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 "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"
22
23enum {
24 /*
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.
28 */
29 STREAM_CHUNK = 512 * 1024, /* in bytes */
30};
31
32typedef struct StreamBlockJob {
33 BlockJob common;
34 BlockDriverState *bottom;
35 BlockdevOnError on_error;
36 char *backing_file_str;
37 bool bs_read_only;
38 bool chain_frozen;
39} StreamBlockJob;
40
41static int coroutine_fn stream_populate(BlockBackend *blk,
42 int64_t offset, uint64_t bytes)
43{
44 assert(bytes < SIZE_MAX);
45
46 return blk_co_preadv(blk, offset, bytes, NULL,
47 BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
48}
49
50static 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;
56 bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->bottom);
57 }
58}
59
60static int stream_prepare(Job *job)
61{
62 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
63 BlockJob *bjob = &s->common;
64 BlockDriverState *bs = blk_bs(bjob->blk);
65 BlockDriverState *base = backing_bs(s->bottom);
66 Error *local_err = NULL;
67 int ret = 0;
68
69 bdrv_unfreeze_backing_chain(bs, s->bottom);
70 s->chain_frozen = false;
71
72 if (bs->backing) {
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 }
80 bdrv_set_backing_hd(bs, base, &local_err);
81 ret = bdrv_change_backing_file(bs, base_id, base_fmt);
82 if (local_err) {
83 error_report_err(local_err);
84 return -EPERM;
85 }
86 }
87
88 return ret;
89}
90
91static 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
97 /* Reopen the image back in read-only mode if necessary */
98 if (s->bs_read_only) {
99 /* Give up write permissions before making it read-only */
100 blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
101 bdrv_reopen_set_read_only(bs, true, NULL);
102 }
103
104 g_free(s->backing_file_str);
105}
106
107static int coroutine_fn stream_run(Job *job, Error **errp)
108{
109 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
110 BlockBackend *blk = s->common.blk;
111 BlockDriverState *bs = blk_bs(blk);
112 bool enable_cor = !backing_bs(s->bottom);
113 int64_t len;
114 int64_t offset = 0;
115 uint64_t delay_ns = 0;
116 int error = 0;
117 int ret = 0;
118 int64_t n = 0; /* bytes */
119
120 if (bs == s->bottom) {
121 /* Nothing to stream */
122 return 0;
123 }
124
125 len = bdrv_getlength(bs);
126 if (len < 0) {
127 return len;
128 }
129 job_progress_set_remaining(&s->common.job, len);
130
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 */
136 if (enable_cor) {
137 bdrv_enable_copy_on_read(bs);
138 }
139
140 for ( ; offset < len; offset += n) {
141 bool copy;
142
143 /* Note that even when no rate limit is applied we need to yield
144 * with no pending I/O here so that bdrv_drain_all() returns.
145 */
146 job_sleep_ns(&s->common.job, delay_ns);
147 if (job_is_cancelled(&s->common.job)) {
148 break;
149 }
150
151 copy = false;
152
153 ret = bdrv_is_allocated(bs, offset, STREAM_CHUNK, &n);
154 if (ret == 1) {
155 /* Allocated in the top, no need to copy. */
156 } else if (ret >= 0) {
157 /* Copy if allocated in the intermediate images. Limit to the
158 * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */
159 ret = bdrv_is_allocated_above(backing_bs(bs), s->bottom, true,
160 offset, n, &n);
161 /* Finish early if end of backing file has been reached */
162 if (ret == 0 && n == 0) {
163 n = len - offset;
164 }
165
166 copy = (ret == 1);
167 }
168 trace_stream_one_iteration(s, offset, n, ret);
169 if (copy) {
170 ret = stream_populate(blk, offset, n);
171 }
172 if (ret < 0) {
173 BlockErrorAction action =
174 block_job_error_action(&s->common, s->on_error, true, -ret);
175 if (action == BLOCK_ERROR_ACTION_STOP) {
176 n = 0;
177 continue;
178 }
179 if (error == 0) {
180 error = ret;
181 }
182 if (action == BLOCK_ERROR_ACTION_REPORT) {
183 break;
184 }
185 }
186 ret = 0;
187
188 /* Publish progress */
189 job_progress_update(&s->common.job, n);
190 if (copy) {
191 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
192 } else {
193 delay_ns = 0;
194 }
195 }
196
197 if (enable_cor) {
198 bdrv_disable_copy_on_read(bs);
199 }
200
201 /* Do not remove the backing file if an error was there but ignored. */
202 return error;
203}
204
205static const BlockJobDriver stream_job_driver = {
206 .job_driver = {
207 .instance_size = sizeof(StreamBlockJob),
208 .job_type = JOB_TYPE_STREAM,
209 .free = block_job_free,
210 .run = stream_run,
211 .prepare = stream_prepare,
212 .abort = stream_abort,
213 .clean = stream_clean,
214 .user_resume = block_job_user_resume,
215 .drain = block_job_drain,
216 },
217};
218
219void stream_start(const char *job_id, BlockDriverState *bs,
220 BlockDriverState *base, const char *backing_file_str,
221 int creation_flags, int64_t speed,
222 BlockdevOnError on_error, Error **errp)
223{
224 StreamBlockJob *s;
225 BlockDriverState *iter;
226 bool bs_read_only;
227 int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
228 BlockDriverState *bottom = bdrv_find_overlay(bs, base);
229
230 if (bdrv_freeze_backing_chain(bs, bottom, errp) < 0) {
231 return;
232 }
233
234 /* Make sure that the image is opened in read-write mode */
235 bs_read_only = bdrv_is_read_only(bs);
236 if (bs_read_only) {
237 if (bdrv_reopen_set_read_only(bs, false, errp) != 0) {
238 bs_read_only = false;
239 goto fail;
240 }
241 }
242
243 /* Prevent concurrent jobs trying to modify the graph structure here, we
244 * already have our own plans. Also don't allow resize as the image size is
245 * queried only at the job start and then cached. */
246 s = block_job_create(job_id, &stream_job_driver, NULL, bs,
247 basic_flags | BLK_PERM_GRAPH_MOD,
248 basic_flags | BLK_PERM_WRITE,
249 speed, creation_flags, NULL, NULL, errp);
250 if (!s) {
251 goto fail;
252 }
253
254 /* Block all intermediate nodes between bs and base, because they will
255 * disappear from the chain after this operation. The streaming job reads
256 * every block only once, assuming that it doesn't change, so forbid writes
257 * and resizes. Reassign the base node pointer because the backing BS of the
258 * bottom node might change after the call to bdrv_reopen_set_read_only()
259 * due to parallel block jobs running.
260 */
261 base = backing_bs(bottom);
262 for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
263 block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
264 basic_flags, &error_abort);
265 }
266
267 s->bottom = bottom;
268 s->backing_file_str = g_strdup(backing_file_str);
269 s->bs_read_only = bs_read_only;
270 s->chain_frozen = true;
271
272 s->on_error = on_error;
273 trace_stream_start(bs, base, s);
274 job_start(&s->common.job);
275 return;
276
277fail:
278 if (bs_read_only) {
279 bdrv_reopen_set_read_only(bs, true, NULL);
280 }
281 bdrv_unfreeze_backing_chain(bs, bottom);
282}
283