]>
Commit | Line | Data |
---|---|---|
98d2c6f2 DM |
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 | ||
80c71a24 | 14 | #include "qemu/osdep.h" |
98d2c6f2 DM |
15 | |
16 | #include "trace.h" | |
17 | #include "block/block.h" | |
18 | #include "block/block_int.h" | |
c87621ea | 19 | #include "block/blockjob_int.h" |
49d3e828 | 20 | #include "block/block_backup.h" |
da34e65c | 21 | #include "qapi/error.h" |
cc7a8ea7 | 22 | #include "qapi/qmp/qerror.h" |
98d2c6f2 | 23 | #include "qemu/ratelimit.h" |
f348b6d1 | 24 | #include "qemu/cutils.h" |
373340b2 | 25 | #include "sysemu/block-backend.h" |
b2f56462 | 26 | #include "qemu/bitmap.h" |
a410a7f1 | 27 | #include "qemu/error-report.h" |
98d2c6f2 | 28 | |
16096a4d | 29 | #define BACKUP_CLUSTER_SIZE_DEFAULT (1 << 16) |
98d2c6f2 | 30 | |
3a75187f VSO |
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; | |
37 | ||
98d2c6f2 DM |
38 | typedef struct BackupBlockJob { |
39 | BlockJob common; | |
5c438bc6 | 40 | BlockBackend *target; |
62aa1fbe | 41 | |
d58d8453 | 42 | BdrvDirtyBitmap *sync_bitmap; |
62aa1fbe JS |
43 | BdrvDirtyBitmap *copy_bitmap; |
44 | ||
fc5d3f84 | 45 | MirrorSyncMode sync_mode; |
c8b56501 | 46 | BitmapSyncMode bitmap_mode; |
98d2c6f2 DM |
47 | BlockdevOnError on_source_error; |
48 | BlockdevOnError on_target_error; | |
49 | CoRwlock flush_rwlock; | |
05df8a6a | 50 | uint64_t len; |
cf79cdf6 | 51 | uint64_t bytes_read; |
16096a4d | 52 | int64_t cluster_size; |
12b3e52e | 53 | NotifierWithReturn before_write; |
98d2c6f2 | 54 | QLIST_HEAD(, CowRequest) inflight_reqs; |
a193b0f0 | 55 | |
9ded4a01 FZ |
56 | bool use_copy_range; |
57 | int64_t copy_range_size; | |
f8d59dfb | 58 | |
a1ed82b4 | 59 | BdrvRequestFlags write_flags; |
7e30dd61 | 60 | bool initializing_bitmap; |
98d2c6f2 DM |
61 | } BackupBlockJob; |
62 | ||
bd21935b KW |
63 | static const BlockJobDriver backup_job_driver; |
64 | ||
98d2c6f2 DM |
65 | /* See if in-flight requests overlap and wait for them to complete */ |
66 | static void coroutine_fn wait_for_overlapping_requests(BackupBlockJob *job, | |
67 | int64_t start, | |
68 | int64_t end) | |
69 | { | |
70 | CowRequest *req; | |
71 | bool retry; | |
72 | ||
73 | do { | |
74 | retry = false; | |
75 | QLIST_FOREACH(req, &job->inflight_reqs, list) { | |
f6ac2078 | 76 | if (end > req->start_byte && start < req->end_byte) { |
1ace7cea | 77 | qemu_co_queue_wait(&req->wait_queue, NULL); |
98d2c6f2 DM |
78 | retry = true; |
79 | break; | |
80 | } | |
81 | } | |
82 | } while (retry); | |
83 | } | |
84 | ||
85 | /* Keep track of an in-flight request */ | |
86 | static void cow_request_begin(CowRequest *req, BackupBlockJob *job, | |
f6ac2078 | 87 | int64_t start, int64_t end) |
98d2c6f2 | 88 | { |
f6ac2078 EB |
89 | req->start_byte = start; |
90 | req->end_byte = end; | |
98d2c6f2 DM |
91 | qemu_co_queue_init(&req->wait_queue); |
92 | QLIST_INSERT_HEAD(&job->inflight_reqs, req, list); | |
93 | } | |
94 | ||
95 | /* Forget about a completed request */ | |
96 | static void cow_request_end(CowRequest *req) | |
97 | { | |
98 | QLIST_REMOVE(req, list); | |
99 | qemu_co_queue_restart_all(&req->wait_queue); | |
100 | } | |
101 | ||
9ded4a01 | 102 | /* Copy range to target with a bounce buffer and return the bytes copied. If |
50d6a8a3 | 103 | * error occurred, return a negative error number */ |
9ded4a01 FZ |
104 | static int coroutine_fn backup_cow_with_bounce_buffer(BackupBlockJob *job, |
105 | int64_t start, | |
106 | int64_t end, | |
107 | bool is_write_notifier, | |
108 | bool *error_is_read, | |
109 | void **bounce_buffer) | |
110 | { | |
111 | int ret; | |
9ded4a01 FZ |
112 | BlockBackend *blk = job->common.blk; |
113 | int nbytes; | |
f8d59dfb | 114 | int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0; |
9ded4a01 | 115 | |
a8389e31 | 116 | assert(QEMU_IS_ALIGNED(start, job->cluster_size)); |
62aa1fbe | 117 | bdrv_reset_dirty_bitmap(job->copy_bitmap, start, job->cluster_size); |
9ded4a01 FZ |
118 | nbytes = MIN(job->cluster_size, job->len - start); |
119 | if (!*bounce_buffer) { | |
120 | *bounce_buffer = blk_blockalign(blk, job->cluster_size); | |
121 | } | |
9ded4a01 | 122 | |
607dbdc4 | 123 | ret = blk_co_pread(blk, start, nbytes, *bounce_buffer, read_flags); |
9ded4a01 FZ |
124 | if (ret < 0) { |
125 | trace_backup_do_cow_read_fail(job, start, ret); | |
126 | if (error_is_read) { | |
127 | *error_is_read = true; | |
128 | } | |
129 | goto fail; | |
130 | } | |
131 | ||
319bd5ed | 132 | ret = blk_co_pwrite(job->target, start, nbytes, *bounce_buffer, |
a1ed82b4 | 133 | job->write_flags); |
9ded4a01 FZ |
134 | if (ret < 0) { |
135 | trace_backup_do_cow_write_fail(job, start, ret); | |
136 | if (error_is_read) { | |
137 | *error_is_read = false; | |
138 | } | |
139 | goto fail; | |
140 | } | |
141 | ||
142 | return nbytes; | |
143 | fail: | |
62aa1fbe | 144 | bdrv_set_dirty_bitmap(job->copy_bitmap, start, job->cluster_size); |
9ded4a01 FZ |
145 | return ret; |
146 | ||
147 | } | |
148 | ||
50d6a8a3 | 149 | /* Copy range to target and return the bytes copied. If error occurred, return a |
9ded4a01 FZ |
150 | * negative error number. */ |
151 | static int coroutine_fn backup_cow_with_offload(BackupBlockJob *job, | |
152 | int64_t start, | |
153 | int64_t end, | |
154 | bool is_write_notifier) | |
155 | { | |
156 | int ret; | |
157 | int nr_clusters; | |
158 | BlockBackend *blk = job->common.blk; | |
159 | int nbytes; | |
f8d59dfb | 160 | int read_flags = is_write_notifier ? BDRV_REQ_NO_SERIALISING : 0; |
9ded4a01 FZ |
161 | |
162 | assert(QEMU_IS_ALIGNED(job->copy_range_size, job->cluster_size)); | |
a8389e31 | 163 | assert(QEMU_IS_ALIGNED(start, job->cluster_size)); |
9ded4a01 FZ |
164 | nbytes = MIN(job->copy_range_size, end - start); |
165 | nr_clusters = DIV_ROUND_UP(nbytes, job->cluster_size); | |
62aa1fbe JS |
166 | bdrv_reset_dirty_bitmap(job->copy_bitmap, start, |
167 | job->cluster_size * nr_clusters); | |
9ded4a01 | 168 | ret = blk_co_copy_range(blk, start, job->target, start, nbytes, |
a1ed82b4 | 169 | read_flags, job->write_flags); |
9ded4a01 FZ |
170 | if (ret < 0) { |
171 | trace_backup_do_cow_copy_range_fail(job, start, ret); | |
62aa1fbe JS |
172 | bdrv_set_dirty_bitmap(job->copy_bitmap, start, |
173 | job->cluster_size * nr_clusters); | |
9ded4a01 FZ |
174 | return ret; |
175 | } | |
176 | ||
177 | return nbytes; | |
178 | } | |
179 | ||
dba8700f JS |
180 | /* |
181 | * Check if the cluster starting at offset is allocated or not. | |
182 | * return via pnum the number of contiguous clusters sharing this allocation. | |
183 | */ | |
184 | static int backup_is_cluster_allocated(BackupBlockJob *s, int64_t offset, | |
185 | int64_t *pnum) | |
186 | { | |
187 | BlockDriverState *bs = blk_bs(s->common.blk); | |
188 | int64_t count, total_count = 0; | |
189 | int64_t bytes = s->len - offset; | |
190 | int ret; | |
191 | ||
192 | assert(QEMU_IS_ALIGNED(offset, s->cluster_size)); | |
193 | ||
194 | while (true) { | |
195 | ret = bdrv_is_allocated(bs, offset, bytes, &count); | |
196 | if (ret < 0) { | |
197 | return ret; | |
198 | } | |
199 | ||
200 | total_count += count; | |
201 | ||
202 | if (ret || count == 0) { | |
203 | /* | |
204 | * ret: partial segment(s) are considered allocated. | |
205 | * otherwise: unallocated tail is treated as an entire segment. | |
206 | */ | |
207 | *pnum = DIV_ROUND_UP(total_count, s->cluster_size); | |
208 | return ret; | |
209 | } | |
210 | ||
211 | /* Unallocated segment(s) with uncertain following segment(s) */ | |
212 | if (total_count >= s->cluster_size) { | |
213 | *pnum = total_count / s->cluster_size; | |
214 | return 0; | |
215 | } | |
216 | ||
217 | offset += count; | |
218 | bytes -= count; | |
219 | } | |
220 | } | |
221 | ||
7e30dd61 JS |
222 | /** |
223 | * Reset bits in copy_bitmap starting at offset if they represent unallocated | |
224 | * data in the image. May reset subsequent contiguous bits. | |
225 | * @return 0 when the cluster at @offset was unallocated, | |
226 | * 1 otherwise, and -ret on error. | |
227 | */ | |
228 | static int64_t backup_bitmap_reset_unallocated(BackupBlockJob *s, | |
229 | int64_t offset, int64_t *count) | |
230 | { | |
231 | int ret; | |
232 | int64_t clusters, bytes, estimate; | |
233 | ||
234 | ret = backup_is_cluster_allocated(s, offset, &clusters); | |
235 | if (ret < 0) { | |
236 | return ret; | |
237 | } | |
238 | ||
239 | bytes = clusters * s->cluster_size; | |
240 | ||
241 | if (!ret) { | |
242 | bdrv_reset_dirty_bitmap(s->copy_bitmap, offset, bytes); | |
243 | estimate = bdrv_get_dirty_count(s->copy_bitmap); | |
244 | job_progress_set_remaining(&s->common.job, estimate); | |
245 | } | |
246 | ||
247 | *count = bytes; | |
248 | return ret; | |
249 | } | |
250 | ||
8543c274 | 251 | static int coroutine_fn backup_do_cow(BackupBlockJob *job, |
03f5d60b | 252 | int64_t offset, uint64_t bytes, |
06c3916b WC |
253 | bool *error_is_read, |
254 | bool is_write_notifier) | |
98d2c6f2 | 255 | { |
98d2c6f2 | 256 | CowRequest cow_request; |
98d2c6f2 | 257 | int ret = 0; |
03f5d60b | 258 | int64_t start, end; /* bytes */ |
9ded4a01 | 259 | void *bounce_buffer = NULL; |
7e30dd61 | 260 | int64_t status_bytes; |
98d2c6f2 DM |
261 | |
262 | qemu_co_rwlock_rdlock(&job->flush_rwlock); | |
263 | ||
03f5d60b EB |
264 | start = QEMU_ALIGN_DOWN(offset, job->cluster_size); |
265 | end = QEMU_ALIGN_UP(bytes + offset, job->cluster_size); | |
98d2c6f2 | 266 | |
03f5d60b | 267 | trace_backup_do_cow_enter(job, start, offset, bytes); |
98d2c6f2 | 268 | |
03f5d60b EB |
269 | wait_for_overlapping_requests(job, start, end); |
270 | cow_request_begin(&cow_request, job, start, end); | |
98d2c6f2 | 271 | |
9ded4a01 | 272 | while (start < end) { |
4a5b91ca HR |
273 | int64_t dirty_end; |
274 | ||
62aa1fbe | 275 | if (!bdrv_dirty_bitmap_get(job->copy_bitmap, start)) { |
03f5d60b | 276 | trace_backup_do_cow_skip(job, start); |
9ded4a01 | 277 | start += job->cluster_size; |
98d2c6f2 DM |
278 | continue; /* already copied */ |
279 | } | |
280 | ||
62aa1fbe JS |
281 | dirty_end = bdrv_dirty_bitmap_next_zero(job->copy_bitmap, start, |
282 | (end - start)); | |
4a5b91ca HR |
283 | if (dirty_end < 0) { |
284 | dirty_end = end; | |
285 | } | |
286 | ||
7e30dd61 JS |
287 | if (job->initializing_bitmap) { |
288 | ret = backup_bitmap_reset_unallocated(job, start, &status_bytes); | |
289 | if (ret == 0) { | |
290 | trace_backup_do_cow_skip_range(job, start, status_bytes); | |
291 | start += status_bytes; | |
292 | continue; | |
293 | } | |
294 | /* Clamp to known allocated region */ | |
295 | dirty_end = MIN(dirty_end, start + status_bytes); | |
296 | } | |
297 | ||
03f5d60b | 298 | trace_backup_do_cow_process(job, start); |
98d2c6f2 | 299 | |
9ded4a01 | 300 | if (job->use_copy_range) { |
4a5b91ca HR |
301 | ret = backup_cow_with_offload(job, start, dirty_end, |
302 | is_write_notifier); | |
9ded4a01 FZ |
303 | if (ret < 0) { |
304 | job->use_copy_range = false; | |
98d2c6f2 | 305 | } |
98d2c6f2 | 306 | } |
9ded4a01 | 307 | if (!job->use_copy_range) { |
4a5b91ca HR |
308 | ret = backup_cow_with_bounce_buffer(job, start, dirty_end, |
309 | is_write_notifier, | |
9ded4a01 | 310 | error_is_read, &bounce_buffer); |
98d2c6f2 DM |
311 | } |
312 | if (ret < 0) { | |
9ded4a01 | 313 | break; |
98d2c6f2 DM |
314 | } |
315 | ||
98d2c6f2 DM |
316 | /* Publish progress, guest I/O counts as progress too. Note that the |
317 | * offset field is an opaque progress value, it is not a disk offset. | |
318 | */ | |
9ded4a01 FZ |
319 | start += ret; |
320 | job->bytes_read += ret; | |
321 | job_progress_update(&job->common.job, ret); | |
322 | ret = 0; | |
98d2c6f2 DM |
323 | } |
324 | ||
98d2c6f2 DM |
325 | if (bounce_buffer) { |
326 | qemu_vfree(bounce_buffer); | |
327 | } | |
328 | ||
329 | cow_request_end(&cow_request); | |
330 | ||
03f5d60b | 331 | trace_backup_do_cow_return(job, offset, bytes, ret); |
98d2c6f2 DM |
332 | |
333 | qemu_co_rwlock_unlock(&job->flush_rwlock); | |
334 | ||
335 | return ret; | |
336 | } | |
337 | ||
338 | static int coroutine_fn backup_before_write_notify( | |
339 | NotifierWithReturn *notifier, | |
340 | void *opaque) | |
341 | { | |
12b3e52e | 342 | BackupBlockJob *job = container_of(notifier, BackupBlockJob, before_write); |
98d2c6f2 DM |
343 | BdrvTrackedRequest *req = opaque; |
344 | ||
5c438bc6 | 345 | assert(req->bs == blk_bs(job->common.blk)); |
03f5d60b EB |
346 | assert(QEMU_IS_ALIGNED(req->offset, BDRV_SECTOR_SIZE)); |
347 | assert(QEMU_IS_ALIGNED(req->bytes, BDRV_SECTOR_SIZE)); | |
793ed47a | 348 | |
03f5d60b | 349 | return backup_do_cow(job, req->offset, req->bytes, NULL, true); |
98d2c6f2 DM |
350 | } |
351 | ||
b976ea3c FZ |
352 | static void backup_cleanup_sync_bitmap(BackupBlockJob *job, int ret) |
353 | { | |
354 | BdrvDirtyBitmap *bm; | |
5c438bc6 | 355 | BlockDriverState *bs = blk_bs(job->common.blk); |
c23909e5 JS |
356 | bool sync = (((ret == 0) || (job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS)) \ |
357 | && (job->bitmap_mode != BITMAP_SYNC_MODE_NEVER)); | |
b976ea3c | 358 | |
c23909e5 | 359 | if (sync) { |
cf0cd293 | 360 | /* |
c23909e5 JS |
361 | * We succeeded, or we always intended to sync the bitmap. |
362 | * Delete this bitmap and install the child. | |
cf0cd293 | 363 | */ |
b976ea3c | 364 | bm = bdrv_dirty_bitmap_abdicate(bs, job->sync_bitmap, NULL); |
c23909e5 JS |
365 | } else { |
366 | /* | |
367 | * We failed, or we never intended to sync the bitmap anyway. | |
368 | * Merge the successor back into the parent, keeping all data. | |
369 | */ | |
370 | bm = bdrv_reclaim_dirty_bitmap(bs, job->sync_bitmap, NULL); | |
371 | } | |
372 | ||
373 | assert(bm); | |
374 | ||
375 | if (ret < 0 && job->bitmap_mode == BITMAP_SYNC_MODE_ALWAYS) { | |
376 | /* If we failed and synced, merge in the bits we didn't copy: */ | |
377 | bdrv_dirty_bitmap_merge_internal(bm, job->copy_bitmap, | |
378 | NULL, true); | |
b976ea3c FZ |
379 | } |
380 | } | |
381 | ||
4ad35181 | 382 | static void backup_commit(Job *job) |
c347b2c6 | 383 | { |
4ad35181 | 384 | BackupBlockJob *s = container_of(job, BackupBlockJob, common.job); |
c347b2c6 JS |
385 | if (s->sync_bitmap) { |
386 | backup_cleanup_sync_bitmap(s, 0); | |
387 | } | |
388 | } | |
389 | ||
4ad35181 | 390 | static void backup_abort(Job *job) |
c347b2c6 | 391 | { |
4ad35181 | 392 | BackupBlockJob *s = container_of(job, BackupBlockJob, common.job); |
c347b2c6 JS |
393 | if (s->sync_bitmap) { |
394 | backup_cleanup_sync_bitmap(s, -1); | |
395 | } | |
396 | } | |
397 | ||
4ad35181 | 398 | static void backup_clean(Job *job) |
e8a40bf7 | 399 | { |
4ad35181 | 400 | BackupBlockJob *s = container_of(job, BackupBlockJob, common.job); |
62aa1fbe | 401 | BlockDriverState *bs = blk_bs(s->common.blk); |
a8389e31 VSO |
402 | |
403 | if (s->copy_bitmap) { | |
62aa1fbe | 404 | bdrv_release_dirty_bitmap(bs, s->copy_bitmap); |
a8389e31 VSO |
405 | s->copy_bitmap = NULL; |
406 | } | |
62aa1fbe JS |
407 | |
408 | assert(s->target); | |
409 | blk_unref(s->target); | |
410 | s->target = NULL; | |
e8a40bf7 JS |
411 | } |
412 | ||
49d3e828 WC |
413 | void backup_do_checkpoint(BlockJob *job, Error **errp) |
414 | { | |
415 | BackupBlockJob *backup_job = container_of(job, BackupBlockJob, common); | |
49d3e828 | 416 | |
bd21935b | 417 | assert(block_job_driver(job) == &backup_job_driver); |
49d3e828 WC |
418 | |
419 | if (backup_job->sync_mode != MIRROR_SYNC_MODE_NONE) { | |
420 | error_setg(errp, "The backup job only supports block checkpoint in" | |
421 | " sync=none mode"); | |
422 | return; | |
423 | } | |
424 | ||
62aa1fbe | 425 | bdrv_set_dirty_bitmap(backup_job->copy_bitmap, 0, backup_job->len); |
49d3e828 WC |
426 | } |
427 | ||
bae8196d PB |
428 | static void backup_drain(BlockJob *job) |
429 | { | |
430 | BackupBlockJob *s = container_of(job, BackupBlockJob, common); | |
431 | ||
432 | /* Need to keep a reference in case blk_drain triggers execution | |
433 | * of backup_complete... | |
434 | */ | |
435 | if (s->target) { | |
436 | BlockBackend *target = s->target; | |
437 | blk_ref(target); | |
438 | blk_drain(target); | |
439 | blk_unref(target); | |
440 | } | |
441 | } | |
442 | ||
98d2c6f2 DM |
443 | static BlockErrorAction backup_error_action(BackupBlockJob *job, |
444 | bool read, int error) | |
445 | { | |
446 | if (read) { | |
81e254dc KW |
447 | return block_job_error_action(&job->common, job->on_source_error, |
448 | true, error); | |
98d2c6f2 | 449 | } else { |
81e254dc KW |
450 | return block_job_error_action(&job->common, job->on_target_error, |
451 | false, error); | |
98d2c6f2 DM |
452 | } |
453 | } | |
454 | ||
d58d8453 JS |
455 | static bool coroutine_fn yield_and_check(BackupBlockJob *job) |
456 | { | |
dee81d51 KW |
457 | uint64_t delay_ns; |
458 | ||
daa7f2f9 | 459 | if (job_is_cancelled(&job->common.job)) { |
d58d8453 JS |
460 | return true; |
461 | } | |
462 | ||
dee81d51 KW |
463 | /* We need to yield even for delay_ns = 0 so that bdrv_drain_all() can |
464 | * return. Without a yield, the VM would not reboot. */ | |
465 | delay_ns = block_job_ratelimit_get_delay(&job->common, job->bytes_read); | |
466 | job->bytes_read = 0; | |
5d43e86e | 467 | job_sleep_ns(&job->common.job, delay_ns); |
d58d8453 | 468 | |
daa7f2f9 | 469 | if (job_is_cancelled(&job->common.job)) { |
d58d8453 JS |
470 | return true; |
471 | } | |
472 | ||
473 | return false; | |
474 | } | |
475 | ||
c334e897 | 476 | static int coroutine_fn backup_loop(BackupBlockJob *job) |
d58d8453 JS |
477 | { |
478 | bool error_is_read; | |
a8389e31 | 479 | int64_t offset; |
62aa1fbe | 480 | BdrvDirtyBitmapIter *bdbi; |
62aa1fbe | 481 | int ret = 0; |
d58d8453 | 482 | |
62aa1fbe JS |
483 | bdbi = bdrv_dirty_iter_new(job->copy_bitmap); |
484 | while ((offset = bdrv_dirty_iter_next(bdbi)) != -1) { | |
53f1c879 VSO |
485 | do { |
486 | if (yield_and_check(job)) { | |
62aa1fbe | 487 | goto out; |
53f1c879 | 488 | } |
a8389e31 | 489 | ret = backup_do_cow(job, offset, |
53f1c879 VSO |
490 | job->cluster_size, &error_is_read, false); |
491 | if (ret < 0 && backup_error_action(job, error_is_read, -ret) == | |
492 | BLOCK_ERROR_ACTION_REPORT) | |
493 | { | |
62aa1fbe | 494 | goto out; |
53f1c879 VSO |
495 | } |
496 | } while (ret < 0); | |
d58d8453 JS |
497 | } |
498 | ||
62aa1fbe JS |
499 | out: |
500 | bdrv_dirty_iter_free(bdbi); | |
501 | return ret; | |
d58d8453 JS |
502 | } |
503 | ||
141cdcdf | 504 | static void backup_init_copy_bitmap(BackupBlockJob *job) |
8cc6dc62 | 505 | { |
141cdcdf JS |
506 | bool ret; |
507 | uint64_t estimate; | |
508 | ||
509 | if (job->sync_mode == MIRROR_SYNC_MODE_BITMAP) { | |
510 | ret = bdrv_dirty_bitmap_merge_internal(job->copy_bitmap, | |
511 | job->sync_bitmap, | |
512 | NULL, true); | |
513 | assert(ret); | |
514 | } else { | |
7e30dd61 JS |
515 | if (job->sync_mode == MIRROR_SYNC_MODE_TOP) { |
516 | /* | |
517 | * We can't hog the coroutine to initialize this thoroughly. | |
518 | * Set a flag and resume work when we are able to yield safely. | |
519 | */ | |
520 | job->initializing_bitmap = true; | |
521 | } | |
141cdcdf JS |
522 | bdrv_set_dirty_bitmap(job->copy_bitmap, 0, job->len); |
523 | } | |
8cc6dc62 | 524 | |
141cdcdf JS |
525 | estimate = bdrv_get_dirty_count(job->copy_bitmap); |
526 | job_progress_set_remaining(&job->common.job, estimate); | |
8cc6dc62 VSO |
527 | } |
528 | ||
68702775 | 529 | static int coroutine_fn backup_run(Job *job, Error **errp) |
98d2c6f2 | 530 | { |
68702775 JS |
531 | BackupBlockJob *s = container_of(job, BackupBlockJob, common.job); |
532 | BlockDriverState *bs = blk_bs(s->common.blk); | |
98d2c6f2 DM |
533 | int ret = 0; |
534 | ||
68702775 JS |
535 | QLIST_INIT(&s->inflight_reqs); |
536 | qemu_co_rwlock_init(&s->flush_rwlock); | |
98d2c6f2 | 537 | |
141cdcdf | 538 | backup_init_copy_bitmap(s); |
8cc6dc62 | 539 | |
68702775 JS |
540 | s->before_write.notify = backup_before_write_notify; |
541 | bdrv_add_before_write_notifier(bs, &s->before_write); | |
98d2c6f2 | 542 | |
7e30dd61 JS |
543 | if (s->sync_mode == MIRROR_SYNC_MODE_TOP) { |
544 | int64_t offset = 0; | |
545 | int64_t count; | |
546 | ||
547 | for (offset = 0; offset < s->len; ) { | |
548 | if (yield_and_check(s)) { | |
549 | ret = -ECANCELED; | |
550 | goto out; | |
551 | } | |
552 | ||
553 | ret = backup_bitmap_reset_unallocated(s, offset, &count); | |
554 | if (ret < 0) { | |
555 | goto out; | |
556 | } | |
557 | ||
558 | offset += count; | |
559 | } | |
560 | s->initializing_bitmap = false; | |
561 | } | |
562 | ||
68702775 | 563 | if (s->sync_mode == MIRROR_SYNC_MODE_NONE) { |
a193b0f0 VSO |
564 | /* All bits are set in copy_bitmap to allow any cluster to be copied. |
565 | * This does not actually require them to be copied. */ | |
68702775 | 566 | while (!job_is_cancelled(job)) { |
fc5d3f84 IM |
567 | /* Yield until the job is cancelled. We just let our before_write |
568 | * notify callback service CoW requests. */ | |
68702775 | 569 | job_yield(job); |
98d2c6f2 | 570 | } |
fc5d3f84 | 571 | } else { |
c334e897 | 572 | ret = backup_loop(s); |
98d2c6f2 DM |
573 | } |
574 | ||
7e30dd61 | 575 | out: |
68702775 | 576 | notifier_with_return_remove(&s->before_write); |
98d2c6f2 DM |
577 | |
578 | /* wait until pending backup_do_cow() calls have completed */ | |
68702775 JS |
579 | qemu_co_rwlock_wrlock(&s->flush_rwlock); |
580 | qemu_co_rwlock_unlock(&s->flush_rwlock); | |
98d2c6f2 | 581 | |
f67432a2 | 582 | return ret; |
98d2c6f2 DM |
583 | } |
584 | ||
a7815a76 | 585 | static const BlockJobDriver backup_job_driver = { |
33e9e9bd KW |
586 | .job_driver = { |
587 | .instance_size = sizeof(BackupBlockJob), | |
252291ea | 588 | .job_type = JOB_TYPE_BACKUP, |
80fa2c75 | 589 | .free = block_job_free, |
b15de828 | 590 | .user_resume = block_job_user_resume, |
b69f777d | 591 | .drain = block_job_drain, |
f67432a2 | 592 | .run = backup_run, |
4ad35181 KW |
593 | .commit = backup_commit, |
594 | .abort = backup_abort, | |
595 | .clean = backup_clean, | |
33e9e9bd | 596 | }, |
a7815a76 JS |
597 | .drain = backup_drain, |
598 | }; | |
599 | ||
ae6b12fa VSO |
600 | static int64_t backup_calculate_cluster_size(BlockDriverState *target, |
601 | Error **errp) | |
602 | { | |
603 | int ret; | |
604 | BlockDriverInfo bdi; | |
605 | ||
606 | /* | |
607 | * If there is no backing file on the target, we cannot rely on COW if our | |
608 | * backup cluster size is smaller than the target cluster size. Even for | |
609 | * targets with a backing file, try to avoid COW if possible. | |
610 | */ | |
611 | ret = bdrv_get_info(target, &bdi); | |
612 | if (ret == -ENOTSUP && !target->backing) { | |
613 | /* Cluster size is not defined */ | |
614 | warn_report("The target block device doesn't provide " | |
615 | "information about the block size and it doesn't have a " | |
616 | "backing file. The default block size of %u bytes is " | |
617 | "used. If the actual block size of the target exceeds " | |
618 | "this default, the backup may be unusable", | |
619 | BACKUP_CLUSTER_SIZE_DEFAULT); | |
620 | return BACKUP_CLUSTER_SIZE_DEFAULT; | |
621 | } else if (ret < 0 && !target->backing) { | |
622 | error_setg_errno(errp, -ret, | |
623 | "Couldn't determine the cluster size of the target image, " | |
624 | "which has no backing file"); | |
625 | error_append_hint(errp, | |
626 | "Aborting, since this may create an unusable destination image\n"); | |
627 | return ret; | |
628 | } else if (ret < 0 && target->backing) { | |
629 | /* Not fatal; just trudge on ahead. */ | |
630 | return BACKUP_CLUSTER_SIZE_DEFAULT; | |
631 | } | |
632 | ||
633 | return MAX(BACKUP_CLUSTER_SIZE_DEFAULT, bdi.cluster_size); | |
634 | } | |
635 | ||
111049a4 | 636 | BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs, |
70559d49 AG |
637 | BlockDriverState *target, int64_t speed, |
638 | MirrorSyncMode sync_mode, BdrvDirtyBitmap *sync_bitmap, | |
c8b56501 | 639 | BitmapSyncMode bitmap_mode, |
13b9414b | 640 | bool compress, |
98d2c6f2 DM |
641 | BlockdevOnError on_source_error, |
642 | BlockdevOnError on_target_error, | |
47970dfb | 643 | int creation_flags, |
097310b5 | 644 | BlockCompletionFunc *cb, void *opaque, |
62c9e416 | 645 | JobTxn *txn, Error **errp) |
98d2c6f2 DM |
646 | { |
647 | int64_t len; | |
91ab6883 | 648 | BackupBlockJob *job = NULL; |
4c9bca7e | 649 | int ret; |
ae6b12fa | 650 | int64_t cluster_size; |
62aa1fbe | 651 | BdrvDirtyBitmap *copy_bitmap = NULL; |
98d2c6f2 DM |
652 | |
653 | assert(bs); | |
654 | assert(target); | |
98d2c6f2 | 655 | |
a6c9365a JS |
656 | /* QMP interface protects us from these cases */ |
657 | assert(sync_mode != MIRROR_SYNC_MODE_INCREMENTAL); | |
658 | assert(sync_bitmap || sync_mode != MIRROR_SYNC_MODE_BITMAP); | |
659 | ||
c29c1dd3 FZ |
660 | if (bs == target) { |
661 | error_setg(errp, "Source and target cannot be the same"); | |
111049a4 | 662 | return NULL; |
c29c1dd3 FZ |
663 | } |
664 | ||
c29c1dd3 FZ |
665 | if (!bdrv_is_inserted(bs)) { |
666 | error_setg(errp, "Device is not inserted: %s", | |
667 | bdrv_get_device_name(bs)); | |
111049a4 | 668 | return NULL; |
c29c1dd3 FZ |
669 | } |
670 | ||
671 | if (!bdrv_is_inserted(target)) { | |
672 | error_setg(errp, "Device is not inserted: %s", | |
673 | bdrv_get_device_name(target)); | |
111049a4 | 674 | return NULL; |
c29c1dd3 FZ |
675 | } |
676 | ||
13b9414b PB |
677 | if (compress && target->drv->bdrv_co_pwritev_compressed == NULL) { |
678 | error_setg(errp, "Compression is not supported for this drive %s", | |
679 | bdrv_get_device_name(target)); | |
111049a4 | 680 | return NULL; |
13b9414b PB |
681 | } |
682 | ||
c29c1dd3 | 683 | if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP_SOURCE, errp)) { |
111049a4 | 684 | return NULL; |
c29c1dd3 FZ |
685 | } |
686 | ||
687 | if (bdrv_op_is_blocked(target, BLOCK_OP_TYPE_BACKUP_TARGET, errp)) { | |
111049a4 | 688 | return NULL; |
c29c1dd3 FZ |
689 | } |
690 | ||
1a2b8b40 | 691 | if (sync_bitmap) { |
b30ffbef JS |
692 | /* If we need to write to this bitmap, check that we can: */ |
693 | if (bitmap_mode != BITMAP_SYNC_MODE_NEVER && | |
694 | bdrv_dirty_bitmap_check(sync_bitmap, BDRV_BITMAP_DEFAULT, errp)) { | |
695 | return NULL; | |
696 | } | |
697 | ||
d58d8453 JS |
698 | /* Create a new bitmap, and freeze/disable this one. */ |
699 | if (bdrv_dirty_bitmap_create_successor(bs, sync_bitmap, errp) < 0) { | |
111049a4 | 700 | return NULL; |
d58d8453 | 701 | } |
d58d8453 JS |
702 | } |
703 | ||
98d2c6f2 DM |
704 | len = bdrv_getlength(bs); |
705 | if (len < 0) { | |
706 | error_setg_errno(errp, -len, "unable to get length for '%s'", | |
707 | bdrv_get_device_name(bs)); | |
d58d8453 | 708 | goto error; |
98d2c6f2 DM |
709 | } |
710 | ||
ae6b12fa VSO |
711 | cluster_size = backup_calculate_cluster_size(target, errp); |
712 | if (cluster_size < 0) { | |
713 | goto error; | |
714 | } | |
715 | ||
62aa1fbe JS |
716 | copy_bitmap = bdrv_create_dirty_bitmap(bs, cluster_size, NULL, errp); |
717 | if (!copy_bitmap) { | |
718 | goto error; | |
719 | } | |
720 | bdrv_disable_dirty_bitmap(copy_bitmap); | |
ae6b12fa | 721 | |
05df8a6a | 722 | /* job->len is fixed, so we can't allow resize */ |
75859b94 | 723 | job = block_job_create(job_id, &backup_job_driver, txn, bs, |
4e9e4323 KW |
724 | BLK_PERM_CONSISTENT_READ, |
725 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | | |
726 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD, | |
c6cc12bf | 727 | speed, creation_flags, cb, opaque, errp); |
98d2c6f2 | 728 | if (!job) { |
d58d8453 | 729 | goto error; |
98d2c6f2 DM |
730 | } |
731 | ||
4e9e4323 | 732 | /* The target must match the source in size, so no resize here either */ |
d861ab3a KW |
733 | job->target = blk_new(job->common.job.aio_context, |
734 | BLK_PERM_WRITE, | |
4e9e4323 KW |
735 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE | |
736 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_GRAPH_MOD); | |
d7086422 KW |
737 | ret = blk_insert_bs(job->target, target, errp); |
738 | if (ret < 0) { | |
739 | goto error; | |
740 | } | |
cf312932 | 741 | blk_set_disable_request_queuing(job->target, true); |
5c438bc6 | 742 | |
98d2c6f2 DM |
743 | job->on_source_error = on_source_error; |
744 | job->on_target_error = on_target_error; | |
fc5d3f84 | 745 | job->sync_mode = sync_mode; |
c8b56501 JS |
746 | job->sync_bitmap = sync_bitmap; |
747 | job->bitmap_mode = bitmap_mode; | |
4c9bca7e | 748 | |
a1ed82b4 VSO |
749 | /* |
750 | * Set write flags: | |
751 | * 1. Detect image-fleecing (and similar) schemes | |
752 | * 2. Handle compression | |
753 | */ | |
754 | job->write_flags = | |
755 | (bdrv_chain_contains(target, bs) ? BDRV_REQ_SERIALISING : 0) | | |
756 | (compress ? BDRV_REQ_WRITE_COMPRESSED : 0); | |
757 | ||
ae6b12fa VSO |
758 | job->cluster_size = cluster_size; |
759 | job->copy_bitmap = copy_bitmap; | |
760 | copy_bitmap = NULL; | |
110571be | 761 | job->use_copy_range = !compress; /* compression isn't supported for it */ |
9ded4a01 FZ |
762 | job->copy_range_size = MIN_NON_ZERO(blk_get_max_transfer(job->common.blk), |
763 | blk_get_max_transfer(job->target)); | |
764 | job->copy_range_size = MAX(job->cluster_size, | |
765 | QEMU_ALIGN_UP(job->copy_range_size, | |
766 | job->cluster_size)); | |
4c9bca7e | 767 | |
4e9e4323 | 768 | /* Required permissions are already taken with target's blk_new() */ |
76d554e2 KW |
769 | block_job_add_bdrv(&job->common, "target", target, 0, BLK_PERM_ALL, |
770 | &error_abort); | |
05df8a6a | 771 | job->len = len; |
111049a4 JS |
772 | |
773 | return &job->common; | |
d58d8453 JS |
774 | |
775 | error: | |
ae6b12fa VSO |
776 | if (copy_bitmap) { |
777 | assert(!job || !job->copy_bitmap); | |
62aa1fbe | 778 | bdrv_release_dirty_bitmap(bs, copy_bitmap); |
ae6b12fa | 779 | } |
d58d8453 JS |
780 | if (sync_bitmap) { |
781 | bdrv_reclaim_dirty_bitmap(bs, sync_bitmap, NULL); | |
782 | } | |
91ab6883 | 783 | if (job) { |
4ad35181 KW |
784 | backup_clean(&job->common.job); |
785 | job_early_fail(&job->common.job); | |
91ab6883 | 786 | } |
111049a4 JS |
787 | |
788 | return NULL; | |
98d2c6f2 | 789 | } |