hw/arm/armsse: Make SRAM bank size configurable
[qemu/ar7.git] / block / backup.c
blob435414e9649d633e0d3ac798fc98644ada8a2777
1 /*
2 * QEMU backup
4 * Copyright (C) 2013 Proxmox Server Solutions
6 * Authors:
7 * Dietmar Maurer (dietmar@proxmox.com)
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"
16 #include "trace.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 {
32 int64_t start_byte;
33 int64_t end_byte;
34 QLIST_ENTRY(CowRequest) list;
35 CoQueue wait_queue; /* coroutines blocked on this request */
36 } CowRequest;
38 typedef struct BackupBlockJob {
39 BlockJob common;
40 BlockBackend *target;
41 /* bitmap for sync=incremental */
42 BdrvDirtyBitmap *sync_bitmap;
43 MirrorSyncMode sync_mode;
44 BlockdevOnError on_source_error;
45 BlockdevOnError on_target_error;
46 CoRwlock flush_rwlock;
47 uint64_t len;
48 uint64_t bytes_read;
49 int64_t cluster_size;
50 bool compress;
51 NotifierWithReturn before_write;
52 QLIST_HEAD(, CowRequest) inflight_reqs;
54 HBitmap *copy_bitmap;
55 bool use_copy_range;
56 int64_t copy_range_size;
58 bool serialize_target_writes;
59 } BackupBlockJob;
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,
65 int64_t start,
66 int64_t end)
68 CowRequest *req;
69 bool retry;
71 do {
72 retry = false;
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);
76 retry = true;
77 break;
80 } while (retry);
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;
88 req->end_byte = end;
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,
103 int64_t start,
104 int64_t end,
105 bool is_write_notifier,
106 bool *error_is_read,
107 void **bounce_buffer)
109 int ret;
110 struct iovec iov;
111 QEMUIOVector qiov;
112 BlockBackend *blk = job->common.blk;
113 int nbytes;
114 int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
115 int write_flags = job->serialize_target_writes ? BDRV_REQ_SERIALISING : 0;
117 hbitmap_reset(job->copy_bitmap, start / job->cluster_size, 1);
118 nbytes = MIN(job->cluster_size, job->len - start);
119 if (!*bounce_buffer) {
120 *bounce_buffer = blk_blockalign(blk, job->cluster_size);
122 iov.iov_base = *bounce_buffer;
123 iov.iov_len = nbytes;
124 qemu_iovec_init_external(&qiov, &iov, 1);
126 ret = blk_co_preadv(blk, start, qiov.size, &qiov, read_flags);
127 if (ret < 0) {
128 trace_backup_do_cow_read_fail(job, start, ret);
129 if (error_is_read) {
130 *error_is_read = true;
132 goto fail;
135 if (qemu_iovec_is_zero(&qiov)) {
136 ret = blk_co_pwrite_zeroes(job->target, start,
137 qiov.size, write_flags | BDRV_REQ_MAY_UNMAP);
138 } else {
139 ret = blk_co_pwritev(job->target, start,
140 qiov.size, &qiov, write_flags |
141 (job->compress ? BDRV_REQ_WRITE_COMPRESSED : 0));
143 if (ret < 0) {
144 trace_backup_do_cow_write_fail(job, start, ret);
145 if (error_is_read) {
146 *error_is_read = false;
148 goto fail;
151 return nbytes;
152 fail:
153 hbitmap_set(job->copy_bitmap, start / job->cluster_size, 1);
154 return ret;
158 /* Copy range to target and return the bytes copied. If error occurred, return a
159 * negative error number. */
160 static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job,
161 int64_t start,
162 int64_t end,
163 bool is_write_notifier)
165 int ret;
166 int nr_clusters;
167 BlockBackend *blk = job->common.blk;
168 int nbytes;
169 int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0;
170 int write_flags = job->serialize_target_writes ? BDRV_REQ_SERIALISING : 0;
172 assert(QEMU_IS_ALIGNED(job->copy_range_size, job->cluster_size));
173 nbytes = MIN(job->copy_range_size, end - start);
174 nr_clusters = DIV_ROUND_UP(nbytes, job->cluster_size);
175 hbitmap_reset(job->copy_bitmap, start / job->cluster_size,
176 nr_clusters);
177 ret = blk_co_copy_range(blk, start, job->target, start, nbytes,
178 read_flags, write_flags);
179 if (ret < 0) {
180 trace_backup_do_cow_copy_range_fail(job, start, ret);
181 hbitmap_set(job->copy_bitmap, start / job->cluster_size,
182 nr_clusters);
183 return ret;
186 return nbytes;
189 static int coroutine_fn backup_do_cow(BackupBlockJob *job,
190 int64_t offset, uint64_t bytes,
191 bool *error_is_read,
192 bool is_write_notifier)
194 CowRequest cow_request;
195 int ret = 0;
196 int64_t start, end; /* bytes */
197 void *bounce_buffer = NULL;
199 qemu_co_rwlock_rdlock(&job->flush_rwlock);
201 start = QEMU_ALIGN_DOWN(offset, job->cluster_size);
202 end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size);
204 trace_backup_do_cow_enter(job, start, offset, bytes);
206 wait_for_overlapping_requests(job, start, end);
207 cow_request_begin(&cow_request, job, start, end);
209 while (start < end) {
210 if (!hbitmap_get(job->copy_bitmap, start / job->cluster_size)) {
211 trace_backup_do_cow_skip(job, start);
212 start += job->cluster_size;
213 continue; /* already copied */
216 trace_backup_do_cow_process(job, start);
218 if (job->use_copy_range) {
219 ret = backup_cow_with_offload(job, start, end, is_write_notifier);
220 if (ret < 0) {
221 job->use_copy_range = false;
224 if (!job->use_copy_range) {
225 ret = backup_cow_with_bounce_buffer(job, start, end, is_write_notifier,
226 error_is_read, &bounce_buffer);
228 if (ret < 0) {
229 break;
232 /* Publish progress, guest I/O counts as progress too. Note that the
233 * offset field is an opaque progress value, it is not a disk offset.
235 start += ret;
236 job->bytes_read += ret;
237 job_progress_update(&job->common.job, ret);
238 ret = 0;
241 if (bounce_buffer) {
242 qemu_vfree(bounce_buffer);
245 cow_request_end(&cow_request);
247 trace_backup_do_cow_return(job, offset, bytes, ret);
249 qemu_co_rwlock_unlock(&job->flush_rwlock);
251 return ret;
254 static int coroutine_fn backup_before_write_notify(
255 NotifierWithReturn *notifier,
256 void *opaque)
258 BackupBlockJob *job = container_of(notifier, BackupBlockJob, before_write);
259 BdrvTrackedRequest *req = opaque;
261 assert(req->bs == blk_bs(job->common.blk));
262 assert(QEMU_IS_ALIGNED(req->offset, BDRV_SECTOR_SIZE));
263 assert(QEMU_IS_ALIGNED(req->bytes, BDRV_SECTOR_SIZE));
265 return backup_do_cow(job, req->offset, req->bytes, NULL, true);
268 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
270 BdrvDirtyBitmap *bm;
271 BlockDriverState *bs = blk_bs(job->common.blk);
273 if (ret < 0) {
274 /* Merge the successor back into the parent, delete nothing. */
275 bm = bdrv_reclaim_dirty_bitmap(bs, job->sync_bitmap, NULL);
276 assert(bm);
277 } else {
278 /* Everything is fine, delete this bitmap and install the backup. */
279 bm = bdrv_dirty_bitmap_abdicate(bs, job->sync_bitmap, NULL);
280 assert(bm);
284 static void backup_commit(Job *job)
286 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
287 if (s->sync_bitmap) {
288 backup_cleanup_sync_bitmap(s, 0);
292 static void backup_abort(Job *job)
294 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
295 if (s->sync_bitmap) {
296 backup_cleanup_sync_bitmap(s, -1);
300 static void backup_clean(Job *job)
302 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
303 assert(s->target);
304 blk_unref(s->target);
305 s->target = NULL;
308 static void backup_attached_aio_context(BlockJob *job, AioContext *aio_context)
310 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
312 blk_set_aio_context(s->target, aio_context);
315 void backup_do_checkpoint(BlockJob *job, Error **errp)
317 BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
318 int64_t len;
320 assert(block_job_driver(job) == &backup_job_driver);
322 if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
323 error_setg(errp, "The backup job only supports block checkpoint in"
324 " sync=none mode");
325 return;
328 len = DIV_ROUND_UP(backup_job->len, backup_job->cluster_size);
329 hbitmap_set(backup_job->copy_bitmap, 0, len);
332 static void backup_drain(BlockJob *job)
334 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
336 /* Need to keep a reference in case blk_drain triggers execution
337 * of backup_complete...
339 if (s->target) {
340 BlockBackend *target = s->target;
341 blk_ref(target);
342 blk_drain(target);
343 blk_unref(target);
347 static BlockErrorAction backup_error_action(BackupBlockJob *job,
348 bool read, int error)
350 if (read) {
351 return block_job_error_action(&job->common, job->on_source_error,
352 true, error);
353 } else {
354 return block_job_error_action(&job->common, job->on_target_error,
355 false, error);
359 static bool coroutine_fn yield_and_check(BackupBlockJob *job)
361 uint64_t delay_ns;
363 if (job_is_cancelled(&job->common.job)) {
364 return true;
367 /* We need to yield even for delay_ns = 0 so that bdrv_drain_all() can
368 * return. Without a yield, the VM would not reboot. */
369 delay_ns = block_job_ratelimit_get_delay(&job->common, job->bytes_read);
370 job->bytes_read = 0;
371 job_sleep_ns(&job->common.job, delay_ns);
373 if (job_is_cancelled(&job->common.job)) {
374 return true;
377 return false;
380 static int coroutine_fn backup_run_incremental(BackupBlockJob *job)
382 int ret;
383 bool error_is_read;
384 int64_t cluster;
385 HBitmapIter hbi;
387 hbitmap_iter_init(&hbi, job->copy_bitmap, 0);
388 while ((cluster = hbitmap_iter_next(&hbi)) != -1) {
389 do {
390 if (yield_and_check(job)) {
391 return 0;
393 ret = backup_do_cow(job, cluster * job->cluster_size,
394 job->cluster_size, &error_is_read, false);
395 if (ret < 0 && backup_error_action(job, error_is_read, -ret) ==
396 BLOCK_ERROR_ACTION_REPORT)
398 return ret;
400 } while (ret < 0);
403 return 0;
406 /* init copy_bitmap from sync_bitmap */
407 static void backup_incremental_init_copy_bitmap(BackupBlockJob *job)
409 BdrvDirtyBitmapIter *dbi;
410 int64_t offset;
411 int64_t end = DIV_ROUND_UP(bdrv_dirty_bitmap_size(job->sync_bitmap),
412 job->cluster_size);
414 dbi = bdrv_dirty_iter_new(job->sync_bitmap);
415 while ((offset = bdrv_dirty_iter_next(dbi)) != -1) {
416 int64_t cluster = offset / job->cluster_size;
417 int64_t next_cluster;
419 offset += bdrv_dirty_bitmap_granularity(job->sync_bitmap);
420 if (offset >= bdrv_dirty_bitmap_size(job->sync_bitmap)) {
421 hbitmap_set(job->copy_bitmap, cluster, end - cluster);
422 break;
425 offset = bdrv_dirty_bitmap_next_zero(job->sync_bitmap, offset,
426 UINT64_MAX);
427 if (offset == -1) {
428 hbitmap_set(job->copy_bitmap, cluster, end - cluster);
429 break;
432 next_cluster = DIV_ROUND_UP(offset, job->cluster_size);
433 hbitmap_set(job->copy_bitmap, cluster, next_cluster - cluster);
434 if (next_cluster >= end) {
435 break;
438 bdrv_set_dirty_iter(dbi, next_cluster * job->cluster_size);
441 /* TODO job_progress_set_remaining() would make more sense */
442 job_progress_update(&job->common.job,
443 job->len - hbitmap_count(job->copy_bitmap) * job->cluster_size);
445 bdrv_dirty_iter_free(dbi);
448 static int coroutine_fn backup_run(Job *job, Error **errp)
450 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
451 BlockDriverState *bs = blk_bs(s->common.blk);
452 int64_t offset, nb_clusters;
453 int ret = 0;
455 QLIST_INIT(&s->inflight_reqs);
456 qemu_co_rwlock_init(&s->flush_rwlock);
458 nb_clusters = DIV_ROUND_UP(s->len, s->cluster_size);
459 job_progress_set_remaining(job, s->len);
461 s->copy_bitmap = hbitmap_alloc(nb_clusters, 0);
462 if (s->sync_mode == MIRROR_SYNC_MODE_INCREMENTAL) {
463 backup_incremental_init_copy_bitmap(s);
464 } else {
465 hbitmap_set(s->copy_bitmap, 0, nb_clusters);
469 s->before_write.notify = backup_before_write_notify;
470 bdrv_add_before_write_notifier(bs, &s->before_write);
472 if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
473 /* All bits are set in copy_bitmap to allow any cluster to be copied.
474 * This does not actually require them to be copied. */
475 while (!job_is_cancelled(job)) {
476 /* Yield until the job is cancelled. We just let our before_write
477 * notify callback service CoW requests. */
478 job_yield(job);
480 } else if (s->sync_mode == MIRROR_SYNC_MODE_INCREMENTAL) {
481 ret = backup_run_incremental(s);
482 } else {
483 /* Both FULL and TOP SYNC_MODE's require copying.. */
484 for (offset = 0; offset < s->len;
485 offset += s->cluster_size) {
486 bool error_is_read;
487 int alloced = 0;
489 if (yield_and_check(s)) {
490 break;
493 if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
494 int i;
495 int64_t n;
497 /* Check to see if these blocks are already in the
498 * backing file. */
500 for (i = 0; i < s->cluster_size;) {
501 /* bdrv_is_allocated() only returns true/false based
502 * on the first set of sectors it comes across that
503 * are are all in the same state.
504 * For that reason we must verify each sector in the
505 * backup cluster length. We end up copying more than
506 * needed but at some point that is always the case. */
507 alloced =
508 bdrv_is_allocated(bs, offset + i,
509 s->cluster_size - i, &n);
510 i += n;
512 if (alloced || n == 0) {
513 break;
517 /* If the above loop never found any sectors that are in
518 * the topmost image, skip this backup. */
519 if (alloced == 0) {
520 continue;
523 /* FULL sync mode we copy the whole drive. */
524 if (alloced < 0) {
525 ret = alloced;
526 } else {
527 ret = backup_do_cow(s, offset, s->cluster_size,
528 &error_is_read, false);
530 if (ret < 0) {
531 /* Depending on error action, fail now or retry cluster */
532 BlockErrorAction action =
533 backup_error_action(s, error_is_read, -ret);
534 if (action == BLOCK_ERROR_ACTION_REPORT) {
535 break;
536 } else {
537 offset -= s->cluster_size;
538 continue;
544 notifier_with_return_remove(&s->before_write);
546 /* wait until pending backup_do_cow() calls have completed */
547 qemu_co_rwlock_wrlock(&s->flush_rwlock);
548 qemu_co_rwlock_unlock(&s->flush_rwlock);
549 hbitmap_free(s->copy_bitmap);
551 return ret;
554 static const BlockJobDriver backup_job_driver = {
555 .job_driver = {
556 .instance_size = sizeof(BackupBlockJob),
557 .job_type = JOB_TYPE_BACKUP,
558 .free = block_job_free,
559 .user_resume = block_job_user_resume,
560 .drain = block_job_drain,
561 .run = backup_run,
562 .commit = backup_commit,
563 .abort = backup_abort,
564 .clean = backup_clean,
566 .attached_aio_context = backup_attached_aio_context,
567 .drain = backup_drain,
570 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
571 BlockDriverState *target, int64_t speed,
572 MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
573 bool compress,
574 BlockdevOnError on_source_error,
575 BlockdevOnError on_target_error,
576 int creation_flags,
577 BlockCompletionFunc *cb, void *opaque,
578 JobTxn *txn, Error **errp)
580 int64_t len;
581 BlockDriverInfo bdi;
582 BackupBlockJob *job = NULL;
583 int ret;
585 assert(bs);
586 assert(target);
588 if (bs == target) {
589 error_setg(errp, "Source and target cannot be the same");
590 return NULL;
593 if (!bdrv_is_inserted(bs)) {
594 error_setg(errp, "Device is not inserted: %s",
595 bdrv_get_device_name(bs));
596 return NULL;
599 if (!bdrv_is_inserted(target)) {
600 error_setg(errp, "Device is not inserted: %s",
601 bdrv_get_device_name(target));
602 return NULL;
605 if (compress && target->drv->bdrv_co_pwritev_compressed == NULL) {
606 error_setg(errp, "Compression is not supported for this drive %s",
607 bdrv_get_device_name(target));
608 return NULL;
611 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
612 return NULL;
615 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
616 return NULL;
619 if (sync_mode == MIRROR_SYNC_MODE_INCREMENTAL) {
620 if (!sync_bitmap) {
621 error_setg(errp, "must provide a valid bitmap name for "
622 "\"incremental\" sync mode");
623 return NULL;
626 /* Create a new bitmap, and freeze/disable this one. */
627 if (bdrv_dirty_bitmap_create_successor(bs, sync_bitmap, errp) < 0) {
628 return NULL;
630 } else if (sync_bitmap) {
631 error_setg(errp,
632 "a sync_bitmap was provided to backup_run, "
633 "but received an incompatible sync_mode (%s)",
634 MirrorSyncMode_str(sync_mode));
635 return NULL;
638 len = bdrv_getlength(bs);
639 if (len < 0) {
640 error_setg_errno(errp, -len, "unable to get length for '%s'",
641 bdrv_get_device_name(bs));
642 goto error;
645 /* job->len is fixed, so we can't allow resize */
646 job = block_job_create(job_id, &backup_job_driver, txn, bs,
647 BLK_PERM_CONSISTENT_READ,
648 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
649 BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD,
650 speed, creation_flags, cb, opaque, errp);
651 if (!job) {
652 goto error;
655 /* The target must match the source in size, so no resize here either */
656 job->target = blk_new(BLK_PERM_WRITE,
657 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE |
658 BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD);
659 ret = blk_insert_bs(job->target, target, errp);
660 if (ret < 0) {
661 goto error;
664 job->on_source_error = on_source_error;
665 job->on_target_error = on_target_error;
666 job->sync_mode = sync_mode;
667 job->sync_bitmap = sync_mode == MIRROR_SYNC_MODE_INCREMENTAL ?
668 sync_bitmap : NULL;
669 job->compress = compress;
671 /* Detect image-fleecing (and similar) schemes */
672 job->serialize_target_writes = bdrv_chain_contains(target, bs);
674 /* If there is no backing file on the target, we cannot rely on COW if our
675 * backup cluster size is smaller than the target cluster size. Even for
676 * targets with a backing file, try to avoid COW if possible. */
677 ret = bdrv_get_info(target, &bdi);
678 if (ret == -ENOTSUP && !target->backing) {
679 /* Cluster size is not defined */
680 warn_report("The target block device doesn't provide "
681 "information about the block size and it doesn't have a "
682 "backing file. The default block size of %u bytes is "
683 "used. If the actual block size of the target exceeds "
684 "this default, the backup may be unusable",
685 BACKUP_CLUSTER_SIZE_DEFAULT);
686 job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
687 } else if (ret < 0 && !target->backing) {
688 error_setg_errno(errp, -ret,
689 "Couldn't determine the cluster size of the target image, "
690 "which has no backing file");
691 error_append_hint(errp,
692 "Aborting, since this may create an unusable destination image\n");
693 goto error;
694 } else if (ret < 0 && target->backing) {
695 /* Not fatal; just trudge on ahead. */
696 job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT;
697 } else {
698 job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size);
700 job->use_copy_range = true;
701 job->copy_range_size = MIN_NON_ZERO(blk_get_max_transfer(job->common.blk),
702 blk_get_max_transfer(job->target));
703 job->copy_range_size = MAX(job->cluster_size,
704 QEMU_ALIGN_UP(job->copy_range_size,
705 job->cluster_size));
707 /* Required permissions are already taken with target's blk_new() */
708 block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
709 &error_abort);
710 job->len = len;
712 return &job->common;
714 error:
715 if (sync_bitmap) {
716 bdrv_reclaim_dirty_bitmap(bs, sync_bitmap, NULL);
718 if (job) {
719 backup_clean(&job->common.job);
720 job_early_fail(&job->common.job);
723 return NULL;