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;
40 static int coroutine_fn stream_populate(BlockBackend *blk,
41 int64_t offset, uint64_t bytes,
50 assert(bytes < SIZE_MAX);
51 qemu_iovec_init_external(&qiov, &iov, 1);
53 /* Copy-on-read the unallocated clusters */
54 return blk_co_preadv(blk, offset, qiov.size, &qiov, BDRV_REQ_COPY_ON_READ);
57 static void stream_exit(Job *job)
59 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
60 BlockJob *bjob = &s->common;
61 BlockDriverState *bs = blk_bs(bjob->blk);
62 BlockDriverState *base = s->base;
63 Error *local_err = NULL;
66 if (!job_is_cancelled(job) && bs->backing && ret == 0) {
67 const char *base_id = NULL, *base_fmt = NULL;
69 base_id = s->backing_file_str;
71 base_fmt = base->drv->format_name;
74 ret = bdrv_change_backing_file(bs, base_id, base_fmt);
75 bdrv_set_backing_hd(bs, base, &local_err);
77 error_report_err(local_err);
84 /* Reopen the image back in read-only mode if necessary */
85 if (s->bs_flags != bdrv_get_flags(bs)) {
86 /* Give up write permissions before making it read-only */
87 blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
88 bdrv_reopen(bs, s->bs_flags, NULL);
91 g_free(s->backing_file_str);
95 static int coroutine_fn stream_run(Job *job, Error **errp)
97 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
98 BlockBackend *blk = s->common.blk;
99 BlockDriverState *bs = blk_bs(blk);
100 BlockDriverState *base = s->base;
103 uint64_t delay_ns = 0;
106 int64_t n = 0; /* bytes */
113 len = bdrv_getlength(bs);
118 job_progress_set_remaining(&s->common.job, len);
120 buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
122 /* Turn on copy-on-read for the whole block device so that guest read
123 * requests help us make progress. Only do this when copying the entire
124 * backing chain since the copy-on-read operation does not take base into
128 bdrv_enable_copy_on_read(bs);
131 for ( ; offset < len; offset += n) {
134 /* Note that even when no rate limit is applied we need to yield
135 * with no pending I/O here so that bdrv_drain_all() returns.
137 job_sleep_ns(&s->common.job, delay_ns);
138 if (job_is_cancelled(&s->common.job)) {
144 ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &n);
146 /* Allocated in the top, no need to copy. */
147 } else if (ret >= 0) {
148 /* Copy if allocated in the intermediate images. Limit to the
149 * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */
150 ret = bdrv_is_allocated_above(backing_bs(bs), base,
153 /* Finish early if end of backing file has been reached */
154 if (ret == 0 && n == 0) {
160 trace_stream_one_iteration(s, offset, n, ret);
162 ret = stream_populate(blk, offset, n, buf);
165 BlockErrorAction action =
166 block_job_error_action(&s->common, s->on_error, true, -ret);
167 if (action == BLOCK_ERROR_ACTION_STOP) {
174 if (action == BLOCK_ERROR_ACTION_REPORT) {
180 /* Publish progress */
181 job_progress_update(&s->common.job, n);
183 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
190 bdrv_disable_copy_on_read(bs);
193 /* Do not remove the backing file if an error was there but ignored. */
199 /* Modify backing chain and close BDSes in main loop */
203 static const BlockJobDriver stream_job_driver = {
205 .instance_size = sizeof(StreamBlockJob),
206 .job_type = JOB_TYPE_STREAM,
207 .free = block_job_free,
210 .user_resume = block_job_user_resume,
211 .drain = block_job_drain,
215 void stream_start(const char *job_id, BlockDriverState *bs,
216 BlockDriverState *base, const char *backing_file_str,
217 int64_t speed, BlockdevOnError on_error, Error **errp)
220 BlockDriverState *iter;
223 /* Make sure that the image is opened in read-write mode */
224 orig_bs_flags = bdrv_get_flags(bs);
225 if (!(orig_bs_flags & BDRV_O_RDWR)) {
226 if (bdrv_reopen(bs, orig_bs_flags | BDRV_O_RDWR, errp) != 0) {
231 /* Prevent concurrent jobs trying to modify the graph structure here, we
232 * already have our own plans. Also don't allow resize as the image size is
233 * queried only at the job start and then cached. */
234 s = block_job_create(job_id, &stream_job_driver, NULL, bs,
235 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
237 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
239 speed, JOB_DEFAULT, NULL, NULL, errp);
244 /* Block all intermediate nodes between bs and base, because they will
245 * disappear from the chain after this operation. The streaming job reads
246 * every block only once, assuming that it doesn't change, so block writes
248 for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
249 block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
250 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED,
255 s->backing_file_str = g_strdup(backing_file_str);
256 s->bs_flags = orig_bs_flags;
258 s->on_error = on_error;
259 trace_stream_start(bs, base, s);
260 job_start(&s->common.job);
264 if (orig_bs_flags != bdrv_get_flags(bs)) {
265 bdrv_reopen(bs, orig_bs_flags, NULL);