]>
Commit | Line | Data |
---|---|---|
61007b31 SH |
1 | /* |
2 | * Block layer I/O functions | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
24 | ||
80c71a24 | 25 | #include "qemu/osdep.h" |
61007b31 | 26 | #include "trace.h" |
7f0e9da6 | 27 | #include "sysemu/block-backend.h" |
61007b31 SH |
28 | #include "block/blockjob.h" |
29 | #include "block/block_int.h" | |
f348b6d1 | 30 | #include "qemu/cutils.h" |
da34e65c | 31 | #include "qapi/error.h" |
d49b6836 | 32 | #include "qemu/error-report.h" |
61007b31 SH |
33 | |
34 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
35 | ||
adad6496 | 36 | static BlockAIOCB *bdrv_co_aio_rw_vector(BdrvChild *child, |
61007b31 SH |
37 | int64_t sector_num, |
38 | QEMUIOVector *qiov, | |
39 | int nb_sectors, | |
40 | BdrvRequestFlags flags, | |
41 | BlockCompletionFunc *cb, | |
42 | void *opaque, | |
43 | bool is_write); | |
44 | static void coroutine_fn bdrv_co_do_rw(void *opaque); | |
d05aa8bb EB |
45 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
46 | int64_t offset, int count, BdrvRequestFlags flags); | |
61007b31 | 47 | |
c2066af0 | 48 | static void bdrv_parent_drained_begin(BlockDriverState *bs) |
61007b31 | 49 | { |
c2066af0 | 50 | BdrvChild *c; |
27ccdd52 | 51 | |
c2066af0 KW |
52 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
53 | if (c->role->drained_begin) { | |
54 | c->role->drained_begin(c); | |
55 | } | |
ce0f1412 PB |
56 | } |
57 | } | |
61007b31 | 58 | |
c2066af0 | 59 | static void bdrv_parent_drained_end(BlockDriverState *bs) |
ce0f1412 | 60 | { |
c2066af0 | 61 | BdrvChild *c; |
27ccdd52 | 62 | |
c2066af0 KW |
63 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
64 | if (c->role->drained_end) { | |
65 | c->role->drained_end(c); | |
66 | } | |
27ccdd52 | 67 | } |
61007b31 SH |
68 | } |
69 | ||
d9e0dfa2 EB |
70 | static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) |
71 | { | |
72 | dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); | |
73 | dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer); | |
74 | dst->opt_mem_alignment = MAX(dst->opt_mem_alignment, | |
75 | src->opt_mem_alignment); | |
76 | dst->min_mem_alignment = MAX(dst->min_mem_alignment, | |
77 | src->min_mem_alignment); | |
78 | dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov); | |
79 | } | |
80 | ||
61007b31 SH |
81 | void bdrv_refresh_limits(BlockDriverState *bs, Error **errp) |
82 | { | |
83 | BlockDriver *drv = bs->drv; | |
84 | Error *local_err = NULL; | |
85 | ||
86 | memset(&bs->bl, 0, sizeof(bs->bl)); | |
87 | ||
88 | if (!drv) { | |
89 | return; | |
90 | } | |
91 | ||
79ba8c98 | 92 | /* Default alignment based on whether driver has byte interface */ |
a5b8dd2c | 93 | bs->bl.request_alignment = drv->bdrv_co_preadv ? 1 : 512; |
79ba8c98 | 94 | |
61007b31 SH |
95 | /* Take some limits from the children as a default */ |
96 | if (bs->file) { | |
9a4f4c31 | 97 | bdrv_refresh_limits(bs->file->bs, &local_err); |
61007b31 SH |
98 | if (local_err) { |
99 | error_propagate(errp, local_err); | |
100 | return; | |
101 | } | |
d9e0dfa2 | 102 | bdrv_merge_limits(&bs->bl, &bs->file->bs->bl); |
61007b31 | 103 | } else { |
4196d2f0 | 104 | bs->bl.min_mem_alignment = 512; |
459b4e66 | 105 | bs->bl.opt_mem_alignment = getpagesize(); |
bd44feb7 SH |
106 | |
107 | /* Safe default since most protocols use readv()/writev()/etc */ | |
108 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
109 | } |
110 | ||
760e0063 KW |
111 | if (bs->backing) { |
112 | bdrv_refresh_limits(bs->backing->bs, &local_err); | |
61007b31 SH |
113 | if (local_err) { |
114 | error_propagate(errp, local_err); | |
115 | return; | |
116 | } | |
d9e0dfa2 | 117 | bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl); |
61007b31 SH |
118 | } |
119 | ||
120 | /* Then let the driver override it */ | |
121 | if (drv->bdrv_refresh_limits) { | |
122 | drv->bdrv_refresh_limits(bs, errp); | |
123 | } | |
124 | } | |
125 | ||
126 | /** | |
127 | * The copy-on-read flag is actually a reference count so multiple users may | |
128 | * use the feature without worrying about clobbering its previous state. | |
129 | * Copy-on-read stays enabled until all users have called to disable it. | |
130 | */ | |
131 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
132 | { | |
133 | bs->copy_on_read++; | |
134 | } | |
135 | ||
136 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
137 | { | |
138 | assert(bs->copy_on_read > 0); | |
139 | bs->copy_on_read--; | |
140 | } | |
141 | ||
142 | /* Check if any requests are in-flight (including throttled requests) */ | |
439db28c | 143 | bool bdrv_requests_pending(BlockDriverState *bs) |
61007b31 | 144 | { |
37a639a7 KW |
145 | BdrvChild *child; |
146 | ||
61007b31 SH |
147 | if (!QLIST_EMPTY(&bs->tracked_requests)) { |
148 | return true; | |
149 | } | |
37a639a7 KW |
150 | |
151 | QLIST_FOREACH(child, &bs->children, next) { | |
152 | if (bdrv_requests_pending(child->bs)) { | |
153 | return true; | |
154 | } | |
61007b31 | 155 | } |
37a639a7 | 156 | |
61007b31 SH |
157 | return false; |
158 | } | |
159 | ||
67da1dc5 FZ |
160 | static void bdrv_drain_recurse(BlockDriverState *bs) |
161 | { | |
162 | BdrvChild *child; | |
163 | ||
164 | if (bs->drv && bs->drv->bdrv_drain) { | |
165 | bs->drv->bdrv_drain(bs); | |
166 | } | |
167 | QLIST_FOREACH(child, &bs->children, next) { | |
168 | bdrv_drain_recurse(child->bs); | |
169 | } | |
170 | } | |
171 | ||
a77fd4bb FZ |
172 | typedef struct { |
173 | Coroutine *co; | |
174 | BlockDriverState *bs; | |
175 | QEMUBH *bh; | |
176 | bool done; | |
177 | } BdrvCoDrainData; | |
178 | ||
b6e84c97 PB |
179 | static void bdrv_drain_poll(BlockDriverState *bs) |
180 | { | |
181 | bool busy = true; | |
182 | ||
183 | while (busy) { | |
184 | /* Keep iterating */ | |
b6e84c97 PB |
185 | busy = bdrv_requests_pending(bs); |
186 | busy |= aio_poll(bdrv_get_aio_context(bs), busy); | |
187 | } | |
188 | } | |
189 | ||
a77fd4bb FZ |
190 | static void bdrv_co_drain_bh_cb(void *opaque) |
191 | { | |
192 | BdrvCoDrainData *data = opaque; | |
193 | Coroutine *co = data->co; | |
194 | ||
195 | qemu_bh_delete(data->bh); | |
b6e84c97 | 196 | bdrv_drain_poll(data->bs); |
a77fd4bb FZ |
197 | data->done = true; |
198 | qemu_coroutine_enter(co, NULL); | |
199 | } | |
200 | ||
b6e84c97 | 201 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs) |
a77fd4bb FZ |
202 | { |
203 | BdrvCoDrainData data; | |
204 | ||
205 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
206 | * other coroutines run if they were queued from | |
207 | * qemu_co_queue_run_restart(). */ | |
208 | ||
209 | assert(qemu_in_coroutine()); | |
210 | data = (BdrvCoDrainData) { | |
211 | .co = qemu_coroutine_self(), | |
212 | .bs = bs, | |
213 | .done = false, | |
214 | .bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_drain_bh_cb, &data), | |
215 | }; | |
216 | qemu_bh_schedule(data.bh); | |
217 | ||
218 | qemu_coroutine_yield(); | |
219 | /* If we are resumed from some other event (such as an aio completion or a | |
220 | * timer callback), it is a bug in the caller that should be fixed. */ | |
221 | assert(data.done); | |
222 | } | |
223 | ||
6820643f KW |
224 | void bdrv_drained_begin(BlockDriverState *bs) |
225 | { | |
226 | if (!bs->quiesce_counter++) { | |
227 | aio_disable_external(bdrv_get_aio_context(bs)); | |
228 | bdrv_parent_drained_begin(bs); | |
229 | } | |
230 | ||
231 | bdrv_io_unplugged_begin(bs); | |
232 | bdrv_drain_recurse(bs); | |
233 | if (qemu_in_coroutine()) { | |
234 | bdrv_co_yield_to_drain(bs); | |
235 | } else { | |
236 | bdrv_drain_poll(bs); | |
237 | } | |
238 | bdrv_io_unplugged_end(bs); | |
239 | } | |
240 | ||
241 | void bdrv_drained_end(BlockDriverState *bs) | |
242 | { | |
243 | assert(bs->quiesce_counter > 0); | |
244 | if (--bs->quiesce_counter > 0) { | |
245 | return; | |
246 | } | |
247 | ||
248 | bdrv_parent_drained_end(bs); | |
249 | aio_enable_external(bdrv_get_aio_context(bs)); | |
250 | } | |
251 | ||
61007b31 | 252 | /* |
67da1dc5 FZ |
253 | * Wait for pending requests to complete on a single BlockDriverState subtree, |
254 | * and suspend block driver's internal I/O until next request arrives. | |
61007b31 | 255 | * |
61007b31 SH |
256 | * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState |
257 | * AioContext. | |
7a63f3cd SH |
258 | * |
259 | * Only this BlockDriverState's AioContext is run, so in-flight requests must | |
260 | * not depend on events in other AioContexts. In that case, use | |
261 | * bdrv_drain_all() instead. | |
61007b31 | 262 | */ |
b6e84c97 | 263 | void coroutine_fn bdrv_co_drain(BlockDriverState *bs) |
61007b31 | 264 | { |
6820643f KW |
265 | assert(qemu_in_coroutine()); |
266 | bdrv_drained_begin(bs); | |
267 | bdrv_drained_end(bs); | |
b6e84c97 | 268 | } |
f406c03c | 269 | |
b6e84c97 PB |
270 | void bdrv_drain(BlockDriverState *bs) |
271 | { | |
6820643f KW |
272 | bdrv_drained_begin(bs); |
273 | bdrv_drained_end(bs); | |
61007b31 SH |
274 | } |
275 | ||
276 | /* | |
277 | * Wait for pending requests to complete across all BlockDriverStates | |
278 | * | |
279 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
280 | * after calling this function. | |
61007b31 SH |
281 | */ |
282 | void bdrv_drain_all(void) | |
283 | { | |
284 | /* Always run first iteration so any pending completion BHs run */ | |
285 | bool busy = true; | |
7c8eece4 | 286 | BlockDriverState *bs; |
88be7b4b | 287 | BdrvNextIterator it; |
eb1364ce | 288 | BlockJob *job = NULL; |
f406c03c | 289 | GSList *aio_ctxs = NULL, *ctx; |
61007b31 | 290 | |
eb1364ce AG |
291 | while ((job = block_job_next(job))) { |
292 | AioContext *aio_context = blk_get_aio_context(job->blk); | |
293 | ||
294 | aio_context_acquire(aio_context); | |
295 | block_job_pause(job); | |
296 | aio_context_release(aio_context); | |
297 | } | |
298 | ||
88be7b4b | 299 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
61007b31 SH |
300 | AioContext *aio_context = bdrv_get_aio_context(bs); |
301 | ||
302 | aio_context_acquire(aio_context); | |
c2066af0 | 303 | bdrv_parent_drained_begin(bs); |
6b98bd64 | 304 | bdrv_io_unplugged_begin(bs); |
9dcf8ecd | 305 | bdrv_drain_recurse(bs); |
61007b31 | 306 | aio_context_release(aio_context); |
f406c03c | 307 | |
764ba3ae | 308 | if (!g_slist_find(aio_ctxs, aio_context)) { |
f406c03c AY |
309 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); |
310 | } | |
61007b31 SH |
311 | } |
312 | ||
7a63f3cd SH |
313 | /* Note that completion of an asynchronous I/O operation can trigger any |
314 | * number of other I/O operations on other devices---for example a | |
315 | * coroutine can submit an I/O request to another device in response to | |
316 | * request completion. Therefore we must keep looping until there was no | |
317 | * more activity rather than simply draining each device independently. | |
318 | */ | |
61007b31 SH |
319 | while (busy) { |
320 | busy = false; | |
61007b31 | 321 | |
f406c03c AY |
322 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
323 | AioContext *aio_context = ctx->data; | |
61007b31 SH |
324 | |
325 | aio_context_acquire(aio_context); | |
88be7b4b | 326 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
f406c03c | 327 | if (aio_context == bdrv_get_aio_context(bs)) { |
f406c03c AY |
328 | if (bdrv_requests_pending(bs)) { |
329 | busy = true; | |
330 | aio_poll(aio_context, busy); | |
331 | } | |
332 | } | |
333 | } | |
334 | busy |= aio_poll(aio_context, false); | |
61007b31 SH |
335 | aio_context_release(aio_context); |
336 | } | |
337 | } | |
338 | ||
88be7b4b | 339 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
61007b31 SH |
340 | AioContext *aio_context = bdrv_get_aio_context(bs); |
341 | ||
342 | aio_context_acquire(aio_context); | |
6b98bd64 | 343 | bdrv_io_unplugged_end(bs); |
c2066af0 | 344 | bdrv_parent_drained_end(bs); |
61007b31 SH |
345 | aio_context_release(aio_context); |
346 | } | |
f406c03c | 347 | g_slist_free(aio_ctxs); |
eb1364ce AG |
348 | |
349 | job = NULL; | |
350 | while ((job = block_job_next(job))) { | |
351 | AioContext *aio_context = blk_get_aio_context(job->blk); | |
352 | ||
353 | aio_context_acquire(aio_context); | |
354 | block_job_resume(job); | |
355 | aio_context_release(aio_context); | |
356 | } | |
61007b31 SH |
357 | } |
358 | ||
359 | /** | |
360 | * Remove an active request from the tracked requests list | |
361 | * | |
362 | * This function should be called when a tracked request is completing. | |
363 | */ | |
364 | static void tracked_request_end(BdrvTrackedRequest *req) | |
365 | { | |
366 | if (req->serialising) { | |
367 | req->bs->serialising_in_flight--; | |
368 | } | |
369 | ||
370 | QLIST_REMOVE(req, list); | |
371 | qemu_co_queue_restart_all(&req->wait_queue); | |
372 | } | |
373 | ||
374 | /** | |
375 | * Add an active request to the tracked requests list | |
376 | */ | |
377 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
378 | BlockDriverState *bs, | |
379 | int64_t offset, | |
ebde595c FZ |
380 | unsigned int bytes, |
381 | enum BdrvTrackedRequestType type) | |
61007b31 SH |
382 | { |
383 | *req = (BdrvTrackedRequest){ | |
384 | .bs = bs, | |
385 | .offset = offset, | |
386 | .bytes = bytes, | |
ebde595c | 387 | .type = type, |
61007b31 SH |
388 | .co = qemu_coroutine_self(), |
389 | .serialising = false, | |
390 | .overlap_offset = offset, | |
391 | .overlap_bytes = bytes, | |
392 | }; | |
393 | ||
394 | qemu_co_queue_init(&req->wait_queue); | |
395 | ||
396 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); | |
397 | } | |
398 | ||
399 | static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align) | |
400 | { | |
401 | int64_t overlap_offset = req->offset & ~(align - 1); | |
402 | unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align) | |
403 | - overlap_offset; | |
404 | ||
405 | if (!req->serialising) { | |
406 | req->bs->serialising_in_flight++; | |
407 | req->serialising = true; | |
408 | } | |
409 | ||
410 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
411 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
412 | } | |
413 | ||
414 | /** | |
244483e6 | 415 | * Round a region to cluster boundaries (sector-based) |
61007b31 | 416 | */ |
244483e6 KW |
417 | void bdrv_round_sectors_to_clusters(BlockDriverState *bs, |
418 | int64_t sector_num, int nb_sectors, | |
419 | int64_t *cluster_sector_num, | |
420 | int *cluster_nb_sectors) | |
61007b31 SH |
421 | { |
422 | BlockDriverInfo bdi; | |
423 | ||
424 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
425 | *cluster_sector_num = sector_num; | |
426 | *cluster_nb_sectors = nb_sectors; | |
427 | } else { | |
428 | int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; | |
429 | *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); | |
430 | *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + | |
431 | nb_sectors, c); | |
432 | } | |
433 | } | |
434 | ||
244483e6 KW |
435 | /** |
436 | * Round a region to cluster boundaries | |
437 | */ | |
438 | void bdrv_round_to_clusters(BlockDriverState *bs, | |
439 | int64_t offset, unsigned int bytes, | |
440 | int64_t *cluster_offset, | |
441 | unsigned int *cluster_bytes) | |
442 | { | |
443 | BlockDriverInfo bdi; | |
444 | ||
445 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
446 | *cluster_offset = offset; | |
447 | *cluster_bytes = bytes; | |
448 | } else { | |
449 | int64_t c = bdi.cluster_size; | |
450 | *cluster_offset = QEMU_ALIGN_DOWN(offset, c); | |
451 | *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c); | |
452 | } | |
453 | } | |
454 | ||
61007b31 SH |
455 | static int bdrv_get_cluster_size(BlockDriverState *bs) |
456 | { | |
457 | BlockDriverInfo bdi; | |
458 | int ret; | |
459 | ||
460 | ret = bdrv_get_info(bs, &bdi); | |
461 | if (ret < 0 || bdi.cluster_size == 0) { | |
a5b8dd2c | 462 | return bs->bl.request_alignment; |
61007b31 SH |
463 | } else { |
464 | return bdi.cluster_size; | |
465 | } | |
466 | } | |
467 | ||
468 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
469 | int64_t offset, unsigned int bytes) | |
470 | { | |
471 | /* aaaa bbbb */ | |
472 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
473 | return false; | |
474 | } | |
475 | /* bbbb aaaa */ | |
476 | if (req->overlap_offset >= offset + bytes) { | |
477 | return false; | |
478 | } | |
479 | return true; | |
480 | } | |
481 | ||
482 | static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self) | |
483 | { | |
484 | BlockDriverState *bs = self->bs; | |
485 | BdrvTrackedRequest *req; | |
486 | bool retry; | |
487 | bool waited = false; | |
488 | ||
489 | if (!bs->serialising_in_flight) { | |
490 | return false; | |
491 | } | |
492 | ||
493 | do { | |
494 | retry = false; | |
495 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
496 | if (req == self || (!req->serialising && !self->serialising)) { | |
497 | continue; | |
498 | } | |
499 | if (tracked_request_overlaps(req, self->overlap_offset, | |
500 | self->overlap_bytes)) | |
501 | { | |
502 | /* Hitting this means there was a reentrant request, for | |
503 | * example, a block driver issuing nested requests. This must | |
504 | * never happen since it means deadlock. | |
505 | */ | |
506 | assert(qemu_coroutine_self() != req->co); | |
507 | ||
508 | /* If the request is already (indirectly) waiting for us, or | |
509 | * will wait for us as soon as it wakes up, then just go on | |
510 | * (instead of producing a deadlock in the former case). */ | |
511 | if (!req->waiting_for) { | |
512 | self->waiting_for = req; | |
513 | qemu_co_queue_wait(&req->wait_queue); | |
514 | self->waiting_for = NULL; | |
515 | retry = true; | |
516 | waited = true; | |
517 | break; | |
518 | } | |
519 | } | |
520 | } | |
521 | } while (retry); | |
522 | ||
523 | return waited; | |
524 | } | |
525 | ||
526 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
527 | size_t size) | |
528 | { | |
529 | if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) { | |
530 | return -EIO; | |
531 | } | |
532 | ||
533 | if (!bdrv_is_inserted(bs)) { | |
534 | return -ENOMEDIUM; | |
535 | } | |
536 | ||
537 | if (offset < 0) { | |
538 | return -EIO; | |
539 | } | |
540 | ||
541 | return 0; | |
542 | } | |
543 | ||
544 | static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num, | |
545 | int nb_sectors) | |
546 | { | |
547 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
548 | return -EIO; | |
549 | } | |
550 | ||
551 | return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE, | |
552 | nb_sectors * BDRV_SECTOR_SIZE); | |
553 | } | |
554 | ||
555 | typedef struct RwCo { | |
556 | BlockDriverState *bs; | |
557 | int64_t offset; | |
558 | QEMUIOVector *qiov; | |
559 | bool is_write; | |
560 | int ret; | |
561 | BdrvRequestFlags flags; | |
562 | } RwCo; | |
563 | ||
564 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
565 | { | |
566 | RwCo *rwco = opaque; | |
567 | ||
568 | if (!rwco->is_write) { | |
cab3a356 KW |
569 | rwco->ret = bdrv_co_preadv(rwco->bs, rwco->offset, |
570 | rwco->qiov->size, rwco->qiov, | |
571 | rwco->flags); | |
61007b31 | 572 | } else { |
cab3a356 KW |
573 | rwco->ret = bdrv_co_pwritev(rwco->bs, rwco->offset, |
574 | rwco->qiov->size, rwco->qiov, | |
575 | rwco->flags); | |
61007b31 SH |
576 | } |
577 | } | |
578 | ||
579 | /* | |
580 | * Process a vectored synchronous request using coroutines | |
581 | */ | |
582 | static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset, | |
583 | QEMUIOVector *qiov, bool is_write, | |
584 | BdrvRequestFlags flags) | |
585 | { | |
586 | Coroutine *co; | |
587 | RwCo rwco = { | |
588 | .bs = bs, | |
589 | .offset = offset, | |
590 | .qiov = qiov, | |
591 | .is_write = is_write, | |
592 | .ret = NOT_DONE, | |
593 | .flags = flags, | |
594 | }; | |
595 | ||
61007b31 SH |
596 | if (qemu_in_coroutine()) { |
597 | /* Fast-path if already in coroutine context */ | |
598 | bdrv_rw_co_entry(&rwco); | |
599 | } else { | |
600 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
601 | ||
602 | co = qemu_coroutine_create(bdrv_rw_co_entry); | |
603 | qemu_coroutine_enter(co, &rwco); | |
604 | while (rwco.ret == NOT_DONE) { | |
605 | aio_poll(aio_context, true); | |
606 | } | |
607 | } | |
608 | return rwco.ret; | |
609 | } | |
610 | ||
611 | /* | |
612 | * Process a synchronous request using coroutines | |
613 | */ | |
614 | static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf, | |
615 | int nb_sectors, bool is_write, BdrvRequestFlags flags) | |
616 | { | |
617 | QEMUIOVector qiov; | |
618 | struct iovec iov = { | |
619 | .iov_base = (void *)buf, | |
620 | .iov_len = nb_sectors * BDRV_SECTOR_SIZE, | |
621 | }; | |
622 | ||
623 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
624 | return -EINVAL; | |
625 | } | |
626 | ||
627 | qemu_iovec_init_external(&qiov, &iov, 1); | |
628 | return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS, | |
629 | &qiov, is_write, flags); | |
630 | } | |
631 | ||
632 | /* return < 0 if error. See bdrv_write() for the return codes */ | |
fbcbbf4e | 633 | int bdrv_read(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
634 | uint8_t *buf, int nb_sectors) |
635 | { | |
fbcbbf4e | 636 | return bdrv_rw_co(child->bs, sector_num, buf, nb_sectors, false, 0); |
61007b31 SH |
637 | } |
638 | ||
61007b31 SH |
639 | /* Return < 0 if error. Important errors are: |
640 | -EIO generic I/O error (may happen for all errors) | |
641 | -ENOMEDIUM No media inserted. | |
642 | -EINVAL Invalid sector number or nb_sectors | |
643 | -EACCES Trying to write a read-only device | |
644 | */ | |
18d51c4b | 645 | int bdrv_write(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
646 | const uint8_t *buf, int nb_sectors) |
647 | { | |
18d51c4b KW |
648 | return bdrv_rw_co(child->bs, sector_num, (uint8_t *)buf, nb_sectors, |
649 | true, 0); | |
61007b31 SH |
650 | } |
651 | ||
74021bc4 EB |
652 | int bdrv_pwrite_zeroes(BlockDriverState *bs, int64_t offset, |
653 | int count, BdrvRequestFlags flags) | |
61007b31 | 654 | { |
74021bc4 EB |
655 | QEMUIOVector qiov; |
656 | struct iovec iov = { | |
657 | .iov_base = NULL, | |
658 | .iov_len = count, | |
659 | }; | |
660 | ||
661 | qemu_iovec_init_external(&qiov, &iov, 1); | |
662 | return bdrv_prwv_co(bs, offset, &qiov, true, | |
663 | BDRV_REQ_ZERO_WRITE | flags); | |
61007b31 SH |
664 | } |
665 | ||
666 | /* | |
74021bc4 | 667 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
668 | * The operation is sped up by checking the block status and only writing |
669 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 670 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 671 | * BDRV_REQ_FUA). |
61007b31 SH |
672 | * |
673 | * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). | |
674 | */ | |
675 | int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags) | |
676 | { | |
677 | int64_t target_sectors, ret, nb_sectors, sector_num = 0; | |
67a0fd2a | 678 | BlockDriverState *file; |
61007b31 SH |
679 | int n; |
680 | ||
681 | target_sectors = bdrv_nb_sectors(bs); | |
682 | if (target_sectors < 0) { | |
683 | return target_sectors; | |
684 | } | |
685 | ||
686 | for (;;) { | |
687 | nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS); | |
688 | if (nb_sectors <= 0) { | |
689 | return 0; | |
690 | } | |
67a0fd2a | 691 | ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n, &file); |
61007b31 SH |
692 | if (ret < 0) { |
693 | error_report("error getting block status at sector %" PRId64 ": %s", | |
694 | sector_num, strerror(-ret)); | |
695 | return ret; | |
696 | } | |
697 | if (ret & BDRV_BLOCK_ZERO) { | |
698 | sector_num += n; | |
699 | continue; | |
700 | } | |
74021bc4 EB |
701 | ret = bdrv_pwrite_zeroes(bs, sector_num << BDRV_SECTOR_BITS, |
702 | n << BDRV_SECTOR_BITS, flags); | |
61007b31 SH |
703 | if (ret < 0) { |
704 | error_report("error writing zeroes at sector %" PRId64 ": %s", | |
705 | sector_num, strerror(-ret)); | |
706 | return ret; | |
707 | } | |
708 | sector_num += n; | |
709 | } | |
710 | } | |
711 | ||
cf2ab8fc | 712 | int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
f1e84741 KW |
713 | { |
714 | int ret; | |
715 | ||
cf2ab8fc | 716 | ret = bdrv_prwv_co(child->bs, offset, qiov, false, 0); |
f1e84741 KW |
717 | if (ret < 0) { |
718 | return ret; | |
719 | } | |
720 | ||
721 | return qiov->size; | |
722 | } | |
723 | ||
cf2ab8fc | 724 | int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes) |
61007b31 SH |
725 | { |
726 | QEMUIOVector qiov; | |
727 | struct iovec iov = { | |
728 | .iov_base = (void *)buf, | |
729 | .iov_len = bytes, | |
730 | }; | |
61007b31 SH |
731 | |
732 | if (bytes < 0) { | |
733 | return -EINVAL; | |
734 | } | |
735 | ||
736 | qemu_iovec_init_external(&qiov, &iov, 1); | |
cf2ab8fc | 737 | return bdrv_preadv(child, offset, &qiov); |
61007b31 SH |
738 | } |
739 | ||
740 | int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov) | |
741 | { | |
742 | int ret; | |
743 | ||
744 | ret = bdrv_prwv_co(bs, offset, qiov, true, 0); | |
745 | if (ret < 0) { | |
746 | return ret; | |
747 | } | |
748 | ||
749 | return qiov->size; | |
750 | } | |
751 | ||
752 | int bdrv_pwrite(BlockDriverState *bs, int64_t offset, | |
753 | const void *buf, int bytes) | |
754 | { | |
755 | QEMUIOVector qiov; | |
756 | struct iovec iov = { | |
757 | .iov_base = (void *) buf, | |
758 | .iov_len = bytes, | |
759 | }; | |
760 | ||
761 | if (bytes < 0) { | |
762 | return -EINVAL; | |
763 | } | |
764 | ||
765 | qemu_iovec_init_external(&qiov, &iov, 1); | |
766 | return bdrv_pwritev(bs, offset, &qiov); | |
767 | } | |
768 | ||
769 | /* | |
770 | * Writes to the file and ensures that no writes are reordered across this | |
771 | * request (acts as a barrier) | |
772 | * | |
773 | * Returns 0 on success, -errno in error cases. | |
774 | */ | |
775 | int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset, | |
776 | const void *buf, int count) | |
777 | { | |
778 | int ret; | |
779 | ||
780 | ret = bdrv_pwrite(bs, offset, buf, count); | |
781 | if (ret < 0) { | |
782 | return ret; | |
783 | } | |
784 | ||
855a6a93 KW |
785 | ret = bdrv_flush(bs); |
786 | if (ret < 0) { | |
787 | return ret; | |
61007b31 SH |
788 | } |
789 | ||
790 | return 0; | |
791 | } | |
792 | ||
08844473 KW |
793 | typedef struct CoroutineIOCompletion { |
794 | Coroutine *coroutine; | |
795 | int ret; | |
796 | } CoroutineIOCompletion; | |
797 | ||
798 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
799 | { | |
800 | CoroutineIOCompletion *co = opaque; | |
801 | ||
802 | co->ret = ret; | |
803 | qemu_coroutine_enter(co->coroutine, NULL); | |
804 | } | |
805 | ||
166fe960 KW |
806 | static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs, |
807 | uint64_t offset, uint64_t bytes, | |
808 | QEMUIOVector *qiov, int flags) | |
809 | { | |
810 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
811 | int64_t sector_num; |
812 | unsigned int nb_sectors; | |
813 | ||
fa166538 EB |
814 | assert(!(flags & ~BDRV_REQ_MASK)); |
815 | ||
3fb06697 KW |
816 | if (drv->bdrv_co_preadv) { |
817 | return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); | |
818 | } | |
819 | ||
820 | sector_num = offset >> BDRV_SECTOR_BITS; | |
821 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
166fe960 KW |
822 | |
823 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
824 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
825 | assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS); | |
826 | ||
08844473 KW |
827 | if (drv->bdrv_co_readv) { |
828 | return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
829 | } else { | |
830 | BlockAIOCB *acb; | |
831 | CoroutineIOCompletion co = { | |
832 | .coroutine = qemu_coroutine_self(), | |
833 | }; | |
834 | ||
835 | acb = bs->drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors, | |
836 | bdrv_co_io_em_complete, &co); | |
837 | if (acb == NULL) { | |
838 | return -EIO; | |
839 | } else { | |
840 | qemu_coroutine_yield(); | |
841 | return co.ret; | |
842 | } | |
843 | } | |
166fe960 KW |
844 | } |
845 | ||
78a07294 KW |
846 | static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs, |
847 | uint64_t offset, uint64_t bytes, | |
848 | QEMUIOVector *qiov, int flags) | |
849 | { | |
850 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
851 | int64_t sector_num; |
852 | unsigned int nb_sectors; | |
78a07294 KW |
853 | int ret; |
854 | ||
fa166538 EB |
855 | assert(!(flags & ~BDRV_REQ_MASK)); |
856 | ||
3fb06697 | 857 | if (drv->bdrv_co_pwritev) { |
515c2f43 KW |
858 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, |
859 | flags & bs->supported_write_flags); | |
860 | flags &= ~bs->supported_write_flags; | |
3fb06697 KW |
861 | goto emulate_flags; |
862 | } | |
863 | ||
864 | sector_num = offset >> BDRV_SECTOR_BITS; | |
865 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
866 | ||
78a07294 KW |
867 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); |
868 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
869 | assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS); | |
870 | ||
871 | if (drv->bdrv_co_writev_flags) { | |
872 | ret = drv->bdrv_co_writev_flags(bs, sector_num, nb_sectors, qiov, | |
4df863f3 EB |
873 | flags & bs->supported_write_flags); |
874 | flags &= ~bs->supported_write_flags; | |
08844473 | 875 | } else if (drv->bdrv_co_writev) { |
4df863f3 | 876 | assert(!bs->supported_write_flags); |
78a07294 | 877 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); |
08844473 KW |
878 | } else { |
879 | BlockAIOCB *acb; | |
880 | CoroutineIOCompletion co = { | |
881 | .coroutine = qemu_coroutine_self(), | |
882 | }; | |
883 | ||
884 | acb = bs->drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors, | |
885 | bdrv_co_io_em_complete, &co); | |
886 | if (acb == NULL) { | |
3fb06697 | 887 | ret = -EIO; |
08844473 KW |
888 | } else { |
889 | qemu_coroutine_yield(); | |
3fb06697 | 890 | ret = co.ret; |
08844473 | 891 | } |
78a07294 KW |
892 | } |
893 | ||
3fb06697 | 894 | emulate_flags: |
4df863f3 | 895 | if (ret == 0 && (flags & BDRV_REQ_FUA)) { |
78a07294 KW |
896 | ret = bdrv_co_flush(bs); |
897 | } | |
898 | ||
899 | return ret; | |
900 | } | |
901 | ||
61007b31 | 902 | static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, |
244483e6 | 903 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov) |
61007b31 SH |
904 | { |
905 | /* Perform I/O through a temporary buffer so that users who scribble over | |
906 | * their read buffer while the operation is in progress do not end up | |
907 | * modifying the image file. This is critical for zero-copy guest I/O | |
908 | * where anything might happen inside guest memory. | |
909 | */ | |
910 | void *bounce_buffer; | |
911 | ||
912 | BlockDriver *drv = bs->drv; | |
913 | struct iovec iov; | |
914 | QEMUIOVector bounce_qiov; | |
244483e6 KW |
915 | int64_t cluster_offset; |
916 | unsigned int cluster_bytes; | |
61007b31 SH |
917 | size_t skip_bytes; |
918 | int ret; | |
919 | ||
920 | /* Cover entire cluster so no additional backing file I/O is required when | |
921 | * allocating cluster in the image file. | |
922 | */ | |
244483e6 | 923 | bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes); |
61007b31 | 924 | |
244483e6 KW |
925 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
926 | cluster_offset, cluster_bytes); | |
61007b31 | 927 | |
244483e6 | 928 | iov.iov_len = cluster_bytes; |
61007b31 SH |
929 | iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len); |
930 | if (bounce_buffer == NULL) { | |
931 | ret = -ENOMEM; | |
932 | goto err; | |
933 | } | |
934 | ||
935 | qemu_iovec_init_external(&bounce_qiov, &iov, 1); | |
936 | ||
244483e6 | 937 | ret = bdrv_driver_preadv(bs, cluster_offset, cluster_bytes, |
166fe960 | 938 | &bounce_qiov, 0); |
61007b31 SH |
939 | if (ret < 0) { |
940 | goto err; | |
941 | } | |
942 | ||
c1499a5e | 943 | if (drv->bdrv_co_pwrite_zeroes && |
61007b31 | 944 | buffer_is_zero(bounce_buffer, iov.iov_len)) { |
a604fa2b EB |
945 | /* FIXME: Should we (perhaps conditionally) be setting |
946 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
947 | * that still correctly reads as zero? */ | |
244483e6 | 948 | ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, cluster_bytes, 0); |
61007b31 SH |
949 | } else { |
950 | /* This does not change the data on the disk, it is not necessary | |
951 | * to flush even in cache=writethrough mode. | |
952 | */ | |
244483e6 | 953 | ret = bdrv_driver_pwritev(bs, cluster_offset, cluster_bytes, |
78a07294 | 954 | &bounce_qiov, 0); |
61007b31 SH |
955 | } |
956 | ||
957 | if (ret < 0) { | |
958 | /* It might be okay to ignore write errors for guest requests. If this | |
959 | * is a deliberate copy-on-read then we don't want to ignore the error. | |
960 | * Simply report it in all cases. | |
961 | */ | |
962 | goto err; | |
963 | } | |
964 | ||
244483e6 KW |
965 | skip_bytes = offset - cluster_offset; |
966 | qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, bytes); | |
61007b31 SH |
967 | |
968 | err: | |
969 | qemu_vfree(bounce_buffer); | |
970 | return ret; | |
971 | } | |
972 | ||
973 | /* | |
974 | * Forwards an already correctly aligned request to the BlockDriver. This | |
975 | * handles copy on read and zeroing after EOF; any other features must be | |
976 | * implemented by the caller. | |
977 | */ | |
978 | static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs, | |
979 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, | |
980 | int64_t align, QEMUIOVector *qiov, int flags) | |
981 | { | |
c9d20029 | 982 | int64_t total_bytes, max_bytes; |
61007b31 SH |
983 | int ret; |
984 | ||
49c07526 KW |
985 | assert(is_power_of_2(align)); |
986 | assert((offset & (align - 1)) == 0); | |
987 | assert((bytes & (align - 1)) == 0); | |
61007b31 | 988 | assert(!qiov || bytes == qiov->size); |
abb06c5a | 989 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
a604fa2b EB |
990 | |
991 | /* TODO: We would need a per-BDS .supported_read_flags and | |
992 | * potential fallback support, if we ever implement any read flags | |
993 | * to pass through to drivers. For now, there aren't any | |
994 | * passthrough flags. */ | |
995 | assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ))); | |
61007b31 SH |
996 | |
997 | /* Handle Copy on Read and associated serialisation */ | |
998 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
999 | /* If we touch the same cluster it counts as an overlap. This | |
1000 | * guarantees that allocating writes will be serialized and not race | |
1001 | * with each other for the same cluster. For example, in copy-on-read | |
1002 | * it ensures that the CoR read and write operations are atomic and | |
1003 | * guest writes cannot interleave between them. */ | |
1004 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
1005 | } | |
1006 | ||
61408b25 FZ |
1007 | if (!(flags & BDRV_REQ_NO_SERIALISING)) { |
1008 | wait_serialising_requests(req); | |
1009 | } | |
61007b31 SH |
1010 | |
1011 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
49c07526 KW |
1012 | int64_t start_sector = offset >> BDRV_SECTOR_BITS; |
1013 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1014 | unsigned int nb_sectors = end_sector - start_sector; | |
61007b31 SH |
1015 | int pnum; |
1016 | ||
49c07526 | 1017 | ret = bdrv_is_allocated(bs, start_sector, nb_sectors, &pnum); |
61007b31 SH |
1018 | if (ret < 0) { |
1019 | goto out; | |
1020 | } | |
1021 | ||
1022 | if (!ret || pnum != nb_sectors) { | |
244483e6 | 1023 | ret = bdrv_co_do_copy_on_readv(bs, offset, bytes, qiov); |
61007b31 SH |
1024 | goto out; |
1025 | } | |
1026 | } | |
1027 | ||
1028 | /* Forward the request to the BlockDriver */ | |
c9d20029 KW |
1029 | total_bytes = bdrv_getlength(bs); |
1030 | if (total_bytes < 0) { | |
1031 | ret = total_bytes; | |
1032 | goto out; | |
1033 | } | |
61007b31 | 1034 | |
c9d20029 | 1035 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
82524274 | 1036 | if (bytes <= max_bytes) { |
c9d20029 KW |
1037 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0); |
1038 | } else if (max_bytes > 0) { | |
1039 | QEMUIOVector local_qiov; | |
61007b31 | 1040 | |
c9d20029 KW |
1041 | qemu_iovec_init(&local_qiov, qiov->niov); |
1042 | qemu_iovec_concat(&local_qiov, qiov, 0, max_bytes); | |
61007b31 | 1043 | |
c9d20029 | 1044 | ret = bdrv_driver_preadv(bs, offset, max_bytes, &local_qiov, 0); |
61007b31 | 1045 | |
c9d20029 KW |
1046 | qemu_iovec_destroy(&local_qiov); |
1047 | } else { | |
1048 | ret = 0; | |
1049 | } | |
61007b31 | 1050 | |
c9d20029 KW |
1051 | /* Reading beyond end of file is supposed to produce zeroes */ |
1052 | if (ret == 0 && total_bytes < offset + bytes) { | |
1053 | uint64_t zero_offset = MAX(0, total_bytes - offset); | |
1054 | uint64_t zero_bytes = offset + bytes - zero_offset; | |
1055 | qemu_iovec_memset(qiov, zero_offset, 0, zero_bytes); | |
61007b31 SH |
1056 | } |
1057 | ||
1058 | out: | |
1059 | return ret; | |
1060 | } | |
1061 | ||
61007b31 SH |
1062 | /* |
1063 | * Handle a read request in coroutine context | |
1064 | */ | |
cab3a356 | 1065 | int coroutine_fn bdrv_co_preadv(BlockDriverState *bs, |
61007b31 SH |
1066 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1067 | BdrvRequestFlags flags) | |
1068 | { | |
1069 | BlockDriver *drv = bs->drv; | |
1070 | BdrvTrackedRequest req; | |
1071 | ||
a5b8dd2c | 1072 | uint64_t align = bs->bl.request_alignment; |
61007b31 SH |
1073 | uint8_t *head_buf = NULL; |
1074 | uint8_t *tail_buf = NULL; | |
1075 | QEMUIOVector local_qiov; | |
1076 | bool use_local_qiov = false; | |
1077 | int ret; | |
1078 | ||
1079 | if (!drv) { | |
1080 | return -ENOMEDIUM; | |
1081 | } | |
1082 | ||
1083 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1084 | if (ret < 0) { | |
1085 | return ret; | |
1086 | } | |
1087 | ||
9568b511 | 1088 | /* Don't do copy-on-read if we read data before write operation */ |
61408b25 | 1089 | if (bs->copy_on_read && !(flags & BDRV_REQ_NO_SERIALISING)) { |
61007b31 SH |
1090 | flags |= BDRV_REQ_COPY_ON_READ; |
1091 | } | |
1092 | ||
61007b31 SH |
1093 | /* Align read if necessary by padding qiov */ |
1094 | if (offset & (align - 1)) { | |
1095 | head_buf = qemu_blockalign(bs, align); | |
1096 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1097 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1098 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1099 | use_local_qiov = true; | |
1100 | ||
1101 | bytes += offset & (align - 1); | |
1102 | offset = offset & ~(align - 1); | |
1103 | } | |
1104 | ||
1105 | if ((offset + bytes) & (align - 1)) { | |
1106 | if (!use_local_qiov) { | |
1107 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1108 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1109 | use_local_qiov = true; | |
1110 | } | |
1111 | tail_buf = qemu_blockalign(bs, align); | |
1112 | qemu_iovec_add(&local_qiov, tail_buf, | |
1113 | align - ((offset + bytes) & (align - 1))); | |
1114 | ||
1115 | bytes = ROUND_UP(bytes, align); | |
1116 | } | |
1117 | ||
ebde595c | 1118 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
61007b31 SH |
1119 | ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align, |
1120 | use_local_qiov ? &local_qiov : qiov, | |
1121 | flags); | |
1122 | tracked_request_end(&req); | |
1123 | ||
1124 | if (use_local_qiov) { | |
1125 | qemu_iovec_destroy(&local_qiov); | |
1126 | qemu_vfree(head_buf); | |
1127 | qemu_vfree(tail_buf); | |
1128 | } | |
1129 | ||
1130 | return ret; | |
1131 | } | |
1132 | ||
adad6496 | 1133 | static int coroutine_fn bdrv_co_do_readv(BdrvChild *child, |
61007b31 SH |
1134 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, |
1135 | BdrvRequestFlags flags) | |
1136 | { | |
1137 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
1138 | return -EINVAL; | |
1139 | } | |
1140 | ||
adad6496 | 1141 | return bdrv_co_preadv(child->bs, sector_num << BDRV_SECTOR_BITS, |
cab3a356 | 1142 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); |
61007b31 SH |
1143 | } |
1144 | ||
28b04a8f KW |
1145 | int coroutine_fn bdrv_co_readv(BdrvChild *child, int64_t sector_num, |
1146 | int nb_sectors, QEMUIOVector *qiov) | |
61007b31 | 1147 | { |
28b04a8f | 1148 | trace_bdrv_co_readv(child->bs, sector_num, nb_sectors); |
61007b31 | 1149 | |
adad6496 | 1150 | return bdrv_co_do_readv(child, sector_num, nb_sectors, qiov, 0); |
61007b31 SH |
1151 | } |
1152 | ||
5def6b80 EB |
1153 | /* Maximum buffer for write zeroes fallback, in bytes */ |
1154 | #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS) | |
61007b31 | 1155 | |
d05aa8bb EB |
1156 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
1157 | int64_t offset, int count, BdrvRequestFlags flags) | |
61007b31 SH |
1158 | { |
1159 | BlockDriver *drv = bs->drv; | |
1160 | QEMUIOVector qiov; | |
1161 | struct iovec iov = {0}; | |
1162 | int ret = 0; | |
465fe887 | 1163 | bool need_flush = false; |
443668ca DL |
1164 | int head = 0; |
1165 | int tail = 0; | |
61007b31 | 1166 | |
cf081fca | 1167 | int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX); |
a5b8dd2c EB |
1168 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1169 | bs->bl.request_alignment); | |
d05aa8bb EB |
1170 | |
1171 | assert(is_power_of_2(alignment)); | |
1172 | head = offset & (alignment - 1); | |
1173 | tail = (offset + count) & (alignment - 1); | |
1174 | max_write_zeroes &= ~(alignment - 1); | |
61007b31 | 1175 | |
d05aa8bb EB |
1176 | while (count > 0 && !ret) { |
1177 | int num = count; | |
61007b31 SH |
1178 | |
1179 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1180 | * to be aligned, and that unaligned requests do not cross cluster |
1181 | * boundaries. | |
61007b31 | 1182 | */ |
443668ca DL |
1183 | if (head) { |
1184 | /* Make a small request up to the first aligned sector. */ | |
d05aa8bb | 1185 | num = MIN(count, alignment - head); |
443668ca | 1186 | head = 0; |
d05aa8bb | 1187 | } else if (tail && num > alignment) { |
443668ca DL |
1188 | /* Shorten the request to the last aligned sector. */ |
1189 | num -= tail; | |
61007b31 SH |
1190 | } |
1191 | ||
1192 | /* limit request size */ | |
1193 | if (num > max_write_zeroes) { | |
1194 | num = max_write_zeroes; | |
1195 | } | |
1196 | ||
1197 | ret = -ENOTSUP; | |
1198 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1199 | if (drv->bdrv_co_pwrite_zeroes) { |
1200 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1201 | flags & bs->supported_zero_flags); | |
1202 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1203 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1204 | need_flush = true; | |
1205 | } | |
465fe887 EB |
1206 | } else { |
1207 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1208 | } |
1209 | ||
1210 | if (ret == -ENOTSUP) { | |
1211 | /* Fall back to bounce buffer if write zeroes is unsupported */ | |
5def6b80 | 1212 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
61007b31 | 1213 | MAX_WRITE_ZEROES_BOUNCE_BUFFER); |
465fe887 EB |
1214 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1215 | ||
1216 | if ((flags & BDRV_REQ_FUA) && | |
1217 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1218 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1219 | * flush on each chunk; use just one at the end */ | |
1220 | write_flags &= ~BDRV_REQ_FUA; | |
1221 | need_flush = true; | |
1222 | } | |
5def6b80 | 1223 | num = MIN(num, max_transfer); |
d05aa8bb | 1224 | iov.iov_len = num; |
61007b31 | 1225 | if (iov.iov_base == NULL) { |
d05aa8bb | 1226 | iov.iov_base = qemu_try_blockalign(bs, num); |
61007b31 SH |
1227 | if (iov.iov_base == NULL) { |
1228 | ret = -ENOMEM; | |
1229 | goto fail; | |
1230 | } | |
d05aa8bb | 1231 | memset(iov.iov_base, 0, num); |
61007b31 SH |
1232 | } |
1233 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1234 | ||
d05aa8bb | 1235 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags); |
61007b31 SH |
1236 | |
1237 | /* Keep bounce buffer around if it is big enough for all | |
1238 | * all future requests. | |
1239 | */ | |
5def6b80 | 1240 | if (num < max_transfer) { |
61007b31 SH |
1241 | qemu_vfree(iov.iov_base); |
1242 | iov.iov_base = NULL; | |
1243 | } | |
1244 | } | |
1245 | ||
d05aa8bb EB |
1246 | offset += num; |
1247 | count -= num; | |
61007b31 SH |
1248 | } |
1249 | ||
1250 | fail: | |
465fe887 EB |
1251 | if (ret == 0 && need_flush) { |
1252 | ret = bdrv_co_flush(bs); | |
1253 | } | |
61007b31 SH |
1254 | qemu_vfree(iov.iov_base); |
1255 | return ret; | |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * Forwards an already correctly aligned write request to the BlockDriver. | |
1260 | */ | |
1261 | static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs, | |
1262 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, | |
cff86b38 | 1263 | int64_t align, QEMUIOVector *qiov, int flags) |
61007b31 SH |
1264 | { |
1265 | BlockDriver *drv = bs->drv; | |
1266 | bool waited; | |
1267 | int ret; | |
1268 | ||
9896c876 KW |
1269 | int64_t start_sector = offset >> BDRV_SECTOR_BITS; |
1270 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
61007b31 | 1271 | |
cff86b38 EB |
1272 | assert(is_power_of_2(align)); |
1273 | assert((offset & (align - 1)) == 0); | |
1274 | assert((bytes & (align - 1)) == 0); | |
61007b31 | 1275 | assert(!qiov || bytes == qiov->size); |
abb06c5a | 1276 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
fa166538 | 1277 | assert(!(flags & ~BDRV_REQ_MASK)); |
61007b31 SH |
1278 | |
1279 | waited = wait_serialising_requests(req); | |
1280 | assert(!waited || !req->serialising); | |
1281 | assert(req->overlap_offset <= offset); | |
1282 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
1283 | ||
1284 | ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req); | |
1285 | ||
1286 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 1287 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
61007b31 SH |
1288 | qemu_iovec_is_zero(qiov)) { |
1289 | flags |= BDRV_REQ_ZERO_WRITE; | |
1290 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
1291 | flags |= BDRV_REQ_MAY_UNMAP; | |
1292 | } | |
1293 | } | |
1294 | ||
1295 | if (ret < 0) { | |
1296 | /* Do nothing, write notifier decided to fail this request */ | |
1297 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
9a4f4c31 | 1298 | bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 1299 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
61007b31 | 1300 | } else { |
9a4f4c31 | 1301 | bdrv_debug_event(bs, BLKDBG_PWRITEV); |
78a07294 | 1302 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags); |
61007b31 | 1303 | } |
9a4f4c31 | 1304 | bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 1305 | |
9896c876 | 1306 | bdrv_set_dirty(bs, start_sector, end_sector - start_sector); |
61007b31 | 1307 | |
53d8f9d8 HR |
1308 | if (bs->wr_highest_offset < offset + bytes) { |
1309 | bs->wr_highest_offset = offset + bytes; | |
1310 | } | |
61007b31 SH |
1311 | |
1312 | if (ret >= 0) { | |
9896c876 | 1313 | bs->total_sectors = MAX(bs->total_sectors, end_sector); |
61007b31 SH |
1314 | } |
1315 | ||
1316 | return ret; | |
1317 | } | |
1318 | ||
9eeb6dd1 FZ |
1319 | static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs, |
1320 | int64_t offset, | |
1321 | unsigned int bytes, | |
1322 | BdrvRequestFlags flags, | |
1323 | BdrvTrackedRequest *req) | |
1324 | { | |
1325 | uint8_t *buf = NULL; | |
1326 | QEMUIOVector local_qiov; | |
1327 | struct iovec iov; | |
a5b8dd2c | 1328 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 FZ |
1329 | unsigned int head_padding_bytes, tail_padding_bytes; |
1330 | int ret = 0; | |
1331 | ||
1332 | head_padding_bytes = offset & (align - 1); | |
1333 | tail_padding_bytes = align - ((offset + bytes) & (align - 1)); | |
1334 | ||
1335 | ||
1336 | assert(flags & BDRV_REQ_ZERO_WRITE); | |
1337 | if (head_padding_bytes || tail_padding_bytes) { | |
1338 | buf = qemu_blockalign(bs, align); | |
1339 | iov = (struct iovec) { | |
1340 | .iov_base = buf, | |
1341 | .iov_len = align, | |
1342 | }; | |
1343 | qemu_iovec_init_external(&local_qiov, &iov, 1); | |
1344 | } | |
1345 | if (head_padding_bytes) { | |
1346 | uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes); | |
1347 | ||
1348 | /* RMW the unaligned part before head. */ | |
1349 | mark_request_serialising(req, align); | |
1350 | wait_serialising_requests(req); | |
9a4f4c31 | 1351 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
9eeb6dd1 FZ |
1352 | ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align, |
1353 | align, &local_qiov, 0); | |
1354 | if (ret < 0) { | |
1355 | goto fail; | |
1356 | } | |
9a4f4c31 | 1357 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
9eeb6dd1 FZ |
1358 | |
1359 | memset(buf + head_padding_bytes, 0, zero_bytes); | |
1360 | ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align, | |
cff86b38 | 1361 | align, &local_qiov, |
9eeb6dd1 FZ |
1362 | flags & ~BDRV_REQ_ZERO_WRITE); |
1363 | if (ret < 0) { | |
1364 | goto fail; | |
1365 | } | |
1366 | offset += zero_bytes; | |
1367 | bytes -= zero_bytes; | |
1368 | } | |
1369 | ||
1370 | assert(!bytes || (offset & (align - 1)) == 0); | |
1371 | if (bytes >= align) { | |
1372 | /* Write the aligned part in the middle. */ | |
1373 | uint64_t aligned_bytes = bytes & ~(align - 1); | |
cff86b38 | 1374 | ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes, align, |
9eeb6dd1 FZ |
1375 | NULL, flags); |
1376 | if (ret < 0) { | |
1377 | goto fail; | |
1378 | } | |
1379 | bytes -= aligned_bytes; | |
1380 | offset += aligned_bytes; | |
1381 | } | |
1382 | ||
1383 | assert(!bytes || (offset & (align - 1)) == 0); | |
1384 | if (bytes) { | |
1385 | assert(align == tail_padding_bytes + bytes); | |
1386 | /* RMW the unaligned part after tail. */ | |
1387 | mark_request_serialising(req, align); | |
1388 | wait_serialising_requests(req); | |
9a4f4c31 | 1389 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
9eeb6dd1 FZ |
1390 | ret = bdrv_aligned_preadv(bs, req, offset, align, |
1391 | align, &local_qiov, 0); | |
1392 | if (ret < 0) { | |
1393 | goto fail; | |
1394 | } | |
9a4f4c31 | 1395 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
9eeb6dd1 FZ |
1396 | |
1397 | memset(buf, 0, bytes); | |
cff86b38 | 1398 | ret = bdrv_aligned_pwritev(bs, req, offset, align, align, |
9eeb6dd1 FZ |
1399 | &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE); |
1400 | } | |
1401 | fail: | |
1402 | qemu_vfree(buf); | |
1403 | return ret; | |
1404 | ||
1405 | } | |
1406 | ||
61007b31 SH |
1407 | /* |
1408 | * Handle a write request in coroutine context | |
1409 | */ | |
cab3a356 | 1410 | int coroutine_fn bdrv_co_pwritev(BlockDriverState *bs, |
61007b31 SH |
1411 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1412 | BdrvRequestFlags flags) | |
1413 | { | |
1414 | BdrvTrackedRequest req; | |
a5b8dd2c | 1415 | uint64_t align = bs->bl.request_alignment; |
61007b31 SH |
1416 | uint8_t *head_buf = NULL; |
1417 | uint8_t *tail_buf = NULL; | |
1418 | QEMUIOVector local_qiov; | |
1419 | bool use_local_qiov = false; | |
1420 | int ret; | |
1421 | ||
1422 | if (!bs->drv) { | |
1423 | return -ENOMEDIUM; | |
1424 | } | |
1425 | if (bs->read_only) { | |
eaf5fe2d | 1426 | return -EPERM; |
61007b31 | 1427 | } |
04c01a5c | 1428 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
61007b31 SH |
1429 | |
1430 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1431 | if (ret < 0) { | |
1432 | return ret; | |
1433 | } | |
1434 | ||
61007b31 SH |
1435 | /* |
1436 | * Align write if necessary by performing a read-modify-write cycle. | |
1437 | * Pad qiov with the read parts and be sure to have a tracked request not | |
1438 | * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle. | |
1439 | */ | |
ebde595c | 1440 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 1441 | |
9eeb6dd1 FZ |
1442 | if (!qiov) { |
1443 | ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req); | |
1444 | goto out; | |
1445 | } | |
1446 | ||
61007b31 SH |
1447 | if (offset & (align - 1)) { |
1448 | QEMUIOVector head_qiov; | |
1449 | struct iovec head_iov; | |
1450 | ||
1451 | mark_request_serialising(&req, align); | |
1452 | wait_serialising_requests(&req); | |
1453 | ||
1454 | head_buf = qemu_blockalign(bs, align); | |
1455 | head_iov = (struct iovec) { | |
1456 | .iov_base = head_buf, | |
1457 | .iov_len = align, | |
1458 | }; | |
1459 | qemu_iovec_init_external(&head_qiov, &head_iov, 1); | |
1460 | ||
9a4f4c31 | 1461 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
61007b31 SH |
1462 | ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align, |
1463 | align, &head_qiov, 0); | |
1464 | if (ret < 0) { | |
1465 | goto fail; | |
1466 | } | |
9a4f4c31 | 1467 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
61007b31 SH |
1468 | |
1469 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1470 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1471 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1472 | use_local_qiov = true; | |
1473 | ||
1474 | bytes += offset & (align - 1); | |
1475 | offset = offset & ~(align - 1); | |
117bc3fa PL |
1476 | |
1477 | /* We have read the tail already if the request is smaller | |
1478 | * than one aligned block. | |
1479 | */ | |
1480 | if (bytes < align) { | |
1481 | qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes); | |
1482 | bytes = align; | |
1483 | } | |
61007b31 SH |
1484 | } |
1485 | ||
1486 | if ((offset + bytes) & (align - 1)) { | |
1487 | QEMUIOVector tail_qiov; | |
1488 | struct iovec tail_iov; | |
1489 | size_t tail_bytes; | |
1490 | bool waited; | |
1491 | ||
1492 | mark_request_serialising(&req, align); | |
1493 | waited = wait_serialising_requests(&req); | |
1494 | assert(!waited || !use_local_qiov); | |
1495 | ||
1496 | tail_buf = qemu_blockalign(bs, align); | |
1497 | tail_iov = (struct iovec) { | |
1498 | .iov_base = tail_buf, | |
1499 | .iov_len = align, | |
1500 | }; | |
1501 | qemu_iovec_init_external(&tail_qiov, &tail_iov, 1); | |
1502 | ||
9a4f4c31 | 1503 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
61007b31 SH |
1504 | ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align, |
1505 | align, &tail_qiov, 0); | |
1506 | if (ret < 0) { | |
1507 | goto fail; | |
1508 | } | |
9a4f4c31 | 1509 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
61007b31 SH |
1510 | |
1511 | if (!use_local_qiov) { | |
1512 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1513 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1514 | use_local_qiov = true; | |
1515 | } | |
1516 | ||
1517 | tail_bytes = (offset + bytes) & (align - 1); | |
1518 | qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes); | |
1519 | ||
1520 | bytes = ROUND_UP(bytes, align); | |
1521 | } | |
1522 | ||
cff86b38 | 1523 | ret = bdrv_aligned_pwritev(bs, &req, offset, bytes, align, |
61007b31 SH |
1524 | use_local_qiov ? &local_qiov : qiov, |
1525 | flags); | |
1526 | ||
1527 | fail: | |
61007b31 SH |
1528 | |
1529 | if (use_local_qiov) { | |
1530 | qemu_iovec_destroy(&local_qiov); | |
1531 | } | |
1532 | qemu_vfree(head_buf); | |
1533 | qemu_vfree(tail_buf); | |
9eeb6dd1 FZ |
1534 | out: |
1535 | tracked_request_end(&req); | |
61007b31 SH |
1536 | return ret; |
1537 | } | |
1538 | ||
adad6496 | 1539 | static int coroutine_fn bdrv_co_do_writev(BdrvChild *child, |
61007b31 SH |
1540 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, |
1541 | BdrvRequestFlags flags) | |
1542 | { | |
1543 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
1544 | return -EINVAL; | |
1545 | } | |
1546 | ||
adad6496 | 1547 | return bdrv_co_pwritev(child->bs, sector_num << BDRV_SECTOR_BITS, |
cab3a356 | 1548 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); |
61007b31 SH |
1549 | } |
1550 | ||
25ec177d | 1551 | int coroutine_fn bdrv_co_writev(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
1552 | int nb_sectors, QEMUIOVector *qiov) |
1553 | { | |
25ec177d | 1554 | trace_bdrv_co_writev(child->bs, sector_num, nb_sectors); |
61007b31 | 1555 | |
adad6496 | 1556 | return bdrv_co_do_writev(child, sector_num, nb_sectors, qiov, 0); |
61007b31 SH |
1557 | } |
1558 | ||
74021bc4 EB |
1559 | int coroutine_fn bdrv_co_pwrite_zeroes(BlockDriverState *bs, |
1560 | int64_t offset, int count, | |
1561 | BdrvRequestFlags flags) | |
61007b31 | 1562 | { |
74021bc4 | 1563 | trace_bdrv_co_pwrite_zeroes(bs, offset, count, flags); |
61007b31 SH |
1564 | |
1565 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
1566 | flags &= ~BDRV_REQ_MAY_UNMAP; | |
1567 | } | |
61007b31 | 1568 | |
74021bc4 EB |
1569 | return bdrv_co_pwritev(bs, offset, count, NULL, |
1570 | BDRV_REQ_ZERO_WRITE | flags); | |
61007b31 SH |
1571 | } |
1572 | ||
61007b31 SH |
1573 | typedef struct BdrvCoGetBlockStatusData { |
1574 | BlockDriverState *bs; | |
1575 | BlockDriverState *base; | |
67a0fd2a | 1576 | BlockDriverState **file; |
61007b31 SH |
1577 | int64_t sector_num; |
1578 | int nb_sectors; | |
1579 | int *pnum; | |
1580 | int64_t ret; | |
1581 | bool done; | |
1582 | } BdrvCoGetBlockStatusData; | |
1583 | ||
1584 | /* | |
1585 | * Returns the allocation status of the specified sectors. | |
1586 | * Drivers not implementing the functionality are assumed to not support | |
1587 | * backing files, hence all their sectors are reported as allocated. | |
1588 | * | |
1589 | * If 'sector_num' is beyond the end of the disk image the return value is 0 | |
1590 | * and 'pnum' is set to 0. | |
1591 | * | |
1592 | * 'pnum' is set to the number of sectors (including and immediately following | |
1593 | * the specified sector) that are known to be in the same | |
1594 | * allocated/unallocated state. | |
1595 | * | |
1596 | * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes | |
1597 | * beyond the end of the disk image it will be clamped. | |
67a0fd2a FZ |
1598 | * |
1599 | * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file' | |
1600 | * points to the BDS which the sector range is allocated in. | |
61007b31 SH |
1601 | */ |
1602 | static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, | |
1603 | int64_t sector_num, | |
67a0fd2a FZ |
1604 | int nb_sectors, int *pnum, |
1605 | BlockDriverState **file) | |
61007b31 SH |
1606 | { |
1607 | int64_t total_sectors; | |
1608 | int64_t n; | |
1609 | int64_t ret, ret2; | |
1610 | ||
1611 | total_sectors = bdrv_nb_sectors(bs); | |
1612 | if (total_sectors < 0) { | |
1613 | return total_sectors; | |
1614 | } | |
1615 | ||
1616 | if (sector_num >= total_sectors) { | |
1617 | *pnum = 0; | |
1618 | return 0; | |
1619 | } | |
1620 | ||
1621 | n = total_sectors - sector_num; | |
1622 | if (n < nb_sectors) { | |
1623 | nb_sectors = n; | |
1624 | } | |
1625 | ||
1626 | if (!bs->drv->bdrv_co_get_block_status) { | |
1627 | *pnum = nb_sectors; | |
1628 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; | |
1629 | if (bs->drv->protocol_name) { | |
1630 | ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE); | |
1631 | } | |
1632 | return ret; | |
1633 | } | |
1634 | ||
67a0fd2a FZ |
1635 | *file = NULL; |
1636 | ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum, | |
1637 | file); | |
61007b31 SH |
1638 | if (ret < 0) { |
1639 | *pnum = 0; | |
1640 | return ret; | |
1641 | } | |
1642 | ||
1643 | if (ret & BDRV_BLOCK_RAW) { | |
1644 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
9a4f4c31 | 1645 | return bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS, |
67a0fd2a | 1646 | *pnum, pnum, file); |
61007b31 SH |
1647 | } |
1648 | ||
1649 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
1650 | ret |= BDRV_BLOCK_ALLOCATED; | |
a53f1a95 | 1651 | } else { |
61007b31 SH |
1652 | if (bdrv_unallocated_blocks_are_zero(bs)) { |
1653 | ret |= BDRV_BLOCK_ZERO; | |
760e0063 KW |
1654 | } else if (bs->backing) { |
1655 | BlockDriverState *bs2 = bs->backing->bs; | |
61007b31 SH |
1656 | int64_t nb_sectors2 = bdrv_nb_sectors(bs2); |
1657 | if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) { | |
1658 | ret |= BDRV_BLOCK_ZERO; | |
1659 | } | |
1660 | } | |
1661 | } | |
1662 | ||
ac987b30 | 1663 | if (*file && *file != bs && |
61007b31 SH |
1664 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
1665 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
67a0fd2a | 1666 | BlockDriverState *file2; |
61007b31 SH |
1667 | int file_pnum; |
1668 | ||
ac987b30 | 1669 | ret2 = bdrv_co_get_block_status(*file, ret >> BDRV_SECTOR_BITS, |
67a0fd2a | 1670 | *pnum, &file_pnum, &file2); |
61007b31 SH |
1671 | if (ret2 >= 0) { |
1672 | /* Ignore errors. This is just providing extra information, it | |
1673 | * is useful but not necessary. | |
1674 | */ | |
1675 | if (!file_pnum) { | |
1676 | /* !file_pnum indicates an offset at or beyond the EOF; it is | |
1677 | * perfectly valid for the format block driver to point to such | |
1678 | * offsets, so catch it and mark everything as zero */ | |
1679 | ret |= BDRV_BLOCK_ZERO; | |
1680 | } else { | |
1681 | /* Limit request to the range reported by the protocol driver */ | |
1682 | *pnum = file_pnum; | |
1683 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
1684 | } | |
1685 | } | |
1686 | } | |
1687 | ||
1688 | return ret; | |
1689 | } | |
1690 | ||
ba3f0e25 FZ |
1691 | static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs, |
1692 | BlockDriverState *base, | |
1693 | int64_t sector_num, | |
1694 | int nb_sectors, | |
67a0fd2a FZ |
1695 | int *pnum, |
1696 | BlockDriverState **file) | |
ba3f0e25 FZ |
1697 | { |
1698 | BlockDriverState *p; | |
1699 | int64_t ret = 0; | |
1700 | ||
1701 | assert(bs != base); | |
760e0063 | 1702 | for (p = bs; p != base; p = backing_bs(p)) { |
67a0fd2a | 1703 | ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum, file); |
ba3f0e25 FZ |
1704 | if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) { |
1705 | break; | |
1706 | } | |
1707 | /* [sector_num, pnum] unallocated on this layer, which could be only | |
1708 | * the first part of [sector_num, nb_sectors]. */ | |
1709 | nb_sectors = MIN(nb_sectors, *pnum); | |
1710 | } | |
1711 | return ret; | |
1712 | } | |
1713 | ||
1714 | /* Coroutine wrapper for bdrv_get_block_status_above() */ | |
1715 | static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque) | |
61007b31 SH |
1716 | { |
1717 | BdrvCoGetBlockStatusData *data = opaque; | |
61007b31 | 1718 | |
ba3f0e25 FZ |
1719 | data->ret = bdrv_co_get_block_status_above(data->bs, data->base, |
1720 | data->sector_num, | |
1721 | data->nb_sectors, | |
67a0fd2a FZ |
1722 | data->pnum, |
1723 | data->file); | |
61007b31 SH |
1724 | data->done = true; |
1725 | } | |
1726 | ||
1727 | /* | |
ba3f0e25 | 1728 | * Synchronous wrapper around bdrv_co_get_block_status_above(). |
61007b31 | 1729 | * |
ba3f0e25 | 1730 | * See bdrv_co_get_block_status_above() for details. |
61007b31 | 1731 | */ |
ba3f0e25 FZ |
1732 | int64_t bdrv_get_block_status_above(BlockDriverState *bs, |
1733 | BlockDriverState *base, | |
1734 | int64_t sector_num, | |
67a0fd2a FZ |
1735 | int nb_sectors, int *pnum, |
1736 | BlockDriverState **file) | |
61007b31 SH |
1737 | { |
1738 | Coroutine *co; | |
1739 | BdrvCoGetBlockStatusData data = { | |
1740 | .bs = bs, | |
ba3f0e25 | 1741 | .base = base, |
67a0fd2a | 1742 | .file = file, |
61007b31 SH |
1743 | .sector_num = sector_num, |
1744 | .nb_sectors = nb_sectors, | |
1745 | .pnum = pnum, | |
1746 | .done = false, | |
1747 | }; | |
1748 | ||
1749 | if (qemu_in_coroutine()) { | |
1750 | /* Fast-path if already in coroutine context */ | |
ba3f0e25 | 1751 | bdrv_get_block_status_above_co_entry(&data); |
61007b31 SH |
1752 | } else { |
1753 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
1754 | ||
ba3f0e25 | 1755 | co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry); |
61007b31 SH |
1756 | qemu_coroutine_enter(co, &data); |
1757 | while (!data.done) { | |
1758 | aio_poll(aio_context, true); | |
1759 | } | |
1760 | } | |
1761 | return data.ret; | |
1762 | } | |
1763 | ||
ba3f0e25 FZ |
1764 | int64_t bdrv_get_block_status(BlockDriverState *bs, |
1765 | int64_t sector_num, | |
67a0fd2a FZ |
1766 | int nb_sectors, int *pnum, |
1767 | BlockDriverState **file) | |
ba3f0e25 | 1768 | { |
760e0063 | 1769 | return bdrv_get_block_status_above(bs, backing_bs(bs), |
67a0fd2a | 1770 | sector_num, nb_sectors, pnum, file); |
ba3f0e25 FZ |
1771 | } |
1772 | ||
61007b31 SH |
1773 | int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, |
1774 | int nb_sectors, int *pnum) | |
1775 | { | |
67a0fd2a FZ |
1776 | BlockDriverState *file; |
1777 | int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum, | |
1778 | &file); | |
61007b31 SH |
1779 | if (ret < 0) { |
1780 | return ret; | |
1781 | } | |
1782 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
1783 | } | |
1784 | ||
1785 | /* | |
1786 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
1787 | * | |
1788 | * Return true if the given sector is allocated in any image between | |
1789 | * BASE and TOP (inclusive). BASE can be NULL to check if the given | |
1790 | * sector is allocated in any image of the chain. Return false otherwise. | |
1791 | * | |
1792 | * 'pnum' is set to the number of sectors (including and immediately following | |
1793 | * the specified sector) that are known to be in the same | |
1794 | * allocated/unallocated state. | |
1795 | * | |
1796 | */ | |
1797 | int bdrv_is_allocated_above(BlockDriverState *top, | |
1798 | BlockDriverState *base, | |
1799 | int64_t sector_num, | |
1800 | int nb_sectors, int *pnum) | |
1801 | { | |
1802 | BlockDriverState *intermediate; | |
1803 | int ret, n = nb_sectors; | |
1804 | ||
1805 | intermediate = top; | |
1806 | while (intermediate && intermediate != base) { | |
1807 | int pnum_inter; | |
1808 | ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors, | |
1809 | &pnum_inter); | |
1810 | if (ret < 0) { | |
1811 | return ret; | |
1812 | } else if (ret) { | |
1813 | *pnum = pnum_inter; | |
1814 | return 1; | |
1815 | } | |
1816 | ||
1817 | /* | |
1818 | * [sector_num, nb_sectors] is unallocated on top but intermediate | |
1819 | * might have | |
1820 | * | |
1821 | * [sector_num+x, nr_sectors] allocated. | |
1822 | */ | |
1823 | if (n > pnum_inter && | |
1824 | (intermediate == top || | |
1825 | sector_num + pnum_inter < intermediate->total_sectors)) { | |
1826 | n = pnum_inter; | |
1827 | } | |
1828 | ||
760e0063 | 1829 | intermediate = backing_bs(intermediate); |
61007b31 SH |
1830 | } |
1831 | ||
1832 | *pnum = n; | |
1833 | return 0; | |
1834 | } | |
1835 | ||
1836 | int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num, | |
1837 | const uint8_t *buf, int nb_sectors) | |
1838 | { | |
1839 | BlockDriver *drv = bs->drv; | |
1840 | int ret; | |
1841 | ||
1842 | if (!drv) { | |
1843 | return -ENOMEDIUM; | |
1844 | } | |
1845 | if (!drv->bdrv_write_compressed) { | |
1846 | return -ENOTSUP; | |
1847 | } | |
1848 | ret = bdrv_check_request(bs, sector_num, nb_sectors); | |
1849 | if (ret < 0) { | |
1850 | return ret; | |
1851 | } | |
1852 | ||
1853 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
1854 | ||
1855 | return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors); | |
1856 | } | |
1857 | ||
1a8ae822 KW |
1858 | typedef struct BdrvVmstateCo { |
1859 | BlockDriverState *bs; | |
1860 | QEMUIOVector *qiov; | |
1861 | int64_t pos; | |
1862 | bool is_read; | |
1863 | int ret; | |
1864 | } BdrvVmstateCo; | |
1865 | ||
1866 | static int coroutine_fn | |
1867 | bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
1868 | bool is_read) | |
1869 | { | |
1870 | BlockDriver *drv = bs->drv; | |
1871 | ||
1872 | if (!drv) { | |
1873 | return -ENOMEDIUM; | |
1874 | } else if (drv->bdrv_load_vmstate) { | |
1875 | return is_read ? drv->bdrv_load_vmstate(bs, qiov, pos) | |
1876 | : drv->bdrv_save_vmstate(bs, qiov, pos); | |
1877 | } else if (bs->file) { | |
1878 | return bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read); | |
1879 | } | |
1880 | ||
1881 | return -ENOTSUP; | |
1882 | } | |
1883 | ||
1884 | static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque) | |
1885 | { | |
1886 | BdrvVmstateCo *co = opaque; | |
1887 | co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read); | |
1888 | } | |
1889 | ||
1890 | static inline int | |
1891 | bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
1892 | bool is_read) | |
1893 | { | |
1894 | if (qemu_in_coroutine()) { | |
1895 | return bdrv_co_rw_vmstate(bs, qiov, pos, is_read); | |
1896 | } else { | |
1897 | BdrvVmstateCo data = { | |
1898 | .bs = bs, | |
1899 | .qiov = qiov, | |
1900 | .pos = pos, | |
1901 | .is_read = is_read, | |
1902 | .ret = -EINPROGRESS, | |
1903 | }; | |
1904 | Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry); | |
1905 | ||
1906 | qemu_coroutine_enter(co, &data); | |
1907 | while (data.ret == -EINPROGRESS) { | |
1908 | aio_poll(bdrv_get_aio_context(bs), true); | |
1909 | } | |
1910 | return data.ret; | |
1911 | } | |
1912 | } | |
1913 | ||
61007b31 SH |
1914 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
1915 | int64_t pos, int size) | |
1916 | { | |
1917 | QEMUIOVector qiov; | |
1918 | struct iovec iov = { | |
1919 | .iov_base = (void *) buf, | |
1920 | .iov_len = size, | |
1921 | }; | |
b433d942 | 1922 | int ret; |
61007b31 SH |
1923 | |
1924 | qemu_iovec_init_external(&qiov, &iov, 1); | |
b433d942 KW |
1925 | |
1926 | ret = bdrv_writev_vmstate(bs, &qiov, pos); | |
1927 | if (ret < 0) { | |
1928 | return ret; | |
1929 | } | |
1930 | ||
1931 | return size; | |
61007b31 SH |
1932 | } |
1933 | ||
1934 | int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
1935 | { | |
1a8ae822 | 1936 | return bdrv_rw_vmstate(bs, qiov, pos, false); |
61007b31 SH |
1937 | } |
1938 | ||
1939 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
1940 | int64_t pos, int size) | |
5ddda0b8 KW |
1941 | { |
1942 | QEMUIOVector qiov; | |
1943 | struct iovec iov = { | |
1944 | .iov_base = buf, | |
1945 | .iov_len = size, | |
1946 | }; | |
b433d942 | 1947 | int ret; |
5ddda0b8 KW |
1948 | |
1949 | qemu_iovec_init_external(&qiov, &iov, 1); | |
b433d942 KW |
1950 | ret = bdrv_readv_vmstate(bs, &qiov, pos); |
1951 | if (ret < 0) { | |
1952 | return ret; | |
1953 | } | |
1954 | ||
1955 | return size; | |
5ddda0b8 KW |
1956 | } |
1957 | ||
1958 | int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 1959 | { |
1a8ae822 | 1960 | return bdrv_rw_vmstate(bs, qiov, pos, true); |
61007b31 SH |
1961 | } |
1962 | ||
1963 | /**************************************************************/ | |
1964 | /* async I/Os */ | |
1965 | ||
ebb7af21 | 1966 | BlockAIOCB *bdrv_aio_readv(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
1967 | QEMUIOVector *qiov, int nb_sectors, |
1968 | BlockCompletionFunc *cb, void *opaque) | |
1969 | { | |
ebb7af21 | 1970 | trace_bdrv_aio_readv(child->bs, sector_num, nb_sectors, opaque); |
61007b31 | 1971 | |
adad6496 | 1972 | return bdrv_co_aio_rw_vector(child, sector_num, qiov, nb_sectors, 0, |
61007b31 SH |
1973 | cb, opaque, false); |
1974 | } | |
1975 | ||
0d1049c7 | 1976 | BlockAIOCB *bdrv_aio_writev(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
1977 | QEMUIOVector *qiov, int nb_sectors, |
1978 | BlockCompletionFunc *cb, void *opaque) | |
1979 | { | |
0d1049c7 | 1980 | trace_bdrv_aio_writev(child->bs, sector_num, nb_sectors, opaque); |
61007b31 | 1981 | |
adad6496 | 1982 | return bdrv_co_aio_rw_vector(child, sector_num, qiov, nb_sectors, 0, |
61007b31 SH |
1983 | cb, opaque, true); |
1984 | } | |
1985 | ||
61007b31 SH |
1986 | void bdrv_aio_cancel(BlockAIOCB *acb) |
1987 | { | |
1988 | qemu_aio_ref(acb); | |
1989 | bdrv_aio_cancel_async(acb); | |
1990 | while (acb->refcnt > 1) { | |
1991 | if (acb->aiocb_info->get_aio_context) { | |
1992 | aio_poll(acb->aiocb_info->get_aio_context(acb), true); | |
1993 | } else if (acb->bs) { | |
1994 | aio_poll(bdrv_get_aio_context(acb->bs), true); | |
1995 | } else { | |
1996 | abort(); | |
1997 | } | |
1998 | } | |
1999 | qemu_aio_unref(acb); | |
2000 | } | |
2001 | ||
2002 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2003 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2004 | * In either case the completion callback must be called. */ | |
2005 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2006 | { | |
2007 | if (acb->aiocb_info->cancel_async) { | |
2008 | acb->aiocb_info->cancel_async(acb); | |
2009 | } | |
2010 | } | |
2011 | ||
2012 | /**************************************************************/ | |
2013 | /* async block device emulation */ | |
2014 | ||
41574268 EB |
2015 | typedef struct BlockRequest { |
2016 | union { | |
2017 | /* Used during read, write, trim */ | |
2018 | struct { | |
2019 | int64_t sector; | |
2020 | int nb_sectors; | |
2021 | int flags; | |
2022 | QEMUIOVector *qiov; | |
2023 | }; | |
2024 | /* Used during ioctl */ | |
2025 | struct { | |
2026 | int req; | |
2027 | void *buf; | |
2028 | }; | |
2029 | }; | |
2030 | BlockCompletionFunc *cb; | |
2031 | void *opaque; | |
2032 | ||
2033 | int error; | |
2034 | } BlockRequest; | |
2035 | ||
61007b31 SH |
2036 | typedef struct BlockAIOCBCoroutine { |
2037 | BlockAIOCB common; | |
adad6496 | 2038 | BdrvChild *child; |
61007b31 SH |
2039 | BlockRequest req; |
2040 | bool is_write; | |
2041 | bool need_bh; | |
2042 | bool *done; | |
2043 | QEMUBH* bh; | |
2044 | } BlockAIOCBCoroutine; | |
2045 | ||
2046 | static const AIOCBInfo bdrv_em_co_aiocb_info = { | |
2047 | .aiocb_size = sizeof(BlockAIOCBCoroutine), | |
2048 | }; | |
2049 | ||
2050 | static void bdrv_co_complete(BlockAIOCBCoroutine *acb) | |
2051 | { | |
2052 | if (!acb->need_bh) { | |
2053 | acb->common.cb(acb->common.opaque, acb->req.error); | |
2054 | qemu_aio_unref(acb); | |
2055 | } | |
2056 | } | |
2057 | ||
2058 | static void bdrv_co_em_bh(void *opaque) | |
2059 | { | |
2060 | BlockAIOCBCoroutine *acb = opaque; | |
2061 | ||
2062 | assert(!acb->need_bh); | |
2063 | qemu_bh_delete(acb->bh); | |
2064 | bdrv_co_complete(acb); | |
2065 | } | |
2066 | ||
2067 | static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb) | |
2068 | { | |
2069 | acb->need_bh = false; | |
2070 | if (acb->req.error != -EINPROGRESS) { | |
2071 | BlockDriverState *bs = acb->common.bs; | |
2072 | ||
2073 | acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb); | |
2074 | qemu_bh_schedule(acb->bh); | |
2075 | } | |
2076 | } | |
2077 | ||
2078 | /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */ | |
2079 | static void coroutine_fn bdrv_co_do_rw(void *opaque) | |
2080 | { | |
2081 | BlockAIOCBCoroutine *acb = opaque; | |
61007b31 SH |
2082 | |
2083 | if (!acb->is_write) { | |
adad6496 | 2084 | acb->req.error = bdrv_co_do_readv(acb->child, acb->req.sector, |
61007b31 SH |
2085 | acb->req.nb_sectors, acb->req.qiov, acb->req.flags); |
2086 | } else { | |
adad6496 | 2087 | acb->req.error = bdrv_co_do_writev(acb->child, acb->req.sector, |
61007b31 SH |
2088 | acb->req.nb_sectors, acb->req.qiov, acb->req.flags); |
2089 | } | |
2090 | ||
2091 | bdrv_co_complete(acb); | |
2092 | } | |
2093 | ||
adad6496 | 2094 | static BlockAIOCB *bdrv_co_aio_rw_vector(BdrvChild *child, |
61007b31 SH |
2095 | int64_t sector_num, |
2096 | QEMUIOVector *qiov, | |
2097 | int nb_sectors, | |
2098 | BdrvRequestFlags flags, | |
2099 | BlockCompletionFunc *cb, | |
2100 | void *opaque, | |
2101 | bool is_write) | |
2102 | { | |
2103 | Coroutine *co; | |
2104 | BlockAIOCBCoroutine *acb; | |
2105 | ||
adad6496 KW |
2106 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, child->bs, cb, opaque); |
2107 | acb->child = child; | |
61007b31 SH |
2108 | acb->need_bh = true; |
2109 | acb->req.error = -EINPROGRESS; | |
2110 | acb->req.sector = sector_num; | |
2111 | acb->req.nb_sectors = nb_sectors; | |
2112 | acb->req.qiov = qiov; | |
2113 | acb->req.flags = flags; | |
2114 | acb->is_write = is_write; | |
2115 | ||
2116 | co = qemu_coroutine_create(bdrv_co_do_rw); | |
2117 | qemu_coroutine_enter(co, acb); | |
2118 | ||
2119 | bdrv_co_maybe_schedule_bh(acb); | |
2120 | return &acb->common; | |
2121 | } | |
2122 | ||
2123 | static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) | |
2124 | { | |
2125 | BlockAIOCBCoroutine *acb = opaque; | |
2126 | BlockDriverState *bs = acb->common.bs; | |
2127 | ||
2128 | acb->req.error = bdrv_co_flush(bs); | |
2129 | bdrv_co_complete(acb); | |
2130 | } | |
2131 | ||
2132 | BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs, | |
2133 | BlockCompletionFunc *cb, void *opaque) | |
2134 | { | |
2135 | trace_bdrv_aio_flush(bs, opaque); | |
2136 | ||
2137 | Coroutine *co; | |
2138 | BlockAIOCBCoroutine *acb; | |
2139 | ||
2140 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
2141 | acb->need_bh = true; | |
2142 | acb->req.error = -EINPROGRESS; | |
2143 | ||
2144 | co = qemu_coroutine_create(bdrv_aio_flush_co_entry); | |
2145 | qemu_coroutine_enter(co, acb); | |
2146 | ||
2147 | bdrv_co_maybe_schedule_bh(acb); | |
2148 | return &acb->common; | |
2149 | } | |
2150 | ||
2151 | static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque) | |
2152 | { | |
2153 | BlockAIOCBCoroutine *acb = opaque; | |
2154 | BlockDriverState *bs = acb->common.bs; | |
2155 | ||
2156 | acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors); | |
2157 | bdrv_co_complete(acb); | |
2158 | } | |
2159 | ||
2160 | BlockAIOCB *bdrv_aio_discard(BlockDriverState *bs, | |
2161 | int64_t sector_num, int nb_sectors, | |
2162 | BlockCompletionFunc *cb, void *opaque) | |
2163 | { | |
2164 | Coroutine *co; | |
2165 | BlockAIOCBCoroutine *acb; | |
2166 | ||
2167 | trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque); | |
2168 | ||
2169 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
2170 | acb->need_bh = true; | |
2171 | acb->req.error = -EINPROGRESS; | |
2172 | acb->req.sector = sector_num; | |
2173 | acb->req.nb_sectors = nb_sectors; | |
2174 | co = qemu_coroutine_create(bdrv_aio_discard_co_entry); | |
2175 | qemu_coroutine_enter(co, acb); | |
2176 | ||
2177 | bdrv_co_maybe_schedule_bh(acb); | |
2178 | return &acb->common; | |
2179 | } | |
2180 | ||
2181 | void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, | |
2182 | BlockCompletionFunc *cb, void *opaque) | |
2183 | { | |
2184 | BlockAIOCB *acb; | |
2185 | ||
c84b3192 | 2186 | acb = g_malloc(aiocb_info->aiocb_size); |
61007b31 SH |
2187 | acb->aiocb_info = aiocb_info; |
2188 | acb->bs = bs; | |
2189 | acb->cb = cb; | |
2190 | acb->opaque = opaque; | |
2191 | acb->refcnt = 1; | |
2192 | return acb; | |
2193 | } | |
2194 | ||
2195 | void qemu_aio_ref(void *p) | |
2196 | { | |
2197 | BlockAIOCB *acb = p; | |
2198 | acb->refcnt++; | |
2199 | } | |
2200 | ||
2201 | void qemu_aio_unref(void *p) | |
2202 | { | |
2203 | BlockAIOCB *acb = p; | |
2204 | assert(acb->refcnt > 0); | |
2205 | if (--acb->refcnt == 0) { | |
c84b3192 | 2206 | g_free(acb); |
61007b31 SH |
2207 | } |
2208 | } | |
2209 | ||
2210 | /**************************************************************/ | |
2211 | /* Coroutine block device emulation */ | |
2212 | ||
61007b31 SH |
2213 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) |
2214 | { | |
2215 | RwCo *rwco = opaque; | |
2216 | ||
2217 | rwco->ret = bdrv_co_flush(rwco->bs); | |
2218 | } | |
2219 | ||
2220 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
2221 | { | |
2222 | int ret; | |
cdb5e315 | 2223 | BdrvTrackedRequest req; |
61007b31 | 2224 | |
1b6bc94d DA |
2225 | if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) || |
2226 | bdrv_is_sg(bs)) { | |
61007b31 SH |
2227 | return 0; |
2228 | } | |
2229 | ||
cdb5e315 | 2230 | tracked_request_begin(&req, bs, 0, 0, BDRV_TRACKED_FLUSH); |
c32b82af PD |
2231 | |
2232 | /* Write back all layers by calling one driver function */ | |
2233 | if (bs->drv->bdrv_co_flush) { | |
2234 | ret = bs->drv->bdrv_co_flush(bs); | |
2235 | goto out; | |
2236 | } | |
2237 | ||
61007b31 SH |
2238 | /* Write back cached data to the OS even with cache=unsafe */ |
2239 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); | |
2240 | if (bs->drv->bdrv_co_flush_to_os) { | |
2241 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
2242 | if (ret < 0) { | |
cdb5e315 | 2243 | goto out; |
61007b31 SH |
2244 | } |
2245 | } | |
2246 | ||
2247 | /* But don't actually force it to the disk with cache=unsafe */ | |
2248 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
2249 | goto flush_parent; | |
2250 | } | |
2251 | ||
2252 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); | |
2253 | if (bs->drv->bdrv_co_flush_to_disk) { | |
2254 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
2255 | } else if (bs->drv->bdrv_aio_flush) { | |
2256 | BlockAIOCB *acb; | |
2257 | CoroutineIOCompletion co = { | |
2258 | .coroutine = qemu_coroutine_self(), | |
2259 | }; | |
2260 | ||
2261 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
2262 | if (acb == NULL) { | |
2263 | ret = -EIO; | |
2264 | } else { | |
2265 | qemu_coroutine_yield(); | |
2266 | ret = co.ret; | |
2267 | } | |
2268 | } else { | |
2269 | /* | |
2270 | * Some block drivers always operate in either writethrough or unsafe | |
2271 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
2272 | * know how the server works (because the behaviour is hardcoded or | |
2273 | * depends on server-side configuration), so we can't ensure that | |
2274 | * everything is safe on disk. Returning an error doesn't work because | |
2275 | * that would break guests even if the server operates in writethrough | |
2276 | * mode. | |
2277 | * | |
2278 | * Let's hope the user knows what he's doing. | |
2279 | */ | |
2280 | ret = 0; | |
2281 | } | |
2282 | if (ret < 0) { | |
cdb5e315 | 2283 | goto out; |
61007b31 SH |
2284 | } |
2285 | ||
2286 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
2287 | * in the case of cache=unsafe, so there are no useless flushes. | |
2288 | */ | |
2289 | flush_parent: | |
cdb5e315 FZ |
2290 | ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0; |
2291 | out: | |
2292 | tracked_request_end(&req); | |
2293 | return ret; | |
61007b31 SH |
2294 | } |
2295 | ||
2296 | int bdrv_flush(BlockDriverState *bs) | |
2297 | { | |
2298 | Coroutine *co; | |
2299 | RwCo rwco = { | |
2300 | .bs = bs, | |
2301 | .ret = NOT_DONE, | |
2302 | }; | |
2303 | ||
2304 | if (qemu_in_coroutine()) { | |
2305 | /* Fast-path if already in coroutine context */ | |
2306 | bdrv_flush_co_entry(&rwco); | |
2307 | } else { | |
2308 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2309 | ||
2310 | co = qemu_coroutine_create(bdrv_flush_co_entry); | |
2311 | qemu_coroutine_enter(co, &rwco); | |
2312 | while (rwco.ret == NOT_DONE) { | |
2313 | aio_poll(aio_context, true); | |
2314 | } | |
2315 | } | |
2316 | ||
2317 | return rwco.ret; | |
2318 | } | |
2319 | ||
2320 | typedef struct DiscardCo { | |
2321 | BlockDriverState *bs; | |
2322 | int64_t sector_num; | |
2323 | int nb_sectors; | |
2324 | int ret; | |
2325 | } DiscardCo; | |
2326 | static void coroutine_fn bdrv_discard_co_entry(void *opaque) | |
2327 | { | |
2328 | DiscardCo *rwco = opaque; | |
2329 | ||
2330 | rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors); | |
2331 | } | |
2332 | ||
2333 | int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num, | |
2334 | int nb_sectors) | |
2335 | { | |
b1066c87 | 2336 | BdrvTrackedRequest req; |
61007b31 SH |
2337 | int max_discard, ret; |
2338 | ||
2339 | if (!bs->drv) { | |
2340 | return -ENOMEDIUM; | |
2341 | } | |
2342 | ||
2343 | ret = bdrv_check_request(bs, sector_num, nb_sectors); | |
2344 | if (ret < 0) { | |
2345 | return ret; | |
2346 | } else if (bs->read_only) { | |
eaf5fe2d | 2347 | return -EPERM; |
61007b31 | 2348 | } |
04c01a5c | 2349 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
61007b31 | 2350 | |
61007b31 SH |
2351 | /* Do nothing if disabled. */ |
2352 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
2353 | return 0; | |
2354 | } | |
2355 | ||
2356 | if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) { | |
2357 | return 0; | |
2358 | } | |
2359 | ||
3a36e474 DL |
2360 | tracked_request_begin(&req, bs, sector_num << BDRV_SECTOR_BITS, |
2361 | nb_sectors << BDRV_SECTOR_BITS, BDRV_TRACKED_DISCARD); | |
50824995 | 2362 | |
ec050f77 DL |
2363 | ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req); |
2364 | if (ret < 0) { | |
2365 | goto out; | |
2366 | } | |
2367 | ||
b9f7855a EB |
2368 | max_discard = MIN_NON_ZERO(bs->bl.max_pdiscard >> BDRV_SECTOR_BITS, |
2369 | BDRV_REQUEST_MAX_SECTORS); | |
61007b31 SH |
2370 | while (nb_sectors > 0) { |
2371 | int ret; | |
2372 | int num = nb_sectors; | |
b9f7855a | 2373 | int discard_alignment = bs->bl.pdiscard_alignment >> BDRV_SECTOR_BITS; |
61007b31 SH |
2374 | |
2375 | /* align request */ | |
b9f7855a EB |
2376 | if (discard_alignment && |
2377 | num >= discard_alignment && | |
2378 | sector_num % discard_alignment) { | |
2379 | if (num > discard_alignment) { | |
2380 | num = discard_alignment; | |
61007b31 | 2381 | } |
b9f7855a | 2382 | num -= sector_num % discard_alignment; |
61007b31 SH |
2383 | } |
2384 | ||
2385 | /* limit request size */ | |
2386 | if (num > max_discard) { | |
2387 | num = max_discard; | |
2388 | } | |
2389 | ||
2390 | if (bs->drv->bdrv_co_discard) { | |
2391 | ret = bs->drv->bdrv_co_discard(bs, sector_num, num); | |
2392 | } else { | |
2393 | BlockAIOCB *acb; | |
2394 | CoroutineIOCompletion co = { | |
2395 | .coroutine = qemu_coroutine_self(), | |
2396 | }; | |
2397 | ||
2398 | acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors, | |
2399 | bdrv_co_io_em_complete, &co); | |
2400 | if (acb == NULL) { | |
b1066c87 FZ |
2401 | ret = -EIO; |
2402 | goto out; | |
61007b31 SH |
2403 | } else { |
2404 | qemu_coroutine_yield(); | |
2405 | ret = co.ret; | |
2406 | } | |
2407 | } | |
2408 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 2409 | goto out; |
61007b31 SH |
2410 | } |
2411 | ||
2412 | sector_num += num; | |
2413 | nb_sectors -= num; | |
2414 | } | |
b1066c87 FZ |
2415 | ret = 0; |
2416 | out: | |
968d8b06 DL |
2417 | bdrv_set_dirty(bs, req.offset >> BDRV_SECTOR_BITS, |
2418 | req.bytes >> BDRV_SECTOR_BITS); | |
b1066c87 FZ |
2419 | tracked_request_end(&req); |
2420 | return ret; | |
61007b31 SH |
2421 | } |
2422 | ||
2423 | int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors) | |
2424 | { | |
2425 | Coroutine *co; | |
2426 | DiscardCo rwco = { | |
2427 | .bs = bs, | |
2428 | .sector_num = sector_num, | |
2429 | .nb_sectors = nb_sectors, | |
2430 | .ret = NOT_DONE, | |
2431 | }; | |
2432 | ||
2433 | if (qemu_in_coroutine()) { | |
2434 | /* Fast-path if already in coroutine context */ | |
2435 | bdrv_discard_co_entry(&rwco); | |
2436 | } else { | |
2437 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2438 | ||
2439 | co = qemu_coroutine_create(bdrv_discard_co_entry); | |
2440 | qemu_coroutine_enter(co, &rwco); | |
2441 | while (rwco.ret == NOT_DONE) { | |
2442 | aio_poll(aio_context, true); | |
2443 | } | |
2444 | } | |
2445 | ||
2446 | return rwco.ret; | |
2447 | } | |
2448 | ||
5c5ae76a | 2449 | static int bdrv_co_do_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
2450 | { |
2451 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
2452 | BdrvTrackedRequest tracked_req; |
2453 | CoroutineIOCompletion co = { | |
2454 | .coroutine = qemu_coroutine_self(), | |
2455 | }; | |
2456 | BlockAIOCB *acb; | |
61007b31 | 2457 | |
5c5ae76a FZ |
2458 | tracked_request_begin(&tracked_req, bs, 0, 0, BDRV_TRACKED_IOCTL); |
2459 | if (!drv || !drv->bdrv_aio_ioctl) { | |
2460 | co.ret = -ENOTSUP; | |
2461 | goto out; | |
2462 | } | |
2463 | ||
2464 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
2465 | if (!acb) { | |
c8a9fd80 FZ |
2466 | co.ret = -ENOTSUP; |
2467 | goto out; | |
5c5ae76a FZ |
2468 | } |
2469 | qemu_coroutine_yield(); | |
2470 | out: | |
2471 | tracked_request_end(&tracked_req); | |
2472 | return co.ret; | |
2473 | } | |
2474 | ||
2475 | typedef struct { | |
2476 | BlockDriverState *bs; | |
2477 | int req; | |
2478 | void *buf; | |
2479 | int ret; | |
2480 | } BdrvIoctlCoData; | |
2481 | ||
2482 | static void coroutine_fn bdrv_co_ioctl_entry(void *opaque) | |
2483 | { | |
2484 | BdrvIoctlCoData *data = opaque; | |
2485 | data->ret = bdrv_co_do_ioctl(data->bs, data->req, data->buf); | |
2486 | } | |
2487 | ||
2488 | /* needed for generic scsi interface */ | |
2489 | int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
2490 | { | |
2491 | BdrvIoctlCoData data = { | |
2492 | .bs = bs, | |
2493 | .req = req, | |
2494 | .buf = buf, | |
2495 | .ret = -EINPROGRESS, | |
2496 | }; | |
2497 | ||
2498 | if (qemu_in_coroutine()) { | |
2499 | /* Fast-path if already in coroutine context */ | |
2500 | bdrv_co_ioctl_entry(&data); | |
2501 | } else { | |
2502 | Coroutine *co = qemu_coroutine_create(bdrv_co_ioctl_entry); | |
ba889444 | 2503 | |
5c5ae76a | 2504 | qemu_coroutine_enter(co, &data); |
ba889444 PB |
2505 | while (data.ret == -EINPROGRESS) { |
2506 | aio_poll(bdrv_get_aio_context(bs), true); | |
2507 | } | |
5c5ae76a FZ |
2508 | } |
2509 | return data.ret; | |
2510 | } | |
2511 | ||
2512 | static void coroutine_fn bdrv_co_aio_ioctl_entry(void *opaque) | |
2513 | { | |
2514 | BlockAIOCBCoroutine *acb = opaque; | |
2515 | acb->req.error = bdrv_co_do_ioctl(acb->common.bs, | |
2516 | acb->req.req, acb->req.buf); | |
2517 | bdrv_co_complete(acb); | |
61007b31 SH |
2518 | } |
2519 | ||
2520 | BlockAIOCB *bdrv_aio_ioctl(BlockDriverState *bs, | |
2521 | unsigned long int req, void *buf, | |
2522 | BlockCompletionFunc *cb, void *opaque) | |
2523 | { | |
5c5ae76a FZ |
2524 | BlockAIOCBCoroutine *acb = qemu_aio_get(&bdrv_em_co_aiocb_info, |
2525 | bs, cb, opaque); | |
2526 | Coroutine *co; | |
61007b31 | 2527 | |
5c5ae76a FZ |
2528 | acb->need_bh = true; |
2529 | acb->req.error = -EINPROGRESS; | |
2530 | acb->req.req = req; | |
2531 | acb->req.buf = buf; | |
2532 | co = qemu_coroutine_create(bdrv_co_aio_ioctl_entry); | |
2533 | qemu_coroutine_enter(co, acb); | |
2534 | ||
2535 | bdrv_co_maybe_schedule_bh(acb); | |
2536 | return &acb->common; | |
61007b31 SH |
2537 | } |
2538 | ||
2539 | void *qemu_blockalign(BlockDriverState *bs, size_t size) | |
2540 | { | |
2541 | return qemu_memalign(bdrv_opt_mem_align(bs), size); | |
2542 | } | |
2543 | ||
2544 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
2545 | { | |
2546 | return memset(qemu_blockalign(bs, size), 0, size); | |
2547 | } | |
2548 | ||
2549 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
2550 | { | |
2551 | size_t align = bdrv_opt_mem_align(bs); | |
2552 | ||
2553 | /* Ensure that NULL is never returned on success */ | |
2554 | assert(align > 0); | |
2555 | if (size == 0) { | |
2556 | size = align; | |
2557 | } | |
2558 | ||
2559 | return qemu_try_memalign(align, size); | |
2560 | } | |
2561 | ||
2562 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
2563 | { | |
2564 | void *mem = qemu_try_blockalign(bs, size); | |
2565 | ||
2566 | if (mem) { | |
2567 | memset(mem, 0, size); | |
2568 | } | |
2569 | ||
2570 | return mem; | |
2571 | } | |
2572 | ||
2573 | /* | |
2574 | * Check if all memory in this vector is sector aligned. | |
2575 | */ | |
2576 | bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
2577 | { | |
2578 | int i; | |
4196d2f0 | 2579 | size_t alignment = bdrv_min_mem_align(bs); |
61007b31 SH |
2580 | |
2581 | for (i = 0; i < qiov->niov; i++) { | |
2582 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
2583 | return false; | |
2584 | } | |
2585 | if (qiov->iov[i].iov_len % alignment) { | |
2586 | return false; | |
2587 | } | |
2588 | } | |
2589 | ||
2590 | return true; | |
2591 | } | |
2592 | ||
2593 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
2594 | NotifierWithReturn *notifier) | |
2595 | { | |
2596 | notifier_with_return_list_add(&bs->before_write_notifiers, notifier); | |
2597 | } | |
2598 | ||
2599 | void bdrv_io_plug(BlockDriverState *bs) | |
2600 | { | |
6b98bd64 PB |
2601 | BdrvChild *child; |
2602 | ||
2603 | QLIST_FOREACH(child, &bs->children, next) { | |
2604 | bdrv_io_plug(child->bs); | |
2605 | } | |
2606 | ||
2607 | if (bs->io_plugged++ == 0 && bs->io_plug_disabled == 0) { | |
2608 | BlockDriver *drv = bs->drv; | |
2609 | if (drv && drv->bdrv_io_plug) { | |
2610 | drv->bdrv_io_plug(bs); | |
2611 | } | |
61007b31 SH |
2612 | } |
2613 | } | |
2614 | ||
2615 | void bdrv_io_unplug(BlockDriverState *bs) | |
2616 | { | |
6b98bd64 PB |
2617 | BdrvChild *child; |
2618 | ||
2619 | assert(bs->io_plugged); | |
2620 | if (--bs->io_plugged == 0 && bs->io_plug_disabled == 0) { | |
2621 | BlockDriver *drv = bs->drv; | |
2622 | if (drv && drv->bdrv_io_unplug) { | |
2623 | drv->bdrv_io_unplug(bs); | |
2624 | } | |
2625 | } | |
2626 | ||
2627 | QLIST_FOREACH(child, &bs->children, next) { | |
2628 | bdrv_io_unplug(child->bs); | |
61007b31 SH |
2629 | } |
2630 | } | |
2631 | ||
6b98bd64 | 2632 | void bdrv_io_unplugged_begin(BlockDriverState *bs) |
61007b31 | 2633 | { |
6b98bd64 PB |
2634 | BdrvChild *child; |
2635 | ||
2636 | if (bs->io_plug_disabled++ == 0 && bs->io_plugged > 0) { | |
2637 | BlockDriver *drv = bs->drv; | |
2638 | if (drv && drv->bdrv_io_unplug) { | |
2639 | drv->bdrv_io_unplug(bs); | |
2640 | } | |
2641 | } | |
2642 | ||
2643 | QLIST_FOREACH(child, &bs->children, next) { | |
2644 | bdrv_io_unplugged_begin(child->bs); | |
2645 | } | |
2646 | } | |
2647 | ||
2648 | void bdrv_io_unplugged_end(BlockDriverState *bs) | |
2649 | { | |
2650 | BdrvChild *child; | |
2651 | ||
2652 | assert(bs->io_plug_disabled); | |
2653 | QLIST_FOREACH(child, &bs->children, next) { | |
2654 | bdrv_io_unplugged_end(child->bs); | |
2655 | } | |
2656 | ||
2657 | if (--bs->io_plug_disabled == 0 && bs->io_plugged > 0) { | |
2658 | BlockDriver *drv = bs->drv; | |
2659 | if (drv && drv->bdrv_io_plug) { | |
2660 | drv->bdrv_io_plug(bs); | |
2661 | } | |
61007b31 SH |
2662 | } |
2663 | } |