]>
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 | ||
b15404e0 EB |
36 | static BlockAIOCB *bdrv_co_aio_prw_vector(BdrvChild *child, |
37 | int64_t offset, | |
38 | QEMUIOVector *qiov, | |
39 | BdrvRequestFlags flags, | |
40 | BlockCompletionFunc *cb, | |
41 | void *opaque, | |
42 | bool is_write); | |
61007b31 | 43 | static void coroutine_fn bdrv_co_do_rw(void *opaque); |
d05aa8bb EB |
44 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
45 | int64_t offset, int count, BdrvRequestFlags flags); | |
61007b31 | 46 | |
c2066af0 | 47 | static void bdrv_parent_drained_begin(BlockDriverState *bs) |
61007b31 | 48 | { |
c2066af0 | 49 | BdrvChild *c; |
27ccdd52 | 50 | |
c2066af0 KW |
51 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
52 | if (c->role->drained_begin) { | |
53 | c->role->drained_begin(c); | |
54 | } | |
ce0f1412 PB |
55 | } |
56 | } | |
61007b31 | 57 | |
c2066af0 | 58 | static void bdrv_parent_drained_end(BlockDriverState *bs) |
ce0f1412 | 59 | { |
c2066af0 | 60 | BdrvChild *c; |
27ccdd52 | 61 | |
c2066af0 KW |
62 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
63 | if (c->role->drained_end) { | |
64 | c->role->drained_end(c); | |
65 | } | |
27ccdd52 | 66 | } |
61007b31 SH |
67 | } |
68 | ||
d9e0dfa2 EB |
69 | static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) |
70 | { | |
71 | dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); | |
72 | dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer); | |
73 | dst->opt_mem_alignment = MAX(dst->opt_mem_alignment, | |
74 | src->opt_mem_alignment); | |
75 | dst->min_mem_alignment = MAX(dst->min_mem_alignment, | |
76 | src->min_mem_alignment); | |
77 | dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov); | |
78 | } | |
79 | ||
61007b31 SH |
80 | void bdrv_refresh_limits(BlockDriverState *bs, Error **errp) |
81 | { | |
82 | BlockDriver *drv = bs->drv; | |
83 | Error *local_err = NULL; | |
84 | ||
85 | memset(&bs->bl, 0, sizeof(bs->bl)); | |
86 | ||
87 | if (!drv) { | |
88 | return; | |
89 | } | |
90 | ||
79ba8c98 | 91 | /* Default alignment based on whether driver has byte interface */ |
a5b8dd2c | 92 | bs->bl.request_alignment = drv->bdrv_co_preadv ? 1 : 512; |
79ba8c98 | 93 | |
61007b31 SH |
94 | /* Take some limits from the children as a default */ |
95 | if (bs->file) { | |
9a4f4c31 | 96 | bdrv_refresh_limits(bs->file->bs, &local_err); |
61007b31 SH |
97 | if (local_err) { |
98 | error_propagate(errp, local_err); | |
99 | return; | |
100 | } | |
d9e0dfa2 | 101 | bdrv_merge_limits(&bs->bl, &bs->file->bs->bl); |
61007b31 | 102 | } else { |
4196d2f0 | 103 | bs->bl.min_mem_alignment = 512; |
459b4e66 | 104 | bs->bl.opt_mem_alignment = getpagesize(); |
bd44feb7 SH |
105 | |
106 | /* Safe default since most protocols use readv()/writev()/etc */ | |
107 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
108 | } |
109 | ||
760e0063 KW |
110 | if (bs->backing) { |
111 | bdrv_refresh_limits(bs->backing->bs, &local_err); | |
61007b31 SH |
112 | if (local_err) { |
113 | error_propagate(errp, local_err); | |
114 | return; | |
115 | } | |
d9e0dfa2 | 116 | bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl); |
61007b31 SH |
117 | } |
118 | ||
119 | /* Then let the driver override it */ | |
120 | if (drv->bdrv_refresh_limits) { | |
121 | drv->bdrv_refresh_limits(bs, errp); | |
122 | } | |
123 | } | |
124 | ||
125 | /** | |
126 | * The copy-on-read flag is actually a reference count so multiple users may | |
127 | * use the feature without worrying about clobbering its previous state. | |
128 | * Copy-on-read stays enabled until all users have called to disable it. | |
129 | */ | |
130 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
131 | { | |
132 | bs->copy_on_read++; | |
133 | } | |
134 | ||
135 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
136 | { | |
137 | assert(bs->copy_on_read > 0); | |
138 | bs->copy_on_read--; | |
139 | } | |
140 | ||
141 | /* Check if any requests are in-flight (including throttled requests) */ | |
439db28c | 142 | bool bdrv_requests_pending(BlockDriverState *bs) |
61007b31 | 143 | { |
37a639a7 KW |
144 | BdrvChild *child; |
145 | ||
99723548 | 146 | if (atomic_read(&bs->in_flight)) { |
61007b31 SH |
147 | return true; |
148 | } | |
37a639a7 KW |
149 | |
150 | QLIST_FOREACH(child, &bs->children, next) { | |
151 | if (bdrv_requests_pending(child->bs)) { | |
152 | return true; | |
153 | } | |
61007b31 | 154 | } |
37a639a7 | 155 | |
61007b31 SH |
156 | return false; |
157 | } | |
158 | ||
d42cf288 | 159 | static bool bdrv_drain_recurse(BlockDriverState *bs) |
67da1dc5 FZ |
160 | { |
161 | BdrvChild *child; | |
d42cf288 PB |
162 | bool waited; |
163 | ||
88b062c2 | 164 | waited = BDRV_POLL_WHILE(bs, atomic_read(&bs->in_flight) > 0); |
67da1dc5 FZ |
165 | |
166 | if (bs->drv && bs->drv->bdrv_drain) { | |
167 | bs->drv->bdrv_drain(bs); | |
168 | } | |
d42cf288 | 169 | |
67da1dc5 | 170 | QLIST_FOREACH(child, &bs->children, next) { |
d42cf288 | 171 | waited |= bdrv_drain_recurse(child->bs); |
67da1dc5 | 172 | } |
d42cf288 PB |
173 | |
174 | return waited; | |
67da1dc5 FZ |
175 | } |
176 | ||
a77fd4bb FZ |
177 | typedef struct { |
178 | Coroutine *co; | |
179 | BlockDriverState *bs; | |
a77fd4bb FZ |
180 | bool done; |
181 | } BdrvCoDrainData; | |
182 | ||
183 | static void bdrv_co_drain_bh_cb(void *opaque) | |
184 | { | |
185 | BdrvCoDrainData *data = opaque; | |
186 | Coroutine *co = data->co; | |
99723548 | 187 | BlockDriverState *bs = data->bs; |
a77fd4bb | 188 | |
99723548 | 189 | bdrv_dec_in_flight(bs); |
d42cf288 | 190 | bdrv_drained_begin(bs); |
a77fd4bb | 191 | data->done = true; |
0b8b8753 | 192 | qemu_coroutine_enter(co); |
a77fd4bb FZ |
193 | } |
194 | ||
b6e84c97 | 195 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs) |
a77fd4bb FZ |
196 | { |
197 | BdrvCoDrainData data; | |
198 | ||
199 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
200 | * other coroutines run if they were queued from | |
201 | * qemu_co_queue_run_restart(). */ | |
202 | ||
203 | assert(qemu_in_coroutine()); | |
204 | data = (BdrvCoDrainData) { | |
205 | .co = qemu_coroutine_self(), | |
206 | .bs = bs, | |
207 | .done = false, | |
a77fd4bb | 208 | }; |
99723548 | 209 | bdrv_inc_in_flight(bs); |
fffb6e12 PB |
210 | aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), |
211 | bdrv_co_drain_bh_cb, &data); | |
a77fd4bb FZ |
212 | |
213 | qemu_coroutine_yield(); | |
214 | /* If we are resumed from some other event (such as an aio completion or a | |
215 | * timer callback), it is a bug in the caller that should be fixed. */ | |
216 | assert(data.done); | |
217 | } | |
218 | ||
6820643f KW |
219 | void bdrv_drained_begin(BlockDriverState *bs) |
220 | { | |
d42cf288 PB |
221 | if (qemu_in_coroutine()) { |
222 | bdrv_co_yield_to_drain(bs); | |
223 | return; | |
224 | } | |
225 | ||
6820643f KW |
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); | |
6820643f KW |
233 | bdrv_io_unplugged_end(bs); |
234 | } | |
235 | ||
236 | void bdrv_drained_end(BlockDriverState *bs) | |
237 | { | |
238 | assert(bs->quiesce_counter > 0); | |
239 | if (--bs->quiesce_counter > 0) { | |
240 | return; | |
241 | } | |
242 | ||
243 | bdrv_parent_drained_end(bs); | |
244 | aio_enable_external(bdrv_get_aio_context(bs)); | |
245 | } | |
246 | ||
61007b31 | 247 | /* |
67da1dc5 FZ |
248 | * Wait for pending requests to complete on a single BlockDriverState subtree, |
249 | * and suspend block driver's internal I/O until next request arrives. | |
61007b31 | 250 | * |
61007b31 SH |
251 | * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState |
252 | * AioContext. | |
7a63f3cd SH |
253 | * |
254 | * Only this BlockDriverState's AioContext is run, so in-flight requests must | |
255 | * not depend on events in other AioContexts. In that case, use | |
256 | * bdrv_drain_all() instead. | |
61007b31 | 257 | */ |
b6e84c97 | 258 | void coroutine_fn bdrv_co_drain(BlockDriverState *bs) |
61007b31 | 259 | { |
6820643f KW |
260 | assert(qemu_in_coroutine()); |
261 | bdrv_drained_begin(bs); | |
262 | bdrv_drained_end(bs); | |
b6e84c97 | 263 | } |
f406c03c | 264 | |
b6e84c97 PB |
265 | void bdrv_drain(BlockDriverState *bs) |
266 | { | |
6820643f KW |
267 | bdrv_drained_begin(bs); |
268 | bdrv_drained_end(bs); | |
61007b31 SH |
269 | } |
270 | ||
271 | /* | |
272 | * Wait for pending requests to complete across all BlockDriverStates | |
273 | * | |
274 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
275 | * after calling this function. | |
c0778f66 AG |
276 | * |
277 | * This pauses all block jobs and disables external clients. It must | |
278 | * be paired with bdrv_drain_all_end(). | |
279 | * | |
280 | * NOTE: no new block jobs or BlockDriverStates can be created between | |
281 | * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls. | |
61007b31 | 282 | */ |
c0778f66 | 283 | void bdrv_drain_all_begin(void) |
61007b31 SH |
284 | { |
285 | /* Always run first iteration so any pending completion BHs run */ | |
99723548 | 286 | bool waited = true; |
7c8eece4 | 287 | BlockDriverState *bs; |
88be7b4b | 288 | BdrvNextIterator it; |
eb1364ce | 289 | BlockJob *job = NULL; |
f406c03c | 290 | GSList *aio_ctxs = NULL, *ctx; |
61007b31 | 291 | |
eb1364ce AG |
292 | while ((job = block_job_next(job))) { |
293 | AioContext *aio_context = blk_get_aio_context(job->blk); | |
294 | ||
295 | aio_context_acquire(aio_context); | |
296 | block_job_pause(job); | |
297 | aio_context_release(aio_context); | |
298 | } | |
299 | ||
88be7b4b | 300 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
61007b31 SH |
301 | AioContext *aio_context = bdrv_get_aio_context(bs); |
302 | ||
303 | aio_context_acquire(aio_context); | |
c2066af0 | 304 | bdrv_parent_drained_begin(bs); |
6b98bd64 | 305 | bdrv_io_unplugged_begin(bs); |
c0778f66 | 306 | aio_disable_external(aio_context); |
61007b31 | 307 | aio_context_release(aio_context); |
f406c03c | 308 | |
764ba3ae | 309 | if (!g_slist_find(aio_ctxs, aio_context)) { |
f406c03c AY |
310 | aio_ctxs = g_slist_prepend(aio_ctxs, aio_context); |
311 | } | |
61007b31 SH |
312 | } |
313 | ||
7a63f3cd SH |
314 | /* Note that completion of an asynchronous I/O operation can trigger any |
315 | * number of other I/O operations on other devices---for example a | |
316 | * coroutine can submit an I/O request to another device in response to | |
317 | * request completion. Therefore we must keep looping until there was no | |
318 | * more activity rather than simply draining each device independently. | |
319 | */ | |
99723548 PB |
320 | while (waited) { |
321 | waited = false; | |
61007b31 | 322 | |
f406c03c AY |
323 | for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) { |
324 | AioContext *aio_context = ctx->data; | |
61007b31 SH |
325 | |
326 | aio_context_acquire(aio_context); | |
88be7b4b | 327 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
f406c03c | 328 | if (aio_context == bdrv_get_aio_context(bs)) { |
d42cf288 | 329 | waited |= bdrv_drain_recurse(bs); |
f406c03c AY |
330 | } |
331 | } | |
61007b31 SH |
332 | aio_context_release(aio_context); |
333 | } | |
334 | } | |
335 | ||
c0778f66 AG |
336 | g_slist_free(aio_ctxs); |
337 | } | |
338 | ||
339 | void bdrv_drain_all_end(void) | |
340 | { | |
341 | BlockDriverState *bs; | |
342 | BdrvNextIterator it; | |
343 | BlockJob *job = NULL; | |
344 | ||
88be7b4b | 345 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
61007b31 SH |
346 | AioContext *aio_context = bdrv_get_aio_context(bs); |
347 | ||
348 | aio_context_acquire(aio_context); | |
c0778f66 | 349 | aio_enable_external(aio_context); |
6b98bd64 | 350 | bdrv_io_unplugged_end(bs); |
c2066af0 | 351 | bdrv_parent_drained_end(bs); |
61007b31 SH |
352 | aio_context_release(aio_context); |
353 | } | |
eb1364ce | 354 | |
eb1364ce AG |
355 | while ((job = block_job_next(job))) { |
356 | AioContext *aio_context = blk_get_aio_context(job->blk); | |
357 | ||
358 | aio_context_acquire(aio_context); | |
359 | block_job_resume(job); | |
360 | aio_context_release(aio_context); | |
361 | } | |
61007b31 SH |
362 | } |
363 | ||
c0778f66 AG |
364 | void bdrv_drain_all(void) |
365 | { | |
366 | bdrv_drain_all_begin(); | |
367 | bdrv_drain_all_end(); | |
368 | } | |
369 | ||
61007b31 SH |
370 | /** |
371 | * Remove an active request from the tracked requests list | |
372 | * | |
373 | * This function should be called when a tracked request is completing. | |
374 | */ | |
375 | static void tracked_request_end(BdrvTrackedRequest *req) | |
376 | { | |
377 | if (req->serialising) { | |
378 | req->bs->serialising_in_flight--; | |
379 | } | |
380 | ||
381 | QLIST_REMOVE(req, list); | |
382 | qemu_co_queue_restart_all(&req->wait_queue); | |
383 | } | |
384 | ||
385 | /** | |
386 | * Add an active request to the tracked requests list | |
387 | */ | |
388 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
389 | BlockDriverState *bs, | |
390 | int64_t offset, | |
ebde595c FZ |
391 | unsigned int bytes, |
392 | enum BdrvTrackedRequestType type) | |
61007b31 SH |
393 | { |
394 | *req = (BdrvTrackedRequest){ | |
395 | .bs = bs, | |
396 | .offset = offset, | |
397 | .bytes = bytes, | |
ebde595c | 398 | .type = type, |
61007b31 SH |
399 | .co = qemu_coroutine_self(), |
400 | .serialising = false, | |
401 | .overlap_offset = offset, | |
402 | .overlap_bytes = bytes, | |
403 | }; | |
404 | ||
405 | qemu_co_queue_init(&req->wait_queue); | |
406 | ||
407 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); | |
408 | } | |
409 | ||
410 | static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align) | |
411 | { | |
412 | int64_t overlap_offset = req->offset & ~(align - 1); | |
413 | unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align) | |
414 | - overlap_offset; | |
415 | ||
416 | if (!req->serialising) { | |
417 | req->bs->serialising_in_flight++; | |
418 | req->serialising = true; | |
419 | } | |
420 | ||
421 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
422 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
423 | } | |
424 | ||
425 | /** | |
244483e6 | 426 | * Round a region to cluster boundaries (sector-based) |
61007b31 | 427 | */ |
244483e6 KW |
428 | void bdrv_round_sectors_to_clusters(BlockDriverState *bs, |
429 | int64_t sector_num, int nb_sectors, | |
430 | int64_t *cluster_sector_num, | |
431 | int *cluster_nb_sectors) | |
61007b31 SH |
432 | { |
433 | BlockDriverInfo bdi; | |
434 | ||
435 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
436 | *cluster_sector_num = sector_num; | |
437 | *cluster_nb_sectors = nb_sectors; | |
438 | } else { | |
439 | int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; | |
440 | *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); | |
441 | *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + | |
442 | nb_sectors, c); | |
443 | } | |
444 | } | |
445 | ||
244483e6 KW |
446 | /** |
447 | * Round a region to cluster boundaries | |
448 | */ | |
449 | void bdrv_round_to_clusters(BlockDriverState *bs, | |
450 | int64_t offset, unsigned int bytes, | |
451 | int64_t *cluster_offset, | |
452 | unsigned int *cluster_bytes) | |
453 | { | |
454 | BlockDriverInfo bdi; | |
455 | ||
456 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
457 | *cluster_offset = offset; | |
458 | *cluster_bytes = bytes; | |
459 | } else { | |
460 | int64_t c = bdi.cluster_size; | |
461 | *cluster_offset = QEMU_ALIGN_DOWN(offset, c); | |
462 | *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c); | |
463 | } | |
464 | } | |
465 | ||
61007b31 SH |
466 | static int bdrv_get_cluster_size(BlockDriverState *bs) |
467 | { | |
468 | BlockDriverInfo bdi; | |
469 | int ret; | |
470 | ||
471 | ret = bdrv_get_info(bs, &bdi); | |
472 | if (ret < 0 || bdi.cluster_size == 0) { | |
a5b8dd2c | 473 | return bs->bl.request_alignment; |
61007b31 SH |
474 | } else { |
475 | return bdi.cluster_size; | |
476 | } | |
477 | } | |
478 | ||
479 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
480 | int64_t offset, unsigned int bytes) | |
481 | { | |
482 | /* aaaa bbbb */ | |
483 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
484 | return false; | |
485 | } | |
486 | /* bbbb aaaa */ | |
487 | if (req->overlap_offset >= offset + bytes) { | |
488 | return false; | |
489 | } | |
490 | return true; | |
491 | } | |
492 | ||
99723548 PB |
493 | void bdrv_inc_in_flight(BlockDriverState *bs) |
494 | { | |
495 | atomic_inc(&bs->in_flight); | |
496 | } | |
497 | ||
c9d1a561 PB |
498 | static void dummy_bh_cb(void *opaque) |
499 | { | |
500 | } | |
501 | ||
502 | void bdrv_wakeup(BlockDriverState *bs) | |
503 | { | |
504 | if (bs->wakeup) { | |
505 | aio_bh_schedule_oneshot(qemu_get_aio_context(), dummy_bh_cb, NULL); | |
506 | } | |
507 | } | |
508 | ||
99723548 PB |
509 | void bdrv_dec_in_flight(BlockDriverState *bs) |
510 | { | |
511 | atomic_dec(&bs->in_flight); | |
c9d1a561 | 512 | bdrv_wakeup(bs); |
99723548 PB |
513 | } |
514 | ||
61007b31 SH |
515 | static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self) |
516 | { | |
517 | BlockDriverState *bs = self->bs; | |
518 | BdrvTrackedRequest *req; | |
519 | bool retry; | |
520 | bool waited = false; | |
521 | ||
522 | if (!bs->serialising_in_flight) { | |
523 | return false; | |
524 | } | |
525 | ||
526 | do { | |
527 | retry = false; | |
528 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
529 | if (req == self || (!req->serialising && !self->serialising)) { | |
530 | continue; | |
531 | } | |
532 | if (tracked_request_overlaps(req, self->overlap_offset, | |
533 | self->overlap_bytes)) | |
534 | { | |
535 | /* Hitting this means there was a reentrant request, for | |
536 | * example, a block driver issuing nested requests. This must | |
537 | * never happen since it means deadlock. | |
538 | */ | |
539 | assert(qemu_coroutine_self() != req->co); | |
540 | ||
541 | /* If the request is already (indirectly) waiting for us, or | |
542 | * will wait for us as soon as it wakes up, then just go on | |
543 | * (instead of producing a deadlock in the former case). */ | |
544 | if (!req->waiting_for) { | |
545 | self->waiting_for = req; | |
546 | qemu_co_queue_wait(&req->wait_queue); | |
547 | self->waiting_for = NULL; | |
548 | retry = true; | |
549 | waited = true; | |
550 | break; | |
551 | } | |
552 | } | |
553 | } | |
554 | } while (retry); | |
555 | ||
556 | return waited; | |
557 | } | |
558 | ||
559 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
560 | size_t size) | |
561 | { | |
562 | if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) { | |
563 | return -EIO; | |
564 | } | |
565 | ||
566 | if (!bdrv_is_inserted(bs)) { | |
567 | return -ENOMEDIUM; | |
568 | } | |
569 | ||
570 | if (offset < 0) { | |
571 | return -EIO; | |
572 | } | |
573 | ||
574 | return 0; | |
575 | } | |
576 | ||
61007b31 | 577 | typedef struct RwCo { |
e293b7a3 | 578 | BdrvChild *child; |
61007b31 SH |
579 | int64_t offset; |
580 | QEMUIOVector *qiov; | |
581 | bool is_write; | |
582 | int ret; | |
583 | BdrvRequestFlags flags; | |
584 | } RwCo; | |
585 | ||
586 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
587 | { | |
588 | RwCo *rwco = opaque; | |
589 | ||
590 | if (!rwco->is_write) { | |
a03ef88f | 591 | rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset, |
cab3a356 KW |
592 | rwco->qiov->size, rwco->qiov, |
593 | rwco->flags); | |
61007b31 | 594 | } else { |
a03ef88f | 595 | rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset, |
cab3a356 KW |
596 | rwco->qiov->size, rwco->qiov, |
597 | rwco->flags); | |
61007b31 SH |
598 | } |
599 | } | |
600 | ||
601 | /* | |
602 | * Process a vectored synchronous request using coroutines | |
603 | */ | |
e293b7a3 | 604 | static int bdrv_prwv_co(BdrvChild *child, int64_t offset, |
61007b31 SH |
605 | QEMUIOVector *qiov, bool is_write, |
606 | BdrvRequestFlags flags) | |
607 | { | |
608 | Coroutine *co; | |
609 | RwCo rwco = { | |
e293b7a3 | 610 | .child = child, |
61007b31 SH |
611 | .offset = offset, |
612 | .qiov = qiov, | |
613 | .is_write = is_write, | |
614 | .ret = NOT_DONE, | |
615 | .flags = flags, | |
616 | }; | |
617 | ||
61007b31 SH |
618 | if (qemu_in_coroutine()) { |
619 | /* Fast-path if already in coroutine context */ | |
620 | bdrv_rw_co_entry(&rwco); | |
621 | } else { | |
0b8b8753 PB |
622 | co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco); |
623 | qemu_coroutine_enter(co); | |
88b062c2 | 624 | BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE); |
61007b31 SH |
625 | } |
626 | return rwco.ret; | |
627 | } | |
628 | ||
629 | /* | |
630 | * Process a synchronous request using coroutines | |
631 | */ | |
e293b7a3 | 632 | static int bdrv_rw_co(BdrvChild *child, int64_t sector_num, uint8_t *buf, |
61007b31 SH |
633 | int nb_sectors, bool is_write, BdrvRequestFlags flags) |
634 | { | |
635 | QEMUIOVector qiov; | |
636 | struct iovec iov = { | |
637 | .iov_base = (void *)buf, | |
638 | .iov_len = nb_sectors * BDRV_SECTOR_SIZE, | |
639 | }; | |
640 | ||
641 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
642 | return -EINVAL; | |
643 | } | |
644 | ||
645 | qemu_iovec_init_external(&qiov, &iov, 1); | |
e293b7a3 | 646 | return bdrv_prwv_co(child, sector_num << BDRV_SECTOR_BITS, |
61007b31 SH |
647 | &qiov, is_write, flags); |
648 | } | |
649 | ||
650 | /* return < 0 if error. See bdrv_write() for the return codes */ | |
fbcbbf4e | 651 | int bdrv_read(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
652 | uint8_t *buf, int nb_sectors) |
653 | { | |
e293b7a3 | 654 | return bdrv_rw_co(child, sector_num, buf, nb_sectors, false, 0); |
61007b31 SH |
655 | } |
656 | ||
61007b31 SH |
657 | /* Return < 0 if error. Important errors are: |
658 | -EIO generic I/O error (may happen for all errors) | |
659 | -ENOMEDIUM No media inserted. | |
660 | -EINVAL Invalid sector number or nb_sectors | |
661 | -EACCES Trying to write a read-only device | |
662 | */ | |
18d51c4b | 663 | int bdrv_write(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
664 | const uint8_t *buf, int nb_sectors) |
665 | { | |
e293b7a3 | 666 | return bdrv_rw_co(child, sector_num, (uint8_t *)buf, nb_sectors, true, 0); |
61007b31 SH |
667 | } |
668 | ||
720ff280 | 669 | int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset, |
74021bc4 | 670 | int count, BdrvRequestFlags flags) |
61007b31 | 671 | { |
74021bc4 EB |
672 | QEMUIOVector qiov; |
673 | struct iovec iov = { | |
674 | .iov_base = NULL, | |
675 | .iov_len = count, | |
676 | }; | |
677 | ||
678 | qemu_iovec_init_external(&qiov, &iov, 1); | |
e293b7a3 | 679 | return bdrv_prwv_co(child, offset, &qiov, true, |
74021bc4 | 680 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
681 | } |
682 | ||
683 | /* | |
74021bc4 | 684 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
685 | * The operation is sped up by checking the block status and only writing |
686 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 687 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 688 | * BDRV_REQ_FUA). |
61007b31 SH |
689 | * |
690 | * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). | |
691 | */ | |
720ff280 | 692 | int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags) |
61007b31 SH |
693 | { |
694 | int64_t target_sectors, ret, nb_sectors, sector_num = 0; | |
720ff280 | 695 | BlockDriverState *bs = child->bs; |
67a0fd2a | 696 | BlockDriverState *file; |
61007b31 SH |
697 | int n; |
698 | ||
699 | target_sectors = bdrv_nb_sectors(bs); | |
700 | if (target_sectors < 0) { | |
701 | return target_sectors; | |
702 | } | |
703 | ||
704 | for (;;) { | |
705 | nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS); | |
706 | if (nb_sectors <= 0) { | |
707 | return 0; | |
708 | } | |
67a0fd2a | 709 | ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n, &file); |
61007b31 SH |
710 | if (ret < 0) { |
711 | error_report("error getting block status at sector %" PRId64 ": %s", | |
712 | sector_num, strerror(-ret)); | |
713 | return ret; | |
714 | } | |
715 | if (ret & BDRV_BLOCK_ZERO) { | |
716 | sector_num += n; | |
717 | continue; | |
718 | } | |
720ff280 | 719 | ret = bdrv_pwrite_zeroes(child, sector_num << BDRV_SECTOR_BITS, |
74021bc4 | 720 | n << BDRV_SECTOR_BITS, flags); |
61007b31 SH |
721 | if (ret < 0) { |
722 | error_report("error writing zeroes at sector %" PRId64 ": %s", | |
723 | sector_num, strerror(-ret)); | |
724 | return ret; | |
725 | } | |
726 | sector_num += n; | |
727 | } | |
728 | } | |
729 | ||
cf2ab8fc | 730 | int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
f1e84741 KW |
731 | { |
732 | int ret; | |
733 | ||
e293b7a3 | 734 | ret = bdrv_prwv_co(child, offset, qiov, false, 0); |
f1e84741 KW |
735 | if (ret < 0) { |
736 | return ret; | |
737 | } | |
738 | ||
739 | return qiov->size; | |
740 | } | |
741 | ||
cf2ab8fc | 742 | int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes) |
61007b31 SH |
743 | { |
744 | QEMUIOVector qiov; | |
745 | struct iovec iov = { | |
746 | .iov_base = (void *)buf, | |
747 | .iov_len = bytes, | |
748 | }; | |
61007b31 SH |
749 | |
750 | if (bytes < 0) { | |
751 | return -EINVAL; | |
752 | } | |
753 | ||
754 | qemu_iovec_init_external(&qiov, &iov, 1); | |
cf2ab8fc | 755 | return bdrv_preadv(child, offset, &qiov); |
61007b31 SH |
756 | } |
757 | ||
d9ca2ea2 | 758 | int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
61007b31 SH |
759 | { |
760 | int ret; | |
761 | ||
e293b7a3 | 762 | ret = bdrv_prwv_co(child, offset, qiov, true, 0); |
61007b31 SH |
763 | if (ret < 0) { |
764 | return ret; | |
765 | } | |
766 | ||
767 | return qiov->size; | |
768 | } | |
769 | ||
d9ca2ea2 | 770 | int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes) |
61007b31 SH |
771 | { |
772 | QEMUIOVector qiov; | |
773 | struct iovec iov = { | |
774 | .iov_base = (void *) buf, | |
775 | .iov_len = bytes, | |
776 | }; | |
777 | ||
778 | if (bytes < 0) { | |
779 | return -EINVAL; | |
780 | } | |
781 | ||
782 | qemu_iovec_init_external(&qiov, &iov, 1); | |
d9ca2ea2 | 783 | return bdrv_pwritev(child, offset, &qiov); |
61007b31 SH |
784 | } |
785 | ||
786 | /* | |
787 | * Writes to the file and ensures that no writes are reordered across this | |
788 | * request (acts as a barrier) | |
789 | * | |
790 | * Returns 0 on success, -errno in error cases. | |
791 | */ | |
d9ca2ea2 KW |
792 | int bdrv_pwrite_sync(BdrvChild *child, int64_t offset, |
793 | const void *buf, int count) | |
61007b31 SH |
794 | { |
795 | int ret; | |
796 | ||
d9ca2ea2 | 797 | ret = bdrv_pwrite(child, offset, buf, count); |
61007b31 SH |
798 | if (ret < 0) { |
799 | return ret; | |
800 | } | |
801 | ||
d9ca2ea2 | 802 | ret = bdrv_flush(child->bs); |
855a6a93 KW |
803 | if (ret < 0) { |
804 | return ret; | |
61007b31 SH |
805 | } |
806 | ||
807 | return 0; | |
808 | } | |
809 | ||
08844473 KW |
810 | typedef struct CoroutineIOCompletion { |
811 | Coroutine *coroutine; | |
812 | int ret; | |
813 | } CoroutineIOCompletion; | |
814 | ||
815 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
816 | { | |
817 | CoroutineIOCompletion *co = opaque; | |
818 | ||
819 | co->ret = ret; | |
0b8b8753 | 820 | qemu_coroutine_enter(co->coroutine); |
08844473 KW |
821 | } |
822 | ||
166fe960 KW |
823 | static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs, |
824 | uint64_t offset, uint64_t bytes, | |
825 | QEMUIOVector *qiov, int flags) | |
826 | { | |
827 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
828 | int64_t sector_num; |
829 | unsigned int nb_sectors; | |
830 | ||
fa166538 EB |
831 | assert(!(flags & ~BDRV_REQ_MASK)); |
832 | ||
3fb06697 KW |
833 | if (drv->bdrv_co_preadv) { |
834 | return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); | |
835 | } | |
836 | ||
837 | sector_num = offset >> BDRV_SECTOR_BITS; | |
838 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
166fe960 KW |
839 | |
840 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
841 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
842 | assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS); | |
843 | ||
08844473 KW |
844 | if (drv->bdrv_co_readv) { |
845 | return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
846 | } else { | |
847 | BlockAIOCB *acb; | |
848 | CoroutineIOCompletion co = { | |
849 | .coroutine = qemu_coroutine_self(), | |
850 | }; | |
851 | ||
852 | acb = bs->drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors, | |
853 | bdrv_co_io_em_complete, &co); | |
854 | if (acb == NULL) { | |
855 | return -EIO; | |
856 | } else { | |
857 | qemu_coroutine_yield(); | |
858 | return co.ret; | |
859 | } | |
860 | } | |
166fe960 KW |
861 | } |
862 | ||
78a07294 KW |
863 | static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs, |
864 | uint64_t offset, uint64_t bytes, | |
865 | QEMUIOVector *qiov, int flags) | |
866 | { | |
867 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
868 | int64_t sector_num; |
869 | unsigned int nb_sectors; | |
78a07294 KW |
870 | int ret; |
871 | ||
fa166538 EB |
872 | assert(!(flags & ~BDRV_REQ_MASK)); |
873 | ||
3fb06697 | 874 | if (drv->bdrv_co_pwritev) { |
515c2f43 KW |
875 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, |
876 | flags & bs->supported_write_flags); | |
877 | flags &= ~bs->supported_write_flags; | |
3fb06697 KW |
878 | goto emulate_flags; |
879 | } | |
880 | ||
881 | sector_num = offset >> BDRV_SECTOR_BITS; | |
882 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
883 | ||
78a07294 KW |
884 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); |
885 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
886 | assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS); | |
887 | ||
888 | if (drv->bdrv_co_writev_flags) { | |
889 | ret = drv->bdrv_co_writev_flags(bs, sector_num, nb_sectors, qiov, | |
4df863f3 EB |
890 | flags & bs->supported_write_flags); |
891 | flags &= ~bs->supported_write_flags; | |
08844473 | 892 | } else if (drv->bdrv_co_writev) { |
4df863f3 | 893 | assert(!bs->supported_write_flags); |
78a07294 | 894 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); |
08844473 KW |
895 | } else { |
896 | BlockAIOCB *acb; | |
897 | CoroutineIOCompletion co = { | |
898 | .coroutine = qemu_coroutine_self(), | |
899 | }; | |
900 | ||
901 | acb = bs->drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors, | |
902 | bdrv_co_io_em_complete, &co); | |
903 | if (acb == NULL) { | |
3fb06697 | 904 | ret = -EIO; |
08844473 KW |
905 | } else { |
906 | qemu_coroutine_yield(); | |
3fb06697 | 907 | ret = co.ret; |
08844473 | 908 | } |
78a07294 KW |
909 | } |
910 | ||
3fb06697 | 911 | emulate_flags: |
4df863f3 | 912 | if (ret == 0 && (flags & BDRV_REQ_FUA)) { |
78a07294 KW |
913 | ret = bdrv_co_flush(bs); |
914 | } | |
915 | ||
916 | return ret; | |
917 | } | |
918 | ||
29a298af PB |
919 | static int coroutine_fn |
920 | bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset, | |
921 | uint64_t bytes, QEMUIOVector *qiov) | |
922 | { | |
923 | BlockDriver *drv = bs->drv; | |
924 | ||
925 | if (!drv->bdrv_co_pwritev_compressed) { | |
926 | return -ENOTSUP; | |
927 | } | |
928 | ||
29a298af PB |
929 | return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov); |
930 | } | |
931 | ||
61007b31 | 932 | static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, |
244483e6 | 933 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov) |
61007b31 SH |
934 | { |
935 | /* Perform I/O through a temporary buffer so that users who scribble over | |
936 | * their read buffer while the operation is in progress do not end up | |
937 | * modifying the image file. This is critical for zero-copy guest I/O | |
938 | * where anything might happen inside guest memory. | |
939 | */ | |
940 | void *bounce_buffer; | |
941 | ||
942 | BlockDriver *drv = bs->drv; | |
943 | struct iovec iov; | |
944 | QEMUIOVector bounce_qiov; | |
244483e6 KW |
945 | int64_t cluster_offset; |
946 | unsigned int cluster_bytes; | |
61007b31 SH |
947 | size_t skip_bytes; |
948 | int ret; | |
949 | ||
950 | /* Cover entire cluster so no additional backing file I/O is required when | |
951 | * allocating cluster in the image file. | |
952 | */ | |
244483e6 | 953 | bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes); |
61007b31 | 954 | |
244483e6 KW |
955 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
956 | cluster_offset, cluster_bytes); | |
61007b31 | 957 | |
244483e6 | 958 | iov.iov_len = cluster_bytes; |
61007b31 SH |
959 | iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len); |
960 | if (bounce_buffer == NULL) { | |
961 | ret = -ENOMEM; | |
962 | goto err; | |
963 | } | |
964 | ||
965 | qemu_iovec_init_external(&bounce_qiov, &iov, 1); | |
966 | ||
244483e6 | 967 | ret = bdrv_driver_preadv(bs, cluster_offset, cluster_bytes, |
166fe960 | 968 | &bounce_qiov, 0); |
61007b31 SH |
969 | if (ret < 0) { |
970 | goto err; | |
971 | } | |
972 | ||
c1499a5e | 973 | if (drv->bdrv_co_pwrite_zeroes && |
61007b31 | 974 | buffer_is_zero(bounce_buffer, iov.iov_len)) { |
a604fa2b EB |
975 | /* FIXME: Should we (perhaps conditionally) be setting |
976 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
977 | * that still correctly reads as zero? */ | |
244483e6 | 978 | ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, cluster_bytes, 0); |
61007b31 SH |
979 | } else { |
980 | /* This does not change the data on the disk, it is not necessary | |
981 | * to flush even in cache=writethrough mode. | |
982 | */ | |
244483e6 | 983 | ret = bdrv_driver_pwritev(bs, cluster_offset, cluster_bytes, |
78a07294 | 984 | &bounce_qiov, 0); |
61007b31 SH |
985 | } |
986 | ||
987 | if (ret < 0) { | |
988 | /* It might be okay to ignore write errors for guest requests. If this | |
989 | * is a deliberate copy-on-read then we don't want to ignore the error. | |
990 | * Simply report it in all cases. | |
991 | */ | |
992 | goto err; | |
993 | } | |
994 | ||
244483e6 KW |
995 | skip_bytes = offset - cluster_offset; |
996 | qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, bytes); | |
61007b31 SH |
997 | |
998 | err: | |
999 | qemu_vfree(bounce_buffer); | |
1000 | return ret; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Forwards an already correctly aligned request to the BlockDriver. This | |
1a62d0ac EB |
1005 | * handles copy on read, zeroing after EOF, and fragmentation of large |
1006 | * reads; any other features must be implemented by the caller. | |
61007b31 SH |
1007 | */ |
1008 | static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs, | |
1009 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, | |
1010 | int64_t align, QEMUIOVector *qiov, int flags) | |
1011 | { | |
c9d20029 | 1012 | int64_t total_bytes, max_bytes; |
1a62d0ac EB |
1013 | int ret = 0; |
1014 | uint64_t bytes_remaining = bytes; | |
1015 | int max_transfer; | |
61007b31 | 1016 | |
49c07526 KW |
1017 | assert(is_power_of_2(align)); |
1018 | assert((offset & (align - 1)) == 0); | |
1019 | assert((bytes & (align - 1)) == 0); | |
61007b31 | 1020 | assert(!qiov || bytes == qiov->size); |
abb06c5a | 1021 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
1a62d0ac EB |
1022 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1023 | align); | |
a604fa2b EB |
1024 | |
1025 | /* TODO: We would need a per-BDS .supported_read_flags and | |
1026 | * potential fallback support, if we ever implement any read flags | |
1027 | * to pass through to drivers. For now, there aren't any | |
1028 | * passthrough flags. */ | |
1029 | assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ))); | |
61007b31 SH |
1030 | |
1031 | /* Handle Copy on Read and associated serialisation */ | |
1032 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
1033 | /* If we touch the same cluster it counts as an overlap. This | |
1034 | * guarantees that allocating writes will be serialized and not race | |
1035 | * with each other for the same cluster. For example, in copy-on-read | |
1036 | * it ensures that the CoR read and write operations are atomic and | |
1037 | * guest writes cannot interleave between them. */ | |
1038 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
1039 | } | |
1040 | ||
61408b25 FZ |
1041 | if (!(flags & BDRV_REQ_NO_SERIALISING)) { |
1042 | wait_serialising_requests(req); | |
1043 | } | |
61007b31 SH |
1044 | |
1045 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
49c07526 KW |
1046 | int64_t start_sector = offset >> BDRV_SECTOR_BITS; |
1047 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1048 | unsigned int nb_sectors = end_sector - start_sector; | |
61007b31 SH |
1049 | int pnum; |
1050 | ||
49c07526 | 1051 | ret = bdrv_is_allocated(bs, start_sector, nb_sectors, &pnum); |
61007b31 SH |
1052 | if (ret < 0) { |
1053 | goto out; | |
1054 | } | |
1055 | ||
1056 | if (!ret || pnum != nb_sectors) { | |
244483e6 | 1057 | ret = bdrv_co_do_copy_on_readv(bs, offset, bytes, qiov); |
61007b31 SH |
1058 | goto out; |
1059 | } | |
1060 | } | |
1061 | ||
1a62d0ac | 1062 | /* Forward the request to the BlockDriver, possibly fragmenting it */ |
c9d20029 KW |
1063 | total_bytes = bdrv_getlength(bs); |
1064 | if (total_bytes < 0) { | |
1065 | ret = total_bytes; | |
1066 | goto out; | |
1067 | } | |
61007b31 | 1068 | |
c9d20029 | 1069 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
1a62d0ac | 1070 | if (bytes <= max_bytes && bytes <= max_transfer) { |
c9d20029 | 1071 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0); |
1a62d0ac EB |
1072 | goto out; |
1073 | } | |
61007b31 | 1074 | |
1a62d0ac EB |
1075 | while (bytes_remaining) { |
1076 | int num; | |
61007b31 | 1077 | |
1a62d0ac EB |
1078 | if (max_bytes) { |
1079 | QEMUIOVector local_qiov; | |
61007b31 | 1080 | |
1a62d0ac EB |
1081 | num = MIN(bytes_remaining, MIN(max_bytes, max_transfer)); |
1082 | assert(num); | |
1083 | qemu_iovec_init(&local_qiov, qiov->niov); | |
1084 | qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num); | |
61007b31 | 1085 | |
1a62d0ac EB |
1086 | ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining, |
1087 | num, &local_qiov, 0); | |
1088 | max_bytes -= num; | |
1089 | qemu_iovec_destroy(&local_qiov); | |
1090 | } else { | |
1091 | num = bytes_remaining; | |
1092 | ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0, | |
1093 | bytes_remaining); | |
1094 | } | |
1095 | if (ret < 0) { | |
1096 | goto out; | |
1097 | } | |
1098 | bytes_remaining -= num; | |
61007b31 SH |
1099 | } |
1100 | ||
1101 | out: | |
1a62d0ac | 1102 | return ret < 0 ? ret : 0; |
61007b31 SH |
1103 | } |
1104 | ||
61007b31 SH |
1105 | /* |
1106 | * Handle a read request in coroutine context | |
1107 | */ | |
a03ef88f | 1108 | int coroutine_fn bdrv_co_preadv(BdrvChild *child, |
61007b31 SH |
1109 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1110 | BdrvRequestFlags flags) | |
1111 | { | |
a03ef88f | 1112 | BlockDriverState *bs = child->bs; |
61007b31 SH |
1113 | BlockDriver *drv = bs->drv; |
1114 | BdrvTrackedRequest req; | |
1115 | ||
a5b8dd2c | 1116 | uint64_t align = bs->bl.request_alignment; |
61007b31 SH |
1117 | uint8_t *head_buf = NULL; |
1118 | uint8_t *tail_buf = NULL; | |
1119 | QEMUIOVector local_qiov; | |
1120 | bool use_local_qiov = false; | |
1121 | int ret; | |
1122 | ||
1123 | if (!drv) { | |
1124 | return -ENOMEDIUM; | |
1125 | } | |
1126 | ||
1127 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1128 | if (ret < 0) { | |
1129 | return ret; | |
1130 | } | |
1131 | ||
99723548 PB |
1132 | bdrv_inc_in_flight(bs); |
1133 | ||
9568b511 | 1134 | /* Don't do copy-on-read if we read data before write operation */ |
61408b25 | 1135 | if (bs->copy_on_read && !(flags & BDRV_REQ_NO_SERIALISING)) { |
61007b31 SH |
1136 | flags |= BDRV_REQ_COPY_ON_READ; |
1137 | } | |
1138 | ||
61007b31 SH |
1139 | /* Align read if necessary by padding qiov */ |
1140 | if (offset & (align - 1)) { | |
1141 | head_buf = qemu_blockalign(bs, align); | |
1142 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1143 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1144 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1145 | use_local_qiov = true; | |
1146 | ||
1147 | bytes += offset & (align - 1); | |
1148 | offset = offset & ~(align - 1); | |
1149 | } | |
1150 | ||
1151 | if ((offset + bytes) & (align - 1)) { | |
1152 | if (!use_local_qiov) { | |
1153 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1154 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1155 | use_local_qiov = true; | |
1156 | } | |
1157 | tail_buf = qemu_blockalign(bs, align); | |
1158 | qemu_iovec_add(&local_qiov, tail_buf, | |
1159 | align - ((offset + bytes) & (align - 1))); | |
1160 | ||
1161 | bytes = ROUND_UP(bytes, align); | |
1162 | } | |
1163 | ||
ebde595c | 1164 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
61007b31 SH |
1165 | ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align, |
1166 | use_local_qiov ? &local_qiov : qiov, | |
1167 | flags); | |
1168 | tracked_request_end(&req); | |
99723548 | 1169 | bdrv_dec_in_flight(bs); |
61007b31 SH |
1170 | |
1171 | if (use_local_qiov) { | |
1172 | qemu_iovec_destroy(&local_qiov); | |
1173 | qemu_vfree(head_buf); | |
1174 | qemu_vfree(tail_buf); | |
1175 | } | |
1176 | ||
1177 | return ret; | |
1178 | } | |
1179 | ||
adad6496 | 1180 | static int coroutine_fn bdrv_co_do_readv(BdrvChild *child, |
61007b31 SH |
1181 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, |
1182 | BdrvRequestFlags flags) | |
1183 | { | |
1184 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
1185 | return -EINVAL; | |
1186 | } | |
1187 | ||
a03ef88f | 1188 | return bdrv_co_preadv(child, sector_num << BDRV_SECTOR_BITS, |
cab3a356 | 1189 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); |
61007b31 SH |
1190 | } |
1191 | ||
28b04a8f KW |
1192 | int coroutine_fn bdrv_co_readv(BdrvChild *child, int64_t sector_num, |
1193 | int nb_sectors, QEMUIOVector *qiov) | |
61007b31 | 1194 | { |
28b04a8f | 1195 | trace_bdrv_co_readv(child->bs, sector_num, nb_sectors); |
61007b31 | 1196 | |
adad6496 | 1197 | return bdrv_co_do_readv(child, sector_num, nb_sectors, qiov, 0); |
61007b31 SH |
1198 | } |
1199 | ||
5def6b80 EB |
1200 | /* Maximum buffer for write zeroes fallback, in bytes */ |
1201 | #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS) | |
61007b31 | 1202 | |
d05aa8bb EB |
1203 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
1204 | int64_t offset, int count, BdrvRequestFlags flags) | |
61007b31 SH |
1205 | { |
1206 | BlockDriver *drv = bs->drv; | |
1207 | QEMUIOVector qiov; | |
1208 | struct iovec iov = {0}; | |
1209 | int ret = 0; | |
465fe887 | 1210 | bool need_flush = false; |
443668ca DL |
1211 | int head = 0; |
1212 | int tail = 0; | |
61007b31 | 1213 | |
cf081fca | 1214 | int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX); |
a5b8dd2c EB |
1215 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1216 | bs->bl.request_alignment); | |
b2f95fee EB |
1217 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
1218 | MAX_WRITE_ZEROES_BOUNCE_BUFFER); | |
d05aa8bb | 1219 | |
b8d0a980 EB |
1220 | assert(alignment % bs->bl.request_alignment == 0); |
1221 | head = offset % alignment; | |
1222 | tail = (offset + count) % alignment; | |
1223 | max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment); | |
1224 | assert(max_write_zeroes >= bs->bl.request_alignment); | |
61007b31 | 1225 | |
d05aa8bb EB |
1226 | while (count > 0 && !ret) { |
1227 | int num = count; | |
61007b31 SH |
1228 | |
1229 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1230 | * to be aligned, and that unaligned requests do not cross cluster |
1231 | * boundaries. | |
61007b31 | 1232 | */ |
443668ca | 1233 | if (head) { |
b2f95fee EB |
1234 | /* Make a small request up to the first aligned sector. For |
1235 | * convenience, limit this request to max_transfer even if | |
1236 | * we don't need to fall back to writes. */ | |
1237 | num = MIN(MIN(count, max_transfer), alignment - head); | |
1238 | head = (head + num) % alignment; | |
1239 | assert(num < max_write_zeroes); | |
d05aa8bb | 1240 | } else if (tail && num > alignment) { |
443668ca DL |
1241 | /* Shorten the request to the last aligned sector. */ |
1242 | num -= tail; | |
61007b31 SH |
1243 | } |
1244 | ||
1245 | /* limit request size */ | |
1246 | if (num > max_write_zeroes) { | |
1247 | num = max_write_zeroes; | |
1248 | } | |
1249 | ||
1250 | ret = -ENOTSUP; | |
1251 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1252 | if (drv->bdrv_co_pwrite_zeroes) { |
1253 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1254 | flags & bs->supported_zero_flags); | |
1255 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1256 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1257 | need_flush = true; | |
1258 | } | |
465fe887 EB |
1259 | } else { |
1260 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1261 | } |
1262 | ||
1263 | if (ret == -ENOTSUP) { | |
1264 | /* Fall back to bounce buffer if write zeroes is unsupported */ | |
465fe887 EB |
1265 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1266 | ||
1267 | if ((flags & BDRV_REQ_FUA) && | |
1268 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1269 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1270 | * flush on each chunk; use just one at the end */ | |
1271 | write_flags &= ~BDRV_REQ_FUA; | |
1272 | need_flush = true; | |
1273 | } | |
5def6b80 | 1274 | num = MIN(num, max_transfer); |
d05aa8bb | 1275 | iov.iov_len = num; |
61007b31 | 1276 | if (iov.iov_base == NULL) { |
d05aa8bb | 1277 | iov.iov_base = qemu_try_blockalign(bs, num); |
61007b31 SH |
1278 | if (iov.iov_base == NULL) { |
1279 | ret = -ENOMEM; | |
1280 | goto fail; | |
1281 | } | |
d05aa8bb | 1282 | memset(iov.iov_base, 0, num); |
61007b31 SH |
1283 | } |
1284 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1285 | ||
d05aa8bb | 1286 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags); |
61007b31 SH |
1287 | |
1288 | /* Keep bounce buffer around if it is big enough for all | |
1289 | * all future requests. | |
1290 | */ | |
5def6b80 | 1291 | if (num < max_transfer) { |
61007b31 SH |
1292 | qemu_vfree(iov.iov_base); |
1293 | iov.iov_base = NULL; | |
1294 | } | |
1295 | } | |
1296 | ||
d05aa8bb EB |
1297 | offset += num; |
1298 | count -= num; | |
61007b31 SH |
1299 | } |
1300 | ||
1301 | fail: | |
465fe887 EB |
1302 | if (ret == 0 && need_flush) { |
1303 | ret = bdrv_co_flush(bs); | |
1304 | } | |
61007b31 SH |
1305 | qemu_vfree(iov.iov_base); |
1306 | return ret; | |
1307 | } | |
1308 | ||
1309 | /* | |
04ed95f4 EB |
1310 | * Forwards an already correctly aligned write request to the BlockDriver, |
1311 | * after possibly fragmenting it. | |
61007b31 SH |
1312 | */ |
1313 | static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs, | |
1314 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, | |
cff86b38 | 1315 | int64_t align, QEMUIOVector *qiov, int flags) |
61007b31 SH |
1316 | { |
1317 | BlockDriver *drv = bs->drv; | |
1318 | bool waited; | |
1319 | int ret; | |
1320 | ||
9896c876 KW |
1321 | int64_t start_sector = offset >> BDRV_SECTOR_BITS; |
1322 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
04ed95f4 EB |
1323 | uint64_t bytes_remaining = bytes; |
1324 | int max_transfer; | |
61007b31 | 1325 | |
cff86b38 EB |
1326 | assert(is_power_of_2(align)); |
1327 | assert((offset & (align - 1)) == 0); | |
1328 | assert((bytes & (align - 1)) == 0); | |
61007b31 | 1329 | assert(!qiov || bytes == qiov->size); |
abb06c5a | 1330 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
fa166538 | 1331 | assert(!(flags & ~BDRV_REQ_MASK)); |
04ed95f4 EB |
1332 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1333 | align); | |
61007b31 SH |
1334 | |
1335 | waited = wait_serialising_requests(req); | |
1336 | assert(!waited || !req->serialising); | |
1337 | assert(req->overlap_offset <= offset); | |
1338 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
1339 | ||
1340 | ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req); | |
1341 | ||
1342 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 1343 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
61007b31 SH |
1344 | qemu_iovec_is_zero(qiov)) { |
1345 | flags |= BDRV_REQ_ZERO_WRITE; | |
1346 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
1347 | flags |= BDRV_REQ_MAY_UNMAP; | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | if (ret < 0) { | |
1352 | /* Do nothing, write notifier decided to fail this request */ | |
1353 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
9a4f4c31 | 1354 | bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 1355 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
3ea1a091 PB |
1356 | } else if (flags & BDRV_REQ_WRITE_COMPRESSED) { |
1357 | ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, qiov); | |
04ed95f4 | 1358 | } else if (bytes <= max_transfer) { |
9a4f4c31 | 1359 | bdrv_debug_event(bs, BLKDBG_PWRITEV); |
78a07294 | 1360 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags); |
04ed95f4 EB |
1361 | } else { |
1362 | bdrv_debug_event(bs, BLKDBG_PWRITEV); | |
1363 | while (bytes_remaining) { | |
1364 | int num = MIN(bytes_remaining, max_transfer); | |
1365 | QEMUIOVector local_qiov; | |
1366 | int local_flags = flags; | |
1367 | ||
1368 | assert(num); | |
1369 | if (num < bytes_remaining && (flags & BDRV_REQ_FUA) && | |
1370 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1371 | /* If FUA is going to be emulated by flush, we only | |
1372 | * need to flush on the last iteration */ | |
1373 | local_flags &= ~BDRV_REQ_FUA; | |
1374 | } | |
1375 | qemu_iovec_init(&local_qiov, qiov->niov); | |
1376 | qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num); | |
1377 | ||
1378 | ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining, | |
1379 | num, &local_qiov, local_flags); | |
1380 | qemu_iovec_destroy(&local_qiov); | |
1381 | if (ret < 0) { | |
1382 | break; | |
1383 | } | |
1384 | bytes_remaining -= num; | |
1385 | } | |
61007b31 | 1386 | } |
9a4f4c31 | 1387 | bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 1388 | |
3ff2f67a | 1389 | ++bs->write_gen; |
9896c876 | 1390 | bdrv_set_dirty(bs, start_sector, end_sector - start_sector); |
61007b31 | 1391 | |
53d8f9d8 HR |
1392 | if (bs->wr_highest_offset < offset + bytes) { |
1393 | bs->wr_highest_offset = offset + bytes; | |
1394 | } | |
61007b31 SH |
1395 | |
1396 | if (ret >= 0) { | |
9896c876 | 1397 | bs->total_sectors = MAX(bs->total_sectors, end_sector); |
04ed95f4 | 1398 | ret = 0; |
61007b31 SH |
1399 | } |
1400 | ||
1401 | return ret; | |
1402 | } | |
1403 | ||
9eeb6dd1 FZ |
1404 | static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs, |
1405 | int64_t offset, | |
1406 | unsigned int bytes, | |
1407 | BdrvRequestFlags flags, | |
1408 | BdrvTrackedRequest *req) | |
1409 | { | |
1410 | uint8_t *buf = NULL; | |
1411 | QEMUIOVector local_qiov; | |
1412 | struct iovec iov; | |
a5b8dd2c | 1413 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 FZ |
1414 | unsigned int head_padding_bytes, tail_padding_bytes; |
1415 | int ret = 0; | |
1416 | ||
1417 | head_padding_bytes = offset & (align - 1); | |
1418 | tail_padding_bytes = align - ((offset + bytes) & (align - 1)); | |
1419 | ||
1420 | ||
1421 | assert(flags & BDRV_REQ_ZERO_WRITE); | |
1422 | if (head_padding_bytes || tail_padding_bytes) { | |
1423 | buf = qemu_blockalign(bs, align); | |
1424 | iov = (struct iovec) { | |
1425 | .iov_base = buf, | |
1426 | .iov_len = align, | |
1427 | }; | |
1428 | qemu_iovec_init_external(&local_qiov, &iov, 1); | |
1429 | } | |
1430 | if (head_padding_bytes) { | |
1431 | uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes); | |
1432 | ||
1433 | /* RMW the unaligned part before head. */ | |
1434 | mark_request_serialising(req, align); | |
1435 | wait_serialising_requests(req); | |
9a4f4c31 | 1436 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
9eeb6dd1 FZ |
1437 | ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align, |
1438 | align, &local_qiov, 0); | |
1439 | if (ret < 0) { | |
1440 | goto fail; | |
1441 | } | |
9a4f4c31 | 1442 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
9eeb6dd1 FZ |
1443 | |
1444 | memset(buf + head_padding_bytes, 0, zero_bytes); | |
1445 | ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align, | |
cff86b38 | 1446 | align, &local_qiov, |
9eeb6dd1 FZ |
1447 | flags & ~BDRV_REQ_ZERO_WRITE); |
1448 | if (ret < 0) { | |
1449 | goto fail; | |
1450 | } | |
1451 | offset += zero_bytes; | |
1452 | bytes -= zero_bytes; | |
1453 | } | |
1454 | ||
1455 | assert(!bytes || (offset & (align - 1)) == 0); | |
1456 | if (bytes >= align) { | |
1457 | /* Write the aligned part in the middle. */ | |
1458 | uint64_t aligned_bytes = bytes & ~(align - 1); | |
cff86b38 | 1459 | ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes, align, |
9eeb6dd1 FZ |
1460 | NULL, flags); |
1461 | if (ret < 0) { | |
1462 | goto fail; | |
1463 | } | |
1464 | bytes -= aligned_bytes; | |
1465 | offset += aligned_bytes; | |
1466 | } | |
1467 | ||
1468 | assert(!bytes || (offset & (align - 1)) == 0); | |
1469 | if (bytes) { | |
1470 | assert(align == tail_padding_bytes + bytes); | |
1471 | /* RMW the unaligned part after tail. */ | |
1472 | mark_request_serialising(req, align); | |
1473 | wait_serialising_requests(req); | |
9a4f4c31 | 1474 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
9eeb6dd1 FZ |
1475 | ret = bdrv_aligned_preadv(bs, req, offset, align, |
1476 | align, &local_qiov, 0); | |
1477 | if (ret < 0) { | |
1478 | goto fail; | |
1479 | } | |
9a4f4c31 | 1480 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
9eeb6dd1 FZ |
1481 | |
1482 | memset(buf, 0, bytes); | |
cff86b38 | 1483 | ret = bdrv_aligned_pwritev(bs, req, offset, align, align, |
9eeb6dd1 FZ |
1484 | &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE); |
1485 | } | |
1486 | fail: | |
1487 | qemu_vfree(buf); | |
1488 | return ret; | |
1489 | ||
1490 | } | |
1491 | ||
61007b31 SH |
1492 | /* |
1493 | * Handle a write request in coroutine context | |
1494 | */ | |
a03ef88f | 1495 | int coroutine_fn bdrv_co_pwritev(BdrvChild *child, |
61007b31 SH |
1496 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1497 | BdrvRequestFlags flags) | |
1498 | { | |
a03ef88f | 1499 | BlockDriverState *bs = child->bs; |
61007b31 | 1500 | BdrvTrackedRequest req; |
a5b8dd2c | 1501 | uint64_t align = bs->bl.request_alignment; |
61007b31 SH |
1502 | uint8_t *head_buf = NULL; |
1503 | uint8_t *tail_buf = NULL; | |
1504 | QEMUIOVector local_qiov; | |
1505 | bool use_local_qiov = false; | |
1506 | int ret; | |
1507 | ||
1508 | if (!bs->drv) { | |
1509 | return -ENOMEDIUM; | |
1510 | } | |
1511 | if (bs->read_only) { | |
eaf5fe2d | 1512 | return -EPERM; |
61007b31 | 1513 | } |
04c01a5c | 1514 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
61007b31 SH |
1515 | |
1516 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1517 | if (ret < 0) { | |
1518 | return ret; | |
1519 | } | |
1520 | ||
99723548 | 1521 | bdrv_inc_in_flight(bs); |
61007b31 SH |
1522 | /* |
1523 | * Align write if necessary by performing a read-modify-write cycle. | |
1524 | * Pad qiov with the read parts and be sure to have a tracked request not | |
1525 | * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle. | |
1526 | */ | |
ebde595c | 1527 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 1528 | |
9eeb6dd1 FZ |
1529 | if (!qiov) { |
1530 | ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req); | |
1531 | goto out; | |
1532 | } | |
1533 | ||
61007b31 SH |
1534 | if (offset & (align - 1)) { |
1535 | QEMUIOVector head_qiov; | |
1536 | struct iovec head_iov; | |
1537 | ||
1538 | mark_request_serialising(&req, align); | |
1539 | wait_serialising_requests(&req); | |
1540 | ||
1541 | head_buf = qemu_blockalign(bs, align); | |
1542 | head_iov = (struct iovec) { | |
1543 | .iov_base = head_buf, | |
1544 | .iov_len = align, | |
1545 | }; | |
1546 | qemu_iovec_init_external(&head_qiov, &head_iov, 1); | |
1547 | ||
9a4f4c31 | 1548 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
61007b31 SH |
1549 | ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align, |
1550 | align, &head_qiov, 0); | |
1551 | if (ret < 0) { | |
1552 | goto fail; | |
1553 | } | |
9a4f4c31 | 1554 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
61007b31 SH |
1555 | |
1556 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1557 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1558 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1559 | use_local_qiov = true; | |
1560 | ||
1561 | bytes += offset & (align - 1); | |
1562 | offset = offset & ~(align - 1); | |
117bc3fa PL |
1563 | |
1564 | /* We have read the tail already if the request is smaller | |
1565 | * than one aligned block. | |
1566 | */ | |
1567 | if (bytes < align) { | |
1568 | qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes); | |
1569 | bytes = align; | |
1570 | } | |
61007b31 SH |
1571 | } |
1572 | ||
1573 | if ((offset + bytes) & (align - 1)) { | |
1574 | QEMUIOVector tail_qiov; | |
1575 | struct iovec tail_iov; | |
1576 | size_t tail_bytes; | |
1577 | bool waited; | |
1578 | ||
1579 | mark_request_serialising(&req, align); | |
1580 | waited = wait_serialising_requests(&req); | |
1581 | assert(!waited || !use_local_qiov); | |
1582 | ||
1583 | tail_buf = qemu_blockalign(bs, align); | |
1584 | tail_iov = (struct iovec) { | |
1585 | .iov_base = tail_buf, | |
1586 | .iov_len = align, | |
1587 | }; | |
1588 | qemu_iovec_init_external(&tail_qiov, &tail_iov, 1); | |
1589 | ||
9a4f4c31 | 1590 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
61007b31 SH |
1591 | ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align, |
1592 | align, &tail_qiov, 0); | |
1593 | if (ret < 0) { | |
1594 | goto fail; | |
1595 | } | |
9a4f4c31 | 1596 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
61007b31 SH |
1597 | |
1598 | if (!use_local_qiov) { | |
1599 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1600 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1601 | use_local_qiov = true; | |
1602 | } | |
1603 | ||
1604 | tail_bytes = (offset + bytes) & (align - 1); | |
1605 | qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes); | |
1606 | ||
1607 | bytes = ROUND_UP(bytes, align); | |
1608 | } | |
1609 | ||
3ea1a091 PB |
1610 | ret = bdrv_aligned_pwritev(bs, &req, offset, bytes, align, |
1611 | use_local_qiov ? &local_qiov : qiov, | |
1612 | flags); | |
61007b31 SH |
1613 | |
1614 | fail: | |
61007b31 SH |
1615 | |
1616 | if (use_local_qiov) { | |
1617 | qemu_iovec_destroy(&local_qiov); | |
1618 | } | |
1619 | qemu_vfree(head_buf); | |
1620 | qemu_vfree(tail_buf); | |
9eeb6dd1 FZ |
1621 | out: |
1622 | tracked_request_end(&req); | |
99723548 | 1623 | bdrv_dec_in_flight(bs); |
61007b31 SH |
1624 | return ret; |
1625 | } | |
1626 | ||
adad6496 | 1627 | static int coroutine_fn bdrv_co_do_writev(BdrvChild *child, |
61007b31 SH |
1628 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, |
1629 | BdrvRequestFlags flags) | |
1630 | { | |
1631 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
1632 | return -EINVAL; | |
1633 | } | |
1634 | ||
a03ef88f | 1635 | return bdrv_co_pwritev(child, sector_num << BDRV_SECTOR_BITS, |
cab3a356 | 1636 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); |
61007b31 SH |
1637 | } |
1638 | ||
25ec177d | 1639 | int coroutine_fn bdrv_co_writev(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
1640 | int nb_sectors, QEMUIOVector *qiov) |
1641 | { | |
25ec177d | 1642 | trace_bdrv_co_writev(child->bs, sector_num, nb_sectors); |
61007b31 | 1643 | |
adad6496 | 1644 | return bdrv_co_do_writev(child, sector_num, nb_sectors, qiov, 0); |
61007b31 SH |
1645 | } |
1646 | ||
a03ef88f KW |
1647 | int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, |
1648 | int count, BdrvRequestFlags flags) | |
61007b31 | 1649 | { |
a03ef88f | 1650 | trace_bdrv_co_pwrite_zeroes(child->bs, offset, count, flags); |
61007b31 | 1651 | |
a03ef88f | 1652 | if (!(child->bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
1653 | flags &= ~BDRV_REQ_MAY_UNMAP; |
1654 | } | |
61007b31 | 1655 | |
a03ef88f | 1656 | return bdrv_co_pwritev(child, offset, count, NULL, |
74021bc4 | 1657 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
1658 | } |
1659 | ||
4085f5c7 JS |
1660 | /* |
1661 | * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not. | |
1662 | */ | |
1663 | int bdrv_flush_all(void) | |
1664 | { | |
1665 | BdrvNextIterator it; | |
1666 | BlockDriverState *bs = NULL; | |
1667 | int result = 0; | |
1668 | ||
1669 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
1670 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
1671 | int ret; | |
1672 | ||
1673 | aio_context_acquire(aio_context); | |
1674 | ret = bdrv_flush(bs); | |
1675 | if (ret < 0 && !result) { | |
1676 | result = ret; | |
1677 | } | |
1678 | aio_context_release(aio_context); | |
1679 | } | |
1680 | ||
1681 | return result; | |
1682 | } | |
1683 | ||
1684 | ||
61007b31 SH |
1685 | typedef struct BdrvCoGetBlockStatusData { |
1686 | BlockDriverState *bs; | |
1687 | BlockDriverState *base; | |
67a0fd2a | 1688 | BlockDriverState **file; |
61007b31 SH |
1689 | int64_t sector_num; |
1690 | int nb_sectors; | |
1691 | int *pnum; | |
1692 | int64_t ret; | |
1693 | bool done; | |
1694 | } BdrvCoGetBlockStatusData; | |
1695 | ||
1696 | /* | |
1697 | * Returns the allocation status of the specified sectors. | |
1698 | * Drivers not implementing the functionality are assumed to not support | |
1699 | * backing files, hence all their sectors are reported as allocated. | |
1700 | * | |
1701 | * If 'sector_num' is beyond the end of the disk image the return value is 0 | |
1702 | * and 'pnum' is set to 0. | |
1703 | * | |
1704 | * 'pnum' is set to the number of sectors (including and immediately following | |
1705 | * the specified sector) that are known to be in the same | |
1706 | * allocated/unallocated state. | |
1707 | * | |
1708 | * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes | |
1709 | * beyond the end of the disk image it will be clamped. | |
67a0fd2a FZ |
1710 | * |
1711 | * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file' | |
1712 | * points to the BDS which the sector range is allocated in. | |
61007b31 SH |
1713 | */ |
1714 | static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, | |
1715 | int64_t sector_num, | |
67a0fd2a FZ |
1716 | int nb_sectors, int *pnum, |
1717 | BlockDriverState **file) | |
61007b31 SH |
1718 | { |
1719 | int64_t total_sectors; | |
1720 | int64_t n; | |
1721 | int64_t ret, ret2; | |
1722 | ||
1723 | total_sectors = bdrv_nb_sectors(bs); | |
1724 | if (total_sectors < 0) { | |
1725 | return total_sectors; | |
1726 | } | |
1727 | ||
1728 | if (sector_num >= total_sectors) { | |
1729 | *pnum = 0; | |
1730 | return 0; | |
1731 | } | |
1732 | ||
1733 | n = total_sectors - sector_num; | |
1734 | if (n < nb_sectors) { | |
1735 | nb_sectors = n; | |
1736 | } | |
1737 | ||
1738 | if (!bs->drv->bdrv_co_get_block_status) { | |
1739 | *pnum = nb_sectors; | |
1740 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; | |
1741 | if (bs->drv->protocol_name) { | |
1742 | ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE); | |
1743 | } | |
1744 | return ret; | |
1745 | } | |
1746 | ||
67a0fd2a | 1747 | *file = NULL; |
99723548 | 1748 | bdrv_inc_in_flight(bs); |
67a0fd2a FZ |
1749 | ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum, |
1750 | file); | |
61007b31 SH |
1751 | if (ret < 0) { |
1752 | *pnum = 0; | |
99723548 | 1753 | goto out; |
61007b31 SH |
1754 | } |
1755 | ||
1756 | if (ret & BDRV_BLOCK_RAW) { | |
1757 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
99723548 PB |
1758 | ret = bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS, |
1759 | *pnum, pnum, file); | |
1760 | goto out; | |
61007b31 SH |
1761 | } |
1762 | ||
1763 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
1764 | ret |= BDRV_BLOCK_ALLOCATED; | |
a53f1a95 | 1765 | } else { |
61007b31 SH |
1766 | if (bdrv_unallocated_blocks_are_zero(bs)) { |
1767 | ret |= BDRV_BLOCK_ZERO; | |
760e0063 KW |
1768 | } else if (bs->backing) { |
1769 | BlockDriverState *bs2 = bs->backing->bs; | |
61007b31 SH |
1770 | int64_t nb_sectors2 = bdrv_nb_sectors(bs2); |
1771 | if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) { | |
1772 | ret |= BDRV_BLOCK_ZERO; | |
1773 | } | |
1774 | } | |
1775 | } | |
1776 | ||
ac987b30 | 1777 | if (*file && *file != bs && |
61007b31 SH |
1778 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
1779 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
67a0fd2a | 1780 | BlockDriverState *file2; |
61007b31 SH |
1781 | int file_pnum; |
1782 | ||
ac987b30 | 1783 | ret2 = bdrv_co_get_block_status(*file, ret >> BDRV_SECTOR_BITS, |
67a0fd2a | 1784 | *pnum, &file_pnum, &file2); |
61007b31 SH |
1785 | if (ret2 >= 0) { |
1786 | /* Ignore errors. This is just providing extra information, it | |
1787 | * is useful but not necessary. | |
1788 | */ | |
1789 | if (!file_pnum) { | |
1790 | /* !file_pnum indicates an offset at or beyond the EOF; it is | |
1791 | * perfectly valid for the format block driver to point to such | |
1792 | * offsets, so catch it and mark everything as zero */ | |
1793 | ret |= BDRV_BLOCK_ZERO; | |
1794 | } else { | |
1795 | /* Limit request to the range reported by the protocol driver */ | |
1796 | *pnum = file_pnum; | |
1797 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
1798 | } | |
1799 | } | |
1800 | } | |
1801 | ||
99723548 PB |
1802 | out: |
1803 | bdrv_dec_in_flight(bs); | |
61007b31 SH |
1804 | return ret; |
1805 | } | |
1806 | ||
ba3f0e25 FZ |
1807 | static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs, |
1808 | BlockDriverState *base, | |
1809 | int64_t sector_num, | |
1810 | int nb_sectors, | |
67a0fd2a FZ |
1811 | int *pnum, |
1812 | BlockDriverState **file) | |
ba3f0e25 FZ |
1813 | { |
1814 | BlockDriverState *p; | |
1815 | int64_t ret = 0; | |
1816 | ||
1817 | assert(bs != base); | |
760e0063 | 1818 | for (p = bs; p != base; p = backing_bs(p)) { |
67a0fd2a | 1819 | ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum, file); |
ba3f0e25 FZ |
1820 | if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) { |
1821 | break; | |
1822 | } | |
1823 | /* [sector_num, pnum] unallocated on this layer, which could be only | |
1824 | * the first part of [sector_num, nb_sectors]. */ | |
1825 | nb_sectors = MIN(nb_sectors, *pnum); | |
1826 | } | |
1827 | return ret; | |
1828 | } | |
1829 | ||
1830 | /* Coroutine wrapper for bdrv_get_block_status_above() */ | |
1831 | static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque) | |
61007b31 SH |
1832 | { |
1833 | BdrvCoGetBlockStatusData *data = opaque; | |
61007b31 | 1834 | |
ba3f0e25 FZ |
1835 | data->ret = bdrv_co_get_block_status_above(data->bs, data->base, |
1836 | data->sector_num, | |
1837 | data->nb_sectors, | |
67a0fd2a FZ |
1838 | data->pnum, |
1839 | data->file); | |
61007b31 SH |
1840 | data->done = true; |
1841 | } | |
1842 | ||
1843 | /* | |
ba3f0e25 | 1844 | * Synchronous wrapper around bdrv_co_get_block_status_above(). |
61007b31 | 1845 | * |
ba3f0e25 | 1846 | * See bdrv_co_get_block_status_above() for details. |
61007b31 | 1847 | */ |
ba3f0e25 FZ |
1848 | int64_t bdrv_get_block_status_above(BlockDriverState *bs, |
1849 | BlockDriverState *base, | |
1850 | int64_t sector_num, | |
67a0fd2a FZ |
1851 | int nb_sectors, int *pnum, |
1852 | BlockDriverState **file) | |
61007b31 SH |
1853 | { |
1854 | Coroutine *co; | |
1855 | BdrvCoGetBlockStatusData data = { | |
1856 | .bs = bs, | |
ba3f0e25 | 1857 | .base = base, |
67a0fd2a | 1858 | .file = file, |
61007b31 SH |
1859 | .sector_num = sector_num, |
1860 | .nb_sectors = nb_sectors, | |
1861 | .pnum = pnum, | |
1862 | .done = false, | |
1863 | }; | |
1864 | ||
1865 | if (qemu_in_coroutine()) { | |
1866 | /* Fast-path if already in coroutine context */ | |
ba3f0e25 | 1867 | bdrv_get_block_status_above_co_entry(&data); |
61007b31 | 1868 | } else { |
0b8b8753 PB |
1869 | co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry, |
1870 | &data); | |
1871 | qemu_coroutine_enter(co); | |
88b062c2 | 1872 | BDRV_POLL_WHILE(bs, !data.done); |
61007b31 SH |
1873 | } |
1874 | return data.ret; | |
1875 | } | |
1876 | ||
ba3f0e25 FZ |
1877 | int64_t bdrv_get_block_status(BlockDriverState *bs, |
1878 | int64_t sector_num, | |
67a0fd2a FZ |
1879 | int nb_sectors, int *pnum, |
1880 | BlockDriverState **file) | |
ba3f0e25 | 1881 | { |
760e0063 | 1882 | return bdrv_get_block_status_above(bs, backing_bs(bs), |
67a0fd2a | 1883 | sector_num, nb_sectors, pnum, file); |
ba3f0e25 FZ |
1884 | } |
1885 | ||
61007b31 SH |
1886 | int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, |
1887 | int nb_sectors, int *pnum) | |
1888 | { | |
67a0fd2a FZ |
1889 | BlockDriverState *file; |
1890 | int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum, | |
1891 | &file); | |
61007b31 SH |
1892 | if (ret < 0) { |
1893 | return ret; | |
1894 | } | |
1895 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
1896 | } | |
1897 | ||
1898 | /* | |
1899 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
1900 | * | |
1901 | * Return true if the given sector is allocated in any image between | |
1902 | * BASE and TOP (inclusive). BASE can be NULL to check if the given | |
1903 | * sector is allocated in any image of the chain. Return false otherwise. | |
1904 | * | |
1905 | * 'pnum' is set to the number of sectors (including and immediately following | |
1906 | * the specified sector) that are known to be in the same | |
1907 | * allocated/unallocated state. | |
1908 | * | |
1909 | */ | |
1910 | int bdrv_is_allocated_above(BlockDriverState *top, | |
1911 | BlockDriverState *base, | |
1912 | int64_t sector_num, | |
1913 | int nb_sectors, int *pnum) | |
1914 | { | |
1915 | BlockDriverState *intermediate; | |
1916 | int ret, n = nb_sectors; | |
1917 | ||
1918 | intermediate = top; | |
1919 | while (intermediate && intermediate != base) { | |
1920 | int pnum_inter; | |
1921 | ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors, | |
1922 | &pnum_inter); | |
1923 | if (ret < 0) { | |
1924 | return ret; | |
1925 | } else if (ret) { | |
1926 | *pnum = pnum_inter; | |
1927 | return 1; | |
1928 | } | |
1929 | ||
1930 | /* | |
1931 | * [sector_num, nb_sectors] is unallocated on top but intermediate | |
1932 | * might have | |
1933 | * | |
1934 | * [sector_num+x, nr_sectors] allocated. | |
1935 | */ | |
1936 | if (n > pnum_inter && | |
1937 | (intermediate == top || | |
1938 | sector_num + pnum_inter < intermediate->total_sectors)) { | |
1939 | n = pnum_inter; | |
1940 | } | |
1941 | ||
760e0063 | 1942 | intermediate = backing_bs(intermediate); |
61007b31 SH |
1943 | } |
1944 | ||
1945 | *pnum = n; | |
1946 | return 0; | |
1947 | } | |
1948 | ||
1a8ae822 KW |
1949 | typedef struct BdrvVmstateCo { |
1950 | BlockDriverState *bs; | |
1951 | QEMUIOVector *qiov; | |
1952 | int64_t pos; | |
1953 | bool is_read; | |
1954 | int ret; | |
1955 | } BdrvVmstateCo; | |
1956 | ||
1957 | static int coroutine_fn | |
1958 | bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
1959 | bool is_read) | |
1960 | { | |
1961 | BlockDriver *drv = bs->drv; | |
1962 | ||
1963 | if (!drv) { | |
1964 | return -ENOMEDIUM; | |
1965 | } else if (drv->bdrv_load_vmstate) { | |
1966 | return is_read ? drv->bdrv_load_vmstate(bs, qiov, pos) | |
1967 | : drv->bdrv_save_vmstate(bs, qiov, pos); | |
1968 | } else if (bs->file) { | |
1969 | return bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read); | |
1970 | } | |
1971 | ||
1972 | return -ENOTSUP; | |
1973 | } | |
1974 | ||
1975 | static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque) | |
1976 | { | |
1977 | BdrvVmstateCo *co = opaque; | |
1978 | co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read); | |
1979 | } | |
1980 | ||
1981 | static inline int | |
1982 | bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
1983 | bool is_read) | |
1984 | { | |
1985 | if (qemu_in_coroutine()) { | |
1986 | return bdrv_co_rw_vmstate(bs, qiov, pos, is_read); | |
1987 | } else { | |
1988 | BdrvVmstateCo data = { | |
1989 | .bs = bs, | |
1990 | .qiov = qiov, | |
1991 | .pos = pos, | |
1992 | .is_read = is_read, | |
1993 | .ret = -EINPROGRESS, | |
1994 | }; | |
0b8b8753 | 1995 | Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data); |
1a8ae822 | 1996 | |
0b8b8753 | 1997 | qemu_coroutine_enter(co); |
1a8ae822 KW |
1998 | while (data.ret == -EINPROGRESS) { |
1999 | aio_poll(bdrv_get_aio_context(bs), true); | |
2000 | } | |
2001 | return data.ret; | |
2002 | } | |
2003 | } | |
2004 | ||
61007b31 SH |
2005 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
2006 | int64_t pos, int size) | |
2007 | { | |
2008 | QEMUIOVector qiov; | |
2009 | struct iovec iov = { | |
2010 | .iov_base = (void *) buf, | |
2011 | .iov_len = size, | |
2012 | }; | |
b433d942 | 2013 | int ret; |
61007b31 SH |
2014 | |
2015 | qemu_iovec_init_external(&qiov, &iov, 1); | |
b433d942 KW |
2016 | |
2017 | ret = bdrv_writev_vmstate(bs, &qiov, pos); | |
2018 | if (ret < 0) { | |
2019 | return ret; | |
2020 | } | |
2021 | ||
2022 | return size; | |
61007b31 SH |
2023 | } |
2024 | ||
2025 | int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
2026 | { | |
1a8ae822 | 2027 | return bdrv_rw_vmstate(bs, qiov, pos, false); |
61007b31 SH |
2028 | } |
2029 | ||
2030 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
2031 | int64_t pos, int size) | |
5ddda0b8 KW |
2032 | { |
2033 | QEMUIOVector qiov; | |
2034 | struct iovec iov = { | |
2035 | .iov_base = buf, | |
2036 | .iov_len = size, | |
2037 | }; | |
b433d942 | 2038 | int ret; |
5ddda0b8 KW |
2039 | |
2040 | qemu_iovec_init_external(&qiov, &iov, 1); | |
b433d942 KW |
2041 | ret = bdrv_readv_vmstate(bs, &qiov, pos); |
2042 | if (ret < 0) { | |
2043 | return ret; | |
2044 | } | |
2045 | ||
2046 | return size; | |
5ddda0b8 KW |
2047 | } |
2048 | ||
2049 | int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 2050 | { |
1a8ae822 | 2051 | return bdrv_rw_vmstate(bs, qiov, pos, true); |
61007b31 SH |
2052 | } |
2053 | ||
2054 | /**************************************************************/ | |
2055 | /* async I/Os */ | |
2056 | ||
ebb7af21 | 2057 | BlockAIOCB *bdrv_aio_readv(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
2058 | QEMUIOVector *qiov, int nb_sectors, |
2059 | BlockCompletionFunc *cb, void *opaque) | |
2060 | { | |
ebb7af21 | 2061 | trace_bdrv_aio_readv(child->bs, sector_num, nb_sectors, opaque); |
61007b31 | 2062 | |
b15404e0 EB |
2063 | assert(nb_sectors << BDRV_SECTOR_BITS == qiov->size); |
2064 | return bdrv_co_aio_prw_vector(child, sector_num << BDRV_SECTOR_BITS, qiov, | |
2065 | 0, cb, opaque, false); | |
61007b31 SH |
2066 | } |
2067 | ||
0d1049c7 | 2068 | BlockAIOCB *bdrv_aio_writev(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
2069 | QEMUIOVector *qiov, int nb_sectors, |
2070 | BlockCompletionFunc *cb, void *opaque) | |
2071 | { | |
0d1049c7 | 2072 | trace_bdrv_aio_writev(child->bs, sector_num, nb_sectors, opaque); |
61007b31 | 2073 | |
b15404e0 EB |
2074 | assert(nb_sectors << BDRV_SECTOR_BITS == qiov->size); |
2075 | return bdrv_co_aio_prw_vector(child, sector_num << BDRV_SECTOR_BITS, qiov, | |
2076 | 0, cb, opaque, true); | |
61007b31 SH |
2077 | } |
2078 | ||
61007b31 SH |
2079 | void bdrv_aio_cancel(BlockAIOCB *acb) |
2080 | { | |
2081 | qemu_aio_ref(acb); | |
2082 | bdrv_aio_cancel_async(acb); | |
2083 | while (acb->refcnt > 1) { | |
2084 | if (acb->aiocb_info->get_aio_context) { | |
2085 | aio_poll(acb->aiocb_info->get_aio_context(acb), true); | |
2086 | } else if (acb->bs) { | |
2087 | aio_poll(bdrv_get_aio_context(acb->bs), true); | |
2088 | } else { | |
2089 | abort(); | |
2090 | } | |
2091 | } | |
2092 | qemu_aio_unref(acb); | |
2093 | } | |
2094 | ||
2095 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2096 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2097 | * In either case the completion callback must be called. */ | |
2098 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2099 | { | |
2100 | if (acb->aiocb_info->cancel_async) { | |
2101 | acb->aiocb_info->cancel_async(acb); | |
2102 | } | |
2103 | } | |
2104 | ||
2105 | /**************************************************************/ | |
2106 | /* async block device emulation */ | |
2107 | ||
41574268 EB |
2108 | typedef struct BlockRequest { |
2109 | union { | |
2110 | /* Used during read, write, trim */ | |
2111 | struct { | |
b15404e0 EB |
2112 | int64_t offset; |
2113 | int bytes; | |
41574268 EB |
2114 | int flags; |
2115 | QEMUIOVector *qiov; | |
2116 | }; | |
2117 | /* Used during ioctl */ | |
2118 | struct { | |
2119 | int req; | |
2120 | void *buf; | |
2121 | }; | |
2122 | }; | |
2123 | BlockCompletionFunc *cb; | |
2124 | void *opaque; | |
2125 | ||
2126 | int error; | |
2127 | } BlockRequest; | |
2128 | ||
61007b31 SH |
2129 | typedef struct BlockAIOCBCoroutine { |
2130 | BlockAIOCB common; | |
adad6496 | 2131 | BdrvChild *child; |
61007b31 SH |
2132 | BlockRequest req; |
2133 | bool is_write; | |
2134 | bool need_bh; | |
2135 | bool *done; | |
61007b31 SH |
2136 | } BlockAIOCBCoroutine; |
2137 | ||
2138 | static const AIOCBInfo bdrv_em_co_aiocb_info = { | |
2139 | .aiocb_size = sizeof(BlockAIOCBCoroutine), | |
2140 | }; | |
2141 | ||
2142 | static void bdrv_co_complete(BlockAIOCBCoroutine *acb) | |
2143 | { | |
2144 | if (!acb->need_bh) { | |
99723548 | 2145 | bdrv_dec_in_flight(acb->common.bs); |
61007b31 SH |
2146 | acb->common.cb(acb->common.opaque, acb->req.error); |
2147 | qemu_aio_unref(acb); | |
2148 | } | |
2149 | } | |
2150 | ||
2151 | static void bdrv_co_em_bh(void *opaque) | |
2152 | { | |
2153 | BlockAIOCBCoroutine *acb = opaque; | |
2154 | ||
2155 | assert(!acb->need_bh); | |
61007b31 SH |
2156 | bdrv_co_complete(acb); |
2157 | } | |
2158 | ||
2159 | static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb) | |
2160 | { | |
2161 | acb->need_bh = false; | |
2162 | if (acb->req.error != -EINPROGRESS) { | |
2163 | BlockDriverState *bs = acb->common.bs; | |
2164 | ||
fffb6e12 | 2165 | aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb); |
61007b31 SH |
2166 | } |
2167 | } | |
2168 | ||
2169 | /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */ | |
2170 | static void coroutine_fn bdrv_co_do_rw(void *opaque) | |
2171 | { | |
2172 | BlockAIOCBCoroutine *acb = opaque; | |
61007b31 SH |
2173 | |
2174 | if (!acb->is_write) { | |
b15404e0 EB |
2175 | acb->req.error = bdrv_co_preadv(acb->child, acb->req.offset, |
2176 | acb->req.qiov->size, acb->req.qiov, acb->req.flags); | |
61007b31 | 2177 | } else { |
b15404e0 EB |
2178 | acb->req.error = bdrv_co_pwritev(acb->child, acb->req.offset, |
2179 | acb->req.qiov->size, acb->req.qiov, acb->req.flags); | |
61007b31 SH |
2180 | } |
2181 | ||
2182 | bdrv_co_complete(acb); | |
2183 | } | |
2184 | ||
b15404e0 EB |
2185 | static BlockAIOCB *bdrv_co_aio_prw_vector(BdrvChild *child, |
2186 | int64_t offset, | |
2187 | QEMUIOVector *qiov, | |
2188 | BdrvRequestFlags flags, | |
2189 | BlockCompletionFunc *cb, | |
2190 | void *opaque, | |
2191 | bool is_write) | |
61007b31 SH |
2192 | { |
2193 | Coroutine *co; | |
2194 | BlockAIOCBCoroutine *acb; | |
2195 | ||
99723548 PB |
2196 | /* Matched by bdrv_co_complete's bdrv_dec_in_flight. */ |
2197 | bdrv_inc_in_flight(child->bs); | |
2198 | ||
adad6496 KW |
2199 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, child->bs, cb, opaque); |
2200 | acb->child = child; | |
61007b31 SH |
2201 | acb->need_bh = true; |
2202 | acb->req.error = -EINPROGRESS; | |
b15404e0 | 2203 | acb->req.offset = offset; |
61007b31 SH |
2204 | acb->req.qiov = qiov; |
2205 | acb->req.flags = flags; | |
2206 | acb->is_write = is_write; | |
2207 | ||
0b8b8753 PB |
2208 | co = qemu_coroutine_create(bdrv_co_do_rw, acb); |
2209 | qemu_coroutine_enter(co); | |
61007b31 SH |
2210 | |
2211 | bdrv_co_maybe_schedule_bh(acb); | |
2212 | return &acb->common; | |
2213 | } | |
2214 | ||
2215 | static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) | |
2216 | { | |
2217 | BlockAIOCBCoroutine *acb = opaque; | |
2218 | BlockDriverState *bs = acb->common.bs; | |
2219 | ||
2220 | acb->req.error = bdrv_co_flush(bs); | |
2221 | bdrv_co_complete(acb); | |
2222 | } | |
2223 | ||
2224 | BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs, | |
2225 | BlockCompletionFunc *cb, void *opaque) | |
2226 | { | |
2227 | trace_bdrv_aio_flush(bs, opaque); | |
2228 | ||
2229 | Coroutine *co; | |
2230 | BlockAIOCBCoroutine *acb; | |
2231 | ||
99723548 PB |
2232 | /* Matched by bdrv_co_complete's bdrv_dec_in_flight. */ |
2233 | bdrv_inc_in_flight(bs); | |
2234 | ||
61007b31 SH |
2235 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); |
2236 | acb->need_bh = true; | |
2237 | acb->req.error = -EINPROGRESS; | |
2238 | ||
0b8b8753 PB |
2239 | co = qemu_coroutine_create(bdrv_aio_flush_co_entry, acb); |
2240 | qemu_coroutine_enter(co); | |
61007b31 SH |
2241 | |
2242 | bdrv_co_maybe_schedule_bh(acb); | |
2243 | return &acb->common; | |
2244 | } | |
2245 | ||
61007b31 SH |
2246 | void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, |
2247 | BlockCompletionFunc *cb, void *opaque) | |
2248 | { | |
2249 | BlockAIOCB *acb; | |
2250 | ||
c84b3192 | 2251 | acb = g_malloc(aiocb_info->aiocb_size); |
61007b31 SH |
2252 | acb->aiocb_info = aiocb_info; |
2253 | acb->bs = bs; | |
2254 | acb->cb = cb; | |
2255 | acb->opaque = opaque; | |
2256 | acb->refcnt = 1; | |
2257 | return acb; | |
2258 | } | |
2259 | ||
2260 | void qemu_aio_ref(void *p) | |
2261 | { | |
2262 | BlockAIOCB *acb = p; | |
2263 | acb->refcnt++; | |
2264 | } | |
2265 | ||
2266 | void qemu_aio_unref(void *p) | |
2267 | { | |
2268 | BlockAIOCB *acb = p; | |
2269 | assert(acb->refcnt > 0); | |
2270 | if (--acb->refcnt == 0) { | |
c84b3192 | 2271 | g_free(acb); |
61007b31 SH |
2272 | } |
2273 | } | |
2274 | ||
2275 | /**************************************************************/ | |
2276 | /* Coroutine block device emulation */ | |
2277 | ||
e293b7a3 KW |
2278 | typedef struct FlushCo { |
2279 | BlockDriverState *bs; | |
2280 | int ret; | |
2281 | } FlushCo; | |
2282 | ||
2283 | ||
61007b31 SH |
2284 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) |
2285 | { | |
e293b7a3 | 2286 | FlushCo *rwco = opaque; |
61007b31 SH |
2287 | |
2288 | rwco->ret = bdrv_co_flush(rwco->bs); | |
2289 | } | |
2290 | ||
2291 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
2292 | { | |
2293 | int ret; | |
2294 | ||
1b6bc94d DA |
2295 | if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) || |
2296 | bdrv_is_sg(bs)) { | |
61007b31 SH |
2297 | return 0; |
2298 | } | |
2299 | ||
99723548 | 2300 | bdrv_inc_in_flight(bs); |
c32b82af | 2301 | |
3ff2f67a EY |
2302 | int current_gen = bs->write_gen; |
2303 | ||
2304 | /* Wait until any previous flushes are completed */ | |
99723548 | 2305 | while (bs->active_flush_req) { |
3ff2f67a EY |
2306 | qemu_co_queue_wait(&bs->flush_queue); |
2307 | } | |
2308 | ||
99723548 | 2309 | bs->active_flush_req = true; |
3ff2f67a | 2310 | |
c32b82af PD |
2311 | /* Write back all layers by calling one driver function */ |
2312 | if (bs->drv->bdrv_co_flush) { | |
2313 | ret = bs->drv->bdrv_co_flush(bs); | |
2314 | goto out; | |
2315 | } | |
2316 | ||
61007b31 SH |
2317 | /* Write back cached data to the OS even with cache=unsafe */ |
2318 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); | |
2319 | if (bs->drv->bdrv_co_flush_to_os) { | |
2320 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
2321 | if (ret < 0) { | |
cdb5e315 | 2322 | goto out; |
61007b31 SH |
2323 | } |
2324 | } | |
2325 | ||
2326 | /* But don't actually force it to the disk with cache=unsafe */ | |
2327 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
2328 | goto flush_parent; | |
2329 | } | |
2330 | ||
3ff2f67a EY |
2331 | /* Check if we really need to flush anything */ |
2332 | if (bs->flushed_gen == current_gen) { | |
2333 | goto flush_parent; | |
2334 | } | |
2335 | ||
61007b31 SH |
2336 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); |
2337 | if (bs->drv->bdrv_co_flush_to_disk) { | |
2338 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
2339 | } else if (bs->drv->bdrv_aio_flush) { | |
2340 | BlockAIOCB *acb; | |
2341 | CoroutineIOCompletion co = { | |
2342 | .coroutine = qemu_coroutine_self(), | |
2343 | }; | |
2344 | ||
2345 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
2346 | if (acb == NULL) { | |
2347 | ret = -EIO; | |
2348 | } else { | |
2349 | qemu_coroutine_yield(); | |
2350 | ret = co.ret; | |
2351 | } | |
2352 | } else { | |
2353 | /* | |
2354 | * Some block drivers always operate in either writethrough or unsafe | |
2355 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
2356 | * know how the server works (because the behaviour is hardcoded or | |
2357 | * depends on server-side configuration), so we can't ensure that | |
2358 | * everything is safe on disk. Returning an error doesn't work because | |
2359 | * that would break guests even if the server operates in writethrough | |
2360 | * mode. | |
2361 | * | |
2362 | * Let's hope the user knows what he's doing. | |
2363 | */ | |
2364 | ret = 0; | |
2365 | } | |
3ff2f67a | 2366 | |
61007b31 | 2367 | if (ret < 0) { |
cdb5e315 | 2368 | goto out; |
61007b31 SH |
2369 | } |
2370 | ||
2371 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
2372 | * in the case of cache=unsafe, so there are no useless flushes. | |
2373 | */ | |
2374 | flush_parent: | |
cdb5e315 FZ |
2375 | ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0; |
2376 | out: | |
3ff2f67a | 2377 | /* Notify any pending flushes that we have completed */ |
e6af1e08 KW |
2378 | if (ret == 0) { |
2379 | bs->flushed_gen = current_gen; | |
2380 | } | |
99723548 | 2381 | bs->active_flush_req = false; |
156af3ac DL |
2382 | /* Return value is ignored - it's ok if wait queue is empty */ |
2383 | qemu_co_queue_next(&bs->flush_queue); | |
3ff2f67a | 2384 | |
99723548 | 2385 | bdrv_dec_in_flight(bs); |
cdb5e315 | 2386 | return ret; |
61007b31 SH |
2387 | } |
2388 | ||
2389 | int bdrv_flush(BlockDriverState *bs) | |
2390 | { | |
2391 | Coroutine *co; | |
e293b7a3 | 2392 | FlushCo flush_co = { |
61007b31 SH |
2393 | .bs = bs, |
2394 | .ret = NOT_DONE, | |
2395 | }; | |
2396 | ||
2397 | if (qemu_in_coroutine()) { | |
2398 | /* Fast-path if already in coroutine context */ | |
e293b7a3 | 2399 | bdrv_flush_co_entry(&flush_co); |
61007b31 | 2400 | } else { |
0b8b8753 PB |
2401 | co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co); |
2402 | qemu_coroutine_enter(co); | |
88b062c2 | 2403 | BDRV_POLL_WHILE(bs, flush_co.ret == NOT_DONE); |
61007b31 SH |
2404 | } |
2405 | ||
e293b7a3 | 2406 | return flush_co.ret; |
61007b31 SH |
2407 | } |
2408 | ||
2409 | typedef struct DiscardCo { | |
2410 | BlockDriverState *bs; | |
0c51a893 EB |
2411 | int64_t offset; |
2412 | int count; | |
61007b31 SH |
2413 | int ret; |
2414 | } DiscardCo; | |
0c51a893 | 2415 | static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque) |
61007b31 SH |
2416 | { |
2417 | DiscardCo *rwco = opaque; | |
2418 | ||
0c51a893 | 2419 | rwco->ret = bdrv_co_pdiscard(rwco->bs, rwco->offset, rwco->count); |
61007b31 SH |
2420 | } |
2421 | ||
9f1963b3 EB |
2422 | int coroutine_fn bdrv_co_pdiscard(BlockDriverState *bs, int64_t offset, |
2423 | int count) | |
61007b31 | 2424 | { |
b1066c87 | 2425 | BdrvTrackedRequest req; |
9f1963b3 | 2426 | int max_pdiscard, ret; |
3482b9bc | 2427 | int head, tail, align; |
61007b31 SH |
2428 | |
2429 | if (!bs->drv) { | |
2430 | return -ENOMEDIUM; | |
2431 | } | |
2432 | ||
9f1963b3 | 2433 | ret = bdrv_check_byte_request(bs, offset, count); |
61007b31 SH |
2434 | if (ret < 0) { |
2435 | return ret; | |
2436 | } else if (bs->read_only) { | |
eaf5fe2d | 2437 | return -EPERM; |
61007b31 | 2438 | } |
04c01a5c | 2439 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
61007b31 | 2440 | |
61007b31 SH |
2441 | /* Do nothing if disabled. */ |
2442 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
2443 | return 0; | |
2444 | } | |
2445 | ||
02aefe43 | 2446 | if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) { |
61007b31 SH |
2447 | return 0; |
2448 | } | |
2449 | ||
3482b9bc EB |
2450 | /* Discard is advisory, but some devices track and coalesce |
2451 | * unaligned requests, so we must pass everything down rather than | |
2452 | * round here. Still, most devices will just silently ignore | |
2453 | * unaligned requests (by returning -ENOTSUP), so we must fragment | |
2454 | * the request accordingly. */ | |
02aefe43 | 2455 | align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment); |
b8d0a980 EB |
2456 | assert(align % bs->bl.request_alignment == 0); |
2457 | head = offset % align; | |
3482b9bc | 2458 | tail = (offset + count) % align; |
9f1963b3 | 2459 | |
99723548 | 2460 | bdrv_inc_in_flight(bs); |
9f1963b3 | 2461 | tracked_request_begin(&req, bs, offset, count, BDRV_TRACKED_DISCARD); |
50824995 | 2462 | |
ec050f77 DL |
2463 | ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req); |
2464 | if (ret < 0) { | |
2465 | goto out; | |
2466 | } | |
2467 | ||
9f1963b3 EB |
2468 | max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX), |
2469 | align); | |
3482b9bc | 2470 | assert(max_pdiscard >= bs->bl.request_alignment); |
61007b31 | 2471 | |
9f1963b3 EB |
2472 | while (count > 0) { |
2473 | int ret; | |
3482b9bc EB |
2474 | int num = count; |
2475 | ||
2476 | if (head) { | |
2477 | /* Make small requests to get to alignment boundaries. */ | |
2478 | num = MIN(count, align - head); | |
2479 | if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) { | |
2480 | num %= bs->bl.request_alignment; | |
2481 | } | |
2482 | head = (head + num) % align; | |
2483 | assert(num < max_pdiscard); | |
2484 | } else if (tail) { | |
2485 | if (num > align) { | |
2486 | /* Shorten the request to the last aligned cluster. */ | |
2487 | num -= tail; | |
2488 | } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) && | |
2489 | tail > bs->bl.request_alignment) { | |
2490 | tail %= bs->bl.request_alignment; | |
2491 | num -= tail; | |
2492 | } | |
2493 | } | |
2494 | /* limit request size */ | |
2495 | if (num > max_pdiscard) { | |
2496 | num = max_pdiscard; | |
2497 | } | |
61007b31 | 2498 | |
47a5486d EB |
2499 | if (bs->drv->bdrv_co_pdiscard) { |
2500 | ret = bs->drv->bdrv_co_pdiscard(bs, offset, num); | |
61007b31 SH |
2501 | } else { |
2502 | BlockAIOCB *acb; | |
2503 | CoroutineIOCompletion co = { | |
2504 | .coroutine = qemu_coroutine_self(), | |
2505 | }; | |
2506 | ||
4da444a0 EB |
2507 | acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num, |
2508 | bdrv_co_io_em_complete, &co); | |
61007b31 | 2509 | if (acb == NULL) { |
b1066c87 FZ |
2510 | ret = -EIO; |
2511 | goto out; | |
61007b31 SH |
2512 | } else { |
2513 | qemu_coroutine_yield(); | |
2514 | ret = co.ret; | |
2515 | } | |
2516 | } | |
2517 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 2518 | goto out; |
61007b31 SH |
2519 | } |
2520 | ||
9f1963b3 EB |
2521 | offset += num; |
2522 | count -= num; | |
61007b31 | 2523 | } |
b1066c87 FZ |
2524 | ret = 0; |
2525 | out: | |
3ff2f67a | 2526 | ++bs->write_gen; |
968d8b06 DL |
2527 | bdrv_set_dirty(bs, req.offset >> BDRV_SECTOR_BITS, |
2528 | req.bytes >> BDRV_SECTOR_BITS); | |
b1066c87 | 2529 | tracked_request_end(&req); |
99723548 | 2530 | bdrv_dec_in_flight(bs); |
b1066c87 | 2531 | return ret; |
61007b31 SH |
2532 | } |
2533 | ||
0c51a893 | 2534 | int bdrv_pdiscard(BlockDriverState *bs, int64_t offset, int count) |
61007b31 SH |
2535 | { |
2536 | Coroutine *co; | |
2537 | DiscardCo rwco = { | |
2538 | .bs = bs, | |
0c51a893 EB |
2539 | .offset = offset, |
2540 | .count = count, | |
61007b31 SH |
2541 | .ret = NOT_DONE, |
2542 | }; | |
2543 | ||
2544 | if (qemu_in_coroutine()) { | |
2545 | /* Fast-path if already in coroutine context */ | |
0c51a893 | 2546 | bdrv_pdiscard_co_entry(&rwco); |
61007b31 | 2547 | } else { |
0c51a893 | 2548 | co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco); |
0b8b8753 | 2549 | qemu_coroutine_enter(co); |
88b062c2 | 2550 | BDRV_POLL_WHILE(bs, rwco.ret == NOT_DONE); |
61007b31 SH |
2551 | } |
2552 | ||
2553 | return rwco.ret; | |
2554 | } | |
2555 | ||
48af776a | 2556 | int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
2557 | { |
2558 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
2559 | CoroutineIOCompletion co = { |
2560 | .coroutine = qemu_coroutine_self(), | |
2561 | }; | |
2562 | BlockAIOCB *acb; | |
61007b31 | 2563 | |
99723548 | 2564 | bdrv_inc_in_flight(bs); |
16a389dc | 2565 | if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) { |
5c5ae76a FZ |
2566 | co.ret = -ENOTSUP; |
2567 | goto out; | |
2568 | } | |
2569 | ||
16a389dc KW |
2570 | if (drv->bdrv_co_ioctl) { |
2571 | co.ret = drv->bdrv_co_ioctl(bs, req, buf); | |
2572 | } else { | |
2573 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
2574 | if (!acb) { | |
2575 | co.ret = -ENOTSUP; | |
2576 | goto out; | |
2577 | } | |
2578 | qemu_coroutine_yield(); | |
5c5ae76a | 2579 | } |
5c5ae76a | 2580 | out: |
99723548 | 2581 | bdrv_dec_in_flight(bs); |
5c5ae76a FZ |
2582 | return co.ret; |
2583 | } | |
2584 | ||
61007b31 SH |
2585 | void *qemu_blockalign(BlockDriverState *bs, size_t size) |
2586 | { | |
2587 | return qemu_memalign(bdrv_opt_mem_align(bs), size); | |
2588 | } | |
2589 | ||
2590 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
2591 | { | |
2592 | return memset(qemu_blockalign(bs, size), 0, size); | |
2593 | } | |
2594 | ||
2595 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
2596 | { | |
2597 | size_t align = bdrv_opt_mem_align(bs); | |
2598 | ||
2599 | /* Ensure that NULL is never returned on success */ | |
2600 | assert(align > 0); | |
2601 | if (size == 0) { | |
2602 | size = align; | |
2603 | } | |
2604 | ||
2605 | return qemu_try_memalign(align, size); | |
2606 | } | |
2607 | ||
2608 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
2609 | { | |
2610 | void *mem = qemu_try_blockalign(bs, size); | |
2611 | ||
2612 | if (mem) { | |
2613 | memset(mem, 0, size); | |
2614 | } | |
2615 | ||
2616 | return mem; | |
2617 | } | |
2618 | ||
2619 | /* | |
2620 | * Check if all memory in this vector is sector aligned. | |
2621 | */ | |
2622 | bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
2623 | { | |
2624 | int i; | |
4196d2f0 | 2625 | size_t alignment = bdrv_min_mem_align(bs); |
61007b31 SH |
2626 | |
2627 | for (i = 0; i < qiov->niov; i++) { | |
2628 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
2629 | return false; | |
2630 | } | |
2631 | if (qiov->iov[i].iov_len % alignment) { | |
2632 | return false; | |
2633 | } | |
2634 | } | |
2635 | ||
2636 | return true; | |
2637 | } | |
2638 | ||
2639 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
2640 | NotifierWithReturn *notifier) | |
2641 | { | |
2642 | notifier_with_return_list_add(&bs->before_write_notifiers, notifier); | |
2643 | } | |
2644 | ||
2645 | void bdrv_io_plug(BlockDriverState *bs) | |
2646 | { | |
6b98bd64 PB |
2647 | BdrvChild *child; |
2648 | ||
2649 | QLIST_FOREACH(child, &bs->children, next) { | |
2650 | bdrv_io_plug(child->bs); | |
2651 | } | |
2652 | ||
2653 | if (bs->io_plugged++ == 0 && bs->io_plug_disabled == 0) { | |
2654 | BlockDriver *drv = bs->drv; | |
2655 | if (drv && drv->bdrv_io_plug) { | |
2656 | drv->bdrv_io_plug(bs); | |
2657 | } | |
61007b31 SH |
2658 | } |
2659 | } | |
2660 | ||
2661 | void bdrv_io_unplug(BlockDriverState *bs) | |
2662 | { | |
6b98bd64 PB |
2663 | BdrvChild *child; |
2664 | ||
2665 | assert(bs->io_plugged); | |
2666 | if (--bs->io_plugged == 0 && bs->io_plug_disabled == 0) { | |
2667 | BlockDriver *drv = bs->drv; | |
2668 | if (drv && drv->bdrv_io_unplug) { | |
2669 | drv->bdrv_io_unplug(bs); | |
2670 | } | |
2671 | } | |
2672 | ||
2673 | QLIST_FOREACH(child, &bs->children, next) { | |
2674 | bdrv_io_unplug(child->bs); | |
61007b31 SH |
2675 | } |
2676 | } | |
2677 | ||
6b98bd64 | 2678 | void bdrv_io_unplugged_begin(BlockDriverState *bs) |
61007b31 | 2679 | { |
6b98bd64 PB |
2680 | BdrvChild *child; |
2681 | ||
2682 | if (bs->io_plug_disabled++ == 0 && bs->io_plugged > 0) { | |
2683 | BlockDriver *drv = bs->drv; | |
2684 | if (drv && drv->bdrv_io_unplug) { | |
2685 | drv->bdrv_io_unplug(bs); | |
2686 | } | |
2687 | } | |
2688 | ||
2689 | QLIST_FOREACH(child, &bs->children, next) { | |
2690 | bdrv_io_unplugged_begin(child->bs); | |
2691 | } | |
2692 | } | |
2693 | ||
2694 | void bdrv_io_unplugged_end(BlockDriverState *bs) | |
2695 | { | |
2696 | BdrvChild *child; | |
2697 | ||
2698 | assert(bs->io_plug_disabled); | |
2699 | QLIST_FOREACH(child, &bs->children, next) { | |
2700 | bdrv_io_unplugged_end(child->bs); | |
2701 | } | |
2702 | ||
2703 | if (--bs->io_plug_disabled == 0 && bs->io_plugged > 0) { | |
2704 | BlockDriver *drv = bs->drv; | |
2705 | if (drv && drv->bdrv_io_plug) { | |
2706 | drv->bdrv_io_plug(bs); | |
2707 | } | |
61007b31 SH |
2708 | } |
2709 | } |