]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU backup | |
3 | * | |
4 | * Copyright (C) 2013 Proxmox Server Solutions | |
5 | * | |
6 | * Authors: | |
7 | * Dietmar Maurer ([email protected]) | |
8 | * | |
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. | |
11 | * | |
12 | */ | |
13 | ||
14 | #include "qemu/osdep.h" | |
15 | ||
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" | |
28 | ||
29 | #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16) | |
30 | ||
31 | typedef struct BackupBlockJob { | |
32 | BlockJob common; | |
33 | BlockBackend *target; | |
34 | /* bitmap for sync=incremental */ | |
35 | BdrvDirtyBitmap *sync_bitmap; | |
36 | MirrorSyncMode sync_mode; | |
37 | BlockdevOnError on_source_error; | |
38 | BlockdevOnError on_target_error; | |
39 | CoRwlock flush_rwlock; | |
40 | uint64_t len; | |
41 | uint64_t bytes_read; | |
42 | int64_t cluster_size; | |
43 | bool compress; | |
44 | NotifierWithReturn before_write; | |
45 | QLIST_HEAD(, CowRequest) inflight_reqs; | |
46 | ||
47 | HBitmap *copy_bitmap; | |
48 | bool use_copy_range; | |
49 | int64_t copy_range_size; | |
50 | ||
51 | bool serialize_target_writes; | |
52 | } BackupBlockJob; | |
53 | ||
54 | static const BlockJobDriver backup_job_driver; | |
55 | ||
56 | /* See if in-flight requests overlap and wait for them to complete */ | |
57 | static void coroutine_fn wait_for_overlapping_requests(BackupBlockJob *job, | |
58 | int64_t start, | |
59 | int64_t end) | |
60 | { | |
61 | CowRequest *req; | |
62 | bool retry; | |
63 | ||
64 | do { | |
65 | retry = false; | |
66 | QLIST_FOREACH(req, &job->inflight_reqs, list) { | |
67 | if (end > req->start_byte && start < req->end_byte) { | |
68 | qemu_co_queue_wait(&req->wait_queue, NULL); | |
69 | retry = true; | |
70 | break; | |
71 | } | |
72 | } | |
73 | } while (retry); | |
74 | } | |
75 | ||
76 | /* Keep track of an in-flight request */ | |
77 | static void cow_request_begin(CowRequest *req, BackupBlockJob *job, | |
78 | int64_t start, int64_t end) | |
79 | { | |
80 | req->start_byte = start; | |
81 | req->end_byte = end; | |
82 | qemu_co_queue_init(&req->wait_queue); | |
83 | QLIST_INSERT_HEAD(&job->inflight_reqs, req, list); | |
84 | } | |
85 | ||
86 | /* Forget about a completed request */ | |
87 | static void cow_request_end(CowRequest *req) | |
88 | { | |
89 | QLIST_REMOVE(req, list); | |
90 | qemu_co_queue_restart_all(&req->wait_queue); | |
91 | } | |
92 | ||
93 | /* Copy range to target with a bounce buffer and return the bytes copied. If | |
94 | * error occurred, return a negative error number */ | |
95 | static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job, | |
96 | int64_t start, | |
97 | int64_t end, | |
98 | bool is_write_notifier, | |
99 | bool *error_is_read, | |
100 | void **bounce_buffer) | |
101 | { | |
102 | int ret; | |
103 | struct iovec iov; | |
104 | QEMUIOVector qiov; | |
105 | BlockBackend *blk = job->common.blk; | |
106 | int nbytes; | |
107 | int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0; | |
108 | int write_flags = job->serialize_target_writes ? BDRV_REQ_SERIALISING : 0; | |
109 | ||
110 | hbitmap_reset(job->copy_bitmap, start / job->cluster_size, 1); | |
111 | nbytes = MIN(job->cluster_size, job->len - start); | |
112 | if (!*bounce_buffer) { | |
113 | *bounce_buffer = blk_blockalign(blk, job->cluster_size); | |
114 | } | |
115 | iov.iov_base = *bounce_buffer; | |
116 | iov.iov_len = nbytes; | |
117 | qemu_iovec_init_external(&qiov, &iov, 1); | |
118 | ||
119 | ret = blk_co_preadv(blk, start, qiov.size, &qiov, read_flags); | |
120 | if (ret < 0) { | |
121 | trace_backup_do_cow_read_fail(job, start, ret); | |
122 | if (error_is_read) { | |
123 | *error_is_read = true; | |
124 | } | |
125 | goto fail; | |
126 | } | |
127 | ||
128 | if (qemu_iovec_is_zero(&qiov)) { | |
129 | ret = blk_co_pwrite_zeroes(job->target, start, | |
130 | qiov.size, write_flags | BDRV_REQ_MAY_UNMAP); | |
131 | } else { | |
132 | ret = blk_co_pwritev(job->target, start, | |
133 | qiov.size, &qiov, write_flags | | |
134 | (job->compress ? BDRV_REQ_WRITE_COMPRESSED : 0)); | |
135 | } | |
136 | if (ret < 0) { | |
137 | trace_backup_do_cow_write_fail(job, start, ret); | |
138 | if (error_is_read) { | |
139 | *error_is_read = false; | |
140 | } | |
141 | goto fail; | |
142 | } | |
143 | ||
144 | return nbytes; | |
145 | fail: | |
146 | hbitmap_set(job->copy_bitmap, start / job->cluster_size, 1); | |
147 | return ret; | |
148 | ||
149 | } | |
150 | ||
151 | /* Copy range to target and return the bytes copied. If error occurred, return a | |
152 | * negative error number. */ | |
153 | static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job, | |
154 | int64_t start, | |
155 | int64_t end, | |
156 | bool is_write_notifier) | |
157 | { | |
158 | int ret; | |
159 | int nr_clusters; | |
160 | BlockBackend *blk = job->common.blk; | |
161 | int nbytes; | |
162 | int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0; | |
163 | int write_flags = job->serialize_target_writes ? BDRV_REQ_SERIALISING : 0; | |
164 | ||
165 | assert(QEMU_IS_ALIGNED(job->copy_range_size, job->cluster_size)); | |
166 | nbytes = MIN(job->copy_range_size, end - start); | |
167 | nr_clusters = DIV_ROUND_UP(nbytes, job->cluster_size); | |
168 | hbitmap_reset(job->copy_bitmap, start / job->cluster_size, | |
169 | nr_clusters); | |
170 | ret = blk_co_copy_range(blk, start, job->target, start, nbytes, | |
171 | read_flags, write_flags); | |
172 | if (ret < 0) { | |
173 | trace_backup_do_cow_copy_range_fail(job, start, ret); | |
174 | hbitmap_set(job->copy_bitmap, start / job->cluster_size, | |
175 | nr_clusters); | |
176 | return ret; | |
177 | } | |
178 | ||
179 | return nbytes; | |
180 | } | |
181 | ||
182 | static int coroutine_fn backup_do_cow(BackupBlockJob *job, | |
183 | int64_t offset, uint64_t bytes, | |
184 | bool *error_is_read, | |
185 | bool is_write_notifier) | |
186 | { | |
187 | CowRequest cow_request; | |
188 | int ret = 0; | |
189 | int64_t start, end; /* bytes */ | |
190 | void *bounce_buffer = NULL; | |
191 | ||
192 | qemu_co_rwlock_rdlock(&job->flush_rwlock); | |
193 | ||
194 | start = QEMU_ALIGN_DOWN(offset, job->cluster_size); | |
195 | end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size); | |
196 | ||
197 | trace_backup_do_cow_enter(job, start, offset, bytes); | |
198 | ||
199 | wait_for_overlapping_requests(job, start, end); | |
200 | cow_request_begin(&cow_request, job, start, end); | |
201 | ||
202 | while (start < end) { | |
203 | if (!hbitmap_get(job->copy_bitmap, start / job->cluster_size)) { | |
204 | trace_backup_do_cow_skip(job, start); | |
205 | start += job->cluster_size; | |
206 | continue; /* already copied */ | |
207 | } | |
208 | ||
209 | trace_backup_do_cow_process(job, start); | |
210 | ||
211 | if (job->use_copy_range) { | |
212 | ret = backup_cow_with_offload(job, start, end, is_write_notifier); | |
213 | if (ret < 0) { | |
214 | job->use_copy_range = false; | |
215 | } | |
216 | } | |
217 | if (!job->use_copy_range) { | |
218 | ret = backup_cow_with_bounce_buffer(job, start, end, is_write_notifier, | |
219 | error_is_read, &bounce_buffer); | |
220 | } | |
221 | if (ret < 0) { | |
222 | break; | |
223 | } | |
224 | ||
225 | /* Publish progress, guest I/O counts as progress too. Note that the | |
226 | * offset field is an opaque progress value, it is not a disk offset. | |
227 | */ | |
228 | start += ret; | |
229 | job->bytes_read += ret; | |
230 | job_progress_update(&job->common.job, ret); | |
231 | ret = 0; | |
232 | } | |
233 | ||
234 | if (bounce_buffer) { | |
235 | qemu_vfree(bounce_buffer); | |
236 | } | |
237 | ||
238 | cow_request_end(&cow_request); | |
239 | ||
240 | trace_backup_do_cow_return(job, offset, bytes, ret); | |
241 | ||
242 | qemu_co_rwlock_unlock(&job->flush_rwlock); | |
243 | ||
244 | return ret; | |
245 | } | |
246 | ||
247 | static int coroutine_fn backup_before_write_notify( | |
248 | NotifierWithReturn *notifier, | |
249 | void *opaque) | |
250 | { | |
251 | BackupBlockJob *job = container_of(notifier, BackupBlockJob, before_write); | |
252 | BdrvTrackedRequest *req = opaque; | |
253 | ||
254 | assert(req->bs == blk_bs(job->common.blk)); | |
255 | assert(QEMU_IS_ALIGNED(req->offset, BDRV_SECTOR_SIZE)); | |
256 | assert(QEMU_IS_ALIGNED(req->bytes, BDRV_SECTOR_SIZE)); | |
257 | ||
258 | return backup_do_cow(job, req->offset, req->bytes, NULL, true); | |
259 | } | |
260 | ||
261 | static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret) | |
262 | { | |
263 | BdrvDirtyBitmap *bm; | |
264 | BlockDriverState *bs = blk_bs(job->common.blk); | |
265 | ||
266 | if (ret < 0) { | |
267 | /* Merge the successor back into the parent, delete nothing. */ | |
268 | bm = bdrv_reclaim_dirty_bitmap(bs, job->sync_bitmap, NULL); | |
269 | assert(bm); | |
270 | } else { | |
271 | /* Everything is fine, delete this bitmap and install the backup. */ | |
272 | bm = bdrv_dirty_bitmap_abdicate(bs, job->sync_bitmap, NULL); | |
273 | assert(bm); | |
274 | } | |
275 | } | |
276 | ||
277 | static void backup_commit(Job *job) | |
278 | { | |
279 | BackupBlockJob *s = container_of(job, BackupBlockJob, common.job); | |
280 | if (s->sync_bitmap) { | |
281 | backup_cleanup_sync_bitmap(s, 0); | |
282 | } | |
283 | } | |
284 | ||
285 | static void backup_abort(Job *job) | |
286 | { | |
287 | BackupBlockJob *s = container_of(job, BackupBlockJob, common.job); | |
288 | if (s->sync_bitmap) { | |
289 | backup_cleanup_sync_bitmap(s, -1); | |
290 | } | |
291 | } | |
292 | ||
293 | static void backup_clean(Job *job) | |
294 | { | |
295 | BackupBlockJob *s = container_of(job, BackupBlockJob, common.job); | |
296 | assert(s->target); | |
297 | blk_unref(s->target); | |
298 | s->target = NULL; | |
299 | } | |
300 | ||
301 | static void backup_attached_aio_context(BlockJob *job, AioContext *aio_context) | |
302 | { | |
303 | BackupBlockJob *s = container_of(job, BackupBlockJob, common); | |
304 | ||
305 | blk_set_aio_context(s->target, aio_context); | |
306 | } | |
307 | ||
308 | void backup_do_checkpoint(BlockJob *job, Error **errp) | |
309 | { | |
310 | BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common); | |
311 | int64_t len; | |
312 | ||
313 | assert(block_job_driver(job) == &backup_job_driver); | |
314 | ||
315 | if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) { | |
316 | error_setg(errp, "The backup job only supports block checkpoint in" | |
317 | " sync=none mode"); | |
318 | return; | |
319 | } | |
320 | ||
321 | len = DIV_ROUND_UP(backup_job->len, backup_job->cluster_size); | |
322 | hbitmap_set(backup_job->copy_bitmap, 0, len); | |
323 | } | |
324 | ||
325 | void backup_wait_for_overlapping_requests(BlockJob *job, int64_t offset, | |
326 | uint64_t bytes) | |
327 | { | |
328 | BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common); | |
329 | int64_t start, end; | |
330 | ||
331 | assert(block_job_driver(job) == &backup_job_driver); | |
332 | ||
333 | start = QEMU_ALIGN_DOWN(offset, backup_job->cluster_size); | |
334 | end = QEMU_ALIGN_UP(offset + bytes, backup_job->cluster_size); | |
335 | wait_for_overlapping_requests(backup_job, start, end); | |
336 | } | |
337 | ||
338 | void backup_cow_request_begin(CowRequest *req, BlockJob *job, | |
339 | int64_t offset, uint64_t bytes) | |
340 | { | |
341 | BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common); | |
342 | int64_t start, end; | |
343 | ||
344 | assert(block_job_driver(job) == &backup_job_driver); | |
345 | ||
346 | start = QEMU_ALIGN_DOWN(offset, backup_job->cluster_size); | |
347 | end = QEMU_ALIGN_UP(offset + bytes, backup_job->cluster_size); | |
348 | cow_request_begin(req, backup_job, start, end); | |
349 | } | |
350 | ||
351 | void backup_cow_request_end(CowRequest *req) | |
352 | { | |
353 | cow_request_end(req); | |
354 | } | |
355 | ||
356 | static void backup_drain(BlockJob *job) | |
357 | { | |
358 | BackupBlockJob *s = container_of(job, BackupBlockJob, common); | |
359 | ||
360 | /* Need to keep a reference in case blk_drain triggers execution | |
361 | * of backup_complete... | |
362 | */ | |
363 | if (s->target) { | |
364 | BlockBackend *target = s->target; | |
365 | blk_ref(target); | |
366 | blk_drain(target); | |
367 | blk_unref(target); | |
368 | } | |
369 | } | |
370 | ||
371 | static BlockErrorAction backup_error_action(BackupBlockJob *job, | |
372 | bool read, int error) | |
373 | { | |
374 | if (read) { | |
375 | return block_job_error_action(&job->common, job->on_source_error, | |
376 | true, error); | |
377 | } else { | |
378 | return block_job_error_action(&job->common, job->on_target_error, | |
379 | false, error); | |
380 | } | |
381 | } | |
382 | ||
383 | typedef struct { | |
384 | int ret; | |
385 | } BackupCompleteData; | |
386 | ||
387 | static void backup_complete(Job *job, void *opaque) | |
388 | { | |
389 | BackupCompleteData *data = opaque; | |
390 | ||
391 | job_completed(job, data->ret, NULL); | |
392 | g_free(data); | |
393 | } | |
394 | ||
395 | static bool coroutine_fn yield_and_check(BackupBlockJob *job) | |
396 | { | |
397 | uint64_t delay_ns; | |
398 | ||
399 | if (job_is_cancelled(&job->common.job)) { | |
400 | return true; | |
401 | } | |
402 | ||
403 | /* We need to yield even for delay_ns = 0 so that bdrv_drain_all() can | |
404 | * return. Without a yield, the VM would not reboot. */ | |
405 | delay_ns = block_job_ratelimit_get_delay(&job->common, job->bytes_read); | |
406 | job->bytes_read = 0; | |
407 | job_sleep_ns(&job->common.job, delay_ns); | |
408 | ||
409 | if (job_is_cancelled(&job->common.job)) { | |
410 | return true; | |
411 | } | |
412 | ||
413 | return false; | |
414 | } | |
415 | ||
416 | static int coroutine_fn backup_run_incremental(BackupBlockJob *job) | |
417 | { | |
418 | int ret; | |
419 | bool error_is_read; | |
420 | int64_t cluster; | |
421 | HBitmapIter hbi; | |
422 | ||
423 | hbitmap_iter_init(&hbi, job->copy_bitmap, 0); | |
424 | while ((cluster = hbitmap_iter_next(&hbi, true)) != -1) { | |
425 | do { | |
426 | if (yield_and_check(job)) { | |
427 | return 0; | |
428 | } | |
429 | ret = backup_do_cow(job, cluster * job->cluster_size, | |
430 | job->cluster_size, &error_is_read, false); | |
431 | if (ret < 0 && backup_error_action(job, error_is_read, -ret) == | |
432 | BLOCK_ERROR_ACTION_REPORT) | |
433 | { | |
434 | return ret; | |
435 | } | |
436 | } while (ret < 0); | |
437 | } | |
438 | ||
439 | return 0; | |
440 | } | |
441 | ||
442 | /* init copy_bitmap from sync_bitmap */ | |
443 | static void backup_incremental_init_copy_bitmap(BackupBlockJob *job) | |
444 | { | |
445 | BdrvDirtyBitmapIter *dbi; | |
446 | int64_t offset; | |
447 | int64_t end = DIV_ROUND_UP(bdrv_dirty_bitmap_size(job->sync_bitmap), | |
448 | job->cluster_size); | |
449 | ||
450 | dbi = bdrv_dirty_iter_new(job->sync_bitmap); | |
451 | while ((offset = bdrv_dirty_iter_next(dbi)) != -1) { | |
452 | int64_t cluster = offset / job->cluster_size; | |
453 | int64_t next_cluster; | |
454 | ||
455 | offset += bdrv_dirty_bitmap_granularity(job->sync_bitmap); | |
456 | if (offset >= bdrv_dirty_bitmap_size(job->sync_bitmap)) { | |
457 | hbitmap_set(job->copy_bitmap, cluster, end - cluster); | |
458 | break; | |
459 | } | |
460 | ||
461 | offset = bdrv_dirty_bitmap_next_zero(job->sync_bitmap, offset); | |
462 | if (offset == -1) { | |
463 | hbitmap_set(job->copy_bitmap, cluster, end - cluster); | |
464 | break; | |
465 | } | |
466 | ||
467 | next_cluster = DIV_ROUND_UP(offset, job->cluster_size); | |
468 | hbitmap_set(job->copy_bitmap, cluster, next_cluster - cluster); | |
469 | if (next_cluster >= end) { | |
470 | break; | |
471 | } | |
472 | ||
473 | bdrv_set_dirty_iter(dbi, next_cluster * job->cluster_size); | |
474 | } | |
475 | ||
476 | /* TODO job_progress_set_remaining() would make more sense */ | |
477 | job_progress_update(&job->common.job, | |
478 | job->len - hbitmap_count(job->copy_bitmap) * job->cluster_size); | |
479 | ||
480 | bdrv_dirty_iter_free(dbi); | |
481 | } | |
482 | ||
483 | static int coroutine_fn backup_run(Job *opaque_job, Error **errp) | |
484 | { | |
485 | BackupBlockJob *job = container_of(opaque_job, BackupBlockJob, common.job); | |
486 | BackupCompleteData *data; | |
487 | BlockDriverState *bs = blk_bs(job->common.blk); | |
488 | int64_t offset, nb_clusters; | |
489 | int ret = 0; | |
490 | ||
491 | QLIST_INIT(&job->inflight_reqs); | |
492 | qemu_co_rwlock_init(&job->flush_rwlock); | |
493 | ||
494 | nb_clusters = DIV_ROUND_UP(job->len, job->cluster_size); | |
495 | job_progress_set_remaining(&job->common.job, job->len); | |
496 | ||
497 | job->copy_bitmap = hbitmap_alloc(nb_clusters, 0); | |
498 | if (job->sync_mode == MIRROR_SYNC_MODE_INCREMENTAL) { | |
499 | backup_incremental_init_copy_bitmap(job); | |
500 | } else { | |
501 | hbitmap_set(job->copy_bitmap, 0, nb_clusters); | |
502 | } | |
503 | ||
504 | ||
505 | job->before_write.notify = backup_before_write_notify; | |
506 | bdrv_add_before_write_notifier(bs, &job->before_write); | |
507 | ||
508 | if (job->sync_mode == MIRROR_SYNC_MODE_NONE) { | |
509 | /* All bits are set in copy_bitmap to allow any cluster to be copied. | |
510 | * This does not actually require them to be copied. */ | |
511 | while (!job_is_cancelled(&job->common.job)) { | |
512 | /* Yield until the job is cancelled. We just let our before_write | |
513 | * notify callback service CoW requests. */ | |
514 | job_yield(&job->common.job); | |
515 | } | |
516 | } else if (job->sync_mode == MIRROR_SYNC_MODE_INCREMENTAL) { | |
517 | ret = backup_run_incremental(job); | |
518 | } else { | |
519 | /* Both FULL and TOP SYNC_MODE's require copying.. */ | |
520 | for (offset = 0; offset < job->len; | |
521 | offset += job->cluster_size) { | |
522 | bool error_is_read; | |
523 | int alloced = 0; | |
524 | ||
525 | if (yield_and_check(job)) { | |
526 | break; | |
527 | } | |
528 | ||
529 | if (job->sync_mode == MIRROR_SYNC_MODE_TOP) { | |
530 | int i; | |
531 | int64_t n; | |
532 | ||
533 | /* Check to see if these blocks are already in the | |
534 | * backing file. */ | |
535 | ||
536 | for (i = 0; i < job->cluster_size;) { | |
537 | /* bdrv_is_allocated() only returns true/false based | |
538 | * on the first set of sectors it comes across that | |
539 | * are are all in the same state. | |
540 | * For that reason we must verify each sector in the | |
541 | * backup cluster length. We end up copying more than | |
542 | * needed but at some point that is always the case. */ | |
543 | alloced = | |
544 | bdrv_is_allocated(bs, offset + i, | |
545 | job->cluster_size - i, &n); | |
546 | i += n; | |
547 | ||
548 | if (alloced || n == 0) { | |
549 | break; | |
550 | } | |
551 | } | |
552 | ||
553 | /* If the above loop never found any sectors that are in | |
554 | * the topmost image, skip this backup. */ | |
555 | if (alloced == 0) { | |
556 | continue; | |
557 | } | |
558 | } | |
559 | /* FULL sync mode we copy the whole drive. */ | |
560 | if (alloced < 0) { | |
561 | ret = alloced; | |
562 | } else { | |
563 | ret = backup_do_cow(job, offset, job->cluster_size, | |
564 | &error_is_read, false); | |
565 | } | |
566 | if (ret < 0) { | |
567 | /* Depending on error action, fail now or retry cluster */ | |
568 | BlockErrorAction action = | |
569 | backup_error_action(job, error_is_read, -ret); | |
570 | if (action == BLOCK_ERROR_ACTION_REPORT) { | |
571 | break; | |
572 | } else { | |
573 | offset -= job->cluster_size; | |
574 | continue; | |
575 | } | |
576 | } | |
577 | } | |
578 | } | |
579 | ||
580 | notifier_with_return_remove(&job->before_write); | |
581 | ||
582 | /* wait until pending backup_do_cow() calls have completed */ | |
583 | qemu_co_rwlock_wrlock(&job->flush_rwlock); | |
584 | qemu_co_rwlock_unlock(&job->flush_rwlock); | |
585 | hbitmap_free(job->copy_bitmap); | |
586 | ||
587 | data = g_malloc(sizeof(*data)); | |
588 | data->ret = ret; | |
589 | job_defer_to_main_loop(&job->common.job, backup_complete, data); | |
590 | return ret; | |
591 | } | |
592 | ||
593 | static const BlockJobDriver backup_job_driver = { | |
594 | .job_driver = { | |
595 | .instance_size = sizeof(BackupBlockJob), | |
596 | .job_type = JOB_TYPE_BACKUP, | |
597 | .free = block_job_free, | |
598 | .user_resume = block_job_user_resume, | |
599 | .drain = block_job_drain, | |
600 | .run = backup_run, | |
601 | .commit = backup_commit, | |
602 | .abort = backup_abort, | |
603 | .clean = backup_clean, | |
604 | }, | |
605 | .attached_aio_context = backup_attached_aio_context, | |
606 | .drain = backup_drain, | |
607 | }; | |
608 | ||
609 | BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs, | |
610 | BlockDriverState *target, int64_t speed, | |
611 | MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap, | |
612 | bool compress, | |
613 | BlockdevOnError on_source_error, | |
614 | BlockdevOnError on_target_error, | |
615 | int creation_flags, | |
616 | BlockCompletionFunc *cb, void *opaque, | |
617 | JobTxn *txn, Error **errp) | |
618 | { | |
619 | int64_t len; | |
620 | BlockDriverInfo bdi; | |
621 | BackupBlockJob *job = NULL; | |
622 | int ret; | |
623 | ||
624 | assert(bs); | |
625 | assert(target); | |
626 | ||
627 | if (bs == target) { | |
628 | error_setg(errp, "Source and target cannot be the same"); | |
629 | return NULL; | |
630 | } | |
631 | ||
632 | if (!bdrv_is_inserted(bs)) { | |
633 | error_setg(errp, "Device is not inserted: %s", | |
634 | bdrv_get_device_name(bs)); | |
635 | return NULL; | |
636 | } | |
637 | ||
638 | if (!bdrv_is_inserted(target)) { | |
639 | error_setg(errp, "Device is not inserted: %s", | |
640 | bdrv_get_device_name(target)); | |
641 | return NULL; | |
642 | } | |
643 | ||
644 | if (compress && target->drv->bdrv_co_pwritev_compressed == NULL) { | |
645 | error_setg(errp, "Compression is not supported for this drive %s", | |
646 | bdrv_get_device_name(target)); | |
647 | return NULL; | |
648 | } | |
649 | ||
650 | if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) { | |
651 | return NULL; | |
652 | } | |
653 | ||
654 | if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) { | |
655 | return NULL; | |
656 | } | |
657 | ||
658 | if (sync_mode == MIRROR_SYNC_MODE_INCREMENTAL) { | |
659 | if (!sync_bitmap) { | |
660 | error_setg(errp, "must provide a valid bitmap name for " | |
661 | "\"incremental\" sync mode"); | |
662 | return NULL; | |
663 | } | |
664 | ||
665 | /* Create a new bitmap, and freeze/disable this one. */ | |
666 | if (bdrv_dirty_bitmap_create_successor(bs, sync_bitmap, errp) < 0) { | |
667 | return NULL; | |
668 | } | |
669 | } else if (sync_bitmap) { | |
670 | error_setg(errp, | |
671 | "a sync_bitmap was provided to backup_run, " | |
672 | "but received an incompatible sync_mode (%s)", | |
673 | MirrorSyncMode_str(sync_mode)); | |
674 | return NULL; | |
675 | } | |
676 | ||
677 | len = bdrv_getlength(bs); | |
678 | if (len < 0) { | |
679 | error_setg_errno(errp, -len, "unable to get length for '%s'", | |
680 | bdrv_get_device_name(bs)); | |
681 | goto error; | |
682 | } | |
683 | ||
684 | /* job->len is fixed, so we can't allow resize */ | |
685 | job = block_job_create(job_id, &backup_job_driver, txn, bs, | |
686 | BLK_PERM_CONSISTENT_READ, | |
687 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | | |
688 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD, | |
689 | speed, creation_flags, cb, opaque, errp); | |
690 | if (!job) { | |
691 | goto error; | |
692 | } | |
693 | ||
694 | /* The target must match the source in size, so no resize here either */ | |
695 | job->target = blk_new(BLK_PERM_WRITE, | |
696 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | | |
697 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD); | |
698 | ret = blk_insert_bs(job->target, target, errp); | |
699 | if (ret < 0) { | |
700 | goto error; | |
701 | } | |
702 | ||
703 | job->on_source_error = on_source_error; | |
704 | job->on_target_error = on_target_error; | |
705 | job->sync_mode = sync_mode; | |
706 | job->sync_bitmap = sync_mode == MIRROR_SYNC_MODE_INCREMENTAL ? | |
707 | sync_bitmap : NULL; | |
708 | job->compress = compress; | |
709 | ||
710 | /* Detect image-fleecing (and similar) schemes */ | |
711 | job->serialize_target_writes = bdrv_chain_contains(target, bs); | |
712 | ||
713 | /* If there is no backing file on the target, we cannot rely on COW if our | |
714 | * backup cluster size is smaller than the target cluster size. Even for | |
715 | * targets with a backing file, try to avoid COW if possible. */ | |
716 | ret = bdrv_get_info(target, &bdi); | |
717 | if (ret == -ENOTSUP && !target->backing) { | |
718 | /* Cluster size is not defined */ | |
719 | warn_report("The target block device doesn't provide " | |
720 | "information about the block size and it doesn't have a " | |
721 | "backing file. The default block size of %u bytes is " | |
722 | "used. If the actual block size of the target exceeds " | |
723 | "this default, the backup may be unusable", | |
724 | BACKUP_CLUSTER_SIZE_DEFAULT); | |
725 | job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT; | |
726 | } else if (ret < 0 && !target->backing) { | |
727 | error_setg_errno(errp, -ret, | |
728 | "Couldn't determine the cluster size of the target image, " | |
729 | "which has no backing file"); | |
730 | error_append_hint(errp, | |
731 | "Aborting, since this may create an unusable destination image\n"); | |
732 | goto error; | |
733 | } else if (ret < 0 && target->backing) { | |
734 | /* Not fatal; just trudge on ahead. */ | |
735 | job->cluster_size = BACKUP_CLUSTER_SIZE_DEFAULT; | |
736 | } else { | |
737 | job->cluster_size = MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size); | |
738 | } | |
739 | job->use_copy_range = true; | |
740 | job->copy_range_size = MIN_NON_ZERO(blk_get_max_transfer(job->common.blk), | |
741 | blk_get_max_transfer(job->target)); | |
742 | job->copy_range_size = MAX(job->cluster_size, | |
743 | QEMU_ALIGN_UP(job->copy_range_size, | |
744 | job->cluster_size)); | |
745 | ||
746 | /* Required permissions are already taken with target's blk_new() */ | |
747 | block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL, | |
748 | &error_abort); | |
749 | job->len = len; | |
750 | ||
751 | return &job->common; | |
752 | ||
753 | error: | |
754 | if (sync_bitmap) { | |
755 | bdrv_reclaim_dirty_bitmap(bs, sync_bitmap, NULL); | |
756 | } | |
757 | if (job) { | |
758 | backup_clean(&job->common.job); | |
759 | job_early_fail(&job->common.job); | |
760 | } | |
761 | ||
762 | return NULL; | |
763 | } |