]>
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 | ||
25 | #include "trace.h" | |
7f0e9da6 | 26 | #include "sysemu/block-backend.h" |
61007b31 SH |
27 | #include "block/blockjob.h" |
28 | #include "block/block_int.h" | |
76f4afb4 | 29 | #include "block/throttle-groups.h" |
d49b6836 | 30 | #include "qemu/error-report.h" |
61007b31 SH |
31 | |
32 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
33 | ||
34 | static BlockAIOCB *bdrv_aio_readv_em(BlockDriverState *bs, | |
35 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
36 | BlockCompletionFunc *cb, void *opaque); | |
37 | static BlockAIOCB *bdrv_aio_writev_em(BlockDriverState *bs, | |
38 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
39 | BlockCompletionFunc *cb, void *opaque); | |
40 | static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, | |
41 | int64_t sector_num, int nb_sectors, | |
42 | QEMUIOVector *iov); | |
43 | static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, | |
44 | int64_t sector_num, int nb_sectors, | |
45 | QEMUIOVector *iov); | |
46 | static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs, | |
47 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, | |
48 | BdrvRequestFlags flags); | |
49 | static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs, | |
50 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, | |
51 | BdrvRequestFlags flags); | |
52 | static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, | |
53 | int64_t sector_num, | |
54 | QEMUIOVector *qiov, | |
55 | int nb_sectors, | |
56 | BdrvRequestFlags flags, | |
57 | BlockCompletionFunc *cb, | |
58 | void *opaque, | |
59 | bool is_write); | |
60 | static void coroutine_fn bdrv_co_do_rw(void *opaque); | |
61 | static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, | |
62 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags); | |
63 | ||
64 | /* throttling disk I/O limits */ | |
65 | void bdrv_set_io_limits(BlockDriverState *bs, | |
66 | ThrottleConfig *cfg) | |
67 | { | |
68 | int i; | |
69 | ||
76f4afb4 | 70 | throttle_group_config(bs, cfg); |
61007b31 SH |
71 | |
72 | for (i = 0; i < 2; i++) { | |
73 | qemu_co_enter_next(&bs->throttled_reqs[i]); | |
74 | } | |
75 | } | |
76 | ||
77 | /* this function drain all the throttled IOs */ | |
78 | static bool bdrv_start_throttled_reqs(BlockDriverState *bs) | |
79 | { | |
80 | bool drained = false; | |
81 | bool enabled = bs->io_limits_enabled; | |
82 | int i; | |
83 | ||
84 | bs->io_limits_enabled = false; | |
85 | ||
86 | for (i = 0; i < 2; i++) { | |
87 | while (qemu_co_enter_next(&bs->throttled_reqs[i])) { | |
88 | drained = true; | |
89 | } | |
90 | } | |
91 | ||
92 | bs->io_limits_enabled = enabled; | |
93 | ||
94 | return drained; | |
95 | } | |
96 | ||
97 | void bdrv_io_limits_disable(BlockDriverState *bs) | |
98 | { | |
99 | bs->io_limits_enabled = false; | |
61007b31 | 100 | bdrv_start_throttled_reqs(bs); |
76f4afb4 | 101 | throttle_group_unregister_bs(bs); |
61007b31 SH |
102 | } |
103 | ||
104 | /* should be called before bdrv_set_io_limits if a limit is set */ | |
76f4afb4 | 105 | void bdrv_io_limits_enable(BlockDriverState *bs, const char *group) |
61007b31 | 106 | { |
61007b31 | 107 | assert(!bs->io_limits_enabled); |
76f4afb4 | 108 | throttle_group_register_bs(bs, group); |
61007b31 SH |
109 | bs->io_limits_enabled = true; |
110 | } | |
111 | ||
76f4afb4 | 112 | void bdrv_io_limits_update_group(BlockDriverState *bs, const char *group) |
61007b31 | 113 | { |
76f4afb4 AG |
114 | /* this bs is not part of any group */ |
115 | if (!bs->throttle_state) { | |
116 | return; | |
61007b31 SH |
117 | } |
118 | ||
76f4afb4 AG |
119 | /* this bs is a part of the same group than the one we want */ |
120 | if (!g_strcmp0(throttle_group_get_name(bs), group)) { | |
61007b31 SH |
121 | return; |
122 | } | |
123 | ||
76f4afb4 AG |
124 | /* need to change the group this bs belong to */ |
125 | bdrv_io_limits_disable(bs); | |
126 | bdrv_io_limits_enable(bs, group); | |
61007b31 SH |
127 | } |
128 | ||
129 | void bdrv_setup_io_funcs(BlockDriver *bdrv) | |
130 | { | |
131 | /* Block drivers without coroutine functions need emulation */ | |
132 | if (!bdrv->bdrv_co_readv) { | |
133 | bdrv->bdrv_co_readv = bdrv_co_readv_em; | |
134 | bdrv->bdrv_co_writev = bdrv_co_writev_em; | |
135 | ||
136 | /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if | |
137 | * the block driver lacks aio we need to emulate that too. | |
138 | */ | |
139 | if (!bdrv->bdrv_aio_readv) { | |
140 | /* add AIO emulation layer */ | |
141 | bdrv->bdrv_aio_readv = bdrv_aio_readv_em; | |
142 | bdrv->bdrv_aio_writev = bdrv_aio_writev_em; | |
143 | } | |
144 | } | |
145 | } | |
146 | ||
147 | void bdrv_refresh_limits(BlockDriverState *bs, Error **errp) | |
148 | { | |
149 | BlockDriver *drv = bs->drv; | |
150 | Error *local_err = NULL; | |
151 | ||
152 | memset(&bs->bl, 0, sizeof(bs->bl)); | |
153 | ||
154 | if (!drv) { | |
155 | return; | |
156 | } | |
157 | ||
158 | /* Take some limits from the children as a default */ | |
159 | if (bs->file) { | |
9a4f4c31 | 160 | bdrv_refresh_limits(bs->file->bs, &local_err); |
61007b31 SH |
161 | if (local_err) { |
162 | error_propagate(errp, local_err); | |
163 | return; | |
164 | } | |
9a4f4c31 KW |
165 | bs->bl.opt_transfer_length = bs->file->bs->bl.opt_transfer_length; |
166 | bs->bl.max_transfer_length = bs->file->bs->bl.max_transfer_length; | |
167 | bs->bl.min_mem_alignment = bs->file->bs->bl.min_mem_alignment; | |
168 | bs->bl.opt_mem_alignment = bs->file->bs->bl.opt_mem_alignment; | |
61007b31 | 169 | } else { |
4196d2f0 | 170 | bs->bl.min_mem_alignment = 512; |
459b4e66 | 171 | bs->bl.opt_mem_alignment = getpagesize(); |
61007b31 SH |
172 | } |
173 | ||
760e0063 KW |
174 | if (bs->backing) { |
175 | bdrv_refresh_limits(bs->backing->bs, &local_err); | |
61007b31 SH |
176 | if (local_err) { |
177 | error_propagate(errp, local_err); | |
178 | return; | |
179 | } | |
180 | bs->bl.opt_transfer_length = | |
181 | MAX(bs->bl.opt_transfer_length, | |
760e0063 | 182 | bs->backing->bs->bl.opt_transfer_length); |
61007b31 SH |
183 | bs->bl.max_transfer_length = |
184 | MIN_NON_ZERO(bs->bl.max_transfer_length, | |
760e0063 | 185 | bs->backing->bs->bl.max_transfer_length); |
61007b31 SH |
186 | bs->bl.opt_mem_alignment = |
187 | MAX(bs->bl.opt_mem_alignment, | |
760e0063 | 188 | bs->backing->bs->bl.opt_mem_alignment); |
4196d2f0 DL |
189 | bs->bl.min_mem_alignment = |
190 | MAX(bs->bl.min_mem_alignment, | |
760e0063 | 191 | bs->backing->bs->bl.min_mem_alignment); |
61007b31 SH |
192 | } |
193 | ||
194 | /* Then let the driver override it */ | |
195 | if (drv->bdrv_refresh_limits) { | |
196 | drv->bdrv_refresh_limits(bs, errp); | |
197 | } | |
198 | } | |
199 | ||
200 | /** | |
201 | * The copy-on-read flag is actually a reference count so multiple users may | |
202 | * use the feature without worrying about clobbering its previous state. | |
203 | * Copy-on-read stays enabled until all users have called to disable it. | |
204 | */ | |
205 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
206 | { | |
207 | bs->copy_on_read++; | |
208 | } | |
209 | ||
210 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
211 | { | |
212 | assert(bs->copy_on_read > 0); | |
213 | bs->copy_on_read--; | |
214 | } | |
215 | ||
216 | /* Check if any requests are in-flight (including throttled requests) */ | |
439db28c | 217 | bool bdrv_requests_pending(BlockDriverState *bs) |
61007b31 SH |
218 | { |
219 | if (!QLIST_EMPTY(&bs->tracked_requests)) { | |
220 | return true; | |
221 | } | |
222 | if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) { | |
223 | return true; | |
224 | } | |
225 | if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) { | |
226 | return true; | |
227 | } | |
9a4f4c31 | 228 | if (bs->file && bdrv_requests_pending(bs->file->bs)) { |
61007b31 SH |
229 | return true; |
230 | } | |
760e0063 | 231 | if (bs->backing && bdrv_requests_pending(bs->backing->bs)) { |
61007b31 SH |
232 | return true; |
233 | } | |
234 | return false; | |
235 | } | |
236 | ||
61007b31 SH |
237 | /* |
238 | * Wait for pending requests to complete on a single BlockDriverState subtree | |
239 | * | |
61007b31 SH |
240 | * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState |
241 | * AioContext. | |
7a63f3cd SH |
242 | * |
243 | * Only this BlockDriverState's AioContext is run, so in-flight requests must | |
244 | * not depend on events in other AioContexts. In that case, use | |
245 | * bdrv_drain_all() instead. | |
61007b31 SH |
246 | */ |
247 | void bdrv_drain(BlockDriverState *bs) | |
248 | { | |
f406c03c AY |
249 | bool busy = true; |
250 | ||
251 | while (busy) { | |
61007b31 | 252 | /* Keep iterating */ |
f406c03c AY |
253 | bdrv_flush_io_queue(bs); |
254 | busy = bdrv_requests_pending(bs); | |
255 | busy |= aio_poll(bdrv_get_aio_context(bs), busy); | |
61007b31 SH |
256 | } |
257 | } | |
258 | ||
259 | /* | |
260 | * Wait for pending requests to complete across all BlockDriverStates | |
261 | * | |
262 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
263 | * after calling this function. | |
61007b31 SH |
264 | */ |
265 | void bdrv_drain_all(void) | |
266 | { | |
267 | /* Always run first iteration so any pending completion BHs run */ | |
268 | bool busy = true; | |
269 | BlockDriverState *bs = NULL; | |
f406c03c | 270 | GSList *aio_ctxs = NULL, *ctx; |
61007b31 SH |
271 | |
272 | while ((bs = bdrv_next(bs))) { | |
273 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
274 | ||
275 | aio_context_acquire(aio_context); | |
276 | if (bs->job) { | |
277 | block_job_pause(bs->job); | |
278 | } | |
279 | aio_context_release(aio_context); | |
f406c03c | 280 | |
764ba3ae | 281 | if (!g_slist_find(aio_ctxs, aio_context)) { |
f406c03c AY |
282 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); |
283 | } | |
61007b31 SH |
284 | } |
285 | ||
7a63f3cd SH |
286 | /* Note that completion of an asynchronous I/O operation can trigger any |
287 | * number of other I/O operations on other devices---for example a | |
288 | * coroutine can submit an I/O request to another device in response to | |
289 | * request completion. Therefore we must keep looping until there was no | |
290 | * more activity rather than simply draining each device independently. | |
291 | */ | |
61007b31 SH |
292 | while (busy) { |
293 | busy = false; | |
61007b31 | 294 | |
f406c03c AY |
295 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
296 | AioContext *aio_context = ctx->data; | |
297 | bs = NULL; | |
61007b31 SH |
298 | |
299 | aio_context_acquire(aio_context); | |
f406c03c AY |
300 | while ((bs = bdrv_next(bs))) { |
301 | if (aio_context == bdrv_get_aio_context(bs)) { | |
302 | bdrv_flush_io_queue(bs); | |
303 | if (bdrv_requests_pending(bs)) { | |
304 | busy = true; | |
305 | aio_poll(aio_context, busy); | |
306 | } | |
307 | } | |
308 | } | |
309 | busy |= aio_poll(aio_context, false); | |
61007b31 SH |
310 | aio_context_release(aio_context); |
311 | } | |
312 | } | |
313 | ||
314 | bs = NULL; | |
315 | while ((bs = bdrv_next(bs))) { | |
316 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
317 | ||
318 | aio_context_acquire(aio_context); | |
319 | if (bs->job) { | |
320 | block_job_resume(bs->job); | |
321 | } | |
322 | aio_context_release(aio_context); | |
323 | } | |
f406c03c | 324 | g_slist_free(aio_ctxs); |
61007b31 SH |
325 | } |
326 | ||
327 | /** | |
328 | * Remove an active request from the tracked requests list | |
329 | * | |
330 | * This function should be called when a tracked request is completing. | |
331 | */ | |
332 | static void tracked_request_end(BdrvTrackedRequest *req) | |
333 | { | |
334 | if (req->serialising) { | |
335 | req->bs->serialising_in_flight--; | |
336 | } | |
337 | ||
338 | QLIST_REMOVE(req, list); | |
339 | qemu_co_queue_restart_all(&req->wait_queue); | |
340 | } | |
341 | ||
342 | /** | |
343 | * Add an active request to the tracked requests list | |
344 | */ | |
345 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
346 | BlockDriverState *bs, | |
347 | int64_t offset, | |
348 | unsigned int bytes, bool is_write) | |
349 | { | |
350 | *req = (BdrvTrackedRequest){ | |
351 | .bs = bs, | |
352 | .offset = offset, | |
353 | .bytes = bytes, | |
354 | .is_write = is_write, | |
355 | .co = qemu_coroutine_self(), | |
356 | .serialising = false, | |
357 | .overlap_offset = offset, | |
358 | .overlap_bytes = bytes, | |
359 | }; | |
360 | ||
361 | qemu_co_queue_init(&req->wait_queue); | |
362 | ||
363 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); | |
364 | } | |
365 | ||
366 | static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align) | |
367 | { | |
368 | int64_t overlap_offset = req->offset & ~(align - 1); | |
369 | unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align) | |
370 | - overlap_offset; | |
371 | ||
372 | if (!req->serialising) { | |
373 | req->bs->serialising_in_flight++; | |
374 | req->serialising = true; | |
375 | } | |
376 | ||
377 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
378 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
379 | } | |
380 | ||
381 | /** | |
382 | * Round a region to cluster boundaries | |
383 | */ | |
384 | void bdrv_round_to_clusters(BlockDriverState *bs, | |
385 | int64_t sector_num, int nb_sectors, | |
386 | int64_t *cluster_sector_num, | |
387 | int *cluster_nb_sectors) | |
388 | { | |
389 | BlockDriverInfo bdi; | |
390 | ||
391 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
392 | *cluster_sector_num = sector_num; | |
393 | *cluster_nb_sectors = nb_sectors; | |
394 | } else { | |
395 | int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; | |
396 | *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); | |
397 | *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + | |
398 | nb_sectors, c); | |
399 | } | |
400 | } | |
401 | ||
402 | static int bdrv_get_cluster_size(BlockDriverState *bs) | |
403 | { | |
404 | BlockDriverInfo bdi; | |
405 | int ret; | |
406 | ||
407 | ret = bdrv_get_info(bs, &bdi); | |
408 | if (ret < 0 || bdi.cluster_size == 0) { | |
409 | return bs->request_alignment; | |
410 | } else { | |
411 | return bdi.cluster_size; | |
412 | } | |
413 | } | |
414 | ||
415 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
416 | int64_t offset, unsigned int bytes) | |
417 | { | |
418 | /* aaaa bbbb */ | |
419 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
420 | return false; | |
421 | } | |
422 | /* bbbb aaaa */ | |
423 | if (req->overlap_offset >= offset + bytes) { | |
424 | return false; | |
425 | } | |
426 | return true; | |
427 | } | |
428 | ||
429 | static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self) | |
430 | { | |
431 | BlockDriverState *bs = self->bs; | |
432 | BdrvTrackedRequest *req; | |
433 | bool retry; | |
434 | bool waited = false; | |
435 | ||
436 | if (!bs->serialising_in_flight) { | |
437 | return false; | |
438 | } | |
439 | ||
440 | do { | |
441 | retry = false; | |
442 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
443 | if (req == self || (!req->serialising && !self->serialising)) { | |
444 | continue; | |
445 | } | |
446 | if (tracked_request_overlaps(req, self->overlap_offset, | |
447 | self->overlap_bytes)) | |
448 | { | |
449 | /* Hitting this means there was a reentrant request, for | |
450 | * example, a block driver issuing nested requests. This must | |
451 | * never happen since it means deadlock. | |
452 | */ | |
453 | assert(qemu_coroutine_self() != req->co); | |
454 | ||
455 | /* If the request is already (indirectly) waiting for us, or | |
456 | * will wait for us as soon as it wakes up, then just go on | |
457 | * (instead of producing a deadlock in the former case). */ | |
458 | if (!req->waiting_for) { | |
459 | self->waiting_for = req; | |
460 | qemu_co_queue_wait(&req->wait_queue); | |
461 | self->waiting_for = NULL; | |
462 | retry = true; | |
463 | waited = true; | |
464 | break; | |
465 | } | |
466 | } | |
467 | } | |
468 | } while (retry); | |
469 | ||
470 | return waited; | |
471 | } | |
472 | ||
473 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
474 | size_t size) | |
475 | { | |
476 | if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) { | |
477 | return -EIO; | |
478 | } | |
479 | ||
480 | if (!bdrv_is_inserted(bs)) { | |
481 | return -ENOMEDIUM; | |
482 | } | |
483 | ||
484 | if (offset < 0) { | |
485 | return -EIO; | |
486 | } | |
487 | ||
488 | return 0; | |
489 | } | |
490 | ||
491 | static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num, | |
492 | int nb_sectors) | |
493 | { | |
494 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
495 | return -EIO; | |
496 | } | |
497 | ||
498 | return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE, | |
499 | nb_sectors * BDRV_SECTOR_SIZE); | |
500 | } | |
501 | ||
502 | typedef struct RwCo { | |
503 | BlockDriverState *bs; | |
504 | int64_t offset; | |
505 | QEMUIOVector *qiov; | |
506 | bool is_write; | |
507 | int ret; | |
508 | BdrvRequestFlags flags; | |
509 | } RwCo; | |
510 | ||
511 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
512 | { | |
513 | RwCo *rwco = opaque; | |
514 | ||
515 | if (!rwco->is_write) { | |
516 | rwco->ret = bdrv_co_do_preadv(rwco->bs, rwco->offset, | |
517 | rwco->qiov->size, rwco->qiov, | |
518 | rwco->flags); | |
519 | } else { | |
520 | rwco->ret = bdrv_co_do_pwritev(rwco->bs, rwco->offset, | |
521 | rwco->qiov->size, rwco->qiov, | |
522 | rwco->flags); | |
523 | } | |
524 | } | |
525 | ||
526 | /* | |
527 | * Process a vectored synchronous request using coroutines | |
528 | */ | |
529 | static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset, | |
530 | QEMUIOVector *qiov, bool is_write, | |
531 | BdrvRequestFlags flags) | |
532 | { | |
533 | Coroutine *co; | |
534 | RwCo rwco = { | |
535 | .bs = bs, | |
536 | .offset = offset, | |
537 | .qiov = qiov, | |
538 | .is_write = is_write, | |
539 | .ret = NOT_DONE, | |
540 | .flags = flags, | |
541 | }; | |
542 | ||
543 | /** | |
544 | * In sync call context, when the vcpu is blocked, this throttling timer | |
545 | * will not fire; so the I/O throttling function has to be disabled here | |
546 | * if it has been enabled. | |
547 | */ | |
548 | if (bs->io_limits_enabled) { | |
549 | fprintf(stderr, "Disabling I/O throttling on '%s' due " | |
550 | "to synchronous I/O.\n", bdrv_get_device_name(bs)); | |
551 | bdrv_io_limits_disable(bs); | |
552 | } | |
553 | ||
554 | if (qemu_in_coroutine()) { | |
555 | /* Fast-path if already in coroutine context */ | |
556 | bdrv_rw_co_entry(&rwco); | |
557 | } else { | |
558 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
559 | ||
560 | co = qemu_coroutine_create(bdrv_rw_co_entry); | |
561 | qemu_coroutine_enter(co, &rwco); | |
562 | while (rwco.ret == NOT_DONE) { | |
563 | aio_poll(aio_context, true); | |
564 | } | |
565 | } | |
566 | return rwco.ret; | |
567 | } | |
568 | ||
569 | /* | |
570 | * Process a synchronous request using coroutines | |
571 | */ | |
572 | static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf, | |
573 | int nb_sectors, bool is_write, BdrvRequestFlags flags) | |
574 | { | |
575 | QEMUIOVector qiov; | |
576 | struct iovec iov = { | |
577 | .iov_base = (void *)buf, | |
578 | .iov_len = nb_sectors * BDRV_SECTOR_SIZE, | |
579 | }; | |
580 | ||
581 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
582 | return -EINVAL; | |
583 | } | |
584 | ||
585 | qemu_iovec_init_external(&qiov, &iov, 1); | |
586 | return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS, | |
587 | &qiov, is_write, flags); | |
588 | } | |
589 | ||
590 | /* return < 0 if error. See bdrv_write() for the return codes */ | |
591 | int bdrv_read(BlockDriverState *bs, int64_t sector_num, | |
592 | uint8_t *buf, int nb_sectors) | |
593 | { | |
594 | return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0); | |
595 | } | |
596 | ||
597 | /* Just like bdrv_read(), but with I/O throttling temporarily disabled */ | |
598 | int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num, | |
599 | uint8_t *buf, int nb_sectors) | |
600 | { | |
601 | bool enabled; | |
602 | int ret; | |
603 | ||
604 | enabled = bs->io_limits_enabled; | |
605 | bs->io_limits_enabled = false; | |
606 | ret = bdrv_read(bs, sector_num, buf, nb_sectors); | |
607 | bs->io_limits_enabled = enabled; | |
608 | return ret; | |
609 | } | |
610 | ||
611 | /* Return < 0 if error. Important errors are: | |
612 | -EIO generic I/O error (may happen for all errors) | |
613 | -ENOMEDIUM No media inserted. | |
614 | -EINVAL Invalid sector number or nb_sectors | |
615 | -EACCES Trying to write a read-only device | |
616 | */ | |
617 | int bdrv_write(BlockDriverState *bs, int64_t sector_num, | |
618 | const uint8_t *buf, int nb_sectors) | |
619 | { | |
620 | return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0); | |
621 | } | |
622 | ||
623 | int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num, | |
624 | int nb_sectors, BdrvRequestFlags flags) | |
625 | { | |
626 | return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true, | |
627 | BDRV_REQ_ZERO_WRITE | flags); | |
628 | } | |
629 | ||
630 | /* | |
631 | * Completely zero out a block device with the help of bdrv_write_zeroes. | |
632 | * The operation is sped up by checking the block status and only writing | |
633 | * zeroes to the device if they currently do not return zeroes. Optional | |
634 | * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP). | |
635 | * | |
636 | * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). | |
637 | */ | |
638 | int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags) | |
639 | { | |
640 | int64_t target_sectors, ret, nb_sectors, sector_num = 0; | |
641 | int n; | |
642 | ||
643 | target_sectors = bdrv_nb_sectors(bs); | |
644 | if (target_sectors < 0) { | |
645 | return target_sectors; | |
646 | } | |
647 | ||
648 | for (;;) { | |
649 | nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS); | |
650 | if (nb_sectors <= 0) { | |
651 | return 0; | |
652 | } | |
653 | ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n); | |
654 | if (ret < 0) { | |
655 | error_report("error getting block status at sector %" PRId64 ": %s", | |
656 | sector_num, strerror(-ret)); | |
657 | return ret; | |
658 | } | |
659 | if (ret & BDRV_BLOCK_ZERO) { | |
660 | sector_num += n; | |
661 | continue; | |
662 | } | |
663 | ret = bdrv_write_zeroes(bs, sector_num, n, flags); | |
664 | if (ret < 0) { | |
665 | error_report("error writing zeroes at sector %" PRId64 ": %s", | |
666 | sector_num, strerror(-ret)); | |
667 | return ret; | |
668 | } | |
669 | sector_num += n; | |
670 | } | |
671 | } | |
672 | ||
673 | int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes) | |
674 | { | |
675 | QEMUIOVector qiov; | |
676 | struct iovec iov = { | |
677 | .iov_base = (void *)buf, | |
678 | .iov_len = bytes, | |
679 | }; | |
680 | int ret; | |
681 | ||
682 | if (bytes < 0) { | |
683 | return -EINVAL; | |
684 | } | |
685 | ||
686 | qemu_iovec_init_external(&qiov, &iov, 1); | |
687 | ret = bdrv_prwv_co(bs, offset, &qiov, false, 0); | |
688 | if (ret < 0) { | |
689 | return ret; | |
690 | } | |
691 | ||
692 | return bytes; | |
693 | } | |
694 | ||
695 | int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov) | |
696 | { | |
697 | int ret; | |
698 | ||
699 | ret = bdrv_prwv_co(bs, offset, qiov, true, 0); | |
700 | if (ret < 0) { | |
701 | return ret; | |
702 | } | |
703 | ||
704 | return qiov->size; | |
705 | } | |
706 | ||
707 | int bdrv_pwrite(BlockDriverState *bs, int64_t offset, | |
708 | const void *buf, int bytes) | |
709 | { | |
710 | QEMUIOVector qiov; | |
711 | struct iovec iov = { | |
712 | .iov_base = (void *) buf, | |
713 | .iov_len = bytes, | |
714 | }; | |
715 | ||
716 | if (bytes < 0) { | |
717 | return -EINVAL; | |
718 | } | |
719 | ||
720 | qemu_iovec_init_external(&qiov, &iov, 1); | |
721 | return bdrv_pwritev(bs, offset, &qiov); | |
722 | } | |
723 | ||
724 | /* | |
725 | * Writes to the file and ensures that no writes are reordered across this | |
726 | * request (acts as a barrier) | |
727 | * | |
728 | * Returns 0 on success, -errno in error cases. | |
729 | */ | |
730 | int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset, | |
731 | const void *buf, int count) | |
732 | { | |
733 | int ret; | |
734 | ||
735 | ret = bdrv_pwrite(bs, offset, buf, count); | |
736 | if (ret < 0) { | |
737 | return ret; | |
738 | } | |
739 | ||
740 | /* No flush needed for cache modes that already do it */ | |
741 | if (bs->enable_write_cache) { | |
742 | bdrv_flush(bs); | |
743 | } | |
744 | ||
745 | return 0; | |
746 | } | |
747 | ||
748 | static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, | |
749 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
750 | { | |
751 | /* Perform I/O through a temporary buffer so that users who scribble over | |
752 | * their read buffer while the operation is in progress do not end up | |
753 | * modifying the image file. This is critical for zero-copy guest I/O | |
754 | * where anything might happen inside guest memory. | |
755 | */ | |
756 | void *bounce_buffer; | |
757 | ||
758 | BlockDriver *drv = bs->drv; | |
759 | struct iovec iov; | |
760 | QEMUIOVector bounce_qiov; | |
761 | int64_t cluster_sector_num; | |
762 | int cluster_nb_sectors; | |
763 | size_t skip_bytes; | |
764 | int ret; | |
765 | ||
766 | /* Cover entire cluster so no additional backing file I/O is required when | |
767 | * allocating cluster in the image file. | |
768 | */ | |
769 | bdrv_round_to_clusters(bs, sector_num, nb_sectors, | |
770 | &cluster_sector_num, &cluster_nb_sectors); | |
771 | ||
772 | trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, | |
773 | cluster_sector_num, cluster_nb_sectors); | |
774 | ||
775 | iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE; | |
776 | iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len); | |
777 | if (bounce_buffer == NULL) { | |
778 | ret = -ENOMEM; | |
779 | goto err; | |
780 | } | |
781 | ||
782 | qemu_iovec_init_external(&bounce_qiov, &iov, 1); | |
783 | ||
784 | ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors, | |
785 | &bounce_qiov); | |
786 | if (ret < 0) { | |
787 | goto err; | |
788 | } | |
789 | ||
790 | if (drv->bdrv_co_write_zeroes && | |
791 | buffer_is_zero(bounce_buffer, iov.iov_len)) { | |
792 | ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num, | |
793 | cluster_nb_sectors, 0); | |
794 | } else { | |
795 | /* This does not change the data on the disk, it is not necessary | |
796 | * to flush even in cache=writethrough mode. | |
797 | */ | |
798 | ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors, | |
799 | &bounce_qiov); | |
800 | } | |
801 | ||
802 | if (ret < 0) { | |
803 | /* It might be okay to ignore write errors for guest requests. If this | |
804 | * is a deliberate copy-on-read then we don't want to ignore the error. | |
805 | * Simply report it in all cases. | |
806 | */ | |
807 | goto err; | |
808 | } | |
809 | ||
810 | skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE; | |
811 | qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, | |
812 | nb_sectors * BDRV_SECTOR_SIZE); | |
813 | ||
814 | err: | |
815 | qemu_vfree(bounce_buffer); | |
816 | return ret; | |
817 | } | |
818 | ||
819 | /* | |
820 | * Forwards an already correctly aligned request to the BlockDriver. This | |
821 | * handles copy on read and zeroing after EOF; any other features must be | |
822 | * implemented by the caller. | |
823 | */ | |
824 | static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs, | |
825 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, | |
826 | int64_t align, QEMUIOVector *qiov, int flags) | |
827 | { | |
828 | BlockDriver *drv = bs->drv; | |
829 | int ret; | |
830 | ||
831 | int64_t sector_num = offset >> BDRV_SECTOR_BITS; | |
832 | unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
833 | ||
834 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
835 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
836 | assert(!qiov || bytes == qiov->size); | |
837 | ||
838 | /* Handle Copy on Read and associated serialisation */ | |
839 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
840 | /* If we touch the same cluster it counts as an overlap. This | |
841 | * guarantees that allocating writes will be serialized and not race | |
842 | * with each other for the same cluster. For example, in copy-on-read | |
843 | * it ensures that the CoR read and write operations are atomic and | |
844 | * guest writes cannot interleave between them. */ | |
845 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
846 | } | |
847 | ||
848 | wait_serialising_requests(req); | |
849 | ||
850 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
851 | int pnum; | |
852 | ||
853 | ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum); | |
854 | if (ret < 0) { | |
855 | goto out; | |
856 | } | |
857 | ||
858 | if (!ret || pnum != nb_sectors) { | |
859 | ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov); | |
860 | goto out; | |
861 | } | |
862 | } | |
863 | ||
864 | /* Forward the request to the BlockDriver */ | |
865 | if (!bs->zero_beyond_eof) { | |
866 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
867 | } else { | |
868 | /* Read zeros after EOF */ | |
869 | int64_t total_sectors, max_nb_sectors; | |
870 | ||
871 | total_sectors = bdrv_nb_sectors(bs); | |
872 | if (total_sectors < 0) { | |
873 | ret = total_sectors; | |
874 | goto out; | |
875 | } | |
876 | ||
877 | max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num), | |
878 | align >> BDRV_SECTOR_BITS); | |
879 | if (nb_sectors < max_nb_sectors) { | |
880 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
881 | } else if (max_nb_sectors > 0) { | |
882 | QEMUIOVector local_qiov; | |
883 | ||
884 | qemu_iovec_init(&local_qiov, qiov->niov); | |
885 | qemu_iovec_concat(&local_qiov, qiov, 0, | |
886 | max_nb_sectors * BDRV_SECTOR_SIZE); | |
887 | ||
888 | ret = drv->bdrv_co_readv(bs, sector_num, max_nb_sectors, | |
889 | &local_qiov); | |
890 | ||
891 | qemu_iovec_destroy(&local_qiov); | |
892 | } else { | |
893 | ret = 0; | |
894 | } | |
895 | ||
896 | /* Reading beyond end of file is supposed to produce zeroes */ | |
897 | if (ret == 0 && total_sectors < sector_num + nb_sectors) { | |
898 | uint64_t offset = MAX(0, total_sectors - sector_num); | |
899 | uint64_t bytes = (sector_num + nb_sectors - offset) * | |
900 | BDRV_SECTOR_SIZE; | |
901 | qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes); | |
902 | } | |
903 | } | |
904 | ||
905 | out: | |
906 | return ret; | |
907 | } | |
908 | ||
61007b31 SH |
909 | /* |
910 | * Handle a read request in coroutine context | |
911 | */ | |
912 | static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs, | |
913 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, | |
914 | BdrvRequestFlags flags) | |
915 | { | |
916 | BlockDriver *drv = bs->drv; | |
917 | BdrvTrackedRequest req; | |
918 | ||
d01c07f2 FZ |
919 | /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */ |
920 | uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment); | |
61007b31 SH |
921 | uint8_t *head_buf = NULL; |
922 | uint8_t *tail_buf = NULL; | |
923 | QEMUIOVector local_qiov; | |
924 | bool use_local_qiov = false; | |
925 | int ret; | |
926 | ||
927 | if (!drv) { | |
928 | return -ENOMEDIUM; | |
929 | } | |
930 | ||
931 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
932 | if (ret < 0) { | |
933 | return ret; | |
934 | } | |
935 | ||
9568b511 WC |
936 | /* Don't do copy-on-read if we read data before write operation */ |
937 | if (bs->copy_on_read && !(flags & BDRV_REQ_NO_COPY_ON_READ)) { | |
61007b31 SH |
938 | flags |= BDRV_REQ_COPY_ON_READ; |
939 | } | |
940 | ||
941 | /* throttling disk I/O */ | |
942 | if (bs->io_limits_enabled) { | |
76f4afb4 | 943 | throttle_group_co_io_limits_intercept(bs, bytes, false); |
61007b31 SH |
944 | } |
945 | ||
946 | /* Align read if necessary by padding qiov */ | |
947 | if (offset & (align - 1)) { | |
948 | head_buf = qemu_blockalign(bs, align); | |
949 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
950 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
951 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
952 | use_local_qiov = true; | |
953 | ||
954 | bytes += offset & (align - 1); | |
955 | offset = offset & ~(align - 1); | |
956 | } | |
957 | ||
958 | if ((offset + bytes) & (align - 1)) { | |
959 | if (!use_local_qiov) { | |
960 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
961 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
962 | use_local_qiov = true; | |
963 | } | |
964 | tail_buf = qemu_blockalign(bs, align); | |
965 | qemu_iovec_add(&local_qiov, tail_buf, | |
966 | align - ((offset + bytes) & (align - 1))); | |
967 | ||
968 | bytes = ROUND_UP(bytes, align); | |
969 | } | |
970 | ||
971 | tracked_request_begin(&req, bs, offset, bytes, false); | |
972 | ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align, | |
973 | use_local_qiov ? &local_qiov : qiov, | |
974 | flags); | |
975 | tracked_request_end(&req); | |
976 | ||
977 | if (use_local_qiov) { | |
978 | qemu_iovec_destroy(&local_qiov); | |
979 | qemu_vfree(head_buf); | |
980 | qemu_vfree(tail_buf); | |
981 | } | |
982 | ||
983 | return ret; | |
984 | } | |
985 | ||
986 | static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, | |
987 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
988 | BdrvRequestFlags flags) | |
989 | { | |
990 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
991 | return -EINVAL; | |
992 | } | |
993 | ||
994 | return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS, | |
995 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); | |
996 | } | |
997 | ||
998 | int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num, | |
999 | int nb_sectors, QEMUIOVector *qiov) | |
1000 | { | |
1001 | trace_bdrv_co_readv(bs, sector_num, nb_sectors); | |
1002 | ||
1003 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0); | |
1004 | } | |
1005 | ||
9568b511 WC |
1006 | int coroutine_fn bdrv_co_no_copy_on_readv(BlockDriverState *bs, |
1007 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
1008 | { | |
1009 | trace_bdrv_co_no_copy_on_readv(bs, sector_num, nb_sectors); | |
1010 | ||
1011 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, | |
1012 | BDRV_REQ_NO_COPY_ON_READ); | |
1013 | } | |
1014 | ||
61007b31 SH |
1015 | int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs, |
1016 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
1017 | { | |
1018 | trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors); | |
1019 | ||
1020 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, | |
1021 | BDRV_REQ_COPY_ON_READ); | |
1022 | } | |
1023 | ||
1024 | #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768 | |
1025 | ||
1026 | static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, | |
1027 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags) | |
1028 | { | |
1029 | BlockDriver *drv = bs->drv; | |
1030 | QEMUIOVector qiov; | |
1031 | struct iovec iov = {0}; | |
1032 | int ret = 0; | |
1033 | ||
1034 | int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_write_zeroes, | |
1035 | BDRV_REQUEST_MAX_SECTORS); | |
1036 | ||
1037 | while (nb_sectors > 0 && !ret) { | |
1038 | int num = nb_sectors; | |
1039 | ||
1040 | /* Align request. Block drivers can expect the "bulk" of the request | |
1041 | * to be aligned. | |
1042 | */ | |
1043 | if (bs->bl.write_zeroes_alignment | |
1044 | && num > bs->bl.write_zeroes_alignment) { | |
1045 | if (sector_num % bs->bl.write_zeroes_alignment != 0) { | |
1046 | /* Make a small request up to the first aligned sector. */ | |
1047 | num = bs->bl.write_zeroes_alignment; | |
1048 | num -= sector_num % bs->bl.write_zeroes_alignment; | |
1049 | } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) { | |
1050 | /* Shorten the request to the last aligned sector. num cannot | |
1051 | * underflow because num > bs->bl.write_zeroes_alignment. | |
1052 | */ | |
1053 | num -= (sector_num + num) % bs->bl.write_zeroes_alignment; | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | /* limit request size */ | |
1058 | if (num > max_write_zeroes) { | |
1059 | num = max_write_zeroes; | |
1060 | } | |
1061 | ||
1062 | ret = -ENOTSUP; | |
1063 | /* First try the efficient write zeroes operation */ | |
1064 | if (drv->bdrv_co_write_zeroes) { | |
1065 | ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags); | |
1066 | } | |
1067 | ||
1068 | if (ret == -ENOTSUP) { | |
1069 | /* Fall back to bounce buffer if write zeroes is unsupported */ | |
1070 | int max_xfer_len = MIN_NON_ZERO(bs->bl.max_transfer_length, | |
1071 | MAX_WRITE_ZEROES_BOUNCE_BUFFER); | |
1072 | num = MIN(num, max_xfer_len); | |
1073 | iov.iov_len = num * BDRV_SECTOR_SIZE; | |
1074 | if (iov.iov_base == NULL) { | |
1075 | iov.iov_base = qemu_try_blockalign(bs, num * BDRV_SECTOR_SIZE); | |
1076 | if (iov.iov_base == NULL) { | |
1077 | ret = -ENOMEM; | |
1078 | goto fail; | |
1079 | } | |
1080 | memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE); | |
1081 | } | |
1082 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1083 | ||
1084 | ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov); | |
1085 | ||
1086 | /* Keep bounce buffer around if it is big enough for all | |
1087 | * all future requests. | |
1088 | */ | |
1089 | if (num < max_xfer_len) { | |
1090 | qemu_vfree(iov.iov_base); | |
1091 | iov.iov_base = NULL; | |
1092 | } | |
1093 | } | |
1094 | ||
1095 | sector_num += num; | |
1096 | nb_sectors -= num; | |
1097 | } | |
1098 | ||
1099 | fail: | |
1100 | qemu_vfree(iov.iov_base); | |
1101 | return ret; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * Forwards an already correctly aligned write request to the BlockDriver. | |
1106 | */ | |
1107 | static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs, | |
1108 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, | |
1109 | QEMUIOVector *qiov, int flags) | |
1110 | { | |
1111 | BlockDriver *drv = bs->drv; | |
1112 | bool waited; | |
1113 | int ret; | |
1114 | ||
1115 | int64_t sector_num = offset >> BDRV_SECTOR_BITS; | |
1116 | unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1117 | ||
1118 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1119 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1120 | assert(!qiov || bytes == qiov->size); | |
1121 | ||
1122 | waited = wait_serialising_requests(req); | |
1123 | assert(!waited || !req->serialising); | |
1124 | assert(req->overlap_offset <= offset); | |
1125 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
1126 | ||
1127 | ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req); | |
1128 | ||
1129 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
1130 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_write_zeroes && | |
1131 | qemu_iovec_is_zero(qiov)) { | |
1132 | flags |= BDRV_REQ_ZERO_WRITE; | |
1133 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
1134 | flags |= BDRV_REQ_MAY_UNMAP; | |
1135 | } | |
1136 | } | |
1137 | ||
1138 | if (ret < 0) { | |
1139 | /* Do nothing, write notifier decided to fail this request */ | |
1140 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
9a4f4c31 | 1141 | bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
61007b31 SH |
1142 | ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags); |
1143 | } else { | |
9a4f4c31 | 1144 | bdrv_debug_event(bs, BLKDBG_PWRITEV); |
61007b31 SH |
1145 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); |
1146 | } | |
9a4f4c31 | 1147 | bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 SH |
1148 | |
1149 | if (ret == 0 && !bs->enable_write_cache) { | |
1150 | ret = bdrv_co_flush(bs); | |
1151 | } | |
1152 | ||
1153 | bdrv_set_dirty(bs, sector_num, nb_sectors); | |
1154 | ||
53d8f9d8 HR |
1155 | if (bs->wr_highest_offset < offset + bytes) { |
1156 | bs->wr_highest_offset = offset + bytes; | |
1157 | } | |
61007b31 SH |
1158 | |
1159 | if (ret >= 0) { | |
1160 | bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors); | |
1161 | } | |
1162 | ||
1163 | return ret; | |
1164 | } | |
1165 | ||
9eeb6dd1 FZ |
1166 | static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs, |
1167 | int64_t offset, | |
1168 | unsigned int bytes, | |
1169 | BdrvRequestFlags flags, | |
1170 | BdrvTrackedRequest *req) | |
1171 | { | |
1172 | uint8_t *buf = NULL; | |
1173 | QEMUIOVector local_qiov; | |
1174 | struct iovec iov; | |
1175 | uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment); | |
1176 | unsigned int head_padding_bytes, tail_padding_bytes; | |
1177 | int ret = 0; | |
1178 | ||
1179 | head_padding_bytes = offset & (align - 1); | |
1180 | tail_padding_bytes = align - ((offset + bytes) & (align - 1)); | |
1181 | ||
1182 | ||
1183 | assert(flags & BDRV_REQ_ZERO_WRITE); | |
1184 | if (head_padding_bytes || tail_padding_bytes) { | |
1185 | buf = qemu_blockalign(bs, align); | |
1186 | iov = (struct iovec) { | |
1187 | .iov_base = buf, | |
1188 | .iov_len = align, | |
1189 | }; | |
1190 | qemu_iovec_init_external(&local_qiov, &iov, 1); | |
1191 | } | |
1192 | if (head_padding_bytes) { | |
1193 | uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes); | |
1194 | ||
1195 | /* RMW the unaligned part before head. */ | |
1196 | mark_request_serialising(req, align); | |
1197 | wait_serialising_requests(req); | |
9a4f4c31 | 1198 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
9eeb6dd1 FZ |
1199 | ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align, |
1200 | align, &local_qiov, 0); | |
1201 | if (ret < 0) { | |
1202 | goto fail; | |
1203 | } | |
9a4f4c31 | 1204 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
9eeb6dd1 FZ |
1205 | |
1206 | memset(buf + head_padding_bytes, 0, zero_bytes); | |
1207 | ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align, | |
1208 | &local_qiov, | |
1209 | flags & ~BDRV_REQ_ZERO_WRITE); | |
1210 | if (ret < 0) { | |
1211 | goto fail; | |
1212 | } | |
1213 | offset += zero_bytes; | |
1214 | bytes -= zero_bytes; | |
1215 | } | |
1216 | ||
1217 | assert(!bytes || (offset & (align - 1)) == 0); | |
1218 | if (bytes >= align) { | |
1219 | /* Write the aligned part in the middle. */ | |
1220 | uint64_t aligned_bytes = bytes & ~(align - 1); | |
1221 | ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes, | |
1222 | NULL, flags); | |
1223 | if (ret < 0) { | |
1224 | goto fail; | |
1225 | } | |
1226 | bytes -= aligned_bytes; | |
1227 | offset += aligned_bytes; | |
1228 | } | |
1229 | ||
1230 | assert(!bytes || (offset & (align - 1)) == 0); | |
1231 | if (bytes) { | |
1232 | assert(align == tail_padding_bytes + bytes); | |
1233 | /* RMW the unaligned part after tail. */ | |
1234 | mark_request_serialising(req, align); | |
1235 | wait_serialising_requests(req); | |
9a4f4c31 | 1236 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
9eeb6dd1 FZ |
1237 | ret = bdrv_aligned_preadv(bs, req, offset, align, |
1238 | align, &local_qiov, 0); | |
1239 | if (ret < 0) { | |
1240 | goto fail; | |
1241 | } | |
9a4f4c31 | 1242 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
9eeb6dd1 FZ |
1243 | |
1244 | memset(buf, 0, bytes); | |
1245 | ret = bdrv_aligned_pwritev(bs, req, offset, align, | |
1246 | &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE); | |
1247 | } | |
1248 | fail: | |
1249 | qemu_vfree(buf); | |
1250 | return ret; | |
1251 | ||
1252 | } | |
1253 | ||
61007b31 SH |
1254 | /* |
1255 | * Handle a write request in coroutine context | |
1256 | */ | |
1257 | static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs, | |
1258 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, | |
1259 | BdrvRequestFlags flags) | |
1260 | { | |
1261 | BdrvTrackedRequest req; | |
d01c07f2 FZ |
1262 | /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */ |
1263 | uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment); | |
61007b31 SH |
1264 | uint8_t *head_buf = NULL; |
1265 | uint8_t *tail_buf = NULL; | |
1266 | QEMUIOVector local_qiov; | |
1267 | bool use_local_qiov = false; | |
1268 | int ret; | |
1269 | ||
1270 | if (!bs->drv) { | |
1271 | return -ENOMEDIUM; | |
1272 | } | |
1273 | if (bs->read_only) { | |
eaf5fe2d | 1274 | return -EPERM; |
61007b31 SH |
1275 | } |
1276 | ||
1277 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1278 | if (ret < 0) { | |
1279 | return ret; | |
1280 | } | |
1281 | ||
1282 | /* throttling disk I/O */ | |
1283 | if (bs->io_limits_enabled) { | |
76f4afb4 | 1284 | throttle_group_co_io_limits_intercept(bs, bytes, true); |
61007b31 SH |
1285 | } |
1286 | ||
1287 | /* | |
1288 | * Align write if necessary by performing a read-modify-write cycle. | |
1289 | * Pad qiov with the read parts and be sure to have a tracked request not | |
1290 | * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle. | |
1291 | */ | |
1292 | tracked_request_begin(&req, bs, offset, bytes, true); | |
1293 | ||
9eeb6dd1 FZ |
1294 | if (!qiov) { |
1295 | ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req); | |
1296 | goto out; | |
1297 | } | |
1298 | ||
61007b31 SH |
1299 | if (offset & (align - 1)) { |
1300 | QEMUIOVector head_qiov; | |
1301 | struct iovec head_iov; | |
1302 | ||
1303 | mark_request_serialising(&req, align); | |
1304 | wait_serialising_requests(&req); | |
1305 | ||
1306 | head_buf = qemu_blockalign(bs, align); | |
1307 | head_iov = (struct iovec) { | |
1308 | .iov_base = head_buf, | |
1309 | .iov_len = align, | |
1310 | }; | |
1311 | qemu_iovec_init_external(&head_qiov, &head_iov, 1); | |
1312 | ||
9a4f4c31 | 1313 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
61007b31 SH |
1314 | ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align, |
1315 | align, &head_qiov, 0); | |
1316 | if (ret < 0) { | |
1317 | goto fail; | |
1318 | } | |
9a4f4c31 | 1319 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
61007b31 SH |
1320 | |
1321 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1322 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1323 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1324 | use_local_qiov = true; | |
1325 | ||
1326 | bytes += offset & (align - 1); | |
1327 | offset = offset & ~(align - 1); | |
1328 | } | |
1329 | ||
1330 | if ((offset + bytes) & (align - 1)) { | |
1331 | QEMUIOVector tail_qiov; | |
1332 | struct iovec tail_iov; | |
1333 | size_t tail_bytes; | |
1334 | bool waited; | |
1335 | ||
1336 | mark_request_serialising(&req, align); | |
1337 | waited = wait_serialising_requests(&req); | |
1338 | assert(!waited || !use_local_qiov); | |
1339 | ||
1340 | tail_buf = qemu_blockalign(bs, align); | |
1341 | tail_iov = (struct iovec) { | |
1342 | .iov_base = tail_buf, | |
1343 | .iov_len = align, | |
1344 | }; | |
1345 | qemu_iovec_init_external(&tail_qiov, &tail_iov, 1); | |
1346 | ||
9a4f4c31 | 1347 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
61007b31 SH |
1348 | ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align, |
1349 | align, &tail_qiov, 0); | |
1350 | if (ret < 0) { | |
1351 | goto fail; | |
1352 | } | |
9a4f4c31 | 1353 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
61007b31 SH |
1354 | |
1355 | if (!use_local_qiov) { | |
1356 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1357 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1358 | use_local_qiov = true; | |
1359 | } | |
1360 | ||
1361 | tail_bytes = (offset + bytes) & (align - 1); | |
1362 | qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes); | |
1363 | ||
1364 | bytes = ROUND_UP(bytes, align); | |
1365 | } | |
1366 | ||
61007b31 SH |
1367 | ret = bdrv_aligned_pwritev(bs, &req, offset, bytes, |
1368 | use_local_qiov ? &local_qiov : qiov, | |
1369 | flags); | |
1370 | ||
1371 | fail: | |
61007b31 SH |
1372 | |
1373 | if (use_local_qiov) { | |
1374 | qemu_iovec_destroy(&local_qiov); | |
1375 | } | |
1376 | qemu_vfree(head_buf); | |
1377 | qemu_vfree(tail_buf); | |
9eeb6dd1 FZ |
1378 | out: |
1379 | tracked_request_end(&req); | |
61007b31 SH |
1380 | return ret; |
1381 | } | |
1382 | ||
1383 | static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, | |
1384 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
1385 | BdrvRequestFlags flags) | |
1386 | { | |
1387 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
1388 | return -EINVAL; | |
1389 | } | |
1390 | ||
1391 | return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS, | |
1392 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); | |
1393 | } | |
1394 | ||
1395 | int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num, | |
1396 | int nb_sectors, QEMUIOVector *qiov) | |
1397 | { | |
1398 | trace_bdrv_co_writev(bs, sector_num, nb_sectors); | |
1399 | ||
1400 | return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0); | |
1401 | } | |
1402 | ||
1403 | int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs, | |
1404 | int64_t sector_num, int nb_sectors, | |
1405 | BdrvRequestFlags flags) | |
1406 | { | |
61007b31 SH |
1407 | trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags); |
1408 | ||
1409 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
1410 | flags &= ~BDRV_REQ_MAY_UNMAP; | |
1411 | } | |
61007b31 | 1412 | |
d01c07f2 FZ |
1413 | return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL, |
1414 | BDRV_REQ_ZERO_WRITE | flags); | |
61007b31 SH |
1415 | } |
1416 | ||
1417 | int bdrv_flush_all(void) | |
1418 | { | |
1419 | BlockDriverState *bs = NULL; | |
1420 | int result = 0; | |
1421 | ||
1422 | while ((bs = bdrv_next(bs))) { | |
1423 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
1424 | int ret; | |
1425 | ||
1426 | aio_context_acquire(aio_context); | |
1427 | ret = bdrv_flush(bs); | |
1428 | if (ret < 0 && !result) { | |
1429 | result = ret; | |
1430 | } | |
1431 | aio_context_release(aio_context); | |
1432 | } | |
1433 | ||
1434 | return result; | |
1435 | } | |
1436 | ||
1437 | typedef struct BdrvCoGetBlockStatusData { | |
1438 | BlockDriverState *bs; | |
1439 | BlockDriverState *base; | |
1440 | int64_t sector_num; | |
1441 | int nb_sectors; | |
1442 | int *pnum; | |
1443 | int64_t ret; | |
1444 | bool done; | |
1445 | } BdrvCoGetBlockStatusData; | |
1446 | ||
1447 | /* | |
1448 | * Returns the allocation status of the specified sectors. | |
1449 | * Drivers not implementing the functionality are assumed to not support | |
1450 | * backing files, hence all their sectors are reported as allocated. | |
1451 | * | |
1452 | * If 'sector_num' is beyond the end of the disk image the return value is 0 | |
1453 | * and 'pnum' is set to 0. | |
1454 | * | |
1455 | * 'pnum' is set to the number of sectors (including and immediately following | |
1456 | * the specified sector) that are known to be in the same | |
1457 | * allocated/unallocated state. | |
1458 | * | |
1459 | * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes | |
1460 | * beyond the end of the disk image it will be clamped. | |
1461 | */ | |
1462 | static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, | |
1463 | int64_t sector_num, | |
1464 | int nb_sectors, int *pnum) | |
1465 | { | |
1466 | int64_t total_sectors; | |
1467 | int64_t n; | |
1468 | int64_t ret, ret2; | |
1469 | ||
1470 | total_sectors = bdrv_nb_sectors(bs); | |
1471 | if (total_sectors < 0) { | |
1472 | return total_sectors; | |
1473 | } | |
1474 | ||
1475 | if (sector_num >= total_sectors) { | |
1476 | *pnum = 0; | |
1477 | return 0; | |
1478 | } | |
1479 | ||
1480 | n = total_sectors - sector_num; | |
1481 | if (n < nb_sectors) { | |
1482 | nb_sectors = n; | |
1483 | } | |
1484 | ||
1485 | if (!bs->drv->bdrv_co_get_block_status) { | |
1486 | *pnum = nb_sectors; | |
1487 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; | |
1488 | if (bs->drv->protocol_name) { | |
1489 | ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE); | |
1490 | } | |
1491 | return ret; | |
1492 | } | |
1493 | ||
1494 | ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum); | |
1495 | if (ret < 0) { | |
1496 | *pnum = 0; | |
1497 | return ret; | |
1498 | } | |
1499 | ||
1500 | if (ret & BDRV_BLOCK_RAW) { | |
1501 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
9a4f4c31 | 1502 | return bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS, |
61007b31 SH |
1503 | *pnum, pnum); |
1504 | } | |
1505 | ||
1506 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
1507 | ret |= BDRV_BLOCK_ALLOCATED; | |
a53f1a95 | 1508 | } else { |
61007b31 SH |
1509 | if (bdrv_unallocated_blocks_are_zero(bs)) { |
1510 | ret |= BDRV_BLOCK_ZERO; | |
760e0063 KW |
1511 | } else if (bs->backing) { |
1512 | BlockDriverState *bs2 = bs->backing->bs; | |
61007b31 SH |
1513 | int64_t nb_sectors2 = bdrv_nb_sectors(bs2); |
1514 | if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) { | |
1515 | ret |= BDRV_BLOCK_ZERO; | |
1516 | } | |
1517 | } | |
1518 | } | |
1519 | ||
1520 | if (bs->file && | |
1521 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && | |
1522 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
1523 | int file_pnum; | |
1524 | ||
9a4f4c31 | 1525 | ret2 = bdrv_co_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS, |
61007b31 SH |
1526 | *pnum, &file_pnum); |
1527 | if (ret2 >= 0) { | |
1528 | /* Ignore errors. This is just providing extra information, it | |
1529 | * is useful but not necessary. | |
1530 | */ | |
1531 | if (!file_pnum) { | |
1532 | /* !file_pnum indicates an offset at or beyond the EOF; it is | |
1533 | * perfectly valid for the format block driver to point to such | |
1534 | * offsets, so catch it and mark everything as zero */ | |
1535 | ret |= BDRV_BLOCK_ZERO; | |
1536 | } else { | |
1537 | /* Limit request to the range reported by the protocol driver */ | |
1538 | *pnum = file_pnum; | |
1539 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
1540 | } | |
1541 | } | |
1542 | } | |
1543 | ||
1544 | return ret; | |
1545 | } | |
1546 | ||
ba3f0e25 FZ |
1547 | static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs, |
1548 | BlockDriverState *base, | |
1549 | int64_t sector_num, | |
1550 | int nb_sectors, | |
1551 | int *pnum) | |
1552 | { | |
1553 | BlockDriverState *p; | |
1554 | int64_t ret = 0; | |
1555 | ||
1556 | assert(bs != base); | |
760e0063 | 1557 | for (p = bs; p != base; p = backing_bs(p)) { |
ba3f0e25 FZ |
1558 | ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum); |
1559 | if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) { | |
1560 | break; | |
1561 | } | |
1562 | /* [sector_num, pnum] unallocated on this layer, which could be only | |
1563 | * the first part of [sector_num, nb_sectors]. */ | |
1564 | nb_sectors = MIN(nb_sectors, *pnum); | |
1565 | } | |
1566 | return ret; | |
1567 | } | |
1568 | ||
1569 | /* Coroutine wrapper for bdrv_get_block_status_above() */ | |
1570 | static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque) | |
61007b31 SH |
1571 | { |
1572 | BdrvCoGetBlockStatusData *data = opaque; | |
61007b31 | 1573 | |
ba3f0e25 FZ |
1574 | data->ret = bdrv_co_get_block_status_above(data->bs, data->base, |
1575 | data->sector_num, | |
1576 | data->nb_sectors, | |
1577 | data->pnum); | |
61007b31 SH |
1578 | data->done = true; |
1579 | } | |
1580 | ||
1581 | /* | |
ba3f0e25 | 1582 | * Synchronous wrapper around bdrv_co_get_block_status_above(). |
61007b31 | 1583 | * |
ba3f0e25 | 1584 | * See bdrv_co_get_block_status_above() for details. |
61007b31 | 1585 | */ |
ba3f0e25 FZ |
1586 | int64_t bdrv_get_block_status_above(BlockDriverState *bs, |
1587 | BlockDriverState *base, | |
1588 | int64_t sector_num, | |
1589 | int nb_sectors, int *pnum) | |
61007b31 SH |
1590 | { |
1591 | Coroutine *co; | |
1592 | BdrvCoGetBlockStatusData data = { | |
1593 | .bs = bs, | |
ba3f0e25 | 1594 | .base = base, |
61007b31 SH |
1595 | .sector_num = sector_num, |
1596 | .nb_sectors = nb_sectors, | |
1597 | .pnum = pnum, | |
1598 | .done = false, | |
1599 | }; | |
1600 | ||
1601 | if (qemu_in_coroutine()) { | |
1602 | /* Fast-path if already in coroutine context */ | |
ba3f0e25 | 1603 | bdrv_get_block_status_above_co_entry(&data); |
61007b31 SH |
1604 | } else { |
1605 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
1606 | ||
ba3f0e25 | 1607 | co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry); |
61007b31 SH |
1608 | qemu_coroutine_enter(co, &data); |
1609 | while (!data.done) { | |
1610 | aio_poll(aio_context, true); | |
1611 | } | |
1612 | } | |
1613 | return data.ret; | |
1614 | } | |
1615 | ||
ba3f0e25 FZ |
1616 | int64_t bdrv_get_block_status(BlockDriverState *bs, |
1617 | int64_t sector_num, | |
1618 | int nb_sectors, int *pnum) | |
1619 | { | |
760e0063 | 1620 | return bdrv_get_block_status_above(bs, backing_bs(bs), |
ba3f0e25 FZ |
1621 | sector_num, nb_sectors, pnum); |
1622 | } | |
1623 | ||
61007b31 SH |
1624 | int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, |
1625 | int nb_sectors, int *pnum) | |
1626 | { | |
1627 | int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum); | |
1628 | if (ret < 0) { | |
1629 | return ret; | |
1630 | } | |
1631 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
1632 | } | |
1633 | ||
1634 | /* | |
1635 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
1636 | * | |
1637 | * Return true if the given sector is allocated in any image between | |
1638 | * BASE and TOP (inclusive). BASE can be NULL to check if the given | |
1639 | * sector is allocated in any image of the chain. Return false otherwise. | |
1640 | * | |
1641 | * 'pnum' is set to the number of sectors (including and immediately following | |
1642 | * the specified sector) that are known to be in the same | |
1643 | * allocated/unallocated state. | |
1644 | * | |
1645 | */ | |
1646 | int bdrv_is_allocated_above(BlockDriverState *top, | |
1647 | BlockDriverState *base, | |
1648 | int64_t sector_num, | |
1649 | int nb_sectors, int *pnum) | |
1650 | { | |
1651 | BlockDriverState *intermediate; | |
1652 | int ret, n = nb_sectors; | |
1653 | ||
1654 | intermediate = top; | |
1655 | while (intermediate && intermediate != base) { | |
1656 | int pnum_inter; | |
1657 | ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors, | |
1658 | &pnum_inter); | |
1659 | if (ret < 0) { | |
1660 | return ret; | |
1661 | } else if (ret) { | |
1662 | *pnum = pnum_inter; | |
1663 | return 1; | |
1664 | } | |
1665 | ||
1666 | /* | |
1667 | * [sector_num, nb_sectors] is unallocated on top but intermediate | |
1668 | * might have | |
1669 | * | |
1670 | * [sector_num+x, nr_sectors] allocated. | |
1671 | */ | |
1672 | if (n > pnum_inter && | |
1673 | (intermediate == top || | |
1674 | sector_num + pnum_inter < intermediate->total_sectors)) { | |
1675 | n = pnum_inter; | |
1676 | } | |
1677 | ||
760e0063 | 1678 | intermediate = backing_bs(intermediate); |
61007b31 SH |
1679 | } |
1680 | ||
1681 | *pnum = n; | |
1682 | return 0; | |
1683 | } | |
1684 | ||
1685 | int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num, | |
1686 | const uint8_t *buf, int nb_sectors) | |
1687 | { | |
1688 | BlockDriver *drv = bs->drv; | |
1689 | int ret; | |
1690 | ||
1691 | if (!drv) { | |
1692 | return -ENOMEDIUM; | |
1693 | } | |
1694 | if (!drv->bdrv_write_compressed) { | |
1695 | return -ENOTSUP; | |
1696 | } | |
1697 | ret = bdrv_check_request(bs, sector_num, nb_sectors); | |
1698 | if (ret < 0) { | |
1699 | return ret; | |
1700 | } | |
1701 | ||
1702 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
1703 | ||
1704 | return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors); | |
1705 | } | |
1706 | ||
1707 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, | |
1708 | int64_t pos, int size) | |
1709 | { | |
1710 | QEMUIOVector qiov; | |
1711 | struct iovec iov = { | |
1712 | .iov_base = (void *) buf, | |
1713 | .iov_len = size, | |
1714 | }; | |
1715 | ||
1716 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1717 | return bdrv_writev_vmstate(bs, &qiov, pos); | |
1718 | } | |
1719 | ||
1720 | int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
1721 | { | |
1722 | BlockDriver *drv = bs->drv; | |
1723 | ||
1724 | if (!drv) { | |
1725 | return -ENOMEDIUM; | |
1726 | } else if (drv->bdrv_save_vmstate) { | |
1727 | return drv->bdrv_save_vmstate(bs, qiov, pos); | |
1728 | } else if (bs->file) { | |
9a4f4c31 | 1729 | return bdrv_writev_vmstate(bs->file->bs, qiov, pos); |
61007b31 SH |
1730 | } |
1731 | ||
1732 | return -ENOTSUP; | |
1733 | } | |
1734 | ||
1735 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
1736 | int64_t pos, int size) | |
1737 | { | |
1738 | BlockDriver *drv = bs->drv; | |
1739 | if (!drv) | |
1740 | return -ENOMEDIUM; | |
1741 | if (drv->bdrv_load_vmstate) | |
1742 | return drv->bdrv_load_vmstate(bs, buf, pos, size); | |
1743 | if (bs->file) | |
9a4f4c31 | 1744 | return bdrv_load_vmstate(bs->file->bs, buf, pos, size); |
61007b31 SH |
1745 | return -ENOTSUP; |
1746 | } | |
1747 | ||
1748 | /**************************************************************/ | |
1749 | /* async I/Os */ | |
1750 | ||
1751 | BlockAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num, | |
1752 | QEMUIOVector *qiov, int nb_sectors, | |
1753 | BlockCompletionFunc *cb, void *opaque) | |
1754 | { | |
1755 | trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque); | |
1756 | ||
1757 | return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0, | |
1758 | cb, opaque, false); | |
1759 | } | |
1760 | ||
1761 | BlockAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num, | |
1762 | QEMUIOVector *qiov, int nb_sectors, | |
1763 | BlockCompletionFunc *cb, void *opaque) | |
1764 | { | |
1765 | trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque); | |
1766 | ||
1767 | return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0, | |
1768 | cb, opaque, true); | |
1769 | } | |
1770 | ||
1771 | BlockAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs, | |
1772 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags, | |
1773 | BlockCompletionFunc *cb, void *opaque) | |
1774 | { | |
1775 | trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque); | |
1776 | ||
1777 | return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors, | |
1778 | BDRV_REQ_ZERO_WRITE | flags, | |
1779 | cb, opaque, true); | |
1780 | } | |
1781 | ||
1782 | ||
1783 | typedef struct MultiwriteCB { | |
1784 | int error; | |
1785 | int num_requests; | |
1786 | int num_callbacks; | |
1787 | struct { | |
1788 | BlockCompletionFunc *cb; | |
1789 | void *opaque; | |
1790 | QEMUIOVector *free_qiov; | |
1791 | } callbacks[]; | |
1792 | } MultiwriteCB; | |
1793 | ||
1794 | static void multiwrite_user_cb(MultiwriteCB *mcb) | |
1795 | { | |
1796 | int i; | |
1797 | ||
1798 | for (i = 0; i < mcb->num_callbacks; i++) { | |
1799 | mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error); | |
1800 | if (mcb->callbacks[i].free_qiov) { | |
1801 | qemu_iovec_destroy(mcb->callbacks[i].free_qiov); | |
1802 | } | |
1803 | g_free(mcb->callbacks[i].free_qiov); | |
1804 | } | |
1805 | } | |
1806 | ||
1807 | static void multiwrite_cb(void *opaque, int ret) | |
1808 | { | |
1809 | MultiwriteCB *mcb = opaque; | |
1810 | ||
1811 | trace_multiwrite_cb(mcb, ret); | |
1812 | ||
1813 | if (ret < 0 && !mcb->error) { | |
1814 | mcb->error = ret; | |
1815 | } | |
1816 | ||
1817 | mcb->num_requests--; | |
1818 | if (mcb->num_requests == 0) { | |
1819 | multiwrite_user_cb(mcb); | |
1820 | g_free(mcb); | |
1821 | } | |
1822 | } | |
1823 | ||
1824 | static int multiwrite_req_compare(const void *a, const void *b) | |
1825 | { | |
1826 | const BlockRequest *req1 = a, *req2 = b; | |
1827 | ||
1828 | /* | |
1829 | * Note that we can't simply subtract req2->sector from req1->sector | |
1830 | * here as that could overflow the return value. | |
1831 | */ | |
1832 | if (req1->sector > req2->sector) { | |
1833 | return 1; | |
1834 | } else if (req1->sector < req2->sector) { | |
1835 | return -1; | |
1836 | } else { | |
1837 | return 0; | |
1838 | } | |
1839 | } | |
1840 | ||
1841 | /* | |
1842 | * Takes a bunch of requests and tries to merge them. Returns the number of | |
1843 | * requests that remain after merging. | |
1844 | */ | |
1845 | static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs, | |
1846 | int num_reqs, MultiwriteCB *mcb) | |
1847 | { | |
1848 | int i, outidx; | |
1849 | ||
1850 | // Sort requests by start sector | |
1851 | qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare); | |
1852 | ||
1853 | // Check if adjacent requests touch the same clusters. If so, combine them, | |
1854 | // filling up gaps with zero sectors. | |
1855 | outidx = 0; | |
1856 | for (i = 1; i < num_reqs; i++) { | |
1857 | int merge = 0; | |
1858 | int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors; | |
1859 | ||
1860 | // Handle exactly sequential writes and overlapping writes. | |
1861 | if (reqs[i].sector <= oldreq_last) { | |
1862 | merge = 1; | |
1863 | } | |
1864 | ||
1865 | if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) { | |
1866 | merge = 0; | |
1867 | } | |
1868 | ||
1869 | if (bs->bl.max_transfer_length && reqs[outidx].nb_sectors + | |
1870 | reqs[i].nb_sectors > bs->bl.max_transfer_length) { | |
1871 | merge = 0; | |
1872 | } | |
1873 | ||
1874 | if (merge) { | |
1875 | size_t size; | |
1876 | QEMUIOVector *qiov = g_malloc0(sizeof(*qiov)); | |
1877 | qemu_iovec_init(qiov, | |
1878 | reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1); | |
1879 | ||
1880 | // Add the first request to the merged one. If the requests are | |
1881 | // overlapping, drop the last sectors of the first request. | |
1882 | size = (reqs[i].sector - reqs[outidx].sector) << 9; | |
1883 | qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size); | |
1884 | ||
1885 | // We should need to add any zeros between the two requests | |
1886 | assert (reqs[i].sector <= oldreq_last); | |
1887 | ||
1888 | // Add the second request | |
1889 | qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size); | |
1890 | ||
1891 | // Add tail of first request, if necessary | |
1892 | if (qiov->size < reqs[outidx].qiov->size) { | |
1893 | qemu_iovec_concat(qiov, reqs[outidx].qiov, qiov->size, | |
1894 | reqs[outidx].qiov->size - qiov->size); | |
1895 | } | |
1896 | ||
1897 | reqs[outidx].nb_sectors = qiov->size >> 9; | |
1898 | reqs[outidx].qiov = qiov; | |
1899 | ||
1900 | mcb->callbacks[i].free_qiov = reqs[outidx].qiov; | |
1901 | } else { | |
1902 | outidx++; | |
1903 | reqs[outidx].sector = reqs[i].sector; | |
1904 | reqs[outidx].nb_sectors = reqs[i].nb_sectors; | |
1905 | reqs[outidx].qiov = reqs[i].qiov; | |
1906 | } | |
1907 | } | |
1908 | ||
7f0e9da6 HR |
1909 | if (bs->blk) { |
1910 | block_acct_merge_done(blk_get_stats(bs->blk), BLOCK_ACCT_WRITE, | |
1911 | num_reqs - outidx - 1); | |
1912 | } | |
61007b31 SH |
1913 | |
1914 | return outidx + 1; | |
1915 | } | |
1916 | ||
1917 | /* | |
1918 | * Submit multiple AIO write requests at once. | |
1919 | * | |
1920 | * On success, the function returns 0 and all requests in the reqs array have | |
1921 | * been submitted. In error case this function returns -1, and any of the | |
1922 | * requests may or may not be submitted yet. In particular, this means that the | |
1923 | * callback will be called for some of the requests, for others it won't. The | |
1924 | * caller must check the error field of the BlockRequest to wait for the right | |
1925 | * callbacks (if error != 0, no callback will be called). | |
1926 | * | |
1927 | * The implementation may modify the contents of the reqs array, e.g. to merge | |
1928 | * requests. However, the fields opaque and error are left unmodified as they | |
1929 | * are used to signal failure for a single request to the caller. | |
1930 | */ | |
1931 | int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs) | |
1932 | { | |
1933 | MultiwriteCB *mcb; | |
1934 | int i; | |
1935 | ||
1936 | /* don't submit writes if we don't have a medium */ | |
1937 | if (bs->drv == NULL) { | |
1938 | for (i = 0; i < num_reqs; i++) { | |
1939 | reqs[i].error = -ENOMEDIUM; | |
1940 | } | |
1941 | return -1; | |
1942 | } | |
1943 | ||
1944 | if (num_reqs == 0) { | |
1945 | return 0; | |
1946 | } | |
1947 | ||
1948 | // Create MultiwriteCB structure | |
1949 | mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks)); | |
1950 | mcb->num_requests = 0; | |
1951 | mcb->num_callbacks = num_reqs; | |
1952 | ||
1953 | for (i = 0; i < num_reqs; i++) { | |
1954 | mcb->callbacks[i].cb = reqs[i].cb; | |
1955 | mcb->callbacks[i].opaque = reqs[i].opaque; | |
1956 | } | |
1957 | ||
1958 | // Check for mergable requests | |
1959 | num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb); | |
1960 | ||
1961 | trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs); | |
1962 | ||
1963 | /* Run the aio requests. */ | |
1964 | mcb->num_requests = num_reqs; | |
1965 | for (i = 0; i < num_reqs; i++) { | |
1966 | bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov, | |
1967 | reqs[i].nb_sectors, reqs[i].flags, | |
1968 | multiwrite_cb, mcb, | |
1969 | true); | |
1970 | } | |
1971 | ||
1972 | return 0; | |
1973 | } | |
1974 | ||
1975 | void bdrv_aio_cancel(BlockAIOCB *acb) | |
1976 | { | |
1977 | qemu_aio_ref(acb); | |
1978 | bdrv_aio_cancel_async(acb); | |
1979 | while (acb->refcnt > 1) { | |
1980 | if (acb->aiocb_info->get_aio_context) { | |
1981 | aio_poll(acb->aiocb_info->get_aio_context(acb), true); | |
1982 | } else if (acb->bs) { | |
1983 | aio_poll(bdrv_get_aio_context(acb->bs), true); | |
1984 | } else { | |
1985 | abort(); | |
1986 | } | |
1987 | } | |
1988 | qemu_aio_unref(acb); | |
1989 | } | |
1990 | ||
1991 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
1992 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
1993 | * In either case the completion callback must be called. */ | |
1994 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
1995 | { | |
1996 | if (acb->aiocb_info->cancel_async) { | |
1997 | acb->aiocb_info->cancel_async(acb); | |
1998 | } | |
1999 | } | |
2000 | ||
2001 | /**************************************************************/ | |
2002 | /* async block device emulation */ | |
2003 | ||
2004 | typedef struct BlockAIOCBSync { | |
2005 | BlockAIOCB common; | |
2006 | QEMUBH *bh; | |
2007 | int ret; | |
2008 | /* vector translation state */ | |
2009 | QEMUIOVector *qiov; | |
2010 | uint8_t *bounce; | |
2011 | int is_write; | |
2012 | } BlockAIOCBSync; | |
2013 | ||
2014 | static const AIOCBInfo bdrv_em_aiocb_info = { | |
2015 | .aiocb_size = sizeof(BlockAIOCBSync), | |
2016 | }; | |
2017 | ||
2018 | static void bdrv_aio_bh_cb(void *opaque) | |
2019 | { | |
2020 | BlockAIOCBSync *acb = opaque; | |
2021 | ||
2022 | if (!acb->is_write && acb->ret >= 0) { | |
2023 | qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size); | |
2024 | } | |
2025 | qemu_vfree(acb->bounce); | |
2026 | acb->common.cb(acb->common.opaque, acb->ret); | |
2027 | qemu_bh_delete(acb->bh); | |
2028 | acb->bh = NULL; | |
2029 | qemu_aio_unref(acb); | |
2030 | } | |
2031 | ||
2032 | static BlockAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs, | |
2033 | int64_t sector_num, | |
2034 | QEMUIOVector *qiov, | |
2035 | int nb_sectors, | |
2036 | BlockCompletionFunc *cb, | |
2037 | void *opaque, | |
2038 | int is_write) | |
2039 | ||
2040 | { | |
2041 | BlockAIOCBSync *acb; | |
2042 | ||
2043 | acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque); | |
2044 | acb->is_write = is_write; | |
2045 | acb->qiov = qiov; | |
2046 | acb->bounce = qemu_try_blockalign(bs, qiov->size); | |
2047 | acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_aio_bh_cb, acb); | |
2048 | ||
2049 | if (acb->bounce == NULL) { | |
2050 | acb->ret = -ENOMEM; | |
2051 | } else if (is_write) { | |
2052 | qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size); | |
2053 | acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors); | |
2054 | } else { | |
2055 | acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors); | |
2056 | } | |
2057 | ||
2058 | qemu_bh_schedule(acb->bh); | |
2059 | ||
2060 | return &acb->common; | |
2061 | } | |
2062 | ||
2063 | static BlockAIOCB *bdrv_aio_readv_em(BlockDriverState *bs, | |
2064 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
2065 | BlockCompletionFunc *cb, void *opaque) | |
2066 | { | |
2067 | return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); | |
2068 | } | |
2069 | ||
2070 | static BlockAIOCB *bdrv_aio_writev_em(BlockDriverState *bs, | |
2071 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
2072 | BlockCompletionFunc *cb, void *opaque) | |
2073 | { | |
2074 | return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); | |
2075 | } | |
2076 | ||
2077 | ||
2078 | typedef struct BlockAIOCBCoroutine { | |
2079 | BlockAIOCB common; | |
2080 | BlockRequest req; | |
2081 | bool is_write; | |
2082 | bool need_bh; | |
2083 | bool *done; | |
2084 | QEMUBH* bh; | |
2085 | } BlockAIOCBCoroutine; | |
2086 | ||
2087 | static const AIOCBInfo bdrv_em_co_aiocb_info = { | |
2088 | .aiocb_size = sizeof(BlockAIOCBCoroutine), | |
2089 | }; | |
2090 | ||
2091 | static void bdrv_co_complete(BlockAIOCBCoroutine *acb) | |
2092 | { | |
2093 | if (!acb->need_bh) { | |
2094 | acb->common.cb(acb->common.opaque, acb->req.error); | |
2095 | qemu_aio_unref(acb); | |
2096 | } | |
2097 | } | |
2098 | ||
2099 | static void bdrv_co_em_bh(void *opaque) | |
2100 | { | |
2101 | BlockAIOCBCoroutine *acb = opaque; | |
2102 | ||
2103 | assert(!acb->need_bh); | |
2104 | qemu_bh_delete(acb->bh); | |
2105 | bdrv_co_complete(acb); | |
2106 | } | |
2107 | ||
2108 | static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb) | |
2109 | { | |
2110 | acb->need_bh = false; | |
2111 | if (acb->req.error != -EINPROGRESS) { | |
2112 | BlockDriverState *bs = acb->common.bs; | |
2113 | ||
2114 | acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb); | |
2115 | qemu_bh_schedule(acb->bh); | |
2116 | } | |
2117 | } | |
2118 | ||
2119 | /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */ | |
2120 | static void coroutine_fn bdrv_co_do_rw(void *opaque) | |
2121 | { | |
2122 | BlockAIOCBCoroutine *acb = opaque; | |
2123 | BlockDriverState *bs = acb->common.bs; | |
2124 | ||
2125 | if (!acb->is_write) { | |
2126 | acb->req.error = bdrv_co_do_readv(bs, acb->req.sector, | |
2127 | acb->req.nb_sectors, acb->req.qiov, acb->req.flags); | |
2128 | } else { | |
2129 | acb->req.error = bdrv_co_do_writev(bs, acb->req.sector, | |
2130 | acb->req.nb_sectors, acb->req.qiov, acb->req.flags); | |
2131 | } | |
2132 | ||
2133 | bdrv_co_complete(acb); | |
2134 | } | |
2135 | ||
2136 | static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, | |
2137 | int64_t sector_num, | |
2138 | QEMUIOVector *qiov, | |
2139 | int nb_sectors, | |
2140 | BdrvRequestFlags flags, | |
2141 | BlockCompletionFunc *cb, | |
2142 | void *opaque, | |
2143 | bool is_write) | |
2144 | { | |
2145 | Coroutine *co; | |
2146 | BlockAIOCBCoroutine *acb; | |
2147 | ||
2148 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
2149 | acb->need_bh = true; | |
2150 | acb->req.error = -EINPROGRESS; | |
2151 | acb->req.sector = sector_num; | |
2152 | acb->req.nb_sectors = nb_sectors; | |
2153 | acb->req.qiov = qiov; | |
2154 | acb->req.flags = flags; | |
2155 | acb->is_write = is_write; | |
2156 | ||
2157 | co = qemu_coroutine_create(bdrv_co_do_rw); | |
2158 | qemu_coroutine_enter(co, acb); | |
2159 | ||
2160 | bdrv_co_maybe_schedule_bh(acb); | |
2161 | return &acb->common; | |
2162 | } | |
2163 | ||
2164 | static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) | |
2165 | { | |
2166 | BlockAIOCBCoroutine *acb = opaque; | |
2167 | BlockDriverState *bs = acb->common.bs; | |
2168 | ||
2169 | acb->req.error = bdrv_co_flush(bs); | |
2170 | bdrv_co_complete(acb); | |
2171 | } | |
2172 | ||
2173 | BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs, | |
2174 | BlockCompletionFunc *cb, void *opaque) | |
2175 | { | |
2176 | trace_bdrv_aio_flush(bs, opaque); | |
2177 | ||
2178 | Coroutine *co; | |
2179 | BlockAIOCBCoroutine *acb; | |
2180 | ||
2181 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
2182 | acb->need_bh = true; | |
2183 | acb->req.error = -EINPROGRESS; | |
2184 | ||
2185 | co = qemu_coroutine_create(bdrv_aio_flush_co_entry); | |
2186 | qemu_coroutine_enter(co, acb); | |
2187 | ||
2188 | bdrv_co_maybe_schedule_bh(acb); | |
2189 | return &acb->common; | |
2190 | } | |
2191 | ||
2192 | static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque) | |
2193 | { | |
2194 | BlockAIOCBCoroutine *acb = opaque; | |
2195 | BlockDriverState *bs = acb->common.bs; | |
2196 | ||
2197 | acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors); | |
2198 | bdrv_co_complete(acb); | |
2199 | } | |
2200 | ||
2201 | BlockAIOCB *bdrv_aio_discard(BlockDriverState *bs, | |
2202 | int64_t sector_num, int nb_sectors, | |
2203 | BlockCompletionFunc *cb, void *opaque) | |
2204 | { | |
2205 | Coroutine *co; | |
2206 | BlockAIOCBCoroutine *acb; | |
2207 | ||
2208 | trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque); | |
2209 | ||
2210 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
2211 | acb->need_bh = true; | |
2212 | acb->req.error = -EINPROGRESS; | |
2213 | acb->req.sector = sector_num; | |
2214 | acb->req.nb_sectors = nb_sectors; | |
2215 | co = qemu_coroutine_create(bdrv_aio_discard_co_entry); | |
2216 | qemu_coroutine_enter(co, acb); | |
2217 | ||
2218 | bdrv_co_maybe_schedule_bh(acb); | |
2219 | return &acb->common; | |
2220 | } | |
2221 | ||
2222 | void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, | |
2223 | BlockCompletionFunc *cb, void *opaque) | |
2224 | { | |
2225 | BlockAIOCB *acb; | |
2226 | ||
c84b3192 | 2227 | acb = g_malloc(aiocb_info->aiocb_size); |
61007b31 SH |
2228 | acb->aiocb_info = aiocb_info; |
2229 | acb->bs = bs; | |
2230 | acb->cb = cb; | |
2231 | acb->opaque = opaque; | |
2232 | acb->refcnt = 1; | |
2233 | return acb; | |
2234 | } | |
2235 | ||
2236 | void qemu_aio_ref(void *p) | |
2237 | { | |
2238 | BlockAIOCB *acb = p; | |
2239 | acb->refcnt++; | |
2240 | } | |
2241 | ||
2242 | void qemu_aio_unref(void *p) | |
2243 | { | |
2244 | BlockAIOCB *acb = p; | |
2245 | assert(acb->refcnt > 0); | |
2246 | if (--acb->refcnt == 0) { | |
c84b3192 | 2247 | g_free(acb); |
61007b31 SH |
2248 | } |
2249 | } | |
2250 | ||
2251 | /**************************************************************/ | |
2252 | /* Coroutine block device emulation */ | |
2253 | ||
2254 | typedef struct CoroutineIOCompletion { | |
2255 | Coroutine *coroutine; | |
2256 | int ret; | |
2257 | } CoroutineIOCompletion; | |
2258 | ||
2259 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
2260 | { | |
2261 | CoroutineIOCompletion *co = opaque; | |
2262 | ||
2263 | co->ret = ret; | |
2264 | qemu_coroutine_enter(co->coroutine, NULL); | |
2265 | } | |
2266 | ||
2267 | static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num, | |
2268 | int nb_sectors, QEMUIOVector *iov, | |
2269 | bool is_write) | |
2270 | { | |
2271 | CoroutineIOCompletion co = { | |
2272 | .coroutine = qemu_coroutine_self(), | |
2273 | }; | |
2274 | BlockAIOCB *acb; | |
2275 | ||
2276 | if (is_write) { | |
2277 | acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors, | |
2278 | bdrv_co_io_em_complete, &co); | |
2279 | } else { | |
2280 | acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors, | |
2281 | bdrv_co_io_em_complete, &co); | |
2282 | } | |
2283 | ||
2284 | trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb); | |
2285 | if (!acb) { | |
2286 | return -EIO; | |
2287 | } | |
2288 | qemu_coroutine_yield(); | |
2289 | ||
2290 | return co.ret; | |
2291 | } | |
2292 | ||
2293 | static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, | |
2294 | int64_t sector_num, int nb_sectors, | |
2295 | QEMUIOVector *iov) | |
2296 | { | |
2297 | return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false); | |
2298 | } | |
2299 | ||
2300 | static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, | |
2301 | int64_t sector_num, int nb_sectors, | |
2302 | QEMUIOVector *iov) | |
2303 | { | |
2304 | return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true); | |
2305 | } | |
2306 | ||
2307 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) | |
2308 | { | |
2309 | RwCo *rwco = opaque; | |
2310 | ||
2311 | rwco->ret = bdrv_co_flush(rwco->bs); | |
2312 | } | |
2313 | ||
2314 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
2315 | { | |
2316 | int ret; | |
2317 | ||
1b6bc94d DA |
2318 | if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) || |
2319 | bdrv_is_sg(bs)) { | |
61007b31 SH |
2320 | return 0; |
2321 | } | |
2322 | ||
2323 | /* Write back cached data to the OS even with cache=unsafe */ | |
2324 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); | |
2325 | if (bs->drv->bdrv_co_flush_to_os) { | |
2326 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
2327 | if (ret < 0) { | |
2328 | return ret; | |
2329 | } | |
2330 | } | |
2331 | ||
2332 | /* But don't actually force it to the disk with cache=unsafe */ | |
2333 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
2334 | goto flush_parent; | |
2335 | } | |
2336 | ||
2337 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); | |
2338 | if (bs->drv->bdrv_co_flush_to_disk) { | |
2339 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
2340 | } else if (bs->drv->bdrv_aio_flush) { | |
2341 | BlockAIOCB *acb; | |
2342 | CoroutineIOCompletion co = { | |
2343 | .coroutine = qemu_coroutine_self(), | |
2344 | }; | |
2345 | ||
2346 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
2347 | if (acb == NULL) { | |
2348 | ret = -EIO; | |
2349 | } else { | |
2350 | qemu_coroutine_yield(); | |
2351 | ret = co.ret; | |
2352 | } | |
2353 | } else { | |
2354 | /* | |
2355 | * Some block drivers always operate in either writethrough or unsafe | |
2356 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
2357 | * know how the server works (because the behaviour is hardcoded or | |
2358 | * depends on server-side configuration), so we can't ensure that | |
2359 | * everything is safe on disk. Returning an error doesn't work because | |
2360 | * that would break guests even if the server operates in writethrough | |
2361 | * mode. | |
2362 | * | |
2363 | * Let's hope the user knows what he's doing. | |
2364 | */ | |
2365 | ret = 0; | |
2366 | } | |
2367 | if (ret < 0) { | |
2368 | return ret; | |
2369 | } | |
2370 | ||
2371 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
2372 | * in the case of cache=unsafe, so there are no useless flushes. | |
2373 | */ | |
2374 | flush_parent: | |
9a4f4c31 | 2375 | return bs->file ? bdrv_co_flush(bs->file->bs) : 0; |
61007b31 SH |
2376 | } |
2377 | ||
2378 | int bdrv_flush(BlockDriverState *bs) | |
2379 | { | |
2380 | Coroutine *co; | |
2381 | RwCo rwco = { | |
2382 | .bs = bs, | |
2383 | .ret = NOT_DONE, | |
2384 | }; | |
2385 | ||
2386 | if (qemu_in_coroutine()) { | |
2387 | /* Fast-path if already in coroutine context */ | |
2388 | bdrv_flush_co_entry(&rwco); | |
2389 | } else { | |
2390 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2391 | ||
2392 | co = qemu_coroutine_create(bdrv_flush_co_entry); | |
2393 | qemu_coroutine_enter(co, &rwco); | |
2394 | while (rwco.ret == NOT_DONE) { | |
2395 | aio_poll(aio_context, true); | |
2396 | } | |
2397 | } | |
2398 | ||
2399 | return rwco.ret; | |
2400 | } | |
2401 | ||
2402 | typedef struct DiscardCo { | |
2403 | BlockDriverState *bs; | |
2404 | int64_t sector_num; | |
2405 | int nb_sectors; | |
2406 | int ret; | |
2407 | } DiscardCo; | |
2408 | static void coroutine_fn bdrv_discard_co_entry(void *opaque) | |
2409 | { | |
2410 | DiscardCo *rwco = opaque; | |
2411 | ||
2412 | rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors); | |
2413 | } | |
2414 | ||
2415 | int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num, | |
2416 | int nb_sectors) | |
2417 | { | |
2418 | int max_discard, ret; | |
2419 | ||
2420 | if (!bs->drv) { | |
2421 | return -ENOMEDIUM; | |
2422 | } | |
2423 | ||
2424 | ret = bdrv_check_request(bs, sector_num, nb_sectors); | |
2425 | if (ret < 0) { | |
2426 | return ret; | |
2427 | } else if (bs->read_only) { | |
eaf5fe2d | 2428 | return -EPERM; |
61007b31 SH |
2429 | } |
2430 | ||
61007b31 SH |
2431 | /* Do nothing if disabled. */ |
2432 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
2433 | return 0; | |
2434 | } | |
2435 | ||
2436 | if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) { | |
2437 | return 0; | |
2438 | } | |
2439 | ||
50824995 FZ |
2440 | bdrv_set_dirty(bs, sector_num, nb_sectors); |
2441 | ||
61007b31 SH |
2442 | max_discard = MIN_NON_ZERO(bs->bl.max_discard, BDRV_REQUEST_MAX_SECTORS); |
2443 | while (nb_sectors > 0) { | |
2444 | int ret; | |
2445 | int num = nb_sectors; | |
2446 | ||
2447 | /* align request */ | |
2448 | if (bs->bl.discard_alignment && | |
2449 | num >= bs->bl.discard_alignment && | |
2450 | sector_num % bs->bl.discard_alignment) { | |
2451 | if (num > bs->bl.discard_alignment) { | |
2452 | num = bs->bl.discard_alignment; | |
2453 | } | |
2454 | num -= sector_num % bs->bl.discard_alignment; | |
2455 | } | |
2456 | ||
2457 | /* limit request size */ | |
2458 | if (num > max_discard) { | |
2459 | num = max_discard; | |
2460 | } | |
2461 | ||
2462 | if (bs->drv->bdrv_co_discard) { | |
2463 | ret = bs->drv->bdrv_co_discard(bs, sector_num, num); | |
2464 | } else { | |
2465 | BlockAIOCB *acb; | |
2466 | CoroutineIOCompletion co = { | |
2467 | .coroutine = qemu_coroutine_self(), | |
2468 | }; | |
2469 | ||
2470 | acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors, | |
2471 | bdrv_co_io_em_complete, &co); | |
2472 | if (acb == NULL) { | |
2473 | return -EIO; | |
2474 | } else { | |
2475 | qemu_coroutine_yield(); | |
2476 | ret = co.ret; | |
2477 | } | |
2478 | } | |
2479 | if (ret && ret != -ENOTSUP) { | |
2480 | return ret; | |
2481 | } | |
2482 | ||
2483 | sector_num += num; | |
2484 | nb_sectors -= num; | |
2485 | } | |
2486 | return 0; | |
2487 | } | |
2488 | ||
2489 | int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors) | |
2490 | { | |
2491 | Coroutine *co; | |
2492 | DiscardCo rwco = { | |
2493 | .bs = bs, | |
2494 | .sector_num = sector_num, | |
2495 | .nb_sectors = nb_sectors, | |
2496 | .ret = NOT_DONE, | |
2497 | }; | |
2498 | ||
2499 | if (qemu_in_coroutine()) { | |
2500 | /* Fast-path if already in coroutine context */ | |
2501 | bdrv_discard_co_entry(&rwco); | |
2502 | } else { | |
2503 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2504 | ||
2505 | co = qemu_coroutine_create(bdrv_discard_co_entry); | |
2506 | qemu_coroutine_enter(co, &rwco); | |
2507 | while (rwco.ret == NOT_DONE) { | |
2508 | aio_poll(aio_context, true); | |
2509 | } | |
2510 | } | |
2511 | ||
2512 | return rwco.ret; | |
2513 | } | |
2514 | ||
2515 | /* needed for generic scsi interface */ | |
2516 | ||
2517 | int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
2518 | { | |
2519 | BlockDriver *drv = bs->drv; | |
2520 | ||
2521 | if (drv && drv->bdrv_ioctl) | |
2522 | return drv->bdrv_ioctl(bs, req, buf); | |
2523 | return -ENOTSUP; | |
2524 | } | |
2525 | ||
2526 | BlockAIOCB *bdrv_aio_ioctl(BlockDriverState *bs, | |
2527 | unsigned long int req, void *buf, | |
2528 | BlockCompletionFunc *cb, void *opaque) | |
2529 | { | |
2530 | BlockDriver *drv = bs->drv; | |
2531 | ||
2532 | if (drv && drv->bdrv_aio_ioctl) | |
2533 | return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque); | |
2534 | return NULL; | |
2535 | } | |
2536 | ||
2537 | void *qemu_blockalign(BlockDriverState *bs, size_t size) | |
2538 | { | |
2539 | return qemu_memalign(bdrv_opt_mem_align(bs), size); | |
2540 | } | |
2541 | ||
2542 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
2543 | { | |
2544 | return memset(qemu_blockalign(bs, size), 0, size); | |
2545 | } | |
2546 | ||
2547 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
2548 | { | |
2549 | size_t align = bdrv_opt_mem_align(bs); | |
2550 | ||
2551 | /* Ensure that NULL is never returned on success */ | |
2552 | assert(align > 0); | |
2553 | if (size == 0) { | |
2554 | size = align; | |
2555 | } | |
2556 | ||
2557 | return qemu_try_memalign(align, size); | |
2558 | } | |
2559 | ||
2560 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
2561 | { | |
2562 | void *mem = qemu_try_blockalign(bs, size); | |
2563 | ||
2564 | if (mem) { | |
2565 | memset(mem, 0, size); | |
2566 | } | |
2567 | ||
2568 | return mem; | |
2569 | } | |
2570 | ||
2571 | /* | |
2572 | * Check if all memory in this vector is sector aligned. | |
2573 | */ | |
2574 | bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
2575 | { | |
2576 | int i; | |
4196d2f0 | 2577 | size_t alignment = bdrv_min_mem_align(bs); |
61007b31 SH |
2578 | |
2579 | for (i = 0; i < qiov->niov; i++) { | |
2580 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
2581 | return false; | |
2582 | } | |
2583 | if (qiov->iov[i].iov_len % alignment) { | |
2584 | return false; | |
2585 | } | |
2586 | } | |
2587 | ||
2588 | return true; | |
2589 | } | |
2590 | ||
2591 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
2592 | NotifierWithReturn *notifier) | |
2593 | { | |
2594 | notifier_with_return_list_add(&bs->before_write_notifiers, notifier); | |
2595 | } | |
2596 | ||
2597 | void bdrv_io_plug(BlockDriverState *bs) | |
2598 | { | |
2599 | BlockDriver *drv = bs->drv; | |
2600 | if (drv && drv->bdrv_io_plug) { | |
2601 | drv->bdrv_io_plug(bs); | |
2602 | } else if (bs->file) { | |
9a4f4c31 | 2603 | bdrv_io_plug(bs->file->bs); |
61007b31 SH |
2604 | } |
2605 | } | |
2606 | ||
2607 | void bdrv_io_unplug(BlockDriverState *bs) | |
2608 | { | |
2609 | BlockDriver *drv = bs->drv; | |
2610 | if (drv && drv->bdrv_io_unplug) { | |
2611 | drv->bdrv_io_unplug(bs); | |
2612 | } else if (bs->file) { | |
9a4f4c31 | 2613 | bdrv_io_unplug(bs->file->bs); |
61007b31 SH |
2614 | } |
2615 | } | |
2616 | ||
2617 | void bdrv_flush_io_queue(BlockDriverState *bs) | |
2618 | { | |
2619 | BlockDriver *drv = bs->drv; | |
2620 | if (drv && drv->bdrv_flush_io_queue) { | |
2621 | drv->bdrv_flush_io_queue(bs); | |
2622 | } else if (bs->file) { | |
9a4f4c31 | 2623 | bdrv_flush_io_queue(bs->file->bs); |
61007b31 | 2624 | } |
f406c03c | 2625 | bdrv_start_throttled_reqs(bs); |
61007b31 | 2626 | } |
51288d79 FZ |
2627 | |
2628 | void bdrv_drained_begin(BlockDriverState *bs) | |
2629 | { | |
2630 | if (!bs->quiesce_counter++) { | |
2631 | aio_disable_external(bdrv_get_aio_context(bs)); | |
2632 | } | |
2633 | bdrv_drain(bs); | |
2634 | } | |
2635 | ||
2636 | void bdrv_drained_end(BlockDriverState *bs) | |
2637 | { | |
2638 | assert(bs->quiesce_counter > 0); | |
2639 | if (--bs->quiesce_counter > 0) { | |
2640 | return; | |
2641 | } | |
2642 | aio_enable_external(bdrv_get_aio_context(bs)); | |
2643 | } |