vmstate_register_with_alias_id: Take an Error **
[qemu.git] / block / stream.c
blob1523ba7dfb8bfa92bf8a25621f38d1cc6c920d9a
1 /*
2 * Image streaming
4 * Copyright IBM, Corp. 2011
6 * Authors:
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"
15 #include "trace.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"
23 enum {
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 {
35 BlockJob common;
36 RateLimit limit;
37 BlockDriverState *base;
38 BlockdevOnError on_error;
39 char *backing_file_str;
40 int bs_flags;
41 } StreamBlockJob;
43 static int coroutine_fn stream_populate(BlockBackend *blk,
44 int64_t sector_num, int nb_sectors,
45 void *buf)
47 struct iovec iov = {
48 .iov_base = buf,
49 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
51 QEMUIOVector qiov;
53 qemu_iovec_init_external(&qiov, &iov, 1);
55 /* Copy-on-read the unallocated clusters */
56 return blk_co_preadv(blk, sector_num * BDRV_SECTOR_SIZE, qiov.size, &qiov,
57 BDRV_REQ_COPY_ON_READ);
60 typedef struct {
61 int ret;
62 bool reached_end;
63 } StreamCompleteData;
65 static void stream_complete(BlockJob *job, void *opaque)
67 StreamBlockJob *s = container_of(job, StreamBlockJob, common);
68 StreamCompleteData *data = opaque;
69 BlockDriverState *bs = blk_bs(job->blk);
70 BlockDriverState *base = s->base;
72 if (!block_job_is_cancelled(&s->common) && data->reached_end &&
73 data->ret == 0) {
74 const char *base_id = NULL, *base_fmt = NULL;
75 if (base) {
76 base_id = s->backing_file_str;
77 if (base->drv) {
78 base_fmt = base->drv->format_name;
81 data->ret = bdrv_change_backing_file(bs, base_id, base_fmt);
82 bdrv_set_backing_hd(bs, base);
85 /* Reopen the image back in read-only mode if necessary */
86 if (s->bs_flags != bdrv_get_flags(bs)) {
87 bdrv_reopen(bs, s->bs_flags, NULL);
90 g_free(s->backing_file_str);
91 block_job_completed(&s->common, data->ret);
92 g_free(data);
95 static void coroutine_fn stream_run(void *opaque)
97 StreamBlockJob *s = opaque;
98 StreamCompleteData *data;
99 BlockBackend *blk = s->common.blk;
100 BlockDriverState *bs = blk_bs(blk);
101 BlockDriverState *base = s->base;
102 int64_t sector_num = 0;
103 int64_t end = -1;
104 uint64_t delay_ns = 0;
105 int error = 0;
106 int ret = 0;
107 int n = 0;
108 void *buf;
110 if (!bs->backing) {
111 goto out;
114 s->common.len = bdrv_getlength(bs);
115 if (s->common.len < 0) {
116 ret = s->common.len;
117 goto out;
120 end = s->common.len >> BDRV_SECTOR_BITS;
121 buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
123 /* Turn on copy-on-read for the whole block device so that guest read
124 * requests help us make progress. Only do this when copying the entire
125 * backing chain since the copy-on-read operation does not take base into
126 * account.
128 if (!base) {
129 bdrv_enable_copy_on_read(bs);
132 for (sector_num = 0; sector_num < end; sector_num += n) {
133 bool copy;
135 /* Note that even when no rate limit is applied we need to yield
136 * with no pending I/O here so that bdrv_drain_all() returns.
138 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
139 if (block_job_is_cancelled(&s->common)) {
140 break;
143 copy = false;
145 ret = bdrv_is_allocated(bs, sector_num,
146 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
147 if (ret == 1) {
148 /* Allocated in the top, no need to copy. */
149 } else if (ret >= 0) {
150 /* Copy if allocated in the intermediate images. Limit to the
151 * known-unallocated area [sector_num, sector_num+n). */
152 ret = bdrv_is_allocated_above(backing_bs(bs), base,
153 sector_num, n, &n);
155 /* Finish early if end of backing file has been reached */
156 if (ret == 0 && n == 0) {
157 n = end - sector_num;
160 copy = (ret == 1);
162 trace_stream_one_iteration(s, sector_num, n, ret);
163 if (copy) {
164 ret = stream_populate(blk, sector_num, n, buf);
166 if (ret < 0) {
167 BlockErrorAction action =
168 block_job_error_action(&s->common, s->on_error, true, -ret);
169 if (action == BLOCK_ERROR_ACTION_STOP) {
170 n = 0;
171 continue;
173 if (error == 0) {
174 error = ret;
176 if (action == BLOCK_ERROR_ACTION_REPORT) {
177 break;
180 ret = 0;
182 /* Publish progress */
183 s->common.offset += n * BDRV_SECTOR_SIZE;
184 if (copy && s->common.speed) {
185 delay_ns = ratelimit_calculate_delay(&s->limit, n);
189 if (!base) {
190 bdrv_disable_copy_on_read(bs);
193 /* Do not remove the backing file if an error was there but ignored. */
194 ret = error;
196 qemu_vfree(buf);
198 out:
199 /* Modify backing chain and close BDSes in main loop */
200 data = g_malloc(sizeof(*data));
201 data->ret = ret;
202 data->reached_end = sector_num == end;
203 block_job_defer_to_main_loop(&s->common, stream_complete, data);
206 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
208 StreamBlockJob *s = container_of(job, StreamBlockJob, common);
210 if (speed < 0) {
211 error_setg(errp, QERR_INVALID_PARAMETER, "speed");
212 return;
214 ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
217 static const BlockJobDriver stream_job_driver = {
218 .instance_size = sizeof(StreamBlockJob),
219 .job_type = BLOCK_JOB_TYPE_STREAM,
220 .set_speed = stream_set_speed,
221 .start = stream_run,
224 void stream_start(const char *job_id, BlockDriverState *bs,
225 BlockDriverState *base, const char *backing_file_str,
226 int64_t speed, BlockdevOnError on_error, Error **errp)
228 StreamBlockJob *s;
229 BlockDriverState *iter;
230 int orig_bs_flags;
232 s = block_job_create(job_id, &stream_job_driver, bs, speed,
233 BLOCK_JOB_DEFAULT, NULL, NULL, errp);
234 if (!s) {
235 return;
238 /* Make sure that the image is opened in read-write mode */
239 orig_bs_flags = bdrv_get_flags(bs);
240 if (!(orig_bs_flags & BDRV_O_RDWR)) {
241 if (bdrv_reopen(bs, orig_bs_flags | BDRV_O_RDWR, errp) != 0) {
242 block_job_unref(&s->common);
243 return;
247 /* Block all intermediate nodes between bs and base, because they
248 * will disappear from the chain after this operation */
249 for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
250 block_job_add_bdrv(&s->common, iter);
253 s->base = base;
254 s->backing_file_str = g_strdup(backing_file_str);
255 s->bs_flags = orig_bs_flags;
257 s->on_error = on_error;
258 trace_stream_start(bs, base, s);
259 block_job_start(&s->common);