |
951 | int ret, Error **local_err) | |
952 | { | |
d9366135 | 953 | assert(local_err && *local_err); |
86f8cdf3 VSO |
954 | assert(ret < 0); |
955 | ||
956 | if (!iter->ret) { | |
957 | iter->ret = ret; | |
958 | error_propagate(&iter->err, *local_err); | |
959 | } else { | |
960 | error_free(*local_err); | |
961 | } | |
962 | ||
963 | *local_err = NULL; | |
964 | } | |
965 | ||
966 | static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret) | |
967 | { | |
968 | assert(ret < 0); | |
969 | ||
970 | if (!iter->request_ret) { | |
971 | iter->request_ret = ret; | |
972 | } | |
973 | } | |
974 | ||
975 | /* | |
976 | * NBD_FOREACH_REPLY_CHUNK | |
977 | * The pointer stored in @payload requires g_free() to free it. | |
978 | */ | |
22efd811 | 979 | #define NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, structured, \ |
86f8cdf3 VSO |
980 | qiov, reply, payload) \ |
981 | for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \ | |
22efd811 | 982 | nbd_reply_chunk_iter_receive(s, &iter, cookie, qiov, reply, payload);) |
86f8cdf3 VSO |
983 | |
984 | /* | |
985 | * nbd_reply_chunk_iter_receive | |
986 | * The pointer stored in @payload requires g_free() to free it. | |
987 | */ | |
8e5a19df PB |
988 | static bool coroutine_fn nbd_reply_chunk_iter_receive(BDRVNBDState *s, |
989 | NBDReplyChunkIter *iter, | |
22efd811 | 990 | uint64_t cookie, |
8e5a19df PB |
991 | QEMUIOVector *qiov, |
992 | NBDReply *reply, | |
993 | void **payload) | |
86f8cdf3 VSO |
994 | { |
995 | int ret, request_ret; | |
996 | NBDReply local_reply; | |
997 | NBDStructuredReplyChunk *chunk; | |
998 | Error *local_err = NULL; | |
86f8cdf3 VSO |
999 | |
1000 | if (iter->done) { | |
1001 | /* Previous iteration was last. */ | |
1002 | goto break_loop; | |
1003 | } | |
1004 | ||
1005 | if (reply == NULL) { | |
1006 | reply = &local_reply; | |
1007 | } | |
1008 | ||
22efd811 | 1009 | ret = nbd_co_receive_one_chunk(s, cookie, iter->only_structured, |
86f8cdf3 VSO |
1010 | &request_ret, qiov, reply, payload, |
1011 | &local_err); | |
1012 | if (ret < 0) { | |
1013 | nbd_iter_channel_error(iter, ret, &local_err); | |
1014 | } else if (request_ret < 0) { | |
1015 | nbd_iter_request_error(iter, request_ret); | |
1016 | } | |
1017 | ||
1018 | /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */ | |
172f5f1a | 1019 | if (nbd_reply_is_simple(reply) || iter->ret < 0) { |
86f8cdf3 VSO |
1020 | goto break_loop; |
1021 | } | |
1022 | ||
1023 | chunk = &reply->structured; | |
1024 | iter->only_structured = true; | |
1025 | ||
1026 | if (chunk->type == NBD_REPLY_TYPE_NONE) { | |
1027 | /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */ | |
1028 | assert(chunk->flags & NBD_REPLY_FLAG_DONE); | |
1029 | goto break_loop; | |
1030 | } | |
1031 | ||
1032 | if (chunk->flags & NBD_REPLY_FLAG_DONE) { | |
1033 | /* This iteration is last. */ | |
1034 | iter->done = true; | |
1035 | } | |
1036 | ||
1037 | /* Execute the loop body */ | |
1038 | return true; | |
1039 | ||
1040 | break_loop: | |
ee19d953 | 1041 | qemu_mutex_lock(&s->requests_lock); |
8cb98a72 | 1042 | s->requests[COOKIE_TO_INDEX(cookie)].coroutine = NULL; |
86f8cdf3 | 1043 | s->in_flight--; |
4ddb5d2f | 1044 | qemu_co_queue_next(&s->free_sema); |
ee19d953 | 1045 | qemu_mutex_unlock(&s->requests_lock); |
86f8cdf3 VSO |
1046 | |
1047 | return false; | |
1048 | } | |
1049 | ||
22efd811 EB |
1050 | static int coroutine_fn |
1051 | nbd_co_receive_return_code(BDRVNBDState *s, uint64_t cookie, | |
1052 | int *request_ret, Error **errp) | |
86f8cdf3 VSO |
1053 | { |
1054 | NBDReplyChunkIter iter; | |
1055 | ||
22efd811 | 1056 | NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, NULL, NULL) { |
86f8cdf3 VSO |
1057 | /* nbd_reply_chunk_iter_receive does all the work */ |
1058 | } | |
1059 | ||
1060 | error_propagate(errp, iter.err); | |
1061 | *request_ret = iter.request_ret; | |
1062 | return iter.ret; | |
1063 | } | |
1064 | ||
22efd811 EB |
1065 | static int coroutine_fn |
1066 | nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t cookie, | |
1067 | uint64_t offset, QEMUIOVector *qiov, | |
1068 | int *request_ret, Error **errp) | |
86f8cdf3 VSO |
1069 | { |
1070 | NBDReplyChunkIter iter; | |
1071 | NBDReply reply; | |
1072 | void *payload = NULL; | |
1073 | Error *local_err = NULL; | |
1074 | ||
ac132d05 EB |
1075 | NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, |
1076 | s->info.mode >= NBD_MODE_STRUCTURED, | |
86f8cdf3 VSO |
1077 | qiov, &reply, &payload) |
1078 | { | |
1079 | int ret; | |
1080 | NBDStructuredReplyChunk *chunk = &reply.structured; | |
1081 | ||
1082 | assert(nbd_reply_is_structured(&reply)); | |
1083 | ||
1084 | switch (chunk->type) { | |
1085 | case NBD_REPLY_TYPE_OFFSET_DATA: | |
1086 | /* | |
1087 | * special cased in nbd_co_receive_one_chunk, data is already | |
1088 | * in qiov | |
1089 | */ | |
1090 | break; | |
1091 | case NBD_REPLY_TYPE_OFFSET_HOLE: | |
1092 | ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload, | |
1093 | offset, qiov, &local_err); | |
1094 | if (ret < 0) { | |
a34b1e5e | 1095 | nbd_channel_error(s, ret); |
86f8cdf3 VSO |
1096 | nbd_iter_channel_error(&iter, ret, &local_err); |
1097 | } | |
1098 | break; | |
1099 | default: | |
1100 | if (!nbd_reply_type_is_error(chunk->type)) { | |
1101 | /* not allowed reply type */ | |
a34b1e5e | 1102 | nbd_channel_error(s, -EINVAL); |
86f8cdf3 VSO |
1103 | error_setg(&local_err, |
1104 | "Unexpected reply type: %d (%s) for CMD_READ", | |
1105 | chunk->type, nbd_reply_type_lookup(chunk->type)); | |
1106 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); | |
1107 | } | |
1108 | } | |
1109 | ||
1110 | g_free(payload); | |
1111 | payload = NULL; | |
1112 | } | |
1113 | ||
1114 | error_propagate(errp, iter.err); | |
1115 | *request_ret = iter.request_ret; | |
1116 | return iter.ret; | |
1117 | } | |
1118 | ||
22efd811 EB |
1119 | static int coroutine_fn |
1120 | nbd_co_receive_blockstatus_reply(BDRVNBDState *s, uint64_t cookie, | |
d95ffb6f | 1121 | uint64_t length, NBDExtent32 *extent, |
22efd811 | 1122 | int *request_ret, Error **errp) |
86f8cdf3 VSO |
1123 | { |
1124 | NBDReplyChunkIter iter; | |
1125 | NBDReply reply; | |
1126 | void *payload = NULL; | |
1127 | Error *local_err = NULL; | |
1128 | bool received = false; | |
1129 | ||
1130 | assert(!extent->length); | |
22efd811 | 1131 | NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, &reply, &payload) { |
86f8cdf3 VSO |
1132 | int ret; |
1133 | NBDStructuredReplyChunk *chunk = &reply.structured; | |
1134 | ||
1135 | assert(nbd_reply_is_structured(&reply)); | |
1136 | ||
1137 | switch (chunk->type) { | |
1138 | case NBD_REPLY_TYPE_BLOCK_STATUS: | |
1139 | if (received) { | |
a34b1e5e | 1140 | nbd_channel_error(s, -EINVAL); |
86f8cdf3 VSO |
1141 | error_setg(&local_err, "Several BLOCK_STATUS chunks in reply"); |
1142 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); | |
1143 | } | |
1144 | received = true; | |
1145 | ||
1146 | ret = nbd_parse_blockstatus_payload(s, &reply.structured, | |
1147 | payload, length, extent, | |
1148 | &local_err); | |
1149 | if (ret < 0) { | |
a34b1e5e | 1150 | nbd_channel_error(s, ret); |
86f8cdf3 VSO |
1151 | nbd_iter_channel_error(&iter, ret, &local_err); |
1152 | } | |
1153 | break; | |
1154 | default: | |
1155 | if (!nbd_reply_type_is_error(chunk->type)) { | |
a34b1e5e | 1156 | nbd_channel_error(s, -EINVAL); |
86f8cdf3 VSO |
1157 | error_setg(&local_err, |
1158 | "Unexpected reply type: %d (%s) " | |
1159 | "for CMD_BLOCK_STATUS", | |
1160 | chunk->type, nbd_reply_type_lookup(chunk->type)); | |
1161 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); | |
1162 | } | |
1163 | } | |
1164 | ||
1165 | g_free(payload); | |
1166 | payload = NULL; | |
1167 | } | |
1168 | ||
1169 | if (!extent->length && !iter.request_ret) { | |
1170 | error_setg(&local_err, "Server did not reply with any status extents"); | |
1171 | nbd_iter_channel_error(&iter, -EIO, &local_err); | |
1172 | } | |
1173 | ||
1174 | error_propagate(errp, iter.err); | |
1175 | *request_ret = iter.request_ret; | |
1176 | return iter.ret; | |
1177 | } | |
1178 | ||
69aa0d37 EGE |
1179 | static int coroutine_fn GRAPH_RDLOCK |
1180 | nbd_co_request(BlockDriverState *bs, NBDRequest *request, | |
1181 | QEMUIOVector *write_qiov) | |
86f8cdf3 VSO |
1182 | { |
1183 | int ret, request_ret; | |
1184 | Error *local_err = NULL; | |
611ae1d7 | 1185 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1186 | |
1187 | assert(request->type != NBD_CMD_READ); | |
1188 | if (write_qiov) { | |
1189 | assert(request->type == NBD_CMD_WRITE); | |
1190 | assert(request->len == iov_size(write_qiov->iov, write_qiov->niov)); | |
1191 | } else { | |
1192 | assert(request->type != NBD_CMD_WRITE); | |
1193 | } | |
86f8cdf3 | 1194 | |
f7651539 VSO |
1195 | do { |
1196 | ret = nbd_co_send_request(bs, request, write_qiov); | |
1197 | if (ret < 0) { | |
1198 | continue; | |
1199 | } | |
1200 | ||
22efd811 | 1201 | ret = nbd_co_receive_return_code(s, request->cookie, |
f7651539 VSO |
1202 | &request_ret, &local_err); |
1203 | if (local_err) { | |
1204 | trace_nbd_co_request_fail(request->from, request->len, | |
22efd811 | 1205 | request->cookie, request->flags, |
f7651539 VSO |
1206 | request->type, |
1207 | nbd_cmd_lookup(request->type), | |
1208 | ret, error_get_pretty(local_err)); | |
1209 | error_free(local_err); | |
1210 | local_err = NULL; | |
1211 | } | |
8d45185c | 1212 | } while (ret < 0 && nbd_client_will_reconnect(s)); |
f7651539 | 1213 | |
86f8cdf3 VSO |
1214 | return ret ? ret : request_ret; |
1215 | } | |
1216 | ||
69aa0d37 EGE |
1217 | static int coroutine_fn GRAPH_RDLOCK |
1218 | nbd_client_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1219 | QEMUIOVector *qiov, BdrvRequestFlags flags) | |
86f8cdf3 VSO |
1220 | { |
1221 | int ret, request_ret; | |
1222 | Error *local_err = NULL; | |
611ae1d7 | 1223 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1224 | NBDRequest request = { |
1225 | .type = NBD_CMD_READ, | |
1226 | .from = offset, | |
1227 | .len = bytes, | |
1228 | }; | |
1229 | ||
1230 | assert(bytes <= NBD_MAX_BUFFER_SIZE); | |
86f8cdf3 VSO |
1231 | |
1232 | if (!bytes) { | |
1233 | return 0; | |
1234 | } | |
1235 | /* | |
1236 | * Work around the fact that the block layer doesn't do | |
1237 | * byte-accurate sizing yet - if the read exceeds the server's | |
1238 | * advertised size because the block layer rounded size up, then | |
1239 | * truncate the request to the server and tail-pad with zero. | |
1240 | */ | |
611ae1d7 | 1241 | if (offset >= s->info.size) { |
86f8cdf3 VSO |
1242 | assert(bytes < BDRV_SECTOR_SIZE); |
1243 | qemu_iovec_memset(qiov, 0, 0, bytes); | |
1244 | return 0; | |
1245 | } | |
611ae1d7 VSO |
1246 | if (offset + bytes > s->info.size) { |
1247 | uint64_t slop = offset + bytes - s->info.size; | |
86f8cdf3 VSO |
1248 | |
1249 | assert(slop < BDRV_SECTOR_SIZE); | |
1250 | qemu_iovec_memset(qiov, bytes - slop, 0, slop); | |
1251 | request.len -= slop; | |
1252 | } | |
1253 | ||
f7651539 VSO |
1254 | do { |
1255 | ret = nbd_co_send_request(bs, &request, NULL); | |
1256 | if (ret < 0) { | |
1257 | continue; | |
1258 | } | |
1259 | ||
22efd811 | 1260 | ret = nbd_co_receive_cmdread_reply(s, request.cookie, offset, qiov, |
f7651539 VSO |
1261 | &request_ret, &local_err); |
1262 | if (local_err) { | |
22efd811 | 1263 | trace_nbd_co_request_fail(request.from, request.len, request.cookie, |
f7651539 VSO |
1264 | request.flags, request.type, |
1265 | nbd_cmd_lookup(request.type), | |
1266 | ret, error_get_pretty(local_err)); | |
1267 | error_free(local_err); | |
1268 | local_err = NULL; | |
1269 | } | |
8d45185c | 1270 | } while (ret < 0 && nbd_client_will_reconnect(s)); |
86f8cdf3 | 1271 | |
86f8cdf3 VSO |
1272 | return ret ? ret : request_ret; |
1273 | } | |
1274 | ||
69aa0d37 EGE |
1275 | static int coroutine_fn GRAPH_RDLOCK |
1276 | nbd_client_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1277 | QEMUIOVector *qiov, BdrvRequestFlags flags) | |
86f8cdf3 | 1278 | { |
611ae1d7 | 1279 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1280 | NBDRequest request = { |
1281 | .type = NBD_CMD_WRITE, | |
1282 | .from = offset, | |
1283 | .len = bytes, | |
1284 | }; | |
1285 | ||
611ae1d7 | 1286 | assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); |
86f8cdf3 | 1287 | if (flags & BDRV_REQ_FUA) { |
611ae1d7 | 1288 | assert(s->info.flags & NBD_FLAG_SEND_FUA); |
86f8cdf3 VSO |
1289 | request.flags |= NBD_CMD_FLAG_FUA; |
1290 | } | |
1291 | ||
1292 | assert(bytes <= NBD_MAX_BUFFER_SIZE); | |
1293 | ||
1294 | if (!bytes) { | |
1295 | return 0; | |
1296 | } | |
1297 | return nbd_co_request(bs, &request, qiov); | |
1298 | } | |
1299 | ||
69aa0d37 EGE |
1300 | static int coroutine_fn GRAPH_RDLOCK |
1301 | nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1302 | BdrvRequestFlags flags) | |
86f8cdf3 | 1303 | { |
611ae1d7 | 1304 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1305 | NBDRequest request = { |
1306 | .type = NBD_CMD_WRITE_ZEROES, | |
1307 | .from = offset, | |
b2578459 | 1308 | .len = bytes, |
86f8cdf3 VSO |
1309 | }; |
1310 | ||
b2578459 EB |
1311 | /* rely on max_pwrite_zeroes */ |
1312 | assert(bytes <= UINT32_MAX || s->info.mode >= NBD_MODE_EXTENDED); | |
f34b2bcf | 1313 | |
611ae1d7 VSO |
1314 | assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); |
1315 | if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) { | |
86f8cdf3 VSO |
1316 | return -ENOTSUP; |
1317 | } | |
1318 | ||
1319 | if (flags & BDRV_REQ_FUA) { | |
611ae1d7 | 1320 | assert(s->info.flags & NBD_FLAG_SEND_FUA); |
86f8cdf3 VSO |
1321 | request.flags |= NBD_CMD_FLAG_FUA; |
1322 | } | |
1323 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
1324 | request.flags |= NBD_CMD_FLAG_NO_HOLE; | |
1325 | } | |
f061656c EB |
1326 | if (flags & BDRV_REQ_NO_FALLBACK) { |
1327 | assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO); | |
1328 | request.flags |= NBD_CMD_FLAG_FAST_ZERO; | |
1329 | } | |
86f8cdf3 VSO |
1330 | |
1331 | if (!bytes) { | |
1332 | return 0; | |
1333 | } | |
1334 | return nbd_co_request(bs, &request, NULL); | |
1335 | } | |
1336 | ||
69aa0d37 | 1337 | static int coroutine_fn GRAPH_RDLOCK nbd_client_co_flush(BlockDriverState *bs) |
86f8cdf3 | 1338 | { |
611ae1d7 | 1339 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1340 | NBDRequest request = { .type = NBD_CMD_FLUSH }; |
1341 | ||
611ae1d7 | 1342 | if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) { |
86f8cdf3 VSO |
1343 | return 0; |
1344 | } | |
1345 | ||
1346 | request.from = 0; | |
1347 | request.len = 0; | |
1348 | ||
1349 | return nbd_co_request(bs, &request, NULL); | |
1350 | } | |
1351 | ||
69aa0d37 EGE |
1352 | static int coroutine_fn GRAPH_RDLOCK |
1353 | nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
86f8cdf3 | 1354 | { |
611ae1d7 | 1355 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1356 | NBDRequest request = { |
1357 | .type = NBD_CMD_TRIM, | |
1358 | .from = offset, | |
b2578459 | 1359 | .len = bytes, |
86f8cdf3 VSO |
1360 | }; |
1361 | ||
b2578459 EB |
1362 | /* rely on max_pdiscard */ |
1363 | assert(bytes <= UINT32_MAX || s->info.mode >= NBD_MODE_EXTENDED); | |
0c802287 | 1364 | |
611ae1d7 VSO |
1365 | assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); |
1366 | if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) { | |
86f8cdf3 VSO |
1367 | return 0; |
1368 | } | |
1369 | ||
1370 | return nbd_co_request(bs, &request, NULL); | |
1371 | } | |
1372 | ||
69aa0d37 | 1373 | static int coroutine_fn GRAPH_RDLOCK nbd_client_co_block_status( |
86f8cdf3 VSO |
1374 | BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes, |
1375 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
1376 | { | |
1377 | int ret, request_ret; | |
d95ffb6f | 1378 | NBDExtent32 extent = { 0 }; |
611ae1d7 | 1379 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
86f8cdf3 VSO |
1380 | Error *local_err = NULL; |
1381 | ||
1382 | NBDRequest request = { | |
1383 | .type = NBD_CMD_BLOCK_STATUS, | |
1384 | .from = offset, | |
b2578459 | 1385 | .len = MIN(bytes, s->info.size - offset), |
86f8cdf3 VSO |
1386 | .flags = NBD_CMD_FLAG_REQ_ONE, |
1387 | }; | |
1388 | ||
611ae1d7 | 1389 | if (!s->info.base_allocation) { |
86f8cdf3 VSO |
1390 | *pnum = bytes; |
1391 | *map = offset; | |
1392 | *file = bs; | |
1393 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
1394 | } | |
b2578459 EB |
1395 | if (s->info.mode < NBD_MODE_EXTENDED) { |
1396 | request.len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment), | |
1397 | request.len); | |
1398 | } | |
86f8cdf3 VSO |
1399 | |
1400 | /* | |
1401 | * Work around the fact that the block layer doesn't do | |
1402 | * byte-accurate sizing yet - if the status request exceeds the | |
1403 | * server's advertised size because the block layer rounded size | |
1404 | * up, we truncated the request to the server (above), or are | |
1405 | * called on just the hole. | |
1406 | */ | |
611ae1d7 | 1407 | if (offset >= s->info.size) { |
86f8cdf3 VSO |
1408 | *pnum = bytes; |
1409 | assert(bytes < BDRV_SECTOR_SIZE); | |
1410 | /* Intentionally don't report offset_valid for the hole */ | |
1411 | return BDRV_BLOCK_ZERO; | |
1412 | } | |
1413 | ||
611ae1d7 VSO |
1414 | if (s->info.min_block) { |
1415 | assert(QEMU_IS_ALIGNED(request.len, s->info.min_block)); | |
86f8cdf3 | 1416 | } |
f7651539 VSO |
1417 | do { |
1418 | ret = nbd_co_send_request(bs, &request, NULL); | |
1419 | if (ret < 0) { | |
1420 | continue; | |
1421 | } | |
1422 | ||
22efd811 | 1423 | ret = nbd_co_receive_blockstatus_reply(s, request.cookie, bytes, |
f7651539 VSO |
1424 | &extent, &request_ret, |
1425 | &local_err); | |
1426 | if (local_err) { | |
22efd811 | 1427 | trace_nbd_co_request_fail(request.from, request.len, request.cookie, |
f7651539 VSO |
1428 | request.flags, request.type, |
1429 | nbd_cmd_lookup(request.type), | |
1430 | ret, error_get_pretty(local_err)); | |
1431 | error_free(local_err); | |
1432 | local_err = NULL; | |
1433 | } | |
8d45185c | 1434 | } while (ret < 0 && nbd_client_will_reconnect(s)); |
86f8cdf3 | 1435 | |
86f8cdf3 VSO |
1436 | if (ret < 0 || request_ret < 0) { |
1437 | return ret ? ret : request_ret; | |
1438 | } | |
1439 | ||
1440 | assert(extent.length); | |
1441 | *pnum = extent.length; | |
1442 | *map = offset; | |
1443 | *file = bs; | |
1444 | return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) | | |
1445 | (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) | | |
1446 | BDRV_BLOCK_OFFSET_VALID; | |
1447 | } | |
1448 | ||
e99754b4 ML |
1449 | static int nbd_client_reopen_prepare(BDRVReopenState *state, |
1450 | BlockReopenQueue *queue, Error **errp) | |
1451 | { | |
1452 | BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque; | |
1453 | ||
1454 | if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) { | |
1455 | error_setg(errp, "Can't reopen read-only NBD mount as read/write"); | |
1456 | return -EACCES; | |
1457 | } | |
1458 | return 0; | |
1459 | } | |
1460 | ||
fee091cd LS |
1461 | static void nbd_yank(void *opaque) |
1462 | { | |
1463 | BlockDriverState *bs = opaque; | |
1464 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
1465 | ||
dba5156c | 1466 | QEMU_LOCK_GUARD(&s->requests_lock); |
7d5b0d68 | 1467 | qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); |
dba5156c | 1468 | s->state = NBD_CLIENT_QUIT; |
fee091cd LS |
1469 | } |
1470 | ||
86f8cdf3 VSO |
1471 | static void nbd_client_close(BlockDriverState *bs) |
1472 | { | |
611ae1d7 | 1473 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
297365b4 | 1474 | NBDRequest request = { .type = NBD_CMD_DISC, .mode = s->info.mode }; |
86f8cdf3 | 1475 | |
f7651539 VSO |
1476 | if (s->ioc) { |
1477 | nbd_send_request(s->ioc, &request); | |
1478 | } | |
86f8cdf3 VSO |
1479 | |
1480 | nbd_teardown_connection(bs); | |
1481 | } | |
1482 | ||
86f8cdf3 | 1483 | |
8f071c9d VSO |
1484 | /* |
1485 | * Parse nbd_open options | |
1486 | */ | |
86f8cdf3 | 1487 | |
f53a1feb | 1488 | static int nbd_parse_uri(const char *filename, QDict *options) |
1d7d2a9d PB |
1489 | { |
1490 | URI *uri; | |
1491 | const char *p; | |
1492 | QueryParams *qp = NULL; | |
1493 | int ret = 0; | |
f53a1feb | 1494 | bool is_unix; |
1d7d2a9d PB |
1495 | |
1496 | uri = uri_parse(filename); | |
1497 | if (!uri) { | |
1498 | return -EINVAL; | |
1499 | } | |
1500 | ||
1501 | /* transport */ | |
f69165a8 | 1502 | if (!g_strcmp0(uri->scheme, "nbd")) { |
f53a1feb | 1503 | is_unix = false; |
f69165a8 | 1504 | } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) { |
f53a1feb | 1505 | is_unix = false; |
f69165a8 | 1506 | } else if (!g_strcmp0(uri->scheme, "nbd+unix")) { |
f53a1feb | 1507 | is_unix = true; |
1d7d2a9d PB |
1508 | } else { |
1509 | ret = -EINVAL; | |
1510 | goto out; | |
1511 | } | |
1512 | ||
2485f22f EB |
1513 | p = uri->path ? uri->path : ""; |
1514 | if (p[0] == '/') { | |
1515 | p++; | |
1516 | } | |
1d7d2a9d | 1517 | if (p[0]) { |
46f5ac20 | 1518 | qdict_put_str(options, "export", p); |
1d7d2a9d PB |
1519 | } |
1520 | ||
1521 | qp = query_params_parse(uri->query); | |
f53a1feb | 1522 | if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) { |
1d7d2a9d PB |
1523 | ret = -EINVAL; |
1524 | goto out; | |
1525 | } | |
1526 | ||
f53a1feb | 1527 | if (is_unix) { |
1d7d2a9d PB |
1528 | /* nbd+unix:///export?socket=path */ |
1529 | if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { | |
1530 | ret = -EINVAL; | |
1531 | goto out; | |
1532 | } | |
46f5ac20 EB |
1533 | qdict_put_str(options, "server.type", "unix"); |
1534 | qdict_put_str(options, "server.path", qp->p[0].value); | |
1d7d2a9d | 1535 | } else { |
23307908 | 1536 | QString *host; |
f84d431b HR |
1537 | char *port_str; |
1538 | ||
bebbf7fa | 1539 | /* nbd[+tcp]://host[:port]/export */ |
1d7d2a9d PB |
1540 | if (!uri->server) { |
1541 | ret = -EINVAL; | |
1542 | goto out; | |
1543 | } | |
f17c90be | 1544 | |
23307908 JT |
1545 | /* strip braces from literal IPv6 address */ |
1546 | if (uri->server[0] == '[') { | |
1547 | host = qstring_from_substr(uri->server, 1, | |
ba891d68 | 1548 | strlen(uri->server) - 1); |
23307908 JT |
1549 | } else { |
1550 | host = qstring_from_str(uri->server); | |
1551 | } | |
1552 | ||
46f5ac20 | 1553 | qdict_put_str(options, "server.type", "inet"); |
9445673e | 1554 | qdict_put(options, "server.host", host); |
f84d431b HR |
1555 | |
1556 | port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT); | |
46f5ac20 | 1557 | qdict_put_str(options, "server.port", port_str); |
f84d431b | 1558 | g_free(port_str); |
1d7d2a9d PB |
1559 | } |
1560 | ||
1561 | out: | |
1562 | if (qp) { | |
1563 | query_params_free(qp); | |
1564 | } | |
1565 | uri_free(uri); | |
1566 | return ret; | |
1567 | } | |
1568 | ||
48c38e0b HR |
1569 | static bool nbd_has_filename_options_conflict(QDict *options, Error **errp) |
1570 | { | |
1571 | const QDictEntry *e; | |
1572 | ||
1573 | for (e = qdict_first(options); e; e = qdict_next(options, e)) { | |
1574 | if (!strcmp(e->key, "host") || | |
1575 | !strcmp(e->key, "port") || | |
1576 | !strcmp(e->key, "path") || | |
491d6c7c HR |
1577 | !strcmp(e->key, "export") || |
1578 | strstart(e->key, "server.", NULL)) | |
48c38e0b HR |
1579 | { |
1580 | error_setg(errp, "Option '%s' cannot be used with a file name", | |
1581 | e->key); | |
1582 | return true; | |
1583 | } | |
1584 | } | |
1585 | ||
1586 | return false; | |
1587 | } | |
1588 | ||
6963a30d KW |
1589 | static void nbd_parse_filename(const char *filename, QDict *options, |
1590 | Error **errp) | |
75818250 | 1591 | { |
df18c04e | 1592 | g_autofree char *file = NULL; |
33897dc7 NT |
1593 | char *export_name; |
1594 | const char *host_spec; | |
75818250 | 1595 | const char *unixpath; |
75818250 | 1596 | |
48c38e0b | 1597 | if (nbd_has_filename_options_conflict(options, errp)) { |
681e7ad0 KW |
1598 | return; |
1599 | } | |
1600 | ||
1d7d2a9d | 1601 | if (strstr(filename, "://")) { |
6963a30d KW |
1602 | int ret = nbd_parse_uri(filename, options); |
1603 | if (ret < 0) { | |
1604 | error_setg(errp, "No valid URL specified"); | |
1605 | } | |
1606 | return; | |
1d7d2a9d PB |
1607 | } |
1608 | ||
7267c094 | 1609 | file = g_strdup(filename); |
1d45f8b5 | 1610 | |
33897dc7 NT |
1611 | export_name = strstr(file, EN_OPTSTR); |
1612 | if (export_name) { | |
1613 | if (export_name[strlen(EN_OPTSTR)] == 0) { | |
df18c04e | 1614 | return; |
1d45f8b5 | 1615 | } |
33897dc7 NT |
1616 | export_name[0] = 0; /* truncate 'file' */ |
1617 | export_name += strlen(EN_OPTSTR); | |
f53a1feb | 1618 | |
46f5ac20 | 1619 | qdict_put_str(options, "export", export_name); |
1d45f8b5 LV |
1620 | } |
1621 | ||
33897dc7 NT |
1622 | /* extract the host_spec - fail if it's not nbd:... */ |
1623 | if (!strstart(file, "nbd:", &host_spec)) { | |
6963a30d | 1624 | error_setg(errp, "File name string for NBD must start with 'nbd:'"); |
df18c04e | 1625 | return; |
1d45f8b5 | 1626 | } |
75818250 | 1627 | |
f53a1feb | 1628 | if (!*host_spec) { |
df18c04e | 1629 | return; |
f53a1feb KW |
1630 | } |
1631 | ||
33897dc7 NT |
1632 | /* are we a UNIX or TCP socket? */ |
1633 | if (strstart(host_spec, "unix:", &unixpath)) { | |
46f5ac20 EB |
1634 | qdict_put_str(options, "server.type", "unix"); |
1635 | qdict_put_str(options, "server.path", unixpath); | |
75818250 | 1636 | } else { |
0785bd7a | 1637 | InetSocketAddress *addr = g_new(InetSocketAddress, 1); |
f53a1feb | 1638 | |
0785bd7a MA |
1639 | if (inet_parse(addr, host_spec, errp)) { |
1640 | goto out_inet; | |
f17c90be | 1641 | } |
75818250 | 1642 | |
46f5ac20 EB |
1643 | qdict_put_str(options, "server.type", "inet"); |
1644 | qdict_put_str(options, "server.host", addr->host); | |
1645 | qdict_put_str(options, "server.port", addr->port); | |
0785bd7a | 1646 | out_inet: |
f53a1feb KW |
1647 | qapi_free_InetSocketAddress(addr); |
1648 | } | |
f53a1feb KW |
1649 | } |
1650 | ||
491d6c7c HR |
1651 | static bool nbd_process_legacy_socket_options(QDict *output_options, |
1652 | QemuOpts *legacy_opts, | |
1653 | Error **errp) | |
f53a1feb | 1654 | { |
491d6c7c HR |
1655 | const char *path = qemu_opt_get(legacy_opts, "path"); |
1656 | const char *host = qemu_opt_get(legacy_opts, "host"); | |
1657 | const char *port = qemu_opt_get(legacy_opts, "port"); | |
1658 | const QDictEntry *e; | |
f53a1feb | 1659 | |
491d6c7c HR |
1660 | if (!path && !host && !port) { |
1661 | return true; | |
1662 | } | |
03504d05 | 1663 | |
491d6c7c HR |
1664 | for (e = qdict_first(output_options); e; e = qdict_next(output_options, e)) |
1665 | { | |
1666 | if (strstart(e->key, "server.", NULL)) { | |
1667 | error_setg(errp, "Cannot use 'server' and path/host/port at the " | |
1668 | "same time"); | |
1669 | return false; | |
681e7ad0 | 1670 | } |
33897dc7 | 1671 | } |
491d6c7c HR |
1672 | |
1673 | if (path && host) { | |
1674 | error_setg(errp, "path and host may not be used at the same time"); | |
1675 | return false; | |
1676 | } else if (path) { | |
1677 | if (port) { | |
1678 | error_setg(errp, "port may not be used without host"); | |
1679 | return false; | |
1680 | } | |
1681 | ||
46f5ac20 EB |
1682 | qdict_put_str(output_options, "server.type", "unix"); |
1683 | qdict_put_str(output_options, "server.path", path); | |
491d6c7c | 1684 | } else if (host) { |
46f5ac20 EB |
1685 | qdict_put_str(output_options, "server.type", "inet"); |
1686 | qdict_put_str(output_options, "server.host", host); | |
1687 | qdict_put_str(output_options, "server.port", | |
1688 | port ?: stringify(NBD_DEFAULT_PORT)); | |
442045cb | 1689 | } |
f53a1feb | 1690 | |
491d6c7c HR |
1691 | return true; |
1692 | } | |
f53a1feb | 1693 | |
62cf396b MA |
1694 | static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options, |
1695 | Error **errp) | |
491d6c7c | 1696 | { |
62cf396b | 1697 | SocketAddress *saddr = NULL; |
491d6c7c | 1698 | QDict *addr = NULL; |
491d6c7c | 1699 | Visitor *iv = NULL; |
491d6c7c HR |
1700 | |
1701 | qdict_extract_subqdict(options, &addr, "server."); | |
1702 | if (!qdict_size(addr)) { | |
1703 | error_setg(errp, "NBD server address missing"); | |
1704 | goto done; | |
f53a1feb KW |
1705 | } |
1706 | ||
af91062e MA |
1707 | iv = qobject_input_visitor_new_flat_confused(addr, errp); |
1708 | if (!iv) { | |
491d6c7c HR |
1709 | goto done; |
1710 | } | |
bebbf7fa | 1711 | |
af175e85 | 1712 | if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) { |
491d6c7c HR |
1713 | goto done; |
1714 | } | |
7a5ed437 | 1715 | |
6cc702be VSO |
1716 | if (socket_address_parse_named_fd(saddr, errp) < 0) { |
1717 | qapi_free_SocketAddress(saddr); | |
1718 | saddr = NULL; | |
1719 | goto done; | |
1720 | } | |
1721 | ||
491d6c7c | 1722 | done: |
cb3e7f08 | 1723 | qobject_unref(addr); |
491d6c7c | 1724 | visit_free(iv); |
7a5ed437 | 1725 | return saddr; |
33897dc7 | 1726 | } |
1d45f8b5 | 1727 | |
75822a12 DB |
1728 | static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp) |
1729 | { | |
1730 | Object *obj; | |
1731 | QCryptoTLSCreds *creds; | |
1732 | ||
1733 | obj = object_resolve_path_component( | |
1734 | object_get_objects_root(), id); | |
1735 | if (!obj) { | |
1736 | error_setg(errp, "No TLS credentials with id '%s'", | |
1737 | id); | |
1738 | return NULL; | |
1739 | } | |
1740 | creds = (QCryptoTLSCreds *) | |
1741 | object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS); | |
1742 | if (!creds) { | |
1743 | error_setg(errp, "Object with id '%s' is not TLS credentials", | |
1744 | id); | |
1745 | return NULL; | |
1746 | } | |
1747 | ||
7b3b6168 PMD |
1748 | if (!qcrypto_tls_creds_check_endpoint(creds, |
1749 | QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT, | |
1750 | errp)) { | |
75822a12 DB |
1751 | return NULL; |
1752 | } | |
1753 | object_ref(obj); | |
1754 | return creds; | |
1755 | } | |
1756 | ||
1757 | ||
7ccc44fd HR |
1758 | static QemuOptsList nbd_runtime_opts = { |
1759 | .name = "nbd", | |
1760 | .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head), | |
1761 | .desc = { | |
1762 | { | |
1763 | .name = "host", | |
1764 | .type = QEMU_OPT_STRING, | |
1765 | .help = "TCP host to connect to", | |
1766 | }, | |
1767 | { | |
1768 | .name = "port", | |
1769 | .type = QEMU_OPT_STRING, | |
1770 | .help = "TCP port to connect to", | |
1771 | }, | |
1772 | { | |
1773 | .name = "path", | |
1774 | .type = QEMU_OPT_STRING, | |
1775 | .help = "Unix socket path to connect to", | |
1776 | }, | |
1777 | { | |
1778 | .name = "export", | |
1779 | .type = QEMU_OPT_STRING, | |
1780 | .help = "Name of the NBD export to open", | |
1781 | }, | |
1782 | { | |
1783 | .name = "tls-creds", | |
1784 | .type = QEMU_OPT_STRING, | |
1785 | .help = "ID of the TLS credentials to use", | |
1786 | }, | |
a0cd6d29 DB |
1787 | { |
1788 | .name = "tls-hostname", | |
1789 | .type = QEMU_OPT_STRING, | |
1790 | .help = "Override hostname for validating TLS x509 certificate", | |
1791 | }, | |
216ee365 EB |
1792 | { |
1793 | .name = "x-dirty-bitmap", | |
1794 | .type = QEMU_OPT_STRING, | |
1795 | .help = "experimental: expose named dirty bitmap in place of " | |
1796 | "block status", | |
1797 | }, | |
b172ae2e VSO |
1798 | { |
1799 | .name = "reconnect-delay", | |
1800 | .type = QEMU_OPT_NUMBER, | |
1801 | .help = "On an unexpected disconnect, the nbd client tries to " | |
1802 | "connect again until succeeding or encountering a serious " | |
1803 | "error. During the first @reconnect-delay seconds, all " | |
1804 | "requests are paused and will be rerun on a successful " | |
1805 | "reconnect. After that time, any delayed requests and all " | |
1806 | "future requests before a successful reconnect will " | |
1807 | "immediately fail. Default 0", | |
1808 | }, | |
be16b8bf VSO |
1809 | { |
1810 | .name = "open-timeout", | |
1811 | .type = QEMU_OPT_NUMBER, | |
1812 | .help = "In seconds. If zero, the nbd driver tries the connection " | |
1813 | "only once, and fails to open if the connection fails. " | |
1814 | "If non-zero, the nbd driver will repeat connection " | |
1815 | "attempts until successful or until @open-timeout seconds " | |
1816 | "have elapsed. Default 0", | |
1817 | }, | |
c4365735 | 1818 | { /* end of list */ } |
7ccc44fd HR |
1819 | }, |
1820 | }; | |
1821 | ||
8f071c9d VSO |
1822 | static int nbd_process_options(BlockDriverState *bs, QDict *options, |
1823 | Error **errp) | |
33897dc7 NT |
1824 | { |
1825 | BDRVNBDState *s = bs->opaque; | |
8f071c9d | 1826 | QemuOpts *opts; |
75822a12 | 1827 | int ret = -EINVAL; |
ae255e52 | 1828 | |
7ccc44fd | 1829 | opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 1830 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
7ccc44fd HR |
1831 | goto error; |
1832 | } | |
1833 | ||
62cf396b | 1834 | /* Translate @host, @port, and @path to a SocketAddress */ |
491d6c7c HR |
1835 | if (!nbd_process_legacy_socket_options(options, opts, errp)) { |
1836 | goto error; | |
1837 | } | |
1838 | ||
33897dc7 | 1839 | /* Pop the config into our state object. Exit if invalid. */ |
491d6c7c HR |
1840 | s->saddr = nbd_config(s, options, errp); |
1841 | if (!s->saddr) { | |
75822a12 DB |
1842 | goto error; |
1843 | } | |
1844 | ||
491d6c7c | 1845 | s->export = g_strdup(qemu_opt_get(opts, "export")); |
93676c88 EB |
1846 | if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) { |
1847 | error_setg(errp, "export name too long to send to server"); | |
1848 | goto error; | |
1849 | } | |
491d6c7c | 1850 | |
03504d05 HR |
1851 | s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds")); |
1852 | if (s->tlscredsid) { | |
8f071c9d VSO |
1853 | s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp); |
1854 | if (!s->tlscreds) { | |
75822a12 DB |
1855 | goto error; |
1856 | } | |
1857 | ||
a0cd6d29 | 1858 | s->tlshostname = g_strdup(qemu_opt_get(opts, "tls-hostname")); |
e8ae8b1a DB |
1859 | if (!s->tlshostname && |
1860 | s->saddr->type == SOCKET_ADDRESS_TYPE_INET) { | |
a0cd6d29 DB |
1861 | s->tlshostname = g_strdup(s->saddr->u.inet.host); |
1862 | } | |
33897dc7 NT |
1863 | } |
1864 | ||
8f071c9d | 1865 | s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap")); |
93676c88 EB |
1866 | if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) { |
1867 | error_setg(errp, "x-dirty-bitmap query too long to send to server"); | |
1868 | goto error; | |
1869 | } | |
1870 | ||
8f071c9d | 1871 | s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0); |
be16b8bf | 1872 | s->open_timeout = qemu_opt_get_number(opts, "open-timeout", 0); |
8f071c9d VSO |
1873 | |
1874 | ret = 0; | |
d42f78e9 | 1875 | |
75822a12 | 1876 | error: |
7ccc44fd | 1877 | qemu_opts_del(opts); |
75822a12 | 1878 | return ret; |
e183ef75 PB |
1879 | } |
1880 | ||
8f071c9d VSO |
1881 | static int nbd_open(BlockDriverState *bs, QDict *options, int flags, |
1882 | Error **errp) | |
1883 | { | |
1884 | int ret; | |
1885 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
1886 | ||
8f071c9d | 1887 | s->bs = bs; |
ee19d953 | 1888 | qemu_mutex_init(&s->requests_lock); |
8f071c9d | 1889 | qemu_co_queue_init(&s->free_sema); |
ee19d953 | 1890 | qemu_co_mutex_init(&s->send_mutex); |
4ddb5d2f | 1891 | qemu_co_mutex_init(&s->receive_mutex); |
8f071c9d | 1892 | |
fee091cd LS |
1893 | if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) { |
1894 | return -EEXIST; | |
1895 | } | |
1896 | ||
bbba1c37 VSO |
1897 | ret = nbd_process_options(bs, options, errp); |
1898 | if (ret < 0) { | |
1899 | goto fail; | |
1900 | } | |
1901 | ||
6d2b0332 | 1902 | s->conn = nbd_client_connection_new(s->saddr, true, s->export, |
046f98d0 DB |
1903 | s->x_dirty_bitmap, s->tlscreds, |
1904 | s->tlshostname); | |
e8b35bf5 | 1905 | |
be16b8bf VSO |
1906 | if (s->open_timeout) { |
1907 | nbd_client_connection_enable_retry(s->conn); | |
1908 | open_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + | |
1909 | s->open_timeout * NANOSECONDS_PER_SECOND); | |
1910 | } | |
1911 | ||
4ddb5d2f | 1912 | s->state = NBD_CLIENT_CONNECTING_WAIT; |
8610b449 | 1913 | ret = nbd_do_establish_connection(bs, true, errp); |
8f071c9d | 1914 | if (ret < 0) { |
bbba1c37 | 1915 | goto fail; |
8f071c9d | 1916 | } |
8f071c9d | 1917 | |
717be964 HR |
1918 | /* |
1919 | * The connect attempt is done, so we no longer need this timer. | |
1920 | * Delete it, because we do not want it to be around when this node | |
1921 | * is drained or closed. | |
1922 | */ | |
1923 | open_timer_del(s); | |
1924 | ||
4ddb5d2f | 1925 | nbd_client_connection_enable_retry(s->conn); |
8f071c9d VSO |
1926 | |
1927 | return 0; | |
bbba1c37 VSO |
1928 | |
1929 | fail: | |
717be964 | 1930 | open_timer_del(s); |
bbba1c37 VSO |
1931 | nbd_clear_bdrvstate(bs); |
1932 | return ret; | |
8f071c9d VSO |
1933 | } |
1934 | ||
fa21e6fa DL |
1935 | static void nbd_refresh_limits(BlockDriverState *bs, Error **errp) |
1936 | { | |
611ae1d7 | 1937 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; |
fd8d372d | 1938 | uint32_t min = s->info.min_block; |
081dd1fe EB |
1939 | uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block); |
1940 | ||
7da537f7 EB |
1941 | /* |
1942 | * If the server did not advertise an alignment: | |
1943 | * - a size that is not sector-aligned implies that an alignment | |
1944 | * of 1 can be used to access those tail bytes | |
1945 | * - advertisement of block status requires an alignment of 1, so | |
1946 | * that we don't violate block layer constraints that block | |
1947 | * status is always aligned (as we can't control whether the | |
1948 | * server will report sub-sector extents, such as a hole at EOF | |
1949 | * on an unaligned POSIX file) | |
1950 | * - otherwise, assume the server is so old that we are safer avoiding | |
1951 | * sub-sector requests | |
1952 | */ | |
1953 | if (!min) { | |
1954 | min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) || | |
1955 | s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE; | |
1956 | } | |
1957 | ||
1958 | bs->bl.request_alignment = min; | |
714eb0db | 1959 | bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min); |
081dd1fe EB |
1960 | bs->bl.max_pwrite_zeroes = max; |
1961 | bs->bl.max_transfer = max; | |
1962 | ||
b2578459 EB |
1963 | /* |
1964 | * Assume that if the server supports extended headers, it also | |
1965 | * supports unlimited size zero and trim commands. | |
1966 | */ | |
1967 | if (s->info.mode >= NBD_MODE_EXTENDED) { | |
1968 | bs->bl.max_pdiscard = bs->bl.max_pwrite_zeroes = 0; | |
1969 | } | |
1970 | ||
081dd1fe EB |
1971 | if (s->info.opt_block && |
1972 | s->info.opt_block > bs->bl.opt_transfer) { | |
1973 | bs->bl.opt_transfer = s->info.opt_block; | |
1974 | } | |
fa21e6fa DL |
1975 | } |
1976 | ||
75818250 TS |
1977 | static void nbd_close(BlockDriverState *bs) |
1978 | { | |
f53a829b | 1979 | nbd_client_close(bs); |
bbba1c37 | 1980 | nbd_clear_bdrvstate(bs); |
75818250 TS |
1981 | } |
1982 | ||
a2b333c0 NS |
1983 | /* |
1984 | * NBD cannot truncate, but if the caller asks to truncate to the same size, or | |
1985 | * to a smaller size with exact=false, there is no reason to fail the | |
1986 | * operation. | |
1987 | * | |
1988 | * Preallocation mode is ignored since it does not seems useful to fail when | |
1989 | * we never change anything. | |
1990 | */ | |
1991 | static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset, | |
1992 | bool exact, PreallocMode prealloc, | |
1993 | BdrvRequestFlags flags, Error **errp) | |
1994 | { | |
1995 | BDRVNBDState *s = bs->opaque; | |
1996 | ||
1997 | if (offset != s->info.size && exact) { | |
1998 | error_setg(errp, "Cannot resize NBD nodes"); | |
1999 | return -ENOTSUP; | |
2000 | } | |
2001 | ||
2002 | if (offset > s->info.size) { | |
2003 | error_setg(errp, "Cannot grow NBD nodes"); | |
2004 | return -EINVAL; | |
2005 | } | |
2006 | ||
2007 | return 0; | |
2008 | } | |
2009 | ||
c86422c5 | 2010 | static int64_t coroutine_fn nbd_co_getlength(BlockDriverState *bs) |
75818250 TS |
2011 | { |
2012 | BDRVNBDState *s = bs->opaque; | |
2013 | ||
611ae1d7 | 2014 | return s->info.size; |
75818250 TS |
2015 | } |
2016 | ||
998b3a1e | 2017 | static void nbd_refresh_filename(BlockDriverState *bs) |
2019d68b | 2018 | { |
03504d05 | 2019 | BDRVNBDState *s = bs->opaque; |
491d6c7c | 2020 | const char *host = NULL, *port = NULL, *path = NULL; |
5c86bdf1 | 2021 | size_t len = 0; |
491d6c7c | 2022 | |
62cf396b | 2023 | if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) { |
9445673e | 2024 | const InetSocketAddress *inet = &s->saddr->u.inet; |
491d6c7c HR |
2025 | if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) { |
2026 | host = inet->host; | |
2027 | port = inet->port; | |
2028 | } | |
62cf396b | 2029 | } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) { |
9445673e MA |
2030 | path = s->saddr->u.q_unix.path; |
2031 | } /* else can't represent as pseudo-filename */ | |
2019d68b | 2032 | |
491d6c7c | 2033 | if (path && s->export) { |
5c86bdf1 EB |
2034 | len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), |
2035 | "nbd+unix:///%s?socket=%s", s->export, path); | |
491d6c7c | 2036 | } else if (path && !s->export) { |
5c86bdf1 EB |
2037 | len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), |
2038 | "nbd+unix://?socket=%s", path); | |
491d6c7c | 2039 | } else if (host && s->export) { |
5c86bdf1 EB |
2040 | len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), |
2041 | "nbd://%s:%s/%s", host, port, s->export); | |
491d6c7c | 2042 | } else if (host && !s->export) { |
5c86bdf1 EB |
2043 | len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), |
2044 | "nbd://%s:%s", host, port); | |
2045 | } | |
00d69986 | 2046 | if (len >= sizeof(bs->exact_filename)) { |
5c86bdf1 EB |
2047 | /* Name is too long to represent exactly, so leave it empty. */ |
2048 | bs->exact_filename[0] = '\0'; | |
ec0de768 | 2049 | } |
2019d68b HR |
2050 | } |
2051 | ||
8a6239c0 HR |
2052 | static char *nbd_dirname(BlockDriverState *bs, Error **errp) |
2053 | { | |
2054 | /* The generic bdrv_dirname() implementation is able to work out some | |
2055 | * directory name for NBD nodes, but that would be wrong. So far there is no | |
2056 | * specification for how "export paths" would work, so NBD does not have | |
2057 | * directory names. */ | |
2058 | error_setg(errp, "Cannot generate a base directory for NBD nodes"); | |
2059 | return NULL; | |
2060 | } | |
2061 | ||
2654267c HR |
2062 | static const char *const nbd_strong_runtime_opts[] = { |
2063 | "path", | |
2064 | "host", | |
2065 | "port", | |
2066 | "export", | |
2067 | "tls-creds", | |
a0cd6d29 | 2068 | "tls-hostname", |
2654267c HR |
2069 | "server.", |
2070 | ||
2071 | NULL | |
2072 | }; | |
2073 | ||
c4f7f24e VSO |
2074 | static void nbd_cancel_in_flight(BlockDriverState *bs) |
2075 | { | |
2076 | BDRVNBDState *s = (BDRVNBDState *)bs->opaque; | |
2077 | ||
2078 | reconnect_delay_timer_del(s); | |
2079 | ||
ee19d953 | 2080 | qemu_mutex_lock(&s->requests_lock); |
c4f7f24e VSO |
2081 | if (s->state == NBD_CLIENT_CONNECTING_WAIT) { |
2082 | s->state = NBD_CLIENT_CONNECTING_NOWAIT; | |
c4f7f24e | 2083 | } |
ee19d953 | 2084 | qemu_mutex_unlock(&s->requests_lock); |
4ddb5d2f VSO |
2085 | |
2086 | nbd_co_establish_connection_cancel(s->conn); | |
c4f7f24e VSO |
2087 | } |
2088 | ||
e15f3a66 HR |
2089 | static void nbd_attach_aio_context(BlockDriverState *bs, |
2090 | AioContext *new_context) | |
2091 | { | |
2092 | BDRVNBDState *s = bs->opaque; | |
2093 | ||
2094 | /* The open_timer is used only during nbd_open() */ | |
2095 | assert(!s->open_timer); | |
2096 | ||
2097 | /* | |
2098 | * The reconnect_delay_timer is scheduled in I/O paths when the | |
2099 | * connection is lost, to cancel the reconnection attempt after a | |
2100 | * given time. Once this attempt is done (successfully or not), | |
2101 | * nbd_reconnect_attempt() ensures the timer is deleted before the | |
2102 | * respective I/O request is resumed. | |
2103 | * Since the AioContext can only be changed when a node is drained, | |
2104 | * the reconnect_delay_timer cannot be active here. | |
2105 | */ | |
2106 | assert(!s->reconnect_delay_timer); | |
e15f3a66 HR |
2107 | } |
2108 | ||
2109 | static void nbd_detach_aio_context(BlockDriverState *bs) | |
2110 | { | |
2111 | BDRVNBDState *s = bs->opaque; | |
2112 | ||
2113 | assert(!s->open_timer); | |
2114 | assert(!s->reconnect_delay_timer); | |
e15f3a66 HR |
2115 | } |
2116 | ||
5efa9d5a | 2117 | static BlockDriver bdrv_nbd = { |
69447cd8 SH |
2118 | .format_name = "nbd", |
2119 | .protocol_name = "nbd", | |
2120 | .instance_size = sizeof(BDRVNBDState), | |
2121 | .bdrv_parse_filename = nbd_parse_filename, | |
5a5e7f8c ML |
2122 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
2123 | .create_opts = &bdrv_create_opts_simple, | |
69447cd8 | 2124 | .bdrv_file_open = nbd_open, |
e99754b4 | 2125 | .bdrv_reopen_prepare = nbd_client_reopen_prepare, |
70c4fb26 EB |
2126 | .bdrv_co_preadv = nbd_client_co_preadv, |
2127 | .bdrv_co_pwritev = nbd_client_co_pwritev, | |
fa778fff | 2128 | .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, |
69447cd8 | 2129 | .bdrv_close = nbd_close, |
5d934513 | 2130 | .bdrv_co_flush_to_os = nbd_client_co_flush, |
447e57c3 | 2131 | .bdrv_co_pdiscard = nbd_client_co_pdiscard, |
fa21e6fa | 2132 | .bdrv_refresh_limits = nbd_refresh_limits, |
a2b333c0 | 2133 | .bdrv_co_truncate = nbd_co_truncate, |
c86422c5 | 2134 | .bdrv_co_getlength = nbd_co_getlength, |
2019d68b | 2135 | .bdrv_refresh_filename = nbd_refresh_filename, |
78a33ab5 | 2136 | .bdrv_co_block_status = nbd_client_co_block_status, |
8a6239c0 | 2137 | .bdrv_dirname = nbd_dirname, |
2654267c | 2138 | .strong_runtime_opts = nbd_strong_runtime_opts, |
c4f7f24e | 2139 | .bdrv_cancel_in_flight = nbd_cancel_in_flight, |
e15f3a66 HR |
2140 | |
2141 | .bdrv_attach_aio_context = nbd_attach_aio_context, | |
2142 | .bdrv_detach_aio_context = nbd_detach_aio_context, | |
1d7d2a9d PB |
2143 | }; |
2144 | ||
2145 | static BlockDriver bdrv_nbd_tcp = { | |
69447cd8 SH |
2146 | .format_name = "nbd", |
2147 | .protocol_name = "nbd+tcp", | |
2148 | .instance_size = sizeof(BDRVNBDState), | |
2149 | .bdrv_parse_filename = nbd_parse_filename, | |
5a5e7f8c ML |
2150 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
2151 | .create_opts = &bdrv_create_opts_simple, | |
69447cd8 | 2152 | .bdrv_file_open = nbd_open, |
e99754b4 | 2153 | .bdrv_reopen_prepare = nbd_client_reopen_prepare, |
70c4fb26 EB |
2154 | .bdrv_co_preadv = nbd_client_co_preadv, |
2155 | .bdrv_co_pwritev = nbd_client_co_pwritev, | |
fa778fff | 2156 | .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, |
69447cd8 | 2157 | .bdrv_close = nbd_close, |
5d934513 | 2158 | .bdrv_co_flush_to_os = nbd_client_co_flush, |
447e57c3 | 2159 | .bdrv_co_pdiscard = nbd_client_co_pdiscard, |
fa21e6fa | 2160 | .bdrv_refresh_limits = nbd_refresh_limits, |
a2b333c0 | 2161 | .bdrv_co_truncate = nbd_co_truncate, |
c86422c5 | 2162 | .bdrv_co_getlength = nbd_co_getlength, |
2019d68b | 2163 | .bdrv_refresh_filename = nbd_refresh_filename, |
78a33ab5 | 2164 | .bdrv_co_block_status = nbd_client_co_block_status, |
8a6239c0 | 2165 | .bdrv_dirname = nbd_dirname, |
2654267c | 2166 | .strong_runtime_opts = nbd_strong_runtime_opts, |
c4f7f24e | 2167 | .bdrv_cancel_in_flight = nbd_cancel_in_flight, |
e15f3a66 HR |
2168 | |
2169 | .bdrv_attach_aio_context = nbd_attach_aio_context, | |
2170 | .bdrv_detach_aio_context = nbd_detach_aio_context, | |
1d7d2a9d PB |
2171 | }; |
2172 | ||
2173 | static BlockDriver bdrv_nbd_unix = { | |
69447cd8 SH |
2174 | .format_name = "nbd", |
2175 | .protocol_name = "nbd+unix", | |
2176 | .instance_size = sizeof(BDRVNBDState), | |
2177 | .bdrv_parse_filename = nbd_parse_filename, | |
5a5e7f8c ML |
2178 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
2179 | .create_opts = &bdrv_create_opts_simple, | |
69447cd8 | 2180 | .bdrv_file_open = nbd_open, |
e99754b4 | 2181 | .bdrv_reopen_prepare = nbd_client_reopen_prepare, |
70c4fb26 EB |
2182 | .bdrv_co_preadv = nbd_client_co_preadv, |
2183 | .bdrv_co_pwritev = nbd_client_co_pwritev, | |
fa778fff | 2184 | .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, |
69447cd8 | 2185 | .bdrv_close = nbd_close, |
5d934513 | 2186 | .bdrv_co_flush_to_os = nbd_client_co_flush, |
447e57c3 | 2187 | .bdrv_co_pdiscard = nbd_client_co_pdiscard, |
fa21e6fa | 2188 | .bdrv_refresh_limits = nbd_refresh_limits, |
a2b333c0 | 2189 | .bdrv_co_truncate = nbd_co_truncate, |
c86422c5 | 2190 | .bdrv_co_getlength = nbd_co_getlength, |
2019d68b | 2191 | .bdrv_refresh_filename = nbd_refresh_filename, |
78a33ab5 | 2192 | .bdrv_co_block_status = nbd_client_co_block_status, |
8a6239c0 | 2193 | .bdrv_dirname = nbd_dirname, |
2654267c | 2194 | .strong_runtime_opts = nbd_strong_runtime_opts, |
c4f7f24e | 2195 | .bdrv_cancel_in_flight = nbd_cancel_in_flight, |
e15f3a66 HR |
2196 | |
2197 | .bdrv_attach_aio_context = nbd_attach_aio_context, | |
2198 | .bdrv_detach_aio_context = nbd_detach_aio_context, | |
75818250 | 2199 | }; |
5efa9d5a AL |
2200 | |
2201 | static void bdrv_nbd_init(void) | |
2202 | { | |
2203 | bdrv_register(&bdrv_nbd); | |
1d7d2a9d PB |
2204 | bdrv_register(&bdrv_nbd_tcp); |
2205 | bdrv_register(&bdrv_nbd_unix); | |
5efa9d5a AL |
2206 | } |
2207 | ||
2208 | block_init(bdrv_nbd_init); |