xref: /openbmc/qemu/block/block-copy.c (revision 4a84e854)
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 "block/reqlist.h"
21 #include "sysemu/block-backend.h"
22 #include "qemu/units.h"
23 #include "qemu/coroutine.h"
24 #include "block/aio_task.h"
25 #include "qemu/error-report.h"
26 #include "qemu/memalign.h"
27 
28 #define BLOCK_COPY_MAX_COPY_RANGE (16 * MiB)
29 #define BLOCK_COPY_MAX_BUFFER (1 * MiB)
30 #define BLOCK_COPY_MAX_MEM (128 * MiB)
31 #define BLOCK_COPY_MAX_WORKERS 64
32 #define BLOCK_COPY_SLICE_TIME 100000000ULL /* ns */
33 #define BLOCK_COPY_CLUSTER_SIZE_DEFAULT (1 << 16)
34 
35 typedef enum {
36     COPY_READ_WRITE_CLUSTER,
37     COPY_READ_WRITE,
38     COPY_WRITE_ZEROES,
39     COPY_RANGE_SMALL,
40     COPY_RANGE_FULL
41 } BlockCopyMethod;
42 
43 static coroutine_fn int block_copy_task_entry(AioTask *task);
44 
45 typedef struct BlockCopyCallState {
46     /* Fields initialized in block_copy_async() and never changed. */
47     BlockCopyState *s;
48     int64_t offset;
49     int64_t bytes;
50     int max_workers;
51     int64_t max_chunk;
52     bool ignore_ratelimit;
53     BlockCopyAsyncCallbackFunc cb;
54     void *cb_opaque;
55     /* Coroutine where async block-copy is running */
56     Coroutine *co;
57 
58     /* Fields whose state changes throughout the execution */
59     bool finished; /* atomic */
60     QemuCoSleep sleep; /* TODO: protect API with a lock */
61     bool cancelled; /* atomic */
62     /* To reference all call states from BlockCopyState */
63     QLIST_ENTRY(BlockCopyCallState) list;
64 
65     /*
66      * Fields that report information about return values and erros.
67      * Protected by lock in BlockCopyState.
68      */
69     bool error_is_read;
70     /*
71      * @ret is set concurrently by tasks under mutex. Only set once by first
72      * failed task (and untouched if no task failed).
73      * After finishing (call_state->finished is true), it is not modified
74      * anymore and may be safely read without mutex.
75      */
76     int ret;
77 } BlockCopyCallState;
78 
79 typedef struct BlockCopyTask {
80     AioTask task;
81 
82     /*
83      * Fields initialized in block_copy_task_create()
84      * and never changed.
85      */
86     BlockCopyState *s;
87     BlockCopyCallState *call_state;
88     /*
89      * @method can also be set again in the while loop of
90      * block_copy_dirty_clusters(), but it is never accessed concurrently
91      * because the only other function that reads it is
92      * block_copy_task_entry() and it is invoked afterwards in the same
93      * iteration.
94      */
95     BlockCopyMethod method;
96 
97     /*
98      * Generally, req is protected by lock in BlockCopyState, Still req.offset
99      * is only set on task creation, so may be read concurrently after creation.
100      * req.bytes is changed at most once, and need only protecting the case of
101      * parallel read while updating @bytes value in block_copy_task_shrink().
102      */
103     BlockReq req;
104 } BlockCopyTask;
105 
106 static int64_t task_end(BlockCopyTask *task)
107 {
108     return task->req.offset + task->req.bytes;
109 }
110 
111 typedef struct BlockCopyState {
112     /*
113      * BdrvChild objects are not owned or managed by block-copy. They are
114      * provided by block-copy user and user is responsible for appropriate
115      * permissions on these children.
116      */
117     BdrvChild *source;
118     BdrvChild *target;
119 
120     /*
121      * Fields initialized in block_copy_state_new()
122      * and never changed.
123      */
124     int64_t cluster_size;
125     int64_t max_transfer;
126     uint64_t len;
127     BdrvRequestFlags write_flags;
128 
129     /*
130      * Fields whose state changes throughout the execution
131      * Protected by lock.
132      */
133     CoMutex lock;
134     int64_t in_flight_bytes;
135     BlockCopyMethod method;
136     BlockReqList reqs;
137     QLIST_HEAD(, BlockCopyCallState) calls;
138     /*
139      * skip_unallocated:
140      *
141      * Used by sync=top jobs, which first scan the source node for unallocated
142      * areas and clear them in the copy_bitmap.  During this process, the bitmap
143      * is thus not fully initialized: It may still have bits set for areas that
144      * are unallocated and should actually not be copied.
145      *
146      * This is indicated by skip_unallocated.
147      *
148      * In this case, block_copy() will query the source’s allocation status,
149      * skip unallocated regions, clear them in the copy_bitmap, and invoke
150      * block_copy_reset_unallocated() every time it does.
151      */
152     bool skip_unallocated; /* atomic */
153     /* State fields that use a thread-safe API */
154     BdrvDirtyBitmap *copy_bitmap;
155     ProgressMeter *progress;
156     SharedResource *mem;
157     RateLimit rate_limit;
158 } BlockCopyState;
159 
160 /* Called with lock held */
161 static int64_t block_copy_chunk_size(BlockCopyState *s)
162 {
163     switch (s->method) {
164     case COPY_READ_WRITE_CLUSTER:
165         return s->cluster_size;
166     case COPY_READ_WRITE:
167     case COPY_RANGE_SMALL:
168         return MIN(MAX(s->cluster_size, BLOCK_COPY_MAX_BUFFER),
169                    s->max_transfer);
170     case COPY_RANGE_FULL:
171         return MIN(MAX(s->cluster_size, BLOCK_COPY_MAX_COPY_RANGE),
172                    s->max_transfer);
173     default:
174         /* Cannot have COPY_WRITE_ZEROES here.  */
175         abort();
176     }
177 }
178 
179 /*
180  * Search for the first dirty area in offset/bytes range and create task at
181  * the beginning of it.
182  */
183 static coroutine_fn BlockCopyTask *
184 block_copy_task_create(BlockCopyState *s, BlockCopyCallState *call_state,
185                        int64_t offset, int64_t bytes)
186 {
187     BlockCopyTask *task;
188     int64_t max_chunk;
189 
190     QEMU_LOCK_GUARD(&s->lock);
191     max_chunk = MIN_NON_ZERO(block_copy_chunk_size(s), call_state->max_chunk);
192     if (!bdrv_dirty_bitmap_next_dirty_area(s->copy_bitmap,
193                                            offset, offset + bytes,
194                                            max_chunk, &offset, &bytes))
195     {
196         return NULL;
197     }
198 
199     assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
200     bytes = QEMU_ALIGN_UP(bytes, s->cluster_size);
201 
202     /* region is dirty, so no existent tasks possible in it */
203     assert(!reqlist_find_conflict(&s->reqs, offset, bytes));
204 
205     bdrv_reset_dirty_bitmap(s->copy_bitmap, offset, bytes);
206     s->in_flight_bytes += bytes;
207 
208     task = g_new(BlockCopyTask, 1);
209     *task = (BlockCopyTask) {
210         .task.func = block_copy_task_entry,
211         .s = s,
212         .call_state = call_state,
213         .method = s->method,
214     };
215     reqlist_init_req(&s->reqs, &task->req, offset, bytes);
216 
217     return task;
218 }
219 
220 /*
221  * block_copy_task_shrink
222  *
223  * Drop the tail of the task to be handled later. Set dirty bits back and
224  * wake up all tasks waiting for us (may be some of them are not intersecting
225  * with shrunk task)
226  */
227 static void coroutine_fn block_copy_task_shrink(BlockCopyTask *task,
228                                                 int64_t new_bytes)
229 {
230     QEMU_LOCK_GUARD(&task->s->lock);
231     if (new_bytes == task->req.bytes) {
232         return;
233     }
234 
235     assert(new_bytes > 0 && new_bytes < task->req.bytes);
236 
237     task->s->in_flight_bytes -= task->req.bytes - new_bytes;
238     bdrv_set_dirty_bitmap(task->s->copy_bitmap,
239                           task->req.offset + new_bytes,
240                           task->req.bytes - new_bytes);
241 
242     reqlist_shrink_req(&task->req, new_bytes);
243 }
244 
245 static void coroutine_fn block_copy_task_end(BlockCopyTask *task, int ret)
246 {
247     QEMU_LOCK_GUARD(&task->s->lock);
248     task->s->in_flight_bytes -= task->req.bytes;
249     if (ret < 0) {
250         bdrv_set_dirty_bitmap(task->s->copy_bitmap, task->req.offset,
251                               task->req.bytes);
252     }
253     if (task->s->progress) {
254         progress_set_remaining(task->s->progress,
255                                bdrv_get_dirty_count(task->s->copy_bitmap) +
256                                task->s->in_flight_bytes);
257     }
258     reqlist_remove_req(&task->req);
259 }
260 
261 void block_copy_state_free(BlockCopyState *s)
262 {
263     if (!s) {
264         return;
265     }
266 
267     ratelimit_destroy(&s->rate_limit);
268     bdrv_release_dirty_bitmap(s->copy_bitmap);
269     shres_destroy(s->mem);
270     g_free(s);
271 }
272 
273 static uint32_t block_copy_max_transfer(BdrvChild *source, BdrvChild *target)
274 {
275     return MIN_NON_ZERO(INT_MAX,
276                         MIN_NON_ZERO(source->bs->bl.max_transfer,
277                                      target->bs->bl.max_transfer));
278 }
279 
280 void block_copy_set_copy_opts(BlockCopyState *s, bool use_copy_range,
281                               bool compress)
282 {
283     /* Keep BDRV_REQ_SERIALISING set (or not set) in block_copy_state_new() */
284     s->write_flags = (s->write_flags & BDRV_REQ_SERIALISING) |
285         (compress ? BDRV_REQ_WRITE_COMPRESSED : 0);
286 
287     if (s->max_transfer < s->cluster_size) {
288         /*
289          * copy_range does not respect max_transfer. We don't want to bother
290          * with requests smaller than block-copy cluster size, so fallback to
291          * buffered copying (read and write respect max_transfer on their
292          * behalf).
293          */
294         s->method = COPY_READ_WRITE_CLUSTER;
295     } else if (compress) {
296         /* Compression supports only cluster-size writes and no copy-range. */
297         s->method = COPY_READ_WRITE_CLUSTER;
298     } else {
299         /*
300          * If copy range enabled, start with COPY_RANGE_SMALL, until first
301          * successful copy_range (look at block_copy_do_copy).
302          */
303         s->method = use_copy_range ? COPY_RANGE_SMALL : COPY_READ_WRITE;
304     }
305 }
306 
307 static int64_t block_copy_calculate_cluster_size(BlockDriverState *target,
308                                                  Error **errp)
309 {
310     int ret;
311     BlockDriverInfo bdi;
312     bool target_does_cow = bdrv_backing_chain_next(target);
313 
314     /*
315      * If there is no backing file on the target, we cannot rely on COW if our
316      * backup cluster size is smaller than the target cluster size. Even for
317      * targets with a backing file, try to avoid COW if possible.
318      */
319     ret = bdrv_get_info(target, &bdi);
320     if (ret == -ENOTSUP && !target_does_cow) {
321         /* Cluster size is not defined */
322         warn_report("The target block device doesn't provide "
323                     "information about the block size and it doesn't have a "
324                     "backing file. The default block size of %u bytes is "
325                     "used. If the actual block size of the target exceeds "
326                     "this default, the backup may be unusable",
327                     BLOCK_COPY_CLUSTER_SIZE_DEFAULT);
328         return BLOCK_COPY_CLUSTER_SIZE_DEFAULT;
329     } else if (ret < 0 && !target_does_cow) {
330         error_setg_errno(errp, -ret,
331             "Couldn't determine the cluster size of the target image, "
332             "which has no backing file");
333         error_append_hint(errp,
334             "Aborting, since this may create an unusable destination image\n");
335         return ret;
336     } else if (ret < 0 && target_does_cow) {
337         /* Not fatal; just trudge on ahead. */
338         return BLOCK_COPY_CLUSTER_SIZE_DEFAULT;
339     }
340 
341     return MAX(BLOCK_COPY_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
342 }
343 
344 BlockCopyState *block_copy_state_new(BdrvChild *source, BdrvChild *target,
345                                      const BdrvDirtyBitmap *bitmap,
346                                      Error **errp)
347 {
348     ERRP_GUARD();
349     BlockCopyState *s;
350     int64_t cluster_size;
351     BdrvDirtyBitmap *copy_bitmap;
352     bool is_fleecing;
353 
354     cluster_size = block_copy_calculate_cluster_size(target->bs, errp);
355     if (cluster_size < 0) {
356         return NULL;
357     }
358 
359     copy_bitmap = bdrv_create_dirty_bitmap(source->bs, cluster_size, NULL,
360                                            errp);
361     if (!copy_bitmap) {
362         return NULL;
363     }
364     bdrv_disable_dirty_bitmap(copy_bitmap);
365     if (bitmap) {
366         if (!bdrv_merge_dirty_bitmap(copy_bitmap, bitmap, NULL, errp)) {
367             error_prepend(errp, "Failed to merge bitmap '%s' to internal "
368                           "copy-bitmap: ", bdrv_dirty_bitmap_name(bitmap));
369             bdrv_release_dirty_bitmap(copy_bitmap);
370             return NULL;
371         }
372     } else {
373         bdrv_set_dirty_bitmap(copy_bitmap, 0,
374                               bdrv_dirty_bitmap_size(copy_bitmap));
375     }
376 
377     /*
378      * If source is in backing chain of target assume that target is going to be
379      * used for "image fleecing", i.e. it should represent a kind of snapshot of
380      * source at backup-start point in time. And target is going to be read by
381      * somebody (for example, used as NBD export) during backup job.
382      *
383      * In this case, we need to add BDRV_REQ_SERIALISING write flag to avoid
384      * intersection of backup writes and third party reads from target,
385      * otherwise reading from target we may occasionally read already updated by
386      * guest data.
387      *
388      * For more information see commit f8d59dfb40bb and test
389      * tests/qemu-iotests/222
390      */
391     is_fleecing = bdrv_chain_contains(target->bs, source->bs);
392 
393     s = g_new(BlockCopyState, 1);
394     *s = (BlockCopyState) {
395         .source = source,
396         .target = target,
397         .copy_bitmap = copy_bitmap,
398         .cluster_size = cluster_size,
399         .len = bdrv_dirty_bitmap_size(copy_bitmap),
400         .write_flags = (is_fleecing ? BDRV_REQ_SERIALISING : 0),
401         .mem = shres_create(BLOCK_COPY_MAX_MEM),
402         .max_transfer = QEMU_ALIGN_DOWN(
403                                     block_copy_max_transfer(source, target),
404                                     cluster_size),
405     };
406 
407     block_copy_set_copy_opts(s, false, false);
408 
409     ratelimit_init(&s->rate_limit);
410     qemu_co_mutex_init(&s->lock);
411     QLIST_INIT(&s->reqs);
412     QLIST_INIT(&s->calls);
413 
414     return s;
415 }
416 
417 /* Only set before running the job, no need for locking. */
418 void block_copy_set_progress_meter(BlockCopyState *s, ProgressMeter *pm)
419 {
420     s->progress = pm;
421 }
422 
423 /*
424  * Takes ownership of @task
425  *
426  * If pool is NULL directly run the task, otherwise schedule it into the pool.
427  *
428  * Returns: task.func return code if pool is NULL
429  *          otherwise -ECANCELED if pool status is bad
430  *          otherwise 0 (successfully scheduled)
431  */
432 static coroutine_fn int block_copy_task_run(AioTaskPool *pool,
433                                             BlockCopyTask *task)
434 {
435     if (!pool) {
436         int ret = task->task.func(&task->task);
437 
438         g_free(task);
439         return ret;
440     }
441 
442     aio_task_pool_wait_slot(pool);
443     if (aio_task_pool_status(pool) < 0) {
444         co_put_to_shres(task->s->mem, task->req.bytes);
445         block_copy_task_end(task, -ECANCELED);
446         g_free(task);
447         return -ECANCELED;
448     }
449 
450     aio_task_pool_start_task(pool, &task->task);
451 
452     return 0;
453 }
454 
455 /*
456  * block_copy_do_copy
457  *
458  * Do copy of cluster-aligned chunk. Requested region is allowed to exceed
459  * s->len only to cover last cluster when s->len is not aligned to clusters.
460  *
461  * No sync here: nor bitmap neighter intersecting requests handling, only copy.
462  *
463  * @method is an in-out argument, so that copy_range can be either extended to
464  * a full-size buffer or disabled if the copy_range attempt fails.  The output
465  * value of @method should be used for subsequent tasks.
466  * Returns 0 on success.
467  */
468 static int coroutine_fn block_copy_do_copy(BlockCopyState *s,
469                                            int64_t offset, int64_t bytes,
470                                            BlockCopyMethod *method,
471                                            bool *error_is_read)
472 {
473     int ret;
474     int64_t nbytes = MIN(offset + bytes, s->len) - offset;
475     void *bounce_buffer = NULL;
476 
477     assert(offset >= 0 && bytes > 0 && INT64_MAX - offset >= bytes);
478     assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
479     assert(QEMU_IS_ALIGNED(bytes, s->cluster_size));
480     assert(offset < s->len);
481     assert(offset + bytes <= s->len ||
482            offset + bytes == QEMU_ALIGN_UP(s->len, s->cluster_size));
483     assert(nbytes < INT_MAX);
484 
485     switch (*method) {
486     case COPY_WRITE_ZEROES:
487         ret = bdrv_co_pwrite_zeroes(s->target, offset, nbytes, s->write_flags &
488                                     ~BDRV_REQ_WRITE_COMPRESSED);
489         if (ret < 0) {
490             trace_block_copy_write_zeroes_fail(s, offset, ret);
491             *error_is_read = false;
492         }
493         return ret;
494 
495     case COPY_RANGE_SMALL:
496     case COPY_RANGE_FULL:
497         ret = bdrv_co_copy_range(s->source, offset, s->target, offset, nbytes,
498                                  0, s->write_flags);
499         if (ret >= 0) {
500             /* Successful copy-range, increase chunk size.  */
501             *method = COPY_RANGE_FULL;
502             return 0;
503         }
504 
505         trace_block_copy_copy_range_fail(s, offset, ret);
506         *method = COPY_READ_WRITE;
507         /* Fall through to read+write with allocated buffer */
508 
509     case COPY_READ_WRITE_CLUSTER:
510     case COPY_READ_WRITE:
511         /*
512          * In case of failed copy_range request above, we may proceed with
513          * buffered request larger than BLOCK_COPY_MAX_BUFFER.
514          * Still, further requests will be properly limited, so don't care too
515          * much. Moreover the most likely case (copy_range is unsupported for
516          * the configuration, so the very first copy_range request fails)
517          * is handled by setting large copy_size only after first successful
518          * copy_range.
519          */
520 
521         bounce_buffer = qemu_blockalign(s->source->bs, nbytes);
522 
523         ret = bdrv_co_pread(s->source, offset, nbytes, bounce_buffer, 0);
524         if (ret < 0) {
525             trace_block_copy_read_fail(s, offset, ret);
526             *error_is_read = true;
527             goto out;
528         }
529 
530         ret = bdrv_co_pwrite(s->target, offset, nbytes, bounce_buffer,
531                              s->write_flags);
532         if (ret < 0) {
533             trace_block_copy_write_fail(s, offset, ret);
534             *error_is_read = false;
535             goto out;
536         }
537 
538     out:
539         qemu_vfree(bounce_buffer);
540         break;
541 
542     default:
543         abort();
544     }
545 
546     return ret;
547 }
548 
549 static coroutine_fn int block_copy_task_entry(AioTask *task)
550 {
551     BlockCopyTask *t = container_of(task, BlockCopyTask, task);
552     BlockCopyState *s = t->s;
553     bool error_is_read = false;
554     BlockCopyMethod method = t->method;
555     int ret;
556 
557     ret = block_copy_do_copy(s, t->req.offset, t->req.bytes, &method,
558                              &error_is_read);
559 
560     WITH_QEMU_LOCK_GUARD(&s->lock) {
561         if (s->method == t->method) {
562             s->method = method;
563         }
564 
565         if (ret < 0) {
566             if (!t->call_state->ret) {
567                 t->call_state->ret = ret;
568                 t->call_state->error_is_read = error_is_read;
569             }
570         } else if (s->progress) {
571             progress_work_done(s->progress, t->req.bytes);
572         }
573     }
574     co_put_to_shres(s->mem, t->req.bytes);
575     block_copy_task_end(t, ret);
576 
577     return ret;
578 }
579 
580 static int block_copy_block_status(BlockCopyState *s, int64_t offset,
581                                    int64_t bytes, int64_t *pnum)
582 {
583     int64_t num;
584     BlockDriverState *base;
585     int ret;
586 
587     if (qatomic_read(&s->skip_unallocated)) {
588         base = bdrv_backing_chain_next(s->source->bs);
589     } else {
590         base = NULL;
591     }
592 
593     ret = bdrv_block_status_above(s->source->bs, base, offset, bytes, &num,
594                                   NULL, NULL);
595     if (ret < 0 || num < s->cluster_size) {
596         /*
597          * On error or if failed to obtain large enough chunk just fallback to
598          * copy one cluster.
599          */
600         num = s->cluster_size;
601         ret = BDRV_BLOCK_ALLOCATED | BDRV_BLOCK_DATA;
602     } else if (offset + num == s->len) {
603         num = QEMU_ALIGN_UP(num, s->cluster_size);
604     } else {
605         num = QEMU_ALIGN_DOWN(num, s->cluster_size);
606     }
607 
608     *pnum = num;
609     return ret;
610 }
611 
612 /*
613  * Check if the cluster starting at offset is allocated or not.
614  * return via pnum the number of contiguous clusters sharing this allocation.
615  */
616 static int block_copy_is_cluster_allocated(BlockCopyState *s, int64_t offset,
617                                            int64_t *pnum)
618 {
619     BlockDriverState *bs = s->source->bs;
620     int64_t count, total_count = 0;
621     int64_t bytes = s->len - offset;
622     int ret;
623 
624     assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
625 
626     while (true) {
627         ret = bdrv_is_allocated(bs, offset, bytes, &count);
628         if (ret < 0) {
629             return ret;
630         }
631 
632         total_count += count;
633 
634         if (ret || count == 0) {
635             /*
636              * ret: partial segment(s) are considered allocated.
637              * otherwise: unallocated tail is treated as an entire segment.
638              */
639             *pnum = DIV_ROUND_UP(total_count, s->cluster_size);
640             return ret;
641         }
642 
643         /* Unallocated segment(s) with uncertain following segment(s) */
644         if (total_count >= s->cluster_size) {
645             *pnum = total_count / s->cluster_size;
646             return 0;
647         }
648 
649         offset += count;
650         bytes -= count;
651     }
652 }
653 
654 void block_copy_reset(BlockCopyState *s, int64_t offset, int64_t bytes)
655 {
656     QEMU_LOCK_GUARD(&s->lock);
657 
658     bdrv_reset_dirty_bitmap(s->copy_bitmap, offset, bytes);
659     if (s->progress) {
660         progress_set_remaining(s->progress,
661                                bdrv_get_dirty_count(s->copy_bitmap) +
662                                s->in_flight_bytes);
663     }
664 }
665 
666 /*
667  * Reset bits in copy_bitmap starting at offset if they represent unallocated
668  * data in the image. May reset subsequent contiguous bits.
669  * @return 0 when the cluster at @offset was unallocated,
670  *         1 otherwise, and -ret on error.
671  */
672 int64_t block_copy_reset_unallocated(BlockCopyState *s,
673                                      int64_t offset, int64_t *count)
674 {
675     int ret;
676     int64_t clusters, bytes;
677 
678     ret = block_copy_is_cluster_allocated(s, offset, &clusters);
679     if (ret < 0) {
680         return ret;
681     }
682 
683     bytes = clusters * s->cluster_size;
684 
685     if (!ret) {
686         block_copy_reset(s, offset, bytes);
687     }
688 
689     *count = bytes;
690     return ret;
691 }
692 
693 /*
694  * block_copy_dirty_clusters
695  *
696  * Copy dirty clusters in @offset/@bytes range.
697  * Returns 1 if dirty clusters found and successfully copied, 0 if no dirty
698  * clusters found and -errno on failure.
699  */
700 static int coroutine_fn
701 block_copy_dirty_clusters(BlockCopyCallState *call_state)
702 {
703     BlockCopyState *s = call_state->s;
704     int64_t offset = call_state->offset;
705     int64_t bytes = call_state->bytes;
706 
707     int ret = 0;
708     bool found_dirty = false;
709     int64_t end = offset + bytes;
710     AioTaskPool *aio = NULL;
711 
712     /*
713      * block_copy() user is responsible for keeping source and target in same
714      * aio context
715      */
716     assert(bdrv_get_aio_context(s->source->bs) ==
717            bdrv_get_aio_context(s->target->bs));
718 
719     assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
720     assert(QEMU_IS_ALIGNED(bytes, s->cluster_size));
721 
722     while (bytes && aio_task_pool_status(aio) == 0 &&
723            !qatomic_read(&call_state->cancelled)) {
724         BlockCopyTask *task;
725         int64_t status_bytes;
726 
727         task = block_copy_task_create(s, call_state, offset, bytes);
728         if (!task) {
729             /* No more dirty bits in the bitmap */
730             trace_block_copy_skip_range(s, offset, bytes);
731             break;
732         }
733         if (task->req.offset > offset) {
734             trace_block_copy_skip_range(s, offset, task->req.offset - offset);
735         }
736 
737         found_dirty = true;
738 
739         ret = block_copy_block_status(s, task->req.offset, task->req.bytes,
740                                       &status_bytes);
741         assert(ret >= 0); /* never fail */
742         if (status_bytes < task->req.bytes) {
743             block_copy_task_shrink(task, status_bytes);
744         }
745         if (qatomic_read(&s->skip_unallocated) &&
746             !(ret & BDRV_BLOCK_ALLOCATED)) {
747             block_copy_task_end(task, 0);
748             trace_block_copy_skip_range(s, task->req.offset, task->req.bytes);
749             offset = task_end(task);
750             bytes = end - offset;
751             g_free(task);
752             continue;
753         }
754         if (ret & BDRV_BLOCK_ZERO) {
755             task->method = COPY_WRITE_ZEROES;
756         }
757 
758         if (!call_state->ignore_ratelimit) {
759             uint64_t ns = ratelimit_calculate_delay(&s->rate_limit, 0);
760             if (ns > 0) {
761                 block_copy_task_end(task, -EAGAIN);
762                 g_free(task);
763                 qemu_co_sleep_ns_wakeable(&call_state->sleep,
764                                           QEMU_CLOCK_REALTIME, ns);
765                 continue;
766             }
767         }
768 
769         ratelimit_calculate_delay(&s->rate_limit, task->req.bytes);
770 
771         trace_block_copy_process(s, task->req.offset);
772 
773         co_get_from_shres(s->mem, task->req.bytes);
774 
775         offset = task_end(task);
776         bytes = end - offset;
777 
778         if (!aio && bytes) {
779             aio = aio_task_pool_new(call_state->max_workers);
780         }
781 
782         ret = block_copy_task_run(aio, task);
783         if (ret < 0) {
784             goto out;
785         }
786     }
787 
788 out:
789     if (aio) {
790         aio_task_pool_wait_all(aio);
791 
792         /*
793          * We are not really interested in -ECANCELED returned from
794          * block_copy_task_run. If it fails, it means some task already failed
795          * for real reason, let's return first failure.
796          * Still, assert that we don't rewrite failure by success.
797          *
798          * Note: ret may be positive here because of block-status result.
799          */
800         assert(ret >= 0 || aio_task_pool_status(aio) < 0);
801         ret = aio_task_pool_status(aio);
802 
803         aio_task_pool_free(aio);
804     }
805 
806     return ret < 0 ? ret : found_dirty;
807 }
808 
809 void block_copy_kick(BlockCopyCallState *call_state)
810 {
811     qemu_co_sleep_wake(&call_state->sleep);
812 }
813 
814 /*
815  * block_copy_common
816  *
817  * Copy requested region, accordingly to dirty bitmap.
818  * Collaborate with parallel block_copy requests: if they succeed it will help
819  * us. If they fail, we will retry not-copied regions. So, if we return error,
820  * it means that some I/O operation failed in context of _this_ block_copy call,
821  * not some parallel operation.
822  */
823 static int coroutine_fn block_copy_common(BlockCopyCallState *call_state)
824 {
825     int ret;
826     BlockCopyState *s = call_state->s;
827 
828     qemu_co_mutex_lock(&s->lock);
829     QLIST_INSERT_HEAD(&s->calls, call_state, list);
830     qemu_co_mutex_unlock(&s->lock);
831 
832     do {
833         ret = block_copy_dirty_clusters(call_state);
834 
835         if (ret == 0 && !qatomic_read(&call_state->cancelled)) {
836             WITH_QEMU_LOCK_GUARD(&s->lock) {
837                 /*
838                  * Check that there is no task we still need to
839                  * wait to complete
840                  */
841                 ret = reqlist_wait_one(&s->reqs, call_state->offset,
842                                        call_state->bytes, &s->lock);
843                 if (ret == 0) {
844                     /*
845                      * No pending tasks, but check again the bitmap in this
846                      * same critical section, since a task might have failed
847                      * between this and the critical section in
848                      * block_copy_dirty_clusters().
849                      *
850                      * reqlist_wait_one return value 0 also means that it
851                      * didn't release the lock. So, we are still in the same
852                      * critical section, not interrupted by any concurrent
853                      * access to state.
854                      */
855                     ret = bdrv_dirty_bitmap_next_dirty(s->copy_bitmap,
856                                                        call_state->offset,
857                                                        call_state->bytes) >= 0;
858                 }
859             }
860         }
861 
862         /*
863          * We retry in two cases:
864          * 1. Some progress done
865          *    Something was copied, which means that there were yield points
866          *    and some new dirty bits may have appeared (due to failed parallel
867          *    block-copy requests).
868          * 2. We have waited for some intersecting block-copy request
869          *    It may have failed and produced new dirty bits.
870          */
871     } while (ret > 0 && !qatomic_read(&call_state->cancelled));
872 
873     qatomic_store_release(&call_state->finished, true);
874 
875     if (call_state->cb) {
876         call_state->cb(call_state->cb_opaque);
877     }
878 
879     qemu_co_mutex_lock(&s->lock);
880     QLIST_REMOVE(call_state, list);
881     qemu_co_mutex_unlock(&s->lock);
882 
883     return ret;
884 }
885 
886 static void coroutine_fn block_copy_async_co_entry(void *opaque)
887 {
888     block_copy_common(opaque);
889 }
890 
891 int coroutine_fn block_copy(BlockCopyState *s, int64_t start, int64_t bytes,
892                             bool ignore_ratelimit, uint64_t timeout_ns,
893                             BlockCopyAsyncCallbackFunc cb,
894                             void *cb_opaque)
895 {
896     int ret;
897     BlockCopyCallState *call_state = g_new(BlockCopyCallState, 1);
898 
899     *call_state = (BlockCopyCallState) {
900         .s = s,
901         .offset = start,
902         .bytes = bytes,
903         .ignore_ratelimit = ignore_ratelimit,
904         .max_workers = BLOCK_COPY_MAX_WORKERS,
905         .cb = cb,
906         .cb_opaque = cb_opaque,
907     };
908 
909     ret = qemu_co_timeout(block_copy_async_co_entry, call_state, timeout_ns,
910                           g_free);
911     if (ret < 0) {
912         assert(ret == -ETIMEDOUT);
913         block_copy_call_cancel(call_state);
914         /* call_state will be freed by running coroutine. */
915         return ret;
916     }
917 
918     ret = call_state->ret;
919     g_free(call_state);
920 
921     return ret;
922 }
923 
924 BlockCopyCallState *block_copy_async(BlockCopyState *s,
925                                      int64_t offset, int64_t bytes,
926                                      int max_workers, int64_t max_chunk,
927                                      BlockCopyAsyncCallbackFunc cb,
928                                      void *cb_opaque)
929 {
930     BlockCopyCallState *call_state = g_new(BlockCopyCallState, 1);
931 
932     *call_state = (BlockCopyCallState) {
933         .s = s,
934         .offset = offset,
935         .bytes = bytes,
936         .max_workers = max_workers,
937         .max_chunk = max_chunk,
938         .cb = cb,
939         .cb_opaque = cb_opaque,
940 
941         .co = qemu_coroutine_create(block_copy_async_co_entry, call_state),
942     };
943 
944     qemu_coroutine_enter(call_state->co);
945 
946     return call_state;
947 }
948 
949 void block_copy_call_free(BlockCopyCallState *call_state)
950 {
951     if (!call_state) {
952         return;
953     }
954 
955     assert(qatomic_read(&call_state->finished));
956     g_free(call_state);
957 }
958 
959 bool block_copy_call_finished(BlockCopyCallState *call_state)
960 {
961     return qatomic_read(&call_state->finished);
962 }
963 
964 bool block_copy_call_succeeded(BlockCopyCallState *call_state)
965 {
966     return qatomic_load_acquire(&call_state->finished) &&
967            !qatomic_read(&call_state->cancelled) &&
968            call_state->ret == 0;
969 }
970 
971 bool block_copy_call_failed(BlockCopyCallState *call_state)
972 {
973     return qatomic_load_acquire(&call_state->finished) &&
974            !qatomic_read(&call_state->cancelled) &&
975            call_state->ret < 0;
976 }
977 
978 bool block_copy_call_cancelled(BlockCopyCallState *call_state)
979 {
980     return qatomic_read(&call_state->cancelled);
981 }
982 
983 int block_copy_call_status(BlockCopyCallState *call_state, bool *error_is_read)
984 {
985     assert(qatomic_load_acquire(&call_state->finished));
986     if (error_is_read) {
987         *error_is_read = call_state->error_is_read;
988     }
989     return call_state->ret;
990 }
991 
992 /*
993  * Note that cancelling and finishing are racy.
994  * User can cancel a block-copy that is already finished.
995  */
996 void block_copy_call_cancel(BlockCopyCallState *call_state)
997 {
998     qatomic_set(&call_state->cancelled, true);
999     block_copy_kick(call_state);
1000 }
1001 
1002 BdrvDirtyBitmap *block_copy_dirty_bitmap(BlockCopyState *s)
1003 {
1004     return s->copy_bitmap;
1005 }
1006 
1007 int64_t block_copy_cluster_size(BlockCopyState *s)
1008 {
1009     return s->cluster_size;
1010 }
1011 
1012 void block_copy_set_skip_unallocated(BlockCopyState *s, bool skip)
1013 {
1014     qatomic_set(&s->skip_unallocated, skip);
1015 }
1016 
1017 void block_copy_set_speed(BlockCopyState *s, uint64_t speed)
1018 {
1019     ratelimit_set_speed(&s->rate_limit, speed, BLOCK_COPY_SLICE_TIME);
1020 
1021     /*
1022      * Note: it's good to kick all call states from here, but it should be done
1023      * only from a coroutine, to not crash if s->calls list changed while
1024      * entering one call. So for now, the only user of this function kicks its
1025      * only one call_state by hand.
1026      */
1027 }
1028