1 /*
2 * Block layer I/O functions
3 *
4 * Copyright (c) 2003 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25 #include "qemu/osdep.h"
26 #include "trace.h"
27 #include "sysemu/block-backend.h"
28 #include "block/aio-wait.h"
29 #include "block/blockjob.h"
30 #include "block/blockjob_int.h"
31 #include "block/block_int.h"
32 #include "qemu/cutils.h"
33 #include "qapi/error.h"
34 #include "qemu/error-report.h"
35
36 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
37
38 /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */
39 #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
40
41 static void bdrv_parent_cb_resize(BlockDriverState *bs);
42 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
43 int64_t offset, int bytes, BdrvRequestFlags flags);
44
bdrv_parent_drained_begin(BlockDriverState * bs,BdrvChild * ignore,bool ignore_bds_parents)45 void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore,
46 bool ignore_bds_parents)
47 {
48 BdrvChild *c, *next;
49
50 QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
51 if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) {
52 continue;
53 }
54 bdrv_parent_drained_begin_single(c, false);
55 }
56 }
57
bdrv_parent_drained_end(BlockDriverState * bs,BdrvChild * ignore,bool ignore_bds_parents)58 void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore,
59 bool ignore_bds_parents)
60 {
61 BdrvChild *c, *next;
62
63 QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
64 if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) {
65 continue;
66 }
67 if (c->role->drained_end) {
68 c->role->drained_end(c);
69 }
70 }
71 }
72
bdrv_parent_drained_poll_single(BdrvChild * c)73 static bool bdrv_parent_drained_poll_single(BdrvChild *c)
74 {
75 if (c->role->drained_poll) {
76 return c->role->drained_poll(c);
77 }
78 return false;
79 }
80
bdrv_parent_drained_poll(BlockDriverState * bs,BdrvChild * ignore,bool ignore_bds_parents)81 static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore,
82 bool ignore_bds_parents)
83 {
84 BdrvChild *c, *next;
85 bool busy = false;
86
87 QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
88 if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) {
89 continue;
90 }
91 busy |= bdrv_parent_drained_poll_single(c);
92 }
93
94 return busy;
95 }
96
bdrv_parent_drained_begin_single(BdrvChild * c,bool poll)97 void bdrv_parent_drained_begin_single(BdrvChild *c, bool poll)
98 {
99 if (c->role->drained_begin) {
100 c->role->drained_begin(c);
101 }
102 if (poll) {
103 BDRV_POLL_WHILE(c->bs, bdrv_parent_drained_poll_single(c));
104 }
105 }
106
bdrv_merge_limits(BlockLimits * dst,const BlockLimits * src)107 static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src)
108 {
109 dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer);
110 dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer);
111 dst->opt_mem_alignment = MAX(dst->opt_mem_alignment,
112 src->opt_mem_alignment);
113 dst->min_mem_alignment = MAX(dst->min_mem_alignment,
114 src->min_mem_alignment);
115 dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov);
116 }
117
bdrv_refresh_limits(BlockDriverState * bs,Error ** errp)118 void bdrv_refresh_limits(BlockDriverState *bs, Error **errp)
119 {
120 BlockDriver *drv = bs->drv;
121 Error *local_err = NULL;
122
123 memset(&bs->bl, 0, sizeof(bs->bl));
124
125 if (!drv) {
126 return;
127 }
128
129 /* Default alignment based on whether driver has byte interface */
130 bs->bl.request_alignment = (drv->bdrv_co_preadv ||
131 drv->bdrv_aio_preadv) ? 1 : 512;
132
133 /* Take some limits from the children as a default */
134 if (bs->file) {
135 bdrv_refresh_limits(bs->file->bs, &local_err);
136 if (local_err) {
137 error_propagate(errp, local_err);
138 return;
139 }
140 bdrv_merge_limits(&bs->bl, &bs->file->bs->bl);
141 } else {
142 bs->bl.min_mem_alignment = 512;
143 bs->bl.opt_mem_alignment = getpagesize();
144
145 /* Safe default since most protocols use readv()/writev()/etc */
146 bs->bl.max_iov = IOV_MAX;
147 }
148
149 if (bs->backing) {
150 bdrv_refresh_limits(bs->backing->bs, &local_err);
151 if (local_err) {
152 error_propagate(errp, local_err);
153 return;
154 }
155 bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl);
156 }
157
158 /* Then let the driver override it */
159 if (drv->bdrv_refresh_limits) {
160 drv->bdrv_refresh_limits(bs, errp);
161 }
162 }
163
164 /**
165 * The copy-on-read flag is actually a reference count so multiple users may
166 * use the feature without worrying about clobbering its previous state.
167 * Copy-on-read stays enabled until all users have called to disable it.
168 */
bdrv_enable_copy_on_read(BlockDriverState * bs)169 void bdrv_enable_copy_on_read(BlockDriverState *bs)
170 {
171 atomic_inc(&bs->copy_on_read);
172 }
173
bdrv_disable_copy_on_read(BlockDriverState * bs)174 void bdrv_disable_copy_on_read(BlockDriverState *bs)
175 {
176 int old = atomic_fetch_dec(&bs->copy_on_read);
177 assert(old >= 1);
178 }
179
180 typedef struct {
181 Coroutine *co;
182 BlockDriverState *bs;
183 bool done;
184 bool begin;
185 bool recursive;
186 bool poll;
187 BdrvChild *parent;
188 bool ignore_bds_parents;
189 } BdrvCoDrainData;
190
bdrv_drain_invoke_entry(void * opaque)191 static void coroutine_fn bdrv_drain_invoke_entry(void *opaque)
192 {
193 BdrvCoDrainData *data = opaque;
194 BlockDriverState *bs = data->bs;
195
196 if (data->begin) {
197 bs->drv->bdrv_co_drain_begin(bs);
198 } else {
199 bs->drv->bdrv_co_drain_end(bs);
200 }
201
202 /* Set data->done before reading bs->wakeup. */
203 atomic_mb_set(&data->done, true);
204 bdrv_dec_in_flight(bs);
205
206 if (data->begin) {
207 g_free(data);
208 }
209 }
210
211 /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */
bdrv_drain_invoke(BlockDriverState * bs,bool begin)212 static void bdrv_drain_invoke(BlockDriverState *bs, bool begin)
213 {
214 BdrvCoDrainData *data;
215
216 if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) ||
217 (!begin && !bs->drv->bdrv_co_drain_end)) {
218 return;
219 }
220
221 data = g_new(BdrvCoDrainData, 1);
222 *data = (BdrvCoDrainData) {
223 .bs = bs,
224 .done = false,
225 .begin = begin
226 };
227
228 /* Make sure the driver callback completes during the polling phase for
229 * drain_begin. */
230 bdrv_inc_in_flight(bs);
231 data->co = qemu_coroutine_create(bdrv_drain_invoke_entry, data);
232 aio_co_schedule(bdrv_get_aio_context(bs), data->co);
233
234 if (!begin) {
235 BDRV_POLL_WHILE(bs, !data->done);
236 g_free(data);
237 }
238 }
239
240 /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */
bdrv_drain_poll(BlockDriverState * bs,bool recursive,BdrvChild * ignore_parent,bool ignore_bds_parents)241 bool bdrv_drain_poll(BlockDriverState *bs, bool recursive,
242 BdrvChild *ignore_parent, bool ignore_bds_parents)
243 {
244 BdrvChild *child, *next;
245
246 if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) {
247 return true;
248 }
249
250 if (atomic_read(&bs->in_flight)) {
251 return true;
252 }
253
254 if (recursive) {
255 assert(!ignore_bds_parents);
256 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
257 if (bdrv_drain_poll(child->bs, recursive, child, false)) {
258 return true;
259 }
260 }
261 }
262
263 return false;
264 }
265
bdrv_drain_poll_top_level(BlockDriverState * bs,bool recursive,BdrvChild * ignore_parent)266 static bool bdrv_drain_poll_top_level(BlockDriverState *bs, bool recursive,
267 BdrvChild *ignore_parent)
268 {
269 return bdrv_drain_poll(bs, recursive, ignore_parent, false);
270 }
271
272 static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
273 BdrvChild *parent, bool ignore_bds_parents,
274 bool poll);
275 static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
276 BdrvChild *parent, bool ignore_bds_parents);
277
bdrv_co_drain_bh_cb(void * opaque)278 static void bdrv_co_drain_bh_cb(void *opaque)
279 {
280 BdrvCoDrainData *data = opaque;
281 Coroutine *co = data->co;
282 BlockDriverState *bs = data->bs;
283
284 if (bs) {
285 AioContext *ctx = bdrv_get_aio_context(bs);
286 AioContext *co_ctx = qemu_coroutine_get_aio_context(co);
287
288 /*
289 * When the coroutine yielded, the lock for its home context was
290 * released, so we need to re-acquire it here. If it explicitly
291 * acquired a different context, the lock is still held and we don't
292 * want to lock it a second time (or AIO_WAIT_WHILE() would hang).
293 */
294 if (ctx == co_ctx) {
295 aio_context_acquire(ctx);
296 }
297 bdrv_dec_in_flight(bs);
298 if (data->begin) {
299 bdrv_do_drained_begin(bs, data->recursive, data->parent,
300 data->ignore_bds_parents, data->poll);
301 } else {
302 bdrv_do_drained_end(bs, data->recursive, data->parent,
303 data->ignore_bds_parents);
304 }
305 if (ctx == co_ctx) {
306 aio_context_release(ctx);
307 }
308 } else {
309 assert(data->begin);
310 bdrv_drain_all_begin();
311 }
312
313 data->done = true;
314 aio_co_wake(co);
315 }
316
bdrv_co_yield_to_drain(BlockDriverState * bs,bool begin,bool recursive,BdrvChild * parent,bool ignore_bds_parents,bool poll)317 static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
318 bool begin, bool recursive,
319 BdrvChild *parent,
320 bool ignore_bds_parents,
321 bool poll)
322 {
323 BdrvCoDrainData data;
324
325 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
326 * other coroutines run if they were queued by aio_co_enter(). */
327
328 assert(qemu_in_coroutine());
329 data = (BdrvCoDrainData) {
330 .co = qemu_coroutine_self(),
331 .bs = bs,
332 .done = false,
333 .begin = begin,
334 .recursive = recursive,
335 .parent = parent,
336 .ignore_bds_parents = ignore_bds_parents,
337 .poll = poll,
338 };
339 if (bs) {
340 bdrv_inc_in_flight(bs);
341 }
342 aio_bh_schedule_oneshot(bdrv_get_aio_context(bs),
343 bdrv_co_drain_bh_cb, &data);
344
345 qemu_coroutine_yield();
346 /* If we are resumed from some other event (such as an aio completion or a
347 * timer callback), it is a bug in the caller that should be fixed. */
348 assert(data.done);
349 }
350
bdrv_do_drained_begin_quiesce(BlockDriverState * bs,BdrvChild * parent,bool ignore_bds_parents)351 void bdrv_do_drained_begin_quiesce(BlockDriverState *bs,
352 BdrvChild *parent, bool ignore_bds_parents)
353 {
354 assert(!qemu_in_coroutine());
355
356 /* Stop things in parent-to-child order */
357 if (atomic_fetch_inc(&bs->quiesce_counter) == 0) {
358 aio_disable_external(bdrv_get_aio_context(bs));
359 }
360
361 bdrv_parent_drained_begin(bs, parent, ignore_bds_parents);
362 bdrv_drain_invoke(bs, true);
363 }
364
bdrv_do_drained_begin(BlockDriverState * bs,bool recursive,BdrvChild * parent,bool ignore_bds_parents,bool poll)365 static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
366 BdrvChild *parent, bool ignore_bds_parents,
367 bool poll)
368 {
369 BdrvChild *child, *next;
370
371 if (qemu_in_coroutine()) {
372 bdrv_co_yield_to_drain(bs, true, recursive, parent, ignore_bds_parents,
373 poll);
374 return;
375 }
376
377 bdrv_do_drained_begin_quiesce(bs, parent, ignore_bds_parents);
378
379 if (recursive) {
380 assert(!ignore_bds_parents);
381 bs->recursive_quiesce_counter++;
382 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
383 bdrv_do_drained_begin(child->bs, true, child, ignore_bds_parents,
384 false);
385 }
386 }
387
388 /*
389 * Wait for drained requests to finish.
390 *
391 * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The
392 * call is needed so things in this AioContext can make progress even
393 * though we don't return to the main AioContext loop - this automatically
394 * includes other nodes in the same AioContext and therefore all child
395 * nodes.
396 */
397 if (poll) {
398 assert(!ignore_bds_parents);
399 BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, recursive, parent));
400 }
401 }
402
bdrv_drained_begin(BlockDriverState * bs)403 void bdrv_drained_begin(BlockDriverState *bs)
404 {
405 bdrv_do_drained_begin(bs, false, NULL, false, true);
406 }
407
bdrv_subtree_drained_begin(BlockDriverState * bs)408 void bdrv_subtree_drained_begin(BlockDriverState *bs)
409 {
410 bdrv_do_drained_begin(bs, true, NULL, false, true);
411 }
412
bdrv_do_drained_end(BlockDriverState * bs,bool recursive,BdrvChild * parent,bool ignore_bds_parents)413 static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
414 BdrvChild *parent, bool ignore_bds_parents)
415 {
416 BdrvChild *child, *next;
417 int old_quiesce_counter;
418
419 if (qemu_in_coroutine()) {
420 bdrv_co_yield_to_drain(bs, false, recursive, parent, ignore_bds_parents,
421 false);
422 return;
423 }
424 assert(bs->quiesce_counter > 0);
425 old_quiesce_counter = atomic_fetch_dec(&bs->quiesce_counter);
426
427 /* Re-enable things in child-to-parent order */
428 bdrv_drain_invoke(bs, false);
429 bdrv_parent_drained_end(bs, parent, ignore_bds_parents);
430 if (old_quiesce_counter == 1) {
431 aio_enable_external(bdrv_get_aio_context(bs));
432 }
433
434 if (recursive) {
435 assert(!ignore_bds_parents);
436 bs->recursive_quiesce_counter--;
437 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
438 bdrv_do_drained_end(child->bs, true, child, ignore_bds_parents);
439 }
440 }
441 }
442
bdrv_drained_end(BlockDriverState * bs)443 void bdrv_drained_end(BlockDriverState *bs)
444 {
445 bdrv_do_drained_end(bs, false, NULL, false);
446 }
447
bdrv_subtree_drained_end(BlockDriverState * bs)448 void bdrv_subtree_drained_end(BlockDriverState *bs)
449 {
450 bdrv_do_drained_end(bs, true, NULL, false);
451 }
452
bdrv_apply_subtree_drain(BdrvChild * child,BlockDriverState * new_parent)453 void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent)
454 {
455 int i;
456
457 for (i = 0; i < new_parent->recursive_quiesce_counter; i++) {
458 bdrv_do_drained_begin(child->bs, true, child, false, true);
459 }
460 }
461
bdrv_unapply_subtree_drain(BdrvChild * child,BlockDriverState * old_parent)462 void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent)
463 {
464 int i;
465
466 for (i = 0; i < old_parent->recursive_quiesce_counter; i++) {
467 bdrv_do_drained_end(child->bs, true, child, false);
468 }
469 }
470
471 /*
472 * Wait for pending requests to complete on a single BlockDriverState subtree,
473 * and suspend block driver's internal I/O until next request arrives.
474 *
475 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
476 * AioContext.
477 */
bdrv_co_drain(BlockDriverState * bs)478 void coroutine_fn bdrv_co_drain(BlockDriverState *bs)
479 {
480 assert(qemu_in_coroutine());
481 bdrv_drained_begin(bs);
482 bdrv_drained_end(bs);
483 }
484
bdrv_drain(BlockDriverState * bs)485 void bdrv_drain(BlockDriverState *bs)
486 {
487 bdrv_drained_begin(bs);
488 bdrv_drained_end(bs);
489 }
490
bdrv_drain_assert_idle(BlockDriverState * bs)491 static void bdrv_drain_assert_idle(BlockDriverState *bs)
492 {
493 BdrvChild *child, *next;
494
495 assert(atomic_read(&bs->in_flight) == 0);
496 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
497 bdrv_drain_assert_idle(child->bs);
498 }
499 }
500
501 unsigned int bdrv_drain_all_count = 0;
502
bdrv_drain_all_poll(void)503 static bool bdrv_drain_all_poll(void)
504 {
505 BlockDriverState *bs = NULL;
506 bool result = false;
507
508 /* bdrv_drain_poll() can't make changes to the graph and we are holding the
509 * main AioContext lock, so iterating bdrv_next_all_states() is safe. */
510 while ((bs = bdrv_next_all_states(bs))) {
511 AioContext *aio_context = bdrv_get_aio_context(bs);
512 aio_context_acquire(aio_context);
513 result |= bdrv_drain_poll(bs, false, NULL, true);
514 aio_context_release(aio_context);
515 }
516
517 return result;
518 }
519
520 /*
521 * Wait for pending requests to complete across all BlockDriverStates
522 *
523 * This function does not flush data to disk, use bdrv_flush_all() for that
524 * after calling this function.
525 *
526 * This pauses all block jobs and disables external clients. It must
527 * be paired with bdrv_drain_all_end().
528 *
529 * NOTE: no new block jobs or BlockDriverStates can be created between
530 * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls.
531 */
bdrv_drain_all_begin(void)532 void bdrv_drain_all_begin(void)
533 {
534 BlockDriverState *bs = NULL;
535
536 if (qemu_in_coroutine()) {
537 bdrv_co_yield_to_drain(NULL, true, false, NULL, true, true);
538 return;
539 }
540
541 /* AIO_WAIT_WHILE() with a NULL context can only be called from the main
542 * loop AioContext, so make sure we're in the main context. */
543 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
544 assert(bdrv_drain_all_count < INT_MAX);
545 bdrv_drain_all_count++;
546
547 /* Quiesce all nodes, without polling in-flight requests yet. The graph
548 * cannot change during this loop. */
549 while ((bs = bdrv_next_all_states(bs))) {
550 AioContext *aio_context = bdrv_get_aio_context(bs);
551
552 aio_context_acquire(aio_context);
553 bdrv_do_drained_begin(bs, false, NULL, true, false);
554 aio_context_release(aio_context);
555 }
556
557 /* Now poll the in-flight requests */
558 AIO_WAIT_WHILE(NULL, bdrv_drain_all_poll());
559
560 while ((bs = bdrv_next_all_states(bs))) {
561 bdrv_drain_assert_idle(bs);
562 }
563 }
564
bdrv_drain_all_end(void)565 void bdrv_drain_all_end(void)
566 {
567 BlockDriverState *bs = NULL;
568
569 while ((bs = bdrv_next_all_states(bs))) {
570 AioContext *aio_context = bdrv_get_aio_context(bs);
571
572 aio_context_acquire(aio_context);
573 bdrv_do_drained_end(bs, false, NULL, true);
574 aio_context_release(aio_context);
575 }
576
577 assert(bdrv_drain_all_count > 0);
578 bdrv_drain_all_count--;
579 }
580
bdrv_drain_all(void)581 void bdrv_drain_all(void)
582 {
583 bdrv_drain_all_begin();
584 bdrv_drain_all_end();
585 }
586
587 /**
588 * Remove an active request from the tracked requests list
589 *
590 * This function should be called when a tracked request is completing.
591 */
tracked_request_end(BdrvTrackedRequest * req)592 static void tracked_request_end(BdrvTrackedRequest *req)
593 {
594 if (req->serialising) {
595 atomic_dec(&req->bs->serialising_in_flight);
596 }
597
598 qemu_co_mutex_lock(&req->bs->reqs_lock);
599 QLIST_REMOVE(req, list);
600 qemu_co_queue_restart_all(&req->wait_queue);
601 qemu_co_mutex_unlock(&req->bs->reqs_lock);
602 }
603
604 /**
605 * Add an active request to the tracked requests list
606 */
tracked_request_begin(BdrvTrackedRequest * req,BlockDriverState * bs,int64_t offset,uint64_t bytes,enum BdrvTrackedRequestType type)607 static void tracked_request_begin(BdrvTrackedRequest *req,
608 BlockDriverState *bs,
609 int64_t offset,
610 uint64_t bytes,
611 enum BdrvTrackedRequestType type)
612 {
613 assert(bytes <= INT64_MAX && offset <= INT64_MAX - bytes);
614
615 *req = (BdrvTrackedRequest){
616 .bs = bs,
617 .offset = offset,
618 .bytes = bytes,
619 .type = type,
620 .co = qemu_coroutine_self(),
621 .serialising = false,
622 .overlap_offset = offset,
623 .overlap_bytes = bytes,
624 };
625
626 qemu_co_queue_init(&req->wait_queue);
627
628 qemu_co_mutex_lock(&bs->reqs_lock);
629 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
630 qemu_co_mutex_unlock(&bs->reqs_lock);
631 }
632
mark_request_serialising(BdrvTrackedRequest * req,uint64_t align)633 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
634 {
635 int64_t overlap_offset = req->offset & ~(align - 1);
636 uint64_t overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
637 - overlap_offset;
638
639 if (!req->serialising) {
640 atomic_inc(&req->bs->serialising_in_flight);
641 req->serialising = true;
642 }
643
644 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
645 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
646 }
647
is_request_serialising_and_aligned(BdrvTrackedRequest * req)648 static bool is_request_serialising_and_aligned(BdrvTrackedRequest *req)
649 {
650 /*
651 * If the request is serialising, overlap_offset and overlap_bytes are set,
652 * so we can check if the request is aligned. Otherwise, don't care and
653 * return false.
654 */
655
656 return req->serialising && (req->offset == req->overlap_offset) &&
657 (req->bytes == req->overlap_bytes);
658 }
659
660 /**
661 * Round a region to cluster boundaries
662 */
bdrv_round_to_clusters(BlockDriverState * bs,int64_t offset,int64_t bytes,int64_t * cluster_offset,int64_t * cluster_bytes)663 void bdrv_round_to_clusters(BlockDriverState *bs,
664 int64_t offset, int64_t bytes,
665 int64_t *cluster_offset,
666 int64_t *cluster_bytes)
667 {
668 BlockDriverInfo bdi;
669
670 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
671 *cluster_offset = offset;
672 *cluster_bytes = bytes;
673 } else {
674 int64_t c = bdi.cluster_size;
675 *cluster_offset = QEMU_ALIGN_DOWN(offset, c);
676 *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c);
677 }
678 }
679
bdrv_get_cluster_size(BlockDriverState * bs)680 static int bdrv_get_cluster_size(BlockDriverState *bs)
681 {
682 BlockDriverInfo bdi;
683 int ret;
684
685 ret = bdrv_get_info(bs, &bdi);
686 if (ret < 0 || bdi.cluster_size == 0) {
687 return bs->bl.request_alignment;
688 } else {
689 return bdi.cluster_size;
690 }
691 }
692
tracked_request_overlaps(BdrvTrackedRequest * req,int64_t offset,uint64_t bytes)693 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
694 int64_t offset, uint64_t bytes)
695 {
696 /* aaaa bbbb */
697 if (offset >= req->overlap_offset + req->overlap_bytes) {
698 return false;
699 }
700 /* bbbb aaaa */
701 if (req->overlap_offset >= offset + bytes) {
702 return false;
703 }
704 return true;
705 }
706
bdrv_inc_in_flight(BlockDriverState * bs)707 void bdrv_inc_in_flight(BlockDriverState *bs)
708 {
709 atomic_inc(&bs->in_flight);
710 }
711
bdrv_wakeup(BlockDriverState * bs)712 void bdrv_wakeup(BlockDriverState *bs)
713 {
714 aio_wait_kick();
715 }
716
bdrv_dec_in_flight(BlockDriverState * bs)717 void bdrv_dec_in_flight(BlockDriverState *bs)
718 {
719 atomic_dec(&bs->in_flight);
720 bdrv_wakeup(bs);
721 }
722
wait_serialising_requests(BdrvTrackedRequest * self)723 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
724 {
725 BlockDriverState *bs = self->bs;
726 BdrvTrackedRequest *req;
727 bool retry;
728 bool waited = false;
729
730 if (!atomic_read(&bs->serialising_in_flight)) {
731 return false;
732 }
733
734 do {
735 retry = false;
736 qemu_co_mutex_lock(&bs->reqs_lock);
737 QLIST_FOREACH(req, &bs->tracked_requests, list) {
738 if (req == self || (!req->serialising && !self->serialising)) {
739 continue;
740 }
741 if (tracked_request_overlaps(req, self->overlap_offset,
742 self->overlap_bytes))
743 {
744 /* Hitting this means there was a reentrant request, for
745 * example, a block driver issuing nested requests. This must
746 * never happen since it means deadlock.
747 */
748 assert(qemu_coroutine_self() != req->co);
749
750 /* If the request is already (indirectly) waiting for us, or
751 * will wait for us as soon as it wakes up, then just go on
752 * (instead of producing a deadlock in the former case). */
753 if (!req->waiting_for) {
754 self->waiting_for = req;
755 qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock);
756 self->waiting_for = NULL;
757 retry = true;
758 waited = true;
759 break;
760 }
761 }
762 }
763 qemu_co_mutex_unlock(&bs->reqs_lock);
764 } while (retry);
765
766 return waited;
767 }
768
bdrv_check_byte_request(BlockDriverState * bs,int64_t offset,size_t size)769 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
770 size_t size)
771 {
772 if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) {
773 return -EIO;
774 }
775
776 if (!bdrv_is_inserted(bs)) {
777 return -ENOMEDIUM;
778 }
779
780 if (offset < 0) {
781 return -EIO;
782 }
783
784 return 0;
785 }
786
787 typedef struct RwCo {
788 BdrvChild *child;
789 int64_t offset;
790 QEMUIOVector *qiov;
791 bool is_write;
792 int ret;
793 BdrvRequestFlags flags;
794 } RwCo;
795
bdrv_rw_co_entry(void * opaque)796 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
797 {
798 RwCo *rwco = opaque;
799
800 if (!rwco->is_write) {
801 rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset,
802 rwco->qiov->size, rwco->qiov,
803 rwco->flags);
804 } else {
805 rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset,
806 rwco->qiov->size, rwco->qiov,
807 rwco->flags);
808 }
809 }
810
811 /*
812 * Process a vectored synchronous request using coroutines
813 */
bdrv_prwv_co(BdrvChild * child,int64_t offset,QEMUIOVector * qiov,bool is_write,BdrvRequestFlags flags)814 static int bdrv_prwv_co(BdrvChild *child, int64_t offset,
815 QEMUIOVector *qiov, bool is_write,
816 BdrvRequestFlags flags)
817 {
818 Coroutine *co;
819 RwCo rwco = {
820 .child = child,
821 .offset = offset,
822 .qiov = qiov,
823 .is_write = is_write,
824 .ret = NOT_DONE,
825 .flags = flags,
826 };
827
828 if (qemu_in_coroutine()) {
829 /* Fast-path if already in coroutine context */
830 bdrv_rw_co_entry(&rwco);
831 } else {
832 co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco);
833 bdrv_coroutine_enter(child->bs, co);
834 BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE);
835 }
836 return rwco.ret;
837 }
838
839 /*
840 * Process a synchronous request using coroutines
841 */
bdrv_rw_co(BdrvChild * child,int64_t sector_num,uint8_t * buf,int nb_sectors,bool is_write,BdrvRequestFlags flags)842 static int bdrv_rw_co(BdrvChild *child, int64_t sector_num, uint8_t *buf,
843 int nb_sectors, bool is_write, BdrvRequestFlags flags)
844 {
845 QEMUIOVector qiov;
846 struct iovec iov = {
847 .iov_base = (void *)buf,
848 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
849 };
850
851 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
852 return -EINVAL;
853 }
854
855 qemu_iovec_init_external(&qiov, &iov, 1);
856 return bdrv_prwv_co(child, sector_num << BDRV_SECTOR_BITS,
857 &qiov, is_write, flags);
858 }
859
860 /* return < 0 if error. See bdrv_write() for the return codes */
bdrv_read(BdrvChild * child,int64_t sector_num,uint8_t * buf,int nb_sectors)861 int bdrv_read(BdrvChild *child, int64_t sector_num,
862 uint8_t *buf, int nb_sectors)
863 {
864 return bdrv_rw_co(child, sector_num, buf, nb_sectors, false, 0);
865 }
866
867 /* Return < 0 if error. Important errors are:
868 -EIO generic I/O error (may happen for all errors)
869 -ENOMEDIUM No media inserted.
870 -EINVAL Invalid sector number or nb_sectors
871 -EACCES Trying to write a read-only device
872 */
bdrv_write(BdrvChild * child,int64_t sector_num,const uint8_t * buf,int nb_sectors)873 int bdrv_write(BdrvChild *child, int64_t sector_num,
874 const uint8_t *buf, int nb_sectors)
875 {
876 return bdrv_rw_co(child, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
877 }
878
bdrv_pwrite_zeroes(BdrvChild * child,int64_t offset,int bytes,BdrvRequestFlags flags)879 int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset,
880 int bytes, BdrvRequestFlags flags)
881 {
882 QEMUIOVector qiov;
883 struct iovec iov = {
884 .iov_base = NULL,
885 .iov_len = bytes,
886 };
887
888 qemu_iovec_init_external(&qiov, &iov, 1);
889 return bdrv_prwv_co(child, offset, &qiov, true,
890 BDRV_REQ_ZERO_WRITE | flags);
891 }
892
893 /*
894 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
895 * The operation is sped up by checking the block status and only writing
896 * zeroes to the device if they currently do not return zeroes. Optional
897 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
898 * BDRV_REQ_FUA).
899 *
900 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
901 */
bdrv_make_zero(BdrvChild * child,BdrvRequestFlags flags)902 int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags)
903 {
904 int ret;
905 int64_t target_size, bytes, offset = 0;
906 BlockDriverState *bs = child->bs;
907
908 target_size = bdrv_getlength(bs);
909 if (target_size < 0) {
910 return target_size;
911 }
912
913 for (;;) {
914 bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES);
915 if (bytes <= 0) {
916 return 0;
917 }
918 ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL);
919 if (ret < 0) {
920 error_report("error getting block status at offset %" PRId64 ": %s",
921 offset, strerror(-ret));
922 return ret;
923 }
924 if (ret & BDRV_BLOCK_ZERO) {
925 offset += bytes;
926 continue;
927 }
928 ret = bdrv_pwrite_zeroes(child, offset, bytes, flags);
929 if (ret < 0) {
930 error_report("error writing zeroes at offset %" PRId64 ": %s",
931 offset, strerror(-ret));
932 return ret;
933 }
934 offset += bytes;
935 }
936 }
937
bdrv_preadv(BdrvChild * child,int64_t offset,QEMUIOVector * qiov)938 int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
939 {
940 int ret;
941
942 ret = bdrv_prwv_co(child, offset, qiov, false, 0);
943 if (ret < 0) {
944 return ret;
945 }
946
947 return qiov->size;
948 }
949
bdrv_pread(BdrvChild * child,int64_t offset,void * buf,int bytes)950 int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes)
951 {
952 QEMUIOVector qiov;
953 struct iovec iov = {
954 .iov_base = (void *)buf,
955 .iov_len = bytes,
956 };
957
958 if (bytes < 0) {
959 return -EINVAL;
960 }
961
962 qemu_iovec_init_external(&qiov, &iov, 1);
963 return bdrv_preadv(child, offset, &qiov);
964 }
965
bdrv_pwritev(BdrvChild * child,int64_t offset,QEMUIOVector * qiov)966 int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
967 {
968 int ret;
969
970 ret = bdrv_prwv_co(child, offset, qiov, true, 0);
971 if (ret < 0) {
972 return ret;
973 }
974
975 return qiov->size;
976 }
977
bdrv_pwrite(BdrvChild * child,int64_t offset,const void * buf,int bytes)978 int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes)
979 {
980 QEMUIOVector qiov;
981 struct iovec iov = {
982 .iov_base = (void *) buf,
983 .iov_len = bytes,
984 };
985
986 if (bytes < 0) {
987 return -EINVAL;
988 }
989
990 qemu_iovec_init_external(&qiov, &iov, 1);
991 return bdrv_pwritev(child, offset, &qiov);
992 }
993
994 /*
995 * Writes to the file and ensures that no writes are reordered across this
996 * request (acts as a barrier)
997 *
998 * Returns 0 on success, -errno in error cases.
999 */
bdrv_pwrite_sync(BdrvChild * child,int64_t offset,const void * buf,int count)1000 int bdrv_pwrite_sync(BdrvChild *child, int64_t offset,
1001 const void *buf, int count)
1002 {
1003 int ret;
1004
1005 ret = bdrv_pwrite(child, offset, buf, count);
1006 if (ret < 0) {
1007 return ret;
1008 }
1009
1010 ret = bdrv_flush(child->bs);
1011 if (ret < 0) {
1012 return ret;
1013 }
1014
1015 return 0;
1016 }
1017
1018 typedef struct CoroutineIOCompletion {
1019 Coroutine *coroutine;
1020 int ret;
1021 } CoroutineIOCompletion;
1022
bdrv_co_io_em_complete(void * opaque,int ret)1023 static void bdrv_co_io_em_complete(void *opaque, int ret)
1024 {
1025 CoroutineIOCompletion *co = opaque;
1026
1027 co->ret = ret;
1028 aio_co_wake(co->coroutine);
1029 }
1030
bdrv_driver_preadv(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,int flags)1031 static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs,
1032 uint64_t offset, uint64_t bytes,
1033 QEMUIOVector *qiov, int flags)
1034 {
1035 BlockDriver *drv = bs->drv;
1036 int64_t sector_num;
1037 unsigned int nb_sectors;
1038
1039 assert(!(flags & ~BDRV_REQ_MASK));
1040
1041 if (!drv) {
1042 return -ENOMEDIUM;
1043 }
1044
1045 if (drv->bdrv_co_preadv) {
1046 return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags);
1047 }
1048
1049 if (drv->bdrv_aio_preadv) {
1050 BlockAIOCB *acb;
1051 CoroutineIOCompletion co = {
1052 .coroutine = qemu_coroutine_self(),
1053 };
1054
1055 acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags,
1056 bdrv_co_io_em_complete, &co);
1057 if (acb == NULL) {
1058 return -EIO;
1059 } else {
1060 qemu_coroutine_yield();
1061 return co.ret;
1062 }
1063 }
1064
1065 sector_num = offset >> BDRV_SECTOR_BITS;
1066 nb_sectors = bytes >> BDRV_SECTOR_BITS;
1067
1068 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1069 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1070 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
1071 assert(drv->bdrv_co_readv);
1072
1073 return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1074 }
1075
bdrv_driver_pwritev(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,int flags)1076 static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs,
1077 uint64_t offset, uint64_t bytes,
1078 QEMUIOVector *qiov, int flags)
1079 {
1080 BlockDriver *drv = bs->drv;
1081 int64_t sector_num;
1082 unsigned int nb_sectors;
1083 int ret;
1084
1085 assert(!(flags & ~BDRV_REQ_MASK));
1086
1087 if (!drv) {
1088 return -ENOMEDIUM;
1089 }
1090
1091 if (drv->bdrv_co_pwritev) {
1092 ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov,
1093 flags & bs->supported_write_flags);
1094 flags &= ~bs->supported_write_flags;
1095 goto emulate_flags;
1096 }
1097
1098 if (drv->bdrv_aio_pwritev) {
1099 BlockAIOCB *acb;
1100 CoroutineIOCompletion co = {
1101 .coroutine = qemu_coroutine_self(),
1102 };
1103
1104 acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov,
1105 flags & bs->supported_write_flags,
1106 bdrv_co_io_em_complete, &co);
1107 flags &= ~bs->supported_write_flags;
1108 if (acb == NULL) {
1109 ret = -EIO;
1110 } else {
1111 qemu_coroutine_yield();
1112 ret = co.ret;
1113 }
1114 goto emulate_flags;
1115 }
1116
1117 sector_num = offset >> BDRV_SECTOR_BITS;
1118 nb_sectors = bytes >> BDRV_SECTOR_BITS;
1119
1120 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1121 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1122 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
1123
1124 assert(drv->bdrv_co_writev);
1125 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov,
1126 flags & bs->supported_write_flags);
1127 flags &= ~bs->supported_write_flags;
1128
1129 emulate_flags:
1130 if (ret == 0 && (flags & BDRV_REQ_FUA)) {
1131 ret = bdrv_co_flush(bs);
1132 }
1133
1134 return ret;
1135 }
1136
1137 static int coroutine_fn
bdrv_driver_pwritev_compressed(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov)1138 bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1139 uint64_t bytes, QEMUIOVector *qiov)
1140 {
1141 BlockDriver *drv = bs->drv;
1142
1143 if (!drv) {
1144 return -ENOMEDIUM;
1145 }
1146
1147 if (!drv->bdrv_co_pwritev_compressed) {
1148 return -ENOTSUP;
1149 }
1150
1151 return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov);
1152 }
1153
bdrv_co_do_copy_on_readv(BdrvChild * child,int64_t offset,unsigned int bytes,QEMUIOVector * qiov)1154 static int coroutine_fn bdrv_co_do_copy_on_readv(BdrvChild *child,
1155 int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
1156 {
1157 BlockDriverState *bs = child->bs;
1158
1159 /* Perform I/O through a temporary buffer so that users who scribble over
1160 * their read buffer while the operation is in progress do not end up
1161 * modifying the image file. This is critical for zero-copy guest I/O
1162 * where anything might happen inside guest memory.
1163 */
1164 void *bounce_buffer;
1165
1166 BlockDriver *drv = bs->drv;
1167 struct iovec iov;
1168 QEMUIOVector local_qiov;
1169 int64_t cluster_offset;
1170 int64_t cluster_bytes;
1171 size_t skip_bytes;
1172 int ret;
1173 int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer,
1174 BDRV_REQUEST_MAX_BYTES);
1175 unsigned int progress = 0;
1176
1177 if (!drv) {
1178 return -ENOMEDIUM;
1179 }
1180
1181 /* FIXME We cannot require callers to have write permissions when all they
1182 * are doing is a read request. If we did things right, write permissions
1183 * would be obtained anyway, but internally by the copy-on-read code. As
1184 * long as it is implemented here rather than in a separate filter driver,
1185 * the copy-on-read code doesn't have its own BdrvChild, however, for which
1186 * it could request permissions. Therefore we have to bypass the permission
1187 * system for the moment. */
1188 // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
1189
1190 /* Cover entire cluster so no additional backing file I/O is required when
1191 * allocating cluster in the image file. Note that this value may exceed
1192 * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which
1193 * is one reason we loop rather than doing it all at once.
1194 */
1195 bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes);
1196 skip_bytes = offset - cluster_offset;
1197
1198 trace_bdrv_co_do_copy_on_readv(bs, offset, bytes,
1199 cluster_offset, cluster_bytes);
1200
1201 bounce_buffer = qemu_try_blockalign(bs,
1202 MIN(MIN(max_transfer, cluster_bytes),
1203 MAX_BOUNCE_BUFFER));
1204 if (bounce_buffer == NULL) {
1205 ret = -ENOMEM;
1206 goto err;
1207 }
1208
1209 while (cluster_bytes) {
1210 int64_t pnum;
1211
1212 ret = bdrv_is_allocated(bs, cluster_offset,
1213 MIN(cluster_bytes, max_transfer), &pnum);
1214 if (ret < 0) {
1215 /* Safe to treat errors in querying allocation as if
1216 * unallocated; we'll probably fail again soon on the
1217 * read, but at least that will set a decent errno.
1218 */
1219 pnum = MIN(cluster_bytes, max_transfer);
1220 }
1221
1222 /* Stop at EOF if the image ends in the middle of the cluster */
1223 if (ret == 0 && pnum == 0) {
1224 assert(progress >= bytes);
1225 break;
1226 }
1227
1228 assert(skip_bytes < pnum);
1229
1230 if (ret <= 0) {
1231 /* Must copy-on-read; use the bounce buffer */
1232 iov.iov_base = bounce_buffer;
1233 iov.iov_len = pnum = MIN(pnum, MAX_BOUNCE_BUFFER);
1234 qemu_iovec_init_external(&local_qiov, &iov, 1);
1235
1236 ret = bdrv_driver_preadv(bs, cluster_offset, pnum,
1237 &local_qiov, 0);
1238 if (ret < 0) {
1239 goto err;
1240 }
1241
1242 bdrv_debug_event(bs, BLKDBG_COR_WRITE);
1243 if (drv->bdrv_co_pwrite_zeroes &&
1244 buffer_is_zero(bounce_buffer, pnum)) {
1245 /* FIXME: Should we (perhaps conditionally) be setting
1246 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
1247 * that still correctly reads as zero? */
1248 ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, pnum,
1249 BDRV_REQ_WRITE_UNCHANGED);
1250 } else {
1251 /* This does not change the data on the disk, it is not
1252 * necessary to flush even in cache=writethrough mode.
1253 */
1254 ret = bdrv_driver_pwritev(bs, cluster_offset, pnum,
1255 &local_qiov,
1256 BDRV_REQ_WRITE_UNCHANGED);
1257 }
1258
1259 if (ret < 0) {
1260 /* It might be okay to ignore write errors for guest
1261 * requests. If this is a deliberate copy-on-read
1262 * then we don't want to ignore the error. Simply
1263 * report it in all cases.
1264 */
1265 goto err;
1266 }
1267
1268 qemu_iovec_from_buf(qiov, progress, bounce_buffer + skip_bytes,
1269 pnum - skip_bytes);
1270 } else {
1271 /* Read directly into the destination */
1272 qemu_iovec_init(&local_qiov, qiov->niov);
1273 qemu_iovec_concat(&local_qiov, qiov, progress, pnum - skip_bytes);
1274 ret = bdrv_driver_preadv(bs, offset + progress, local_qiov.size,
1275 &local_qiov, 0);
1276 qemu_iovec_destroy(&local_qiov);
1277 if (ret < 0) {
1278 goto err;
1279 }
1280 }
1281
1282 cluster_offset += pnum;
1283 cluster_bytes -= pnum;
1284 progress += pnum - skip_bytes;
1285 skip_bytes = 0;
1286 }
1287 ret = 0;
1288
1289 err:
1290 qemu_vfree(bounce_buffer);
1291 return ret;
1292 }
1293
1294 /*
1295 * Forwards an already correctly aligned request to the BlockDriver. This
1296 * handles copy on read, zeroing after EOF, and fragmentation of large
1297 * reads; any other features must be implemented by the caller.
1298 */
bdrv_aligned_preadv(BdrvChild * child,BdrvTrackedRequest * req,int64_t offset,unsigned int bytes,int64_t align,QEMUIOVector * qiov,int flags)1299 static int coroutine_fn bdrv_aligned_preadv(BdrvChild *child,
1300 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1301 int64_t align, QEMUIOVector *qiov, int flags)
1302 {
1303 BlockDriverState *bs = child->bs;
1304 int64_t total_bytes, max_bytes;
1305 int ret = 0;
1306 uint64_t bytes_remaining = bytes;
1307 int max_transfer;
1308
1309 assert(is_power_of_2(align));
1310 assert((offset & (align - 1)) == 0);
1311 assert((bytes & (align - 1)) == 0);
1312 assert(!qiov || bytes == qiov->size);
1313 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
1314 max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
1315 align);
1316
1317 /* TODO: We would need a per-BDS .supported_read_flags and
1318 * potential fallback support, if we ever implement any read flags
1319 * to pass through to drivers. For now, there aren't any
1320 * passthrough flags. */
1321 assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ)));
1322
1323 /* Handle Copy on Read and associated serialisation */
1324 if (flags & BDRV_REQ_COPY_ON_READ) {
1325 /* If we touch the same cluster it counts as an overlap. This
1326 * guarantees that allocating writes will be serialized and not race
1327 * with each other for the same cluster. For example, in copy-on-read
1328 * it ensures that the CoR read and write operations are atomic and
1329 * guest writes cannot interleave between them. */
1330 mark_request_serialising(req, bdrv_get_cluster_size(bs));
1331 }
1332
1333 /* BDRV_REQ_SERIALISING is only for write operation */
1334 assert(!(flags & BDRV_REQ_SERIALISING));
1335
1336 if (!(flags & BDRV_REQ_NO_SERIALISING)) {
1337 wait_serialising_requests(req);
1338 }
1339
1340 if (flags & BDRV_REQ_COPY_ON_READ) {
1341 int64_t pnum;
1342
1343 ret = bdrv_is_allocated(bs, offset, bytes, &pnum);
1344 if (ret < 0) {
1345 goto out;
1346 }
1347
1348 if (!ret || pnum != bytes) {
1349 ret = bdrv_co_do_copy_on_readv(child, offset, bytes, qiov);
1350 goto out;
1351 }
1352 }
1353
1354 /* Forward the request to the BlockDriver, possibly fragmenting it */
1355 total_bytes = bdrv_getlength(bs);
1356 if (total_bytes < 0) {
1357 ret = total_bytes;
1358 goto out;
1359 }
1360
1361 max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align);
1362 if (bytes <= max_bytes && bytes <= max_transfer) {
1363 ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0);
1364 goto out;
1365 }
1366
1367 while (bytes_remaining) {
1368 int num;
1369
1370 if (max_bytes) {
1371 QEMUIOVector local_qiov;
1372
1373 num = MIN(bytes_remaining, MIN(max_bytes, max_transfer));
1374 assert(num);
1375 qemu_iovec_init(&local_qiov, qiov->niov);
1376 qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
1377
1378 ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining,
1379 num, &local_qiov, 0);
1380 max_bytes -= num;
1381 qemu_iovec_destroy(&local_qiov);
1382 } else {
1383 num = bytes_remaining;
1384 ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0,
1385 bytes_remaining);
1386 }
1387 if (ret < 0) {
1388 goto out;
1389 }
1390 bytes_remaining -= num;
1391 }
1392
1393 out:
1394 return ret < 0 ? ret : 0;
1395 }
1396
1397 /*
1398 * Handle a read request in coroutine context
1399 */
bdrv_co_preadv(BdrvChild * child,int64_t offset,unsigned int bytes,QEMUIOVector * qiov,BdrvRequestFlags flags)1400 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
1401 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1402 BdrvRequestFlags flags)
1403 {
1404 BlockDriverState *bs = child->bs;
1405 BlockDriver *drv = bs->drv;
1406 BdrvTrackedRequest req;
1407
1408 uint64_t align = bs->bl.request_alignment;
1409 uint8_t *head_buf = NULL;
1410 uint8_t *tail_buf = NULL;
1411 QEMUIOVector local_qiov;
1412 bool use_local_qiov = false;
1413 int ret;
1414
1415 trace_bdrv_co_preadv(child->bs, offset, bytes, flags);
1416
1417 if (!drv) {
1418 return -ENOMEDIUM;
1419 }
1420
1421 ret = bdrv_check_byte_request(bs, offset, bytes);
1422 if (ret < 0) {
1423 return ret;
1424 }
1425
1426 bdrv_inc_in_flight(bs);
1427
1428 /* Don't do copy-on-read if we read data before write operation */
1429 if (atomic_read(&bs->copy_on_read) && !(flags & BDRV_REQ_NO_SERIALISING)) {
1430 flags |= BDRV_REQ_COPY_ON_READ;
1431 }
1432
1433 /* Align read if necessary by padding qiov */
1434 if (offset & (align - 1)) {
1435 head_buf = qemu_blockalign(bs, align);
1436 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1437 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1438 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1439 use_local_qiov = true;
1440
1441 bytes += offset & (align - 1);
1442 offset = offset & ~(align - 1);
1443 }
1444
1445 if ((offset + bytes) & (align - 1)) {
1446 if (!use_local_qiov) {
1447 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1448 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1449 use_local_qiov = true;
1450 }
1451 tail_buf = qemu_blockalign(bs, align);
1452 qemu_iovec_add(&local_qiov, tail_buf,
1453 align - ((offset + bytes) & (align - 1)));
1454
1455 bytes = ROUND_UP(bytes, align);
1456 }
1457
1458 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ);
1459 ret = bdrv_aligned_preadv(child, &req, offset, bytes, align,
1460 use_local_qiov ? &local_qiov : qiov,
1461 flags);
1462 tracked_request_end(&req);
1463 bdrv_dec_in_flight(bs);
1464
1465 if (use_local_qiov) {
1466 qemu_iovec_destroy(&local_qiov);
1467 qemu_vfree(head_buf);
1468 qemu_vfree(tail_buf);
1469 }
1470
1471 return ret;
1472 }
1473
bdrv_co_do_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int bytes,BdrvRequestFlags flags)1474 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
1475 int64_t offset, int bytes, BdrvRequestFlags flags)
1476 {
1477 BlockDriver *drv = bs->drv;
1478 QEMUIOVector qiov;
1479 struct iovec iov = {0};
1480 int ret = 0;
1481 bool need_flush = false;
1482 int head = 0;
1483 int tail = 0;
1484
1485 int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX);
1486 int alignment = MAX(bs->bl.pwrite_zeroes_alignment,
1487 bs->bl.request_alignment);
1488 int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER);
1489
1490 if (!drv) {
1491 return -ENOMEDIUM;
1492 }
1493
1494 assert(alignment % bs->bl.request_alignment == 0);
1495 head = offset % alignment;
1496 tail = (offset + bytes) % alignment;
1497 max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment);
1498 assert(max_write_zeroes >= bs->bl.request_alignment);
1499
1500 while (bytes > 0 && !ret) {
1501 int num = bytes;
1502
1503 /* Align request. Block drivers can expect the "bulk" of the request
1504 * to be aligned, and that unaligned requests do not cross cluster
1505 * boundaries.
1506 */
1507 if (head) {
1508 /* Make a small request up to the first aligned sector. For
1509 * convenience, limit this request to max_transfer even if
1510 * we don't need to fall back to writes. */
1511 num = MIN(MIN(bytes, max_transfer), alignment - head);
1512 head = (head + num) % alignment;
1513 assert(num < max_write_zeroes);
1514 } else if (tail && num > alignment) {
1515 /* Shorten the request to the last aligned sector. */
1516 num -= tail;
1517 }
1518
1519 /* limit request size */
1520 if (num > max_write_zeroes) {
1521 num = max_write_zeroes;
1522 }
1523
1524 ret = -ENOTSUP;
1525 /* First try the efficient write zeroes operation */
1526 if (drv->bdrv_co_pwrite_zeroes) {
1527 ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num,
1528 flags & bs->supported_zero_flags);
1529 if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) &&
1530 !(bs->supported_zero_flags & BDRV_REQ_FUA)) {
1531 need_flush = true;
1532 }
1533 } else {
1534 assert(!bs->supported_zero_flags);
1535 }
1536
1537 if (ret == -ENOTSUP) {
1538 /* Fall back to bounce buffer if write zeroes is unsupported */
1539 BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE;
1540
1541 if ((flags & BDRV_REQ_FUA) &&
1542 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1543 /* No need for bdrv_driver_pwrite() to do a fallback
1544 * flush on each chunk; use just one at the end */
1545 write_flags &= ~BDRV_REQ_FUA;
1546 need_flush = true;
1547 }
1548 num = MIN(num, max_transfer);
1549 iov.iov_len = num;
1550 if (iov.iov_base == NULL) {
1551 iov.iov_base = qemu_try_blockalign(bs, num);
1552 if (iov.iov_base == NULL) {
1553 ret = -ENOMEM;
1554 goto fail;
1555 }
1556 memset(iov.iov_base, 0, num);
1557 }
1558 qemu_iovec_init_external(&qiov, &iov, 1);
1559
1560 ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags);
1561
1562 /* Keep bounce buffer around if it is big enough for all
1563 * all future requests.
1564 */
1565 if (num < max_transfer) {
1566 qemu_vfree(iov.iov_base);
1567 iov.iov_base = NULL;
1568 }
1569 }
1570
1571 offset += num;
1572 bytes -= num;
1573 }
1574
1575 fail:
1576 if (ret == 0 && need_flush) {
1577 ret = bdrv_co_flush(bs);
1578 }
1579 qemu_vfree(iov.iov_base);
1580 return ret;
1581 }
1582
1583 static inline int coroutine_fn
bdrv_co_write_req_prepare(BdrvChild * child,int64_t offset,uint64_t bytes,BdrvTrackedRequest * req,int flags)1584 bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes,
1585 BdrvTrackedRequest *req, int flags)
1586 {
1587 BlockDriverState *bs = child->bs;
1588 bool waited;
1589 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1590
1591 if (bs->read_only) {
1592 return -EPERM;
1593 }
1594
1595 /* BDRV_REQ_NO_SERIALISING is only for read operation */
1596 assert(!(flags & BDRV_REQ_NO_SERIALISING));
1597 assert(!(bs->open_flags & BDRV_O_INACTIVE));
1598 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
1599 assert(!(flags & ~BDRV_REQ_MASK));
1600
1601 if (flags & BDRV_REQ_SERIALISING) {
1602 mark_request_serialising(req, bdrv_get_cluster_size(bs));
1603 }
1604
1605 waited = wait_serialising_requests(req);
1606
1607 assert(!waited || !req->serialising ||
1608 is_request_serialising_and_aligned(req));
1609 assert(req->overlap_offset <= offset);
1610 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
1611 assert(end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE);
1612
1613 switch (req->type) {
1614 case BDRV_TRACKED_WRITE:
1615 case BDRV_TRACKED_DISCARD:
1616 if (flags & BDRV_REQ_WRITE_UNCHANGED) {
1617 assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
1618 } else {
1619 assert(child->perm & BLK_PERM_WRITE);
1620 }
1621 return notifier_with_return_list_notify(&bs->before_write_notifiers,
1622 req);
1623 case BDRV_TRACKED_TRUNCATE:
1624 assert(child->perm & BLK_PERM_RESIZE);
1625 return 0;
1626 default:
1627 abort();
1628 }
1629 }
1630
1631 static inline void coroutine_fn
bdrv_co_write_req_finish(BdrvChild * child,int64_t offset,uint64_t bytes,BdrvTrackedRequest * req,int ret)1632 bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, uint64_t bytes,
1633 BdrvTrackedRequest *req, int ret)
1634 {
1635 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1636 BlockDriverState *bs = child->bs;
1637
1638 atomic_inc(&bs->write_gen);
1639
1640 /*
1641 * Discard cannot extend the image, but in error handling cases, such as
1642 * when reverting a qcow2 cluster allocation, the discarded range can pass
1643 * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD
1644 * here. Instead, just skip it, since semantically a discard request
1645 * beyond EOF cannot expand the image anyway.
1646 */
1647 if (ret == 0 &&
1648 (req->type == BDRV_TRACKED_TRUNCATE ||
1649 end_sector > bs->total_sectors) &&
1650 req->type != BDRV_TRACKED_DISCARD) {
1651 bs->total_sectors = end_sector;
1652 bdrv_parent_cb_resize(bs);
1653 bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS);
1654 }
1655 if (req->bytes) {
1656 switch (req->type) {
1657 case BDRV_TRACKED_WRITE:
1658 stat64_max(&bs->wr_highest_offset, offset + bytes);
1659 /* fall through, to set dirty bits */
1660 case BDRV_TRACKED_DISCARD:
1661 bdrv_set_dirty(bs, offset, bytes);
1662 break;
1663 default:
1664 break;
1665 }
1666 }
1667 }
1668
1669 /*
1670 * Forwards an already correctly aligned write request to the BlockDriver,
1671 * after possibly fragmenting it.
1672 */
bdrv_aligned_pwritev(BdrvChild * child,BdrvTrackedRequest * req,int64_t offset,unsigned int bytes,int64_t align,QEMUIOVector * qiov,int flags)1673 static int coroutine_fn bdrv_aligned_pwritev(BdrvChild *child,
1674 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1675 int64_t align, QEMUIOVector *qiov, int flags)
1676 {
1677 BlockDriverState *bs = child->bs;
1678 BlockDriver *drv = bs->drv;
1679 int ret;
1680
1681 uint64_t bytes_remaining = bytes;
1682 int max_transfer;
1683
1684 if (!drv) {
1685 return -ENOMEDIUM;
1686 }
1687
1688 if (bdrv_has_readonly_bitmaps(bs)) {
1689 return -EPERM;
1690 }
1691
1692 assert(is_power_of_2(align));
1693 assert((offset & (align - 1)) == 0);
1694 assert((bytes & (align - 1)) == 0);
1695 assert(!qiov || bytes == qiov->size);
1696 max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
1697 align);
1698
1699 ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags);
1700
1701 if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
1702 !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes &&
1703 qemu_iovec_is_zero(qiov)) {
1704 flags |= BDRV_REQ_ZERO_WRITE;
1705 if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
1706 flags |= BDRV_REQ_MAY_UNMAP;
1707 }
1708 }
1709
1710 if (ret < 0) {
1711 /* Do nothing, write notifier decided to fail this request */
1712 } else if (flags & BDRV_REQ_ZERO_WRITE) {
1713 bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO);
1714 ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags);
1715 } else if (flags & BDRV_REQ_WRITE_COMPRESSED) {
1716 ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, qiov);
1717 } else if (bytes <= max_transfer) {
1718 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1719 ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags);
1720 } else {
1721 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1722 while (bytes_remaining) {
1723 int num = MIN(bytes_remaining, max_transfer);
1724 QEMUIOVector local_qiov;
1725 int local_flags = flags;
1726
1727 assert(num);
1728 if (num < bytes_remaining && (flags & BDRV_REQ_FUA) &&
1729 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1730 /* If FUA is going to be emulated by flush, we only
1731 * need to flush on the last iteration */
1732 local_flags &= ~BDRV_REQ_FUA;
1733 }
1734 qemu_iovec_init(&local_qiov, qiov->niov);
1735 qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
1736
1737 ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining,
1738 num, &local_qiov, local_flags);
1739 qemu_iovec_destroy(&local_qiov);
1740 if (ret < 0) {
1741 break;
1742 }
1743 bytes_remaining -= num;
1744 }
1745 }
1746 bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
1747
1748 if (ret >= 0) {
1749 ret = 0;
1750 }
1751 bdrv_co_write_req_finish(child, offset, bytes, req, ret);
1752
1753 return ret;
1754 }
1755
bdrv_co_do_zero_pwritev(BdrvChild * child,int64_t offset,unsigned int bytes,BdrvRequestFlags flags,BdrvTrackedRequest * req)1756 static int coroutine_fn bdrv_co_do_zero_pwritev(BdrvChild *child,
1757 int64_t offset,
1758 unsigned int bytes,
1759 BdrvRequestFlags flags,
1760 BdrvTrackedRequest *req)
1761 {
1762 BlockDriverState *bs = child->bs;
1763 uint8_t *buf = NULL;
1764 QEMUIOVector local_qiov;
1765 struct iovec iov;
1766 uint64_t align = bs->bl.request_alignment;
1767 unsigned int head_padding_bytes, tail_padding_bytes;
1768 int ret = 0;
1769
1770 head_padding_bytes = offset & (align - 1);
1771 tail_padding_bytes = (align - (offset + bytes)) & (align - 1);
1772
1773
1774 assert(flags & BDRV_REQ_ZERO_WRITE);
1775 if (head_padding_bytes || tail_padding_bytes) {
1776 buf = qemu_blockalign(bs, align);
1777 iov = (struct iovec) {
1778 .iov_base = buf,
1779 .iov_len = align,
1780 };
1781 qemu_iovec_init_external(&local_qiov, &iov, 1);
1782 }
1783 if (head_padding_bytes) {
1784 uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes);
1785
1786 /* RMW the unaligned part before head. */
1787 mark_request_serialising(req, align);
1788 wait_serialising_requests(req);
1789 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1790 ret = bdrv_aligned_preadv(child, req, offset & ~(align - 1), align,
1791 align, &local_qiov, 0);
1792 if (ret < 0) {
1793 goto fail;
1794 }
1795 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1796
1797 memset(buf + head_padding_bytes, 0, zero_bytes);
1798 ret = bdrv_aligned_pwritev(child, req, offset & ~(align - 1), align,
1799 align, &local_qiov,
1800 flags & ~BDRV_REQ_ZERO_WRITE);
1801 if (ret < 0) {
1802 goto fail;
1803 }
1804 offset += zero_bytes;
1805 bytes -= zero_bytes;
1806 }
1807
1808 assert(!bytes || (offset & (align - 1)) == 0);
1809 if (bytes >= align) {
1810 /* Write the aligned part in the middle. */
1811 uint64_t aligned_bytes = bytes & ~(align - 1);
1812 ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align,
1813 NULL, flags);
1814 if (ret < 0) {
1815 goto fail;
1816 }
1817 bytes -= aligned_bytes;
1818 offset += aligned_bytes;
1819 }
1820
1821 assert(!bytes || (offset & (align - 1)) == 0);
1822 if (bytes) {
1823 assert(align == tail_padding_bytes + bytes);
1824 /* RMW the unaligned part after tail. */
1825 mark_request_serialising(req, align);
1826 wait_serialising_requests(req);
1827 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1828 ret = bdrv_aligned_preadv(child, req, offset, align,
1829 align, &local_qiov, 0);
1830 if (ret < 0) {
1831 goto fail;
1832 }
1833 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1834
1835 memset(buf, 0, bytes);
1836 ret = bdrv_aligned_pwritev(child, req, offset, align, align,
1837 &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE);
1838 }
1839 fail:
1840 qemu_vfree(buf);
1841 return ret;
1842
1843 }
1844
1845 /*
1846 * Handle a write request in coroutine context
1847 */
bdrv_co_pwritev(BdrvChild * child,int64_t offset,unsigned int bytes,QEMUIOVector * qiov,BdrvRequestFlags flags)1848 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
1849 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1850 BdrvRequestFlags flags)
1851 {
1852 BlockDriverState *bs = child->bs;
1853 BdrvTrackedRequest req;
1854 uint64_t align = bs->bl.request_alignment;
1855 uint8_t *head_buf = NULL;
1856 uint8_t *tail_buf = NULL;
1857 QEMUIOVector local_qiov;
1858 bool use_local_qiov = false;
1859 int ret;
1860
1861 trace_bdrv_co_pwritev(child->bs, offset, bytes, flags);
1862
1863 if (!bs->drv) {
1864 return -ENOMEDIUM;
1865 }
1866
1867 ret = bdrv_check_byte_request(bs, offset, bytes);
1868 if (ret < 0) {
1869 return ret;
1870 }
1871
1872 bdrv_inc_in_flight(bs);
1873 /*
1874 * Align write if necessary by performing a read-modify-write cycle.
1875 * Pad qiov with the read parts and be sure to have a tracked request not
1876 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1877 */
1878 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE);
1879
1880 if (flags & BDRV_REQ_ZERO_WRITE) {
1881 ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req);
1882 goto out;
1883 }
1884
1885 if (offset & (align - 1)) {
1886 QEMUIOVector head_qiov;
1887 struct iovec head_iov;
1888
1889 mark_request_serialising(&req, align);
1890 wait_serialising_requests(&req);
1891
1892 head_buf = qemu_blockalign(bs, align);
1893 head_iov = (struct iovec) {
1894 .iov_base = head_buf,
1895 .iov_len = align,
1896 };
1897 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
1898
1899 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1900 ret = bdrv_aligned_preadv(child, &req, offset & ~(align - 1), align,
1901 align, &head_qiov, 0);
1902 if (ret < 0) {
1903 goto fail;
1904 }
1905 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1906
1907 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1908 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1909 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1910 use_local_qiov = true;
1911
1912 bytes += offset & (align - 1);
1913 offset = offset & ~(align - 1);
1914
1915 /* We have read the tail already if the request is smaller
1916 * than one aligned block.
1917 */
1918 if (bytes < align) {
1919 qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes);
1920 bytes = align;
1921 }
1922 }
1923
1924 if ((offset + bytes) & (align - 1)) {
1925 QEMUIOVector tail_qiov;
1926 struct iovec tail_iov;
1927 size_t tail_bytes;
1928 bool waited;
1929
1930 mark_request_serialising(&req, align);
1931 waited = wait_serialising_requests(&req);
1932 assert(!waited || !use_local_qiov);
1933
1934 tail_buf = qemu_blockalign(bs, align);
1935 tail_iov = (struct iovec) {
1936 .iov_base = tail_buf,
1937 .iov_len = align,
1938 };
1939 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
1940
1941 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1942 ret = bdrv_aligned_preadv(child, &req, (offset + bytes) & ~(align - 1),
1943 align, align, &tail_qiov, 0);
1944 if (ret < 0) {
1945 goto fail;
1946 }
1947 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1948
1949 if (!use_local_qiov) {
1950 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1951 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1952 use_local_qiov = true;
1953 }
1954
1955 tail_bytes = (offset + bytes) & (align - 1);
1956 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
1957
1958 bytes = ROUND_UP(bytes, align);
1959 }
1960
1961 ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align,
1962 use_local_qiov ? &local_qiov : qiov,
1963 flags);
1964
1965 fail:
1966
1967 if (use_local_qiov) {
1968 qemu_iovec_destroy(&local_qiov);
1969 }
1970 qemu_vfree(head_buf);
1971 qemu_vfree(tail_buf);
1972 out:
1973 tracked_request_end(&req);
1974 bdrv_dec_in_flight(bs);
1975 return ret;
1976 }
1977
bdrv_co_pwrite_zeroes(BdrvChild * child,int64_t offset,int bytes,BdrvRequestFlags flags)1978 int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset,
1979 int bytes, BdrvRequestFlags flags)
1980 {
1981 trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags);
1982
1983 if (!(child->bs->open_flags & BDRV_O_UNMAP)) {
1984 flags &= ~BDRV_REQ_MAY_UNMAP;
1985 }
1986
1987 return bdrv_co_pwritev(child, offset, bytes, NULL,
1988 BDRV_REQ_ZERO_WRITE | flags);
1989 }
1990
1991 /*
1992 * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
1993 */
bdrv_flush_all(void)1994 int bdrv_flush_all(void)
1995 {
1996 BdrvNextIterator it;
1997 BlockDriverState *bs = NULL;
1998 int result = 0;
1999
2000 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
2001 AioContext *aio_context = bdrv_get_aio_context(bs);
2002 int ret;
2003
2004 aio_context_acquire(aio_context);
2005 ret = bdrv_flush(bs);
2006 if (ret < 0 && !result) {
2007 result = ret;
2008 }
2009 aio_context_release(aio_context);
2010 }
2011
2012 return result;
2013 }
2014
2015
2016 typedef struct BdrvCoBlockStatusData {
2017 BlockDriverState *bs;
2018 BlockDriverState *base;
2019 bool want_zero;
2020 int64_t offset;
2021 int64_t bytes;
2022 int64_t *pnum;
2023 int64_t *map;
2024 BlockDriverState **file;
2025 int ret;
2026 bool done;
2027 } BdrvCoBlockStatusData;
2028
bdrv_co_block_status_from_file(BlockDriverState * bs,bool want_zero,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)2029 int coroutine_fn bdrv_co_block_status_from_file(BlockDriverState *bs,
2030 bool want_zero,
2031 int64_t offset,
2032 int64_t bytes,
2033 int64_t *pnum,
2034 int64_t *map,
2035 BlockDriverState **file)
2036 {
2037 assert(bs->file && bs->file->bs);
2038 *pnum = bytes;
2039 *map = offset;
2040 *file = bs->file->bs;
2041 return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
2042 }
2043
bdrv_co_block_status_from_backing(BlockDriverState * bs,bool want_zero,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)2044 int coroutine_fn bdrv_co_block_status_from_backing(BlockDriverState *bs,
2045 bool want_zero,
2046 int64_t offset,
2047 int64_t bytes,
2048 int64_t *pnum,
2049 int64_t *map,
2050 BlockDriverState **file)
2051 {
2052 assert(bs->backing && bs->backing->bs);
2053 *pnum = bytes;
2054 *map = offset;
2055 *file = bs->backing->bs;
2056 return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
2057 }
2058
2059 /*
2060 * Returns the allocation status of the specified sectors.
2061 * Drivers not implementing the functionality are assumed to not support
2062 * backing files, hence all their sectors are reported as allocated.
2063 *
2064 * If 'want_zero' is true, the caller is querying for mapping
2065 * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and
2066 * _ZERO where possible; otherwise, the result favors larger 'pnum',
2067 * with a focus on accurate BDRV_BLOCK_ALLOCATED.
2068 *
2069 * If 'offset' is beyond the end of the disk image the return value is
2070 * BDRV_BLOCK_EOF and 'pnum' is set to 0.
2071 *
2072 * 'bytes' is the max value 'pnum' should be set to. If bytes goes
2073 * beyond the end of the disk image it will be clamped; if 'pnum' is set to
2074 * the end of the image, then the returned value will include BDRV_BLOCK_EOF.
2075 *
2076 * 'pnum' is set to the number of bytes (including and immediately
2077 * following the specified offset) that are easily known to be in the
2078 * same allocated/unallocated state. Note that a second call starting
2079 * at the original offset plus returned pnum may have the same status.
2080 * The returned value is non-zero on success except at end-of-file.
2081 *
2082 * Returns negative errno on failure. Otherwise, if the
2083 * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are
2084 * set to the host mapping and BDS corresponding to the guest offset.
2085 */
bdrv_co_block_status(BlockDriverState * bs,bool want_zero,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)2086 static int coroutine_fn bdrv_co_block_status(BlockDriverState *bs,
2087 bool want_zero,
2088 int64_t offset, int64_t bytes,
2089 int64_t *pnum, int64_t *map,
2090 BlockDriverState **file)
2091 {
2092 int64_t total_size;
2093 int64_t n; /* bytes */
2094 int ret;
2095 int64_t local_map = 0;
2096 BlockDriverState *local_file = NULL;
2097 int64_t aligned_offset, aligned_bytes;
2098 uint32_t align;
2099
2100 assert(pnum);
2101 *pnum = 0;
2102 total_size = bdrv_getlength(bs);
2103 if (total_size < 0) {
2104 ret = total_size;
2105 goto early_out;
2106 }
2107
2108 if (offset >= total_size) {
2109 ret = BDRV_BLOCK_EOF;
2110 goto early_out;
2111 }
2112 if (!bytes) {
2113 ret = 0;
2114 goto early_out;
2115 }
2116
2117 n = total_size - offset;
2118 if (n < bytes) {
2119 bytes = n;
2120 }
2121
2122 /* Must be non-NULL or bdrv_getlength() would have failed */
2123 assert(bs->drv);
2124 if (!bs->drv->bdrv_co_block_status) {
2125 *pnum = bytes;
2126 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
2127 if (offset + bytes == total_size) {
2128 ret |= BDRV_BLOCK_EOF;
2129 }
2130 if (bs->drv->protocol_name) {
2131 ret |= BDRV_BLOCK_OFFSET_VALID;
2132 local_map = offset;
2133 local_file = bs;
2134 }
2135 goto early_out;
2136 }
2137
2138 bdrv_inc_in_flight(bs);
2139
2140 /* Round out to request_alignment boundaries */
2141 align = bs->bl.request_alignment;
2142 aligned_offset = QEMU_ALIGN_DOWN(offset, align);
2143 aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset;
2144
2145 ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset,
2146 aligned_bytes, pnum, &local_map,
2147 &local_file);
2148 if (ret < 0) {
2149 *pnum = 0;
2150 goto out;
2151 }
2152
2153 /*
2154 * The driver's result must be a non-zero multiple of request_alignment.
2155 * Clamp pnum and adjust map to original request.
2156 */
2157 assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) &&
2158 align > offset - aligned_offset);
2159 *pnum -= offset - aligned_offset;
2160 if (*pnum > bytes) {
2161 *pnum = bytes;
2162 }
2163 if (ret & BDRV_BLOCK_OFFSET_VALID) {
2164 local_map += offset - aligned_offset;
2165 }
2166
2167 if (ret & BDRV_BLOCK_RAW) {
2168 assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file);
2169 ret = bdrv_co_block_status(local_file, want_zero, local_map,
2170 *pnum, pnum, &local_map, &local_file);
2171 goto out;
2172 }
2173
2174 if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
2175 ret |= BDRV_BLOCK_ALLOCATED;
2176 } else if (want_zero) {
2177 if (bdrv_unallocated_blocks_are_zero(bs)) {
2178 ret |= BDRV_BLOCK_ZERO;
2179 } else if (bs->backing) {
2180 BlockDriverState *bs2 = bs->backing->bs;
2181 int64_t size2 = bdrv_getlength(bs2);
2182
2183 if (size2 >= 0 && offset >= size2) {
2184 ret |= BDRV_BLOCK_ZERO;
2185 }
2186 }
2187 }
2188
2189 if (want_zero && local_file && local_file != bs &&
2190 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
2191 (ret & BDRV_BLOCK_OFFSET_VALID)) {
2192 int64_t file_pnum;
2193 int ret2;
2194
2195 ret2 = bdrv_co_block_status(local_file, want_zero, local_map,
2196 *pnum, &file_pnum, NULL, NULL);
2197 if (ret2 >= 0) {
2198 /* Ignore errors. This is just providing extra information, it
2199 * is useful but not necessary.
2200 */
2201 if (ret2 & BDRV_BLOCK_EOF &&
2202 (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) {
2203 /*
2204 * It is valid for the format block driver to read
2205 * beyond the end of the underlying file's current
2206 * size; such areas read as zero.
2207 */
2208 ret |= BDRV_BLOCK_ZERO;
2209 } else {
2210 /* Limit request to the range reported by the protocol driver */
2211 *pnum = file_pnum;
2212 ret |= (ret2 & BDRV_BLOCK_ZERO);
2213 }
2214 }
2215 }
2216
2217 out:
2218 bdrv_dec_in_flight(bs);
2219 if (ret >= 0 && offset + *pnum == total_size) {
2220 ret |= BDRV_BLOCK_EOF;
2221 }
2222 early_out:
2223 if (file) {
2224 *file = local_file;
2225 }
2226 if (map) {
2227 *map = local_map;
2228 }
2229 return ret;
2230 }
2231
bdrv_co_block_status_above(BlockDriverState * bs,BlockDriverState * base,bool want_zero,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)2232 static int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs,
2233 BlockDriverState *base,
2234 bool want_zero,
2235 int64_t offset,
2236 int64_t bytes,
2237 int64_t *pnum,
2238 int64_t *map,
2239 BlockDriverState **file)
2240 {
2241 BlockDriverState *p;
2242 int ret = 0;
2243 bool first = true;
2244
2245 assert(bs != base);
2246 for (p = bs; p != base; p = backing_bs(p)) {
2247 ret = bdrv_co_block_status(p, want_zero, offset, bytes, pnum, map,
2248 file);
2249 if (ret < 0) {
2250 break;
2251 }
2252 if (ret & BDRV_BLOCK_ZERO && ret & BDRV_BLOCK_EOF && !first) {
2253 /*
2254 * Reading beyond the end of the file continues to read
2255 * zeroes, but we can only widen the result to the
2256 * unallocated length we learned from an earlier
2257 * iteration.
2258 */
2259 *pnum = bytes;
2260 }
2261 if (ret & (BDRV_BLOCK_ZERO | BDRV_BLOCK_DATA)) {
2262 break;
2263 }
2264 /* [offset, pnum] unallocated on this layer, which could be only
2265 * the first part of [offset, bytes]. */
2266 bytes = MIN(bytes, *pnum);
2267 first = false;
2268 }
2269 return ret;
2270 }
2271
2272 /* Coroutine wrapper for bdrv_block_status_above() */
bdrv_block_status_above_co_entry(void * opaque)2273 static void coroutine_fn bdrv_block_status_above_co_entry(void *opaque)
2274 {
2275 BdrvCoBlockStatusData *data = opaque;
2276
2277 data->ret = bdrv_co_block_status_above(data->bs, data->base,
2278 data->want_zero,
2279 data->offset, data->bytes,
2280 data->pnum, data->map, data->file);
2281 data->done = true;
2282 }
2283
2284 /*
2285 * Synchronous wrapper around bdrv_co_block_status_above().
2286 *
2287 * See bdrv_co_block_status_above() for details.
2288 */
bdrv_common_block_status_above(BlockDriverState * bs,BlockDriverState * base,bool want_zero,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)2289 static int bdrv_common_block_status_above(BlockDriverState *bs,
2290 BlockDriverState *base,
2291 bool want_zero, int64_t offset,
2292 int64_t bytes, int64_t *pnum,
2293 int64_t *map,
2294 BlockDriverState **file)
2295 {
2296 Coroutine *co;
2297 BdrvCoBlockStatusData data = {
2298 .bs = bs,
2299 .base = base,
2300 .want_zero = want_zero,
2301 .offset = offset,
2302 .bytes = bytes,
2303 .pnum = pnum,
2304 .map = map,
2305 .file = file,
2306 .done = false,
2307 };
2308
2309 if (qemu_in_coroutine()) {
2310 /* Fast-path if already in coroutine context */
2311 bdrv_block_status_above_co_entry(&data);
2312 } else {
2313 co = qemu_coroutine_create(bdrv_block_status_above_co_entry, &data);
2314 bdrv_coroutine_enter(bs, co);
2315 BDRV_POLL_WHILE(bs, !data.done);
2316 }
2317 return data.ret;
2318 }
2319
bdrv_block_status_above(BlockDriverState * bs,BlockDriverState * base,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)2320 int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base,
2321 int64_t offset, int64_t bytes, int64_t *pnum,
2322 int64_t *map, BlockDriverState **file)
2323 {
2324 return bdrv_common_block_status_above(bs, base, true, offset, bytes,
2325 pnum, map, file);
2326 }
2327
bdrv_block_status(BlockDriverState * bs,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)2328 int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes,
2329 int64_t *pnum, int64_t *map, BlockDriverState **file)
2330 {
2331 return bdrv_block_status_above(bs, backing_bs(bs),
2332 offset, bytes, pnum, map, file);
2333 }
2334
bdrv_is_allocated(BlockDriverState * bs,int64_t offset,int64_t bytes,int64_t * pnum)2335 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t offset,
2336 int64_t bytes, int64_t *pnum)
2337 {
2338 int ret;
2339 int64_t dummy;
2340
2341 ret = bdrv_common_block_status_above(bs, backing_bs(bs), false, offset,
2342 bytes, pnum ? pnum : &dummy, NULL,
2343 NULL);
2344 if (ret < 0) {
2345 return ret;
2346 }
2347 return !!(ret & BDRV_BLOCK_ALLOCATED);
2348 }
2349
2350 /*
2351 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2352 *
2353 * Return true if (a prefix of) the given range is allocated in any image
2354 * between BASE and TOP (inclusive). BASE can be NULL to check if the given
2355 * offset is allocated in any image of the chain. Return false otherwise,
2356 * or negative errno on failure.
2357 *
2358 * 'pnum' is set to the number of bytes (including and immediately
2359 * following the specified offset) that are known to be in the same
2360 * allocated/unallocated state. Note that a subsequent call starting
2361 * at 'offset + *pnum' may return the same allocation status (in other
2362 * words, the result is not necessarily the maximum possible range);
2363 * but 'pnum' will only be 0 when end of file is reached.
2364 *
2365 */
bdrv_is_allocated_above(BlockDriverState * top,BlockDriverState * base,int64_t offset,int64_t bytes,int64_t * pnum)2366 int bdrv_is_allocated_above(BlockDriverState *top,
2367 BlockDriverState *base,
2368 int64_t offset, int64_t bytes, int64_t *pnum)
2369 {
2370 BlockDriverState *intermediate;
2371 int ret;
2372 int64_t n = bytes;
2373
2374 intermediate = top;
2375 while (intermediate && intermediate != base) {
2376 int64_t pnum_inter;
2377 int64_t size_inter;
2378
2379 ret = bdrv_is_allocated(intermediate, offset, bytes, &pnum_inter);
2380 if (ret < 0) {
2381 return ret;
2382 }
2383 if (ret) {
2384 *pnum = pnum_inter;
2385 return 1;
2386 }
2387
2388 size_inter = bdrv_getlength(intermediate);
2389 if (size_inter < 0) {
2390 return size_inter;
2391 }
2392 if (n > pnum_inter &&
2393 (intermediate == top || offset + pnum_inter < size_inter)) {
2394 n = pnum_inter;
2395 }
2396
2397 intermediate = backing_bs(intermediate);
2398 }
2399
2400 *pnum = n;
2401 return 0;
2402 }
2403
2404 typedef struct BdrvVmstateCo {
2405 BlockDriverState *bs;
2406 QEMUIOVector *qiov;
2407 int64_t pos;
2408 bool is_read;
2409 int ret;
2410 } BdrvVmstateCo;
2411
2412 static int coroutine_fn
bdrv_co_rw_vmstate(BlockDriverState * bs,QEMUIOVector * qiov,int64_t pos,bool is_read)2413 bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
2414 bool is_read)
2415 {
2416 BlockDriver *drv = bs->drv;
2417 int ret = -ENOTSUP;
2418
2419 bdrv_inc_in_flight(bs);
2420
2421 if (!drv) {
2422 ret = -ENOMEDIUM;
2423 } else if (drv->bdrv_load_vmstate) {
2424 if (is_read) {
2425 ret = drv->bdrv_load_vmstate(bs, qiov, pos);
2426 } else {
2427 ret = drv->bdrv_save_vmstate(bs, qiov, pos);
2428 }
2429 } else if (bs->file) {
2430 ret = bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read);
2431 }
2432
2433 bdrv_dec_in_flight(bs);
2434 return ret;
2435 }
2436
bdrv_co_rw_vmstate_entry(void * opaque)2437 static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque)
2438 {
2439 BdrvVmstateCo *co = opaque;
2440 co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read);
2441 }
2442
2443 static inline int
bdrv_rw_vmstate(BlockDriverState * bs,QEMUIOVector * qiov,int64_t pos,bool is_read)2444 bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
2445 bool is_read)
2446 {
2447 if (qemu_in_coroutine()) {
2448 return bdrv_co_rw_vmstate(bs, qiov, pos, is_read);
2449 } else {
2450 BdrvVmstateCo data = {
2451 .bs = bs,
2452 .qiov = qiov,
2453 .pos = pos,
2454 .is_read = is_read,
2455 .ret = -EINPROGRESS,
2456 };
2457 Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data);
2458
2459 bdrv_coroutine_enter(bs, co);
2460 BDRV_POLL_WHILE(bs, data.ret == -EINPROGRESS);
2461 return data.ret;
2462 }
2463 }
2464
bdrv_save_vmstate(BlockDriverState * bs,const uint8_t * buf,int64_t pos,int size)2465 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2466 int64_t pos, int size)
2467 {
2468 QEMUIOVector qiov;
2469 struct iovec iov = {
2470 .iov_base = (void *) buf,
2471 .iov_len = size,
2472 };
2473 int ret;
2474
2475 qemu_iovec_init_external(&qiov, &iov, 1);
2476
2477 ret = bdrv_writev_vmstate(bs, &qiov, pos);
2478 if (ret < 0) {
2479 return ret;
2480 }
2481
2482 return size;
2483 }
2484
bdrv_writev_vmstate(BlockDriverState * bs,QEMUIOVector * qiov,int64_t pos)2485 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
2486 {
2487 return bdrv_rw_vmstate(bs, qiov, pos, false);
2488 }
2489
bdrv_load_vmstate(BlockDriverState * bs,uint8_t * buf,int64_t pos,int size)2490 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2491 int64_t pos, int size)
2492 {
2493 QEMUIOVector qiov;
2494 struct iovec iov = {
2495 .iov_base = buf,
2496 .iov_len = size,
2497 };
2498 int ret;
2499
2500 qemu_iovec_init_external(&qiov, &iov, 1);
2501 ret = bdrv_readv_vmstate(bs, &qiov, pos);
2502 if (ret < 0) {
2503 return ret;
2504 }
2505
2506 return size;
2507 }
2508
bdrv_readv_vmstate(BlockDriverState * bs,QEMUIOVector * qiov,int64_t pos)2509 int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
2510 {
2511 return bdrv_rw_vmstate(bs, qiov, pos, true);
2512 }
2513
2514 /**************************************************************/
2515 /* async I/Os */
2516
bdrv_aio_cancel(BlockAIOCB * acb)2517 void bdrv_aio_cancel(BlockAIOCB *acb)
2518 {
2519 qemu_aio_ref(acb);
2520 bdrv_aio_cancel_async(acb);
2521 while (acb->refcnt > 1) {
2522 if (acb->aiocb_info->get_aio_context) {
2523 aio_poll(acb->aiocb_info->get_aio_context(acb), true);
2524 } else if (acb->bs) {
2525 /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
2526 * assert that we're not using an I/O thread. Thread-safe
2527 * code should use bdrv_aio_cancel_async exclusively.
2528 */
2529 assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context());
2530 aio_poll(bdrv_get_aio_context(acb->bs), true);
2531 } else {
2532 abort();
2533 }
2534 }
2535 qemu_aio_unref(acb);
2536 }
2537
2538 /* Async version of aio cancel. The caller is not blocked if the acb implements
2539 * cancel_async, otherwise we do nothing and let the request normally complete.
2540 * In either case the completion callback must be called. */
bdrv_aio_cancel_async(BlockAIOCB * acb)2541 void bdrv_aio_cancel_async(BlockAIOCB *acb)
2542 {
2543 if (acb->aiocb_info->cancel_async) {
2544 acb->aiocb_info->cancel_async(acb);
2545 }
2546 }
2547
2548 /**************************************************************/
2549 /* Coroutine block device emulation */
2550
2551 typedef struct FlushCo {
2552 BlockDriverState *bs;
2553 int ret;
2554 } FlushCo;
2555
2556
bdrv_flush_co_entry(void * opaque)2557 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
2558 {
2559 FlushCo *rwco = opaque;
2560
2561 rwco->ret = bdrv_co_flush(rwco->bs);
2562 }
2563
bdrv_co_flush(BlockDriverState * bs)2564 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
2565 {
2566 int current_gen;
2567 int ret = 0;
2568
2569 bdrv_inc_in_flight(bs);
2570
2571 if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs) ||
2572 bdrv_is_sg(bs)) {
2573 goto early_exit;
2574 }
2575
2576 qemu_co_mutex_lock(&bs->reqs_lock);
2577 current_gen = atomic_read(&bs->write_gen);
2578
2579 /* Wait until any previous flushes are completed */
2580 while (bs->active_flush_req) {
2581 qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock);
2582 }
2583
2584 /* Flushes reach this point in nondecreasing current_gen order. */
2585 bs->active_flush_req = true;
2586 qemu_co_mutex_unlock(&bs->reqs_lock);
2587
2588 /* Write back all layers by calling one driver function */
2589 if (bs->drv->bdrv_co_flush) {
2590 ret = bs->drv->bdrv_co_flush(bs);
2591 goto out;
2592 }
2593
2594 /* Write back cached data to the OS even with cache=unsafe */
2595 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
2596 if (bs->drv->bdrv_co_flush_to_os) {
2597 ret = bs->drv->bdrv_co_flush_to_os(bs);
2598 if (ret < 0) {
2599 goto out;
2600 }
2601 }
2602
2603 /* But don't actually force it to the disk with cache=unsafe */
2604 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2605 goto flush_parent;
2606 }
2607
2608 /* Check if we really need to flush anything */
2609 if (bs->flushed_gen == current_gen) {
2610 goto flush_parent;
2611 }
2612
2613 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
2614 if (!bs->drv) {
2615 /* bs->drv->bdrv_co_flush() might have ejected the BDS
2616 * (even in case of apparent success) */
2617 ret = -ENOMEDIUM;
2618 goto out;
2619 }
2620 if (bs->drv->bdrv_co_flush_to_disk) {
2621 ret = bs->drv->bdrv_co_flush_to_disk(bs);
2622 } else if (bs->drv->bdrv_aio_flush) {
2623 BlockAIOCB *acb;
2624 CoroutineIOCompletion co = {
2625 .coroutine = qemu_coroutine_self(),
2626 };
2627
2628 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
2629 if (acb == NULL) {
2630 ret = -EIO;
2631 } else {
2632 qemu_coroutine_yield();
2633 ret = co.ret;
2634 }
2635 } else {
2636 /*
2637 * Some block drivers always operate in either writethrough or unsafe
2638 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2639 * know how the server works (because the behaviour is hardcoded or
2640 * depends on server-side configuration), so we can't ensure that
2641 * everything is safe on disk. Returning an error doesn't work because
2642 * that would break guests even if the server operates in writethrough
2643 * mode.
2644 *
2645 * Let's hope the user knows what he's doing.
2646 */
2647 ret = 0;
2648 }
2649
2650 if (ret < 0) {
2651 goto out;
2652 }
2653
2654 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2655 * in the case of cache=unsafe, so there are no useless flushes.
2656 */
2657 flush_parent:
2658 ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0;
2659 out:
2660 /* Notify any pending flushes that we have completed */
2661 if (ret == 0) {
2662 bs->flushed_gen = current_gen;
2663 }
2664
2665 qemu_co_mutex_lock(&bs->reqs_lock);
2666 bs->active_flush_req = false;
2667 /* Return value is ignored - it's ok if wait queue is empty */
2668 qemu_co_queue_next(&bs->flush_queue);
2669 qemu_co_mutex_unlock(&bs->reqs_lock);
2670
2671 early_exit:
2672 bdrv_dec_in_flight(bs);
2673 return ret;
2674 }
2675
bdrv_flush(BlockDriverState * bs)2676 int bdrv_flush(BlockDriverState *bs)
2677 {
2678 Coroutine *co;
2679 FlushCo flush_co = {
2680 .bs = bs,
2681 .ret = NOT_DONE,
2682 };
2683
2684 if (qemu_in_coroutine()) {
2685 /* Fast-path if already in coroutine context */
2686 bdrv_flush_co_entry(&flush_co);
2687 } else {
2688 co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co);
2689 bdrv_coroutine_enter(bs, co);
2690 BDRV_POLL_WHILE(bs, flush_co.ret == NOT_DONE);
2691 }
2692
2693 return flush_co.ret;
2694 }
2695
2696 typedef struct DiscardCo {
2697 BdrvChild *child;
2698 int64_t offset;
2699 int bytes;
2700 int ret;
2701 } DiscardCo;
bdrv_pdiscard_co_entry(void * opaque)2702 static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque)
2703 {
2704 DiscardCo *rwco = opaque;
2705
2706 rwco->ret = bdrv_co_pdiscard(rwco->child, rwco->offset, rwco->bytes);
2707 }
2708
bdrv_co_pdiscard(BdrvChild * child,int64_t offset,int bytes)2709 int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, int bytes)
2710 {
2711 BdrvTrackedRequest req;
2712 int max_pdiscard, ret;
2713 int head, tail, align;
2714 BlockDriverState *bs = child->bs;
2715
2716 if (!bs || !bs->drv) {
2717 return -ENOMEDIUM;
2718 }
2719
2720 if (bdrv_has_readonly_bitmaps(bs)) {
2721 return -EPERM;
2722 }
2723
2724 ret = bdrv_check_byte_request(bs, offset, bytes);
2725 if (ret < 0) {
2726 return ret;
2727 }
2728
2729 /* Do nothing if disabled. */
2730 if (!(bs->open_flags & BDRV_O_UNMAP)) {
2731 return 0;
2732 }
2733
2734 if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) {
2735 return 0;
2736 }
2737
2738 /* Discard is advisory, but some devices track and coalesce
2739 * unaligned requests, so we must pass everything down rather than
2740 * round here. Still, most devices will just silently ignore
2741 * unaligned requests (by returning -ENOTSUP), so we must fragment
2742 * the request accordingly. */
2743 align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment);
2744 assert(align % bs->bl.request_alignment == 0);
2745 head = offset % align;
2746 tail = (offset + bytes) % align;
2747
2748 bdrv_inc_in_flight(bs);
2749 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD);
2750
2751 ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0);
2752 if (ret < 0) {
2753 goto out;
2754 }
2755
2756 max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX),
2757 align);
2758 assert(max_pdiscard >= bs->bl.request_alignment);
2759
2760 while (bytes > 0) {
2761 int num = bytes;
2762
2763 if (head) {
2764 /* Make small requests to get to alignment boundaries. */
2765 num = MIN(bytes, align - head);
2766 if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) {
2767 num %= bs->bl.request_alignment;
2768 }
2769 head = (head + num) % align;
2770 assert(num < max_pdiscard);
2771 } else if (tail) {
2772 if (num > align) {
2773 /* Shorten the request to the last aligned cluster. */
2774 num -= tail;
2775 } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) &&
2776 tail > bs->bl.request_alignment) {
2777 tail %= bs->bl.request_alignment;
2778 num -= tail;
2779 }
2780 }
2781 /* limit request size */
2782 if (num > max_pdiscard) {
2783 num = max_pdiscard;
2784 }
2785
2786 if (!bs->drv) {
2787 ret = -ENOMEDIUM;
2788 goto out;
2789 }
2790 if (bs->drv->bdrv_co_pdiscard) {
2791 ret = bs->drv->bdrv_co_pdiscard(bs, offset, num);
2792 } else {
2793 BlockAIOCB *acb;
2794 CoroutineIOCompletion co = {
2795 .coroutine = qemu_coroutine_self(),
2796 };
2797
2798 acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num,
2799 bdrv_co_io_em_complete, &co);
2800 if (acb == NULL) {
2801 ret = -EIO;
2802 goto out;
2803 } else {
2804 qemu_coroutine_yield();
2805 ret = co.ret;
2806 }
2807 }
2808 if (ret && ret != -ENOTSUP) {
2809 goto out;
2810 }
2811
2812 offset += num;
2813 bytes -= num;
2814 }
2815 ret = 0;
2816 out:
2817 bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret);
2818 tracked_request_end(&req);
2819 bdrv_dec_in_flight(bs);
2820 return ret;
2821 }
2822
bdrv_pdiscard(BdrvChild * child,int64_t offset,int bytes)2823 int bdrv_pdiscard(BdrvChild *child, int64_t offset, int bytes)
2824 {
2825 Coroutine *co;
2826 DiscardCo rwco = {
2827 .child = child,
2828 .offset = offset,
2829 .bytes = bytes,
2830 .ret = NOT_DONE,
2831 };
2832
2833 if (qemu_in_coroutine()) {
2834 /* Fast-path if already in coroutine context */
2835 bdrv_pdiscard_co_entry(&rwco);
2836 } else {
2837 co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco);
2838 bdrv_coroutine_enter(child->bs, co);
2839 BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE);
2840 }
2841
2842 return rwco.ret;
2843 }
2844
bdrv_co_ioctl(BlockDriverState * bs,int req,void * buf)2845 int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf)
2846 {
2847 BlockDriver *drv = bs->drv;
2848 CoroutineIOCompletion co = {
2849 .coroutine = qemu_coroutine_self(),
2850 };
2851 BlockAIOCB *acb;
2852
2853 bdrv_inc_in_flight(bs);
2854 if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) {
2855 co.ret = -ENOTSUP;
2856 goto out;
2857 }
2858
2859 if (drv->bdrv_co_ioctl) {
2860 co.ret = drv->bdrv_co_ioctl(bs, req, buf);
2861 } else {
2862 acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co);
2863 if (!acb) {
2864 co.ret = -ENOTSUP;
2865 goto out;
2866 }
2867 qemu_coroutine_yield();
2868 }
2869 out:
2870 bdrv_dec_in_flight(bs);
2871 return co.ret;
2872 }
2873
qemu_blockalign(BlockDriverState * bs,size_t size)2874 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2875 {
2876 return qemu_memalign(bdrv_opt_mem_align(bs), size);
2877 }
2878
qemu_blockalign0(BlockDriverState * bs,size_t size)2879 void *qemu_blockalign0(BlockDriverState *bs, size_t size)
2880 {
2881 return memset(qemu_blockalign(bs, size), 0, size);
2882 }
2883
qemu_try_blockalign(BlockDriverState * bs,size_t size)2884 void *qemu_try_blockalign(BlockDriverState *bs, size_t size)
2885 {
2886 size_t align = bdrv_opt_mem_align(bs);
2887
2888 /* Ensure that NULL is never returned on success */
2889 assert(align > 0);
2890 if (size == 0) {
2891 size = align;
2892 }
2893
2894 return qemu_try_memalign(align, size);
2895 }
2896
qemu_try_blockalign0(BlockDriverState * bs,size_t size)2897 void *qemu_try_blockalign0(BlockDriverState *bs, size_t size)
2898 {
2899 void *mem = qemu_try_blockalign(bs, size);
2900
2901 if (mem) {
2902 memset(mem, 0, size);
2903 }
2904
2905 return mem;
2906 }
2907
2908 /*
2909 * Check if all memory in this vector is sector aligned.
2910 */
bdrv_qiov_is_aligned(BlockDriverState * bs,QEMUIOVector * qiov)2911 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
2912 {
2913 int i;
2914 size_t alignment = bdrv_min_mem_align(bs);
2915
2916 for (i = 0; i < qiov->niov; i++) {
2917 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
2918 return false;
2919 }
2920 if (qiov->iov[i].iov_len % alignment) {
2921 return false;
2922 }
2923 }
2924
2925 return true;
2926 }
2927
bdrv_add_before_write_notifier(BlockDriverState * bs,NotifierWithReturn * notifier)2928 void bdrv_add_before_write_notifier(BlockDriverState *bs,
2929 NotifierWithReturn *notifier)
2930 {
2931 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
2932 }
2933
bdrv_io_plug(BlockDriverState * bs)2934 void bdrv_io_plug(BlockDriverState *bs)
2935 {
2936 BdrvChild *child;
2937
2938 QLIST_FOREACH(child, &bs->children, next) {
2939 bdrv_io_plug(child->bs);
2940 }
2941
2942 if (atomic_fetch_inc(&bs->io_plugged) == 0) {
2943 BlockDriver *drv = bs->drv;
2944 if (drv && drv->bdrv_io_plug) {
2945 drv->bdrv_io_plug(bs);
2946 }
2947 }
2948 }
2949
bdrv_io_unplug(BlockDriverState * bs)2950 void bdrv_io_unplug(BlockDriverState *bs)
2951 {
2952 BdrvChild *child;
2953
2954 assert(bs->io_plugged);
2955 if (atomic_fetch_dec(&bs->io_plugged) == 1) {
2956 BlockDriver *drv = bs->drv;
2957 if (drv && drv->bdrv_io_unplug) {
2958 drv->bdrv_io_unplug(bs);
2959 }
2960 }
2961
2962 QLIST_FOREACH(child, &bs->children, next) {
2963 bdrv_io_unplug(child->bs);
2964 }
2965 }
2966
bdrv_register_buf(BlockDriverState * bs,void * host,size_t size)2967 void bdrv_register_buf(BlockDriverState *bs, void *host, size_t size)
2968 {
2969 BdrvChild *child;
2970
2971 if (bs->drv && bs->drv->bdrv_register_buf) {
2972 bs->drv->bdrv_register_buf(bs, host, size);
2973 }
2974 QLIST_FOREACH(child, &bs->children, next) {
2975 bdrv_register_buf(child->bs, host, size);
2976 }
2977 }
2978
bdrv_unregister_buf(BlockDriverState * bs,void * host)2979 void bdrv_unregister_buf(BlockDriverState *bs, void *host)
2980 {
2981 BdrvChild *child;
2982
2983 if (bs->drv && bs->drv->bdrv_unregister_buf) {
2984 bs->drv->bdrv_unregister_buf(bs, host);
2985 }
2986 QLIST_FOREACH(child, &bs->children, next) {
2987 bdrv_unregister_buf(child->bs, host);
2988 }
2989 }
2990
bdrv_co_copy_range_internal(BdrvChild * src,uint64_t src_offset,BdrvChild * dst,uint64_t dst_offset,uint64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags,bool recurse_src)2991 static int coroutine_fn bdrv_co_copy_range_internal(
2992 BdrvChild *src, uint64_t src_offset, BdrvChild *dst,
2993 uint64_t dst_offset, uint64_t bytes,
2994 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags,
2995 bool recurse_src)
2996 {
2997 BdrvTrackedRequest req;
2998 int ret;
2999
3000 if (!dst || !dst->bs) {
3001 return -ENOMEDIUM;
3002 }
3003 ret = bdrv_check_byte_request(dst->bs, dst_offset, bytes);
3004 if (ret) {
3005 return ret;
3006 }
3007 if (write_flags & BDRV_REQ_ZERO_WRITE) {
3008 return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags);
3009 }
3010
3011 if (!src || !src->bs) {
3012 return -ENOMEDIUM;
3013 }
3014 ret = bdrv_check_byte_request(src->bs, src_offset, bytes);
3015 if (ret) {
3016 return ret;
3017 }
3018
3019 if (!src->bs->drv->bdrv_co_copy_range_from
3020 || !dst->bs->drv->bdrv_co_copy_range_to
3021 || src->bs->encrypted || dst->bs->encrypted) {
3022 return -ENOTSUP;
3023 }
3024
3025 if (recurse_src) {
3026 bdrv_inc_in_flight(src->bs);
3027 tracked_request_begin(&req, src->bs, src_offset, bytes,
3028 BDRV_TRACKED_READ);
3029
3030 /* BDRV_REQ_SERIALISING is only for write operation */
3031 assert(!(read_flags & BDRV_REQ_SERIALISING));
3032 if (!(read_flags & BDRV_REQ_NO_SERIALISING)) {
3033 wait_serialising_requests(&req);
3034 }
3035
3036 ret = src->bs->drv->bdrv_co_copy_range_from(src->bs,
3037 src, src_offset,
3038 dst, dst_offset,
3039 bytes,
3040 read_flags, write_flags);
3041
3042 tracked_request_end(&req);
3043 bdrv_dec_in_flight(src->bs);
3044 } else {
3045 bdrv_inc_in_flight(dst->bs);
3046 tracked_request_begin(&req, dst->bs, dst_offset, bytes,
3047 BDRV_TRACKED_WRITE);
3048 ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req,
3049 write_flags);
3050 if (!ret) {
3051 ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs,
3052 src, src_offset,
3053 dst, dst_offset,
3054 bytes,
3055 read_flags, write_flags);
3056 }
3057 bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret);
3058 tracked_request_end(&req);
3059 bdrv_dec_in_flight(dst->bs);
3060 }
3061
3062 return ret;
3063 }
3064
3065 /* Copy range from @src to @dst.
3066 *
3067 * See the comment of bdrv_co_copy_range for the parameter and return value
3068 * semantics. */
bdrv_co_copy_range_from(BdrvChild * src,uint64_t src_offset,BdrvChild * dst,uint64_t dst_offset,uint64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)3069 int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, uint64_t src_offset,
3070 BdrvChild *dst, uint64_t dst_offset,
3071 uint64_t bytes,
3072 BdrvRequestFlags read_flags,
3073 BdrvRequestFlags write_flags)
3074 {
3075 trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes,
3076 read_flags, write_flags);
3077 return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset,
3078 bytes, read_flags, write_flags, true);
3079 }
3080
3081 /* Copy range from @src to @dst.
3082 *
3083 * See the comment of bdrv_co_copy_range for the parameter and return value
3084 * semantics. */
bdrv_co_copy_range_to(BdrvChild * src,uint64_t src_offset,BdrvChild * dst,uint64_t dst_offset,uint64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)3085 int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, uint64_t src_offset,
3086 BdrvChild *dst, uint64_t dst_offset,
3087 uint64_t bytes,
3088 BdrvRequestFlags read_flags,
3089 BdrvRequestFlags write_flags)
3090 {
3091 trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
3092 read_flags, write_flags);
3093 return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset,
3094 bytes, read_flags, write_flags, false);
3095 }
3096
bdrv_co_copy_range(BdrvChild * src,uint64_t src_offset,BdrvChild * dst,uint64_t dst_offset,uint64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)3097 int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset,
3098 BdrvChild *dst, uint64_t dst_offset,
3099 uint64_t bytes, BdrvRequestFlags read_flags,
3100 BdrvRequestFlags write_flags)
3101 {
3102 return bdrv_co_copy_range_from(src, src_offset,
3103 dst, dst_offset,
3104 bytes, read_flags, write_flags);
3105 }
3106
bdrv_parent_cb_resize(BlockDriverState * bs)3107 static void bdrv_parent_cb_resize(BlockDriverState *bs)
3108 {
3109 BdrvChild *c;
3110 QLIST_FOREACH(c, &bs->parents, next_parent) {
3111 if (c->role->resize) {
3112 c->role->resize(c);
3113 }
3114 }
3115 }
3116
3117 /**
3118 * Truncate file to 'offset' bytes (needed only for file protocols)
3119 */
bdrv_co_truncate(BdrvChild * child,int64_t offset,PreallocMode prealloc,Error ** errp)3120 int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset,
3121 PreallocMode prealloc, Error **errp)
3122 {
3123 BlockDriverState *bs = child->bs;
3124 BlockDriver *drv = bs->drv;
3125 BdrvTrackedRequest req;
3126 int64_t old_size, new_bytes;
3127 int ret;
3128
3129
3130 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3131 if (!drv) {
3132 error_setg(errp, "No medium inserted");
3133 return -ENOMEDIUM;
3134 }
3135 if (offset < 0) {
3136 error_setg(errp, "Image size cannot be negative");
3137 return -EINVAL;
3138 }
3139
3140 old_size = bdrv_getlength(bs);
3141 if (old_size < 0) {
3142 error_setg_errno(errp, -old_size, "Failed to get old image size");
3143 return old_size;
3144 }
3145
3146 if (offset > old_size) {
3147 new_bytes = offset - old_size;
3148 } else {
3149 new_bytes = 0;
3150 }
3151
3152 bdrv_inc_in_flight(bs);
3153 tracked_request_begin(&req, bs, offset - new_bytes, new_bytes,
3154 BDRV_TRACKED_TRUNCATE);
3155
3156 /* If we are growing the image and potentially using preallocation for the
3157 * new area, we need to make sure that no write requests are made to it
3158 * concurrently or they might be overwritten by preallocation. */
3159 if (new_bytes) {
3160 mark_request_serialising(&req, 1);
3161 }
3162 if (bs->read_only) {
3163 error_setg(errp, "Image is read-only");
3164 ret = -EACCES;
3165 goto out;
3166 }
3167 ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req,
3168 0);
3169 if (ret < 0) {
3170 error_setg_errno(errp, -ret,
3171 "Failed to prepare request for truncation");
3172 goto out;
3173 }
3174
3175 if (!drv->bdrv_co_truncate) {
3176 if (bs->file && drv->is_filter) {
3177 ret = bdrv_co_truncate(bs->file, offset, prealloc, errp);
3178 goto out;
3179 }
3180 error_setg(errp, "Image format driver does not support resize");
3181 ret = -ENOTSUP;
3182 goto out;
3183 }
3184
3185 ret = drv->bdrv_co_truncate(bs, offset, prealloc, errp);
3186 if (ret < 0) {
3187 goto out;
3188 }
3189 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3190 if (ret < 0) {
3191 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3192 } else {
3193 offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3194 }
3195 /* It's possible that truncation succeeded but refresh_total_sectors
3196 * failed, but the latter doesn't affect how we should finish the request.
3197 * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. */
3198 bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0);
3199
3200 out:
3201 tracked_request_end(&req);
3202 bdrv_dec_in_flight(bs);
3203
3204 return ret;
3205 }
3206
3207 typedef struct TruncateCo {
3208 BdrvChild *child;
3209 int64_t offset;
3210 PreallocMode prealloc;
3211 Error **errp;
3212 int ret;
3213 } TruncateCo;
3214
bdrv_truncate_co_entry(void * opaque)3215 static void coroutine_fn bdrv_truncate_co_entry(void *opaque)
3216 {
3217 TruncateCo *tco = opaque;
3218 tco->ret = bdrv_co_truncate(tco->child, tco->offset, tco->prealloc,
3219 tco->errp);
3220 }
3221
bdrv_truncate(BdrvChild * child,int64_t offset,PreallocMode prealloc,Error ** errp)3222 int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
3223 Error **errp)
3224 {
3225 Coroutine *co;
3226 TruncateCo tco = {
3227 .child = child,
3228 .offset = offset,
3229 .prealloc = prealloc,
3230 .errp = errp,
3231 .ret = NOT_DONE,
3232 };
3233
3234 if (qemu_in_coroutine()) {
3235 /* Fast-path if already in coroutine context */
3236 bdrv_truncate_co_entry(&tco);
3237 } else {
3238 co = qemu_coroutine_create(bdrv_truncate_co_entry, &tco);
3239 qemu_coroutine_enter(co);
3240 BDRV_POLL_WHILE(child->bs, tco.ret == NOT_DONE);
3241 }
3242
3243 return tco.ret;
3244 }
3245