]>
Commit | Line | Data |
---|---|---|
4f1043b4 SH |
1 | /* |
2 | * Image streaming | |
3 | * | |
4 | * Copyright IBM, Corp. 2011 | |
5 | * | |
6 | * Authors: | |
7 | * Stefan Hajnoczi <[email protected]> | |
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 | /* | |
99136607 VSO |
25 | * Maximum chunk size to feed to copy-on-read. This should be |
26 | * large enough to process multiple clusters in a single call, so | |
27 | * that populating contiguous regions of the image is efficient. | |
4f1043b4 | 28 | */ |
99136607 | 29 | STREAM_CHUNK = 512 * 1024, /* in bytes */ |
4f1043b4 SH |
30 | }; |
31 | ||
32 | typedef struct StreamBlockJob { | |
33 | BlockJob common; | |
c624b015 | 34 | BlockDriverState *bottom; |
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, |
99136607 | 42 | int64_t offset, uint64_t bytes) |
4f1043b4 | 43 | { |
8493211c | 44 | assert(bytes < SIZE_MAX); |
4f1043b4 | 45 | |
99136607 VSO |
46 | return blk_co_preadv(blk, offset, bytes, NULL, |
47 | BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH); | |
4f1043b4 SH |
48 | } |
49 | ||
65854933 AG |
50 | static void stream_abort(Job *job) |
51 | { | |
52 | StreamBlockJob *s = container_of(job, StreamBlockJob, common.job); | |
53 | ||
54 | if (s->chain_frozen) { | |
55 | BlockJob *bjob = &s->common; | |
c624b015 | 56 | bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->bottom); |
65854933 AG |
57 | } |
58 | } | |
59 | ||
1b57488a | 60 | static int stream_prepare(Job *job) |
f3e69beb | 61 | { |
1908a559 KW |
62 | StreamBlockJob *s = container_of(job, StreamBlockJob, common.job); |
63 | BlockJob *bjob = &s->common; | |
1908a559 | 64 | BlockDriverState *bs = blk_bs(bjob->blk); |
c624b015 | 65 | BlockDriverState *base = backing_bs(s->bottom); |
12fa4af6 | 66 | Error *local_err = NULL; |
1b57488a | 67 | int ret = 0; |
f3e69beb | 68 | |
c624b015 | 69 | bdrv_unfreeze_backing_chain(bs, s->bottom); |
65854933 AG |
70 | s->chain_frozen = false; |
71 | ||
1b57488a | 72 | if (bs->backing) { |
f3e69beb SH |
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 | } | |
12fa4af6 | 80 | bdrv_set_backing_hd(bs, base, &local_err); |
8441d82d | 81 | ret = bdrv_change_backing_file(bs, base_id, base_fmt); |
12fa4af6 KW |
82 | if (local_err) { |
83 | error_report_err(local_err); | |
1b57488a | 84 | return -EPERM; |
12fa4af6 | 85 | } |
f3e69beb SH |
86 | } |
87 | ||
1b57488a JS |
88 | return ret; |
89 | } | |
90 | ||
91 | static void stream_clean(Job *job) | |
92 | { | |
93 | StreamBlockJob *s = container_of(job, StreamBlockJob, common.job); | |
94 | BlockJob *bjob = &s->common; | |
95 | BlockDriverState *bs = blk_bs(bjob->blk); | |
96 | ||
61b49e48 | 97 | /* Reopen the image back in read-only mode if necessary */ |
e7d22f8b | 98 | if (s->bs_read_only) { |
a170a91f | 99 | /* Give up write permissions before making it read-only */ |
1908a559 | 100 | blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort); |
e7d22f8b | 101 | bdrv_reopen_set_read_only(bs, true, NULL); |
61b49e48 AG |
102 | } |
103 | ||
f3e69beb | 104 | g_free(s->backing_file_str); |
f3e69beb SH |
105 | } |
106 | ||
f67432a2 | 107 | static int coroutine_fn stream_run(Job *job, Error **errp) |
4f1043b4 | 108 | { |
f67432a2 | 109 | StreamBlockJob *s = container_of(job, StreamBlockJob, common.job); |
03e35d82 KW |
110 | BlockBackend *blk = s->common.blk; |
111 | BlockDriverState *bs = blk_bs(blk); | |
c624b015 | 112 | bool enable_cor = !backing_bs(s->bottom); |
05df8a6a | 113 | int64_t len; |
d535435f | 114 | int64_t offset = 0; |
f14a39cc | 115 | uint64_t delay_ns = 0; |
1d809098 | 116 | int error = 0; |
4f1043b4 | 117 | int ret = 0; |
51b0a488 | 118 | int64_t n = 0; /* bytes */ |
4f1043b4 | 119 | |
c624b015 AS |
120 | if (bs == s->bottom) { |
121 | /* Nothing to stream */ | |
96a07d5b | 122 | return 0; |
f4a193e7 HR |
123 | } |
124 | ||
05df8a6a KW |
125 | len = bdrv_getlength(bs); |
126 | if (len < 0) { | |
96a07d5b | 127 | return len; |
4f1043b4 | 128 | } |
30a5c887 | 129 | job_progress_set_remaining(&s->common.job, len); |
4f1043b4 | 130 | |
4f1043b4 SH |
131 | /* Turn on copy-on-read for the whole block device so that guest read |
132 | * requests help us make progress. Only do this when copying the entire | |
133 | * backing chain since the copy-on-read operation does not take base into | |
134 | * account. | |
135 | */ | |
c624b015 | 136 | if (enable_cor) { |
4f1043b4 SH |
137 | bdrv_enable_copy_on_read(bs); |
138 | } | |
139 | ||
05df8a6a | 140 | for ( ; offset < len; offset += n) { |
f9749f28 | 141 | bool copy; |
4513eafe | 142 | |
4513eafe | 143 | /* Note that even when no rate limit is applied we need to yield |
c57b6656 | 144 | * with no pending I/O here so that bdrv_drain_all() returns. |
4513eafe | 145 | */ |
5d43e86e | 146 | job_sleep_ns(&s->common.job, delay_ns); |
daa7f2f9 | 147 | if (job_is_cancelled(&s->common.job)) { |
4f1043b4 SH |
148 | break; |
149 | } | |
150 | ||
c3e4f43a SW |
151 | copy = false; |
152 | ||
99136607 | 153 | ret = bdrv_is_allocated(bs, offset, STREAM_CHUNK, &n); |
f9749f28 PB |
154 | if (ret == 1) { |
155 | /* Allocated in the top, no need to copy. */ | |
d663640c | 156 | } else if (ret >= 0) { |
f9749f28 | 157 | /* Copy if allocated in the intermediate images. Limit to the |
d535435f | 158 | * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */ |
c624b015 | 159 | ret = bdrv_is_allocated_above(backing_bs(bs), s->bottom, true, |
51b0a488 | 160 | offset, n, &n); |
571cd9dc SH |
161 | /* Finish early if end of backing file has been reached */ |
162 | if (ret == 0 && n == 0) { | |
05df8a6a | 163 | n = len - offset; |
571cd9dc SH |
164 | } |
165 | ||
f9749f28 PB |
166 | copy = (ret == 1); |
167 | } | |
51b0a488 | 168 | trace_stream_one_iteration(s, offset, n, ret); |
c3e4f43a | 169 | if (copy) { |
99136607 | 170 | ret = stream_populate(blk, offset, n); |
4f1043b4 SH |
171 | } |
172 | if (ret < 0) { | |
1d809098 | 173 | BlockErrorAction action = |
81e254dc | 174 | block_job_error_action(&s->common, s->on_error, true, -ret); |
a589569f | 175 | if (action == BLOCK_ERROR_ACTION_STOP) { |
1d809098 PB |
176 | n = 0; |
177 | continue; | |
178 | } | |
179 | if (error == 0) { | |
180 | error = ret; | |
181 | } | |
a589569f | 182 | if (action == BLOCK_ERROR_ACTION_REPORT) { |
1d809098 PB |
183 | break; |
184 | } | |
4f1043b4 | 185 | } |
c8c3080f | 186 | ret = 0; |
4f1043b4 SH |
187 | |
188 | /* Publish progress */ | |
30a5c887 | 189 | job_progress_update(&s->common.job, n); |
dee81d51 KW |
190 | if (copy) { |
191 | delay_ns = block_job_ratelimit_get_delay(&s->common, n); | |
2fe4bba1 KW |
192 | } else { |
193 | delay_ns = 0; | |
f14a39cc | 194 | } |
4f1043b4 SH |
195 | } |
196 | ||
c624b015 | 197 | if (enable_cor) { |
4f1043b4 SH |
198 | bdrv_disable_copy_on_read(bs); |
199 | } | |
200 | ||
96a07d5b AS |
201 | /* Do not remove the backing file if an error was there but ignored. */ |
202 | return error; | |
4f1043b4 SH |
203 | } |
204 | ||
3fc4b10a | 205 | static const BlockJobDriver stream_job_driver = { |
33e9e9bd KW |
206 | .job_driver = { |
207 | .instance_size = sizeof(StreamBlockJob), | |
252291ea | 208 | .job_type = JOB_TYPE_STREAM, |
80fa2c75 | 209 | .free = block_job_free, |
f67432a2 | 210 | .run = stream_run, |
1b57488a | 211 | .prepare = stream_prepare, |
65854933 | 212 | .abort = stream_abort, |
1b57488a | 213 | .clean = stream_clean, |
b15de828 | 214 | .user_resume = block_job_user_resume, |
33e9e9bd | 215 | }, |
4f1043b4 SH |
216 | }; |
217 | ||
2323322e AG |
218 | void stream_start(const char *job_id, BlockDriverState *bs, |
219 | BlockDriverState *base, const char *backing_file_str, | |
cf6320df JS |
220 | int creation_flags, int64_t speed, |
221 | BlockdevOnError on_error, Error **errp) | |
4f1043b4 SH |
222 | { |
223 | StreamBlockJob *s; | |
61b49e48 | 224 | BlockDriverState *iter; |
e7d22f8b | 225 | bool bs_read_only; |
c624b015 AS |
226 | int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED; |
227 | BlockDriverState *bottom = bdrv_find_overlay(bs, base); | |
4f1043b4 | 228 | |
c624b015 | 229 | if (bdrv_freeze_backing_chain(bs, bottom, errp) < 0) { |
20509c4b AG |
230 | return; |
231 | } | |
232 | ||
61b49e48 | 233 | /* Make sure that the image is opened in read-write mode */ |
e7d22f8b AG |
234 | bs_read_only = bdrv_is_read_only(bs); |
235 | if (bs_read_only) { | |
236 | if (bdrv_reopen_set_read_only(bs, false, errp) != 0) { | |
20509c4b AG |
237 | bs_read_only = false; |
238 | goto fail; | |
61b49e48 AG |
239 | } |
240 | } | |
241 | ||
a170a91f KW |
242 | /* Prevent concurrent jobs trying to modify the graph structure here, we |
243 | * already have our own plans. Also don't allow resize as the image size is | |
244 | * queried only at the job start and then cached. */ | |
75859b94 | 245 | s = block_job_create(job_id, &stream_job_driver, NULL, bs, |
c624b015 AS |
246 | basic_flags | BLK_PERM_GRAPH_MOD, |
247 | basic_flags | BLK_PERM_WRITE, | |
cf6320df | 248 | speed, creation_flags, NULL, NULL, errp); |
a170a91f KW |
249 | if (!s) { |
250 | goto fail; | |
251 | } | |
252 | ||
253 | /* Block all intermediate nodes between bs and base, because they will | |
254 | * disappear from the chain after this operation. The streaming job reads | |
c624b015 AS |
255 | * every block only once, assuming that it doesn't change, so forbid writes |
256 | * and resizes. Reassign the base node pointer because the backing BS of the | |
257 | * bottom node might change after the call to bdrv_reopen_set_read_only() | |
258 | * due to parallel block jobs running. | |
259 | */ | |
260 | base = backing_bs(bottom); | |
61b49e48 | 261 | for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) { |
76d554e2 | 262 | block_job_add_bdrv(&s->common, "intermediate node", iter, 0, |
c624b015 | 263 | basic_flags, &error_abort); |
61b49e48 AG |
264 | } |
265 | ||
c624b015 | 266 | s->bottom = bottom; |
13d8cc51 | 267 | s->backing_file_str = g_strdup(backing_file_str); |
e7d22f8b | 268 | s->bs_read_only = bs_read_only; |
65854933 | 269 | s->chain_frozen = true; |
4f1043b4 | 270 | |
1d809098 | 271 | s->on_error = on_error; |
5ccac6f1 | 272 | trace_stream_start(bs, base, s); |
da01ff7f | 273 | job_start(&s->common.job); |
a170a91f KW |
274 | return; |
275 | ||
276 | fail: | |
e7d22f8b AG |
277 | if (bs_read_only) { |
278 | bdrv_reopen_set_read_only(bs, true, NULL); | |
a170a91f | 279 | } |
17a7c392 | 280 | bdrv_unfreeze_backing_chain(bs, bottom); |
4f1043b4 | 281 | } |