]>
Commit | Line | Data |
---|---|---|
747ff602 JC |
1 | /* |
2 | * Live block commit | |
3 | * | |
4 | * Copyright Red Hat, Inc. 2012 | |
5 | * | |
6 | * Authors: | |
7 | * Jeff Cody <jcody@redhat.com> | |
8 | * Based on stream.c by Stefan Hajnoczi | |
9 | * | |
10 | * This work is licensed under the terms of the GNU LGPL, version 2 or later. | |
11 | * See the COPYING.LIB file in the top-level directory. | |
12 | * | |
13 | */ | |
14 | ||
80c71a24 | 15 | #include "qemu/osdep.h" |
dcbf37ce | 16 | #include "qemu/cutils.h" |
747ff602 | 17 | #include "trace.h" |
737e150e | 18 | #include "block/block_int.h" |
c87621ea | 19 | #include "block/blockjob_int.h" |
da34e65c | 20 | #include "qapi/error.h" |
747ff602 | 21 | #include "qemu/ratelimit.h" |
5df022cf | 22 | #include "qemu/memalign.h" |
373340b2 | 23 | #include "sysemu/block-backend.h" |
747ff602 JC |
24 | |
25 | enum { | |
26 | /* | |
27 | * Size of data buffer for populating the image file. This should be large | |
28 | * enough to process multiple clusters in a single call, so that populating | |
29 | * contiguous regions of the image is efficient. | |
30 | */ | |
31 | COMMIT_BUFFER_SIZE = 512 * 1024, /* in bytes */ | |
32 | }; | |
33 | ||
747ff602 JC |
34 | typedef struct CommitBlockJob { |
35 | BlockJob common; | |
8dfba279 | 36 | BlockDriverState *commit_top_bs; |
4653456a KW |
37 | BlockBackend *top; |
38 | BlockBackend *base; | |
22dffcbe | 39 | BlockDriverState *base_bs; |
9a71b9de | 40 | BlockDriverState *base_overlay; |
92aa5c6d | 41 | BlockdevOnError on_error; |
e70cdc57 | 42 | bool base_read_only; |
df827336 | 43 | bool chain_frozen; |
54e26900 | 44 | char *backing_file_str; |
747ff602 JC |
45 | } CommitBlockJob; |
46 | ||
22dffcbe | 47 | static int commit_prepare(Job *job) |
747ff602 | 48 | { |
1908a559 | 49 | CommitBlockJob *s = container_of(job, CommitBlockJob, common.job); |
19ebd13e | 50 | |
9275fc72 | 51 | bdrv_graph_rdlock_main_loop(); |
df827336 AG |
52 | bdrv_unfreeze_backing_chain(s->commit_top_bs, s->base_bs); |
53 | s->chain_frozen = false; | |
9275fc72 | 54 | bdrv_graph_rdunlock_main_loop(); |
df827336 | 55 | |
8dfba279 KW |
56 | /* Remove base node parent that still uses BLK_PERM_WRITE/RESIZE before |
57 | * the normal backing chain can be restored. */ | |
58 | blk_unref(s->base); | |
22dffcbe JS |
59 | s->base = NULL; |
60 | ||
61 | /* FIXME: bdrv_drop_intermediate treats total failures and partial failures | |
62 | * identically. Further work is needed to disambiguate these cases. */ | |
63 | return bdrv_drop_intermediate(s->commit_top_bs, s->base_bs, | |
64 | s->backing_file_str); | |
65 | } | |
9e85cd5c | 66 | |
22dffcbe JS |
67 | static void commit_abort(Job *job) |
68 | { | |
69 | CommitBlockJob *s = container_of(job, CommitBlockJob, common.job); | |
70 | BlockDriverState *top_bs = blk_bs(s->top); | |
ccd6a379 | 71 | BlockDriverState *commit_top_backing_bs; |
22dffcbe | 72 | |
df827336 | 73 | if (s->chain_frozen) { |
9275fc72 | 74 | bdrv_graph_rdlock_main_loop(); |
df827336 | 75 | bdrv_unfreeze_backing_chain(s->commit_top_bs, s->base_bs); |
9275fc72 | 76 | bdrv_graph_rdunlock_main_loop(); |
df827336 AG |
77 | } |
78 | ||
22dffcbe JS |
79 | /* Make sure commit_top_bs and top stay around until bdrv_replace_node() */ |
80 | bdrv_ref(top_bs); | |
81 | bdrv_ref(s->commit_top_bs); | |
82 | ||
83 | if (s->base) { | |
84 | blk_unref(s->base); | |
9e85cd5c SH |
85 | } |
86 | ||
22dffcbe JS |
87 | /* free the blockers on the intermediate nodes so that bdrv_replace_nodes |
88 | * can succeed */ | |
89 | block_job_remove_all_bdrv(&s->common); | |
90 | ||
91 | /* If bdrv_drop_intermediate() failed (or was not invoked), remove the | |
92 | * commit filter driver from the backing chain now. Do this as the final | |
93 | * step so that the 'consistent read' permission can be granted. | |
94 | * | |
95 | * XXX Can (or should) we somehow keep 'consistent read' blocked even | |
96 | * after the failed/cancelled commit job is gone? If we already wrote | |
97 | * something to base, the intermediate images aren't valid any more. */ | |
004915a9 | 98 | bdrv_graph_rdlock_main_loop(); |
ccd6a379 | 99 | commit_top_backing_bs = s->commit_top_bs->backing->bs; |
004915a9 KW |
100 | bdrv_graph_rdunlock_main_loop(); |
101 | ||
ccd6a379 KW |
102 | bdrv_drained_begin(commit_top_backing_bs); |
103 | bdrv_graph_wrlock(commit_top_backing_bs); | |
104 | bdrv_replace_node(s->commit_top_bs, commit_top_backing_bs, &error_abort); | |
6bc0bcc8 | 105 | bdrv_graph_wrunlock(commit_top_backing_bs); |
ccd6a379 | 106 | bdrv_drained_end(commit_top_backing_bs); |
22dffcbe JS |
107 | |
108 | bdrv_unref(s->commit_top_bs); | |
109 | bdrv_unref(top_bs); | |
110 | } | |
111 | ||
112 | static void commit_clean(Job *job) | |
113 | { | |
114 | CommitBlockJob *s = container_of(job, CommitBlockJob, common.job); | |
115 | ||
9e85cd5c SH |
116 | /* restore base open flags here if appropriate (e.g., change the base back |
117 | * to r/o). These reopens do not need to be atomic, since we won't abort | |
118 | * even on failure here */ | |
e70cdc57 AG |
119 | if (s->base_read_only) { |
120 | bdrv_reopen_set_read_only(s->base_bs, true, NULL); | |
9e85cd5c | 121 | } |
22dffcbe | 122 | |
9e85cd5c | 123 | g_free(s->backing_file_str); |
4653456a | 124 | blk_unref(s->top); |
9e85cd5c SH |
125 | } |
126 | ||
f67432a2 | 127 | static int coroutine_fn commit_run(Job *job, Error **errp) |
9e85cd5c | 128 | { |
f67432a2 | 129 | CommitBlockJob *s = container_of(job, CommitBlockJob, common.job); |
317a6676 | 130 | int64_t offset; |
747ff602 | 131 | int ret = 0; |
51b0a488 | 132 | int64_t n = 0; /* bytes */ |
71701708 | 133 | QEMU_AUTO_VFREE void *buf = NULL; |
05df8a6a | 134 | int64_t len, base_len; |
747ff602 | 135 | |
c86422c5 | 136 | len = blk_co_getlength(s->top); |
05df8a6a | 137 | if (len < 0) { |
71701708 | 138 | return len; |
747ff602 | 139 | } |
30a5c887 | 140 | job_progress_set_remaining(&s->common.job, len); |
747ff602 | 141 | |
c86422c5 | 142 | base_len = blk_co_getlength(s->base); |
747ff602 | 143 | if (base_len < 0) { |
71701708 | 144 | return base_len; |
747ff602 JC |
145 | } |
146 | ||
05df8a6a | 147 | if (base_len < len) { |
a0667887 | 148 | ret = blk_co_truncate(s->base, len, false, PREALLOC_MODE_OFF, 0, NULL); |
747ff602 | 149 | if (ret) { |
71701708 | 150 | return ret; |
747ff602 JC |
151 | } |
152 | } | |
153 | ||
4653456a | 154 | buf = blk_blockalign(s->top, COMMIT_BUFFER_SIZE); |
747ff602 | 155 | |
05df8a6a | 156 | for (offset = 0; offset < len; offset += n) { |
747ff602 | 157 | bool copy; |
9ad1e79f | 158 | bool error_in_source = true; |
747ff602 | 159 | |
747ff602 | 160 | /* Note that even when no rate limit is applied we need to yield |
c57b6656 | 161 | * with no pending I/O here so that bdrv_drain_all() returns. |
747ff602 | 162 | */ |
018e5987 | 163 | block_job_ratelimit_sleep(&s->common); |
daa7f2f9 | 164 | if (job_is_cancelled(&s->common.job)) { |
747ff602 JC |
165 | break; |
166 | } | |
167 | /* Copy if allocated above the base */ | |
ff7e261b EGE |
168 | ret = blk_co_is_allocated_above(s->top, s->base_overlay, true, |
169 | offset, COMMIT_BUFFER_SIZE, &n); | |
a92b1b06 | 170 | copy = (ret > 0); |
51b0a488 | 171 | trace_commit_one_iteration(s, offset, n, ret); |
747ff602 | 172 | if (copy) { |
0c42e175 KW |
173 | assert(n < SIZE_MAX); |
174 | ||
175 | ret = blk_co_pread(s->top, offset, n, buf, 0); | |
176 | if (ret >= 0) { | |
177 | ret = blk_co_pwrite(s->base, offset, n, buf, 0); | |
9ad1e79f KW |
178 | if (ret < 0) { |
179 | error_in_source = false; | |
180 | } | |
0c42e175 | 181 | } |
747ff602 JC |
182 | } |
183 | if (ret < 0) { | |
1e8fb7f1 | 184 | BlockErrorAction action = |
9ad1e79f KW |
185 | block_job_error_action(&s->common, s->on_error, |
186 | error_in_source, -ret); | |
1e8fb7f1 | 187 | if (action == BLOCK_ERROR_ACTION_REPORT) { |
71701708 | 188 | return ret; |
747ff602 JC |
189 | } else { |
190 | n = 0; | |
191 | continue; | |
192 | } | |
193 | } | |
194 | /* Publish progress */ | |
30a5c887 | 195 | job_progress_update(&s->common.job, n); |
f14a39cc | 196 | |
dee81d51 | 197 | if (copy) { |
018e5987 | 198 | block_job_ratelimit_processed_bytes(&s->common, n); |
f14a39cc | 199 | } |
747ff602 JC |
200 | } |
201 | ||
71701708 | 202 | return 0; |
747ff602 JC |
203 | } |
204 | ||
3fc4b10a | 205 | static const BlockJobDriver commit_job_driver = { |
33e9e9bd KW |
206 | .job_driver = { |
207 | .instance_size = sizeof(CommitBlockJob), | |
252291ea | 208 | .job_type = JOB_TYPE_COMMIT, |
80fa2c75 | 209 | .free = block_job_free, |
b15de828 | 210 | .user_resume = block_job_user_resume, |
f67432a2 | 211 | .run = commit_run, |
22dffcbe JS |
212 | .prepare = commit_prepare, |
213 | .abort = commit_abort, | |
214 | .clean = commit_clean | |
33e9e9bd | 215 | }, |
747ff602 JC |
216 | }; |
217 | ||
b9b10c35 KW |
218 | static int coroutine_fn GRAPH_RDLOCK |
219 | bdrv_commit_top_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
220 | QEMUIOVector *qiov, BdrvRequestFlags flags) | |
8dfba279 KW |
221 | { |
222 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
223 | } | |
224 | ||
004915a9 | 225 | static GRAPH_RDLOCK void bdrv_commit_top_refresh_filename(BlockDriverState *bs) |
dcbf37ce | 226 | { |
dcbf37ce KW |
227 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), |
228 | bs->backing->bs->filename); | |
229 | } | |
91965658 | 230 | |
8dfba279 | 231 | static void bdrv_commit_top_child_perm(BlockDriverState *bs, BdrvChild *c, |
bf8e925e | 232 | BdrvChildRole role, |
e0995dc3 | 233 | BlockReopenQueue *reopen_queue, |
8dfba279 KW |
234 | uint64_t perm, uint64_t shared, |
235 | uint64_t *nperm, uint64_t *nshared) | |
236 | { | |
237 | *nperm = 0; | |
238 | *nshared = BLK_PERM_ALL; | |
239 | } | |
240 | ||
241 | /* Dummy node that provides consistent read to its users without requiring it | |
242 | * from its backing file and that allows writes on the backing file chain. */ | |
243 | static BlockDriver bdrv_commit_top = { | |
91965658 KW |
244 | .format_name = "commit_top", |
245 | .bdrv_co_preadv = bdrv_commit_top_preadv, | |
dcbf37ce | 246 | .bdrv_refresh_filename = bdrv_commit_top_refresh_filename, |
91965658 | 247 | .bdrv_child_perm = bdrv_commit_top_child_perm, |
6540fd15 HR |
248 | |
249 | .is_filter = true, | |
046fd84f | 250 | .filtered_child_is_backing = true, |
8dfba279 KW |
251 | }; |
252 | ||
fd62c609 | 253 | void commit_start(const char *job_id, BlockDriverState *bs, |
5360782d JS |
254 | BlockDriverState *base, BlockDriverState *top, |
255 | int creation_flags, int64_t speed, | |
8254b6d9 | 256 | BlockdevOnError on_error, const char *backing_file_str, |
0db832f4 | 257 | const char *filter_node_name, Error **errp) |
747ff602 JC |
258 | { |
259 | CommitBlockJob *s; | |
3e4c5122 | 260 | BlockDriverState *iter; |
8dfba279 | 261 | BlockDriverState *commit_top_bs = NULL; |
9a71b9de | 262 | BlockDriverState *filtered_base; |
9a71b9de HR |
263 | int64_t base_size, top_size; |
264 | uint64_t base_perms, iter_shared_perms; | |
d7086422 | 265 | int ret; |
747ff602 | 266 | |
b4ad82aa EGE |
267 | GLOBAL_STATE_CODE(); |
268 | ||
18da7f94 | 269 | assert(top != bs); |
ad74751f | 270 | bdrv_graph_rdlock_main_loop(); |
9a71b9de | 271 | if (bdrv_skip_filters(top) == bdrv_skip_filters(base)) { |
747ff602 | 272 | error_setg(errp, "Invalid files for merge: top and base are the same"); |
ad74751f | 273 | bdrv_graph_rdunlock_main_loop(); |
747ff602 JC |
274 | return; |
275 | } | |
ad74751f | 276 | bdrv_graph_rdunlock_main_loop(); |
747ff602 | 277 | |
9a71b9de HR |
278 | base_size = bdrv_getlength(base); |
279 | if (base_size < 0) { | |
280 | error_setg_errno(errp, -base_size, "Could not inquire base image size"); | |
281 | return; | |
282 | } | |
283 | ||
284 | top_size = bdrv_getlength(top); | |
285 | if (top_size < 0) { | |
286 | error_setg_errno(errp, -top_size, "Could not inquire top image size"); | |
287 | return; | |
288 | } | |
289 | ||
290 | base_perms = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE; | |
291 | if (base_size < top_size) { | |
292 | base_perms |= BLK_PERM_RESIZE; | |
293 | } | |
294 | ||
75859b94 | 295 | s = block_job_create(job_id, &commit_job_driver, NULL, bs, 0, BLK_PERM_ALL, |
5360782d | 296 | speed, creation_flags, NULL, NULL, errp); |
834fe28d AG |
297 | if (!s) { |
298 | return; | |
299 | } | |
300 | ||
bde70715 | 301 | /* convert base to r/w, if necessary */ |
e70cdc57 AG |
302 | s->base_read_only = bdrv_is_read_only(base); |
303 | if (s->base_read_only) { | |
304 | if (bdrv_reopen_set_read_only(base, false, errp) != 0) { | |
d7086422 | 305 | goto fail; |
747ff602 JC |
306 | } |
307 | } | |
308 | ||
8dfba279 KW |
309 | /* Insert commit_top block node above top, so we can block consistent read |
310 | * on the backing chain below it */ | |
0db832f4 KW |
311 | commit_top_bs = bdrv_new_open_driver(&bdrv_commit_top, filter_node_name, 0, |
312 | errp); | |
8dfba279 KW |
313 | if (commit_top_bs == NULL) { |
314 | goto fail; | |
315 | } | |
d3c8c674 KW |
316 | if (!filter_node_name) { |
317 | commit_top_bs->implicit = true; | |
318 | } | |
e5182c1c HR |
319 | |
320 | /* So that we can always drop this node */ | |
321 | commit_top_bs->never_freeze = true; | |
322 | ||
0d0676a1 | 323 | commit_top_bs->total_sectors = top->total_sectors; |
8dfba279 | 324 | |
934aee14 | 325 | ret = bdrv_append(commit_top_bs, top, errp); |
ae9d4417 | 326 | bdrv_unref(commit_top_bs); /* referenced by new parents or failed */ |
934aee14 | 327 | if (ret < 0) { |
b69f00dd | 328 | commit_top_bs = NULL; |
b69f00dd FZ |
329 | goto fail; |
330 | } | |
8dfba279 KW |
331 | |
332 | s->commit_top_bs = commit_top_bs; | |
747ff602 | 333 | |
9a71b9de HR |
334 | /* |
335 | * Block all nodes between top and base, because they will | |
336 | * disappear from the chain after this operation. | |
337 | * Note that this assumes that the user is fine with removing all | |
338 | * nodes (including R/W filters) between top and base. Assuring | |
339 | * this is the responsibility of the interface (i.e. whoever calls | |
340 | * commit_start()). | |
341 | */ | |
ad74751f | 342 | bdrv_graph_wrlock(top); |
9a71b9de HR |
343 | s->base_overlay = bdrv_find_overlay(top, base); |
344 | assert(s->base_overlay); | |
345 | ||
346 | /* | |
347 | * The topmost node with | |
348 | * bdrv_skip_filters(filtered_base) == bdrv_skip_filters(base) | |
349 | */ | |
350 | filtered_base = bdrv_cow_bs(s->base_overlay); | |
351 | assert(bdrv_skip_filters(filtered_base) == bdrv_skip_filters(base)); | |
352 | ||
353 | /* | |
354 | * XXX BLK_PERM_WRITE needs to be allowed so we don't block ourselves | |
355 | * at s->base (if writes are blocked for a node, they are also blocked | |
356 | * for its backing file). The other options would be a second filter | |
357 | * driver above s->base. | |
358 | */ | |
359 | iter_shared_perms = BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE; | |
360 | ||
361 | for (iter = top; iter != base; iter = bdrv_filter_or_cow_bs(iter)) { | |
362 | if (iter == filtered_base) { | |
363 | /* | |
364 | * From here on, all nodes are filters on the base. This | |
365 | * allows us to share BLK_PERM_CONSISTENT_READ. | |
366 | */ | |
367 | iter_shared_perms |= BLK_PERM_CONSISTENT_READ; | |
368 | } | |
369 | ||
8dfba279 | 370 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, |
9a71b9de | 371 | iter_shared_perms, errp); |
8dfba279 | 372 | if (ret < 0) { |
6bc0bcc8 | 373 | bdrv_graph_wrunlock(top); |
8dfba279 KW |
374 | goto fail; |
375 | } | |
3e4c5122 | 376 | } |
8dfba279 | 377 | |
df827336 | 378 | if (bdrv_freeze_backing_chain(commit_top_bs, base, errp) < 0) { |
6bc0bcc8 | 379 | bdrv_graph_wrunlock(top); |
df827336 AG |
380 | goto fail; |
381 | } | |
382 | s->chain_frozen = true; | |
383 | ||
8dfba279 | 384 | ret = block_job_add_bdrv(&s->common, "base", base, 0, BLK_PERM_ALL, errp); |
6bc0bcc8 | 385 | bdrv_graph_wrunlock(top); |
f3bbc53d | 386 | |
8dfba279 KW |
387 | if (ret < 0) { |
388 | goto fail; | |
389 | } | |
390 | ||
d861ab3a | 391 | s->base = blk_new(s->common.job.aio_context, |
9a71b9de | 392 | base_perms, |
8dfba279 | 393 | BLK_PERM_CONSISTENT_READ |
8dfba279 | 394 | | BLK_PERM_WRITE_UNCHANGED); |
d7086422 KW |
395 | ret = blk_insert_bs(s->base, base, errp); |
396 | if (ret < 0) { | |
397 | goto fail; | |
398 | } | |
cf312932 | 399 | blk_set_disable_request_queuing(s->base, true); |
22dffcbe | 400 | s->base_bs = base; |
4653456a | 401 | |
8dfba279 | 402 | /* Required permissions are already taken with block_job_add_bdrv() */ |
d861ab3a | 403 | s->top = blk_new(s->common.job.aio_context, 0, BLK_PERM_ALL); |
b247767a | 404 | ret = blk_insert_bs(s->top, top, errp); |
d7086422 KW |
405 | if (ret < 0) { |
406 | goto fail; | |
407 | } | |
cf312932 | 408 | blk_set_disable_request_queuing(s->top, true); |
4653456a | 409 | |
54e26900 | 410 | s->backing_file_str = g_strdup(backing_file_str); |
747ff602 | 411 | s->on_error = on_error; |
747ff602 | 412 | |
5ccac6f1 | 413 | trace_commit_start(bs, base, top, s); |
da01ff7f | 414 | job_start(&s->common.job); |
d7086422 KW |
415 | return; |
416 | ||
417 | fail: | |
df827336 | 418 | if (s->chain_frozen) { |
9275fc72 | 419 | bdrv_graph_rdlock_main_loop(); |
df827336 | 420 | bdrv_unfreeze_backing_chain(commit_top_bs, base); |
9275fc72 | 421 | bdrv_graph_rdunlock_main_loop(); |
df827336 | 422 | } |
d7086422 KW |
423 | if (s->base) { |
424 | blk_unref(s->base); | |
425 | } | |
426 | if (s->top) { | |
427 | blk_unref(s->top); | |
428 | } | |
065abf9f AG |
429 | if (s->base_read_only) { |
430 | bdrv_reopen_set_read_only(base, true, NULL); | |
431 | } | |
2468eed3 AG |
432 | job_early_fail(&s->common.job); |
433 | /* commit_top_bs has to be replaced after deleting the block job, | |
434 | * otherwise this would fail because of lack of permissions. */ | |
8dfba279 | 435 | if (commit_top_bs) { |
ccd6a379 KW |
436 | bdrv_drained_begin(top); |
437 | bdrv_graph_wrlock(top); | |
bde70715 | 438 | bdrv_replace_node(commit_top_bs, top, &error_abort); |
6bc0bcc8 | 439 | bdrv_graph_wrunlock(top); |
ccd6a379 | 440 | bdrv_drained_end(top); |
8dfba279 | 441 | } |
747ff602 | 442 | } |
83fd6dd3 KW |
443 | |
444 | ||
d6a644bb | 445 | #define COMMIT_BUF_SIZE (2048 * BDRV_SECTOR_SIZE) |
83fd6dd3 KW |
446 | |
447 | /* commit COW file into the raw image */ | |
448 | int bdrv_commit(BlockDriverState *bs) | |
449 | { | |
f8e2bd53 | 450 | BlockBackend *src, *backing; |
d3f06759 KW |
451 | BlockDriverState *backing_file_bs = NULL; |
452 | BlockDriverState *commit_top_bs = NULL; | |
83fd6dd3 | 453 | BlockDriver *drv = bs->drv; |
d861ab3a | 454 | AioContext *ctx; |
d6a644bb | 455 | int64_t offset, length, backing_length; |
c742a364 | 456 | int ro; |
d6a644bb | 457 | int64_t n; |
83fd6dd3 | 458 | int ret = 0; |
71701708 | 459 | QEMU_AUTO_VFREE uint8_t *buf = NULL; |
d3f06759 | 460 | Error *local_err = NULL; |
83fd6dd3 | 461 | |
f791bf7f | 462 | GLOBAL_STATE_CODE(); |
277f2007 | 463 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
f791bf7f | 464 | |
83fd6dd3 KW |
465 | if (!drv) |
466 | return -ENOMEDIUM; | |
467 | ||
9a71b9de HR |
468 | backing_file_bs = bdrv_cow_bs(bs); |
469 | ||
470 | if (!backing_file_bs) { | |
83fd6dd3 KW |
471 | return -ENOTSUP; |
472 | } | |
473 | ||
474 | if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, NULL) || | |
9a71b9de HR |
475 | bdrv_op_is_blocked(backing_file_bs, BLOCK_OP_TYPE_COMMIT_TARGET, NULL)) |
476 | { | |
83fd6dd3 KW |
477 | return -EBUSY; |
478 | } | |
479 | ||
307261b2 | 480 | ro = bdrv_is_read_only(backing_file_bs); |
83fd6dd3 KW |
481 | |
482 | if (ro) { | |
9a71b9de | 483 | if (bdrv_reopen_set_read_only(backing_file_bs, false, NULL)) { |
83fd6dd3 KW |
484 | return -EACCES; |
485 | } | |
486 | } | |
487 | ||
d861ab3a | 488 | ctx = bdrv_get_aio_context(bs); |
2d97fde4 HR |
489 | /* WRITE_UNCHANGED is required for bdrv_make_empty() */ |
490 | src = blk_new(ctx, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED, | |
491 | BLK_PERM_ALL); | |
d861ab3a | 492 | backing = blk_new(ctx, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); |
d7086422 | 493 | |
d3f06759 | 494 | ret = blk_insert_bs(src, bs, &local_err); |
d7086422 | 495 | if (ret < 0) { |
d3f06759 KW |
496 | error_report_err(local_err); |
497 | goto ro_cleanup; | |
498 | } | |
499 | ||
500 | /* Insert commit_top block node above backing, so we can write to it */ | |
d3f06759 KW |
501 | commit_top_bs = bdrv_new_open_driver(&bdrv_commit_top, NULL, BDRV_O_RDWR, |
502 | &local_err); | |
503 | if (commit_top_bs == NULL) { | |
504 | error_report_err(local_err); | |
d7086422 KW |
505 | goto ro_cleanup; |
506 | } | |
507 | ||
12fa4af6 KW |
508 | bdrv_set_backing_hd(commit_top_bs, backing_file_bs, &error_abort); |
509 | bdrv_set_backing_hd(bs, commit_top_bs, &error_abort); | |
d3f06759 KW |
510 | |
511 | ret = blk_insert_bs(backing, backing_file_bs, &local_err); | |
d7086422 | 512 | if (ret < 0) { |
d3f06759 | 513 | error_report_err(local_err); |
d7086422 KW |
514 | goto ro_cleanup; |
515 | } | |
f8e2bd53 KW |
516 | |
517 | length = blk_getlength(src); | |
83fd6dd3 KW |
518 | if (length < 0) { |
519 | ret = length; | |
520 | goto ro_cleanup; | |
521 | } | |
522 | ||
f8e2bd53 | 523 | backing_length = blk_getlength(backing); |
83fd6dd3 KW |
524 | if (backing_length < 0) { |
525 | ret = backing_length; | |
526 | goto ro_cleanup; | |
527 | } | |
528 | ||
529 | /* If our top snapshot is larger than the backing file image, | |
530 | * grow the backing file image if possible. If not possible, | |
531 | * we must return an error */ | |
532 | if (length > backing_length) { | |
8c6242b6 | 533 | ret = blk_truncate(backing, length, false, PREALLOC_MODE_OFF, 0, |
c80d8b06 | 534 | &local_err); |
83fd6dd3 | 535 | if (ret < 0) { |
ed3d2ec9 | 536 | error_report_err(local_err); |
83fd6dd3 KW |
537 | goto ro_cleanup; |
538 | } | |
539 | } | |
540 | ||
f8e2bd53 KW |
541 | /* blk_try_blockalign() for src will choose an alignment that works for |
542 | * backing as well, so no need to compare the alignment manually. */ | |
d6a644bb | 543 | buf = blk_try_blockalign(src, COMMIT_BUF_SIZE); |
83fd6dd3 KW |
544 | if (buf == NULL) { |
545 | ret = -ENOMEM; | |
546 | goto ro_cleanup; | |
547 | } | |
548 | ||
d6a644bb EB |
549 | for (offset = 0; offset < length; offset += n) { |
550 | ret = bdrv_is_allocated(bs, offset, COMMIT_BUF_SIZE, &n); | |
83fd6dd3 KW |
551 | if (ret < 0) { |
552 | goto ro_cleanup; | |
553 | } | |
554 | if (ret) { | |
a9262f55 | 555 | ret = blk_pread(src, offset, n, buf, 0); |
83fd6dd3 KW |
556 | if (ret < 0) { |
557 | goto ro_cleanup; | |
558 | } | |
559 | ||
a9262f55 | 560 | ret = blk_pwrite(backing, offset, n, buf, 0); |
83fd6dd3 KW |
561 | if (ret < 0) { |
562 | goto ro_cleanup; | |
563 | } | |
564 | } | |
565 | } | |
566 | ||
2d97fde4 HR |
567 | ret = blk_make_empty(src, NULL); |
568 | /* Ignore -ENOTSUP */ | |
569 | if (ret < 0 && ret != -ENOTSUP) { | |
570 | goto ro_cleanup; | |
83fd6dd3 KW |
571 | } |
572 | ||
2d97fde4 HR |
573 | blk_flush(src); |
574 | ||
83fd6dd3 KW |
575 | /* |
576 | * Make sure all data we wrote to the backing device is actually | |
577 | * stable on disk. | |
578 | */ | |
f8e2bd53 | 579 | blk_flush(backing); |
83fd6dd3 KW |
580 | |
581 | ret = 0; | |
582 | ro_cleanup: | |
f8e2bd53 | 583 | blk_unref(backing); |
9a71b9de | 584 | if (bdrv_cow_bs(bs) != backing_file_bs) { |
12fa4af6 | 585 | bdrv_set_backing_hd(bs, backing_file_bs, &error_abort); |
d3f06759 KW |
586 | } |
587 | bdrv_unref(commit_top_bs); | |
588 | blk_unref(src); | |
f8e2bd53 | 589 | |
83fd6dd3 KW |
590 | if (ro) { |
591 | /* ignoring error return here */ | |
9a71b9de | 592 | bdrv_reopen_set_read_only(backing_file_bs, true, NULL); |
83fd6dd3 KW |
593 | } |
594 | ||
595 | return ret; | |
596 | } |