qemu-ga: document vsock-listen in the man page
[qemu/ar7.git] / block / commit.c
blob8e672799afb0d1f6ce88808f890ab191864704fd
1 /*
2 * Live block commit
4 * Copyright Red Hat, Inc. 2012
6 * Authors:
7 * Jeff Cody <jcody@redhat.com>
8 * Based on stream.c by Stefan Hajnoczi
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.
15 #include "qemu/osdep.h"
16 #include "qemu/cutils.h"
17 #include "trace.h"
18 #include "block/block_int.h"
19 #include "block/blockjob_int.h"
20 #include "qapi/error.h"
21 #include "qapi/qmp/qerror.h"
22 #include "qemu/ratelimit.h"
23 #include "sysemu/block-backend.h"
25 enum {
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.
31 COMMIT_BUFFER_SIZE = 512 * 1024, /* in bytes */
34 typedef struct CommitBlockJob {
35 BlockJob common;
36 BlockDriverState *commit_top_bs;
37 BlockBackend *top;
38 BlockBackend *base;
39 BlockDriverState *base_bs;
40 BlockdevOnError on_error;
41 bool base_read_only;
42 bool chain_frozen;
43 char *backing_file_str;
44 } CommitBlockJob;
46 static int commit_prepare(Job *job)
48 CommitBlockJob *s = container_of(job, CommitBlockJob, common.job);
50 bdrv_unfreeze_backing_chain(s->commit_top_bs, s->base_bs);
51 s->chain_frozen = false;
53 /* Remove base node parent that still uses BLK_PERM_WRITE/RESIZE before
54 * the normal backing chain can be restored. */
55 blk_unref(s->base);
56 s->base = NULL;
58 /* FIXME: bdrv_drop_intermediate treats total failures and partial failures
59 * identically. Further work is needed to disambiguate these cases. */
60 return bdrv_drop_intermediate(s->commit_top_bs, s->base_bs,
61 s->backing_file_str);
64 static void commit_abort(Job *job)
66 CommitBlockJob *s = container_of(job, CommitBlockJob, common.job);
67 BlockDriverState *top_bs = blk_bs(s->top);
69 if (s->chain_frozen) {
70 bdrv_unfreeze_backing_chain(s->commit_top_bs, s->base_bs);
73 /* Make sure commit_top_bs and top stay around until bdrv_replace_node() */
74 bdrv_ref(top_bs);
75 bdrv_ref(s->commit_top_bs);
77 if (s->base) {
78 blk_unref(s->base);
81 /* free the blockers on the intermediate nodes so that bdrv_replace_nodes
82 * can succeed */
83 block_job_remove_all_bdrv(&s->common);
85 /* If bdrv_drop_intermediate() failed (or was not invoked), remove the
86 * commit filter driver from the backing chain now. Do this as the final
87 * step so that the 'consistent read' permission can be granted.
89 * XXX Can (or should) we somehow keep 'consistent read' blocked even
90 * after the failed/cancelled commit job is gone? If we already wrote
91 * something to base, the intermediate images aren't valid any more. */
92 bdrv_replace_node(s->commit_top_bs, backing_bs(s->commit_top_bs),
93 &error_abort);
95 bdrv_unref(s->commit_top_bs);
96 bdrv_unref(top_bs);
99 static void commit_clean(Job *job)
101 CommitBlockJob *s = container_of(job, CommitBlockJob, common.job);
103 /* restore base open flags here if appropriate (e.g., change the base back
104 * to r/o). These reopens do not need to be atomic, since we won't abort
105 * even on failure here */
106 if (s->base_read_only) {
107 bdrv_reopen_set_read_only(s->base_bs, true, NULL);
110 g_free(s->backing_file_str);
111 blk_unref(s->top);
114 static int coroutine_fn commit_run(Job *job, Error **errp)
116 CommitBlockJob *s = container_of(job, CommitBlockJob, common.job);
117 int64_t offset;
118 uint64_t delay_ns = 0;
119 int ret = 0;
120 int64_t n = 0; /* bytes */
121 void *buf = NULL;
122 int64_t len, base_len;
124 ret = len = blk_getlength(s->top);
125 if (len < 0) {
126 goto out;
128 job_progress_set_remaining(&s->common.job, len);
130 ret = base_len = blk_getlength(s->base);
131 if (base_len < 0) {
132 goto out;
135 if (base_len < len) {
136 ret = blk_truncate(s->base, len, false, PREALLOC_MODE_OFF, NULL);
137 if (ret) {
138 goto out;
142 buf = blk_blockalign(s->top, COMMIT_BUFFER_SIZE);
144 for (offset = 0; offset < len; offset += n) {
145 bool copy;
146 bool error_in_source = true;
148 /* Note that even when no rate limit is applied we need to yield
149 * with no pending I/O here so that bdrv_drain_all() returns.
151 job_sleep_ns(&s->common.job, delay_ns);
152 if (job_is_cancelled(&s->common.job)) {
153 break;
155 /* Copy if allocated above the base */
156 ret = bdrv_is_allocated_above(blk_bs(s->top), blk_bs(s->base), false,
157 offset, COMMIT_BUFFER_SIZE, &n);
158 copy = (ret == 1);
159 trace_commit_one_iteration(s, offset, n, ret);
160 if (copy) {
161 assert(n < SIZE_MAX);
163 ret = blk_co_pread(s->top, offset, n, buf, 0);
164 if (ret >= 0) {
165 ret = blk_co_pwrite(s->base, offset, n, buf, 0);
166 if (ret < 0) {
167 error_in_source = false;
171 if (ret < 0) {
172 BlockErrorAction action =
173 block_job_error_action(&s->common, s->on_error,
174 error_in_source, -ret);
175 if (action == BLOCK_ERROR_ACTION_REPORT) {
176 goto out;
177 } else {
178 n = 0;
179 continue;
182 /* Publish progress */
183 job_progress_update(&s->common.job, n);
185 if (copy) {
186 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
187 } else {
188 delay_ns = 0;
192 ret = 0;
194 out:
195 qemu_vfree(buf);
197 return ret;
200 static const BlockJobDriver commit_job_driver = {
201 .job_driver = {
202 .instance_size = sizeof(CommitBlockJob),
203 .job_type = JOB_TYPE_COMMIT,
204 .free = block_job_free,
205 .user_resume = block_job_user_resume,
206 .run = commit_run,
207 .prepare = commit_prepare,
208 .abort = commit_abort,
209 .clean = commit_clean
213 static int coroutine_fn bdrv_commit_top_preadv(BlockDriverState *bs,
214 uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags)
216 return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags);
219 static void bdrv_commit_top_refresh_filename(BlockDriverState *bs)
221 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
222 bs->backing->bs->filename);
225 static void bdrv_commit_top_child_perm(BlockDriverState *bs, BdrvChild *c,
226 const BdrvChildRole *role,
227 BlockReopenQueue *reopen_queue,
228 uint64_t perm, uint64_t shared,
229 uint64_t *nperm, uint64_t *nshared)
231 *nperm = 0;
232 *nshared = BLK_PERM_ALL;
235 /* Dummy node that provides consistent read to its users without requiring it
236 * from its backing file and that allows writes on the backing file chain. */
237 static BlockDriver bdrv_commit_top = {
238 .format_name = "commit_top",
239 .bdrv_co_preadv = bdrv_commit_top_preadv,
240 .bdrv_co_block_status = bdrv_co_block_status_from_backing,
241 .bdrv_refresh_filename = bdrv_commit_top_refresh_filename,
242 .bdrv_child_perm = bdrv_commit_top_child_perm,
245 void commit_start(const char *job_id, BlockDriverState *bs,
246 BlockDriverState *base, BlockDriverState *top,
247 int creation_flags, int64_t speed,
248 BlockdevOnError on_error, const char *backing_file_str,
249 const char *filter_node_name, Error **errp)
251 CommitBlockJob *s;
252 BlockDriverState *iter;
253 BlockDriverState *commit_top_bs = NULL;
254 Error *local_err = NULL;
255 int ret;
257 assert(top != bs);
258 if (top == base) {
259 error_setg(errp, "Invalid files for merge: top and base are the same");
260 return;
263 s = block_job_create(job_id, &commit_job_driver, NULL, bs, 0, BLK_PERM_ALL,
264 speed, creation_flags, NULL, NULL, errp);
265 if (!s) {
266 return;
269 /* convert base to r/w, if necessary */
270 s->base_read_only = bdrv_is_read_only(base);
271 if (s->base_read_only) {
272 if (bdrv_reopen_set_read_only(base, false, errp) != 0) {
273 goto fail;
277 /* Insert commit_top block node above top, so we can block consistent read
278 * on the backing chain below it */
279 commit_top_bs = bdrv_new_open_driver(&bdrv_commit_top, filter_node_name, 0,
280 errp);
281 if (commit_top_bs == NULL) {
282 goto fail;
284 if (!filter_node_name) {
285 commit_top_bs->implicit = true;
288 /* So that we can always drop this node */
289 commit_top_bs->never_freeze = true;
291 commit_top_bs->total_sectors = top->total_sectors;
293 bdrv_append(commit_top_bs, top, &local_err);
294 if (local_err) {
295 commit_top_bs = NULL;
296 error_propagate(errp, local_err);
297 goto fail;
300 s->commit_top_bs = commit_top_bs;
302 /* Block all nodes between top and base, because they will
303 * disappear from the chain after this operation. */
304 assert(bdrv_chain_contains(top, base));
305 for (iter = top; iter != base; iter = backing_bs(iter)) {
306 /* XXX BLK_PERM_WRITE needs to be allowed so we don't block ourselves
307 * at s->base (if writes are blocked for a node, they are also blocked
308 * for its backing file). The other options would be a second filter
309 * driver above s->base. */
310 ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
311 BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE,
312 errp);
313 if (ret < 0) {
314 goto fail;
318 if (bdrv_freeze_backing_chain(commit_top_bs, base, errp) < 0) {
319 goto fail;
321 s->chain_frozen = true;
323 ret = block_job_add_bdrv(&s->common, "base", base, 0, BLK_PERM_ALL, errp);
324 if (ret < 0) {
325 goto fail;
328 s->base = blk_new(s->common.job.aio_context,
329 BLK_PERM_CONSISTENT_READ
330 | BLK_PERM_WRITE
331 | BLK_PERM_RESIZE,
332 BLK_PERM_CONSISTENT_READ
333 | BLK_PERM_GRAPH_MOD
334 | BLK_PERM_WRITE_UNCHANGED);
335 ret = blk_insert_bs(s->base, base, errp);
336 if (ret < 0) {
337 goto fail;
339 blk_set_disable_request_queuing(s->base, true);
340 s->base_bs = base;
342 /* Required permissions are already taken with block_job_add_bdrv() */
343 s->top = blk_new(s->common.job.aio_context, 0, BLK_PERM_ALL);
344 ret = blk_insert_bs(s->top, top, errp);
345 if (ret < 0) {
346 goto fail;
348 blk_set_disable_request_queuing(s->top, true);
350 s->backing_file_str = g_strdup(backing_file_str);
351 s->on_error = on_error;
353 trace_commit_start(bs, base, top, s);
354 job_start(&s->common.job);
355 return;
357 fail:
358 if (s->chain_frozen) {
359 bdrv_unfreeze_backing_chain(commit_top_bs, base);
361 if (s->base) {
362 blk_unref(s->base);
364 if (s->top) {
365 blk_unref(s->top);
367 if (s->base_read_only) {
368 bdrv_reopen_set_read_only(base, true, NULL);
370 job_early_fail(&s->common.job);
371 /* commit_top_bs has to be replaced after deleting the block job,
372 * otherwise this would fail because of lack of permissions. */
373 if (commit_top_bs) {
374 bdrv_replace_node(commit_top_bs, top, &error_abort);
379 #define COMMIT_BUF_SIZE (2048 * BDRV_SECTOR_SIZE)
381 /* commit COW file into the raw image */
382 int bdrv_commit(BlockDriverState *bs)
384 BlockBackend *src, *backing;
385 BlockDriverState *backing_file_bs = NULL;
386 BlockDriverState *commit_top_bs = NULL;
387 BlockDriver *drv = bs->drv;
388 AioContext *ctx;
389 int64_t offset, length, backing_length;
390 int ro;
391 int64_t n;
392 int ret = 0;
393 uint8_t *buf = NULL;
394 Error *local_err = NULL;
396 if (!drv)
397 return -ENOMEDIUM;
399 if (!bs->backing) {
400 return -ENOTSUP;
403 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, NULL) ||
404 bdrv_op_is_blocked(bs->backing->bs, BLOCK_OP_TYPE_COMMIT_TARGET, NULL)) {
405 return -EBUSY;
408 ro = bs->backing->bs->read_only;
410 if (ro) {
411 if (bdrv_reopen_set_read_only(bs->backing->bs, false, NULL)) {
412 return -EACCES;
416 ctx = bdrv_get_aio_context(bs);
417 src = blk_new(ctx, BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
418 backing = blk_new(ctx, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
420 ret = blk_insert_bs(src, bs, &local_err);
421 if (ret < 0) {
422 error_report_err(local_err);
423 goto ro_cleanup;
426 /* Insert commit_top block node above backing, so we can write to it */
427 backing_file_bs = backing_bs(bs);
429 commit_top_bs = bdrv_new_open_driver(&bdrv_commit_top, NULL, BDRV_O_RDWR,
430 &local_err);
431 if (commit_top_bs == NULL) {
432 error_report_err(local_err);
433 goto ro_cleanup;
436 bdrv_set_backing_hd(commit_top_bs, backing_file_bs, &error_abort);
437 bdrv_set_backing_hd(bs, commit_top_bs, &error_abort);
439 ret = blk_insert_bs(backing, backing_file_bs, &local_err);
440 if (ret < 0) {
441 error_report_err(local_err);
442 goto ro_cleanup;
445 length = blk_getlength(src);
446 if (length < 0) {
447 ret = length;
448 goto ro_cleanup;
451 backing_length = blk_getlength(backing);
452 if (backing_length < 0) {
453 ret = backing_length;
454 goto ro_cleanup;
457 /* If our top snapshot is larger than the backing file image,
458 * grow the backing file image if possible. If not possible,
459 * we must return an error */
460 if (length > backing_length) {
461 ret = blk_truncate(backing, length, false, PREALLOC_MODE_OFF,
462 &local_err);
463 if (ret < 0) {
464 error_report_err(local_err);
465 goto ro_cleanup;
469 /* blk_try_blockalign() for src will choose an alignment that works for
470 * backing as well, so no need to compare the alignment manually. */
471 buf = blk_try_blockalign(src, COMMIT_BUF_SIZE);
472 if (buf == NULL) {
473 ret = -ENOMEM;
474 goto ro_cleanup;
477 for (offset = 0; offset < length; offset += n) {
478 ret = bdrv_is_allocated(bs, offset, COMMIT_BUF_SIZE, &n);
479 if (ret < 0) {
480 goto ro_cleanup;
482 if (ret) {
483 ret = blk_pread(src, offset, buf, n);
484 if (ret < 0) {
485 goto ro_cleanup;
488 ret = blk_pwrite(backing, offset, buf, n, 0);
489 if (ret < 0) {
490 goto ro_cleanup;
495 if (drv->bdrv_make_empty) {
496 ret = drv->bdrv_make_empty(bs);
497 if (ret < 0) {
498 goto ro_cleanup;
500 blk_flush(src);
504 * Make sure all data we wrote to the backing device is actually
505 * stable on disk.
507 blk_flush(backing);
509 ret = 0;
510 ro_cleanup:
511 qemu_vfree(buf);
513 blk_unref(backing);
514 if (backing_file_bs) {
515 bdrv_set_backing_hd(bs, backing_file_bs, &error_abort);
517 bdrv_unref(commit_top_bs);
518 blk_unref(src);
520 if (ro) {
521 /* ignoring error return here */
522 bdrv_reopen_set_read_only(bs->backing->bs, true, NULL);
525 return ret;