]>
Commit | Line | Data |
---|---|---|
61007b31 SH |
1 | /* |
2 | * Block layer I/O functions | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
24 | ||
80c71a24 | 25 | #include "qemu/osdep.h" |
61007b31 | 26 | #include "trace.h" |
7f0e9da6 | 27 | #include "sysemu/block-backend.h" |
61007b31 SH |
28 | #include "block/blockjob.h" |
29 | #include "block/block_int.h" | |
f348b6d1 | 30 | #include "qemu/cutils.h" |
da34e65c | 31 | #include "qapi/error.h" |
d49b6836 | 32 | #include "qemu/error-report.h" |
61007b31 SH |
33 | |
34 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
35 | ||
61007b31 SH |
36 | static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, |
37 | int64_t sector_num, | |
38 | QEMUIOVector *qiov, | |
39 | int nb_sectors, | |
40 | BdrvRequestFlags flags, | |
41 | BlockCompletionFunc *cb, | |
42 | void *opaque, | |
43 | bool is_write); | |
44 | static void coroutine_fn bdrv_co_do_rw(void *opaque); | |
45 | static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, | |
46 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags); | |
47 | ||
c2066af0 | 48 | static void bdrv_parent_drained_begin(BlockDriverState *bs) |
61007b31 | 49 | { |
c2066af0 | 50 | BdrvChild *c; |
27ccdd52 | 51 | |
c2066af0 KW |
52 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
53 | if (c->role->drained_begin) { | |
54 | c->role->drained_begin(c); | |
55 | } | |
ce0f1412 PB |
56 | } |
57 | } | |
61007b31 | 58 | |
c2066af0 | 59 | static void bdrv_parent_drained_end(BlockDriverState *bs) |
ce0f1412 | 60 | { |
c2066af0 | 61 | BdrvChild *c; |
27ccdd52 | 62 | |
c2066af0 KW |
63 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
64 | if (c->role->drained_end) { | |
65 | c->role->drained_end(c); | |
66 | } | |
27ccdd52 | 67 | } |
61007b31 SH |
68 | } |
69 | ||
61007b31 SH |
70 | void bdrv_refresh_limits(BlockDriverState *bs, Error **errp) |
71 | { | |
72 | BlockDriver *drv = bs->drv; | |
73 | Error *local_err = NULL; | |
74 | ||
75 | memset(&bs->bl, 0, sizeof(bs->bl)); | |
76 | ||
77 | if (!drv) { | |
78 | return; | |
79 | } | |
80 | ||
81 | /* Take some limits from the children as a default */ | |
82 | if (bs->file) { | |
9a4f4c31 | 83 | bdrv_refresh_limits(bs->file->bs, &local_err); |
61007b31 SH |
84 | if (local_err) { |
85 | error_propagate(errp, local_err); | |
86 | return; | |
87 | } | |
9a4f4c31 KW |
88 | bs->bl.opt_transfer_length = bs->file->bs->bl.opt_transfer_length; |
89 | bs->bl.max_transfer_length = bs->file->bs->bl.max_transfer_length; | |
90 | bs->bl.min_mem_alignment = bs->file->bs->bl.min_mem_alignment; | |
91 | bs->bl.opt_mem_alignment = bs->file->bs->bl.opt_mem_alignment; | |
bd44feb7 | 92 | bs->bl.max_iov = bs->file->bs->bl.max_iov; |
61007b31 | 93 | } else { |
4196d2f0 | 94 | bs->bl.min_mem_alignment = 512; |
459b4e66 | 95 | bs->bl.opt_mem_alignment = getpagesize(); |
bd44feb7 SH |
96 | |
97 | /* Safe default since most protocols use readv()/writev()/etc */ | |
98 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
99 | } |
100 | ||
760e0063 KW |
101 | if (bs->backing) { |
102 | bdrv_refresh_limits(bs->backing->bs, &local_err); | |
61007b31 SH |
103 | if (local_err) { |
104 | error_propagate(errp, local_err); | |
105 | return; | |
106 | } | |
107 | bs->bl.opt_transfer_length = | |
108 | MAX(bs->bl.opt_transfer_length, | |
760e0063 | 109 | bs->backing->bs->bl.opt_transfer_length); |
61007b31 SH |
110 | bs->bl.max_transfer_length = |
111 | MIN_NON_ZERO(bs->bl.max_transfer_length, | |
760e0063 | 112 | bs->backing->bs->bl.max_transfer_length); |
61007b31 SH |
113 | bs->bl.opt_mem_alignment = |
114 | MAX(bs->bl.opt_mem_alignment, | |
760e0063 | 115 | bs->backing->bs->bl.opt_mem_alignment); |
4196d2f0 DL |
116 | bs->bl.min_mem_alignment = |
117 | MAX(bs->bl.min_mem_alignment, | |
760e0063 | 118 | bs->backing->bs->bl.min_mem_alignment); |
bd44feb7 SH |
119 | bs->bl.max_iov = |
120 | MIN(bs->bl.max_iov, | |
121 | bs->backing->bs->bl.max_iov); | |
61007b31 SH |
122 | } |
123 | ||
124 | /* Then let the driver override it */ | |
125 | if (drv->bdrv_refresh_limits) { | |
126 | drv->bdrv_refresh_limits(bs, errp); | |
127 | } | |
128 | } | |
129 | ||
130 | /** | |
131 | * The copy-on-read flag is actually a reference count so multiple users may | |
132 | * use the feature without worrying about clobbering its previous state. | |
133 | * Copy-on-read stays enabled until all users have called to disable it. | |
134 | */ | |
135 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
136 | { | |
137 | bs->copy_on_read++; | |
138 | } | |
139 | ||
140 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
141 | { | |
142 | assert(bs->copy_on_read > 0); | |
143 | bs->copy_on_read--; | |
144 | } | |
145 | ||
146 | /* Check if any requests are in-flight (including throttled requests) */ | |
439db28c | 147 | bool bdrv_requests_pending(BlockDriverState *bs) |
61007b31 | 148 | { |
37a639a7 KW |
149 | BdrvChild *child; |
150 | ||
61007b31 SH |
151 | if (!QLIST_EMPTY(&bs->tracked_requests)) { |
152 | return true; | |
153 | } | |
37a639a7 KW |
154 | |
155 | QLIST_FOREACH(child, &bs->children, next) { | |
156 | if (bdrv_requests_pending(child->bs)) { | |
157 | return true; | |
158 | } | |
61007b31 | 159 | } |
37a639a7 | 160 | |
61007b31 SH |
161 | return false; |
162 | } | |
163 | ||
67da1dc5 FZ |
164 | static void bdrv_drain_recurse(BlockDriverState *bs) |
165 | { | |
166 | BdrvChild *child; | |
167 | ||
168 | if (bs->drv && bs->drv->bdrv_drain) { | |
169 | bs->drv->bdrv_drain(bs); | |
170 | } | |
171 | QLIST_FOREACH(child, &bs->children, next) { | |
172 | bdrv_drain_recurse(child->bs); | |
173 | } | |
174 | } | |
175 | ||
a77fd4bb FZ |
176 | typedef struct { |
177 | Coroutine *co; | |
178 | BlockDriverState *bs; | |
179 | QEMUBH *bh; | |
180 | bool done; | |
181 | } BdrvCoDrainData; | |
182 | ||
b6e84c97 PB |
183 | static void bdrv_drain_poll(BlockDriverState *bs) |
184 | { | |
185 | bool busy = true; | |
186 | ||
187 | while (busy) { | |
188 | /* Keep iterating */ | |
b6e84c97 PB |
189 | busy = bdrv_requests_pending(bs); |
190 | busy |= aio_poll(bdrv_get_aio_context(bs), busy); | |
191 | } | |
192 | } | |
193 | ||
a77fd4bb FZ |
194 | static void bdrv_co_drain_bh_cb(void *opaque) |
195 | { | |
196 | BdrvCoDrainData *data = opaque; | |
197 | Coroutine *co = data->co; | |
198 | ||
199 | qemu_bh_delete(data->bh); | |
b6e84c97 | 200 | bdrv_drain_poll(data->bs); |
a77fd4bb FZ |
201 | data->done = true; |
202 | qemu_coroutine_enter(co, NULL); | |
203 | } | |
204 | ||
b6e84c97 | 205 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs) |
a77fd4bb FZ |
206 | { |
207 | BdrvCoDrainData data; | |
208 | ||
209 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
210 | * other coroutines run if they were queued from | |
211 | * qemu_co_queue_run_restart(). */ | |
212 | ||
213 | assert(qemu_in_coroutine()); | |
214 | data = (BdrvCoDrainData) { | |
215 | .co = qemu_coroutine_self(), | |
216 | .bs = bs, | |
217 | .done = false, | |
218 | .bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_drain_bh_cb, &data), | |
219 | }; | |
220 | qemu_bh_schedule(data.bh); | |
221 | ||
222 | qemu_coroutine_yield(); | |
223 | /* If we are resumed from some other event (such as an aio completion or a | |
224 | * timer callback), it is a bug in the caller that should be fixed. */ | |
225 | assert(data.done); | |
226 | } | |
227 | ||
61007b31 | 228 | /* |
67da1dc5 FZ |
229 | * Wait for pending requests to complete on a single BlockDriverState subtree, |
230 | * and suspend block driver's internal I/O until next request arrives. | |
61007b31 | 231 | * |
61007b31 SH |
232 | * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState |
233 | * AioContext. | |
7a63f3cd SH |
234 | * |
235 | * Only this BlockDriverState's AioContext is run, so in-flight requests must | |
236 | * not depend on events in other AioContexts. In that case, use | |
237 | * bdrv_drain_all() instead. | |
61007b31 | 238 | */ |
b6e84c97 | 239 | void coroutine_fn bdrv_co_drain(BlockDriverState *bs) |
61007b31 | 240 | { |
c2066af0 | 241 | bdrv_parent_drained_begin(bs); |
6b98bd64 | 242 | bdrv_io_unplugged_begin(bs); |
b6e84c97 PB |
243 | bdrv_drain_recurse(bs); |
244 | bdrv_co_yield_to_drain(bs); | |
6b98bd64 | 245 | bdrv_io_unplugged_end(bs); |
c2066af0 | 246 | bdrv_parent_drained_end(bs); |
b6e84c97 | 247 | } |
f406c03c | 248 | |
b6e84c97 PB |
249 | void bdrv_drain(BlockDriverState *bs) |
250 | { | |
c2066af0 | 251 | bdrv_parent_drained_begin(bs); |
6b98bd64 | 252 | bdrv_io_unplugged_begin(bs); |
67da1dc5 | 253 | bdrv_drain_recurse(bs); |
a77fd4bb | 254 | if (qemu_in_coroutine()) { |
b6e84c97 PB |
255 | bdrv_co_yield_to_drain(bs); |
256 | } else { | |
257 | bdrv_drain_poll(bs); | |
61007b31 | 258 | } |
6b98bd64 | 259 | bdrv_io_unplugged_end(bs); |
c2066af0 | 260 | bdrv_parent_drained_end(bs); |
61007b31 SH |
261 | } |
262 | ||
263 | /* | |
264 | * Wait for pending requests to complete across all BlockDriverStates | |
265 | * | |
266 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
267 | * after calling this function. | |
61007b31 SH |
268 | */ |
269 | void bdrv_drain_all(void) | |
270 | { | |
271 | /* Always run first iteration so any pending completion BHs run */ | |
272 | bool busy = true; | |
7c8eece4 | 273 | BlockDriverState *bs; |
88be7b4b | 274 | BdrvNextIterator it; |
f406c03c | 275 | GSList *aio_ctxs = NULL, *ctx; |
61007b31 | 276 | |
88be7b4b | 277 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
61007b31 SH |
278 | AioContext *aio_context = bdrv_get_aio_context(bs); |
279 | ||
280 | aio_context_acquire(aio_context); | |
281 | if (bs->job) { | |
282 | block_job_pause(bs->job); | |
283 | } | |
c2066af0 | 284 | bdrv_parent_drained_begin(bs); |
6b98bd64 | 285 | bdrv_io_unplugged_begin(bs); |
9dcf8ecd | 286 | bdrv_drain_recurse(bs); |
61007b31 | 287 | aio_context_release(aio_context); |
f406c03c | 288 | |
764ba3ae | 289 | if (!g_slist_find(aio_ctxs, aio_context)) { |
f406c03c AY |
290 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); |
291 | } | |
61007b31 SH |
292 | } |
293 | ||
7a63f3cd SH |
294 | /* Note that completion of an asynchronous I/O operation can trigger any |
295 | * number of other I/O operations on other devices---for example a | |
296 | * coroutine can submit an I/O request to another device in response to | |
297 | * request completion. Therefore we must keep looping until there was no | |
298 | * more activity rather than simply draining each device independently. | |
299 | */ | |
61007b31 SH |
300 | while (busy) { |
301 | busy = false; | |
61007b31 | 302 | |
f406c03c AY |
303 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
304 | AioContext *aio_context = ctx->data; | |
61007b31 SH |
305 | |
306 | aio_context_acquire(aio_context); | |
88be7b4b | 307 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
f406c03c | 308 | if (aio_context == bdrv_get_aio_context(bs)) { |
f406c03c AY |
309 | if (bdrv_requests_pending(bs)) { |
310 | busy = true; | |
311 | aio_poll(aio_context, busy); | |
312 | } | |
313 | } | |
314 | } | |
315 | busy |= aio_poll(aio_context, false); | |
61007b31 SH |
316 | aio_context_release(aio_context); |
317 | } | |
318 | } | |
319 | ||
88be7b4b | 320 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
61007b31 SH |
321 | AioContext *aio_context = bdrv_get_aio_context(bs); |
322 | ||
323 | aio_context_acquire(aio_context); | |
6b98bd64 | 324 | bdrv_io_unplugged_end(bs); |
c2066af0 | 325 | bdrv_parent_drained_end(bs); |
61007b31 SH |
326 | if (bs->job) { |
327 | block_job_resume(bs->job); | |
328 | } | |
329 | aio_context_release(aio_context); | |
330 | } | |
f406c03c | 331 | g_slist_free(aio_ctxs); |
61007b31 SH |
332 | } |
333 | ||
334 | /** | |
335 | * Remove an active request from the tracked requests list | |
336 | * | |
337 | * This function should be called when a tracked request is completing. | |
338 | */ | |
339 | static void tracked_request_end(BdrvTrackedRequest *req) | |
340 | { | |
341 | if (req->serialising) { | |
342 | req->bs->serialising_in_flight--; | |
343 | } | |
344 | ||
345 | QLIST_REMOVE(req, list); | |
346 | qemu_co_queue_restart_all(&req->wait_queue); | |
347 | } | |
348 | ||
349 | /** | |
350 | * Add an active request to the tracked requests list | |
351 | */ | |
352 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
353 | BlockDriverState *bs, | |
354 | int64_t offset, | |
ebde595c FZ |
355 | unsigned int bytes, |
356 | enum BdrvTrackedRequestType type) | |
61007b31 SH |
357 | { |
358 | *req = (BdrvTrackedRequest){ | |
359 | .bs = bs, | |
360 | .offset = offset, | |
361 | .bytes = bytes, | |
ebde595c | 362 | .type = type, |
61007b31 SH |
363 | .co = qemu_coroutine_self(), |
364 | .serialising = false, | |
365 | .overlap_offset = offset, | |
366 | .overlap_bytes = bytes, | |
367 | }; | |
368 | ||
369 | qemu_co_queue_init(&req->wait_queue); | |
370 | ||
371 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); | |
372 | } | |
373 | ||
374 | static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align) | |
375 | { | |
376 | int64_t overlap_offset = req->offset & ~(align - 1); | |
377 | unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align) | |
378 | - overlap_offset; | |
379 | ||
380 | if (!req->serialising) { | |
381 | req->bs->serialising_in_flight++; | |
382 | req->serialising = true; | |
383 | } | |
384 | ||
385 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
386 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
387 | } | |
388 | ||
389 | /** | |
390 | * Round a region to cluster boundaries | |
391 | */ | |
392 | void bdrv_round_to_clusters(BlockDriverState *bs, | |
393 | int64_t sector_num, int nb_sectors, | |
394 | int64_t *cluster_sector_num, | |
395 | int *cluster_nb_sectors) | |
396 | { | |
397 | BlockDriverInfo bdi; | |
398 | ||
399 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
400 | *cluster_sector_num = sector_num; | |
401 | *cluster_nb_sectors = nb_sectors; | |
402 | } else { | |
403 | int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; | |
404 | *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); | |
405 | *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + | |
406 | nb_sectors, c); | |
407 | } | |
408 | } | |
409 | ||
410 | static int bdrv_get_cluster_size(BlockDriverState *bs) | |
411 | { | |
412 | BlockDriverInfo bdi; | |
413 | int ret; | |
414 | ||
415 | ret = bdrv_get_info(bs, &bdi); | |
416 | if (ret < 0 || bdi.cluster_size == 0) { | |
417 | return bs->request_alignment; | |
418 | } else { | |
419 | return bdi.cluster_size; | |
420 | } | |
421 | } | |
422 | ||
423 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
424 | int64_t offset, unsigned int bytes) | |
425 | { | |
426 | /* aaaa bbbb */ | |
427 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
428 | return false; | |
429 | } | |
430 | /* bbbb aaaa */ | |
431 | if (req->overlap_offset >= offset + bytes) { | |
432 | return false; | |
433 | } | |
434 | return true; | |
435 | } | |
436 | ||
437 | static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self) | |
438 | { | |
439 | BlockDriverState *bs = self->bs; | |
440 | BdrvTrackedRequest *req; | |
441 | bool retry; | |
442 | bool waited = false; | |
443 | ||
444 | if (!bs->serialising_in_flight) { | |
445 | return false; | |
446 | } | |
447 | ||
448 | do { | |
449 | retry = false; | |
450 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
451 | if (req == self || (!req->serialising && !self->serialising)) { | |
452 | continue; | |
453 | } | |
454 | if (tracked_request_overlaps(req, self->overlap_offset, | |
455 | self->overlap_bytes)) | |
456 | { | |
457 | /* Hitting this means there was a reentrant request, for | |
458 | * example, a block driver issuing nested requests. This must | |
459 | * never happen since it means deadlock. | |
460 | */ | |
461 | assert(qemu_coroutine_self() != req->co); | |
462 | ||
463 | /* If the request is already (indirectly) waiting for us, or | |
464 | * will wait for us as soon as it wakes up, then just go on | |
465 | * (instead of producing a deadlock in the former case). */ | |
466 | if (!req->waiting_for) { | |
467 | self->waiting_for = req; | |
468 | qemu_co_queue_wait(&req->wait_queue); | |
469 | self->waiting_for = NULL; | |
470 | retry = true; | |
471 | waited = true; | |
472 | break; | |
473 | } | |
474 | } | |
475 | } | |
476 | } while (retry); | |
477 | ||
478 | return waited; | |
479 | } | |
480 | ||
481 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
482 | size_t size) | |
483 | { | |
484 | if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) { | |
485 | return -EIO; | |
486 | } | |
487 | ||
488 | if (!bdrv_is_inserted(bs)) { | |
489 | return -ENOMEDIUM; | |
490 | } | |
491 | ||
492 | if (offset < 0) { | |
493 | return -EIO; | |
494 | } | |
495 | ||
496 | return 0; | |
497 | } | |
498 | ||
499 | static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num, | |
500 | int nb_sectors) | |
501 | { | |
502 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
503 | return -EIO; | |
504 | } | |
505 | ||
506 | return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE, | |
507 | nb_sectors * BDRV_SECTOR_SIZE); | |
508 | } | |
509 | ||
510 | typedef struct RwCo { | |
511 | BlockDriverState *bs; | |
512 | int64_t offset; | |
513 | QEMUIOVector *qiov; | |
514 | bool is_write; | |
515 | int ret; | |
516 | BdrvRequestFlags flags; | |
517 | } RwCo; | |
518 | ||
519 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
520 | { | |
521 | RwCo *rwco = opaque; | |
522 | ||
523 | if (!rwco->is_write) { | |
cab3a356 KW |
524 | rwco->ret = bdrv_co_preadv(rwco->bs, rwco->offset, |
525 | rwco->qiov->size, rwco->qiov, | |
526 | rwco->flags); | |
61007b31 | 527 | } else { |
cab3a356 KW |
528 | rwco->ret = bdrv_co_pwritev(rwco->bs, rwco->offset, |
529 | rwco->qiov->size, rwco->qiov, | |
530 | rwco->flags); | |
61007b31 SH |
531 | } |
532 | } | |
533 | ||
534 | /* | |
535 | * Process a vectored synchronous request using coroutines | |
536 | */ | |
537 | static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset, | |
538 | QEMUIOVector *qiov, bool is_write, | |
539 | BdrvRequestFlags flags) | |
540 | { | |
541 | Coroutine *co; | |
542 | RwCo rwco = { | |
543 | .bs = bs, | |
544 | .offset = offset, | |
545 | .qiov = qiov, | |
546 | .is_write = is_write, | |
547 | .ret = NOT_DONE, | |
548 | .flags = flags, | |
549 | }; | |
550 | ||
61007b31 SH |
551 | if (qemu_in_coroutine()) { |
552 | /* Fast-path if already in coroutine context */ | |
553 | bdrv_rw_co_entry(&rwco); | |
554 | } else { | |
555 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
556 | ||
557 | co = qemu_coroutine_create(bdrv_rw_co_entry); | |
558 | qemu_coroutine_enter(co, &rwco); | |
559 | while (rwco.ret == NOT_DONE) { | |
560 | aio_poll(aio_context, true); | |
561 | } | |
562 | } | |
563 | return rwco.ret; | |
564 | } | |
565 | ||
566 | /* | |
567 | * Process a synchronous request using coroutines | |
568 | */ | |
569 | static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf, | |
570 | int nb_sectors, bool is_write, BdrvRequestFlags flags) | |
571 | { | |
572 | QEMUIOVector qiov; | |
573 | struct iovec iov = { | |
574 | .iov_base = (void *)buf, | |
575 | .iov_len = nb_sectors * BDRV_SECTOR_SIZE, | |
576 | }; | |
577 | ||
578 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
579 | return -EINVAL; | |
580 | } | |
581 | ||
582 | qemu_iovec_init_external(&qiov, &iov, 1); | |
583 | return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS, | |
584 | &qiov, is_write, flags); | |
585 | } | |
586 | ||
587 | /* return < 0 if error. See bdrv_write() for the return codes */ | |
588 | int bdrv_read(BlockDriverState *bs, int64_t sector_num, | |
589 | uint8_t *buf, int nb_sectors) | |
590 | { | |
591 | return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0); | |
592 | } | |
593 | ||
61007b31 SH |
594 | /* Return < 0 if error. Important errors are: |
595 | -EIO generic I/O error (may happen for all errors) | |
596 | -ENOMEDIUM No media inserted. | |
597 | -EINVAL Invalid sector number or nb_sectors | |
598 | -EACCES Trying to write a read-only device | |
599 | */ | |
600 | int bdrv_write(BlockDriverState *bs, int64_t sector_num, | |
601 | const uint8_t *buf, int nb_sectors) | |
602 | { | |
603 | return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0); | |
604 | } | |
605 | ||
606 | int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num, | |
607 | int nb_sectors, BdrvRequestFlags flags) | |
608 | { | |
609 | return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true, | |
610 | BDRV_REQ_ZERO_WRITE | flags); | |
611 | } | |
612 | ||
613 | /* | |
614 | * Completely zero out a block device with the help of bdrv_write_zeroes. | |
615 | * The operation is sped up by checking the block status and only writing | |
616 | * zeroes to the device if they currently do not return zeroes. Optional | |
465fe887 EB |
617 | * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
618 | * BDRV_REQ_FUA). | |
61007b31 SH |
619 | * |
620 | * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). | |
621 | */ | |
622 | int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags) | |
623 | { | |
624 | int64_t target_sectors, ret, nb_sectors, sector_num = 0; | |
67a0fd2a | 625 | BlockDriverState *file; |
61007b31 SH |
626 | int n; |
627 | ||
628 | target_sectors = bdrv_nb_sectors(bs); | |
629 | if (target_sectors < 0) { | |
630 | return target_sectors; | |
631 | } | |
632 | ||
633 | for (;;) { | |
634 | nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS); | |
635 | if (nb_sectors <= 0) { | |
636 | return 0; | |
637 | } | |
67a0fd2a | 638 | ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n, &file); |
61007b31 SH |
639 | if (ret < 0) { |
640 | error_report("error getting block status at sector %" PRId64 ": %s", | |
641 | sector_num, strerror(-ret)); | |
642 | return ret; | |
643 | } | |
644 | if (ret & BDRV_BLOCK_ZERO) { | |
645 | sector_num += n; | |
646 | continue; | |
647 | } | |
648 | ret = bdrv_write_zeroes(bs, sector_num, n, flags); | |
649 | if (ret < 0) { | |
650 | error_report("error writing zeroes at sector %" PRId64 ": %s", | |
651 | sector_num, strerror(-ret)); | |
652 | return ret; | |
653 | } | |
654 | sector_num += n; | |
655 | } | |
656 | } | |
657 | ||
658 | int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes) | |
659 | { | |
660 | QEMUIOVector qiov; | |
661 | struct iovec iov = { | |
662 | .iov_base = (void *)buf, | |
663 | .iov_len = bytes, | |
664 | }; | |
665 | int ret; | |
666 | ||
667 | if (bytes < 0) { | |
668 | return -EINVAL; | |
669 | } | |
670 | ||
671 | qemu_iovec_init_external(&qiov, &iov, 1); | |
672 | ret = bdrv_prwv_co(bs, offset, &qiov, false, 0); | |
673 | if (ret < 0) { | |
674 | return ret; | |
675 | } | |
676 | ||
677 | return bytes; | |
678 | } | |
679 | ||
680 | int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov) | |
681 | { | |
682 | int ret; | |
683 | ||
684 | ret = bdrv_prwv_co(bs, offset, qiov, true, 0); | |
685 | if (ret < 0) { | |
686 | return ret; | |
687 | } | |
688 | ||
689 | return qiov->size; | |
690 | } | |
691 | ||
692 | int bdrv_pwrite(BlockDriverState *bs, int64_t offset, | |
693 | const void *buf, int bytes) | |
694 | { | |
695 | QEMUIOVector qiov; | |
696 | struct iovec iov = { | |
697 | .iov_base = (void *) buf, | |
698 | .iov_len = bytes, | |
699 | }; | |
700 | ||
701 | if (bytes < 0) { | |
702 | return -EINVAL; | |
703 | } | |
704 | ||
705 | qemu_iovec_init_external(&qiov, &iov, 1); | |
706 | return bdrv_pwritev(bs, offset, &qiov); | |
707 | } | |
708 | ||
709 | /* | |
710 | * Writes to the file and ensures that no writes are reordered across this | |
711 | * request (acts as a barrier) | |
712 | * | |
713 | * Returns 0 on success, -errno in error cases. | |
714 | */ | |
715 | int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset, | |
716 | const void *buf, int count) | |
717 | { | |
718 | int ret; | |
719 | ||
720 | ret = bdrv_pwrite(bs, offset, buf, count); | |
721 | if (ret < 0) { | |
722 | return ret; | |
723 | } | |
724 | ||
855a6a93 KW |
725 | ret = bdrv_flush(bs); |
726 | if (ret < 0) { | |
727 | return ret; | |
61007b31 SH |
728 | } |
729 | ||
730 | return 0; | |
731 | } | |
732 | ||
08844473 KW |
733 | typedef struct CoroutineIOCompletion { |
734 | Coroutine *coroutine; | |
735 | int ret; | |
736 | } CoroutineIOCompletion; | |
737 | ||
738 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
739 | { | |
740 | CoroutineIOCompletion *co = opaque; | |
741 | ||
742 | co->ret = ret; | |
743 | qemu_coroutine_enter(co->coroutine, NULL); | |
744 | } | |
745 | ||
166fe960 KW |
746 | static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs, |
747 | uint64_t offset, uint64_t bytes, | |
748 | QEMUIOVector *qiov, int flags) | |
749 | { | |
750 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
751 | int64_t sector_num; |
752 | unsigned int nb_sectors; | |
753 | ||
754 | if (drv->bdrv_co_preadv) { | |
755 | return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); | |
756 | } | |
757 | ||
758 | sector_num = offset >> BDRV_SECTOR_BITS; | |
759 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
166fe960 KW |
760 | |
761 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
762 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
763 | assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS); | |
764 | ||
08844473 KW |
765 | if (drv->bdrv_co_readv) { |
766 | return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
767 | } else { | |
768 | BlockAIOCB *acb; | |
769 | CoroutineIOCompletion co = { | |
770 | .coroutine = qemu_coroutine_self(), | |
771 | }; | |
772 | ||
773 | acb = bs->drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors, | |
774 | bdrv_co_io_em_complete, &co); | |
775 | if (acb == NULL) { | |
776 | return -EIO; | |
777 | } else { | |
778 | qemu_coroutine_yield(); | |
779 | return co.ret; | |
780 | } | |
781 | } | |
166fe960 KW |
782 | } |
783 | ||
78a07294 KW |
784 | static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs, |
785 | uint64_t offset, uint64_t bytes, | |
786 | QEMUIOVector *qiov, int flags) | |
787 | { | |
788 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
789 | int64_t sector_num; |
790 | unsigned int nb_sectors; | |
78a07294 KW |
791 | int ret; |
792 | ||
3fb06697 KW |
793 | if (drv->bdrv_co_pwritev) { |
794 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, flags); | |
795 | goto emulate_flags; | |
796 | } | |
797 | ||
798 | sector_num = offset >> BDRV_SECTOR_BITS; | |
799 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
800 | ||
78a07294 KW |
801 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); |
802 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
803 | assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS); | |
804 | ||
805 | if (drv->bdrv_co_writev_flags) { | |
806 | ret = drv->bdrv_co_writev_flags(bs, sector_num, nb_sectors, qiov, | |
4df863f3 EB |
807 | flags & bs->supported_write_flags); |
808 | flags &= ~bs->supported_write_flags; | |
08844473 | 809 | } else if (drv->bdrv_co_writev) { |
4df863f3 | 810 | assert(!bs->supported_write_flags); |
78a07294 | 811 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); |
08844473 KW |
812 | } else { |
813 | BlockAIOCB *acb; | |
814 | CoroutineIOCompletion co = { | |
815 | .coroutine = qemu_coroutine_self(), | |
816 | }; | |
817 | ||
818 | acb = bs->drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors, | |
819 | bdrv_co_io_em_complete, &co); | |
820 | if (acb == NULL) { | |
3fb06697 | 821 | ret = -EIO; |
08844473 KW |
822 | } else { |
823 | qemu_coroutine_yield(); | |
3fb06697 | 824 | ret = co.ret; |
08844473 | 825 | } |
78a07294 KW |
826 | } |
827 | ||
3fb06697 | 828 | emulate_flags: |
4df863f3 | 829 | if (ret == 0 && (flags & BDRV_REQ_FUA)) { |
78a07294 KW |
830 | ret = bdrv_co_flush(bs); |
831 | } | |
832 | ||
833 | return ret; | |
834 | } | |
835 | ||
61007b31 SH |
836 | static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, |
837 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
838 | { | |
839 | /* Perform I/O through a temporary buffer so that users who scribble over | |
840 | * their read buffer while the operation is in progress do not end up | |
841 | * modifying the image file. This is critical for zero-copy guest I/O | |
842 | * where anything might happen inside guest memory. | |
843 | */ | |
844 | void *bounce_buffer; | |
845 | ||
846 | BlockDriver *drv = bs->drv; | |
847 | struct iovec iov; | |
848 | QEMUIOVector bounce_qiov; | |
849 | int64_t cluster_sector_num; | |
850 | int cluster_nb_sectors; | |
851 | size_t skip_bytes; | |
852 | int ret; | |
853 | ||
854 | /* Cover entire cluster so no additional backing file I/O is required when | |
855 | * allocating cluster in the image file. | |
856 | */ | |
857 | bdrv_round_to_clusters(bs, sector_num, nb_sectors, | |
858 | &cluster_sector_num, &cluster_nb_sectors); | |
859 | ||
860 | trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, | |
861 | cluster_sector_num, cluster_nb_sectors); | |
862 | ||
863 | iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE; | |
864 | iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len); | |
865 | if (bounce_buffer == NULL) { | |
866 | ret = -ENOMEM; | |
867 | goto err; | |
868 | } | |
869 | ||
870 | qemu_iovec_init_external(&bounce_qiov, &iov, 1); | |
871 | ||
166fe960 KW |
872 | ret = bdrv_driver_preadv(bs, cluster_sector_num * BDRV_SECTOR_SIZE, |
873 | cluster_nb_sectors * BDRV_SECTOR_SIZE, | |
874 | &bounce_qiov, 0); | |
61007b31 SH |
875 | if (ret < 0) { |
876 | goto err; | |
877 | } | |
878 | ||
879 | if (drv->bdrv_co_write_zeroes && | |
880 | buffer_is_zero(bounce_buffer, iov.iov_len)) { | |
881 | ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num, | |
882 | cluster_nb_sectors, 0); | |
883 | } else { | |
884 | /* This does not change the data on the disk, it is not necessary | |
885 | * to flush even in cache=writethrough mode. | |
886 | */ | |
78a07294 KW |
887 | ret = bdrv_driver_pwritev(bs, cluster_sector_num * BDRV_SECTOR_SIZE, |
888 | cluster_nb_sectors * BDRV_SECTOR_SIZE, | |
889 | &bounce_qiov, 0); | |
61007b31 SH |
890 | } |
891 | ||
892 | if (ret < 0) { | |
893 | /* It might be okay to ignore write errors for guest requests. If this | |
894 | * is a deliberate copy-on-read then we don't want to ignore the error. | |
895 | * Simply report it in all cases. | |
896 | */ | |
897 | goto err; | |
898 | } | |
899 | ||
900 | skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE; | |
901 | qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, | |
902 | nb_sectors * BDRV_SECTOR_SIZE); | |
903 | ||
904 | err: | |
905 | qemu_vfree(bounce_buffer); | |
906 | return ret; | |
907 | } | |
908 | ||
909 | /* | |
910 | * Forwards an already correctly aligned request to the BlockDriver. This | |
911 | * handles copy on read and zeroing after EOF; any other features must be | |
912 | * implemented by the caller. | |
913 | */ | |
914 | static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs, | |
915 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, | |
916 | int64_t align, QEMUIOVector *qiov, int flags) | |
917 | { | |
61007b31 SH |
918 | int ret; |
919 | ||
920 | int64_t sector_num = offset >> BDRV_SECTOR_BITS; | |
921 | unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
922 | ||
923 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
924 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
925 | assert(!qiov || bytes == qiov->size); | |
abb06c5a | 926 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
61007b31 SH |
927 | |
928 | /* Handle Copy on Read and associated serialisation */ | |
929 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
930 | /* If we touch the same cluster it counts as an overlap. This | |
931 | * guarantees that allocating writes will be serialized and not race | |
932 | * with each other for the same cluster. For example, in copy-on-read | |
933 | * it ensures that the CoR read and write operations are atomic and | |
934 | * guest writes cannot interleave between them. */ | |
935 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
936 | } | |
937 | ||
61408b25 FZ |
938 | if (!(flags & BDRV_REQ_NO_SERIALISING)) { |
939 | wait_serialising_requests(req); | |
940 | } | |
61007b31 SH |
941 | |
942 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
943 | int pnum; | |
944 | ||
945 | ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum); | |
946 | if (ret < 0) { | |
947 | goto out; | |
948 | } | |
949 | ||
950 | if (!ret || pnum != nb_sectors) { | |
951 | ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov); | |
952 | goto out; | |
953 | } | |
954 | } | |
955 | ||
956 | /* Forward the request to the BlockDriver */ | |
957 | if (!bs->zero_beyond_eof) { | |
166fe960 | 958 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0); |
61007b31 SH |
959 | } else { |
960 | /* Read zeros after EOF */ | |
961 | int64_t total_sectors, max_nb_sectors; | |
962 | ||
963 | total_sectors = bdrv_nb_sectors(bs); | |
964 | if (total_sectors < 0) { | |
965 | ret = total_sectors; | |
966 | goto out; | |
967 | } | |
968 | ||
969 | max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num), | |
970 | align >> BDRV_SECTOR_BITS); | |
971 | if (nb_sectors < max_nb_sectors) { | |
166fe960 | 972 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0); |
61007b31 SH |
973 | } else if (max_nb_sectors > 0) { |
974 | QEMUIOVector local_qiov; | |
975 | ||
976 | qemu_iovec_init(&local_qiov, qiov->niov); | |
977 | qemu_iovec_concat(&local_qiov, qiov, 0, | |
978 | max_nb_sectors * BDRV_SECTOR_SIZE); | |
979 | ||
166fe960 KW |
980 | ret = bdrv_driver_preadv(bs, offset, |
981 | max_nb_sectors * BDRV_SECTOR_SIZE, | |
982 | &local_qiov, 0); | |
61007b31 SH |
983 | |
984 | qemu_iovec_destroy(&local_qiov); | |
985 | } else { | |
986 | ret = 0; | |
987 | } | |
988 | ||
989 | /* Reading beyond end of file is supposed to produce zeroes */ | |
990 | if (ret == 0 && total_sectors < sector_num + nb_sectors) { | |
991 | uint64_t offset = MAX(0, total_sectors - sector_num); | |
992 | uint64_t bytes = (sector_num + nb_sectors - offset) * | |
993 | BDRV_SECTOR_SIZE; | |
994 | qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes); | |
995 | } | |
996 | } | |
997 | ||
998 | out: | |
999 | return ret; | |
1000 | } | |
1001 | ||
61007b31 SH |
1002 | /* |
1003 | * Handle a read request in coroutine context | |
1004 | */ | |
cab3a356 | 1005 | int coroutine_fn bdrv_co_preadv(BlockDriverState *bs, |
61007b31 SH |
1006 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1007 | BdrvRequestFlags flags) | |
1008 | { | |
1009 | BlockDriver *drv = bs->drv; | |
1010 | BdrvTrackedRequest req; | |
1011 | ||
d01c07f2 FZ |
1012 | /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */ |
1013 | uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment); | |
61007b31 SH |
1014 | uint8_t *head_buf = NULL; |
1015 | uint8_t *tail_buf = NULL; | |
1016 | QEMUIOVector local_qiov; | |
1017 | bool use_local_qiov = false; | |
1018 | int ret; | |
1019 | ||
1020 | if (!drv) { | |
1021 | return -ENOMEDIUM; | |
1022 | } | |
1023 | ||
1024 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1025 | if (ret < 0) { | |
1026 | return ret; | |
1027 | } | |
1028 | ||
9568b511 | 1029 | /* Don't do copy-on-read if we read data before write operation */ |
61408b25 | 1030 | if (bs->copy_on_read && !(flags & BDRV_REQ_NO_SERIALISING)) { |
61007b31 SH |
1031 | flags |= BDRV_REQ_COPY_ON_READ; |
1032 | } | |
1033 | ||
61007b31 SH |
1034 | /* Align read if necessary by padding qiov */ |
1035 | if (offset & (align - 1)) { | |
1036 | head_buf = qemu_blockalign(bs, align); | |
1037 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1038 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1039 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1040 | use_local_qiov = true; | |
1041 | ||
1042 | bytes += offset & (align - 1); | |
1043 | offset = offset & ~(align - 1); | |
1044 | } | |
1045 | ||
1046 | if ((offset + bytes) & (align - 1)) { | |
1047 | if (!use_local_qiov) { | |
1048 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1049 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1050 | use_local_qiov = true; | |
1051 | } | |
1052 | tail_buf = qemu_blockalign(bs, align); | |
1053 | qemu_iovec_add(&local_qiov, tail_buf, | |
1054 | align - ((offset + bytes) & (align - 1))); | |
1055 | ||
1056 | bytes = ROUND_UP(bytes, align); | |
1057 | } | |
1058 | ||
ebde595c | 1059 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
61007b31 SH |
1060 | ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align, |
1061 | use_local_qiov ? &local_qiov : qiov, | |
1062 | flags); | |
1063 | tracked_request_end(&req); | |
1064 | ||
1065 | if (use_local_qiov) { | |
1066 | qemu_iovec_destroy(&local_qiov); | |
1067 | qemu_vfree(head_buf); | |
1068 | qemu_vfree(tail_buf); | |
1069 | } | |
1070 | ||
1071 | return ret; | |
1072 | } | |
1073 | ||
1074 | static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, | |
1075 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
1076 | BdrvRequestFlags flags) | |
1077 | { | |
1078 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
1079 | return -EINVAL; | |
1080 | } | |
1081 | ||
cab3a356 KW |
1082 | return bdrv_co_preadv(bs, sector_num << BDRV_SECTOR_BITS, |
1083 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); | |
61007b31 SH |
1084 | } |
1085 | ||
1086 | int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num, | |
1087 | int nb_sectors, QEMUIOVector *qiov) | |
1088 | { | |
1089 | trace_bdrv_co_readv(bs, sector_num, nb_sectors); | |
1090 | ||
1091 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0); | |
1092 | } | |
1093 | ||
61408b25 | 1094 | int coroutine_fn bdrv_co_readv_no_serialising(BlockDriverState *bs, |
9568b511 WC |
1095 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) |
1096 | { | |
61408b25 | 1097 | trace_bdrv_co_readv_no_serialising(bs, sector_num, nb_sectors); |
9568b511 WC |
1098 | |
1099 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, | |
61408b25 | 1100 | BDRV_REQ_NO_SERIALISING); |
9568b511 WC |
1101 | } |
1102 | ||
61007b31 SH |
1103 | int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs, |
1104 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
1105 | { | |
1106 | trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors); | |
1107 | ||
1108 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, | |
1109 | BDRV_REQ_COPY_ON_READ); | |
1110 | } | |
1111 | ||
1112 | #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768 | |
1113 | ||
1114 | static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, | |
1115 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags) | |
1116 | { | |
1117 | BlockDriver *drv = bs->drv; | |
1118 | QEMUIOVector qiov; | |
1119 | struct iovec iov = {0}; | |
1120 | int ret = 0; | |
465fe887 | 1121 | bool need_flush = false; |
61007b31 SH |
1122 | |
1123 | int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_write_zeroes, | |
1124 | BDRV_REQUEST_MAX_SECTORS); | |
1125 | ||
1126 | while (nb_sectors > 0 && !ret) { | |
1127 | int num = nb_sectors; | |
1128 | ||
1129 | /* Align request. Block drivers can expect the "bulk" of the request | |
1130 | * to be aligned. | |
1131 | */ | |
1132 | if (bs->bl.write_zeroes_alignment | |
1133 | && num > bs->bl.write_zeroes_alignment) { | |
1134 | if (sector_num % bs->bl.write_zeroes_alignment != 0) { | |
1135 | /* Make a small request up to the first aligned sector. */ | |
1136 | num = bs->bl.write_zeroes_alignment; | |
1137 | num -= sector_num % bs->bl.write_zeroes_alignment; | |
1138 | } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) { | |
1139 | /* Shorten the request to the last aligned sector. num cannot | |
1140 | * underflow because num > bs->bl.write_zeroes_alignment. | |
1141 | */ | |
1142 | num -= (sector_num + num) % bs->bl.write_zeroes_alignment; | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | /* limit request size */ | |
1147 | if (num > max_write_zeroes) { | |
1148 | num = max_write_zeroes; | |
1149 | } | |
1150 | ||
1151 | ret = -ENOTSUP; | |
1152 | /* First try the efficient write zeroes operation */ | |
1153 | if (drv->bdrv_co_write_zeroes) { | |
465fe887 EB |
1154 | ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, |
1155 | flags & bs->supported_zero_flags); | |
1156 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1157 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1158 | need_flush = true; | |
1159 | } | |
1160 | } else { | |
1161 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1162 | } |
1163 | ||
1164 | if (ret == -ENOTSUP) { | |
1165 | /* Fall back to bounce buffer if write zeroes is unsupported */ | |
1166 | int max_xfer_len = MIN_NON_ZERO(bs->bl.max_transfer_length, | |
1167 | MAX_WRITE_ZEROES_BOUNCE_BUFFER); | |
465fe887 EB |
1168 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1169 | ||
1170 | if ((flags & BDRV_REQ_FUA) && | |
1171 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1172 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1173 | * flush on each chunk; use just one at the end */ | |
1174 | write_flags &= ~BDRV_REQ_FUA; | |
1175 | need_flush = true; | |
1176 | } | |
61007b31 SH |
1177 | num = MIN(num, max_xfer_len); |
1178 | iov.iov_len = num * BDRV_SECTOR_SIZE; | |
1179 | if (iov.iov_base == NULL) { | |
1180 | iov.iov_base = qemu_try_blockalign(bs, num * BDRV_SECTOR_SIZE); | |
1181 | if (iov.iov_base == NULL) { | |
1182 | ret = -ENOMEM; | |
1183 | goto fail; | |
1184 | } | |
1185 | memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE); | |
1186 | } | |
1187 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1188 | ||
78a07294 | 1189 | ret = bdrv_driver_pwritev(bs, sector_num * BDRV_SECTOR_SIZE, |
465fe887 EB |
1190 | num * BDRV_SECTOR_SIZE, &qiov, |
1191 | write_flags); | |
61007b31 SH |
1192 | |
1193 | /* Keep bounce buffer around if it is big enough for all | |
1194 | * all future requests. | |
1195 | */ | |
1196 | if (num < max_xfer_len) { | |
1197 | qemu_vfree(iov.iov_base); | |
1198 | iov.iov_base = NULL; | |
1199 | } | |
1200 | } | |
1201 | ||
1202 | sector_num += num; | |
1203 | nb_sectors -= num; | |
1204 | } | |
1205 | ||
1206 | fail: | |
465fe887 EB |
1207 | if (ret == 0 && need_flush) { |
1208 | ret = bdrv_co_flush(bs); | |
1209 | } | |
61007b31 SH |
1210 | qemu_vfree(iov.iov_base); |
1211 | return ret; | |
1212 | } | |
1213 | ||
1214 | /* | |
1215 | * Forwards an already correctly aligned write request to the BlockDriver. | |
1216 | */ | |
1217 | static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs, | |
1218 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, | |
1219 | QEMUIOVector *qiov, int flags) | |
1220 | { | |
1221 | BlockDriver *drv = bs->drv; | |
1222 | bool waited; | |
1223 | int ret; | |
1224 | ||
1225 | int64_t sector_num = offset >> BDRV_SECTOR_BITS; | |
1226 | unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1227 | ||
1228 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1229 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1230 | assert(!qiov || bytes == qiov->size); | |
abb06c5a | 1231 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
61007b31 SH |
1232 | |
1233 | waited = wait_serialising_requests(req); | |
1234 | assert(!waited || !req->serialising); | |
1235 | assert(req->overlap_offset <= offset); | |
1236 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
1237 | ||
1238 | ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req); | |
1239 | ||
1240 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
1241 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_write_zeroes && | |
1242 | qemu_iovec_is_zero(qiov)) { | |
1243 | flags |= BDRV_REQ_ZERO_WRITE; | |
1244 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
1245 | flags |= BDRV_REQ_MAY_UNMAP; | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | if (ret < 0) { | |
1250 | /* Do nothing, write notifier decided to fail this request */ | |
1251 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
9a4f4c31 | 1252 | bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
61007b31 SH |
1253 | ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags); |
1254 | } else { | |
9a4f4c31 | 1255 | bdrv_debug_event(bs, BLKDBG_PWRITEV); |
78a07294 | 1256 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags); |
61007b31 | 1257 | } |
9a4f4c31 | 1258 | bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 1259 | |
61007b31 SH |
1260 | bdrv_set_dirty(bs, sector_num, nb_sectors); |
1261 | ||
53d8f9d8 HR |
1262 | if (bs->wr_highest_offset < offset + bytes) { |
1263 | bs->wr_highest_offset = offset + bytes; | |
1264 | } | |
61007b31 SH |
1265 | |
1266 | if (ret >= 0) { | |
1267 | bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors); | |
1268 | } | |
1269 | ||
1270 | return ret; | |
1271 | } | |
1272 | ||
9eeb6dd1 FZ |
1273 | static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs, |
1274 | int64_t offset, | |
1275 | unsigned int bytes, | |
1276 | BdrvRequestFlags flags, | |
1277 | BdrvTrackedRequest *req) | |
1278 | { | |
1279 | uint8_t *buf = NULL; | |
1280 | QEMUIOVector local_qiov; | |
1281 | struct iovec iov; | |
1282 | uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment); | |
1283 | unsigned int head_padding_bytes, tail_padding_bytes; | |
1284 | int ret = 0; | |
1285 | ||
1286 | head_padding_bytes = offset & (align - 1); | |
1287 | tail_padding_bytes = align - ((offset + bytes) & (align - 1)); | |
1288 | ||
1289 | ||
1290 | assert(flags & BDRV_REQ_ZERO_WRITE); | |
1291 | if (head_padding_bytes || tail_padding_bytes) { | |
1292 | buf = qemu_blockalign(bs, align); | |
1293 | iov = (struct iovec) { | |
1294 | .iov_base = buf, | |
1295 | .iov_len = align, | |
1296 | }; | |
1297 | qemu_iovec_init_external(&local_qiov, &iov, 1); | |
1298 | } | |
1299 | if (head_padding_bytes) { | |
1300 | uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes); | |
1301 | ||
1302 | /* RMW the unaligned part before head. */ | |
1303 | mark_request_serialising(req, align); | |
1304 | wait_serialising_requests(req); | |
9a4f4c31 | 1305 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
9eeb6dd1 FZ |
1306 | ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align, |
1307 | align, &local_qiov, 0); | |
1308 | if (ret < 0) { | |
1309 | goto fail; | |
1310 | } | |
9a4f4c31 | 1311 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
9eeb6dd1 FZ |
1312 | |
1313 | memset(buf + head_padding_bytes, 0, zero_bytes); | |
1314 | ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align, | |
1315 | &local_qiov, | |
1316 | flags & ~BDRV_REQ_ZERO_WRITE); | |
1317 | if (ret < 0) { | |
1318 | goto fail; | |
1319 | } | |
1320 | offset += zero_bytes; | |
1321 | bytes -= zero_bytes; | |
1322 | } | |
1323 | ||
1324 | assert(!bytes || (offset & (align - 1)) == 0); | |
1325 | if (bytes >= align) { | |
1326 | /* Write the aligned part in the middle. */ | |
1327 | uint64_t aligned_bytes = bytes & ~(align - 1); | |
1328 | ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes, | |
1329 | NULL, flags); | |
1330 | if (ret < 0) { | |
1331 | goto fail; | |
1332 | } | |
1333 | bytes -= aligned_bytes; | |
1334 | offset += aligned_bytes; | |
1335 | } | |
1336 | ||
1337 | assert(!bytes || (offset & (align - 1)) == 0); | |
1338 | if (bytes) { | |
1339 | assert(align == tail_padding_bytes + bytes); | |
1340 | /* RMW the unaligned part after tail. */ | |
1341 | mark_request_serialising(req, align); | |
1342 | wait_serialising_requests(req); | |
9a4f4c31 | 1343 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
9eeb6dd1 FZ |
1344 | ret = bdrv_aligned_preadv(bs, req, offset, align, |
1345 | align, &local_qiov, 0); | |
1346 | if (ret < 0) { | |
1347 | goto fail; | |
1348 | } | |
9a4f4c31 | 1349 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
9eeb6dd1 FZ |
1350 | |
1351 | memset(buf, 0, bytes); | |
1352 | ret = bdrv_aligned_pwritev(bs, req, offset, align, | |
1353 | &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE); | |
1354 | } | |
1355 | fail: | |
1356 | qemu_vfree(buf); | |
1357 | return ret; | |
1358 | ||
1359 | } | |
1360 | ||
61007b31 SH |
1361 | /* |
1362 | * Handle a write request in coroutine context | |
1363 | */ | |
cab3a356 | 1364 | int coroutine_fn bdrv_co_pwritev(BlockDriverState *bs, |
61007b31 SH |
1365 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1366 | BdrvRequestFlags flags) | |
1367 | { | |
1368 | BdrvTrackedRequest req; | |
d01c07f2 FZ |
1369 | /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */ |
1370 | uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment); | |
61007b31 SH |
1371 | uint8_t *head_buf = NULL; |
1372 | uint8_t *tail_buf = NULL; | |
1373 | QEMUIOVector local_qiov; | |
1374 | bool use_local_qiov = false; | |
1375 | int ret; | |
1376 | ||
1377 | if (!bs->drv) { | |
1378 | return -ENOMEDIUM; | |
1379 | } | |
1380 | if (bs->read_only) { | |
eaf5fe2d | 1381 | return -EPERM; |
61007b31 | 1382 | } |
04c01a5c | 1383 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
61007b31 SH |
1384 | |
1385 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1386 | if (ret < 0) { | |
1387 | return ret; | |
1388 | } | |
1389 | ||
61007b31 SH |
1390 | /* |
1391 | * Align write if necessary by performing a read-modify-write cycle. | |
1392 | * Pad qiov with the read parts and be sure to have a tracked request not | |
1393 | * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle. | |
1394 | */ | |
ebde595c | 1395 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 1396 | |
9eeb6dd1 FZ |
1397 | if (!qiov) { |
1398 | ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req); | |
1399 | goto out; | |
1400 | } | |
1401 | ||
61007b31 SH |
1402 | if (offset & (align - 1)) { |
1403 | QEMUIOVector head_qiov; | |
1404 | struct iovec head_iov; | |
1405 | ||
1406 | mark_request_serialising(&req, align); | |
1407 | wait_serialising_requests(&req); | |
1408 | ||
1409 | head_buf = qemu_blockalign(bs, align); | |
1410 | head_iov = (struct iovec) { | |
1411 | .iov_base = head_buf, | |
1412 | .iov_len = align, | |
1413 | }; | |
1414 | qemu_iovec_init_external(&head_qiov, &head_iov, 1); | |
1415 | ||
9a4f4c31 | 1416 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
61007b31 SH |
1417 | ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align, |
1418 | align, &head_qiov, 0); | |
1419 | if (ret < 0) { | |
1420 | goto fail; | |
1421 | } | |
9a4f4c31 | 1422 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
61007b31 SH |
1423 | |
1424 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1425 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1426 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1427 | use_local_qiov = true; | |
1428 | ||
1429 | bytes += offset & (align - 1); | |
1430 | offset = offset & ~(align - 1); | |
1431 | } | |
1432 | ||
1433 | if ((offset + bytes) & (align - 1)) { | |
1434 | QEMUIOVector tail_qiov; | |
1435 | struct iovec tail_iov; | |
1436 | size_t tail_bytes; | |
1437 | bool waited; | |
1438 | ||
1439 | mark_request_serialising(&req, align); | |
1440 | waited = wait_serialising_requests(&req); | |
1441 | assert(!waited || !use_local_qiov); | |
1442 | ||
1443 | tail_buf = qemu_blockalign(bs, align); | |
1444 | tail_iov = (struct iovec) { | |
1445 | .iov_base = tail_buf, | |
1446 | .iov_len = align, | |
1447 | }; | |
1448 | qemu_iovec_init_external(&tail_qiov, &tail_iov, 1); | |
1449 | ||
9a4f4c31 | 1450 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
61007b31 SH |
1451 | ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align, |
1452 | align, &tail_qiov, 0); | |
1453 | if (ret < 0) { | |
1454 | goto fail; | |
1455 | } | |
9a4f4c31 | 1456 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
61007b31 SH |
1457 | |
1458 | if (!use_local_qiov) { | |
1459 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1460 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1461 | use_local_qiov = true; | |
1462 | } | |
1463 | ||
1464 | tail_bytes = (offset + bytes) & (align - 1); | |
1465 | qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes); | |
1466 | ||
1467 | bytes = ROUND_UP(bytes, align); | |
1468 | } | |
1469 | ||
61007b31 SH |
1470 | ret = bdrv_aligned_pwritev(bs, &req, offset, bytes, |
1471 | use_local_qiov ? &local_qiov : qiov, | |
1472 | flags); | |
1473 | ||
1474 | fail: | |
61007b31 SH |
1475 | |
1476 | if (use_local_qiov) { | |
1477 | qemu_iovec_destroy(&local_qiov); | |
1478 | } | |
1479 | qemu_vfree(head_buf); | |
1480 | qemu_vfree(tail_buf); | |
9eeb6dd1 FZ |
1481 | out: |
1482 | tracked_request_end(&req); | |
61007b31 SH |
1483 | return ret; |
1484 | } | |
1485 | ||
1486 | static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, | |
1487 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
1488 | BdrvRequestFlags flags) | |
1489 | { | |
1490 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
1491 | return -EINVAL; | |
1492 | } | |
1493 | ||
cab3a356 KW |
1494 | return bdrv_co_pwritev(bs, sector_num << BDRV_SECTOR_BITS, |
1495 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); | |
61007b31 SH |
1496 | } |
1497 | ||
1498 | int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num, | |
1499 | int nb_sectors, QEMUIOVector *qiov) | |
1500 | { | |
1501 | trace_bdrv_co_writev(bs, sector_num, nb_sectors); | |
1502 | ||
1503 | return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0); | |
1504 | } | |
1505 | ||
1506 | int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs, | |
1507 | int64_t sector_num, int nb_sectors, | |
1508 | BdrvRequestFlags flags) | |
1509 | { | |
61007b31 SH |
1510 | trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags); |
1511 | ||
1512 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
1513 | flags &= ~BDRV_REQ_MAY_UNMAP; | |
1514 | } | |
61007b31 | 1515 | |
d01c07f2 FZ |
1516 | return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL, |
1517 | BDRV_REQ_ZERO_WRITE | flags); | |
61007b31 SH |
1518 | } |
1519 | ||
61007b31 SH |
1520 | typedef struct BdrvCoGetBlockStatusData { |
1521 | BlockDriverState *bs; | |
1522 | BlockDriverState *base; | |
67a0fd2a | 1523 | BlockDriverState **file; |
61007b31 SH |
1524 | int64_t sector_num; |
1525 | int nb_sectors; | |
1526 | int *pnum; | |
1527 | int64_t ret; | |
1528 | bool done; | |
1529 | } BdrvCoGetBlockStatusData; | |
1530 | ||
1531 | /* | |
1532 | * Returns the allocation status of the specified sectors. | |
1533 | * Drivers not implementing the functionality are assumed to not support | |
1534 | * backing files, hence all their sectors are reported as allocated. | |
1535 | * | |
1536 | * If 'sector_num' is beyond the end of the disk image the return value is 0 | |
1537 | * and 'pnum' is set to 0. | |
1538 | * | |
1539 | * 'pnum' is set to the number of sectors (including and immediately following | |
1540 | * the specified sector) that are known to be in the same | |
1541 | * allocated/unallocated state. | |
1542 | * | |
1543 | * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes | |
1544 | * beyond the end of the disk image it will be clamped. | |
67a0fd2a FZ |
1545 | * |
1546 | * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file' | |
1547 | * points to the BDS which the sector range is allocated in. | |
61007b31 SH |
1548 | */ |
1549 | static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, | |
1550 | int64_t sector_num, | |
67a0fd2a FZ |
1551 | int nb_sectors, int *pnum, |
1552 | BlockDriverState **file) | |
61007b31 SH |
1553 | { |
1554 | int64_t total_sectors; | |
1555 | int64_t n; | |
1556 | int64_t ret, ret2; | |
1557 | ||
1558 | total_sectors = bdrv_nb_sectors(bs); | |
1559 | if (total_sectors < 0) { | |
1560 | return total_sectors; | |
1561 | } | |
1562 | ||
1563 | if (sector_num >= total_sectors) { | |
1564 | *pnum = 0; | |
1565 | return 0; | |
1566 | } | |
1567 | ||
1568 | n = total_sectors - sector_num; | |
1569 | if (n < nb_sectors) { | |
1570 | nb_sectors = n; | |
1571 | } | |
1572 | ||
1573 | if (!bs->drv->bdrv_co_get_block_status) { | |
1574 | *pnum = nb_sectors; | |
1575 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; | |
1576 | if (bs->drv->protocol_name) { | |
1577 | ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE); | |
1578 | } | |
1579 | return ret; | |
1580 | } | |
1581 | ||
67a0fd2a FZ |
1582 | *file = NULL; |
1583 | ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum, | |
1584 | file); | |
61007b31 SH |
1585 | if (ret < 0) { |
1586 | *pnum = 0; | |
1587 | return ret; | |
1588 | } | |
1589 | ||
1590 | if (ret & BDRV_BLOCK_RAW) { | |
1591 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
9a4f4c31 | 1592 | return bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS, |
67a0fd2a | 1593 | *pnum, pnum, file); |
61007b31 SH |
1594 | } |
1595 | ||
1596 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
1597 | ret |= BDRV_BLOCK_ALLOCATED; | |
a53f1a95 | 1598 | } else { |
61007b31 SH |
1599 | if (bdrv_unallocated_blocks_are_zero(bs)) { |
1600 | ret |= BDRV_BLOCK_ZERO; | |
760e0063 KW |
1601 | } else if (bs->backing) { |
1602 | BlockDriverState *bs2 = bs->backing->bs; | |
61007b31 SH |
1603 | int64_t nb_sectors2 = bdrv_nb_sectors(bs2); |
1604 | if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) { | |
1605 | ret |= BDRV_BLOCK_ZERO; | |
1606 | } | |
1607 | } | |
1608 | } | |
1609 | ||
ac987b30 | 1610 | if (*file && *file != bs && |
61007b31 SH |
1611 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
1612 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
67a0fd2a | 1613 | BlockDriverState *file2; |
61007b31 SH |
1614 | int file_pnum; |
1615 | ||
ac987b30 | 1616 | ret2 = bdrv_co_get_block_status(*file, ret >> BDRV_SECTOR_BITS, |
67a0fd2a | 1617 | *pnum, &file_pnum, &file2); |
61007b31 SH |
1618 | if (ret2 >= 0) { |
1619 | /* Ignore errors. This is just providing extra information, it | |
1620 | * is useful but not necessary. | |
1621 | */ | |
1622 | if (!file_pnum) { | |
1623 | /* !file_pnum indicates an offset at or beyond the EOF; it is | |
1624 | * perfectly valid for the format block driver to point to such | |
1625 | * offsets, so catch it and mark everything as zero */ | |
1626 | ret |= BDRV_BLOCK_ZERO; | |
1627 | } else { | |
1628 | /* Limit request to the range reported by the protocol driver */ | |
1629 | *pnum = file_pnum; | |
1630 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
1631 | } | |
1632 | } | |
1633 | } | |
1634 | ||
1635 | return ret; | |
1636 | } | |
1637 | ||
ba3f0e25 FZ |
1638 | static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs, |
1639 | BlockDriverState *base, | |
1640 | int64_t sector_num, | |
1641 | int nb_sectors, | |
67a0fd2a FZ |
1642 | int *pnum, |
1643 | BlockDriverState **file) | |
ba3f0e25 FZ |
1644 | { |
1645 | BlockDriverState *p; | |
1646 | int64_t ret = 0; | |
1647 | ||
1648 | assert(bs != base); | |
760e0063 | 1649 | for (p = bs; p != base; p = backing_bs(p)) { |
67a0fd2a | 1650 | ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum, file); |
ba3f0e25 FZ |
1651 | if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) { |
1652 | break; | |
1653 | } | |
1654 | /* [sector_num, pnum] unallocated on this layer, which could be only | |
1655 | * the first part of [sector_num, nb_sectors]. */ | |
1656 | nb_sectors = MIN(nb_sectors, *pnum); | |
1657 | } | |
1658 | return ret; | |
1659 | } | |
1660 | ||
1661 | /* Coroutine wrapper for bdrv_get_block_status_above() */ | |
1662 | static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque) | |
61007b31 SH |
1663 | { |
1664 | BdrvCoGetBlockStatusData *data = opaque; | |
61007b31 | 1665 | |
ba3f0e25 FZ |
1666 | data->ret = bdrv_co_get_block_status_above(data->bs, data->base, |
1667 | data->sector_num, | |
1668 | data->nb_sectors, | |
67a0fd2a FZ |
1669 | data->pnum, |
1670 | data->file); | |
61007b31 SH |
1671 | data->done = true; |
1672 | } | |
1673 | ||
1674 | /* | |
ba3f0e25 | 1675 | * Synchronous wrapper around bdrv_co_get_block_status_above(). |
61007b31 | 1676 | * |
ba3f0e25 | 1677 | * See bdrv_co_get_block_status_above() for details. |
61007b31 | 1678 | */ |
ba3f0e25 FZ |
1679 | int64_t bdrv_get_block_status_above(BlockDriverState *bs, |
1680 | BlockDriverState *base, | |
1681 | int64_t sector_num, | |
67a0fd2a FZ |
1682 | int nb_sectors, int *pnum, |
1683 | BlockDriverState **file) | |
61007b31 SH |
1684 | { |
1685 | Coroutine *co; | |
1686 | BdrvCoGetBlockStatusData data = { | |
1687 | .bs = bs, | |
ba3f0e25 | 1688 | .base = base, |
67a0fd2a | 1689 | .file = file, |
61007b31 SH |
1690 | .sector_num = sector_num, |
1691 | .nb_sectors = nb_sectors, | |
1692 | .pnum = pnum, | |
1693 | .done = false, | |
1694 | }; | |
1695 | ||
1696 | if (qemu_in_coroutine()) { | |
1697 | /* Fast-path if already in coroutine context */ | |
ba3f0e25 | 1698 | bdrv_get_block_status_above_co_entry(&data); |
61007b31 SH |
1699 | } else { |
1700 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
1701 | ||
ba3f0e25 | 1702 | co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry); |
61007b31 SH |
1703 | qemu_coroutine_enter(co, &data); |
1704 | while (!data.done) { | |
1705 | aio_poll(aio_context, true); | |
1706 | } | |
1707 | } | |
1708 | return data.ret; | |
1709 | } | |
1710 | ||
ba3f0e25 FZ |
1711 | int64_t bdrv_get_block_status(BlockDriverState *bs, |
1712 | int64_t sector_num, | |
67a0fd2a FZ |
1713 | int nb_sectors, int *pnum, |
1714 | BlockDriverState **file) | |
ba3f0e25 | 1715 | { |
760e0063 | 1716 | return bdrv_get_block_status_above(bs, backing_bs(bs), |
67a0fd2a | 1717 | sector_num, nb_sectors, pnum, file); |
ba3f0e25 FZ |
1718 | } |
1719 | ||
61007b31 SH |
1720 | int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, |
1721 | int nb_sectors, int *pnum) | |
1722 | { | |
67a0fd2a FZ |
1723 | BlockDriverState *file; |
1724 | int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum, | |
1725 | &file); | |
61007b31 SH |
1726 | if (ret < 0) { |
1727 | return ret; | |
1728 | } | |
1729 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
1730 | } | |
1731 | ||
1732 | /* | |
1733 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
1734 | * | |
1735 | * Return true if the given sector is allocated in any image between | |
1736 | * BASE and TOP (inclusive). BASE can be NULL to check if the given | |
1737 | * sector is allocated in any image of the chain. Return false otherwise. | |
1738 | * | |
1739 | * 'pnum' is set to the number of sectors (including and immediately following | |
1740 | * the specified sector) that are known to be in the same | |
1741 | * allocated/unallocated state. | |
1742 | * | |
1743 | */ | |
1744 | int bdrv_is_allocated_above(BlockDriverState *top, | |
1745 | BlockDriverState *base, | |
1746 | int64_t sector_num, | |
1747 | int nb_sectors, int *pnum) | |
1748 | { | |
1749 | BlockDriverState *intermediate; | |
1750 | int ret, n = nb_sectors; | |
1751 | ||
1752 | intermediate = top; | |
1753 | while (intermediate && intermediate != base) { | |
1754 | int pnum_inter; | |
1755 | ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors, | |
1756 | &pnum_inter); | |
1757 | if (ret < 0) { | |
1758 | return ret; | |
1759 | } else if (ret) { | |
1760 | *pnum = pnum_inter; | |
1761 | return 1; | |
1762 | } | |
1763 | ||
1764 | /* | |
1765 | * [sector_num, nb_sectors] is unallocated on top but intermediate | |
1766 | * might have | |
1767 | * | |
1768 | * [sector_num+x, nr_sectors] allocated. | |
1769 | */ | |
1770 | if (n > pnum_inter && | |
1771 | (intermediate == top || | |
1772 | sector_num + pnum_inter < intermediate->total_sectors)) { | |
1773 | n = pnum_inter; | |
1774 | } | |
1775 | ||
760e0063 | 1776 | intermediate = backing_bs(intermediate); |
61007b31 SH |
1777 | } |
1778 | ||
1779 | *pnum = n; | |
1780 | return 0; | |
1781 | } | |
1782 | ||
1783 | int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num, | |
1784 | const uint8_t *buf, int nb_sectors) | |
1785 | { | |
1786 | BlockDriver *drv = bs->drv; | |
1787 | int ret; | |
1788 | ||
1789 | if (!drv) { | |
1790 | return -ENOMEDIUM; | |
1791 | } | |
1792 | if (!drv->bdrv_write_compressed) { | |
1793 | return -ENOTSUP; | |
1794 | } | |
1795 | ret = bdrv_check_request(bs, sector_num, nb_sectors); | |
1796 | if (ret < 0) { | |
1797 | return ret; | |
1798 | } | |
1799 | ||
1800 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
1801 | ||
1802 | return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors); | |
1803 | } | |
1804 | ||
1805 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, | |
1806 | int64_t pos, int size) | |
1807 | { | |
1808 | QEMUIOVector qiov; | |
1809 | struct iovec iov = { | |
1810 | .iov_base = (void *) buf, | |
1811 | .iov_len = size, | |
1812 | }; | |
1813 | ||
1814 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1815 | return bdrv_writev_vmstate(bs, &qiov, pos); | |
1816 | } | |
1817 | ||
1818 | int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
1819 | { | |
1820 | BlockDriver *drv = bs->drv; | |
1821 | ||
1822 | if (!drv) { | |
1823 | return -ENOMEDIUM; | |
1824 | } else if (drv->bdrv_save_vmstate) { | |
1825 | return drv->bdrv_save_vmstate(bs, qiov, pos); | |
1826 | } else if (bs->file) { | |
9a4f4c31 | 1827 | return bdrv_writev_vmstate(bs->file->bs, qiov, pos); |
61007b31 SH |
1828 | } |
1829 | ||
1830 | return -ENOTSUP; | |
1831 | } | |
1832 | ||
1833 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
1834 | int64_t pos, int size) | |
1835 | { | |
1836 | BlockDriver *drv = bs->drv; | |
1837 | if (!drv) | |
1838 | return -ENOMEDIUM; | |
1839 | if (drv->bdrv_load_vmstate) | |
1840 | return drv->bdrv_load_vmstate(bs, buf, pos, size); | |
1841 | if (bs->file) | |
9a4f4c31 | 1842 | return bdrv_load_vmstate(bs->file->bs, buf, pos, size); |
61007b31 SH |
1843 | return -ENOTSUP; |
1844 | } | |
1845 | ||
1846 | /**************************************************************/ | |
1847 | /* async I/Os */ | |
1848 | ||
1849 | BlockAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num, | |
1850 | QEMUIOVector *qiov, int nb_sectors, | |
1851 | BlockCompletionFunc *cb, void *opaque) | |
1852 | { | |
1853 | trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque); | |
1854 | ||
1855 | return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0, | |
1856 | cb, opaque, false); | |
1857 | } | |
1858 | ||
1859 | BlockAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num, | |
1860 | QEMUIOVector *qiov, int nb_sectors, | |
1861 | BlockCompletionFunc *cb, void *opaque) | |
1862 | { | |
1863 | trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque); | |
1864 | ||
1865 | return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0, | |
1866 | cb, opaque, true); | |
1867 | } | |
1868 | ||
1869 | BlockAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs, | |
1870 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags, | |
1871 | BlockCompletionFunc *cb, void *opaque) | |
1872 | { | |
1873 | trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque); | |
1874 | ||
1875 | return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors, | |
1876 | BDRV_REQ_ZERO_WRITE | flags, | |
1877 | cb, opaque, true); | |
1878 | } | |
1879 | ||
61007b31 SH |
1880 | void bdrv_aio_cancel(BlockAIOCB *acb) |
1881 | { | |
1882 | qemu_aio_ref(acb); | |
1883 | bdrv_aio_cancel_async(acb); | |
1884 | while (acb->refcnt > 1) { | |
1885 | if (acb->aiocb_info->get_aio_context) { | |
1886 | aio_poll(acb->aiocb_info->get_aio_context(acb), true); | |
1887 | } else if (acb->bs) { | |
1888 | aio_poll(bdrv_get_aio_context(acb->bs), true); | |
1889 | } else { | |
1890 | abort(); | |
1891 | } | |
1892 | } | |
1893 | qemu_aio_unref(acb); | |
1894 | } | |
1895 | ||
1896 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
1897 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
1898 | * In either case the completion callback must be called. */ | |
1899 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
1900 | { | |
1901 | if (acb->aiocb_info->cancel_async) { | |
1902 | acb->aiocb_info->cancel_async(acb); | |
1903 | } | |
1904 | } | |
1905 | ||
1906 | /**************************************************************/ | |
1907 | /* async block device emulation */ | |
1908 | ||
61007b31 SH |
1909 | typedef struct BlockAIOCBCoroutine { |
1910 | BlockAIOCB common; | |
1911 | BlockRequest req; | |
1912 | bool is_write; | |
1913 | bool need_bh; | |
1914 | bool *done; | |
1915 | QEMUBH* bh; | |
1916 | } BlockAIOCBCoroutine; | |
1917 | ||
1918 | static const AIOCBInfo bdrv_em_co_aiocb_info = { | |
1919 | .aiocb_size = sizeof(BlockAIOCBCoroutine), | |
1920 | }; | |
1921 | ||
1922 | static void bdrv_co_complete(BlockAIOCBCoroutine *acb) | |
1923 | { | |
1924 | if (!acb->need_bh) { | |
1925 | acb->common.cb(acb->common.opaque, acb->req.error); | |
1926 | qemu_aio_unref(acb); | |
1927 | } | |
1928 | } | |
1929 | ||
1930 | static void bdrv_co_em_bh(void *opaque) | |
1931 | { | |
1932 | BlockAIOCBCoroutine *acb = opaque; | |
1933 | ||
1934 | assert(!acb->need_bh); | |
1935 | qemu_bh_delete(acb->bh); | |
1936 | bdrv_co_complete(acb); | |
1937 | } | |
1938 | ||
1939 | static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb) | |
1940 | { | |
1941 | acb->need_bh = false; | |
1942 | if (acb->req.error != -EINPROGRESS) { | |
1943 | BlockDriverState *bs = acb->common.bs; | |
1944 | ||
1945 | acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb); | |
1946 | qemu_bh_schedule(acb->bh); | |
1947 | } | |
1948 | } | |
1949 | ||
1950 | /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */ | |
1951 | static void coroutine_fn bdrv_co_do_rw(void *opaque) | |
1952 | { | |
1953 | BlockAIOCBCoroutine *acb = opaque; | |
1954 | BlockDriverState *bs = acb->common.bs; | |
1955 | ||
1956 | if (!acb->is_write) { | |
1957 | acb->req.error = bdrv_co_do_readv(bs, acb->req.sector, | |
1958 | acb->req.nb_sectors, acb->req.qiov, acb->req.flags); | |
1959 | } else { | |
1960 | acb->req.error = bdrv_co_do_writev(bs, acb->req.sector, | |
1961 | acb->req.nb_sectors, acb->req.qiov, acb->req.flags); | |
1962 | } | |
1963 | ||
1964 | bdrv_co_complete(acb); | |
1965 | } | |
1966 | ||
1967 | static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, | |
1968 | int64_t sector_num, | |
1969 | QEMUIOVector *qiov, | |
1970 | int nb_sectors, | |
1971 | BdrvRequestFlags flags, | |
1972 | BlockCompletionFunc *cb, | |
1973 | void *opaque, | |
1974 | bool is_write) | |
1975 | { | |
1976 | Coroutine *co; | |
1977 | BlockAIOCBCoroutine *acb; | |
1978 | ||
1979 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
1980 | acb->need_bh = true; | |
1981 | acb->req.error = -EINPROGRESS; | |
1982 | acb->req.sector = sector_num; | |
1983 | acb->req.nb_sectors = nb_sectors; | |
1984 | acb->req.qiov = qiov; | |
1985 | acb->req.flags = flags; | |
1986 | acb->is_write = is_write; | |
1987 | ||
1988 | co = qemu_coroutine_create(bdrv_co_do_rw); | |
1989 | qemu_coroutine_enter(co, acb); | |
1990 | ||
1991 | bdrv_co_maybe_schedule_bh(acb); | |
1992 | return &acb->common; | |
1993 | } | |
1994 | ||
1995 | static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) | |
1996 | { | |
1997 | BlockAIOCBCoroutine *acb = opaque; | |
1998 | BlockDriverState *bs = acb->common.bs; | |
1999 | ||
2000 | acb->req.error = bdrv_co_flush(bs); | |
2001 | bdrv_co_complete(acb); | |
2002 | } | |
2003 | ||
2004 | BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs, | |
2005 | BlockCompletionFunc *cb, void *opaque) | |
2006 | { | |
2007 | trace_bdrv_aio_flush(bs, opaque); | |
2008 | ||
2009 | Coroutine *co; | |
2010 | BlockAIOCBCoroutine *acb; | |
2011 | ||
2012 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
2013 | acb->need_bh = true; | |
2014 | acb->req.error = -EINPROGRESS; | |
2015 | ||
2016 | co = qemu_coroutine_create(bdrv_aio_flush_co_entry); | |
2017 | qemu_coroutine_enter(co, acb); | |
2018 | ||
2019 | bdrv_co_maybe_schedule_bh(acb); | |
2020 | return &acb->common; | |
2021 | } | |
2022 | ||
2023 | static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque) | |
2024 | { | |
2025 | BlockAIOCBCoroutine *acb = opaque; | |
2026 | BlockDriverState *bs = acb->common.bs; | |
2027 | ||
2028 | acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors); | |
2029 | bdrv_co_complete(acb); | |
2030 | } | |
2031 | ||
2032 | BlockAIOCB *bdrv_aio_discard(BlockDriverState *bs, | |
2033 | int64_t sector_num, int nb_sectors, | |
2034 | BlockCompletionFunc *cb, void *opaque) | |
2035 | { | |
2036 | Coroutine *co; | |
2037 | BlockAIOCBCoroutine *acb; | |
2038 | ||
2039 | trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque); | |
2040 | ||
2041 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
2042 | acb->need_bh = true; | |
2043 | acb->req.error = -EINPROGRESS; | |
2044 | acb->req.sector = sector_num; | |
2045 | acb->req.nb_sectors = nb_sectors; | |
2046 | co = qemu_coroutine_create(bdrv_aio_discard_co_entry); | |
2047 | qemu_coroutine_enter(co, acb); | |
2048 | ||
2049 | bdrv_co_maybe_schedule_bh(acb); | |
2050 | return &acb->common; | |
2051 | } | |
2052 | ||
2053 | void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, | |
2054 | BlockCompletionFunc *cb, void *opaque) | |
2055 | { | |
2056 | BlockAIOCB *acb; | |
2057 | ||
c84b3192 | 2058 | acb = g_malloc(aiocb_info->aiocb_size); |
61007b31 SH |
2059 | acb->aiocb_info = aiocb_info; |
2060 | acb->bs = bs; | |
2061 | acb->cb = cb; | |
2062 | acb->opaque = opaque; | |
2063 | acb->refcnt = 1; | |
2064 | return acb; | |
2065 | } | |
2066 | ||
2067 | void qemu_aio_ref(void *p) | |
2068 | { | |
2069 | BlockAIOCB *acb = p; | |
2070 | acb->refcnt++; | |
2071 | } | |
2072 | ||
2073 | void qemu_aio_unref(void *p) | |
2074 | { | |
2075 | BlockAIOCB *acb = p; | |
2076 | assert(acb->refcnt > 0); | |
2077 | if (--acb->refcnt == 0) { | |
c84b3192 | 2078 | g_free(acb); |
61007b31 SH |
2079 | } |
2080 | } | |
2081 | ||
2082 | /**************************************************************/ | |
2083 | /* Coroutine block device emulation */ | |
2084 | ||
61007b31 SH |
2085 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) |
2086 | { | |
2087 | RwCo *rwco = opaque; | |
2088 | ||
2089 | rwco->ret = bdrv_co_flush(rwco->bs); | |
2090 | } | |
2091 | ||
2092 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
2093 | { | |
2094 | int ret; | |
cdb5e315 | 2095 | BdrvTrackedRequest req; |
61007b31 | 2096 | |
1b6bc94d DA |
2097 | if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) || |
2098 | bdrv_is_sg(bs)) { | |
61007b31 SH |
2099 | return 0; |
2100 | } | |
2101 | ||
cdb5e315 | 2102 | tracked_request_begin(&req, bs, 0, 0, BDRV_TRACKED_FLUSH); |
c32b82af PD |
2103 | |
2104 | /* Write back all layers by calling one driver function */ | |
2105 | if (bs->drv->bdrv_co_flush) { | |
2106 | ret = bs->drv->bdrv_co_flush(bs); | |
2107 | goto out; | |
2108 | } | |
2109 | ||
61007b31 SH |
2110 | /* Write back cached data to the OS even with cache=unsafe */ |
2111 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); | |
2112 | if (bs->drv->bdrv_co_flush_to_os) { | |
2113 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
2114 | if (ret < 0) { | |
cdb5e315 | 2115 | goto out; |
61007b31 SH |
2116 | } |
2117 | } | |
2118 | ||
2119 | /* But don't actually force it to the disk with cache=unsafe */ | |
2120 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
2121 | goto flush_parent; | |
2122 | } | |
2123 | ||
2124 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); | |
2125 | if (bs->drv->bdrv_co_flush_to_disk) { | |
2126 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
2127 | } else if (bs->drv->bdrv_aio_flush) { | |
2128 | BlockAIOCB *acb; | |
2129 | CoroutineIOCompletion co = { | |
2130 | .coroutine = qemu_coroutine_self(), | |
2131 | }; | |
2132 | ||
2133 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
2134 | if (acb == NULL) { | |
2135 | ret = -EIO; | |
2136 | } else { | |
2137 | qemu_coroutine_yield(); | |
2138 | ret = co.ret; | |
2139 | } | |
2140 | } else { | |
2141 | /* | |
2142 | * Some block drivers always operate in either writethrough or unsafe | |
2143 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
2144 | * know how the server works (because the behaviour is hardcoded or | |
2145 | * depends on server-side configuration), so we can't ensure that | |
2146 | * everything is safe on disk. Returning an error doesn't work because | |
2147 | * that would break guests even if the server operates in writethrough | |
2148 | * mode. | |
2149 | * | |
2150 | * Let's hope the user knows what he's doing. | |
2151 | */ | |
2152 | ret = 0; | |
2153 | } | |
2154 | if (ret < 0) { | |
cdb5e315 | 2155 | goto out; |
61007b31 SH |
2156 | } |
2157 | ||
2158 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
2159 | * in the case of cache=unsafe, so there are no useless flushes. | |
2160 | */ | |
2161 | flush_parent: | |
cdb5e315 FZ |
2162 | ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0; |
2163 | out: | |
2164 | tracked_request_end(&req); | |
2165 | return ret; | |
61007b31 SH |
2166 | } |
2167 | ||
2168 | int bdrv_flush(BlockDriverState *bs) | |
2169 | { | |
2170 | Coroutine *co; | |
2171 | RwCo rwco = { | |
2172 | .bs = bs, | |
2173 | .ret = NOT_DONE, | |
2174 | }; | |
2175 | ||
2176 | if (qemu_in_coroutine()) { | |
2177 | /* Fast-path if already in coroutine context */ | |
2178 | bdrv_flush_co_entry(&rwco); | |
2179 | } else { | |
2180 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2181 | ||
2182 | co = qemu_coroutine_create(bdrv_flush_co_entry); | |
2183 | qemu_coroutine_enter(co, &rwco); | |
2184 | while (rwco.ret == NOT_DONE) { | |
2185 | aio_poll(aio_context, true); | |
2186 | } | |
2187 | } | |
2188 | ||
2189 | return rwco.ret; | |
2190 | } | |
2191 | ||
2192 | typedef struct DiscardCo { | |
2193 | BlockDriverState *bs; | |
2194 | int64_t sector_num; | |
2195 | int nb_sectors; | |
2196 | int ret; | |
2197 | } DiscardCo; | |
2198 | static void coroutine_fn bdrv_discard_co_entry(void *opaque) | |
2199 | { | |
2200 | DiscardCo *rwco = opaque; | |
2201 | ||
2202 | rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors); | |
2203 | } | |
2204 | ||
2205 | int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num, | |
2206 | int nb_sectors) | |
2207 | { | |
b1066c87 | 2208 | BdrvTrackedRequest req; |
61007b31 SH |
2209 | int max_discard, ret; |
2210 | ||
2211 | if (!bs->drv) { | |
2212 | return -ENOMEDIUM; | |
2213 | } | |
2214 | ||
2215 | ret = bdrv_check_request(bs, sector_num, nb_sectors); | |
2216 | if (ret < 0) { | |
2217 | return ret; | |
2218 | } else if (bs->read_only) { | |
eaf5fe2d | 2219 | return -EPERM; |
61007b31 | 2220 | } |
04c01a5c | 2221 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
61007b31 | 2222 | |
61007b31 SH |
2223 | /* Do nothing if disabled. */ |
2224 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
2225 | return 0; | |
2226 | } | |
2227 | ||
2228 | if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) { | |
2229 | return 0; | |
2230 | } | |
2231 | ||
b1066c87 FZ |
2232 | tracked_request_begin(&req, bs, sector_num, nb_sectors, |
2233 | BDRV_TRACKED_DISCARD); | |
50824995 FZ |
2234 | bdrv_set_dirty(bs, sector_num, nb_sectors); |
2235 | ||
61007b31 SH |
2236 | max_discard = MIN_NON_ZERO(bs->bl.max_discard, BDRV_REQUEST_MAX_SECTORS); |
2237 | while (nb_sectors > 0) { | |
2238 | int ret; | |
2239 | int num = nb_sectors; | |
2240 | ||
2241 | /* align request */ | |
2242 | if (bs->bl.discard_alignment && | |
2243 | num >= bs->bl.discard_alignment && | |
2244 | sector_num % bs->bl.discard_alignment) { | |
2245 | if (num > bs->bl.discard_alignment) { | |
2246 | num = bs->bl.discard_alignment; | |
2247 | } | |
2248 | num -= sector_num % bs->bl.discard_alignment; | |
2249 | } | |
2250 | ||
2251 | /* limit request size */ | |
2252 | if (num > max_discard) { | |
2253 | num = max_discard; | |
2254 | } | |
2255 | ||
2256 | if (bs->drv->bdrv_co_discard) { | |
2257 | ret = bs->drv->bdrv_co_discard(bs, sector_num, num); | |
2258 | } else { | |
2259 | BlockAIOCB *acb; | |
2260 | CoroutineIOCompletion co = { | |
2261 | .coroutine = qemu_coroutine_self(), | |
2262 | }; | |
2263 | ||
2264 | acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors, | |
2265 | bdrv_co_io_em_complete, &co); | |
2266 | if (acb == NULL) { | |
b1066c87 FZ |
2267 | ret = -EIO; |
2268 | goto out; | |
61007b31 SH |
2269 | } else { |
2270 | qemu_coroutine_yield(); | |
2271 | ret = co.ret; | |
2272 | } | |
2273 | } | |
2274 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 2275 | goto out; |
61007b31 SH |
2276 | } |
2277 | ||
2278 | sector_num += num; | |
2279 | nb_sectors -= num; | |
2280 | } | |
b1066c87 FZ |
2281 | ret = 0; |
2282 | out: | |
2283 | tracked_request_end(&req); | |
2284 | return ret; | |
61007b31 SH |
2285 | } |
2286 | ||
2287 | int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors) | |
2288 | { | |
2289 | Coroutine *co; | |
2290 | DiscardCo rwco = { | |
2291 | .bs = bs, | |
2292 | .sector_num = sector_num, | |
2293 | .nb_sectors = nb_sectors, | |
2294 | .ret = NOT_DONE, | |
2295 | }; | |
2296 | ||
2297 | if (qemu_in_coroutine()) { | |
2298 | /* Fast-path if already in coroutine context */ | |
2299 | bdrv_discard_co_entry(&rwco); | |
2300 | } else { | |
2301 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2302 | ||
2303 | co = qemu_coroutine_create(bdrv_discard_co_entry); | |
2304 | qemu_coroutine_enter(co, &rwco); | |
2305 | while (rwco.ret == NOT_DONE) { | |
2306 | aio_poll(aio_context, true); | |
2307 | } | |
2308 | } | |
2309 | ||
2310 | return rwco.ret; | |
2311 | } | |
2312 | ||
5c5ae76a FZ |
2313 | typedef struct { |
2314 | CoroutineIOCompletion *co; | |
2315 | QEMUBH *bh; | |
2316 | } BdrvIoctlCompletionData; | |
61007b31 | 2317 | |
5c5ae76a FZ |
2318 | static void bdrv_ioctl_bh_cb(void *opaque) |
2319 | { | |
2320 | BdrvIoctlCompletionData *data = opaque; | |
2321 | ||
2322 | bdrv_co_io_em_complete(data->co, -ENOTSUP); | |
2323 | qemu_bh_delete(data->bh); | |
2324 | } | |
2325 | ||
2326 | static int bdrv_co_do_ioctl(BlockDriverState *bs, int req, void *buf) | |
61007b31 SH |
2327 | { |
2328 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
2329 | BdrvTrackedRequest tracked_req; |
2330 | CoroutineIOCompletion co = { | |
2331 | .coroutine = qemu_coroutine_self(), | |
2332 | }; | |
2333 | BlockAIOCB *acb; | |
61007b31 | 2334 | |
5c5ae76a FZ |
2335 | tracked_request_begin(&tracked_req, bs, 0, 0, BDRV_TRACKED_IOCTL); |
2336 | if (!drv || !drv->bdrv_aio_ioctl) { | |
2337 | co.ret = -ENOTSUP; | |
2338 | goto out; | |
2339 | } | |
2340 | ||
2341 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
2342 | if (!acb) { | |
2343 | BdrvIoctlCompletionData *data = g_new(BdrvIoctlCompletionData, 1); | |
2344 | data->bh = aio_bh_new(bdrv_get_aio_context(bs), | |
2345 | bdrv_ioctl_bh_cb, data); | |
2346 | data->co = &co; | |
2347 | qemu_bh_schedule(data->bh); | |
2348 | } | |
2349 | qemu_coroutine_yield(); | |
2350 | out: | |
2351 | tracked_request_end(&tracked_req); | |
2352 | return co.ret; | |
2353 | } | |
2354 | ||
2355 | typedef struct { | |
2356 | BlockDriverState *bs; | |
2357 | int req; | |
2358 | void *buf; | |
2359 | int ret; | |
2360 | } BdrvIoctlCoData; | |
2361 | ||
2362 | static void coroutine_fn bdrv_co_ioctl_entry(void *opaque) | |
2363 | { | |
2364 | BdrvIoctlCoData *data = opaque; | |
2365 | data->ret = bdrv_co_do_ioctl(data->bs, data->req, data->buf); | |
2366 | } | |
2367 | ||
2368 | /* needed for generic scsi interface */ | |
2369 | int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
2370 | { | |
2371 | BdrvIoctlCoData data = { | |
2372 | .bs = bs, | |
2373 | .req = req, | |
2374 | .buf = buf, | |
2375 | .ret = -EINPROGRESS, | |
2376 | }; | |
2377 | ||
2378 | if (qemu_in_coroutine()) { | |
2379 | /* Fast-path if already in coroutine context */ | |
2380 | bdrv_co_ioctl_entry(&data); | |
2381 | } else { | |
2382 | Coroutine *co = qemu_coroutine_create(bdrv_co_ioctl_entry); | |
ba889444 | 2383 | |
5c5ae76a | 2384 | qemu_coroutine_enter(co, &data); |
ba889444 PB |
2385 | while (data.ret == -EINPROGRESS) { |
2386 | aio_poll(bdrv_get_aio_context(bs), true); | |
2387 | } | |
5c5ae76a FZ |
2388 | } |
2389 | return data.ret; | |
2390 | } | |
2391 | ||
2392 | static void coroutine_fn bdrv_co_aio_ioctl_entry(void *opaque) | |
2393 | { | |
2394 | BlockAIOCBCoroutine *acb = opaque; | |
2395 | acb->req.error = bdrv_co_do_ioctl(acb->common.bs, | |
2396 | acb->req.req, acb->req.buf); | |
2397 | bdrv_co_complete(acb); | |
61007b31 SH |
2398 | } |
2399 | ||
2400 | BlockAIOCB *bdrv_aio_ioctl(BlockDriverState *bs, | |
2401 | unsigned long int req, void *buf, | |
2402 | BlockCompletionFunc *cb, void *opaque) | |
2403 | { | |
5c5ae76a FZ |
2404 | BlockAIOCBCoroutine *acb = qemu_aio_get(&bdrv_em_co_aiocb_info, |
2405 | bs, cb, opaque); | |
2406 | Coroutine *co; | |
61007b31 | 2407 | |
5c5ae76a FZ |
2408 | acb->need_bh = true; |
2409 | acb->req.error = -EINPROGRESS; | |
2410 | acb->req.req = req; | |
2411 | acb->req.buf = buf; | |
2412 | co = qemu_coroutine_create(bdrv_co_aio_ioctl_entry); | |
2413 | qemu_coroutine_enter(co, acb); | |
2414 | ||
2415 | bdrv_co_maybe_schedule_bh(acb); | |
2416 | return &acb->common; | |
61007b31 SH |
2417 | } |
2418 | ||
2419 | void *qemu_blockalign(BlockDriverState *bs, size_t size) | |
2420 | { | |
2421 | return qemu_memalign(bdrv_opt_mem_align(bs), size); | |
2422 | } | |
2423 | ||
2424 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
2425 | { | |
2426 | return memset(qemu_blockalign(bs, size), 0, size); | |
2427 | } | |
2428 | ||
2429 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
2430 | { | |
2431 | size_t align = bdrv_opt_mem_align(bs); | |
2432 | ||
2433 | /* Ensure that NULL is never returned on success */ | |
2434 | assert(align > 0); | |
2435 | if (size == 0) { | |
2436 | size = align; | |
2437 | } | |
2438 | ||
2439 | return qemu_try_memalign(align, size); | |
2440 | } | |
2441 | ||
2442 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
2443 | { | |
2444 | void *mem = qemu_try_blockalign(bs, size); | |
2445 | ||
2446 | if (mem) { | |
2447 | memset(mem, 0, size); | |
2448 | } | |
2449 | ||
2450 | return mem; | |
2451 | } | |
2452 | ||
2453 | /* | |
2454 | * Check if all memory in this vector is sector aligned. | |
2455 | */ | |
2456 | bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
2457 | { | |
2458 | int i; | |
4196d2f0 | 2459 | size_t alignment = bdrv_min_mem_align(bs); |
61007b31 SH |
2460 | |
2461 | for (i = 0; i < qiov->niov; i++) { | |
2462 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
2463 | return false; | |
2464 | } | |
2465 | if (qiov->iov[i].iov_len % alignment) { | |
2466 | return false; | |
2467 | } | |
2468 | } | |
2469 | ||
2470 | return true; | |
2471 | } | |
2472 | ||
2473 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
2474 | NotifierWithReturn *notifier) | |
2475 | { | |
2476 | notifier_with_return_list_add(&bs->before_write_notifiers, notifier); | |
2477 | } | |
2478 | ||
2479 | void bdrv_io_plug(BlockDriverState *bs) | |
2480 | { | |
6b98bd64 PB |
2481 | BdrvChild *child; |
2482 | ||
2483 | QLIST_FOREACH(child, &bs->children, next) { | |
2484 | bdrv_io_plug(child->bs); | |
2485 | } | |
2486 | ||
2487 | if (bs->io_plugged++ == 0 && bs->io_plug_disabled == 0) { | |
2488 | BlockDriver *drv = bs->drv; | |
2489 | if (drv && drv->bdrv_io_plug) { | |
2490 | drv->bdrv_io_plug(bs); | |
2491 | } | |
61007b31 SH |
2492 | } |
2493 | } | |
2494 | ||
2495 | void bdrv_io_unplug(BlockDriverState *bs) | |
2496 | { | |
6b98bd64 PB |
2497 | BdrvChild *child; |
2498 | ||
2499 | assert(bs->io_plugged); | |
2500 | if (--bs->io_plugged == 0 && bs->io_plug_disabled == 0) { | |
2501 | BlockDriver *drv = bs->drv; | |
2502 | if (drv && drv->bdrv_io_unplug) { | |
2503 | drv->bdrv_io_unplug(bs); | |
2504 | } | |
2505 | } | |
2506 | ||
2507 | QLIST_FOREACH(child, &bs->children, next) { | |
2508 | bdrv_io_unplug(child->bs); | |
61007b31 SH |
2509 | } |
2510 | } | |
2511 | ||
6b98bd64 | 2512 | void bdrv_io_unplugged_begin(BlockDriverState *bs) |
61007b31 | 2513 | { |
6b98bd64 PB |
2514 | BdrvChild *child; |
2515 | ||
2516 | if (bs->io_plug_disabled++ == 0 && bs->io_plugged > 0) { | |
2517 | BlockDriver *drv = bs->drv; | |
2518 | if (drv && drv->bdrv_io_unplug) { | |
2519 | drv->bdrv_io_unplug(bs); | |
2520 | } | |
2521 | } | |
2522 | ||
2523 | QLIST_FOREACH(child, &bs->children, next) { | |
2524 | bdrv_io_unplugged_begin(child->bs); | |
2525 | } | |
2526 | } | |
2527 | ||
2528 | void bdrv_io_unplugged_end(BlockDriverState *bs) | |
2529 | { | |
2530 | BdrvChild *child; | |
2531 | ||
2532 | assert(bs->io_plug_disabled); | |
2533 | QLIST_FOREACH(child, &bs->children, next) { | |
2534 | bdrv_io_unplugged_end(child->bs); | |
2535 | } | |
2536 | ||
2537 | if (--bs->io_plug_disabled == 0 && bs->io_plugged > 0) { | |
2538 | BlockDriver *drv = bs->drv; | |
2539 | if (drv && drv->bdrv_io_plug) { | |
2540 | drv->bdrv_io_plug(bs); | |
2541 | } | |
61007b31 SH |
2542 | } |
2543 | } | |
51288d79 FZ |
2544 | |
2545 | void bdrv_drained_begin(BlockDriverState *bs) | |
2546 | { | |
2547 | if (!bs->quiesce_counter++) { | |
2548 | aio_disable_external(bdrv_get_aio_context(bs)); | |
2549 | } | |
bb9aaeca | 2550 | bdrv_parent_drained_begin(bs); |
51288d79 FZ |
2551 | bdrv_drain(bs); |
2552 | } | |
2553 | ||
2554 | void bdrv_drained_end(BlockDriverState *bs) | |
2555 | { | |
bb9aaeca KW |
2556 | bdrv_parent_drained_end(bs); |
2557 | ||
51288d79 FZ |
2558 | assert(bs->quiesce_counter > 0); |
2559 | if (--bs->quiesce_counter > 0) { | |
2560 | return; | |
2561 | } | |
2562 | aio_enable_external(bdrv_get_aio_context(bs)); | |
2563 | } |