]> git.proxmox.com Git - mirror_qemu.git/blob - block/backup.c
MAINTAINERS: Change my email address
[mirror_qemu.git] / block / backup.c
1 /*
2 * QEMU backup
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 *
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.
12 *
13 */
14
15 #include "qemu/osdep.h"
16
17 #include "trace.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/cutils.h"
26 #include "sysemu/block-backend.h"
27 #include "qemu/bitmap.h"
28 #include "qemu/error-report.h"
29
30 #include "block/copy-before-write.h"
31
32 typedef struct BackupBlockJob {
33 BlockJob common;
34 BlockDriverState *cbw;
35 BlockDriverState *source_bs;
36 BlockDriverState *target_bs;
37
38 BdrvDirtyBitmap *sync_bitmap;
39
40 MirrorSyncMode sync_mode;
41 BitmapSyncMode bitmap_mode;
42 BlockdevOnError on_source_error;
43 BlockdevOnError on_target_error;
44 uint64_t len;
45 int64_t cluster_size;
46 BackupPerf perf;
47
48 BlockCopyState *bcs;
49
50 bool wait;
51 BlockCopyCallState *bg_bcs_call;
52 } BackupBlockJob;
53
54 static const BlockJobDriver backup_job_driver;
55
56 static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret)
57 {
58 BdrvDirtyBitmap *bm;
59 bool sync = (((ret == 0) || (job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS)) \
60 && (job->bitmap_mode != BITMAP_SYNC_MODE_NEVER));
61
62 if (sync) {
63 /*
64 * We succeeded, or we always intended to sync the bitmap.
65 * Delete this bitmap and install the child.
66 */
67 bm = bdrv_dirty_bitmap_abdicate(job->sync_bitmap, NULL);
68 } else {
69 /*
70 * We failed, or we never intended to sync the bitmap anyway.
71 * Merge the successor back into the parent, keeping all data.
72 */
73 bm = bdrv_reclaim_dirty_bitmap(job->sync_bitmap, NULL);
74 }
75
76 assert(bm);
77
78 if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) {
79 /* If we failed and synced, merge in the bits we didn't copy: */
80 bdrv_dirty_bitmap_merge_internal(bm, block_copy_dirty_bitmap(job->bcs),
81 NULL, true);
82 }
83 }
84
85 static void backup_commit(Job *job)
86 {
87 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
88 if (s->sync_bitmap) {
89 backup_cleanup_sync_bitmap(s, 0);
90 }
91 }
92
93 static void backup_abort(Job *job)
94 {
95 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
96 if (s->sync_bitmap) {
97 backup_cleanup_sync_bitmap(s, -1);
98 }
99 }
100
101 static void backup_clean(Job *job)
102 {
103 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
104 block_job_remove_all_bdrv(&s->common);
105 bdrv_cbw_drop(s->cbw);
106 }
107
108 void backup_do_checkpoint(BlockJob *job, Error **errp)
109 {
110 BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common);
111
112 assert(block_job_driver(job) == &backup_job_driver);
113
114 if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) {
115 error_setg(errp, "The backup job only supports block checkpoint in"
116 " sync=none mode");
117 return;
118 }
119
120 bdrv_set_dirty_bitmap(block_copy_dirty_bitmap(backup_job->bcs), 0,
121 backup_job->len);
122 }
123
124 static BlockErrorAction backup_error_action(BackupBlockJob *job,
125 bool read, int error)
126 {
127 if (read) {
128 return block_job_error_action(&job->common, job->on_source_error,
129 true, error);
130 } else {
131 return block_job_error_action(&job->common, job->on_target_error,
132 false, error);
133 }
134 }
135
136 static void coroutine_fn backup_block_copy_callback(void *opaque)
137 {
138 BackupBlockJob *s = opaque;
139
140 if (s->wait) {
141 s->wait = false;
142 aio_co_wake(s->common.job.co);
143 } else {
144 job_enter(&s->common.job);
145 }
146 }
147
148 static int coroutine_fn backup_loop(BackupBlockJob *job)
149 {
150 BlockCopyCallState *s = NULL;
151 int ret = 0;
152 bool error_is_read;
153 BlockErrorAction act;
154
155 while (true) { /* retry loop */
156 job->bg_bcs_call = s = block_copy_async(job->bcs, 0,
157 QEMU_ALIGN_UP(job->len, job->cluster_size),
158 job->perf.max_workers, job->perf.max_chunk,
159 backup_block_copy_callback, job);
160
161 while (!block_copy_call_finished(s) &&
162 !job_is_cancelled(&job->common.job))
163 {
164 job_yield(&job->common.job);
165 }
166
167 if (!block_copy_call_finished(s)) {
168 assert(job_is_cancelled(&job->common.job));
169 /*
170 * Note that we can't use job_yield() here, as it doesn't work for
171 * cancelled job.
172 */
173 block_copy_call_cancel(s);
174 job->wait = true;
175 qemu_coroutine_yield();
176 assert(block_copy_call_finished(s));
177 ret = 0;
178 goto out;
179 }
180
181 if (job_is_cancelled(&job->common.job) ||
182 block_copy_call_succeeded(s))
183 {
184 ret = 0;
185 goto out;
186 }
187
188 if (block_copy_call_cancelled(s)) {
189 /*
190 * Job is not cancelled but only block-copy call. This is possible
191 * after job pause. Now the pause is finished, start new block-copy
192 * iteration.
193 */
194 block_copy_call_free(s);
195 continue;
196 }
197
198 /* The only remaining case is failed block-copy call. */
199 assert(block_copy_call_failed(s));
200
201 ret = block_copy_call_status(s, &error_is_read);
202 act = backup_error_action(job, error_is_read, -ret);
203 switch (act) {
204 case BLOCK_ERROR_ACTION_REPORT:
205 goto out;
206 case BLOCK_ERROR_ACTION_STOP:
207 /*
208 * Go to pause prior to starting new block-copy call on the next
209 * iteration.
210 */
211 job_pause_point(&job->common.job);
212 break;
213 case BLOCK_ERROR_ACTION_IGNORE:
214 /* Proceed to new block-copy call to retry. */
215 break;
216 default:
217 abort();
218 }
219
220 block_copy_call_free(s);
221 }
222
223 out:
224 block_copy_call_free(s);
225 job->bg_bcs_call = NULL;
226 return ret;
227 }
228
229 static void backup_init_bcs_bitmap(BackupBlockJob *job)
230 {
231 bool ret;
232 uint64_t estimate;
233 BdrvDirtyBitmap *bcs_bitmap = block_copy_dirty_bitmap(job->bcs);
234
235 if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) {
236 bdrv_clear_dirty_bitmap(bcs_bitmap, NULL);
237 ret = bdrv_dirty_bitmap_merge_internal(bcs_bitmap, job->sync_bitmap,
238 NULL, true);
239 assert(ret);
240 } else if (job->sync_mode == MIRROR_SYNC_MODE_TOP) {
241 /*
242 * We can't hog the coroutine to initialize this thoroughly.
243 * Set a flag and resume work when we are able to yield safely.
244 */
245 block_copy_set_skip_unallocated(job->bcs, true);
246 }
247
248 estimate = bdrv_get_dirty_count(bcs_bitmap);
249 job_progress_set_remaining(&job->common.job, estimate);
250 }
251
252 static int coroutine_fn backup_run(Job *job, Error **errp)
253 {
254 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
255 int ret;
256
257 backup_init_bcs_bitmap(s);
258
259 if (s->sync_mode == MIRROR_SYNC_MODE_TOP) {
260 int64_t offset = 0;
261 int64_t count;
262
263 for (offset = 0; offset < s->len; ) {
264 if (job_is_cancelled(job)) {
265 return -ECANCELED;
266 }
267
268 job_pause_point(job);
269
270 if (job_is_cancelled(job)) {
271 return -ECANCELED;
272 }
273
274 ret = block_copy_reset_unallocated(s->bcs, offset, &count);
275 if (ret < 0) {
276 return ret;
277 }
278
279 offset += count;
280 }
281 block_copy_set_skip_unallocated(s->bcs, false);
282 }
283
284 if (s->sync_mode == MIRROR_SYNC_MODE_NONE) {
285 /*
286 * All bits are set in bcs bitmap to allow any cluster to be copied.
287 * This does not actually require them to be copied.
288 */
289 while (!job_is_cancelled(job)) {
290 /*
291 * Yield until the job is cancelled. We just let our before_write
292 * notify callback service CoW requests.
293 */
294 job_yield(job);
295 }
296 } else {
297 return backup_loop(s);
298 }
299
300 return 0;
301 }
302
303 static void coroutine_fn backup_pause(Job *job)
304 {
305 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
306
307 if (s->bg_bcs_call && !block_copy_call_finished(s->bg_bcs_call)) {
308 block_copy_call_cancel(s->bg_bcs_call);
309 s->wait = true;
310 qemu_coroutine_yield();
311 }
312 }
313
314 static void coroutine_fn backup_set_speed(BlockJob *job, int64_t speed)
315 {
316 BackupBlockJob *s = container_of(job, BackupBlockJob, common);
317
318 /*
319 * block_job_set_speed() is called first from block_job_create(), when we
320 * don't yet have s->bcs.
321 */
322 if (s->bcs) {
323 block_copy_set_speed(s->bcs, speed);
324 if (s->bg_bcs_call) {
325 block_copy_kick(s->bg_bcs_call);
326 }
327 }
328 }
329
330 static bool backup_cancel(Job *job, bool force)
331 {
332 BackupBlockJob *s = container_of(job, BackupBlockJob, common.job);
333
334 bdrv_cancel_in_flight(s->target_bs);
335 return true;
336 }
337
338 static const BlockJobDriver backup_job_driver = {
339 .job_driver = {
340 .instance_size = sizeof(BackupBlockJob),
341 .job_type = JOB_TYPE_BACKUP,
342 .free = block_job_free,
343 .user_resume = block_job_user_resume,
344 .run = backup_run,
345 .commit = backup_commit,
346 .abort = backup_abort,
347 .clean = backup_clean,
348 .pause = backup_pause,
349 .cancel = backup_cancel,
350 },
351 .set_speed = backup_set_speed,
352 };
353
354 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
355 BlockDriverState *target, int64_t speed,
356 MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap,
357 BitmapSyncMode bitmap_mode,
358 bool compress,
359 const char *filter_node_name,
360 BackupPerf *perf,
361 BlockdevOnError on_source_error,
362 BlockdevOnError on_target_error,
363 int creation_flags,
364 BlockCompletionFunc *cb, void *opaque,
365 JobTxn *txn, Error **errp)
366 {
367 int64_t len, target_len;
368 BackupBlockJob *job = NULL;
369 int64_t cluster_size;
370 BlockDriverState *cbw = NULL;
371 BlockCopyState *bcs = NULL;
372
373 assert(bs);
374 assert(target);
375
376 /* QMP interface protects us from these cases */
377 assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL);
378 assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP);
379
380 if (bs == target) {
381 error_setg(errp, "Source and target cannot be the same");
382 return NULL;
383 }
384
385 if (!bdrv_is_inserted(bs)) {
386 error_setg(errp, "Device is not inserted: %s",
387 bdrv_get_device_name(bs));
388 return NULL;
389 }
390
391 if (!bdrv_is_inserted(target)) {
392 error_setg(errp, "Device is not inserted: %s",
393 bdrv_get_device_name(target));
394 return NULL;
395 }
396
397 if (compress && !bdrv_supports_compressed_writes(target)) {
398 error_setg(errp, "Compression is not supported for this drive %s",
399 bdrv_get_device_name(target));
400 return NULL;
401 }
402
403 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) {
404 return NULL;
405 }
406
407 if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) {
408 return NULL;
409 }
410
411 if (perf->max_workers < 1 || perf->max_workers > INT_MAX) {
412 error_setg(errp, "max-workers must be between 1 and %d", INT_MAX);
413 return NULL;
414 }
415
416 if (perf->max_chunk < 0) {
417 error_setg(errp, "max-chunk must be zero (which means no limit) or "
418 "positive");
419 return NULL;
420 }
421
422 if (sync_bitmap) {
423 /* If we need to write to this bitmap, check that we can: */
424 if (bitmap_mode != BITMAP_SYNC_MODE_NEVER &&
425 bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) {
426 return NULL;
427 }
428
429 /* Create a new bitmap, and freeze/disable this one. */
430 if (bdrv_dirty_bitmap_create_successor(sync_bitmap, errp) < 0) {
431 return NULL;
432 }
433 }
434
435 len = bdrv_getlength(bs);
436 if (len < 0) {
437 error_setg_errno(errp, -len, "Unable to get length for '%s'",
438 bdrv_get_device_or_node_name(bs));
439 goto error;
440 }
441
442 target_len = bdrv_getlength(target);
443 if (target_len < 0) {
444 error_setg_errno(errp, -target_len, "Unable to get length for '%s'",
445 bdrv_get_device_or_node_name(bs));
446 goto error;
447 }
448
449 if (target_len != len) {
450 error_setg(errp, "Source and target image have different sizes");
451 goto error;
452 }
453
454 cbw = bdrv_cbw_append(bs, target, filter_node_name, &bcs, errp);
455 if (!cbw) {
456 goto error;
457 }
458
459 cluster_size = block_copy_cluster_size(bcs);
460
461 if (perf->max_chunk && perf->max_chunk < cluster_size) {
462 error_setg(errp, "Required max-chunk (%" PRIi64 ") is less than backup "
463 "cluster size (%" PRIi64 ")", perf->max_chunk, cluster_size);
464 goto error;
465 }
466
467 /* job->len is fixed, so we can't allow resize */
468 job = block_job_create(job_id, &backup_job_driver, txn, cbw,
469 0, BLK_PERM_ALL,
470 speed, creation_flags, cb, opaque, errp);
471 if (!job) {
472 goto error;
473 }
474
475 job->cbw = cbw;
476 job->source_bs = bs;
477 job->target_bs = target;
478 job->on_source_error = on_source_error;
479 job->on_target_error = on_target_error;
480 job->sync_mode = sync_mode;
481 job->sync_bitmap = sync_bitmap;
482 job->bitmap_mode = bitmap_mode;
483 job->bcs = bcs;
484 job->cluster_size = cluster_size;
485 job->len = len;
486 job->perf = *perf;
487
488 block_copy_set_copy_opts(bcs, perf->use_copy_range, compress);
489 block_copy_set_progress_meter(bcs, &job->common.job.progress);
490 block_copy_set_speed(bcs, speed);
491
492 /* Required permissions are taken by copy-before-write filter target */
493 block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL,
494 &error_abort);
495
496 return &job->common;
497
498 error:
499 if (sync_bitmap) {
500 bdrv_reclaim_dirty_bitmap(sync_bitmap, NULL);
501 }
502 if (cbw) {
503 bdrv_cbw_drop(cbw);
504 }
505
506 return NULL;
507 }