]>
Commit | Line | Data |
---|---|---|
2302c1ca MAL |
1 | /* |
2 | * QEMU Block driver for NBD | |
3 | * | |
b626b51a | 4 | * Copyright (C) 2016 Red Hat, Inc. |
2302c1ca MAL |
5 | * Copyright (C) 2008 Bull S.A.S. |
6 | * Author: Laurent Vivier <Laurent.Vivier@bull.net> | |
7 | * | |
8 | * Some parts: | |
9 | * Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws> | |
10 | * | |
11 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
12 | * of this software and associated documentation files (the "Software"), to deal | |
13 | * in the Software without restriction, including without limitation the rights | |
14 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
15 | * copies of the Software, and to permit persons to whom the Software is | |
16 | * furnished to do so, subject to the following conditions: | |
17 | * | |
18 | * The above copyright notice and this permission notice shall be included in | |
19 | * all copies or substantial portions of the Software. | |
20 | * | |
21 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
22 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
23 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
24 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
25 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
26 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
27 | * THE SOFTWARE. | |
28 | */ | |
29 | ||
80c71a24 | 30 | #include "qemu/osdep.h" |
d8b4bad8 VSO |
31 | |
32 | #include "trace.h" | |
be41c100 | 33 | #include "qapi/error.h" |
2302c1ca | 34 | #include "nbd-client.h" |
2302c1ca | 35 | |
cfa3ad63 EB |
36 | #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs)) |
37 | #define INDEX_TO_HANDLE(bs, index) ((index) ^ (uint64_t)(intptr_t)(bs)) | |
2302c1ca | 38 | |
07b1b99c | 39 | static void nbd_recv_coroutines_wake_all(NBDClientSession *s) |
69152c09 MAL |
40 | { |
41 | int i; | |
42 | ||
43 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { | |
40f4a218 SH |
44 | NBDClientRequest *req = &s->requests[i]; |
45 | ||
46 | if (req->coroutine && req->receiving) { | |
47 | aio_co_wake(req->coroutine); | |
69152c09 MAL |
48 | } |
49 | } | |
50 | } | |
51 | ||
f53a829b | 52 | static void nbd_teardown_connection(BlockDriverState *bs) |
4a41a2d6 | 53 | { |
10676b81 | 54 | NBDClientSession *client = nbd_get_client_session(bs); |
f53a829b | 55 | |
88ed4e1b | 56 | assert(client->ioc); |
064097d9 | 57 | |
4a41a2d6 | 58 | /* finish any pending coroutines */ |
064097d9 DB |
59 | qio_channel_shutdown(client->ioc, |
60 | QIO_CHANNEL_SHUTDOWN_BOTH, | |
61 | NULL); | |
bc5a0335 | 62 | BDRV_POLL_WHILE(bs, client->connection_co); |
4a41a2d6 | 63 | |
f53a829b | 64 | nbd_client_detach_aio_context(bs); |
064097d9 DB |
65 | object_unref(OBJECT(client->sioc)); |
66 | client->sioc = NULL; | |
67 | object_unref(OBJECT(client->ioc)); | |
68 | client->ioc = NULL; | |
4a41a2d6 SH |
69 | } |
70 | ||
bc5a0335 | 71 | static coroutine_fn void nbd_connection_entry(void *opaque) |
2302c1ca | 72 | { |
ff82911c | 73 | NBDClientSession *s = opaque; |
2302c1ca | 74 | uint64_t i; |
d0a18013 | 75 | int ret = 0; |
be41c100 | 76 | Error *local_err = NULL; |
2302c1ca | 77 | |
72b6ffc7 | 78 | while (!s->quit) { |
5ad81b49 KW |
79 | /* |
80 | * The NBD client can only really be considered idle when it has | |
81 | * yielded from qio_channel_readv_all_eof(), waiting for data. This is | |
82 | * the point where the additional scheduled coroutine entry happens | |
83 | * after nbd_client_attach_aio_context(). | |
84 | * | |
85 | * Therefore we keep an additional in_flight reference all the time and | |
86 | * only drop it temporarily here. | |
5ad81b49 | 87 | */ |
ff82911c | 88 | assert(s->reply.handle == 0); |
d3bd5b90 | 89 | ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, &local_err); |
5ad81b49 | 90 | |
08ace1d7 | 91 | if (local_err) { |
d8b4bad8 VSO |
92 | trace_nbd_read_reply_entry_fail(ret, error_get_pretty(local_err)); |
93 | error_free(local_err); | |
be41c100 | 94 | } |
a12a712a | 95 | if (ret <= 0) { |
ff82911c | 96 | break; |
2302c1ca | 97 | } |
2302c1ca | 98 | |
ff82911c PB |
99 | /* There's no need for a mutex on the receive side, because the |
100 | * handler acts as a synchronization point and ensures that only | |
101 | * one coroutine is called until the reply finishes. | |
102 | */ | |
103 | i = HANDLE_TO_INDEX(s, s->reply.handle); | |
40f4a218 SH |
104 | if (i >= MAX_NBD_REQUESTS || |
105 | !s->requests[i].coroutine || | |
d2febedb | 106 | !s->requests[i].receiving || |
f140e300 | 107 | (nbd_reply_is_structured(&s->reply) && !s->info.structured_reply)) |
d2febedb | 108 | { |
ff82911c PB |
109 | break; |
110 | } | |
2302c1ca | 111 | |
40f4a218 | 112 | /* We're woken up again by the request itself. Note that there |
bc5a0335 | 113 | * is no race between yielding and reentering connection_co. This |
ff82911c PB |
114 | * is because: |
115 | * | |
40f4a218 | 116 | * - if the request runs on the same AioContext, it is only |
ff82911c PB |
117 | * entered after we yield |
118 | * | |
40f4a218 | 119 | * - if the request runs on a different AioContext, reentering |
bc5a0335 | 120 | * connection_co happens through a bottom half, which can only |
ff82911c PB |
121 | * run after we yield. |
122 | */ | |
40f4a218 | 123 | aio_co_wake(s->requests[i].coroutine); |
ff82911c | 124 | qemu_coroutine_yield(); |
2302c1ca | 125 | } |
a12a712a | 126 | |
40f4a218 | 127 | s->quit = true; |
07b1b99c | 128 | nbd_recv_coroutines_wake_all(s); |
5ad81b49 KW |
129 | bdrv_dec_in_flight(s->bs); |
130 | ||
bc5a0335 | 131 | s->connection_co = NULL; |
4720cbee | 132 | aio_wait_kick(); |
2302c1ca MAL |
133 | } |
134 | ||
f53a829b | 135 | static int nbd_co_send_request(BlockDriverState *bs, |
ed2dd912 | 136 | NBDRequest *request, |
1e2a77a8 | 137 | QEMUIOVector *qiov) |
2302c1ca | 138 | { |
10676b81 | 139 | NBDClientSession *s = nbd_get_client_session(bs); |
030fa7f6 | 140 | int rc, i; |
2302c1ca MAL |
141 | |
142 | qemu_co_mutex_lock(&s->send_mutex); | |
6bdcc018 PB |
143 | while (s->in_flight == MAX_NBD_REQUESTS) { |
144 | qemu_co_queue_wait(&s->free_sema, &s->send_mutex); | |
145 | } | |
146 | s->in_flight++; | |
141cabe6 BW |
147 | |
148 | for (i = 0; i < MAX_NBD_REQUESTS; i++) { | |
40f4a218 | 149 | if (s->requests[i].coroutine == NULL) { |
141cabe6 BW |
150 | break; |
151 | } | |
152 | } | |
153 | ||
1c778ef7 | 154 | g_assert(qemu_in_coroutine()); |
141cabe6 | 155 | assert(i < MAX_NBD_REQUESTS); |
40f4a218 SH |
156 | |
157 | s->requests[i].coroutine = qemu_coroutine_self(); | |
f140e300 | 158 | s->requests[i].offset = request->from; |
40f4a218 SH |
159 | s->requests[i].receiving = false; |
160 | ||
141cabe6 | 161 | request->handle = INDEX_TO_HANDLE(s, i); |
064097d9 | 162 | |
72b6ffc7 | 163 | if (s->quit) { |
3c2d5183 SH |
164 | rc = -EIO; |
165 | goto err; | |
72b6ffc7 | 166 | } |
88ed4e1b | 167 | assert(s->ioc); |
064097d9 | 168 | |
2302c1ca | 169 | if (qiov) { |
064097d9 | 170 | qio_channel_set_cork(s->ioc, true); |
1c778ef7 | 171 | rc = nbd_send_request(s->ioc, request); |
72b6ffc7 | 172 | if (rc >= 0 && !s->quit) { |
030fa7f6 EB |
173 | if (qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov, |
174 | NULL) < 0) { | |
2302c1ca MAL |
175 | rc = -EIO; |
176 | } | |
a6934370 VSO |
177 | } else if (rc >= 0) { |
178 | rc = -EIO; | |
2302c1ca | 179 | } |
064097d9 | 180 | qio_channel_set_cork(s->ioc, false); |
2302c1ca | 181 | } else { |
1c778ef7 | 182 | rc = nbd_send_request(s->ioc, request); |
2302c1ca | 183 | } |
3c2d5183 SH |
184 | |
185 | err: | |
72b6ffc7 EB |
186 | if (rc < 0) { |
187 | s->quit = true; | |
3c2d5183 SH |
188 | s->requests[i].coroutine = NULL; |
189 | s->in_flight--; | |
190 | qemu_co_queue_next(&s->free_sema); | |
72b6ffc7 | 191 | } |
2302c1ca MAL |
192 | qemu_co_mutex_unlock(&s->send_mutex); |
193 | return rc; | |
194 | } | |
195 | ||
f140e300 VSO |
196 | static inline uint16_t payload_advance16(uint8_t **payload) |
197 | { | |
198 | *payload += 2; | |
199 | return lduw_be_p(*payload - 2); | |
200 | } | |
201 | ||
202 | static inline uint32_t payload_advance32(uint8_t **payload) | |
203 | { | |
204 | *payload += 4; | |
205 | return ldl_be_p(*payload - 4); | |
206 | } | |
207 | ||
208 | static inline uint64_t payload_advance64(uint8_t **payload) | |
209 | { | |
210 | *payload += 8; | |
211 | return ldq_be_p(*payload - 8); | |
212 | } | |
213 | ||
75d34eb9 EB |
214 | static int nbd_parse_offset_hole_payload(NBDClientSession *client, |
215 | NBDStructuredReplyChunk *chunk, | |
f140e300 VSO |
216 | uint8_t *payload, uint64_t orig_offset, |
217 | QEMUIOVector *qiov, Error **errp) | |
218 | { | |
219 | uint64_t offset; | |
220 | uint32_t hole_size; | |
221 | ||
222 | if (chunk->length != sizeof(offset) + sizeof(hole_size)) { | |
223 | error_setg(errp, "Protocol error: invalid payload for " | |
224 | "NBD_REPLY_TYPE_OFFSET_HOLE"); | |
225 | return -EINVAL; | |
226 | } | |
227 | ||
228 | offset = payload_advance64(&payload); | |
229 | hole_size = payload_advance32(&payload); | |
230 | ||
b4176cb3 | 231 | if (!hole_size || offset < orig_offset || hole_size > qiov->size || |
f140e300 VSO |
232 | offset > orig_offset + qiov->size - hole_size) { |
233 | error_setg(errp, "Protocol error: server sent chunk exceeding requested" | |
234 | " region"); | |
235 | return -EINVAL; | |
236 | } | |
75d34eb9 EB |
237 | if (client->info.min_block && |
238 | !QEMU_IS_ALIGNED(hole_size, client->info.min_block)) { | |
239 | trace_nbd_structured_read_compliance("hole"); | |
240 | } | |
f140e300 VSO |
241 | |
242 | qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size); | |
243 | ||
244 | return 0; | |
245 | } | |
246 | ||
78a33ab5 | 247 | /* nbd_parse_blockstatus_payload |
a39286dd EB |
248 | * Based on our request, we expect only one extent in reply, for the |
249 | * base:allocation context. | |
78a33ab5 VSO |
250 | */ |
251 | static int nbd_parse_blockstatus_payload(NBDClientSession *client, | |
252 | NBDStructuredReplyChunk *chunk, | |
253 | uint8_t *payload, uint64_t orig_length, | |
254 | NBDExtent *extent, Error **errp) | |
255 | { | |
256 | uint32_t context_id; | |
257 | ||
a39286dd EB |
258 | /* The server succeeded, so it must have sent [at least] one extent */ |
259 | if (chunk->length < sizeof(context_id) + sizeof(*extent)) { | |
78a33ab5 VSO |
260 | error_setg(errp, "Protocol error: invalid payload for " |
261 | "NBD_REPLY_TYPE_BLOCK_STATUS"); | |
262 | return -EINVAL; | |
263 | } | |
264 | ||
265 | context_id = payload_advance32(&payload); | |
2df94eb5 | 266 | if (client->info.context_id != context_id) { |
78a33ab5 VSO |
267 | error_setg(errp, "Protocol error: unexpected context id %d for " |
268 | "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context " | |
269 | "id is %d", context_id, | |
2df94eb5 | 270 | client->info.context_id); |
78a33ab5 VSO |
271 | return -EINVAL; |
272 | } | |
273 | ||
274 | extent->length = payload_advance32(&payload); | |
275 | extent->flags = payload_advance32(&payload); | |
276 | ||
737d3f52 | 277 | if (extent->length == 0) { |
78a33ab5 | 278 | error_setg(errp, "Protocol error: server sent status chunk with " |
737d3f52 | 279 | "zero length"); |
78a33ab5 VSO |
280 | return -EINVAL; |
281 | } | |
282 | ||
737d3f52 EB |
283 | /* |
284 | * A server sending unaligned block status is in violation of the | |
285 | * protocol, but as qemu-nbd 3.1 is such a server (at least for | |
286 | * POSIX files that are not a multiple of 512 bytes, since qemu | |
287 | * rounds files up to 512-byte multiples but lseek(SEEK_HOLE) | |
288 | * still sees an implicit hole beyond the real EOF), it's nicer to | |
289 | * work around the misbehaving server. If the request included | |
290 | * more than the final unaligned block, truncate it back to an | |
291 | * aligned result; if the request was only the final block, round | |
292 | * up to the full block and change the status to fully-allocated | |
293 | * (always a safe status, even if it loses information). | |
294 | */ | |
295 | if (client->info.min_block && !QEMU_IS_ALIGNED(extent->length, | |
296 | client->info.min_block)) { | |
297 | trace_nbd_parse_blockstatus_compliance("extent length is unaligned"); | |
298 | if (extent->length > client->info.min_block) { | |
299 | extent->length = QEMU_ALIGN_DOWN(extent->length, | |
300 | client->info.min_block); | |
301 | } else { | |
302 | extent->length = client->info.min_block; | |
303 | extent->flags = 0; | |
304 | } | |
305 | } | |
306 | ||
a39286dd EB |
307 | /* |
308 | * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have | |
309 | * sent us any more than one extent, nor should it have included | |
310 | * status beyond our request in that extent. However, it's easy | |
311 | * enough to ignore the server's noncompliance without killing the | |
312 | * connection; just ignore trailing extents, and clamp things to | |
313 | * the length of our request. | |
314 | */ | |
315 | if (chunk->length > sizeof(context_id) + sizeof(*extent)) { | |
316 | trace_nbd_parse_blockstatus_compliance("more than one extent"); | |
317 | } | |
acfd8f7a EB |
318 | if (extent->length > orig_length) { |
319 | extent->length = orig_length; | |
a39286dd | 320 | trace_nbd_parse_blockstatus_compliance("extent length too large"); |
acfd8f7a EB |
321 | } |
322 | ||
78a33ab5 VSO |
323 | return 0; |
324 | } | |
325 | ||
f140e300 VSO |
326 | /* nbd_parse_error_payload |
327 | * on success @errp contains message describing nbd error reply | |
328 | */ | |
329 | static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk, | |
330 | uint8_t *payload, int *request_ret, | |
331 | Error **errp) | |
332 | { | |
333 | uint32_t error; | |
334 | uint16_t message_size; | |
335 | ||
336 | assert(chunk->type & (1 << 15)); | |
337 | ||
338 | if (chunk->length < sizeof(error) + sizeof(message_size)) { | |
339 | error_setg(errp, | |
340 | "Protocol error: invalid payload for structured error"); | |
341 | return -EINVAL; | |
342 | } | |
343 | ||
344 | error = nbd_errno_to_system_errno(payload_advance32(&payload)); | |
345 | if (error == 0) { | |
e659fb3b | 346 | error_setg(errp, "Protocol error: server sent structured error chunk " |
f140e300 VSO |
347 | "with error = 0"); |
348 | return -EINVAL; | |
349 | } | |
350 | ||
351 | *request_ret = -error; | |
352 | message_size = payload_advance16(&payload); | |
353 | ||
354 | if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) { | |
e659fb3b | 355 | error_setg(errp, "Protocol error: server sent structured error chunk " |
f140e300 VSO |
356 | "with incorrect message size"); |
357 | return -EINVAL; | |
358 | } | |
359 | ||
360 | /* TODO: Add a trace point to mention the server complaint */ | |
361 | ||
362 | /* TODO handle ERROR_OFFSET */ | |
363 | ||
364 | return 0; | |
365 | } | |
366 | ||
367 | static int nbd_co_receive_offset_data_payload(NBDClientSession *s, | |
368 | uint64_t orig_offset, | |
369 | QEMUIOVector *qiov, Error **errp) | |
370 | { | |
371 | QEMUIOVector sub_qiov; | |
372 | uint64_t offset; | |
373 | size_t data_size; | |
374 | int ret; | |
375 | NBDStructuredReplyChunk *chunk = &s->reply.structured; | |
376 | ||
377 | assert(nbd_reply_is_structured(&s->reply)); | |
378 | ||
b4176cb3 EB |
379 | /* The NBD spec requires at least one byte of payload */ |
380 | if (chunk->length <= sizeof(offset)) { | |
f140e300 VSO |
381 | error_setg(errp, "Protocol error: invalid payload for " |
382 | "NBD_REPLY_TYPE_OFFSET_DATA"); | |
383 | return -EINVAL; | |
384 | } | |
385 | ||
e6798f06 | 386 | if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) { |
f140e300 VSO |
387 | return -EIO; |
388 | } | |
f140e300 VSO |
389 | |
390 | data_size = chunk->length - sizeof(offset); | |
b4176cb3 | 391 | assert(data_size); |
f140e300 VSO |
392 | if (offset < orig_offset || data_size > qiov->size || |
393 | offset > orig_offset + qiov->size - data_size) { | |
394 | error_setg(errp, "Protocol error: server sent chunk exceeding requested" | |
395 | " region"); | |
396 | return -EINVAL; | |
397 | } | |
75d34eb9 EB |
398 | if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) { |
399 | trace_nbd_structured_read_compliance("data"); | |
400 | } | |
f140e300 VSO |
401 | |
402 | qemu_iovec_init(&sub_qiov, qiov->niov); | |
403 | qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size); | |
404 | ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp); | |
405 | qemu_iovec_destroy(&sub_qiov); | |
406 | ||
407 | return ret < 0 ? -EIO : 0; | |
408 | } | |
409 | ||
410 | #define NBD_MAX_MALLOC_PAYLOAD 1000 | |
411 | /* nbd_co_receive_structured_payload | |
412 | */ | |
413 | static coroutine_fn int nbd_co_receive_structured_payload( | |
414 | NBDClientSession *s, void **payload, Error **errp) | |
415 | { | |
416 | int ret; | |
417 | uint32_t len; | |
418 | ||
419 | assert(nbd_reply_is_structured(&s->reply)); | |
420 | ||
421 | len = s->reply.structured.length; | |
422 | ||
423 | if (len == 0) { | |
424 | return 0; | |
425 | } | |
426 | ||
427 | if (payload == NULL) { | |
428 | error_setg(errp, "Unexpected structured payload"); | |
429 | return -EINVAL; | |
430 | } | |
431 | ||
432 | if (len > NBD_MAX_MALLOC_PAYLOAD) { | |
433 | error_setg(errp, "Payload too large"); | |
434 | return -EINVAL; | |
435 | } | |
436 | ||
437 | *payload = g_new(char, len); | |
e6798f06 | 438 | ret = nbd_read(s->ioc, *payload, len, "structured payload", errp); |
f140e300 VSO |
439 | if (ret < 0) { |
440 | g_free(*payload); | |
441 | *payload = NULL; | |
442 | return ret; | |
443 | } | |
444 | ||
445 | return 0; | |
446 | } | |
447 | ||
448 | /* nbd_co_do_receive_one_chunk | |
449 | * for simple reply: | |
450 | * set request_ret to received reply error | |
451 | * if qiov is not NULL: read payload to @qiov | |
452 | * for structured reply chunk: | |
453 | * if error chunk: read payload, set @request_ret, do not set @payload | |
454 | * else if offset_data chunk: read payload data to @qiov, do not set @payload | |
455 | * else: read payload to @payload | |
456 | * | |
457 | * If function fails, @errp contains corresponding error message, and the | |
458 | * connection with the server is suspect. If it returns 0, then the | |
459 | * transaction succeeded (although @request_ret may be a negative errno | |
460 | * corresponding to the server's error reply), and errp is unchanged. | |
461 | */ | |
462 | static coroutine_fn int nbd_co_do_receive_one_chunk( | |
463 | NBDClientSession *s, uint64_t handle, bool only_structured, | |
464 | int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp) | |
2302c1ca | 465 | { |
319a56cd | 466 | int ret; |
ed397b2f | 467 | int i = HANDLE_TO_INDEX(s, handle); |
f140e300 VSO |
468 | void *local_payload = NULL; |
469 | NBDStructuredReplyChunk *chunk; | |
470 | ||
471 | if (payload) { | |
472 | *payload = NULL; | |
473 | } | |
474 | *request_ret = 0; | |
2302c1ca | 475 | |
bc5a0335 | 476 | /* Wait until we're woken up by nbd_connection_entry. */ |
40f4a218 | 477 | s->requests[i].receiving = true; |
2302c1ca | 478 | qemu_coroutine_yield(); |
40f4a218 | 479 | s->requests[i].receiving = false; |
88ed4e1b | 480 | if (s->quit) { |
f140e300 VSO |
481 | error_setg(errp, "Connection closed"); |
482 | return -EIO; | |
483 | } | |
88ed4e1b | 484 | assert(s->ioc); |
f140e300 VSO |
485 | |
486 | assert(s->reply.handle == handle); | |
487 | ||
488 | if (nbd_reply_is_simple(&s->reply)) { | |
489 | if (only_structured) { | |
490 | error_setg(errp, "Protocol error: simple reply when structured " | |
491 | "reply chunk was expected"); | |
492 | return -EINVAL; | |
2302c1ca MAL |
493 | } |
494 | ||
f140e300 VSO |
495 | *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error); |
496 | if (*request_ret < 0 || !qiov) { | |
497 | return 0; | |
498 | } | |
499 | ||
500 | return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov, | |
501 | errp) < 0 ? -EIO : 0; | |
502 | } | |
503 | ||
504 | /* handle structured reply chunk */ | |
505 | assert(s->info.structured_reply); | |
506 | chunk = &s->reply.structured; | |
507 | ||
508 | if (chunk->type == NBD_REPLY_TYPE_NONE) { | |
509 | if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) { | |
510 | error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without" | |
e659fb3b | 511 | " NBD_REPLY_FLAG_DONE flag set"); |
f140e300 VSO |
512 | return -EINVAL; |
513 | } | |
b4176cb3 EB |
514 | if (chunk->length) { |
515 | error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with" | |
516 | " nonzero length"); | |
517 | return -EINVAL; | |
518 | } | |
f140e300 VSO |
519 | return 0; |
520 | } | |
521 | ||
522 | if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) { | |
523 | if (!qiov) { | |
524 | error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk"); | |
525 | return -EINVAL; | |
526 | } | |
527 | ||
528 | return nbd_co_receive_offset_data_payload(s, s->requests[i].offset, | |
529 | qiov, errp); | |
530 | } | |
531 | ||
532 | if (nbd_reply_type_is_error(chunk->type)) { | |
533 | payload = &local_payload; | |
534 | } | |
535 | ||
536 | ret = nbd_co_receive_structured_payload(s, payload, errp); | |
537 | if (ret < 0) { | |
538 | return ret; | |
2302c1ca | 539 | } |
ff82911c | 540 | |
f140e300 VSO |
541 | if (nbd_reply_type_is_error(chunk->type)) { |
542 | ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp); | |
543 | g_free(local_payload); | |
544 | return ret; | |
545 | } | |
546 | ||
547 | return 0; | |
548 | } | |
549 | ||
550 | /* nbd_co_receive_one_chunk | |
bc5a0335 | 551 | * Read reply, wake up connection_co and set s->quit if needed. |
f140e300 VSO |
552 | * Return value is a fatal error code or normal nbd reply error code |
553 | */ | |
554 | static coroutine_fn int nbd_co_receive_one_chunk( | |
555 | NBDClientSession *s, uint64_t handle, bool only_structured, | |
7f86068d VSO |
556 | int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload, |
557 | Error **errp) | |
f140e300 | 558 | { |
f140e300 | 559 | int ret = nbd_co_do_receive_one_chunk(s, handle, only_structured, |
7f86068d | 560 | request_ret, qiov, payload, errp); |
f140e300 VSO |
561 | |
562 | if (ret < 0) { | |
563 | s->quit = true; | |
564 | } else { | |
bc5a0335 | 565 | /* For assert at loop start in nbd_connection_entry */ |
f140e300 VSO |
566 | if (reply) { |
567 | *reply = s->reply; | |
568 | } | |
569 | s->reply.handle = 0; | |
f140e300 | 570 | } |
ff82911c | 571 | |
bc5a0335 VSO |
572 | if (s->connection_co) { |
573 | aio_co_wake(s->connection_co); | |
2302c1ca | 574 | } |
6bdcc018 | 575 | |
f140e300 VSO |
576 | return ret; |
577 | } | |
578 | ||
579 | typedef struct NBDReplyChunkIter { | |
580 | int ret; | |
7f86068d | 581 | int request_ret; |
f140e300 VSO |
582 | Error *err; |
583 | bool done, only_structured; | |
584 | } NBDReplyChunkIter; | |
585 | ||
7f86068d VSO |
586 | static void nbd_iter_channel_error(NBDReplyChunkIter *iter, |
587 | int ret, Error **local_err) | |
f140e300 VSO |
588 | { |
589 | assert(ret < 0); | |
590 | ||
7f86068d | 591 | if (!iter->ret) { |
f140e300 VSO |
592 | iter->ret = ret; |
593 | error_propagate(&iter->err, *local_err); | |
594 | } else { | |
595 | error_free(*local_err); | |
596 | } | |
597 | ||
598 | *local_err = NULL; | |
599 | } | |
600 | ||
7f86068d VSO |
601 | static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret) |
602 | { | |
603 | assert(ret < 0); | |
604 | ||
605 | if (!iter->request_ret) { | |
606 | iter->request_ret = ret; | |
607 | } | |
608 | } | |
609 | ||
f140e300 VSO |
610 | /* NBD_FOREACH_REPLY_CHUNK |
611 | */ | |
612 | #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \ | |
613 | qiov, reply, payload) \ | |
614 | for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \ | |
615 | nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);) | |
616 | ||
617 | /* nbd_reply_chunk_iter_receive | |
618 | */ | |
619 | static bool nbd_reply_chunk_iter_receive(NBDClientSession *s, | |
620 | NBDReplyChunkIter *iter, | |
621 | uint64_t handle, | |
622 | QEMUIOVector *qiov, NBDReply *reply, | |
623 | void **payload) | |
624 | { | |
7f86068d | 625 | int ret, request_ret; |
f140e300 VSO |
626 | NBDReply local_reply; |
627 | NBDStructuredReplyChunk *chunk; | |
628 | Error *local_err = NULL; | |
629 | if (s->quit) { | |
630 | error_setg(&local_err, "Connection closed"); | |
7f86068d | 631 | nbd_iter_channel_error(iter, -EIO, &local_err); |
f140e300 VSO |
632 | goto break_loop; |
633 | } | |
634 | ||
635 | if (iter->done) { | |
636 | /* Previous iteration was last. */ | |
637 | goto break_loop; | |
638 | } | |
639 | ||
640 | if (reply == NULL) { | |
641 | reply = &local_reply; | |
642 | } | |
643 | ||
644 | ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured, | |
7f86068d VSO |
645 | &request_ret, qiov, reply, payload, |
646 | &local_err); | |
f140e300 | 647 | if (ret < 0) { |
7f86068d VSO |
648 | nbd_iter_channel_error(iter, ret, &local_err); |
649 | } else if (request_ret < 0) { | |
650 | nbd_iter_request_error(iter, request_ret); | |
f140e300 VSO |
651 | } |
652 | ||
653 | /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */ | |
65e01d47 | 654 | if (nbd_reply_is_simple(reply) || s->quit) { |
f140e300 VSO |
655 | goto break_loop; |
656 | } | |
657 | ||
658 | chunk = &reply->structured; | |
659 | iter->only_structured = true; | |
660 | ||
661 | if (chunk->type == NBD_REPLY_TYPE_NONE) { | |
662 | /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */ | |
663 | assert(chunk->flags & NBD_REPLY_FLAG_DONE); | |
664 | goto break_loop; | |
665 | } | |
666 | ||
667 | if (chunk->flags & NBD_REPLY_FLAG_DONE) { | |
668 | /* This iteration is last. */ | |
669 | iter->done = true; | |
670 | } | |
671 | ||
672 | /* Execute the loop body */ | |
673 | return true; | |
674 | ||
675 | break_loop: | |
676 | s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL; | |
677 | ||
6bdcc018 PB |
678 | qemu_co_mutex_lock(&s->send_mutex); |
679 | s->in_flight--; | |
680 | qemu_co_queue_next(&s->free_sema); | |
681 | qemu_co_mutex_unlock(&s->send_mutex); | |
319a56cd | 682 | |
f140e300 | 683 | return false; |
2302c1ca MAL |
684 | } |
685 | ||
f140e300 | 686 | static int nbd_co_receive_return_code(NBDClientSession *s, uint64_t handle, |
7f86068d | 687 | int *request_ret, Error **errp) |
f140e300 VSO |
688 | { |
689 | NBDReplyChunkIter iter; | |
690 | ||
691 | NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) { | |
692 | /* nbd_reply_chunk_iter_receive does all the work */ | |
693 | } | |
694 | ||
695 | error_propagate(errp, iter.err); | |
7f86068d | 696 | *request_ret = iter.request_ret; |
f140e300 VSO |
697 | return iter.ret; |
698 | } | |
699 | ||
700 | static int nbd_co_receive_cmdread_reply(NBDClientSession *s, uint64_t handle, | |
701 | uint64_t offset, QEMUIOVector *qiov, | |
7f86068d | 702 | int *request_ret, Error **errp) |
f140e300 VSO |
703 | { |
704 | NBDReplyChunkIter iter; | |
705 | NBDReply reply; | |
706 | void *payload = NULL; | |
707 | Error *local_err = NULL; | |
708 | ||
709 | NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply, | |
710 | qiov, &reply, &payload) | |
711 | { | |
712 | int ret; | |
713 | NBDStructuredReplyChunk *chunk = &reply.structured; | |
714 | ||
715 | assert(nbd_reply_is_structured(&reply)); | |
716 | ||
717 | switch (chunk->type) { | |
718 | case NBD_REPLY_TYPE_OFFSET_DATA: | |
719 | /* special cased in nbd_co_receive_one_chunk, data is already | |
720 | * in qiov */ | |
721 | break; | |
722 | case NBD_REPLY_TYPE_OFFSET_HOLE: | |
75d34eb9 | 723 | ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload, |
f140e300 VSO |
724 | offset, qiov, &local_err); |
725 | if (ret < 0) { | |
726 | s->quit = true; | |
7f86068d | 727 | nbd_iter_channel_error(&iter, ret, &local_err); |
f140e300 VSO |
728 | } |
729 | break; | |
730 | default: | |
731 | if (!nbd_reply_type_is_error(chunk->type)) { | |
732 | /* not allowed reply type */ | |
733 | s->quit = true; | |
734 | error_setg(&local_err, | |
735 | "Unexpected reply type: %d (%s) for CMD_READ", | |
736 | chunk->type, nbd_reply_type_lookup(chunk->type)); | |
7f86068d | 737 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); |
f140e300 VSO |
738 | } |
739 | } | |
740 | ||
741 | g_free(payload); | |
742 | payload = NULL; | |
743 | } | |
744 | ||
745 | error_propagate(errp, iter.err); | |
7f86068d | 746 | *request_ret = iter.request_ret; |
f140e300 VSO |
747 | return iter.ret; |
748 | } | |
749 | ||
78a33ab5 VSO |
750 | static int nbd_co_receive_blockstatus_reply(NBDClientSession *s, |
751 | uint64_t handle, uint64_t length, | |
7f86068d VSO |
752 | NBDExtent *extent, |
753 | int *request_ret, Error **errp) | |
78a33ab5 VSO |
754 | { |
755 | NBDReplyChunkIter iter; | |
756 | NBDReply reply; | |
757 | void *payload = NULL; | |
758 | Error *local_err = NULL; | |
759 | bool received = false; | |
760 | ||
761 | assert(!extent->length); | |
ebd82cd8 | 762 | NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, &reply, &payload) { |
78a33ab5 VSO |
763 | int ret; |
764 | NBDStructuredReplyChunk *chunk = &reply.structured; | |
765 | ||
766 | assert(nbd_reply_is_structured(&reply)); | |
767 | ||
768 | switch (chunk->type) { | |
769 | case NBD_REPLY_TYPE_BLOCK_STATUS: | |
770 | if (received) { | |
771 | s->quit = true; | |
772 | error_setg(&local_err, "Several BLOCK_STATUS chunks in reply"); | |
7f86068d | 773 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); |
78a33ab5 VSO |
774 | } |
775 | received = true; | |
776 | ||
777 | ret = nbd_parse_blockstatus_payload(s, &reply.structured, | |
778 | payload, length, extent, | |
779 | &local_err); | |
780 | if (ret < 0) { | |
781 | s->quit = true; | |
7f86068d | 782 | nbd_iter_channel_error(&iter, ret, &local_err); |
78a33ab5 VSO |
783 | } |
784 | break; | |
785 | default: | |
786 | if (!nbd_reply_type_is_error(chunk->type)) { | |
787 | s->quit = true; | |
788 | error_setg(&local_err, | |
789 | "Unexpected reply type: %d (%s) " | |
790 | "for CMD_BLOCK_STATUS", | |
791 | chunk->type, nbd_reply_type_lookup(chunk->type)); | |
7f86068d | 792 | nbd_iter_channel_error(&iter, -EINVAL, &local_err); |
78a33ab5 VSO |
793 | } |
794 | } | |
795 | ||
796 | g_free(payload); | |
797 | payload = NULL; | |
798 | } | |
799 | ||
b29f3a3d EB |
800 | if (!extent->length && !iter.request_ret) { |
801 | error_setg(&local_err, "Server did not reply with any status extents"); | |
802 | nbd_iter_channel_error(&iter, -EIO, &local_err); | |
78a33ab5 | 803 | } |
7f86068d | 804 | |
78a33ab5 | 805 | error_propagate(errp, iter.err); |
7f86068d | 806 | *request_ret = iter.request_ret; |
78a33ab5 VSO |
807 | return iter.ret; |
808 | } | |
809 | ||
f140e300 VSO |
810 | static int nbd_co_request(BlockDriverState *bs, NBDRequest *request, |
811 | QEMUIOVector *write_qiov) | |
f35dff7e | 812 | { |
7f86068d | 813 | int ret, request_ret; |
f140e300 VSO |
814 | Error *local_err = NULL; |
815 | NBDClientSession *client = nbd_get_client_session(bs); | |
f35dff7e | 816 | |
f140e300 VSO |
817 | assert(request->type != NBD_CMD_READ); |
818 | if (write_qiov) { | |
819 | assert(request->type == NBD_CMD_WRITE); | |
820 | assert(request->len == iov_size(write_qiov->iov, write_qiov->niov)); | |
4bfe4478 | 821 | } else { |
f140e300 | 822 | assert(request->type != NBD_CMD_WRITE); |
4bfe4478 | 823 | } |
f140e300 | 824 | ret = nbd_co_send_request(bs, request, write_qiov); |
f35dff7e | 825 | if (ret < 0) { |
319a56cd | 826 | return ret; |
f35dff7e | 827 | } |
319a56cd | 828 | |
7f86068d VSO |
829 | ret = nbd_co_receive_return_code(client, request->handle, |
830 | &request_ret, &local_err); | |
f140e300 | 831 | if (local_err) { |
d8b4bad8 VSO |
832 | trace_nbd_co_request_fail(request->from, request->len, request->handle, |
833 | request->flags, request->type, | |
834 | nbd_cmd_lookup(request->type), | |
835 | ret, error_get_pretty(local_err)); | |
836 | error_free(local_err); | |
f140e300 | 837 | } |
7f86068d | 838 | return ret ? ret : request_ret; |
f35dff7e VSO |
839 | } |
840 | ||
70c4fb26 EB |
841 | int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset, |
842 | uint64_t bytes, QEMUIOVector *qiov, int flags) | |
2302c1ca | 843 | { |
7f86068d | 844 | int ret, request_ret; |
f140e300 VSO |
845 | Error *local_err = NULL; |
846 | NBDClientSession *client = nbd_get_client_session(bs); | |
ed2dd912 | 847 | NBDRequest request = { |
70c4fb26 EB |
848 | .type = NBD_CMD_READ, |
849 | .from = offset, | |
850 | .len = bytes, | |
851 | }; | |
2302c1ca | 852 | |
70c4fb26 EB |
853 | assert(bytes <= NBD_MAX_BUFFER_SIZE); |
854 | assert(!flags); | |
2302c1ca | 855 | |
9d8f818c EB |
856 | if (!bytes) { |
857 | return 0; | |
858 | } | |
9cf63850 EB |
859 | /* |
860 | * Work around the fact that the block layer doesn't do | |
861 | * byte-accurate sizing yet - if the read exceeds the server's | |
862 | * advertised size because the block layer rounded size up, then | |
863 | * truncate the request to the server and tail-pad with zero. | |
864 | */ | |
865 | if (offset >= client->info.size) { | |
866 | assert(bytes < BDRV_SECTOR_SIZE); | |
867 | qemu_iovec_memset(qiov, 0, 0, bytes); | |
868 | return 0; | |
869 | } | |
870 | if (offset + bytes > client->info.size) { | |
871 | uint64_t slop = offset + bytes - client->info.size; | |
872 | ||
873 | assert(slop < BDRV_SECTOR_SIZE); | |
874 | qemu_iovec_memset(qiov, bytes - slop, 0, slop); | |
875 | request.len -= slop; | |
876 | } | |
877 | ||
f140e300 VSO |
878 | ret = nbd_co_send_request(bs, &request, NULL); |
879 | if (ret < 0) { | |
880 | return ret; | |
881 | } | |
882 | ||
883 | ret = nbd_co_receive_cmdread_reply(client, request.handle, offset, qiov, | |
7f86068d | 884 | &request_ret, &local_err); |
08ace1d7 | 885 | if (local_err) { |
d8b4bad8 VSO |
886 | trace_nbd_co_request_fail(request.from, request.len, request.handle, |
887 | request.flags, request.type, | |
888 | nbd_cmd_lookup(request.type), | |
889 | ret, error_get_pretty(local_err)); | |
890 | error_free(local_err); | |
f140e300 | 891 | } |
7f86068d | 892 | return ret ? ret : request_ret; |
2302c1ca MAL |
893 | } |
894 | ||
70c4fb26 EB |
895 | int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset, |
896 | uint64_t bytes, QEMUIOVector *qiov, int flags) | |
2302c1ca | 897 | { |
10676b81 | 898 | NBDClientSession *client = nbd_get_client_session(bs); |
ed2dd912 | 899 | NBDRequest request = { |
70c4fb26 EB |
900 | .type = NBD_CMD_WRITE, |
901 | .from = offset, | |
902 | .len = bytes, | |
903 | }; | |
2302c1ca | 904 | |
1104d83c | 905 | assert(!(client->info.flags & NBD_FLAG_READ_ONLY)); |
52a46505 | 906 | if (flags & BDRV_REQ_FUA) { |
004a89fc | 907 | assert(client->info.flags & NBD_FLAG_SEND_FUA); |
b626b51a | 908 | request.flags |= NBD_CMD_FLAG_FUA; |
2302c1ca MAL |
909 | } |
910 | ||
70c4fb26 | 911 | assert(bytes <= NBD_MAX_BUFFER_SIZE); |
2302c1ca | 912 | |
9d8f818c EB |
913 | if (!bytes) { |
914 | return 0; | |
915 | } | |
f35dff7e | 916 | return nbd_co_request(bs, &request, qiov); |
2302c1ca MAL |
917 | } |
918 | ||
fa778fff | 919 | int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, |
f5a5ca79 | 920 | int bytes, BdrvRequestFlags flags) |
fa778fff | 921 | { |
fa778fff EB |
922 | NBDClientSession *client = nbd_get_client_session(bs); |
923 | NBDRequest request = { | |
924 | .type = NBD_CMD_WRITE_ZEROES, | |
925 | .from = offset, | |
f5a5ca79 | 926 | .len = bytes, |
fa778fff | 927 | }; |
fa778fff | 928 | |
1104d83c | 929 | assert(!(client->info.flags & NBD_FLAG_READ_ONLY)); |
004a89fc | 930 | if (!(client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) { |
fa778fff EB |
931 | return -ENOTSUP; |
932 | } | |
933 | ||
934 | if (flags & BDRV_REQ_FUA) { | |
004a89fc | 935 | assert(client->info.flags & NBD_FLAG_SEND_FUA); |
fa778fff EB |
936 | request.flags |= NBD_CMD_FLAG_FUA; |
937 | } | |
938 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
939 | request.flags |= NBD_CMD_FLAG_NO_HOLE; | |
940 | } | |
941 | ||
9d8f818c EB |
942 | if (!bytes) { |
943 | return 0; | |
944 | } | |
f35dff7e | 945 | return nbd_co_request(bs, &request, NULL); |
fa778fff EB |
946 | } |
947 | ||
f53a829b | 948 | int nbd_client_co_flush(BlockDriverState *bs) |
2302c1ca | 949 | { |
10676b81 | 950 | NBDClientSession *client = nbd_get_client_session(bs); |
ed2dd912 | 951 | NBDRequest request = { .type = NBD_CMD_FLUSH }; |
2302c1ca | 952 | |
004a89fc | 953 | if (!(client->info.flags & NBD_FLAG_SEND_FLUSH)) { |
2302c1ca MAL |
954 | return 0; |
955 | } | |
956 | ||
2302c1ca MAL |
957 | request.from = 0; |
958 | request.len = 0; | |
959 | ||
f35dff7e | 960 | return nbd_co_request(bs, &request, NULL); |
2302c1ca MAL |
961 | } |
962 | ||
f5a5ca79 | 963 | int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) |
2302c1ca | 964 | { |
10676b81 | 965 | NBDClientSession *client = nbd_get_client_session(bs); |
ed2dd912 | 966 | NBDRequest request = { |
447e57c3 EB |
967 | .type = NBD_CMD_TRIM, |
968 | .from = offset, | |
f5a5ca79 | 969 | .len = bytes, |
447e57c3 | 970 | }; |
2302c1ca | 971 | |
1104d83c | 972 | assert(!(client->info.flags & NBD_FLAG_READ_ONLY)); |
9d8f818c | 973 | if (!(client->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) { |
2302c1ca MAL |
974 | return 0; |
975 | } | |
2302c1ca | 976 | |
f35dff7e | 977 | return nbd_co_request(bs, &request, NULL); |
2302c1ca MAL |
978 | } |
979 | ||
78a33ab5 VSO |
980 | int coroutine_fn nbd_client_co_block_status(BlockDriverState *bs, |
981 | bool want_zero, | |
982 | int64_t offset, int64_t bytes, | |
983 | int64_t *pnum, int64_t *map, | |
984 | BlockDriverState **file) | |
985 | { | |
7f86068d | 986 | int ret, request_ret; |
78a33ab5 VSO |
987 | NBDExtent extent = { 0 }; |
988 | NBDClientSession *client = nbd_get_client_session(bs); | |
989 | Error *local_err = NULL; | |
990 | ||
991 | NBDRequest request = { | |
992 | .type = NBD_CMD_BLOCK_STATUS, | |
993 | .from = offset, | |
994 | .len = MIN(MIN_NON_ZERO(QEMU_ALIGN_DOWN(INT_MAX, | |
995 | bs->bl.request_alignment), | |
9cf63850 EB |
996 | client->info.max_block), |
997 | MIN(bytes, client->info.size - offset)), | |
78a33ab5 VSO |
998 | .flags = NBD_CMD_FLAG_REQ_ONE, |
999 | }; | |
1000 | ||
1001 | if (!client->info.base_allocation) { | |
1002 | *pnum = bytes; | |
a62a85ef EB |
1003 | *map = offset; |
1004 | *file = bs; | |
1005 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
78a33ab5 VSO |
1006 | } |
1007 | ||
9cf63850 EB |
1008 | /* |
1009 | * Work around the fact that the block layer doesn't do | |
1010 | * byte-accurate sizing yet - if the status request exceeds the | |
1011 | * server's advertised size because the block layer rounded size | |
1012 | * up, we truncated the request to the server (above), or are | |
1013 | * called on just the hole. | |
1014 | */ | |
1015 | if (offset >= client->info.size) { | |
1016 | *pnum = bytes; | |
1017 | assert(bytes < BDRV_SECTOR_SIZE); | |
1018 | /* Intentionally don't report offset_valid for the hole */ | |
1019 | return BDRV_BLOCK_ZERO; | |
1020 | } | |
1021 | ||
1022 | if (client->info.min_block) { | |
1023 | assert(QEMU_IS_ALIGNED(request.len, client->info.min_block)); | |
1024 | } | |
78a33ab5 VSO |
1025 | ret = nbd_co_send_request(bs, &request, NULL); |
1026 | if (ret < 0) { | |
1027 | return ret; | |
1028 | } | |
1029 | ||
1030 | ret = nbd_co_receive_blockstatus_reply(client, request.handle, bytes, | |
7f86068d | 1031 | &extent, &request_ret, &local_err); |
78a33ab5 | 1032 | if (local_err) { |
d8b4bad8 VSO |
1033 | trace_nbd_co_request_fail(request.from, request.len, request.handle, |
1034 | request.flags, request.type, | |
1035 | nbd_cmd_lookup(request.type), | |
1036 | ret, error_get_pretty(local_err)); | |
1037 | error_free(local_err); | |
78a33ab5 | 1038 | } |
7f86068d VSO |
1039 | if (ret < 0 || request_ret < 0) { |
1040 | return ret ? ret : request_ret; | |
78a33ab5 VSO |
1041 | } |
1042 | ||
1043 | assert(extent.length); | |
1044 | *pnum = extent.length; | |
a62a85ef EB |
1045 | *map = offset; |
1046 | *file = bs; | |
78a33ab5 | 1047 | return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) | |
a62a85ef EB |
1048 | (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) | |
1049 | BDRV_BLOCK_OFFSET_VALID; | |
78a33ab5 VSO |
1050 | } |
1051 | ||
f53a829b | 1052 | void nbd_client_detach_aio_context(BlockDriverState *bs) |
69447cd8 | 1053 | { |
ff82911c | 1054 | NBDClientSession *client = nbd_get_client_session(bs); |
96d06835 | 1055 | qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc)); |
69447cd8 SH |
1056 | } |
1057 | ||
28e0b2d2 KW |
1058 | static void nbd_client_attach_aio_context_bh(void *opaque) |
1059 | { | |
1060 | BlockDriverState *bs = opaque; | |
1061 | NBDClientSession *client = nbd_get_client_session(bs); | |
1062 | ||
1063 | /* The node is still drained, so we know the coroutine has yielded in | |
1064 | * nbd_read_eof(), the only place where bs->in_flight can reach 0, or it is | |
1065 | * entered for the first time. Both places are safe for entering the | |
1066 | * coroutine.*/ | |
1067 | qemu_aio_coroutine_enter(bs->aio_context, client->connection_co); | |
1068 | bdrv_dec_in_flight(bs); | |
1069 | } | |
1070 | ||
f53a829b HR |
1071 | void nbd_client_attach_aio_context(BlockDriverState *bs, |
1072 | AioContext *new_context) | |
69447cd8 | 1073 | { |
ff82911c | 1074 | NBDClientSession *client = nbd_get_client_session(bs); |
96d06835 | 1075 | qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc), new_context); |
5ad81b49 | 1076 | |
28e0b2d2 KW |
1077 | bdrv_inc_in_flight(bs); |
1078 | ||
1079 | /* Need to wait here for the BH to run because the BH must run while the | |
1080 | * node is still drained. */ | |
1081 | aio_wait_bh_oneshot(new_context, nbd_client_attach_aio_context_bh, bs); | |
69447cd8 SH |
1082 | } |
1083 | ||
f53a829b | 1084 | void nbd_client_close(BlockDriverState *bs) |
2302c1ca | 1085 | { |
10676b81 | 1086 | NBDClientSession *client = nbd_get_client_session(bs); |
ed2dd912 | 1087 | NBDRequest request = { .type = NBD_CMD_DISC }; |
2302c1ca | 1088 | |
88ed4e1b | 1089 | assert(client->ioc); |
4a41a2d6 | 1090 | |
1c778ef7 | 1091 | nbd_send_request(client->ioc, &request); |
5ad283eb | 1092 | |
f53a829b | 1093 | nbd_teardown_connection(bs); |
2302c1ca MAL |
1094 | } |
1095 | ||
d42f78e9 VSO |
1096 | static QIOChannelSocket *nbd_establish_connection(SocketAddress *saddr, |
1097 | Error **errp) | |
1098 | { | |
1099 | QIOChannelSocket *sioc; | |
1100 | Error *local_err = NULL; | |
1101 | ||
1102 | sioc = qio_channel_socket_new(); | |
1103 | qio_channel_set_name(QIO_CHANNEL(sioc), "nbd-client"); | |
1104 | ||
1105 | qio_channel_socket_connect_sync(sioc, saddr, &local_err); | |
1106 | if (local_err) { | |
1107 | object_unref(OBJECT(sioc)); | |
1108 | error_propagate(errp, local_err); | |
1109 | return NULL; | |
1110 | } | |
1111 | ||
1112 | qio_channel_set_delay(QIO_CHANNEL(sioc), false); | |
1113 | ||
1114 | return sioc; | |
1115 | } | |
1116 | ||
b0e4b5a5 VSO |
1117 | static int nbd_client_connect(BlockDriverState *bs, |
1118 | SocketAddress *saddr, | |
1119 | const char *export, | |
1120 | QCryptoTLSCreds *tlscreds, | |
1121 | const char *hostname, | |
1122 | const char *x_dirty_bitmap, | |
1123 | Error **errp) | |
2302c1ca | 1124 | { |
10676b81 | 1125 | NBDClientSession *client = nbd_get_client_session(bs); |
2302c1ca MAL |
1126 | int ret; |
1127 | ||
d42f78e9 VSO |
1128 | /* |
1129 | * establish TCP connection, return error if it fails | |
1130 | * TODO: Configurable retry-until-timeout behaviour. | |
1131 | */ | |
1132 | QIOChannelSocket *sioc = nbd_establish_connection(saddr, errp); | |
1133 | ||
1134 | if (!sioc) { | |
1135 | return -ECONNREFUSED; | |
1136 | } | |
1137 | ||
2302c1ca | 1138 | /* NBD handshake */ |
e2bc625f | 1139 | logout("session init %s\n", export); |
064097d9 DB |
1140 | qio_channel_set_blocking(QIO_CHANNEL(sioc), true, NULL); |
1141 | ||
081dd1fe | 1142 | client->info.request_sizes = true; |
f140e300 | 1143 | client->info.structured_reply = true; |
78a33ab5 | 1144 | client->info.base_allocation = true; |
216ee365 | 1145 | client->info.x_dirty_bitmap = g_strdup(x_dirty_bitmap); |
6dc1667d EB |
1146 | client->info.name = g_strdup(export ?: ""); |
1147 | ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), tlscreds, hostname, | |
004a89fc | 1148 | &client->ioc, &client->info, errp); |
216ee365 | 1149 | g_free(client->info.x_dirty_bitmap); |
6dc1667d | 1150 | g_free(client->info.name); |
2302c1ca MAL |
1151 | if (ret < 0) { |
1152 | logout("Failed to negotiate with the NBD server\n"); | |
d42f78e9 | 1153 | object_unref(OBJECT(sioc)); |
2302c1ca MAL |
1154 | return ret; |
1155 | } | |
47829c40 EB |
1156 | if (x_dirty_bitmap && !client->info.base_allocation) { |
1157 | error_setg(errp, "requested x-dirty-bitmap %s not found", | |
1158 | x_dirty_bitmap); | |
c688e6ca EB |
1159 | ret = -EINVAL; |
1160 | goto fail; | |
47829c40 | 1161 | } |
6c2e581d KW |
1162 | if (client->info.flags & NBD_FLAG_READ_ONLY) { |
1163 | ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp); | |
1164 | if (ret < 0) { | |
c688e6ca | 1165 | goto fail; |
6c2e581d | 1166 | } |
1104d83c | 1167 | } |
004a89fc | 1168 | if (client->info.flags & NBD_FLAG_SEND_FUA) { |
4df863f3 | 1169 | bs->supported_write_flags = BDRV_REQ_FUA; |
169407e1 EB |
1170 | bs->supported_zero_flags |= BDRV_REQ_FUA; |
1171 | } | |
004a89fc | 1172 | if (client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) { |
169407e1 | 1173 | bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP; |
4df863f3 | 1174 | } |
2302c1ca | 1175 | |
064097d9 | 1176 | client->sioc = sioc; |
f95910fe DB |
1177 | |
1178 | if (!client->ioc) { | |
1179 | client->ioc = QIO_CHANNEL(sioc); | |
1180 | object_ref(OBJECT(client->ioc)); | |
1181 | } | |
2302c1ca MAL |
1182 | |
1183 | /* Now that we're connected, set the socket to be non-blocking and | |
1184 | * kick the reply mechanism. */ | |
064097d9 | 1185 | qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL); |
bc5a0335 | 1186 | client->connection_co = qemu_coroutine_create(nbd_connection_entry, client); |
5ad81b49 | 1187 | bdrv_inc_in_flight(bs); |
f53a829b | 1188 | nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs)); |
2302c1ca MAL |
1189 | |
1190 | logout("Established connection with NBD server\n"); | |
1191 | return 0; | |
c688e6ca EB |
1192 | |
1193 | fail: | |
1194 | /* | |
1195 | * We have connected, but must fail for other reasons. The | |
1196 | * connection is still blocking; send NBD_CMD_DISC as a courtesy | |
1197 | * to the server. | |
1198 | */ | |
1199 | { | |
1200 | NBDRequest request = { .type = NBD_CMD_DISC }; | |
1201 | ||
1202 | nbd_send_request(client->ioc ?: QIO_CHANNEL(sioc), &request); | |
d42f78e9 VSO |
1203 | |
1204 | object_unref(OBJECT(sioc)); | |
1205 | ||
c688e6ca EB |
1206 | return ret; |
1207 | } | |
2302c1ca | 1208 | } |
b0e4b5a5 VSO |
1209 | |
1210 | int nbd_client_init(BlockDriverState *bs, | |
1211 | SocketAddress *saddr, | |
1212 | const char *export, | |
1213 | QCryptoTLSCreds *tlscreds, | |
1214 | const char *hostname, | |
1215 | const char *x_dirty_bitmap, | |
1216 | Error **errp) | |
1217 | { | |
1218 | NBDClientSession *client = nbd_get_client_session(bs); | |
1219 | ||
5ad81b49 | 1220 | client->bs = bs; |
b0e4b5a5 VSO |
1221 | qemu_co_mutex_init(&client->send_mutex); |
1222 | qemu_co_queue_init(&client->free_sema); | |
1223 | ||
1224 | return nbd_client_connect(bs, saddr, export, tlscreds, hostname, | |
1225 | x_dirty_bitmap, errp); | |
1226 | } |