]>
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 | 16 | #include "block/block_int.h" |
c87621ea | 17 | #include "block/blockjob_int.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 | ||
32 | typedef struct StreamBlockJob { | |
33 | BlockJob common; | |
34 | BlockDriverState *base; | |
1d809098 | 35 | BlockdevOnError on_error; |
13d8cc51 | 36 | char *backing_file_str; |
e7d22f8b | 37 | bool bs_read_only; |
65854933 | 38 | bool chain_frozen; |
4f1043b4 SH |
39 | } StreamBlockJob; |
40 | ||
03e35d82 | 41 | static int coroutine_fn stream_populate(BlockBackend *blk, |
8493211c | 42 | int64_t offset, uint64_t bytes, |
4f1043b4 SH |
43 | void *buf) |
44 | { | |
8040446d | 45 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes); |
4f1043b4 | 46 | |
8493211c | 47 | assert(bytes < SIZE_MAX); |
4f1043b4 SH |
48 | |
49 | /* Copy-on-read the unallocated clusters */ | |
8493211c | 50 | return blk_co_preadv(blk, offset, qiov.size, &qiov, BDRV_REQ_COPY_ON_READ); |
4f1043b4 SH |
51 | } |
52 | ||
65854933 AG |
53 | static void stream_abort(Job *job) |
54 | { | |
55 | StreamBlockJob *s = container_of(job, StreamBlockJob, common.job); | |
56 | ||
57 | if (s->chain_frozen) { | |
58 | BlockJob *bjob = &s->common; | |
59 | bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->base); | |
60 | } | |
61 | } | |
62 | ||
1b57488a | 63 | static int stream_prepare(Job *job) |
f3e69beb | 64 | { |
1908a559 KW |
65 | StreamBlockJob *s = container_of(job, StreamBlockJob, common.job); |
66 | BlockJob *bjob = &s->common; | |
1908a559 | 67 | BlockDriverState *bs = blk_bs(bjob->blk); |
f3e69beb | 68 | BlockDriverState *base = s->base; |
12fa4af6 | 69 | Error *local_err = NULL; |
1b57488a | 70 | int ret = 0; |
f3e69beb | 71 | |
65854933 AG |
72 | bdrv_unfreeze_backing_chain(bs, base); |
73 | s->chain_frozen = false; | |
74 | ||
1b57488a | 75 | if (bs->backing) { |
f3e69beb SH |
76 | const char *base_id = NULL, *base_fmt = NULL; |
77 | if (base) { | |
78 | base_id = s->backing_file_str; | |
79 | if (base->drv) { | |
80 | base_fmt = base->drv->format_name; | |
81 | } | |
82 | } | |
eb23654d | 83 | ret = bdrv_change_backing_file(bs, base_id, base_fmt); |
12fa4af6 KW |
84 | bdrv_set_backing_hd(bs, base, &local_err); |
85 | if (local_err) { | |
86 | error_report_err(local_err); | |
1b57488a | 87 | return -EPERM; |
12fa4af6 | 88 | } |
f3e69beb SH |
89 | } |
90 | ||
1b57488a JS |
91 | return ret; |
92 | } | |
93 | ||
94 | static void stream_clean(Job *job) | |
95 | { | |
96 | StreamBlockJob *s = container_of(job, StreamBlockJob, common.job); | |
97 | BlockJob *bjob = &s->common; | |
98 | BlockDriverState *bs = blk_bs(bjob->blk); | |
99 | ||
61b49e48 | 100 | /* Reopen the image back in read-only mode if necessary */ |
e7d22f8b | 101 | if (s->bs_read_only) { |
a170a91f | 102 | /* Give up write permissions before making it read-only */ |
1908a559 | 103 | blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort); |
e7d22f8b | 104 | bdrv_reopen_set_read_only(bs, true, NULL); |
61b49e48 AG |
105 | } |
106 | ||
f3e69beb | 107 | g_free(s->backing_file_str); |
f3e69beb SH |
108 | } |
109 | ||
f67432a2 | 110 | static int coroutine_fn stream_run(Job *job, Error **errp) |
4f1043b4 | 111 | { |
f67432a2 | 112 | StreamBlockJob *s = container_of(job, StreamBlockJob, common.job); |
03e35d82 KW |
113 | BlockBackend *blk = s->common.blk; |
114 | BlockDriverState *bs = blk_bs(blk); | |
c8c3080f | 115 | BlockDriverState *base = s->base; |
05df8a6a | 116 | int64_t len; |
d535435f | 117 | int64_t offset = 0; |
f14a39cc | 118 | uint64_t delay_ns = 0; |
1d809098 | 119 | int error = 0; |
4f1043b4 | 120 | int ret = 0; |
51b0a488 | 121 | int64_t n = 0; /* bytes */ |
4f1043b4 SH |
122 | void *buf; |
123 | ||
760e0063 | 124 | if (!bs->backing) { |
6578629e | 125 | goto out; |
f4a193e7 HR |
126 | } |
127 | ||
05df8a6a KW |
128 | len = bdrv_getlength(bs); |
129 | if (len < 0) { | |
130 | ret = len; | |
6578629e | 131 | goto out; |
4f1043b4 | 132 | } |
30a5c887 | 133 | job_progress_set_remaining(&s->common.job, len); |
4f1043b4 | 134 | |
4f1043b4 SH |
135 | buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE); |
136 | ||
137 | /* Turn on copy-on-read for the whole block device so that guest read | |
138 | * requests help us make progress. Only do this when copying the entire | |
139 | * backing chain since the copy-on-read operation does not take base into | |
140 | * account. | |
141 | */ | |
142 | if (!base) { | |
143 | bdrv_enable_copy_on_read(bs); | |
144 | } | |
145 | ||
05df8a6a | 146 | for ( ; offset < len; offset += n) { |
f9749f28 | 147 | bool copy; |
4513eafe | 148 | |
4513eafe | 149 | /* Note that even when no rate limit is applied we need to yield |
c57b6656 | 150 | * with no pending I/O here so that bdrv_drain_all() returns. |
4513eafe | 151 | */ |
5d43e86e | 152 | job_sleep_ns(&s->common.job, delay_ns); |
daa7f2f9 | 153 | if (job_is_cancelled(&s->common.job)) { |
4f1043b4 SH |
154 | break; |
155 | } | |
156 | ||
c3e4f43a SW |
157 | copy = false; |
158 | ||
51b0a488 | 159 | ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &n); |
f9749f28 PB |
160 | if (ret == 1) { |
161 | /* Allocated in the top, no need to copy. */ | |
d663640c | 162 | } else if (ret >= 0) { |
f9749f28 | 163 | /* Copy if allocated in the intermediate images. Limit to the |
d535435f | 164 | * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */ |
760e0063 | 165 | ret = bdrv_is_allocated_above(backing_bs(bs), base, |
51b0a488 | 166 | offset, n, &n); |
571cd9dc SH |
167 | |
168 | /* Finish early if end of backing file has been reached */ | |
169 | if (ret == 0 && n == 0) { | |
05df8a6a | 170 | n = len - offset; |
571cd9dc SH |
171 | } |
172 | ||
f9749f28 PB |
173 | copy = (ret == 1); |
174 | } | |
51b0a488 | 175 | trace_stream_one_iteration(s, offset, n, ret); |
c3e4f43a | 176 | if (copy) { |
51b0a488 | 177 | ret = stream_populate(blk, offset, n, buf); |
4f1043b4 SH |
178 | } |
179 | if (ret < 0) { | |
1d809098 | 180 | BlockErrorAction action = |
81e254dc | 181 | block_job_error_action(&s->common, s->on_error, true, -ret); |
a589569f | 182 | if (action == BLOCK_ERROR_ACTION_STOP) { |
1d809098 PB |
183 | n = 0; |
184 | continue; | |
185 | } | |
186 | if (error == 0) { | |
187 | error = ret; | |
188 | } | |
a589569f | 189 | if (action == BLOCK_ERROR_ACTION_REPORT) { |
1d809098 PB |
190 | break; |
191 | } | |
4f1043b4 | 192 | } |
c8c3080f | 193 | ret = 0; |
4f1043b4 SH |
194 | |
195 | /* Publish progress */ | |
30a5c887 | 196 | job_progress_update(&s->common.job, n); |
dee81d51 KW |
197 | if (copy) { |
198 | delay_ns = block_job_ratelimit_get_delay(&s->common, n); | |
2fe4bba1 KW |
199 | } else { |
200 | delay_ns = 0; | |
f14a39cc | 201 | } |
4f1043b4 SH |
202 | } |
203 | ||
204 | if (!base) { | |
205 | bdrv_disable_copy_on_read(bs); | |
206 | } | |
207 | ||
1d809098 PB |
208 | /* Do not remove the backing file if an error was there but ignored. */ |
209 | ret = error; | |
210 | ||
4f1043b4 | 211 | qemu_vfree(buf); |
f3e69beb | 212 | |
6578629e | 213 | out: |
f3e69beb | 214 | /* Modify backing chain and close BDSes in main loop */ |
f67432a2 | 215 | return ret; |
4f1043b4 SH |
216 | } |
217 | ||
3fc4b10a | 218 | static const BlockJobDriver stream_job_driver = { |
33e9e9bd KW |
219 | .job_driver = { |
220 | .instance_size = sizeof(StreamBlockJob), | |
252291ea | 221 | .job_type = JOB_TYPE_STREAM, |
80fa2c75 | 222 | .free = block_job_free, |
f67432a2 | 223 | .run = stream_run, |
1b57488a | 224 | .prepare = stream_prepare, |
65854933 | 225 | .abort = stream_abort, |
1b57488a | 226 | .clean = stream_clean, |
b15de828 | 227 | .user_resume = block_job_user_resume, |
b69f777d | 228 | .drain = block_job_drain, |
33e9e9bd | 229 | }, |
4f1043b4 SH |
230 | }; |
231 | ||
2323322e AG |
232 | void stream_start(const char *job_id, BlockDriverState *bs, |
233 | BlockDriverState *base, const char *backing_file_str, | |
cf6320df JS |
234 | int creation_flags, int64_t speed, |
235 | BlockdevOnError on_error, Error **errp) | |
4f1043b4 SH |
236 | { |
237 | StreamBlockJob *s; | |
61b49e48 | 238 | BlockDriverState *iter; |
e7d22f8b | 239 | bool bs_read_only; |
4f1043b4 | 240 | |
20509c4b AG |
241 | if (bdrv_freeze_backing_chain(bs, base, errp) < 0) { |
242 | return; | |
243 | } | |
244 | ||
61b49e48 | 245 | /* Make sure that the image is opened in read-write mode */ |
e7d22f8b AG |
246 | bs_read_only = bdrv_is_read_only(bs); |
247 | if (bs_read_only) { | |
248 | if (bdrv_reopen_set_read_only(bs, false, errp) != 0) { | |
20509c4b AG |
249 | bs_read_only = false; |
250 | goto fail; | |
61b49e48 AG |
251 | } |
252 | } | |
253 | ||
a170a91f KW |
254 | /* Prevent concurrent jobs trying to modify the graph structure here, we |
255 | * already have our own plans. Also don't allow resize as the image size is | |
256 | * queried only at the job start and then cached. */ | |
75859b94 | 257 | s = block_job_create(job_id, &stream_job_driver, NULL, bs, |
a170a91f KW |
258 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | |
259 | BLK_PERM_GRAPH_MOD, | |
260 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
261 | BLK_PERM_WRITE, | |
cf6320df | 262 | speed, creation_flags, NULL, NULL, errp); |
a170a91f KW |
263 | if (!s) { |
264 | goto fail; | |
265 | } | |
266 | ||
267 | /* Block all intermediate nodes between bs and base, because they will | |
268 | * disappear from the chain after this operation. The streaming job reads | |
269 | * every block only once, assuming that it doesn't change, so block writes | |
270 | * and resizes. */ | |
61b49e48 | 271 | for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) { |
76d554e2 | 272 | block_job_add_bdrv(&s->common, "intermediate node", iter, 0, |
a170a91f KW |
273 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED, |
274 | &error_abort); | |
61b49e48 AG |
275 | } |
276 | ||
4f1043b4 | 277 | s->base = base; |
13d8cc51 | 278 | s->backing_file_str = g_strdup(backing_file_str); |
e7d22f8b | 279 | s->bs_read_only = bs_read_only; |
65854933 | 280 | s->chain_frozen = true; |
4f1043b4 | 281 | |
1d809098 | 282 | s->on_error = on_error; |
5ccac6f1 | 283 | trace_stream_start(bs, base, s); |
da01ff7f | 284 | job_start(&s->common.job); |
a170a91f KW |
285 | return; |
286 | ||
287 | fail: | |
e7d22f8b AG |
288 | if (bs_read_only) { |
289 | bdrv_reopen_set_read_only(bs, true, NULL); | |
a170a91f | 290 | } |
20509c4b | 291 | bdrv_unfreeze_backing_chain(bs, base); |
4f1043b4 | 292 | } |