travis: add whitespace between each major section & matrix entry
[qemu/ar7.git] / block / commit.c
blob53148e610bceeca0bc52c1edad961c6fc0da3177
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 char *backing_file_str;
43 } CommitBlockJob;
45 static int coroutine_fn commit_populate(BlockBackend *bs, BlockBackend *base,
46 int64_t offset, uint64_t bytes,
47 void *buf)
49 int ret = 0;
50 QEMUIOVector qiov;
51 struct iovec iov = {
52 .iov_base = buf,
53 .iov_len = bytes,
56 assert(bytes < SIZE_MAX);
57 qemu_iovec_init_external(&qiov, &iov, 1);
59 ret = blk_co_preadv(bs, offset, qiov.size, &qiov, 0);
60 if (ret < 0) {
61 return ret;
64 ret = blk_co_pwritev(base, offset, qiov.size, &qiov, 0);
65 if (ret < 0) {
66 return ret;
69 return 0;
72 static int commit_prepare(Job *job)
74 CommitBlockJob *s = container_of(job, CommitBlockJob, common.job);
76 /* Remove base node parent that still uses BLK_PERM_WRITE/RESIZE before
77 * the normal backing chain can be restored. */
78 blk_unref(s->base);
79 s->base = NULL;
81 /* FIXME: bdrv_drop_intermediate treats total failures and partial failures
82 * identically. Further work is needed to disambiguate these cases. */
83 return bdrv_drop_intermediate(s->commit_top_bs, s->base_bs,
84 s->backing_file_str);
87 static void commit_abort(Job *job)
89 CommitBlockJob *s = container_of(job, CommitBlockJob, common.job);
90 BlockDriverState *top_bs = blk_bs(s->top);
92 /* Make sure commit_top_bs and top stay around until bdrv_replace_node() */
93 bdrv_ref(top_bs);
94 bdrv_ref(s->commit_top_bs);
96 if (s->base) {
97 blk_unref(s->base);
100 /* free the blockers on the intermediate nodes so that bdrv_replace_nodes
101 * can succeed */
102 block_job_remove_all_bdrv(&s->common);
104 /* If bdrv_drop_intermediate() failed (or was not invoked), remove the
105 * commit filter driver from the backing chain now. Do this as the final
106 * step so that the 'consistent read' permission can be granted.
108 * XXX Can (or should) we somehow keep 'consistent read' blocked even
109 * after the failed/cancelled commit job is gone? If we already wrote
110 * something to base, the intermediate images aren't valid any more. */
111 bdrv_child_try_set_perm(s->commit_top_bs->backing, 0, BLK_PERM_ALL,
112 &error_abort);
113 bdrv_replace_node(s->commit_top_bs, backing_bs(s->commit_top_bs),
114 &error_abort);
116 bdrv_unref(s->commit_top_bs);
117 bdrv_unref(top_bs);
120 static void commit_clean(Job *job)
122 CommitBlockJob *s = container_of(job, CommitBlockJob, common.job);
124 /* restore base open flags here if appropriate (e.g., change the base back
125 * to r/o). These reopens do not need to be atomic, since we won't abort
126 * even on failure here */
127 if (s->base_read_only) {
128 bdrv_reopen_set_read_only(s->base_bs, true, NULL);
131 g_free(s->backing_file_str);
132 blk_unref(s->top);
135 static int coroutine_fn commit_run(Job *job, Error **errp)
137 CommitBlockJob *s = container_of(job, CommitBlockJob, common.job);
138 int64_t offset;
139 uint64_t delay_ns = 0;
140 int ret = 0;
141 int64_t n = 0; /* bytes */
142 void *buf = NULL;
143 int bytes_written = 0;
144 int64_t len, base_len;
146 ret = len = blk_getlength(s->top);
147 if (len < 0) {
148 goto out;
150 job_progress_set_remaining(&s->common.job, len);
152 ret = base_len = blk_getlength(s->base);
153 if (base_len < 0) {
154 goto out;
157 if (base_len < len) {
158 ret = blk_truncate(s->base, len, PREALLOC_MODE_OFF, NULL);
159 if (ret) {
160 goto out;
164 buf = blk_blockalign(s->top, COMMIT_BUFFER_SIZE);
166 for (offset = 0; offset < len; offset += n) {
167 bool copy;
169 /* Note that even when no rate limit is applied we need to yield
170 * with no pending I/O here so that bdrv_drain_all() returns.
172 job_sleep_ns(&s->common.job, delay_ns);
173 if (job_is_cancelled(&s->common.job)) {
174 break;
176 /* Copy if allocated above the base */
177 ret = bdrv_is_allocated_above(blk_bs(s->top), blk_bs(s->base),
178 offset, COMMIT_BUFFER_SIZE, &n);
179 copy = (ret == 1);
180 trace_commit_one_iteration(s, offset, n, ret);
181 if (copy) {
182 ret = commit_populate(s->top, s->base, offset, n, buf);
183 bytes_written += n;
185 if (ret < 0) {
186 BlockErrorAction action =
187 block_job_error_action(&s->common, false, s->on_error, -ret);
188 if (action == BLOCK_ERROR_ACTION_REPORT) {
189 goto out;
190 } else {
191 n = 0;
192 continue;
195 /* Publish progress */
196 job_progress_update(&s->common.job, n);
198 if (copy) {
199 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
200 } else {
201 delay_ns = 0;
205 ret = 0;
207 out:
208 qemu_vfree(buf);
210 return ret;
213 static const BlockJobDriver commit_job_driver = {
214 .job_driver = {
215 .instance_size = sizeof(CommitBlockJob),
216 .job_type = JOB_TYPE_COMMIT,
217 .free = block_job_free,
218 .user_resume = block_job_user_resume,
219 .drain = block_job_drain,
220 .run = commit_run,
221 .prepare = commit_prepare,
222 .abort = commit_abort,
223 .clean = commit_clean
227 static int coroutine_fn bdrv_commit_top_preadv(BlockDriverState *bs,
228 uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags)
230 return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags);
233 static void bdrv_commit_top_refresh_filename(BlockDriverState *bs, QDict *opts)
235 bdrv_refresh_filename(bs->backing->bs);
236 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
237 bs->backing->bs->filename);
240 static void bdrv_commit_top_child_perm(BlockDriverState *bs, BdrvChild *c,
241 const BdrvChildRole *role,
242 BlockReopenQueue *reopen_queue,
243 uint64_t perm, uint64_t shared,
244 uint64_t *nperm, uint64_t *nshared)
246 *nperm = 0;
247 *nshared = BLK_PERM_ALL;
250 /* Dummy node that provides consistent read to its users without requiring it
251 * from its backing file and that allows writes on the backing file chain. */
252 static BlockDriver bdrv_commit_top = {
253 .format_name = "commit_top",
254 .bdrv_co_preadv = bdrv_commit_top_preadv,
255 .bdrv_co_block_status = bdrv_co_block_status_from_backing,
256 .bdrv_refresh_filename = bdrv_commit_top_refresh_filename,
257 .bdrv_child_perm = bdrv_commit_top_child_perm,
260 void commit_start(const char *job_id, BlockDriverState *bs,
261 BlockDriverState *base, BlockDriverState *top,
262 int creation_flags, int64_t speed,
263 BlockdevOnError on_error, const char *backing_file_str,
264 const char *filter_node_name, Error **errp)
266 CommitBlockJob *s;
267 BlockDriverState *iter;
268 BlockDriverState *commit_top_bs = NULL;
269 Error *local_err = NULL;
270 int ret;
272 assert(top != bs);
273 if (top == base) {
274 error_setg(errp, "Invalid files for merge: top and base are the same");
275 return;
278 s = block_job_create(job_id, &commit_job_driver, NULL, bs, 0, BLK_PERM_ALL,
279 speed, creation_flags, NULL, NULL, errp);
280 if (!s) {
281 return;
284 /* convert base to r/w, if necessary */
285 s->base_read_only = bdrv_is_read_only(base);
286 if (s->base_read_only) {
287 if (bdrv_reopen_set_read_only(base, false, errp) != 0) {
288 goto fail;
292 /* Insert commit_top block node above top, so we can block consistent read
293 * on the backing chain below it */
294 commit_top_bs = bdrv_new_open_driver(&bdrv_commit_top, filter_node_name, 0,
295 errp);
296 if (commit_top_bs == NULL) {
297 goto fail;
299 if (!filter_node_name) {
300 commit_top_bs->implicit = true;
302 commit_top_bs->total_sectors = top->total_sectors;
303 bdrv_set_aio_context(commit_top_bs, bdrv_get_aio_context(top));
305 bdrv_set_backing_hd(commit_top_bs, top, &local_err);
306 if (local_err) {
307 bdrv_unref(commit_top_bs);
308 commit_top_bs = NULL;
309 error_propagate(errp, local_err);
310 goto fail;
312 bdrv_replace_node(top, commit_top_bs, &local_err);
313 if (local_err) {
314 bdrv_unref(commit_top_bs);
315 commit_top_bs = NULL;
316 error_propagate(errp, local_err);
317 goto fail;
320 s->commit_top_bs = commit_top_bs;
321 bdrv_unref(commit_top_bs);
323 /* Block all nodes between top and base, because they will
324 * disappear from the chain after this operation. */
325 assert(bdrv_chain_contains(top, base));
326 for (iter = top; iter != base; iter = backing_bs(iter)) {
327 /* XXX BLK_PERM_WRITE needs to be allowed so we don't block ourselves
328 * at s->base (if writes are blocked for a node, they are also blocked
329 * for its backing file). The other options would be a second filter
330 * driver above s->base. */
331 ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
332 BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE,
333 errp);
334 if (ret < 0) {
335 goto fail;
339 ret = block_job_add_bdrv(&s->common, "base", base, 0, BLK_PERM_ALL, errp);
340 if (ret < 0) {
341 goto fail;
344 s->base = blk_new(BLK_PERM_CONSISTENT_READ
345 | BLK_PERM_WRITE
346 | BLK_PERM_RESIZE,
347 BLK_PERM_CONSISTENT_READ
348 | BLK_PERM_GRAPH_MOD
349 | BLK_PERM_WRITE_UNCHANGED);
350 ret = blk_insert_bs(s->base, base, errp);
351 if (ret < 0) {
352 goto fail;
354 s->base_bs = base;
356 /* Required permissions are already taken with block_job_add_bdrv() */
357 s->top = blk_new(0, BLK_PERM_ALL);
358 ret = blk_insert_bs(s->top, top, errp);
359 if (ret < 0) {
360 goto fail;
363 s->backing_file_str = g_strdup(backing_file_str);
364 s->on_error = on_error;
366 trace_commit_start(bs, base, top, s);
367 job_start(&s->common.job);
368 return;
370 fail:
371 if (s->base) {
372 blk_unref(s->base);
374 if (s->top) {
375 blk_unref(s->top);
377 if (commit_top_bs) {
378 bdrv_replace_node(commit_top_bs, top, &error_abort);
380 job_early_fail(&s->common.job);
384 #define COMMIT_BUF_SIZE (2048 * BDRV_SECTOR_SIZE)
386 /* commit COW file into the raw image */
387 int bdrv_commit(BlockDriverState *bs)
389 BlockBackend *src, *backing;
390 BlockDriverState *backing_file_bs = NULL;
391 BlockDriverState *commit_top_bs = NULL;
392 BlockDriver *drv = bs->drv;
393 int64_t offset, length, backing_length;
394 int ro;
395 int64_t n;
396 int ret = 0;
397 uint8_t *buf = NULL;
398 Error *local_err = NULL;
400 if (!drv)
401 return -ENOMEDIUM;
403 if (!bs->backing) {
404 return -ENOTSUP;
407 if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT_SOURCE, NULL) ||
408 bdrv_op_is_blocked(bs->backing->bs, BLOCK_OP_TYPE_COMMIT_TARGET, NULL)) {
409 return -EBUSY;
412 ro = bs->backing->bs->read_only;
414 if (ro) {
415 if (bdrv_reopen_set_read_only(bs->backing->bs, false, NULL)) {
416 return -EACCES;
420 src = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
421 backing = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
423 ret = blk_insert_bs(src, bs, &local_err);
424 if (ret < 0) {
425 error_report_err(local_err);
426 goto ro_cleanup;
429 /* Insert commit_top block node above backing, so we can write to it */
430 backing_file_bs = backing_bs(bs);
432 commit_top_bs = bdrv_new_open_driver(&bdrv_commit_top, NULL, BDRV_O_RDWR,
433 &local_err);
434 if (commit_top_bs == NULL) {
435 error_report_err(local_err);
436 goto ro_cleanup;
438 bdrv_set_aio_context(commit_top_bs, bdrv_get_aio_context(backing_file_bs));
440 bdrv_set_backing_hd(commit_top_bs, backing_file_bs, &error_abort);
441 bdrv_set_backing_hd(bs, commit_top_bs, &error_abort);
443 ret = blk_insert_bs(backing, backing_file_bs, &local_err);
444 if (ret < 0) {
445 error_report_err(local_err);
446 goto ro_cleanup;
449 length = blk_getlength(src);
450 if (length < 0) {
451 ret = length;
452 goto ro_cleanup;
455 backing_length = blk_getlength(backing);
456 if (backing_length < 0) {
457 ret = backing_length;
458 goto ro_cleanup;
461 /* If our top snapshot is larger than the backing file image,
462 * grow the backing file image if possible. If not possible,
463 * we must return an error */
464 if (length > backing_length) {
465 ret = blk_truncate(backing, length, PREALLOC_MODE_OFF, &local_err);
466 if (ret < 0) {
467 error_report_err(local_err);
468 goto ro_cleanup;
472 /* blk_try_blockalign() for src will choose an alignment that works for
473 * backing as well, so no need to compare the alignment manually. */
474 buf = blk_try_blockalign(src, COMMIT_BUF_SIZE);
475 if (buf == NULL) {
476 ret = -ENOMEM;
477 goto ro_cleanup;
480 for (offset = 0; offset < length; offset += n) {
481 ret = bdrv_is_allocated(bs, offset, COMMIT_BUF_SIZE, &n);
482 if (ret < 0) {
483 goto ro_cleanup;
485 if (ret) {
486 ret = blk_pread(src, offset, buf, n);
487 if (ret < 0) {
488 goto ro_cleanup;
491 ret = blk_pwrite(backing, offset, buf, n, 0);
492 if (ret < 0) {
493 goto ro_cleanup;
498 if (drv->bdrv_make_empty) {
499 ret = drv->bdrv_make_empty(bs);
500 if (ret < 0) {
501 goto ro_cleanup;
503 blk_flush(src);
507 * Make sure all data we wrote to the backing device is actually
508 * stable on disk.
510 blk_flush(backing);
512 ret = 0;
513 ro_cleanup:
514 qemu_vfree(buf);
516 blk_unref(backing);
517 if (backing_file_bs) {
518 bdrv_set_backing_hd(bs, backing_file_bs, &error_abort);
520 bdrv_unref(commit_top_bs);
521 blk_unref(src);
523 if (ro) {
524 /* ignoring error return here */
525 bdrv_reopen_set_read_only(bs->backing->bs, true, NULL);
528 return ret;