4 * Copyright (C) 2013 Proxmox Server Solutions
5 * Copyright (c) 2019 Virtuozzo International GmbH.
8 * Dietmar Maurer (dietmar@proxmox.com)
10 * This work is licensed under the terms of the GNU GPL, version 2 or later.
11 * See the COPYING file in the top-level directory.
15 #include "qemu/osdep.h"
18 #include "block/block.h"
19 #include "block/block_int.h"
20 #include "block/blockjob_int.h"
21 #include "block/block_backup.h"
22 #include "block/block-copy.h"
23 #include "qapi/error.h"
24 #include "qapi/qmp/qerror.h"
25 #include "qemu/ratelimit.h"
26 #include "qemu/cutils.h"
27 #include "sysemu/block-backend.h"
28 #include "qemu/bitmap.h"
29 #include "qemu/error-report.h"
31 #include "block/backup-top.h"
33 #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16)
35 typedef struct BackupBlockJob {
37 BlockDriverState *backup_top;
38 BlockDriverState *source_bs;
40 BdrvDirtyBitmap *sync_bitmap;
42 MirrorSyncMode sync_mode;
43 BitmapSyncMode bitmap_mode;
44 BlockdevOnError on_source_error;
45 BlockdevOnError on_target_error;
53 static const BlockJobDriver backup_job_driver;
55 static void backup_progress_bytes_callback(int64_t bytes, void *opaque)
57 BackupBlockJob *s = opaque;
59 s->bytes_read += bytes;
62 static int coroutine_fn backup_do_cow(BackupBlockJob *job,
63 int64_t offset, uint64_t bytes,
67 int64_t start, end; /* bytes */
69 start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
70 end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
72 trace_backup_do_cow_enter(job, start, offset, bytes);
74 ret = block_copy(job->bcs, start, end - start, error_is_read);
76 trace_backup_do_cow_return(job, offset, bytes, ret);
81 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
84 bool sync = (((ret == 0) || (job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS)) \
85 && (job->bitmap_mode != BITMAP_SYNC_MODE_NEVER));
89 * We succeeded, or we always intended to sync the bitmap.
90 * Delete this bitmap and install the child.
92 bm = bdrv_dirty_bitmap_abdicate(job->sync_bitmap, NULL);
95 * We failed, or we never intended to sync the bitmap anyway.
96 * Merge the successor back into the parent, keeping all data.
98 bm = bdrv_reclaim_dirty_bitmap(job->sync_bitmap, NULL);
103 if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) {
104 /* If we failed and synced, merge in the bits we didn't copy: */
105 bdrv_dirty_bitmap_merge_internal(bm, block_copy_dirty_bitmap(job->bcs),
110 static void backup_commit(Job *job)
112 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
113 if (s->sync_bitmap) {
114 backup_cleanup_sync_bitmap(s, 0);
118 static void backup_abort(Job *job)
120 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
121 if (s->sync_bitmap) {
122 backup_cleanup_sync_bitmap(s, -1);
126 static void backup_clean(Job *job)
128 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
129 AioContext *aio_context = bdrv_get_aio_context(s->backup_top);
131 aio_context_acquire(aio_context);
132 bdrv_backup_top_drop(s->backup_top);
133 aio_context_release(aio_context);
136 void backup_do_checkpoint(BlockJob *job, Error **errp)
138 BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
140 assert(block_job_driver(job) == &backup_job_driver);
142 if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
143 error_setg(errp, "The backup job only supports block checkpoint in"
148 bdrv_set_dirty_bitmap(block_copy_dirty_bitmap(backup_job->bcs), 0,
152 static BlockErrorAction backup_error_action(BackupBlockJob *job,
153 bool read, int error)
156 return block_job_error_action(&job->common, job->on_source_error,
159 return block_job_error_action(&job->common, job->on_target_error,
164 static bool coroutine_fn yield_and_check(BackupBlockJob *job)
168 if (job_is_cancelled(&job->common.job)) {
173 * We need to yield even for delay_ns = 0 so that bdrv_drain_all() can
174 * return. Without a yield, the VM would not reboot.
176 delay_ns = block_job_ratelimit_get_delay(&job->common, job->bytes_read);
178 job_sleep_ns(&job->common.job, delay_ns);
180 if (job_is_cancelled(&job->common.job)) {
187 static int coroutine_fn backup_loop(BackupBlockJob *job)
191 BdrvDirtyBitmapIter *bdbi;
194 bdbi = bdrv_dirty_iter_new(block_copy_dirty_bitmap(job->bcs));
195 while ((offset = bdrv_dirty_iter_next(bdbi)) != -1) {
197 if (yield_and_check(job)) {
200 ret = backup_do_cow(job, offset, job->cluster_size, &error_is_read);
201 if (ret < 0 && backup_error_action(job, error_is_read, -ret) ==
202 BLOCK_ERROR_ACTION_REPORT)
210 bdrv_dirty_iter_free(bdbi);
214 static void backup_init_bcs_bitmap(BackupBlockJob *job)
218 BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs);
220 if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
221 ret = bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap,
225 if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
227 * We can't hog the coroutine to initialize this thoroughly.
228 * Set a flag and resume work when we are able to yield safely.
230 block_copy_set_skip_unallocated(job->bcs, true);
232 bdrv_set_dirty_bitmap(bcs_bitmap, 0, job->len);
235 estimate = bdrv_get_dirty_count(bcs_bitmap);
236 job_progress_set_remaining(&job->common.job, estimate);
239 static int coroutine_fn backup_run(Job *job, Error **errp)
241 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
244 backup_init_bcs_bitmap(s);
246 if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
250 for (offset = 0; offset < s->len; ) {
251 if (yield_and_check(s)) {
256 ret = block_copy_reset_unallocated(s->bcs, offset, &count);
263 block_copy_set_skip_unallocated(s->bcs, false);
266 if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
268 * All bits are set in bcs bitmap to allow any cluster to be copied.
269 * This does not actually require them to be copied.
271 while (!job_is_cancelled(job)) {
273 * Yield until the job is cancelled. We just let our before_write
274 * notify callback service CoW requests.
279 ret = backup_loop(s);
286 static const BlockJobDriver backup_job_driver = {
288 .instance_size = sizeof(BackupBlockJob),
289 .job_type = JOB_TYPE_BACKUP,
290 .free = block_job_free,
291 .user_resume = block_job_user_resume,
293 .commit = backup_commit,
294 .abort = backup_abort,
295 .clean = backup_clean,
299 static int64_t backup_calculate_cluster_size(BlockDriverState *target,
306 * If there is no backing file on the target, we cannot rely on COW if our
307 * backup cluster size is smaller than the target cluster size. Even for
308 * targets with a backing file, try to avoid COW if possible.
310 ret = bdrv_get_info(target, &bdi);
311 if (ret == -ENOTSUP && !target->backing) {
312 /* Cluster size is not defined */
313 warn_report("The target block device doesn't provide "
314 "information about the block size and it doesn't have a "
315 "backing file. The default block size of %u bytes is "
316 "used. If the actual block size of the target exceeds "
317 "this default, the backup may be unusable",
318 BACKUP_CLUSTER_SIZE_DEFAULT);
319 return BACKUP_CLUSTER_SIZE_DEFAULT;
320 } else if (ret < 0 && !target->backing) {
321 error_setg_errno(errp, -ret,
322 "Couldn't determine the cluster size of the target image, "
323 "which has no backing file");
324 error_append_hint(errp,
325 "Aborting, since this may create an unusable destination image\n");
327 } else if (ret < 0 && target->backing) {
328 /* Not fatal; just trudge on ahead. */
329 return BACKUP_CLUSTER_SIZE_DEFAULT;
332 return MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
335 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
336 BlockDriverState *target, int64_t speed,
337 MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
338 BitmapSyncMode bitmap_mode,
340 const char *filter_node_name,
341 BlockdevOnError on_source_error,
342 BlockdevOnError on_target_error,
344 BlockCompletionFunc *cb, void *opaque,
345 JobTxn *txn, Error **errp)
348 BackupBlockJob *job = NULL;
349 int64_t cluster_size;
350 BdrvRequestFlags write_flags;
351 BlockDriverState *backup_top = NULL;
352 BlockCopyState *bcs = NULL;
357 /* QMP interface protects us from these cases */
358 assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
359 assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP);
362 error_setg(errp, "Source and target cannot be the same");
366 if (!bdrv_is_inserted(bs)) {
367 error_setg(errp, "Device is not inserted: %s",
368 bdrv_get_device_name(bs));
372 if (!bdrv_is_inserted(target)) {
373 error_setg(errp, "Device is not inserted: %s",
374 bdrv_get_device_name(target));
378 if (compress && !block_driver_can_compress(target->drv)) {
379 error_setg(errp, "Compression is not supported for this drive %s",
380 bdrv_get_device_name(target));
384 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
388 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
393 /* If we need to write to this bitmap, check that we can: */
394 if (bitmap_mode != BITMAP_SYNC_MODE_NEVER &&
395 bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) {
399 /* Create a new bitmap, and freeze/disable this one. */
400 if (bdrv_dirty_bitmap_create_successor(sync_bitmap, errp) < 0) {
405 len = bdrv_getlength(bs);
407 error_setg_errno(errp, -len, "unable to get length for '%s'",
408 bdrv_get_device_name(bs));
412 cluster_size = backup_calculate_cluster_size(target, errp);
413 if (cluster_size < 0) {
418 * If source is in backing chain of target assume that target is going to be
419 * used for "image fleecing", i.e. it should represent a kind of snapshot of
420 * source at backup-start point in time. And target is going to be read by
421 * somebody (for example, used as NBD export) during backup job.
423 * In this case, we need to add BDRV_REQ_SERIALISING write flag to avoid
424 * intersection of backup writes and third party reads from target,
425 * otherwise reading from target we may occasionally read already updated by
428 * For more information see commit f8d59dfb40bb and test
429 * tests/qemu-iotests/222
431 write_flags = (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) |
432 (compress ? BDRV_REQ_WRITE_COMPRESSED : 0),
434 backup_top = bdrv_backup_top_append(bs, target, filter_node_name,
435 cluster_size, write_flags, &bcs, errp);
440 /* job->len is fixed, so we can't allow resize */
441 job = block_job_create(job_id, &backup_job_driver, txn, backup_top,
443 speed, creation_flags, cb, opaque, errp);
448 job->backup_top = backup_top;
450 job->on_source_error = on_source_error;
451 job->on_target_error = on_target_error;
452 job->sync_mode = sync_mode;
453 job->sync_bitmap = sync_bitmap;
454 job->bitmap_mode = bitmap_mode;
456 job->cluster_size = cluster_size;
459 block_copy_set_progress_callback(bcs, backup_progress_bytes_callback, job);
460 block_copy_set_progress_meter(bcs, &job->common.job.progress);
462 /* Required permissions are already taken by backup-top target */
463 block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
470 bdrv_reclaim_dirty_bitmap(sync_bitmap, NULL);
473 bdrv_backup_top_drop(backup_top);