4 * Copyright IBM, Corp. 2011
9 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
10 * See the COPYING.LIB file in the top-level directory.
14 #include "qemu/osdep.h"
16 #include "block/block_int.h"
17 #include "block/blockjob_int.h"
18 #include "qapi/error.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qemu/ratelimit.h"
21 #include "sysemu/block-backend.h"
25 * Size of data buffer for populating the image file. This should be large
26 * enough to process multiple clusters in a single call, so that populating
27 * contiguous regions of the image is efficient.
29 STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
32 typedef struct StreamBlockJob {
34 BlockDriverState *base;
35 BlockdevOnError on_error;
36 char *backing_file_str;
41 static int coroutine_fn stream_populate(BlockBackend *blk,
42 int64_t offset, uint64_t bytes,
45 QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
47 assert(bytes < SIZE_MAX);
49 /* Copy-on-read the unallocated clusters */
50 return blk_co_preadv(blk, offset, qiov.size, &qiov, BDRV_REQ_COPY_ON_READ);
53 static void stream_abort(Job *job)
55 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
57 if (s->chain_frozen) {
58 BlockJob *bjob = &s->common;
59 bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->base);
63 static int stream_prepare(Job *job)
65 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
66 BlockJob *bjob = &s->common;
67 BlockDriverState *bs = blk_bs(bjob->blk);
68 BlockDriverState *base = s->base;
69 Error *local_err = NULL;
72 bdrv_unfreeze_backing_chain(bs, base);
73 s->chain_frozen = false;
76 const char *base_id = NULL, *base_fmt = NULL;
78 base_id = s->backing_file_str;
80 base_fmt = base->drv->format_name;
83 ret = bdrv_change_backing_file(bs, base_id, base_fmt);
84 bdrv_set_backing_hd(bs, base, &local_err);
86 error_report_err(local_err);
94 static void stream_clean(Job *job)
96 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
97 BlockJob *bjob = &s->common;
98 BlockDriverState *bs = blk_bs(bjob->blk);
100 /* Reopen the image back in read-only mode if necessary */
101 if (s->bs_read_only) {
102 /* Give up write permissions before making it read-only */
103 blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
104 bdrv_reopen_set_read_only(bs, true, NULL);
107 g_free(s->backing_file_str);
110 static int coroutine_fn stream_run(Job *job, Error **errp)
112 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
113 BlockBackend *blk = s->common.blk;
114 BlockDriverState *bs = blk_bs(blk);
115 BlockDriverState *base = s->base;
118 uint64_t delay_ns = 0;
121 int64_t n = 0; /* bytes */
128 len = bdrv_getlength(bs);
133 job_progress_set_remaining(&s->common.job, len);
135 buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
137 /* Turn on copy-on-read for the whole block device so that guest read
138 * requests help us make progress. Only do this when copying the entire
139 * backing chain since the copy-on-read operation does not take base into
143 bdrv_enable_copy_on_read(bs);
146 for ( ; offset < len; offset += n) {
149 /* Note that even when no rate limit is applied we need to yield
150 * with no pending I/O here so that bdrv_drain_all() returns.
152 job_sleep_ns(&s->common.job, delay_ns);
153 if (job_is_cancelled(&s->common.job)) {
159 ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &n);
161 /* Allocated in the top, no need to copy. */
162 } else if (ret >= 0) {
163 /* Copy if allocated in the intermediate images. Limit to the
164 * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */
165 ret = bdrv_is_allocated_above(backing_bs(bs), base,
168 /* Finish early if end of backing file has been reached */
169 if (ret == 0 && n == 0) {
175 trace_stream_one_iteration(s, offset, n, ret);
177 ret = stream_populate(blk, offset, n, buf);
180 BlockErrorAction action =
181 block_job_error_action(&s->common, s->on_error, true, -ret);
182 if (action == BLOCK_ERROR_ACTION_STOP) {
189 if (action == BLOCK_ERROR_ACTION_REPORT) {
195 /* Publish progress */
196 job_progress_update(&s->common.job, n);
198 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
205 bdrv_disable_copy_on_read(bs);
208 /* Do not remove the backing file if an error was there but ignored. */
214 /* Modify backing chain and close BDSes in main loop */
218 static const BlockJobDriver stream_job_driver = {
220 .instance_size = sizeof(StreamBlockJob),
221 .job_type = JOB_TYPE_STREAM,
222 .free = block_job_free,
224 .prepare = stream_prepare,
225 .abort = stream_abort,
226 .clean = stream_clean,
227 .user_resume = block_job_user_resume,
228 .drain = block_job_drain,
232 void stream_start(const char *job_id, BlockDriverState *bs,
233 BlockDriverState *base, const char *backing_file_str,
234 int creation_flags, int64_t speed,
235 BlockdevOnError on_error, Error **errp)
238 BlockDriverState *iter;
241 /* Make sure that the image is opened in read-write mode */
242 bs_read_only = bdrv_is_read_only(bs);
244 if (bdrv_reopen_set_read_only(bs, false, errp) != 0) {
249 /* Prevent concurrent jobs trying to modify the graph structure here, we
250 * already have our own plans. Also don't allow resize as the image size is
251 * queried only at the job start and then cached. */
252 s = block_job_create(job_id, &stream_job_driver, NULL, bs,
253 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
255 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
257 speed, creation_flags, NULL, NULL, errp);
262 /* Block all intermediate nodes between bs and base, because they will
263 * disappear from the chain after this operation. The streaming job reads
264 * every block only once, assuming that it doesn't change, so block writes
266 for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
267 block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
268 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED,
272 if (bdrv_freeze_backing_chain(bs, base, errp) < 0) {
273 job_early_fail(&s->common.job);
278 s->backing_file_str = g_strdup(backing_file_str);
279 s->bs_read_only = bs_read_only;
280 s->chain_frozen = true;
282 s->on_error = on_error;
283 trace_stream_start(bs, base, s);
284 job_start(&s->common.job);
289 bdrv_reopen_set_read_only(bs, true, NULL);