4 * Copyright IBM, Corp. 2011
7 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
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.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 #define SLICE_TIME 100000000ULL /* ns */
34 typedef struct StreamBlockJob {
37 BlockDriverState *base;
38 BlockdevOnError on_error;
39 char *backing_file_str;
42 static int coroutine_fn stream_populate(BlockDriverState *bs,
43 int64_t sector_num, int nb_sectors,
48 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
52 qemu_iovec_init_external(&qiov, &iov, 1);
54 /* Copy-on-read the unallocated clusters */
55 return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
63 static void stream_complete(BlockJob *job, void *opaque)
65 StreamBlockJob *s = container_of(job, StreamBlockJob, common);
66 StreamCompleteData *data = opaque;
67 BlockDriverState *base = s->base;
69 if (!block_job_is_cancelled(&s->common) && data->reached_end &&
71 const char *base_id = NULL, *base_fmt = NULL;
73 base_id = s->backing_file_str;
75 base_fmt = base->drv->format_name;
78 data->ret = bdrv_change_backing_file(job->bs, base_id, base_fmt);
79 bdrv_set_backing_hd(job->bs, base);
82 g_free(s->backing_file_str);
83 block_job_completed(&s->common, data->ret);
87 static void coroutine_fn stream_run(void *opaque)
89 StreamBlockJob *s = opaque;
90 StreamCompleteData *data;
91 BlockDriverState *bs = s->common.bs;
92 BlockDriverState *base = s->base;
93 int64_t sector_num = 0;
104 s->common.len = bdrv_getlength(bs);
105 if (s->common.len < 0) {
110 end = s->common.len >> BDRV_SECTOR_BITS;
111 buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
113 /* Turn on copy-on-read for the whole block device so that guest read
114 * requests help us make progress. Only do this when copying the entire
115 * backing chain since the copy-on-read operation does not take base into
119 bdrv_enable_copy_on_read(bs);
122 for (sector_num = 0; sector_num < end; sector_num += n) {
123 uint64_t delay_ns = 0;
127 /* Note that even when no rate limit is applied we need to yield
128 * with no pending I/O here so that bdrv_drain_all() returns.
130 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
131 if (block_job_is_cancelled(&s->common)) {
137 ret = bdrv_is_allocated(bs, sector_num,
138 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
140 /* Allocated in the top, no need to copy. */
141 } else if (ret >= 0) {
142 /* Copy if allocated in the intermediate images. Limit to the
143 * known-unallocated area [sector_num, sector_num+n). */
144 ret = bdrv_is_allocated_above(backing_bs(bs), base,
147 /* Finish early if end of backing file has been reached */
148 if (ret == 0 && n == 0) {
149 n = end - sector_num;
154 trace_stream_one_iteration(s, sector_num, n, ret);
156 if (s->common.speed) {
157 delay_ns = ratelimit_calculate_delay(&s->limit, n);
162 ret = stream_populate(bs, sector_num, n, buf);
165 BlockErrorAction action =
166 block_job_error_action(&s->common, s->common.bs, s->on_error,
168 if (action == BLOCK_ERROR_ACTION_STOP) {
175 if (action == BLOCK_ERROR_ACTION_REPORT) {
181 /* Publish progress */
182 s->common.offset += n * BDRV_SECTOR_SIZE;
186 bdrv_disable_copy_on_read(bs);
189 /* Do not remove the backing file if an error was there but ignored. */
195 /* Modify backing chain and close BDSes in main loop */
196 data = g_malloc(sizeof(*data));
198 data->reached_end = sector_num == end;
199 block_job_defer_to_main_loop(&s->common, stream_complete, data);
202 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
204 StreamBlockJob *s = container_of(job, StreamBlockJob, common);
207 error_setg(errp, QERR_INVALID_PARAMETER, "speed");
210 ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
213 static const BlockJobDriver stream_job_driver = {
214 .instance_size = sizeof(StreamBlockJob),
215 .job_type = BLOCK_JOB_TYPE_STREAM,
216 .set_speed = stream_set_speed,
219 void stream_start(BlockDriverState *bs, BlockDriverState *base,
220 const char *backing_file_str, int64_t speed,
221 BlockdevOnError on_error,
222 BlockCompletionFunc *cb,
223 void *opaque, Error **errp)
227 if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
228 on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
229 (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) {
230 error_setg(errp, QERR_INVALID_PARAMETER, "on-error");
234 s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp);
240 s->backing_file_str = g_strdup(backing_file_str);
242 s->on_error = on_error;
243 s->common.co = qemu_coroutine_create(stream_run);
244 trace_stream_start(bs, base, s, s->common.co, opaque);
245 qemu_coroutine_enter(s->common.co, s);