]>
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 | ||
14 | #include "trace.h" | |
737e150e PB |
15 | #include "block/block_int.h" |
16 | #include "block/blockjob.h" | |
6ef228fc | 17 | #include "qemu/ratelimit.h" |
4f1043b4 SH |
18 | |
19 | enum { | |
20 | /* | |
21 | * Size of data buffer for populating the image file. This should be large | |
22 | * enough to process multiple clusters in a single call, so that populating | |
23 | * contiguous regions of the image is efficient. | |
24 | */ | |
25 | STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */ | |
26 | }; | |
27 | ||
5094a6c0 SH |
28 | #define SLICE_TIME 100000000ULL /* ns */ |
29 | ||
4f1043b4 SH |
30 | typedef struct StreamBlockJob { |
31 | BlockJob common; | |
5094a6c0 | 32 | RateLimit limit; |
4f1043b4 | 33 | BlockDriverState *base; |
1d809098 | 34 | BlockdevOnError on_error; |
c8c3080f | 35 | char backing_file_id[1024]; |
4f1043b4 SH |
36 | } StreamBlockJob; |
37 | ||
38 | static int coroutine_fn stream_populate(BlockDriverState *bs, | |
39 | int64_t sector_num, int nb_sectors, | |
40 | void *buf) | |
41 | { | |
42 | struct iovec iov = { | |
43 | .iov_base = buf, | |
44 | .iov_len = nb_sectors * BDRV_SECTOR_SIZE, | |
45 | }; | |
46 | QEMUIOVector qiov; | |
47 | ||
48 | qemu_iovec_init_external(&qiov, &iov, 1); | |
49 | ||
50 | /* Copy-on-read the unallocated clusters */ | |
51 | return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov); | |
52 | } | |
53 | ||
5a67a104 MT |
54 | static void close_unused_images(BlockDriverState *top, BlockDriverState *base, |
55 | const char *base_id) | |
56 | { | |
57 | BlockDriverState *intermediate; | |
58 | intermediate = top->backing_hd; | |
59 | ||
88266f5a SH |
60 | /* Must assign before bdrv_delete() to prevent traversing dangling pointer |
61 | * while we delete backing image instances. | |
62 | */ | |
63 | top->backing_hd = base; | |
64 | ||
5a67a104 MT |
65 | while (intermediate) { |
66 | BlockDriverState *unused; | |
67 | ||
68 | /* reached base */ | |
69 | if (intermediate == base) { | |
70 | break; | |
71 | } | |
72 | ||
73 | unused = intermediate; | |
74 | intermediate = intermediate->backing_hd; | |
75 | unused->backing_hd = NULL; | |
4f6fd349 | 76 | bdrv_unref(unused); |
5a67a104 | 77 | } |
5a67a104 MT |
78 | } |
79 | ||
4f1043b4 SH |
80 | static void coroutine_fn stream_run(void *opaque) |
81 | { | |
82 | StreamBlockJob *s = opaque; | |
83 | BlockDriverState *bs = s->common.bs; | |
c8c3080f | 84 | BlockDriverState *base = s->base; |
4f1043b4 | 85 | int64_t sector_num, end; |
1d809098 | 86 | int error = 0; |
4f1043b4 | 87 | int ret = 0; |
04120e3b | 88 | int n = 0; |
4f1043b4 SH |
89 | void *buf; |
90 | ||
91 | s->common.len = bdrv_getlength(bs); | |
92 | if (s->common.len < 0) { | |
65f46322 | 93 | block_job_completed(&s->common, s->common.len); |
4f1043b4 SH |
94 | return; |
95 | } | |
96 | ||
97 | end = s->common.len >> BDRV_SECTOR_BITS; | |
98 | buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE); | |
99 | ||
100 | /* Turn on copy-on-read for the whole block device so that guest read | |
101 | * requests help us make progress. Only do this when copying the entire | |
102 | * backing chain since the copy-on-read operation does not take base into | |
103 | * account. | |
104 | */ | |
105 | if (!base) { | |
106 | bdrv_enable_copy_on_read(bs); | |
107 | } | |
108 | ||
109 | for (sector_num = 0; sector_num < end; sector_num += n) { | |
4513eafe | 110 | uint64_t delay_ns = 0; |
f9749f28 | 111 | bool copy; |
4513eafe PB |
112 | |
113 | wait: | |
114 | /* Note that even when no rate limit is applied we need to yield | |
c57b6656 | 115 | * with no pending I/O here so that bdrv_drain_all() returns. |
4513eafe | 116 | */ |
7483d1e5 | 117 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
4f1043b4 SH |
118 | if (block_job_is_cancelled(&s->common)) { |
119 | break; | |
120 | } | |
121 | ||
c3e4f43a SW |
122 | copy = false; |
123 | ||
bdad13b9 PB |
124 | ret = bdrv_is_allocated(bs, sector_num, |
125 | STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n); | |
f9749f28 PB |
126 | if (ret == 1) { |
127 | /* Allocated in the top, no need to copy. */ | |
d663640c | 128 | } else if (ret >= 0) { |
f9749f28 PB |
129 | /* Copy if allocated in the intermediate images. Limit to the |
130 | * known-unallocated area [sector_num, sector_num+n). */ | |
4f578637 PB |
131 | ret = bdrv_is_allocated_above(bs->backing_hd, base, |
132 | sector_num, n, &n); | |
571cd9dc SH |
133 | |
134 | /* Finish early if end of backing file has been reached */ | |
135 | if (ret == 0 && n == 0) { | |
136 | n = end - sector_num; | |
137 | } | |
138 | ||
f9749f28 PB |
139 | copy = (ret == 1); |
140 | } | |
4f1043b4 | 141 | trace_stream_one_iteration(s, sector_num, n, ret); |
c3e4f43a | 142 | if (copy) { |
5094a6c0 | 143 | if (s->common.speed) { |
4513eafe | 144 | delay_ns = ratelimit_calculate_delay(&s->limit, n); |
5094a6c0 | 145 | if (delay_ns > 0) { |
4513eafe | 146 | goto wait; |
5094a6c0 SH |
147 | } |
148 | } | |
4f1043b4 SH |
149 | ret = stream_populate(bs, sector_num, n, buf); |
150 | } | |
151 | if (ret < 0) { | |
1d809098 PB |
152 | BlockErrorAction action = |
153 | block_job_error_action(&s->common, s->common.bs, s->on_error, | |
154 | true, -ret); | |
155 | if (action == BDRV_ACTION_STOP) { | |
156 | n = 0; | |
157 | continue; | |
158 | } | |
159 | if (error == 0) { | |
160 | error = ret; | |
161 | } | |
162 | if (action == BDRV_ACTION_REPORT) { | |
163 | break; | |
164 | } | |
4f1043b4 | 165 | } |
c8c3080f | 166 | ret = 0; |
4f1043b4 SH |
167 | |
168 | /* Publish progress */ | |
169 | s->common.offset += n * BDRV_SECTOR_SIZE; | |
170 | } | |
171 | ||
172 | if (!base) { | |
173 | bdrv_disable_copy_on_read(bs); | |
174 | } | |
175 | ||
1d809098 PB |
176 | /* Do not remove the backing file if an error was there but ignored. */ |
177 | ret = error; | |
178 | ||
3e914655 | 179 | if (!block_job_is_cancelled(&s->common) && sector_num == end && ret == 0) { |
f6133def | 180 | const char *base_id = NULL, *base_fmt = NULL; |
c8c3080f MT |
181 | if (base) { |
182 | base_id = s->backing_file_id; | |
f6133def PB |
183 | if (base->drv) { |
184 | base_fmt = base->drv->format_name; | |
185 | } | |
c8c3080f | 186 | } |
f6133def | 187 | ret = bdrv_change_backing_file(bs, base_id, base_fmt); |
5a67a104 | 188 | close_unused_images(bs, base, base_id); |
4f1043b4 SH |
189 | } |
190 | ||
191 | qemu_vfree(buf); | |
65f46322 | 192 | block_job_completed(&s->common, ret); |
4f1043b4 SH |
193 | } |
194 | ||
882ec7ce | 195 | static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp) |
5094a6c0 SH |
196 | { |
197 | StreamBlockJob *s = container_of(job, StreamBlockJob, common); | |
198 | ||
882ec7ce SH |
199 | if (speed < 0) { |
200 | error_set(errp, QERR_INVALID_PARAMETER, "speed"); | |
9e6636c7 | 201 | return; |
5094a6c0 | 202 | } |
6ef228fc | 203 | ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); |
5094a6c0 SH |
204 | } |
205 | ||
3fc4b10a | 206 | static const BlockJobDriver stream_job_driver = { |
4f1043b4 | 207 | .instance_size = sizeof(StreamBlockJob), |
79e14bf7 | 208 | .job_type = BLOCK_JOB_TYPE_STREAM, |
5094a6c0 | 209 | .set_speed = stream_set_speed, |
4f1043b4 SH |
210 | }; |
211 | ||
fd7f8c65 | 212 | void stream_start(BlockDriverState *bs, BlockDriverState *base, |
c83c66c3 | 213 | const char *base_id, int64_t speed, |
1d809098 | 214 | BlockdevOnError on_error, |
c83c66c3 | 215 | BlockDriverCompletionFunc *cb, |
fd7f8c65 | 216 | void *opaque, Error **errp) |
4f1043b4 SH |
217 | { |
218 | StreamBlockJob *s; | |
4f1043b4 | 219 | |
1d809098 PB |
220 | if ((on_error == BLOCKDEV_ON_ERROR_STOP || |
221 | on_error == BLOCKDEV_ON_ERROR_ENOSPC) && | |
222 | !bdrv_iostatus_is_enabled(bs)) { | |
223 | error_set(errp, QERR_INVALID_PARAMETER, "on-error"); | |
224 | return; | |
225 | } | |
226 | ||
3fc4b10a | 227 | s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp); |
4f1043b4 | 228 | if (!s) { |
fd7f8c65 | 229 | return; |
4f1043b4 SH |
230 | } |
231 | ||
232 | s->base = base; | |
c8c3080f MT |
233 | if (base_id) { |
234 | pstrcpy(s->backing_file_id, sizeof(s->backing_file_id), base_id); | |
235 | } | |
4f1043b4 | 236 | |
1d809098 | 237 | s->on_error = on_error; |
fa4478d5 PB |
238 | s->common.co = qemu_coroutine_create(stream_run); |
239 | trace_stream_start(bs, base, s, s->common.co, opaque); | |
240 | qemu_coroutine_enter(s->common.co, s); | |
4f1043b4 | 241 | } |