]> git.proxmox.com Git - mirror_qemu.git/blob - block/stream.c
Merge tag 'qemu-slof-20211112' of github.com:aik/qemu into ppc-next
[mirror_qemu.git] / block / stream.c
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 "qapi/qmp/qdict.h"
21 #include "qemu/ratelimit.h"
22 #include "sysemu/block-backend.h"
23 #include "block/copy-on-read.h"
24
25 enum {
26 /*
27 * Maximum chunk size to feed to copy-on-read. This should be
28 * large enough to process multiple clusters in a single call, so
29 * that populating contiguous regions of the image is efficient.
30 */
31 STREAM_CHUNK = 512 * 1024, /* in bytes */
32 };
33
34 typedef struct StreamBlockJob {
35 BlockJob common;
36 BlockDriverState *base_overlay; /* COW overlay (stream from this) */
37 BlockDriverState *above_base; /* Node directly above the base */
38 BlockDriverState *cor_filter_bs;
39 BlockDriverState *target_bs;
40 BlockdevOnError on_error;
41 char *backing_file_str;
42 bool bs_read_only;
43 } StreamBlockJob;
44
45 static int coroutine_fn stream_populate(BlockBackend *blk,
46 int64_t offset, uint64_t bytes)
47 {
48 assert(bytes < SIZE_MAX);
49
50 return blk_co_preadv(blk, offset, bytes, NULL, BDRV_REQ_PREFETCH);
51 }
52
53 static int stream_prepare(Job *job)
54 {
55 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
56 BlockDriverState *unfiltered_bs = bdrv_skip_filters(s->target_bs);
57 BlockDriverState *base;
58 BlockDriverState *unfiltered_base;
59 Error *local_err = NULL;
60 int ret = 0;
61
62 /* We should drop filter at this point, as filter hold the backing chain */
63 bdrv_cor_filter_drop(s->cor_filter_bs);
64 s->cor_filter_bs = NULL;
65
66 base = bdrv_filter_or_cow_bs(s->above_base);
67 unfiltered_base = bdrv_skip_filters(base);
68
69 if (bdrv_cow_child(unfiltered_bs)) {
70 const char *base_id = NULL, *base_fmt = NULL;
71 if (unfiltered_base) {
72 base_id = s->backing_file_str ?: unfiltered_base->filename;
73 if (unfiltered_base->drv) {
74 base_fmt = unfiltered_base->drv->format_name;
75 }
76 }
77 bdrv_set_backing_hd(unfiltered_bs, base, &local_err);
78 ret = bdrv_change_backing_file(unfiltered_bs, base_id, base_fmt, false);
79 if (local_err) {
80 error_report_err(local_err);
81 return -EPERM;
82 }
83 }
84
85 return ret;
86 }
87
88 static void stream_clean(Job *job)
89 {
90 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
91 BlockJob *bjob = &s->common;
92
93 if (s->cor_filter_bs) {
94 bdrv_cor_filter_drop(s->cor_filter_bs);
95 s->cor_filter_bs = NULL;
96 }
97
98 /* Reopen the image back in read-only mode if necessary */
99 if (s->bs_read_only) {
100 /* Give up write permissions before making it read-only */
101 blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
102 bdrv_reopen_set_read_only(s->target_bs, true, NULL);
103 }
104
105 g_free(s->backing_file_str);
106 }
107
108 static int coroutine_fn stream_run(Job *job, Error **errp)
109 {
110 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
111 BlockBackend *blk = s->common.blk;
112 BlockDriverState *unfiltered_bs = bdrv_skip_filters(s->target_bs);
113 int64_t len;
114 int64_t offset = 0;
115 uint64_t delay_ns = 0;
116 int error = 0;
117 int64_t n = 0; /* bytes */
118
119 if (unfiltered_bs == s->base_overlay) {
120 /* Nothing to stream */
121 return 0;
122 }
123
124 len = bdrv_getlength(s->target_bs);
125 if (len < 0) {
126 return len;
127 }
128 job_progress_set_remaining(&s->common.job, len);
129
130 for ( ; offset < len; offset += n) {
131 bool copy;
132 int ret;
133
134 /* Note that even when no rate limit is applied we need to yield
135 * with no pending I/O here so that bdrv_drain_all() returns.
136 */
137 job_sleep_ns(&s->common.job, delay_ns);
138 if (job_is_cancelled(&s->common.job)) {
139 break;
140 }
141
142 copy = false;
143
144 ret = bdrv_is_allocated(unfiltered_bs, offset, STREAM_CHUNK, &n);
145 if (ret == 1) {
146 /* Allocated in the top, no need to copy. */
147 } else if (ret >= 0) {
148 /* Copy if allocated in the intermediate images. Limit to the
149 * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */
150 ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
151 s->base_overlay, true,
152 offset, n, &n);
153 /* Finish early if end of backing file has been reached */
154 if (ret == 0 && n == 0) {
155 n = len - offset;
156 }
157
158 copy = (ret > 0);
159 }
160 trace_stream_one_iteration(s, offset, n, ret);
161 if (copy) {
162 ret = stream_populate(blk, offset, n);
163 }
164 if (ret < 0) {
165 BlockErrorAction action =
166 block_job_error_action(&s->common, s->on_error, true, -ret);
167 if (action == BLOCK_ERROR_ACTION_STOP) {
168 n = 0;
169 continue;
170 }
171 if (error == 0) {
172 error = ret;
173 }
174 if (action == BLOCK_ERROR_ACTION_REPORT) {
175 break;
176 }
177 }
178
179 /* Publish progress */
180 job_progress_update(&s->common.job, n);
181 if (copy) {
182 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
183 } else {
184 delay_ns = 0;
185 }
186 }
187
188 /* Do not remove the backing file if an error was there but ignored. */
189 return error;
190 }
191
192 static const BlockJobDriver stream_job_driver = {
193 .job_driver = {
194 .instance_size = sizeof(StreamBlockJob),
195 .job_type = JOB_TYPE_STREAM,
196 .free = block_job_free,
197 .run = stream_run,
198 .prepare = stream_prepare,
199 .clean = stream_clean,
200 .user_resume = block_job_user_resume,
201 },
202 };
203
204 void stream_start(const char *job_id, BlockDriverState *bs,
205 BlockDriverState *base, const char *backing_file_str,
206 BlockDriverState *bottom,
207 int creation_flags, int64_t speed,
208 BlockdevOnError on_error,
209 const char *filter_node_name,
210 Error **errp)
211 {
212 StreamBlockJob *s = NULL;
213 BlockDriverState *iter;
214 bool bs_read_only;
215 int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
216 BlockDriverState *base_overlay;
217 BlockDriverState *cor_filter_bs = NULL;
218 BlockDriverState *above_base;
219 QDict *opts;
220 int ret;
221
222 assert(!(base && bottom));
223 assert(!(backing_file_str && bottom));
224
225 if (bottom) {
226 /*
227 * New simple interface. The code is written in terms of old interface
228 * with @base parameter (still, it doesn't freeze link to base, so in
229 * this mean old code is correct for new interface). So, for now, just
230 * emulate base_overlay and above_base. Still, when old interface
231 * finally removed, we should refactor code to use only "bottom", but
232 * not "*base*" things.
233 */
234 assert(!bottom->drv->is_filter);
235 base_overlay = above_base = bottom;
236 } else {
237 base_overlay = bdrv_find_overlay(bs, base);
238 if (!base_overlay) {
239 error_setg(errp, "'%s' is not in the backing chain of '%s'",
240 base->node_name, bs->node_name);
241 return;
242 }
243
244 /*
245 * Find the node directly above @base. @base_overlay is a COW overlay,
246 * so it must have a bdrv_cow_child(), but it is the immediate overlay
247 * of @base, so between the two there can only be filters.
248 */
249 above_base = base_overlay;
250 if (bdrv_cow_bs(above_base) != base) {
251 above_base = bdrv_cow_bs(above_base);
252 while (bdrv_filter_bs(above_base) != base) {
253 above_base = bdrv_filter_bs(above_base);
254 }
255 }
256 }
257
258 /* Make sure that the image is opened in read-write mode */
259 bs_read_only = bdrv_is_read_only(bs);
260 if (bs_read_only) {
261 int ret;
262 /* Hold the chain during reopen */
263 if (bdrv_freeze_backing_chain(bs, above_base, errp) < 0) {
264 return;
265 }
266
267 ret = bdrv_reopen_set_read_only(bs, false, errp);
268
269 /* failure, or cor-filter will hold the chain */
270 bdrv_unfreeze_backing_chain(bs, above_base);
271
272 if (ret < 0) {
273 return;
274 }
275 }
276
277 opts = qdict_new();
278
279 qdict_put_str(opts, "driver", "copy-on-read");
280 qdict_put_str(opts, "file", bdrv_get_node_name(bs));
281 /* Pass the base_overlay node name as 'bottom' to COR driver */
282 qdict_put_str(opts, "bottom", base_overlay->node_name);
283 if (filter_node_name) {
284 qdict_put_str(opts, "node-name", filter_node_name);
285 }
286
287 cor_filter_bs = bdrv_insert_node(bs, opts, BDRV_O_RDWR, errp);
288 if (!cor_filter_bs) {
289 goto fail;
290 }
291
292 if (!filter_node_name) {
293 cor_filter_bs->implicit = true;
294 }
295
296 s = block_job_create(job_id, &stream_job_driver, NULL, cor_filter_bs,
297 BLK_PERM_CONSISTENT_READ,
298 basic_flags | BLK_PERM_WRITE,
299 speed, creation_flags, NULL, NULL, errp);
300 if (!s) {
301 goto fail;
302 }
303
304 /*
305 * Prevent concurrent jobs trying to modify the graph structure here, we
306 * already have our own plans. Also don't allow resize as the image size is
307 * queried only at the job start and then cached.
308 */
309 if (block_job_add_bdrv(&s->common, "active node", bs, 0,
310 basic_flags | BLK_PERM_WRITE, errp)) {
311 goto fail;
312 }
313
314 /* Block all intermediate nodes between bs and base, because they will
315 * disappear from the chain after this operation. The streaming job reads
316 * every block only once, assuming that it doesn't change, so forbid writes
317 * and resizes. Reassign the base node pointer because the backing BS of the
318 * bottom node might change after the call to bdrv_reopen_set_read_only()
319 * due to parallel block jobs running.
320 * above_base node might change after the call to
321 * bdrv_reopen_set_read_only() due to parallel block jobs running.
322 */
323 base = bdrv_filter_or_cow_bs(above_base);
324 for (iter = bdrv_filter_or_cow_bs(bs); iter != base;
325 iter = bdrv_filter_or_cow_bs(iter))
326 {
327 ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
328 basic_flags, errp);
329 if (ret < 0) {
330 goto fail;
331 }
332 }
333
334 s->base_overlay = base_overlay;
335 s->above_base = above_base;
336 s->backing_file_str = g_strdup(backing_file_str);
337 s->cor_filter_bs = cor_filter_bs;
338 s->target_bs = bs;
339 s->bs_read_only = bs_read_only;
340
341 s->on_error = on_error;
342 trace_stream_start(bs, base, s);
343 job_start(&s->common.job);
344 return;
345
346 fail:
347 if (s) {
348 job_early_fail(&s->common.job);
349 }
350 if (cor_filter_bs) {
351 bdrv_cor_filter_drop(cor_filter_bs);
352 }
353 if (bs_read_only) {
354 bdrv_reopen_set_read_only(bs, true, NULL);
355 }
356 }