]>
Commit | Line | Data |
---|---|---|
4f1043b4 SH |
1 | /* |
2 | * Image streaming | |
3 | * | |
4 | * Copyright IBM, Corp. 2011 | |
5 | * | |
6 | * Authors: | |
7 | * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com> | |
8 | * | |
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. | |
11 | * | |
12 | */ | |
13 | ||
80c71a24 | 14 | #include "qemu/osdep.h" |
4f1043b4 | 15 | #include "trace.h" |
737e150e PB |
16 | #include "block/block_int.h" |
17 | #include "block/blockjob.h" | |
da34e65c | 18 | #include "qapi/error.h" |
cc7a8ea7 | 19 | #include "qapi/qmp/qerror.h" |
6ef228fc | 20 | #include "qemu/ratelimit.h" |
373340b2 | 21 | #include "sysemu/block-backend.h" |
4f1043b4 SH |
22 | |
23 | enum { | |
24 | /* | |
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. | |
28 | */ | |
29 | STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */ | |
30 | }; | |
31 | ||
5094a6c0 SH |
32 | #define SLICE_TIME 100000000ULL /* ns */ |
33 | ||
4f1043b4 SH |
34 | typedef struct StreamBlockJob { |
35 | BlockJob common; | |
5094a6c0 | 36 | RateLimit limit; |
4f1043b4 | 37 | BlockDriverState *base; |
1d809098 | 38 | BlockdevOnError on_error; |
13d8cc51 | 39 | char *backing_file_str; |
4f1043b4 SH |
40 | } StreamBlockJob; |
41 | ||
03e35d82 | 42 | static int coroutine_fn stream_populate(BlockBackend *blk, |
4f1043b4 SH |
43 | int64_t sector_num, int nb_sectors, |
44 | void *buf) | |
45 | { | |
46 | struct iovec iov = { | |
47 | .iov_base = buf, | |
48 | .iov_len = nb_sectors * BDRV_SECTOR_SIZE, | |
49 | }; | |
50 | QEMUIOVector qiov; | |
51 | ||
52 | qemu_iovec_init_external(&qiov, &iov, 1); | |
53 | ||
54 | /* Copy-on-read the unallocated clusters */ | |
03e35d82 KW |
55 | return blk_co_preadv(blk, sector_num * BDRV_SECTOR_SIZE, qiov.size, &qiov, |
56 | BDRV_REQ_COPY_ON_READ); | |
4f1043b4 SH |
57 | } |
58 | ||
f3e69beb SH |
59 | typedef struct { |
60 | int ret; | |
61 | bool reached_end; | |
62 | } StreamCompleteData; | |
63 | ||
64 | static void stream_complete(BlockJob *job, void *opaque) | |
65 | { | |
66 | StreamBlockJob *s = container_of(job, StreamBlockJob, common); | |
67 | StreamCompleteData *data = opaque; | |
03e35d82 | 68 | BlockDriverState *bs = blk_bs(job->blk); |
f3e69beb SH |
69 | BlockDriverState *base = s->base; |
70 | ||
71 | if (!block_job_is_cancelled(&s->common) && data->reached_end && | |
72 | data->ret == 0) { | |
73 | const char *base_id = NULL, *base_fmt = NULL; | |
74 | if (base) { | |
75 | base_id = s->backing_file_str; | |
76 | if (base->drv) { | |
77 | base_fmt = base->drv->format_name; | |
78 | } | |
79 | } | |
03e35d82 KW |
80 | data->ret = bdrv_change_backing_file(bs, base_id, base_fmt); |
81 | bdrv_set_backing_hd(bs, base); | |
f3e69beb SH |
82 | } |
83 | ||
84 | g_free(s->backing_file_str); | |
85 | block_job_completed(&s->common, data->ret); | |
86 | g_free(data); | |
87 | } | |
88 | ||
4f1043b4 SH |
89 | static void coroutine_fn stream_run(void *opaque) |
90 | { | |
91 | StreamBlockJob *s = opaque; | |
f3e69beb | 92 | StreamCompleteData *data; |
03e35d82 KW |
93 | BlockBackend *blk = s->common.blk; |
94 | BlockDriverState *bs = blk_bs(blk); | |
c8c3080f | 95 | BlockDriverState *base = s->base; |
6578629e AG |
96 | int64_t sector_num = 0; |
97 | int64_t end = -1; | |
f14a39cc | 98 | uint64_t delay_ns = 0; |
1d809098 | 99 | int error = 0; |
4f1043b4 | 100 | int ret = 0; |
04120e3b | 101 | int n = 0; |
4f1043b4 SH |
102 | void *buf; |
103 | ||
760e0063 | 104 | if (!bs->backing) { |
6578629e | 105 | goto out; |
f4a193e7 HR |
106 | } |
107 | ||
4f1043b4 SH |
108 | s->common.len = bdrv_getlength(bs); |
109 | if (s->common.len < 0) { | |
6578629e AG |
110 | ret = s->common.len; |
111 | goto out; | |
4f1043b4 SH |
112 | } |
113 | ||
114 | end = s->common.len >> BDRV_SECTOR_BITS; | |
115 | buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE); | |
116 | ||
117 | /* Turn on copy-on-read for the whole block device so that guest read | |
118 | * requests help us make progress. Only do this when copying the entire | |
119 | * backing chain since the copy-on-read operation does not take base into | |
120 | * account. | |
121 | */ | |
122 | if (!base) { | |
123 | bdrv_enable_copy_on_read(bs); | |
124 | } | |
125 | ||
126 | for (sector_num = 0; sector_num < end; sector_num += n) { | |
f9749f28 | 127 | bool copy; |
4513eafe | 128 | |
4513eafe | 129 | /* Note that even when no rate limit is applied we need to yield |
c57b6656 | 130 | * with no pending I/O here so that bdrv_drain_all() returns. |
4513eafe | 131 | */ |
7483d1e5 | 132 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
4f1043b4 SH |
133 | if (block_job_is_cancelled(&s->common)) { |
134 | break; | |
135 | } | |
136 | ||
c3e4f43a SW |
137 | copy = false; |
138 | ||
bdad13b9 PB |
139 | ret = bdrv_is_allocated(bs, sector_num, |
140 | STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n); | |
f9749f28 PB |
141 | if (ret == 1) { |
142 | /* Allocated in the top, no need to copy. */ | |
d663640c | 143 | } else if (ret >= 0) { |
f9749f28 PB |
144 | /* Copy if allocated in the intermediate images. Limit to the |
145 | * known-unallocated area [sector_num, sector_num+n). */ | |
760e0063 | 146 | ret = bdrv_is_allocated_above(backing_bs(bs), base, |
4f578637 | 147 | sector_num, n, &n); |
571cd9dc SH |
148 | |
149 | /* Finish early if end of backing file has been reached */ | |
150 | if (ret == 0 && n == 0) { | |
151 | n = end - sector_num; | |
152 | } | |
153 | ||
f9749f28 PB |
154 | copy = (ret == 1); |
155 | } | |
4f1043b4 | 156 | trace_stream_one_iteration(s, sector_num, n, ret); |
c3e4f43a | 157 | if (copy) { |
03e35d82 | 158 | ret = stream_populate(blk, sector_num, n, buf); |
4f1043b4 SH |
159 | } |
160 | if (ret < 0) { | |
1d809098 | 161 | BlockErrorAction action = |
81e254dc | 162 | block_job_error_action(&s->common, s->on_error, true, -ret); |
a589569f | 163 | if (action == BLOCK_ERROR_ACTION_STOP) { |
1d809098 PB |
164 | n = 0; |
165 | continue; | |
166 | } | |
167 | if (error == 0) { | |
168 | error = ret; | |
169 | } | |
a589569f | 170 | if (action == BLOCK_ERROR_ACTION_REPORT) { |
1d809098 PB |
171 | break; |
172 | } | |
4f1043b4 | 173 | } |
c8c3080f | 174 | ret = 0; |
4f1043b4 SH |
175 | |
176 | /* Publish progress */ | |
177 | s->common.offset += n * BDRV_SECTOR_SIZE; | |
f14a39cc SS |
178 | if (copy && s->common.speed) { |
179 | delay_ns = ratelimit_calculate_delay(&s->limit, n); | |
180 | } | |
4f1043b4 SH |
181 | } |
182 | ||
183 | if (!base) { | |
184 | bdrv_disable_copy_on_read(bs); | |
185 | } | |
186 | ||
1d809098 PB |
187 | /* Do not remove the backing file if an error was there but ignored. */ |
188 | ret = error; | |
189 | ||
4f1043b4 | 190 | qemu_vfree(buf); |
f3e69beb | 191 | |
6578629e | 192 | out: |
f3e69beb SH |
193 | /* Modify backing chain and close BDSes in main loop */ |
194 | data = g_malloc(sizeof(*data)); | |
195 | data->ret = ret; | |
196 | data->reached_end = sector_num == end; | |
197 | block_job_defer_to_main_loop(&s->common, stream_complete, data); | |
4f1043b4 SH |
198 | } |
199 | ||
882ec7ce | 200 | static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp) |
5094a6c0 SH |
201 | { |
202 | StreamBlockJob *s = container_of(job, StreamBlockJob, common); | |
203 | ||
882ec7ce | 204 | if (speed < 0) { |
c6bd8c70 | 205 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
9e6636c7 | 206 | return; |
5094a6c0 | 207 | } |
6ef228fc | 208 | ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); |
5094a6c0 SH |
209 | } |
210 | ||
3fc4b10a | 211 | static const BlockJobDriver stream_job_driver = { |
4f1043b4 | 212 | .instance_size = sizeof(StreamBlockJob), |
79e14bf7 | 213 | .job_type = BLOCK_JOB_TYPE_STREAM, |
5094a6c0 | 214 | .set_speed = stream_set_speed, |
4f1043b4 SH |
215 | }; |
216 | ||
2323322e AG |
217 | void stream_start(const char *job_id, BlockDriverState *bs, |
218 | BlockDriverState *base, const char *backing_file_str, | |
219 | int64_t speed, BlockdevOnError on_error, | |
220 | BlockCompletionFunc *cb, void *opaque, Error **errp) | |
4f1043b4 SH |
221 | { |
222 | StreamBlockJob *s; | |
4f1043b4 | 223 | |
2323322e AG |
224 | s = block_job_create(job_id, &stream_job_driver, bs, speed, |
225 | cb, opaque, errp); | |
4f1043b4 | 226 | if (!s) { |
fd7f8c65 | 227 | return; |
4f1043b4 SH |
228 | } |
229 | ||
230 | s->base = base; | |
13d8cc51 | 231 | s->backing_file_str = g_strdup(backing_file_str); |
4f1043b4 | 232 | |
1d809098 | 233 | s->on_error = on_error; |
0b8b8753 | 234 | s->common.co = qemu_coroutine_create(stream_run, s); |
fa4478d5 | 235 | trace_stream_start(bs, base, s, s->common.co, opaque); |
0b8b8753 | 236 | qemu_coroutine_enter(s->common.co); |
4f1043b4 | 237 | } |