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.
15 #include "block/block_int.h"
16 #include "block/blockjob.h"
17 #include "qapi/qmp/qerror.h"
18 #include "qemu/ratelimit.h"
22 * Size of data buffer for populating the image file. This should be large
23 * enough to process multiple clusters in a single call, so that populating
24 * contiguous regions of the image is efficient.
26 STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
29 #define SLICE_TIME 100000000ULL /* ns */
31 typedef struct StreamBlockJob {
34 BlockDriverState *base;
35 BlockdevOnError on_error;
36 char *backing_file_str;
39 static int coroutine_fn stream_populate(BlockDriverState *bs,
40 int64_t sector_num, int nb_sectors,
45 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
49 qemu_iovec_init_external(&qiov, &iov, 1);
51 /* Copy-on-read the unallocated clusters */
52 return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
55 static void close_unused_images(BlockDriverState *top, BlockDriverState *base,
58 BlockDriverState *intermediate;
59 intermediate = top->backing_hd;
61 /* Must assign before bdrv_delete() to prevent traversing dangling pointer
62 * while we delete backing image instances.
64 bdrv_set_backing_hd(top, base);
66 while (intermediate) {
67 BlockDriverState *unused;
70 if (intermediate == base) {
74 unused = intermediate;
75 intermediate = intermediate->backing_hd;
76 bdrv_set_backing_hd(unused, NULL);
80 bdrv_refresh_limits(top, NULL);
88 static void stream_complete(BlockJob *job, void *opaque)
90 StreamBlockJob *s = container_of(job, StreamBlockJob, common);
91 StreamCompleteData *data = opaque;
92 BlockDriverState *base = s->base;
94 if (!block_job_is_cancelled(&s->common) && data->reached_end &&
96 const char *base_id = NULL, *base_fmt = NULL;
98 base_id = s->backing_file_str;
100 base_fmt = base->drv->format_name;
103 data->ret = bdrv_change_backing_file(job->bs, base_id, base_fmt);
104 close_unused_images(job->bs, base, base_id);
107 g_free(s->backing_file_str);
108 block_job_completed(&s->common, data->ret);
112 static void coroutine_fn stream_run(void *opaque)
114 StreamBlockJob *s = opaque;
115 StreamCompleteData *data;
116 BlockDriverState *bs = s->common.bs;
117 BlockDriverState *base = s->base;
118 int64_t sector_num, end;
124 if (!bs->backing_hd) {
125 block_job_completed(&s->common, 0);
129 s->common.len = bdrv_getlength(bs);
130 if (s->common.len < 0) {
131 block_job_completed(&s->common, s->common.len);
135 end = s->common.len >> BDRV_SECTOR_BITS;
136 buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
138 /* Turn on copy-on-read for the whole block device so that guest read
139 * requests help us make progress. Only do this when copying the entire
140 * backing chain since the copy-on-read operation does not take base into
144 bdrv_enable_copy_on_read(bs);
147 for (sector_num = 0; sector_num < end; sector_num += n) {
148 uint64_t delay_ns = 0;
152 /* Note that even when no rate limit is applied we need to yield
153 * with no pending I/O here so that bdrv_drain_all() returns.
155 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
156 if (block_job_is_cancelled(&s->common)) {
162 ret = bdrv_is_allocated(bs, sector_num,
163 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
165 /* Allocated in the top, no need to copy. */
166 } else if (ret >= 0) {
167 /* Copy if allocated in the intermediate images. Limit to the
168 * known-unallocated area [sector_num, sector_num+n). */
169 ret = bdrv_is_allocated_above(bs->backing_hd, base,
172 /* Finish early if end of backing file has been reached */
173 if (ret == 0 && n == 0) {
174 n = end - sector_num;
179 trace_stream_one_iteration(s, sector_num, n, ret);
181 if (s->common.speed) {
182 delay_ns = ratelimit_calculate_delay(&s->limit, n);
187 ret = stream_populate(bs, sector_num, n, buf);
190 BlockErrorAction action =
191 block_job_error_action(&s->common, s->common.bs, s->on_error,
193 if (action == BLOCK_ERROR_ACTION_STOP) {
200 if (action == BLOCK_ERROR_ACTION_REPORT) {
206 /* Publish progress */
207 s->common.offset += n * BDRV_SECTOR_SIZE;
211 bdrv_disable_copy_on_read(bs);
214 /* Do not remove the backing file if an error was there but ignored. */
219 /* Modify backing chain and close BDSes in main loop */
220 data = g_malloc(sizeof(*data));
222 data->reached_end = sector_num == end;
223 block_job_defer_to_main_loop(&s->common, stream_complete, data);
226 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
228 StreamBlockJob *s = container_of(job, StreamBlockJob, common);
231 error_setg(errp, QERR_INVALID_PARAMETER, "speed");
234 ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
237 static const BlockJobDriver stream_job_driver = {
238 .instance_size = sizeof(StreamBlockJob),
239 .job_type = BLOCK_JOB_TYPE_STREAM,
240 .set_speed = stream_set_speed,
243 void stream_start(BlockDriverState *bs, BlockDriverState *base,
244 const char *backing_file_str, int64_t speed,
245 BlockdevOnError on_error,
246 BlockCompletionFunc *cb,
247 void *opaque, Error **errp)
251 if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
252 on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
253 !bdrv_iostatus_is_enabled(bs)) {
254 error_setg(errp, QERR_INVALID_PARAMETER, "on-error");
258 s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp);
264 s->backing_file_str = g_strdup(backing_file_str);
266 s->on_error = on_error;
267 s->common.co = qemu_coroutine_create(stream_run);
268 trace_stream_start(bs, base, s, s->common.co, opaque);
269 qemu_coroutine_enter(s->common.co, s);