]> git.proxmox.com Git - mirror_qemu.git/blame - block/block-copy.c
block-copy: fix block_copy_task_entry() progress update
[mirror_qemu.git] / block / block-copy.c
CommitLineData
beb5f545
VSO
1/*
2 * block_copy API
3 *
4 * Copyright (C) 2013 Proxmox Server Solutions
5 * Copyright (c) 2019 Virtuozzo International GmbH.
6 *
7 * Authors:
8 * Dietmar Maurer (dietmar@proxmox.com)
9 * Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
10 *
11 * This work is licensed under the terms of the GNU GPL, version 2 or later.
12 * See the COPYING file in the top-level directory.
13 */
14
15#include "qemu/osdep.h"
16
17#include "trace.h"
18#include "qapi/error.h"
19#include "block/block-copy.h"
20#include "sysemu/block-backend.h"
b3b7036a 21#include "qemu/units.h"
4ce5dd3e
VSO
22#include "qemu/coroutine.h"
23#include "block/aio_task.h"
b3b7036a
VSO
24
25#define BLOCK_COPY_MAX_COPY_RANGE (16 * MiB)
0e240245 26#define BLOCK_COPY_MAX_BUFFER (1 * MiB)
7f739d0e 27#define BLOCK_COPY_MAX_MEM (128 * MiB)
4ce5dd3e 28#define BLOCK_COPY_MAX_WORKERS 64
7e032df0 29#define BLOCK_COPY_SLICE_TIME 100000000ULL /* ns */
4ce5dd3e
VSO
30
31static coroutine_fn int block_copy_task_entry(AioTask *task);
32
33typedef struct BlockCopyCallState {
de4641b4 34 /* IN parameters. Initialized in block_copy_async() and never changed. */
3b8c2329
VSO
35 BlockCopyState *s;
36 int64_t offset;
37 int64_t bytes;
26be9d62
VSO
38 int max_workers;
39 int64_t max_chunk;
7e032df0 40 bool ignore_ratelimit;
de4641b4
VSO
41 BlockCopyAsyncCallbackFunc cb;
42 void *cb_opaque;
43
44 /* Coroutine where async block-copy is running */
45 Coroutine *co;
3b8c2329 46
2e099a9d
VSO
47 /* To reference all call states from BlockCopyState */
48 QLIST_ENTRY(BlockCopyCallState) list;
49
3b8c2329 50 /* State */
de4641b4
VSO
51 int ret;
52 bool finished;
29a6ea24 53 QemuCoSleep sleep;
a6d23d56 54 bool cancelled;
3b8c2329
VSO
55
56 /* OUT parameters */
4ce5dd3e
VSO
57 bool error_is_read;
58} BlockCopyCallState;
beb5f545 59
e9407785 60typedef struct BlockCopyTask {
4ce5dd3e
VSO
61 AioTask task;
62
1348a657 63 BlockCopyState *s;
4ce5dd3e 64 BlockCopyCallState *call_state;
397f4e9d
VSO
65 int64_t offset;
66 int64_t bytes;
4ce5dd3e 67 bool zeroes;
e9407785
VSO
68 QLIST_ENTRY(BlockCopyTask) list;
69 CoQueue wait_queue; /* coroutines blocked on this task */
70} BlockCopyTask;
397f4e9d 71
42ac2144
VSO
72static int64_t task_end(BlockCopyTask *task)
73{
74 return task->offset + task->bytes;
75}
76
397f4e9d
VSO
77typedef struct BlockCopyState {
78 /*
79 * BdrvChild objects are not owned or managed by block-copy. They are
80 * provided by block-copy user and user is responsible for appropriate
81 * permissions on these children.
82 */
83 BdrvChild *source;
84 BdrvChild *target;
85 BdrvDirtyBitmap *copy_bitmap;
86 int64_t in_flight_bytes;
87 int64_t cluster_size;
88 bool use_copy_range;
89 int64_t copy_size;
90 uint64_t len;
2e099a9d
VSO
91 QLIST_HEAD(, BlockCopyTask) tasks; /* All tasks from all block-copy calls */
92 QLIST_HEAD(, BlockCopyCallState) calls;
397f4e9d
VSO
93
94 BdrvRequestFlags write_flags;
95
96 /*
97 * skip_unallocated:
98 *
99 * Used by sync=top jobs, which first scan the source node for unallocated
100 * areas and clear them in the copy_bitmap. During this process, the bitmap
101 * is thus not fully initialized: It may still have bits set for areas that
102 * are unallocated and should actually not be copied.
103 *
104 * This is indicated by skip_unallocated.
105 *
106 * In this case, block_copy() will query the source’s allocation status,
107 * skip unallocated regions, clear them in the copy_bitmap, and invoke
108 * block_copy_reset_unallocated() every time it does.
109 */
110 bool skip_unallocated;
111
112 ProgressMeter *progress;
397f4e9d
VSO
113
114 SharedResource *mem;
7e032df0
VSO
115
116 uint64_t speed;
117 RateLimit rate_limit;
397f4e9d
VSO
118} BlockCopyState;
119
e9407785
VSO
120static BlockCopyTask *find_conflicting_task(BlockCopyState *s,
121 int64_t offset, int64_t bytes)
17187cb6 122{
e9407785 123 BlockCopyTask *t;
17187cb6 124
e9407785
VSO
125 QLIST_FOREACH(t, &s->tasks, list) {
126 if (offset + bytes > t->offset && offset < t->offset + t->bytes) {
127 return t;
17187cb6
VSO
128 }
129 }
130
131 return NULL;
132}
133
5332e5d2 134/*
e9407785
VSO
135 * If there are no intersecting tasks return false. Otherwise, wait for the
136 * first found intersecting tasks to finish and return true.
5332e5d2
VSO
137 */
138static bool coroutine_fn block_copy_wait_one(BlockCopyState *s, int64_t offset,
139 int64_t bytes)
a6ffe199 140{
e9407785 141 BlockCopyTask *task = find_conflicting_task(s, offset, bytes);
17187cb6 142
e9407785 143 if (!task) {
5332e5d2 144 return false;
17187cb6 145 }
5332e5d2 146
e9407785 147 qemu_co_queue_wait(&task->wait_queue, NULL);
5332e5d2
VSO
148
149 return true;
a6ffe199
VSO
150}
151
42ac2144
VSO
152/*
153 * Search for the first dirty area in offset/bytes range and create task at
154 * the beginning of it.
155 */
f13e60a9 156static BlockCopyTask *block_copy_task_create(BlockCopyState *s,
4ce5dd3e 157 BlockCopyCallState *call_state,
f13e60a9 158 int64_t offset, int64_t bytes)
a6ffe199 159{
42ac2144 160 BlockCopyTask *task;
26be9d62 161 int64_t max_chunk = MIN_NON_ZERO(s->copy_size, call_state->max_chunk);
f13e60a9 162
42ac2144
VSO
163 if (!bdrv_dirty_bitmap_next_dirty_area(s->copy_bitmap,
164 offset, offset + bytes,
26be9d62 165 max_chunk, &offset, &bytes))
42ac2144
VSO
166 {
167 return NULL;
168 }
169
7661a886
SR
170 assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
171 bytes = QEMU_ALIGN_UP(bytes, s->cluster_size);
172
42ac2144 173 /* region is dirty, so no existent tasks possible in it */
e9407785 174 assert(!find_conflicting_task(s, offset, bytes));
5332e5d2
VSO
175
176 bdrv_reset_dirty_bitmap(s->copy_bitmap, offset, bytes);
177 s->in_flight_bytes += bytes;
178
42ac2144 179 task = g_new(BlockCopyTask, 1);
1348a657 180 *task = (BlockCopyTask) {
4ce5dd3e 181 .task.func = block_copy_task_entry,
1348a657 182 .s = s,
4ce5dd3e 183 .call_state = call_state,
1348a657
VSO
184 .offset = offset,
185 .bytes = bytes,
186 };
e9407785
VSO
187 qemu_co_queue_init(&task->wait_queue);
188 QLIST_INSERT_HEAD(&s->tasks, task, list);
f13e60a9
VSO
189
190 return task;
a6ffe199
VSO
191}
192
5332e5d2 193/*
e9407785 194 * block_copy_task_shrink
5332e5d2 195 *
e9407785
VSO
196 * Drop the tail of the task to be handled later. Set dirty bits back and
197 * wake up all tasks waiting for us (may be some of them are not intersecting
198 * with shrunk task)
5332e5d2 199 */
1348a657 200static void coroutine_fn block_copy_task_shrink(BlockCopyTask *task,
e9407785 201 int64_t new_bytes)
a6ffe199 202{
e9407785 203 if (new_bytes == task->bytes) {
5332e5d2
VSO
204 return;
205 }
206
e9407785 207 assert(new_bytes > 0 && new_bytes < task->bytes);
5332e5d2 208
1348a657
VSO
209 task->s->in_flight_bytes -= task->bytes - new_bytes;
210 bdrv_set_dirty_bitmap(task->s->copy_bitmap,
e9407785 211 task->offset + new_bytes, task->bytes - new_bytes);
5332e5d2 212
e9407785
VSO
213 task->bytes = new_bytes;
214 qemu_co_queue_restart_all(&task->wait_queue);
5332e5d2
VSO
215}
216
1348a657 217static void coroutine_fn block_copy_task_end(BlockCopyTask *task, int ret)
5332e5d2 218{
1348a657 219 task->s->in_flight_bytes -= task->bytes;
5332e5d2 220 if (ret < 0) {
1348a657 221 bdrv_set_dirty_bitmap(task->s->copy_bitmap, task->offset, task->bytes);
5332e5d2 222 }
e9407785
VSO
223 QLIST_REMOVE(task, list);
224 qemu_co_queue_restart_all(&task->wait_queue);
a6ffe199
VSO
225}
226
beb5f545
VSO
227void block_copy_state_free(BlockCopyState *s)
228{
229 if (!s) {
230 return;
231 }
232
4951967d 233 ratelimit_destroy(&s->rate_limit);
5deb6cbd 234 bdrv_release_dirty_bitmap(s->copy_bitmap);
7f739d0e 235 shres_destroy(s->mem);
beb5f545
VSO
236 g_free(s);
237}
238
9d31bc53
VSO
239static uint32_t block_copy_max_transfer(BdrvChild *source, BdrvChild *target)
240{
241 return MIN_NON_ZERO(INT_MAX,
242 MIN_NON_ZERO(source->bs->bl.max_transfer,
243 target->bs->bl.max_transfer));
244}
245
00e30f05 246BlockCopyState *block_copy_state_new(BdrvChild *source, BdrvChild *target,
86c6a3b6 247 int64_t cluster_size, bool use_copy_range,
0f4b02b7 248 BdrvRequestFlags write_flags, Error **errp)
beb5f545
VSO
249{
250 BlockCopyState *s;
beb5f545
VSO
251 BdrvDirtyBitmap *copy_bitmap;
252
00e30f05
VSO
253 copy_bitmap = bdrv_create_dirty_bitmap(source->bs, cluster_size, NULL,
254 errp);
beb5f545
VSO
255 if (!copy_bitmap) {
256 return NULL;
257 }
258 bdrv_disable_dirty_bitmap(copy_bitmap);
259
260 s = g_new(BlockCopyState, 1);
261 *s = (BlockCopyState) {
00e30f05
VSO
262 .source = source,
263 .target = target,
beb5f545
VSO
264 .copy_bitmap = copy_bitmap,
265 .cluster_size = cluster_size,
266 .len = bdrv_dirty_bitmap_size(copy_bitmap),
267 .write_flags = write_flags,
7f739d0e 268 .mem = shres_create(BLOCK_COPY_MAX_MEM),
beb5f545
VSO
269 };
270
9d31bc53 271 if (block_copy_max_transfer(source, target) < cluster_size) {
0e240245
VSO
272 /*
273 * copy_range does not respect max_transfer. We don't want to bother
274 * with requests smaller than block-copy cluster size, so fallback to
275 * buffered copying (read and write respect max_transfer on their
276 * behalf).
277 */
278 s->use_copy_range = false;
279 s->copy_size = cluster_size;
280 } else if (write_flags & BDRV_REQ_WRITE_COMPRESSED) {
dcfbece6 281 /* Compression supports only cluster-size writes and no copy-range. */
0e240245 282 s->use_copy_range = false;
dcfbece6 283 s->copy_size = cluster_size;
0e240245
VSO
284 } else {
285 /*
9d31bc53
VSO
286 * We enable copy-range, but keep small copy_size, until first
287 * successful copy_range (look at block_copy_do_copy).
0e240245 288 */
86c6a3b6 289 s->use_copy_range = use_copy_range;
9d31bc53 290 s->copy_size = MAX(s->cluster_size, BLOCK_COPY_MAX_BUFFER);
0e240245 291 }
beb5f545 292
4951967d 293 ratelimit_init(&s->rate_limit);
e9407785 294 QLIST_INIT(&s->tasks);
2e099a9d 295 QLIST_INIT(&s->calls);
a6ffe199 296
beb5f545 297 return s;
beb5f545
VSO
298}
299
d0ebeca1
VSO
300void block_copy_set_progress_meter(BlockCopyState *s, ProgressMeter *pm)
301{
302 s->progress = pm;
303}
304
4ce5dd3e
VSO
305/*
306 * Takes ownership of @task
307 *
308 * If pool is NULL directly run the task, otherwise schedule it into the pool.
309 *
310 * Returns: task.func return code if pool is NULL
311 * otherwise -ECANCELED if pool status is bad
312 * otherwise 0 (successfully scheduled)
313 */
314static coroutine_fn int block_copy_task_run(AioTaskPool *pool,
315 BlockCopyTask *task)
316{
317 if (!pool) {
318 int ret = task->task.func(&task->task);
319
320 g_free(task);
321 return ret;
322 }
323
324 aio_task_pool_wait_slot(pool);
325 if (aio_task_pool_status(pool) < 0) {
326 co_put_to_shres(task->s->mem, task->bytes);
327 block_copy_task_end(task, -ECANCELED);
328 g_free(task);
329 return -ECANCELED;
330 }
331
332 aio_task_pool_start_task(pool, &task->task);
333
334 return 0;
335}
336
beb5f545 337/*
e332a726
VSO
338 * block_copy_do_copy
339 *
dafaf135
VSO
340 * Do copy of cluster-aligned chunk. Requested region is allowed to exceed
341 * s->len only to cover last cluster when s->len is not aligned to clusters.
e332a726
VSO
342 *
343 * No sync here: nor bitmap neighter intersecting requests handling, only copy.
344 *
345 * Returns 0 on success.
beb5f545 346 */
e332a726 347static int coroutine_fn block_copy_do_copy(BlockCopyState *s,
8719091f 348 int64_t offset, int64_t bytes,
2d57511a 349 bool zeroes, bool *error_is_read)
beb5f545
VSO
350{
351 int ret;
8719091f 352 int64_t nbytes = MIN(offset + bytes, s->len) - offset;
e332a726 353 void *bounce_buffer = NULL;
beb5f545 354
8719091f
VSO
355 assert(offset >= 0 && bytes > 0 && INT64_MAX - offset >= bytes);
356 assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
dafaf135 357 assert(QEMU_IS_ALIGNED(bytes, s->cluster_size));
8719091f
VSO
358 assert(offset < s->len);
359 assert(offset + bytes <= s->len ||
360 offset + bytes == QEMU_ALIGN_UP(s->len, s->cluster_size));
dafaf135 361 assert(nbytes < INT_MAX);
e332a726 362
2d57511a 363 if (zeroes) {
8719091f 364 ret = bdrv_co_pwrite_zeroes(s->target, offset, nbytes, s->write_flags &
2d57511a
VSO
365 ~BDRV_REQ_WRITE_COMPRESSED);
366 if (ret < 0) {
8719091f 367 trace_block_copy_write_zeroes_fail(s, offset, ret);
d7eca542 368 *error_is_read = false;
2d57511a
VSO
369 }
370 return ret;
371 }
372
e332a726 373 if (s->use_copy_range) {
8719091f 374 ret = bdrv_co_copy_range(s->source, offset, s->target, offset, nbytes,
e332a726
VSO
375 0, s->write_flags);
376 if (ret < 0) {
8719091f 377 trace_block_copy_copy_range_fail(s, offset, ret);
e332a726 378 s->use_copy_range = false;
0e240245 379 s->copy_size = MAX(s->cluster_size, BLOCK_COPY_MAX_BUFFER);
e332a726
VSO
380 /* Fallback to read+write with allocated buffer */
381 } else {
9d31bc53
VSO
382 if (s->use_copy_range) {
383 /*
384 * Successful copy-range. Now increase copy_size. copy_range
385 * does not respect max_transfer (it's a TODO), so we factor
386 * that in here.
387 *
388 * Note: we double-check s->use_copy_range for the case when
389 * parallel block-copy request unsets it during previous
390 * bdrv_co_copy_range call.
391 */
392 s->copy_size =
393 MIN(MAX(s->cluster_size, BLOCK_COPY_MAX_COPY_RANGE),
394 QEMU_ALIGN_DOWN(block_copy_max_transfer(s->source,
395 s->target),
396 s->cluster_size));
397 }
e332a726
VSO
398 goto out;
399 }
400 }
401
0e240245
VSO
402 /*
403 * In case of failed copy_range request above, we may proceed with buffered
404 * request larger than BLOCK_COPY_MAX_BUFFER. Still, further requests will
9d31bc53
VSO
405 * be properly limited, so don't care too much. Moreover the most likely
406 * case (copy_range is unsupported for the configuration, so the very first
407 * copy_range request fails) is handled by setting large copy_size only
408 * after first successful copy_range.
0e240245
VSO
409 */
410
e332a726 411 bounce_buffer = qemu_blockalign(s->source->bs, nbytes);
beb5f545 412
8719091f 413 ret = bdrv_co_pread(s->source, offset, nbytes, bounce_buffer, 0);
beb5f545 414 if (ret < 0) {
8719091f 415 trace_block_copy_read_fail(s, offset, ret);
d7eca542 416 *error_is_read = true;
e332a726 417 goto out;
beb5f545
VSO
418 }
419
8719091f 420 ret = bdrv_co_pwrite(s->target, offset, nbytes, bounce_buffer,
00e30f05 421 s->write_flags);
beb5f545 422 if (ret < 0) {
8719091f 423 trace_block_copy_write_fail(s, offset, ret);
d7eca542 424 *error_is_read = false;
e332a726 425 goto out;
beb5f545
VSO
426 }
427
e332a726 428out:
3816edd2
VSO
429 qemu_vfree(bounce_buffer);
430
beb5f545 431 return ret;
beb5f545
VSO
432}
433
4ce5dd3e
VSO
434static coroutine_fn int block_copy_task_entry(AioTask *task)
435{
436 BlockCopyTask *t = container_of(task, BlockCopyTask, task);
c78dd00e 437 bool error_is_read = false;
4ce5dd3e
VSO
438 int ret;
439
440 ret = block_copy_do_copy(t->s, t->offset, t->bytes, t->zeroes,
441 &error_is_read);
8146b357
VSO
442 if (ret < 0) {
443 if (!t->call_state->ret) {
444 t->call_state->ret = ret;
445 t->call_state->error_is_read = error_is_read;
446 }
4ce5dd3e
VSO
447 } else {
448 progress_work_done(t->s->progress, t->bytes);
4ce5dd3e
VSO
449 }
450 co_put_to_shres(t->s->mem, t->bytes);
451 block_copy_task_end(t, ret);
452
453 return ret;
454}
455
2d57511a
VSO
456static int block_copy_block_status(BlockCopyState *s, int64_t offset,
457 int64_t bytes, int64_t *pnum)
458{
459 int64_t num;
460 BlockDriverState *base;
461 int ret;
462
c6f6d846
HR
463 if (s->skip_unallocated) {
464 base = bdrv_backing_chain_next(s->source->bs);
2d57511a
VSO
465 } else {
466 base = NULL;
467 }
468
469 ret = bdrv_block_status_above(s->source->bs, base, offset, bytes, &num,
470 NULL, NULL);
471 if (ret < 0 || num < s->cluster_size) {
472 /*
473 * On error or if failed to obtain large enough chunk just fallback to
474 * copy one cluster.
475 */
476 num = s->cluster_size;
477 ret = BDRV_BLOCK_ALLOCATED | BDRV_BLOCK_DATA;
478 } else if (offset + num == s->len) {
479 num = QEMU_ALIGN_UP(num, s->cluster_size);
480 } else {
481 num = QEMU_ALIGN_DOWN(num, s->cluster_size);
482 }
483
484 *pnum = num;
485 return ret;
486}
487
beb5f545
VSO
488/*
489 * Check if the cluster starting at offset is allocated or not.
490 * return via pnum the number of contiguous clusters sharing this allocation.
491 */
492static int block_copy_is_cluster_allocated(BlockCopyState *s, int64_t offset,
493 int64_t *pnum)
494{
00e30f05 495 BlockDriverState *bs = s->source->bs;
beb5f545
VSO
496 int64_t count, total_count = 0;
497 int64_t bytes = s->len - offset;
498 int ret;
499
500 assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
501
502 while (true) {
503 ret = bdrv_is_allocated(bs, offset, bytes, &count);
504 if (ret < 0) {
505 return ret;
506 }
507
508 total_count += count;
509
510 if (ret || count == 0) {
511 /*
512 * ret: partial segment(s) are considered allocated.
513 * otherwise: unallocated tail is treated as an entire segment.
514 */
515 *pnum = DIV_ROUND_UP(total_count, s->cluster_size);
516 return ret;
517 }
518
519 /* Unallocated segment(s) with uncertain following segment(s) */
520 if (total_count >= s->cluster_size) {
521 *pnum = total_count / s->cluster_size;
522 return 0;
523 }
524
525 offset += count;
526 bytes -= count;
527 }
528}
529
530/*
531 * Reset bits in copy_bitmap starting at offset if they represent unallocated
532 * data in the image. May reset subsequent contiguous bits.
533 * @return 0 when the cluster at @offset was unallocated,
534 * 1 otherwise, and -ret on error.
535 */
536int64_t block_copy_reset_unallocated(BlockCopyState *s,
537 int64_t offset, int64_t *count)
538{
539 int ret;
540 int64_t clusters, bytes;
541
542 ret = block_copy_is_cluster_allocated(s, offset, &clusters);
543 if (ret < 0) {
544 return ret;
545 }
546
547 bytes = clusters * s->cluster_size;
548
549 if (!ret) {
550 bdrv_reset_dirty_bitmap(s->copy_bitmap, offset, bytes);
d0ebeca1
VSO
551 progress_set_remaining(s->progress,
552 bdrv_get_dirty_count(s->copy_bitmap) +
553 s->in_flight_bytes);
beb5f545
VSO
554 }
555
556 *count = bytes;
557 return ret;
558}
559
5332e5d2
VSO
560/*
561 * block_copy_dirty_clusters
562 *
563 * Copy dirty clusters in @offset/@bytes range.
564 * Returns 1 if dirty clusters found and successfully copied, 0 if no dirty
565 * clusters found and -errno on failure.
566 */
3b8c2329
VSO
567static int coroutine_fn
568block_copy_dirty_clusters(BlockCopyCallState *call_state)
beb5f545 569{
3b8c2329
VSO
570 BlockCopyState *s = call_state->s;
571 int64_t offset = call_state->offset;
572 int64_t bytes = call_state->bytes;
573
beb5f545 574 int ret = 0;
5332e5d2 575 bool found_dirty = false;
42ac2144 576 int64_t end = offset + bytes;
4ce5dd3e 577 AioTaskPool *aio = NULL;
beb5f545
VSO
578
579 /*
580 * block_copy() user is responsible for keeping source and target in same
581 * aio context
582 */
00e30f05
VSO
583 assert(bdrv_get_aio_context(s->source->bs) ==
584 bdrv_get_aio_context(s->target->bs));
beb5f545 585
8719091f 586 assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
dafaf135 587 assert(QEMU_IS_ALIGNED(bytes, s->cluster_size));
beb5f545 588
a6d23d56 589 while (bytes && aio_task_pool_status(aio) == 0 && !call_state->cancelled) {
4ce5dd3e 590 BlockCopyTask *task;
42ac2144 591 int64_t status_bytes;
beb5f545 592
3b8c2329 593 task = block_copy_task_create(s, call_state, offset, bytes);
42ac2144
VSO
594 if (!task) {
595 /* No more dirty bits in the bitmap */
596 trace_block_copy_skip_range(s, offset, bytes);
597 break;
598 }
599 if (task->offset > offset) {
600 trace_block_copy_skip_range(s, offset, task->offset - offset);
beb5f545
VSO
601 }
602
5332e5d2
VSO
603 found_dirty = true;
604
42ac2144
VSO
605 ret = block_copy_block_status(s, task->offset, task->bytes,
606 &status_bytes);
5332e5d2 607 assert(ret >= 0); /* never fail */
42ac2144
VSO
608 if (status_bytes < task->bytes) {
609 block_copy_task_shrink(task, status_bytes);
610 }
2d57511a 611 if (s->skip_unallocated && !(ret & BDRV_BLOCK_ALLOCATED)) {
1348a657 612 block_copy_task_end(task, 0);
2d57511a
VSO
613 progress_set_remaining(s->progress,
614 bdrv_get_dirty_count(s->copy_bitmap) +
615 s->in_flight_bytes);
42ac2144
VSO
616 trace_block_copy_skip_range(s, task->offset, task->bytes);
617 offset = task_end(task);
618 bytes = end - offset;
fc9aefc8 619 g_free(task);
2d57511a 620 continue;
beb5f545 621 }
4ce5dd3e 622 task->zeroes = ret & BDRV_BLOCK_ZERO;
beb5f545 623
7e032df0
VSO
624 if (s->speed) {
625 if (!call_state->ignore_ratelimit) {
626 uint64_t ns = ratelimit_calculate_delay(&s->rate_limit, 0);
627 if (ns > 0) {
628 block_copy_task_end(task, -EAGAIN);
629 g_free(task);
29a6ea24
PB
630 qemu_co_sleep_ns_wakeable(&call_state->sleep,
631 QEMU_CLOCK_REALTIME, ns);
7e032df0
VSO
632 continue;
633 }
634 }
635
636 ratelimit_calculate_delay(&s->rate_limit, task->bytes);
637 }
638
42ac2144 639 trace_block_copy_process(s, task->offset);
beb5f545 640
42ac2144 641 co_get_from_shres(s->mem, task->bytes);
beb5f545 642
42ac2144
VSO
643 offset = task_end(task);
644 bytes = end - offset;
4ce5dd3e
VSO
645
646 if (!aio && bytes) {
26be9d62 647 aio = aio_task_pool_new(call_state->max_workers);
4ce5dd3e
VSO
648 }
649
650 ret = block_copy_task_run(aio, task);
651 if (ret < 0) {
652 goto out;
653 }
654 }
655
656out:
657 if (aio) {
658 aio_task_pool_wait_all(aio);
659
660 /*
661 * We are not really interested in -ECANCELED returned from
662 * block_copy_task_run. If it fails, it means some task already failed
663 * for real reason, let's return first failure.
664 * Still, assert that we don't rewrite failure by success.
e8de7ba9
VSO
665 *
666 * Note: ret may be positive here because of block-status result.
4ce5dd3e 667 */
e8de7ba9 668 assert(ret >= 0 || aio_task_pool_status(aio) < 0);
4ce5dd3e
VSO
669 ret = aio_task_pool_status(aio);
670
671 aio_task_pool_free(aio);
672 }
beb5f545 673
4ce5dd3e 674 return ret < 0 ? ret : found_dirty;
5332e5d2
VSO
675}
676
7e032df0
VSO
677void block_copy_kick(BlockCopyCallState *call_state)
678{
29a6ea24 679 qemu_co_sleep_wake(&call_state->sleep);
7e032df0
VSO
680}
681
5332e5d2 682/*
3b8c2329 683 * block_copy_common
5332e5d2
VSO
684 *
685 * Copy requested region, accordingly to dirty bitmap.
686 * Collaborate with parallel block_copy requests: if they succeed it will help
687 * us. If they fail, we will retry not-copied regions. So, if we return error,
688 * it means that some I/O operation failed in context of _this_ block_copy call,
689 * not some parallel operation.
690 */
3b8c2329 691static int coroutine_fn block_copy_common(BlockCopyCallState *call_state)
5332e5d2
VSO
692{
693 int ret;
694
2e099a9d
VSO
695 QLIST_INSERT_HEAD(&call_state->s->calls, call_state, list);
696
5332e5d2 697 do {
3b8c2329 698 ret = block_copy_dirty_clusters(call_state);
5332e5d2 699
a6d23d56 700 if (ret == 0 && !call_state->cancelled) {
3b8c2329
VSO
701 ret = block_copy_wait_one(call_state->s, call_state->offset,
702 call_state->bytes);
5332e5d2
VSO
703 }
704
705 /*
706 * We retry in two cases:
707 * 1. Some progress done
708 * Something was copied, which means that there were yield points
709 * and some new dirty bits may have appeared (due to failed parallel
710 * block-copy requests).
711 * 2. We have waited for some intersecting block-copy request
712 * It may have failed and produced new dirty bits.
713 */
a6d23d56 714 } while (ret > 0 && !call_state->cancelled);
a6ffe199 715
de4641b4
VSO
716 call_state->finished = true;
717
718 if (call_state->cb) {
719 call_state->cb(call_state->cb_opaque);
720 }
721
2e099a9d
VSO
722 QLIST_REMOVE(call_state, list);
723
beb5f545
VSO
724 return ret;
725}
397f4e9d 726
3b8c2329 727int coroutine_fn block_copy(BlockCopyState *s, int64_t start, int64_t bytes,
143a6384 728 bool ignore_ratelimit)
3b8c2329
VSO
729{
730 BlockCopyCallState call_state = {
731 .s = s,
732 .offset = start,
733 .bytes = bytes,
7e032df0 734 .ignore_ratelimit = ignore_ratelimit,
26be9d62 735 .max_workers = BLOCK_COPY_MAX_WORKERS,
3b8c2329
VSO
736 };
737
143a6384 738 return block_copy_common(&call_state);
3b8c2329
VSO
739}
740
de4641b4
VSO
741static void coroutine_fn block_copy_async_co_entry(void *opaque)
742{
743 block_copy_common(opaque);
744}
745
746BlockCopyCallState *block_copy_async(BlockCopyState *s,
747 int64_t offset, int64_t bytes,
26be9d62 748 int max_workers, int64_t max_chunk,
de4641b4
VSO
749 BlockCopyAsyncCallbackFunc cb,
750 void *cb_opaque)
751{
752 BlockCopyCallState *call_state = g_new(BlockCopyCallState, 1);
753
754 *call_state = (BlockCopyCallState) {
755 .s = s,
756 .offset = offset,
757 .bytes = bytes,
26be9d62
VSO
758 .max_workers = max_workers,
759 .max_chunk = max_chunk,
de4641b4
VSO
760 .cb = cb,
761 .cb_opaque = cb_opaque,
762
763 .co = qemu_coroutine_create(block_copy_async_co_entry, call_state),
764 };
765
766 qemu_coroutine_enter(call_state->co);
767
768 return call_state;
769}
770
771void block_copy_call_free(BlockCopyCallState *call_state)
772{
773 if (!call_state) {
774 return;
775 }
776
777 assert(call_state->finished);
778 g_free(call_state);
779}
780
781bool block_copy_call_finished(BlockCopyCallState *call_state)
782{
783 return call_state->finished;
784}
785
786bool block_copy_call_succeeded(BlockCopyCallState *call_state)
787{
a6d23d56
VSO
788 return call_state->finished && !call_state->cancelled &&
789 call_state->ret == 0;
de4641b4
VSO
790}
791
792bool block_copy_call_failed(BlockCopyCallState *call_state)
793{
a6d23d56
VSO
794 return call_state->finished && !call_state->cancelled &&
795 call_state->ret < 0;
796}
797
798bool block_copy_call_cancelled(BlockCopyCallState *call_state)
799{
800 return call_state->cancelled;
de4641b4
VSO
801}
802
803int block_copy_call_status(BlockCopyCallState *call_state, bool *error_is_read)
804{
805 assert(call_state->finished);
806 if (error_is_read) {
807 *error_is_read = call_state->error_is_read;
808 }
809 return call_state->ret;
810}
811
a6d23d56
VSO
812void block_copy_call_cancel(BlockCopyCallState *call_state)
813{
814 call_state->cancelled = true;
815 block_copy_kick(call_state);
816}
817
397f4e9d
VSO
818BdrvDirtyBitmap *block_copy_dirty_bitmap(BlockCopyState *s)
819{
820 return s->copy_bitmap;
821}
822
823void block_copy_set_skip_unallocated(BlockCopyState *s, bool skip)
824{
825 s->skip_unallocated = skip;
826}
7e032df0
VSO
827
828void block_copy_set_speed(BlockCopyState *s, uint64_t speed)
829{
830 s->speed = speed;
831 if (speed > 0) {
832 ratelimit_set_speed(&s->rate_limit, speed, BLOCK_COPY_SLICE_TIME);
833 }
834
835 /*
836 * Note: it's good to kick all call states from here, but it should be done
837 * only from a coroutine, to not crash if s->calls list changed while
838 * entering one call. So for now, the only user of this function kicks its
839 * only one call_state by hand.
840 */
841}