4 * Copyright (C) 2013 Proxmox Server Solutions
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
14 #include "qemu/osdep.h"
17 #include "block/block.h"
18 #include "block/block_int.h"
19 #include "block/blockjob_int.h"
20 #include "block/block_backup.h"
21 #include "qapi/error.h"
22 #include "qapi/qmp/qerror.h"
23 #include "qemu/ratelimit.h"
24 #include "qemu/cutils.h"
25 #include "sysemu/block-backend.h"
26 #include "qemu/bitmap.h"
27 #include "qemu/error-report.h"
29 #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16)
31 typedef struct CowRequest {
34 QLIST_ENTRY(CowRequest) list;
35 CoQueue wait_queue; /* coroutines blocked on this request */
38 typedef struct BackupBlockJob {
41 BdrvDirtyBitmap *sync_bitmap;
42 MirrorSyncMode sync_mode;
43 BitmapSyncMode bitmap_mode;
44 BlockdevOnError on_source_error;
45 BlockdevOnError on_target_error;
46 CoRwlock flush_rwlock;
51 NotifierWithReturn before_write;
52 QLIST_HEAD(, CowRequest) inflight_reqs;
56 int64_t copy_range_size;
58 bool serialize_target_writes;
61 static const BlockJobDriver backup_job_driver;
63 /* See if in-flight requests overlap and wait for them to complete */
64 static void coroutine_fn wait_for_overlapping_requests(BackupBlockJob *job,
73 QLIST_FOREACH(req, &job->inflight_reqs, list) {
74 if (end > req->start_byte && start < req->end_byte) {
75 qemu_co_queue_wait(&req->wait_queue, NULL);
83 /* Keep track of an in-flight request */
84 static void cow_request_begin(CowRequest *req, BackupBlockJob *job,
85 int64_t start, int64_t end)
87 req->start_byte = start;
89 qemu_co_queue_init(&req->wait_queue);
90 QLIST_INSERT_HEAD(&job->inflight_reqs, req, list);
93 /* Forget about a completed request */
94 static void cow_request_end(CowRequest *req)
96 QLIST_REMOVE(req, list);
97 qemu_co_queue_restart_all(&req->wait_queue);
100 /* Copy range to target with a bounce buffer and return the bytes copied. If
101 * error occurred, return a negative error number */
102 static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job,
105 bool is_write_notifier,
107 void **bounce_buffer)
110 BlockBackend *blk = job->common.blk;
112 int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
113 int write_flags = job->serialize_target_writes ? BDRV_REQ_SERIALISING : 0;
115 assert(QEMU_IS_ALIGNED(start, job->cluster_size));
116 hbitmap_reset(job->copy_bitmap, start, job->cluster_size);
117 nbytes = MIN(job->cluster_size, job->len - start);
118 if (!*bounce_buffer) {
119 *bounce_buffer = blk_blockalign(blk, job->cluster_size);
122 ret = blk_co_pread(blk, start, nbytes, *bounce_buffer, read_flags);
124 trace_backup_do_cow_read_fail(job, start, ret);
126 *error_is_read = true;
131 if (buffer_is_zero(*bounce_buffer, nbytes)) {
132 ret = blk_co_pwrite_zeroes(job->target, start,
133 nbytes, write_flags | BDRV_REQ_MAY_UNMAP);
135 ret = blk_co_pwrite(job->target, start,
136 nbytes, *bounce_buffer, write_flags |
137 (job->compress ? BDRV_REQ_WRITE_COMPRESSED : 0));
140 trace_backup_do_cow_write_fail(job, start, ret);
142 *error_is_read = false;
149 hbitmap_set(job->copy_bitmap, start, job->cluster_size);
154 /* Copy range to target and return the bytes copied. If error occurred, return a
155 * negative error number. */
156 static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job,
159 bool is_write_notifier)
163 BlockBackend *blk = job->common.blk;
165 int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
166 int write_flags = job->serialize_target_writes ? BDRV_REQ_SERIALISING : 0;
168 assert(QEMU_IS_ALIGNED(job->copy_range_size, job->cluster_size));
169 assert(QEMU_IS_ALIGNED(start, job->cluster_size));
170 nbytes = MIN(job->copy_range_size, end - start);
171 nr_clusters = DIV_ROUND_UP(nbytes, job->cluster_size);
172 hbitmap_reset(job->copy_bitmap, start, job->cluster_size * nr_clusters);
173 ret = blk_co_copy_range(blk, start, job->target, start, nbytes,
174 read_flags, write_flags);
176 trace_backup_do_cow_copy_range_fail(job, start, ret);
177 hbitmap_set(job->copy_bitmap, start, job->cluster_size * nr_clusters);
184 static int coroutine_fn backup_do_cow(BackupBlockJob *job,
185 int64_t offset, uint64_t bytes,
187 bool is_write_notifier)
189 CowRequest cow_request;
191 int64_t start, end; /* bytes */
192 void *bounce_buffer = NULL;
194 qemu_co_rwlock_rdlock(&job->flush_rwlock);
196 start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
197 end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
199 trace_backup_do_cow_enter(job, start, offset, bytes);
201 wait_for_overlapping_requests(job, start, end);
202 cow_request_begin(&cow_request, job, start, end);
204 while (start < end) {
207 if (!hbitmap_get(job->copy_bitmap, start)) {
208 trace_backup_do_cow_skip(job, start);
209 start += job->cluster_size;
210 continue; /* already copied */
213 dirty_end = hbitmap_next_zero(job->copy_bitmap, start, (end - start));
218 trace_backup_do_cow_process(job, start);
220 if (job->use_copy_range) {
221 ret = backup_cow_with_offload(job, start, dirty_end,
224 job->use_copy_range = false;
227 if (!job->use_copy_range) {
228 ret = backup_cow_with_bounce_buffer(job, start, dirty_end,
230 error_is_read, &bounce_buffer);
236 /* Publish progress, guest I/O counts as progress too. Note that the
237 * offset field is an opaque progress value, it is not a disk offset.
240 job->bytes_read += ret;
241 job_progress_update(&job->common.job, ret);
246 qemu_vfree(bounce_buffer);
249 cow_request_end(&cow_request);
251 trace_backup_do_cow_return(job, offset, bytes, ret);
253 qemu_co_rwlock_unlock(&job->flush_rwlock);
258 static int coroutine_fn backup_before_write_notify(
259 NotifierWithReturn *notifier,
262 BackupBlockJob *job = container_of(notifier, BackupBlockJob, before_write);
263 BdrvTrackedRequest *req = opaque;
265 assert(req->bs == blk_bs(job->common.blk));
266 assert(QEMU_IS_ALIGNED(req->offset, BDRV_SECTOR_SIZE));
267 assert(QEMU_IS_ALIGNED(req->bytes, BDRV_SECTOR_SIZE));
269 return backup_do_cow(job, req->offset, req->bytes, NULL, true);
272 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
275 BlockDriverState *bs = blk_bs(job->common.blk);
277 if (ret < 0 || job->bitmap_mode == BITMAP_SYNC_MODE_NEVER) {
279 * Failure, or we don't want to synchronize the bitmap.
280 * Merge the successor back into the parent, delete nothing.
282 bm = bdrv_reclaim_dirty_bitmap(bs, job->sync_bitmap, NULL);
285 /* Everything is fine, delete this bitmap and install the backup. */
286 bm = bdrv_dirty_bitmap_abdicate(bs, job->sync_bitmap, NULL);
291 static void backup_commit(Job *job)
293 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
294 if (s->sync_bitmap) {
295 backup_cleanup_sync_bitmap(s, 0);
299 static void backup_abort(Job *job)
301 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
302 if (s->sync_bitmap) {
303 backup_cleanup_sync_bitmap(s, -1);
307 static void backup_clean(Job *job)
309 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
311 blk_unref(s->target);
314 if (s->copy_bitmap) {
315 hbitmap_free(s->copy_bitmap);
316 s->copy_bitmap = NULL;
320 void backup_do_checkpoint(BlockJob *job, Error **errp)
322 BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
324 assert(block_job_driver(job) == &backup_job_driver);
326 if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
327 error_setg(errp, "The backup job only supports block checkpoint in"
332 hbitmap_set(backup_job->copy_bitmap, 0, backup_job->len);
335 static void backup_drain(BlockJob *job)
337 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
339 /* Need to keep a reference in case blk_drain triggers execution
340 * of backup_complete...
343 BlockBackend *target = s->target;
350 static BlockErrorAction backup_error_action(BackupBlockJob *job,
351 bool read, int error)
354 return block_job_error_action(&job->common, job->on_source_error,
357 return block_job_error_action(&job->common, job->on_target_error,
362 static bool coroutine_fn yield_and_check(BackupBlockJob *job)
366 if (job_is_cancelled(&job->common.job)) {
370 /* We need to yield even for delay_ns = 0 so that bdrv_drain_all() can
371 * return. Without a yield, the VM would not reboot. */
372 delay_ns = block_job_ratelimit_get_delay(&job->common, job->bytes_read);
374 job_sleep_ns(&job->common.job, delay_ns);
376 if (job_is_cancelled(&job->common.job)) {
383 static bool bdrv_is_unallocated_range(BlockDriverState *bs,
384 int64_t offset, int64_t bytes)
386 int64_t end = offset + bytes;
388 while (offset < end && !bdrv_is_allocated(bs, offset, bytes, &bytes)) {
393 bytes = end - offset;
396 return offset >= end;
399 static int coroutine_fn backup_loop(BackupBlockJob *job)
405 BlockDriverState *bs = blk_bs(job->common.blk);
407 hbitmap_iter_init(&hbi, job->copy_bitmap, 0);
408 while ((offset = hbitmap_iter_next(&hbi)) != -1) {
409 if (job->sync_mode == MIRROR_SYNC_MODE_TOP &&
410 bdrv_is_unallocated_range(bs, offset, job->cluster_size))
412 hbitmap_reset(job->copy_bitmap, offset, job->cluster_size);
417 if (yield_and_check(job)) {
420 ret = backup_do_cow(job, offset,
421 job->cluster_size, &error_is_read, false);
422 if (ret < 0 && backup_error_action(job, error_is_read, -ret) ==
423 BLOCK_ERROR_ACTION_REPORT)
433 /* init copy_bitmap from sync_bitmap */
434 static void backup_incremental_init_copy_bitmap(BackupBlockJob *job)
437 uint64_t bytes = job->len;
439 while (bdrv_dirty_bitmap_next_dirty_area(job->sync_bitmap,
442 hbitmap_set(job->copy_bitmap, offset, bytes);
445 if (offset >= job->len) {
448 bytes = job->len - offset;
451 /* TODO job_progress_set_remaining() would make more sense */
452 job_progress_update(&job->common.job,
453 job->len - hbitmap_count(job->copy_bitmap));
456 static int coroutine_fn backup_run(Job *job, Error **errp)
458 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
459 BlockDriverState *bs = blk_bs(s->common.blk);
462 QLIST_INIT(&s->inflight_reqs);
463 qemu_co_rwlock_init(&s->flush_rwlock);
465 job_progress_set_remaining(job, s->len);
467 if (s->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
468 backup_incremental_init_copy_bitmap(s);
470 hbitmap_set(s->copy_bitmap, 0, s->len);
473 s->before_write.notify = backup_before_write_notify;
474 bdrv_add_before_write_notifier(bs, &s->before_write);
476 if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
477 /* All bits are set in copy_bitmap to allow any cluster to be copied.
478 * This does not actually require them to be copied. */
479 while (!job_is_cancelled(job)) {
480 /* Yield until the job is cancelled. We just let our before_write
481 * notify callback service CoW requests. */
485 ret = backup_loop(s);
488 notifier_with_return_remove(&s->before_write);
490 /* wait until pending backup_do_cow() calls have completed */
491 qemu_co_rwlock_wrlock(&s->flush_rwlock);
492 qemu_co_rwlock_unlock(&s->flush_rwlock);
497 static const BlockJobDriver backup_job_driver = {
499 .instance_size = sizeof(BackupBlockJob),
500 .job_type = JOB_TYPE_BACKUP,
501 .free = block_job_free,
502 .user_resume = block_job_user_resume,
503 .drain = block_job_drain,
505 .commit = backup_commit,
506 .abort = backup_abort,
507 .clean = backup_clean,
509 .drain = backup_drain,
512 static int64_t backup_calculate_cluster_size(BlockDriverState *target,
519 * If there is no backing file on the target, we cannot rely on COW if our
520 * backup cluster size is smaller than the target cluster size. Even for
521 * targets with a backing file, try to avoid COW if possible.
523 ret = bdrv_get_info(target, &bdi);
524 if (ret == -ENOTSUP && !target->backing) {
525 /* Cluster size is not defined */
526 warn_report("The target block device doesn't provide "
527 "information about the block size and it doesn't have a "
528 "backing file. The default block size of %u bytes is "
529 "used. If the actual block size of the target exceeds "
530 "this default, the backup may be unusable",
531 BACKUP_CLUSTER_SIZE_DEFAULT);
532 return BACKUP_CLUSTER_SIZE_DEFAULT;
533 } else if (ret < 0 && !target->backing) {
534 error_setg_errno(errp, -ret,
535 "Couldn't determine the cluster size of the target image, "
536 "which has no backing file");
537 error_append_hint(errp,
538 "Aborting, since this may create an unusable destination image\n");
540 } else if (ret < 0 && target->backing) {
541 /* Not fatal; just trudge on ahead. */
542 return BACKUP_CLUSTER_SIZE_DEFAULT;
545 return MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
548 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
549 BlockDriverState *target, int64_t speed,
550 MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
551 BitmapSyncMode bitmap_mode,
553 BlockdevOnError on_source_error,
554 BlockdevOnError on_target_error,
556 BlockCompletionFunc *cb, void *opaque,
557 JobTxn *txn, Error **errp)
560 BackupBlockJob *job = NULL;
562 int64_t cluster_size;
563 HBitmap *copy_bitmap = NULL;
569 error_setg(errp, "Source and target cannot be the same");
573 if (!bdrv_is_inserted(bs)) {
574 error_setg(errp, "Device is not inserted: %s",
575 bdrv_get_device_name(bs));
579 if (!bdrv_is_inserted(target)) {
580 error_setg(errp, "Device is not inserted: %s",
581 bdrv_get_device_name(target));
585 if (compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
586 error_setg(errp, "Compression is not supported for this drive %s",
587 bdrv_get_device_name(target));
591 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
595 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
599 /* QMP interface should have handled translating this to bitmap mode */
600 assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
602 if (sync_mode == MIRROR_SYNC_MODE_BITMAP) {
604 error_setg(errp, "must provide a valid bitmap name for "
605 "'%s' sync mode", MirrorSyncMode_str(sync_mode));
609 /* Create a new bitmap, and freeze/disable this one. */
610 if (bdrv_dirty_bitmap_create_successor(bs, sync_bitmap, errp) < 0) {
613 } else if (sync_bitmap) {
615 "a bitmap was given to backup_job_create, "
616 "but it received an incompatible sync_mode (%s)",
617 MirrorSyncMode_str(sync_mode));
621 len = bdrv_getlength(bs);
623 error_setg_errno(errp, -len, "unable to get length for '%s'",
624 bdrv_get_device_name(bs));
628 cluster_size = backup_calculate_cluster_size(target, errp);
629 if (cluster_size < 0) {
633 copy_bitmap = hbitmap_alloc(len, ctz32(cluster_size));
635 /* job->len is fixed, so we can't allow resize */
636 job = block_job_create(job_id, &backup_job_driver, txn, bs,
637 BLK_PERM_CONSISTENT_READ,
638 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
639 BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD,
640 speed, creation_flags, cb, opaque, errp);
645 /* The target must match the source in size, so no resize here either */
646 job->target = blk_new(job->common.job.aio_context,
648 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
649 BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
650 ret = blk_insert_bs(job->target, target, errp);
654 blk_set_disable_request_queuing(job->target, true);
656 job->on_source_error = on_source_error;
657 job->on_target_error = on_target_error;
658 job->sync_mode = sync_mode;
659 job->sync_bitmap = sync_bitmap;
660 job->bitmap_mode = bitmap_mode;
661 job->compress = compress;
663 /* Detect image-fleecing (and similar) schemes */
664 job->serialize_target_writes = bdrv_chain_contains(target, bs);
665 job->cluster_size = cluster_size;
666 job->copy_bitmap = copy_bitmap;
668 job->use_copy_range = !compress; /* compression isn't supported for it */
669 job->copy_range_size = MIN_NON_ZERO(blk_get_max_transfer(job->common.blk),
670 blk_get_max_transfer(job->target));
671 job->copy_range_size = MAX(job->cluster_size,
672 QEMU_ALIGN_UP(job->copy_range_size,
675 /* Required permissions are already taken with target's blk_new() */
676 block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
684 assert(!job || !job->copy_bitmap);
685 hbitmap_free(copy_bitmap);
688 bdrv_reclaim_dirty_bitmap(bs, sync_bitmap, NULL);
691 backup_clean(&job->common.job);
692 job_early_fail(&job->common.job);