]>
Commit | Line | Data |
---|---|---|
893f7eba PB |
1 | /* |
2 | * Image mirroring | |
3 | * | |
4 | * Copyright Red Hat, Inc. 2012 | |
5 | * | |
6 | * Authors: | |
7 | * Paolo Bonzini <pbonzini@redhat.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/blockjob.h" |
16 | #include "block/block_int.h" | |
373340b2 | 17 | #include "sysemu/block-backend.h" |
cc7a8ea7 | 18 | #include "qapi/qmp/qerror.h" |
893f7eba | 19 | #include "qemu/ratelimit.h" |
b812f671 | 20 | #include "qemu/bitmap.h" |
893f7eba | 21 | |
402a4741 PB |
22 | #define SLICE_TIME 100000000ULL /* ns */ |
23 | #define MAX_IN_FLIGHT 16 | |
48ac0a4d | 24 | #define DEFAULT_MIRROR_BUF_SIZE (10 << 20) |
402a4741 PB |
25 | |
26 | /* The mirroring buffer is a list of granularity-sized chunks. | |
27 | * Free chunks are organized in a list. | |
28 | */ | |
29 | typedef struct MirrorBuffer { | |
30 | QSIMPLEQ_ENTRY(MirrorBuffer) next; | |
31 | } MirrorBuffer; | |
893f7eba PB |
32 | |
33 | typedef struct MirrorBlockJob { | |
34 | BlockJob common; | |
35 | RateLimit limit; | |
36 | BlockDriverState *target; | |
5bc361b8 | 37 | BlockDriverState *base; |
09158f00 BC |
38 | /* The name of the graph node to replace */ |
39 | char *replaces; | |
40 | /* The BDS to replace */ | |
41 | BlockDriverState *to_replace; | |
42 | /* Used to block operations on the drive-mirror-replace target */ | |
43 | Error *replace_blocker; | |
03544a6e | 44 | bool is_none_mode; |
b952b558 | 45 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 PB |
46 | bool synced; |
47 | bool should_complete; | |
893f7eba | 48 | int64_t sector_num; |
eee13dfe | 49 | int64_t granularity; |
b812f671 | 50 | size_t buf_size; |
b21c7652 | 51 | int64_t bdev_length; |
b812f671 | 52 | unsigned long *cow_bitmap; |
e4654d2d | 53 | BdrvDirtyBitmap *dirty_bitmap; |
8f0720ec | 54 | HBitmapIter hbi; |
893f7eba | 55 | uint8_t *buf; |
402a4741 PB |
56 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
57 | int buf_free_count; | |
bd48bde8 | 58 | |
402a4741 | 59 | unsigned long *in_flight_bitmap; |
bd48bde8 | 60 | int in_flight; |
b21c7652 | 61 | int sectors_in_flight; |
bd48bde8 | 62 | int ret; |
0fc9f8ea | 63 | bool unmap; |
e424aff5 | 64 | bool waiting_for_io; |
893f7eba PB |
65 | } MirrorBlockJob; |
66 | ||
bd48bde8 PB |
67 | typedef struct MirrorOp { |
68 | MirrorBlockJob *s; | |
69 | QEMUIOVector qiov; | |
bd48bde8 PB |
70 | int64_t sector_num; |
71 | int nb_sectors; | |
72 | } MirrorOp; | |
73 | ||
b952b558 PB |
74 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
75 | int error) | |
76 | { | |
77 | s->synced = false; | |
78 | if (read) { | |
79 | return block_job_error_action(&s->common, s->common.bs, | |
80 | s->on_source_error, true, error); | |
81 | } else { | |
82 | return block_job_error_action(&s->common, s->target, | |
83 | s->on_target_error, false, error); | |
84 | } | |
85 | } | |
86 | ||
bd48bde8 PB |
87 | static void mirror_iteration_done(MirrorOp *op, int ret) |
88 | { | |
89 | MirrorBlockJob *s = op->s; | |
402a4741 | 90 | struct iovec *iov; |
bd48bde8 | 91 | int64_t chunk_num; |
402a4741 | 92 | int i, nb_chunks, sectors_per_chunk; |
bd48bde8 PB |
93 | |
94 | trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret); | |
95 | ||
96 | s->in_flight--; | |
b21c7652 | 97 | s->sectors_in_flight -= op->nb_sectors; |
402a4741 PB |
98 | iov = op->qiov.iov; |
99 | for (i = 0; i < op->qiov.niov; i++) { | |
100 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
101 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
102 | s->buf_free_count++; | |
103 | } | |
104 | ||
bd48bde8 PB |
105 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
106 | chunk_num = op->sector_num / sectors_per_chunk; | |
107 | nb_chunks = op->nb_sectors / sectors_per_chunk; | |
402a4741 | 108 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
b21c7652 HR |
109 | if (ret >= 0) { |
110 | if (s->cow_bitmap) { | |
111 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
112 | } | |
113 | s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
114 | } |
115 | ||
6df3bf8e | 116 | qemu_iovec_destroy(&op->qiov); |
c84b3192 | 117 | g_free(op); |
7b770c72 | 118 | |
e424aff5 | 119 | if (s->waiting_for_io) { |
7b770c72 SH |
120 | qemu_coroutine_enter(s->common.co, NULL); |
121 | } | |
bd48bde8 PB |
122 | } |
123 | ||
124 | static void mirror_write_complete(void *opaque, int ret) | |
125 | { | |
126 | MirrorOp *op = opaque; | |
127 | MirrorBlockJob *s = op->s; | |
128 | if (ret < 0) { | |
bd48bde8 PB |
129 | BlockErrorAction action; |
130 | ||
20dca810 | 131 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 132 | action = mirror_error_action(s, false, -ret); |
a589569f | 133 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
134 | s->ret = ret; |
135 | } | |
136 | } | |
137 | mirror_iteration_done(op, ret); | |
138 | } | |
139 | ||
140 | static void mirror_read_complete(void *opaque, int ret) | |
141 | { | |
142 | MirrorOp *op = opaque; | |
143 | MirrorBlockJob *s = op->s; | |
144 | if (ret < 0) { | |
bd48bde8 PB |
145 | BlockErrorAction action; |
146 | ||
20dca810 | 147 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 148 | action = mirror_error_action(s, true, -ret); |
a589569f | 149 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
150 | s->ret = ret; |
151 | } | |
152 | ||
153 | mirror_iteration_done(op, ret); | |
154 | return; | |
155 | } | |
156 | bdrv_aio_writev(s->target, op->sector_num, &op->qiov, op->nb_sectors, | |
157 | mirror_write_complete, op); | |
158 | } | |
159 | ||
cc8c9d6c | 160 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) |
893f7eba PB |
161 | { |
162 | BlockDriverState *source = s->common.bs; | |
402a4741 | 163 | int nb_sectors, sectors_per_chunk, nb_chunks; |
884fea4e | 164 | int64_t end, sector_num, next_chunk, next_sector, hbitmap_next_sector; |
6d0de8eb | 165 | uint64_t delay_ns = 0; |
bd48bde8 | 166 | MirrorOp *op; |
dcfb3beb FZ |
167 | int pnum; |
168 | int64_t ret; | |
893f7eba | 169 | |
8f0720ec PB |
170 | s->sector_num = hbitmap_iter_next(&s->hbi); |
171 | if (s->sector_num < 0) { | |
20dca810 | 172 | bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi); |
8f0720ec | 173 | s->sector_num = hbitmap_iter_next(&s->hbi); |
20dca810 | 174 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
8f0720ec PB |
175 | assert(s->sector_num >= 0); |
176 | } | |
177 | ||
402a4741 | 178 | hbitmap_next_sector = s->sector_num; |
884fea4e PB |
179 | sector_num = s->sector_num; |
180 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; | |
b21c7652 | 181 | end = s->bdev_length / BDRV_SECTOR_SIZE; |
402a4741 | 182 | |
884fea4e PB |
183 | /* Extend the QEMUIOVector to include all adjacent blocks that will |
184 | * be copied in this operation. | |
b812f671 | 185 | * |
884fea4e PB |
186 | * We have to do this if we have no backing file yet in the destination, |
187 | * and the cluster size is very large. Then we need to do COW ourselves. | |
188 | * The first time a cluster is copied, copy it entirely. Note that, | |
189 | * because both the granularity and the cluster size are powers of two, | |
190 | * the number of sectors to copy cannot exceed one cluster. | |
191 | * | |
192 | * We also want to extend the QEMUIOVector to include more adjacent | |
193 | * dirty blocks if possible, to limit the number of I/O operations and | |
194 | * run efficiently even with a small granularity. | |
b812f671 | 195 | */ |
884fea4e PB |
196 | nb_chunks = 0; |
197 | nb_sectors = 0; | |
198 | next_sector = sector_num; | |
199 | next_chunk = sector_num / sectors_per_chunk; | |
402a4741 PB |
200 | |
201 | /* Wait for I/O to this cluster (from a previous iteration) to be done. */ | |
884fea4e | 202 | while (test_bit(next_chunk, s->in_flight_bitmap)) { |
402a4741 | 203 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); |
e424aff5 | 204 | s->waiting_for_io = true; |
402a4741 | 205 | qemu_coroutine_yield(); |
e424aff5 | 206 | s->waiting_for_io = false; |
b812f671 PB |
207 | } |
208 | ||
884fea4e PB |
209 | do { |
210 | int added_sectors, added_chunks; | |
211 | ||
e4654d2d | 212 | if (!bdrv_get_dirty(source, s->dirty_bitmap, next_sector) || |
884fea4e PB |
213 | test_bit(next_chunk, s->in_flight_bitmap)) { |
214 | assert(nb_sectors > 0); | |
215 | break; | |
216 | } | |
217 | ||
218 | added_sectors = sectors_per_chunk; | |
219 | if (s->cow_bitmap && !test_bit(next_chunk, s->cow_bitmap)) { | |
220 | bdrv_round_to_clusters(s->target, | |
221 | next_sector, added_sectors, | |
222 | &next_sector, &added_sectors); | |
223 | ||
224 | /* On the first iteration, the rounding may make us copy | |
225 | * sectors before the first dirty one. | |
226 | */ | |
227 | if (next_sector < sector_num) { | |
228 | assert(nb_sectors == 0); | |
229 | sector_num = next_sector; | |
230 | next_chunk = next_sector / sectors_per_chunk; | |
231 | } | |
232 | } | |
233 | ||
234 | added_sectors = MIN(added_sectors, end - (sector_num + nb_sectors)); | |
235 | added_chunks = (added_sectors + sectors_per_chunk - 1) / sectors_per_chunk; | |
236 | ||
237 | /* When doing COW, it may happen that there is not enough space for | |
238 | * a full cluster. Wait if that is the case. | |
239 | */ | |
240 | while (nb_chunks == 0 && s->buf_free_count < added_chunks) { | |
241 | trace_mirror_yield_buf_busy(s, nb_chunks, s->in_flight); | |
e424aff5 | 242 | s->waiting_for_io = true; |
884fea4e | 243 | qemu_coroutine_yield(); |
e424aff5 | 244 | s->waiting_for_io = false; |
884fea4e PB |
245 | } |
246 | if (s->buf_free_count < nb_chunks + added_chunks) { | |
247 | trace_mirror_break_buf_busy(s, nb_chunks, s->in_flight); | |
248 | break; | |
249 | } | |
cae98cb8 SH |
250 | if (IOV_MAX < nb_chunks + added_chunks) { |
251 | trace_mirror_break_iov_max(s, nb_chunks, added_chunks); | |
252 | break; | |
253 | } | |
884fea4e PB |
254 | |
255 | /* We have enough free space to copy these sectors. */ | |
256 | bitmap_set(s->in_flight_bitmap, next_chunk, added_chunks); | |
402a4741 | 257 | |
884fea4e PB |
258 | nb_sectors += added_sectors; |
259 | nb_chunks += added_chunks; | |
260 | next_sector += added_sectors; | |
261 | next_chunk += added_chunks; | |
cc8c9d6c PB |
262 | if (!s->synced && s->common.speed) { |
263 | delay_ns = ratelimit_calculate_delay(&s->limit, added_sectors); | |
cc8c9d6c PB |
264 | } |
265 | } while (delay_ns == 0 && next_sector < end); | |
bd48bde8 PB |
266 | |
267 | /* Allocate a MirrorOp that is used as an AIO callback. */ | |
c84b3192 | 268 | op = g_new(MirrorOp, 1); |
bd48bde8 | 269 | op->s = s; |
bd48bde8 PB |
270 | op->sector_num = sector_num; |
271 | op->nb_sectors = nb_sectors; | |
402a4741 PB |
272 | |
273 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
274 | * from s->buf_free. | |
275 | */ | |
276 | qemu_iovec_init(&op->qiov, nb_chunks); | |
277 | next_sector = sector_num; | |
278 | while (nb_chunks-- > 0) { | |
279 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
5a0f6fd5 KW |
280 | size_t remaining = (nb_sectors * BDRV_SECTOR_SIZE) - op->qiov.size; |
281 | ||
402a4741 PB |
282 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
283 | s->buf_free_count--; | |
5a0f6fd5 | 284 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 PB |
285 | |
286 | /* Advance the HBitmapIter in parallel, so that we do not examine | |
287 | * the same sector twice. | |
288 | */ | |
e4654d2d FZ |
289 | if (next_sector > hbitmap_next_sector |
290 | && bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) { | |
402a4741 PB |
291 | hbitmap_next_sector = hbitmap_iter_next(&s->hbi); |
292 | } | |
293 | ||
294 | next_sector += sectors_per_chunk; | |
295 | } | |
bd48bde8 | 296 | |
20dca810 | 297 | bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, nb_sectors); |
893f7eba PB |
298 | |
299 | /* Copy the dirty cluster. */ | |
bd48bde8 | 300 | s->in_flight++; |
b21c7652 | 301 | s->sectors_in_flight += nb_sectors; |
b812f671 | 302 | trace_mirror_one_iteration(s, sector_num, nb_sectors); |
dcfb3beb FZ |
303 | |
304 | ret = bdrv_get_block_status_above(source, NULL, sector_num, | |
305 | nb_sectors, &pnum); | |
306 | if (ret < 0 || pnum < nb_sectors || | |
307 | (ret & BDRV_BLOCK_DATA && !(ret & BDRV_BLOCK_ZERO))) { | |
308 | bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors, | |
309 | mirror_read_complete, op); | |
310 | } else if (ret & BDRV_BLOCK_ZERO) { | |
311 | bdrv_aio_write_zeroes(s->target, sector_num, op->nb_sectors, | |
312 | s->unmap ? BDRV_REQ_MAY_UNMAP : 0, | |
313 | mirror_write_complete, op); | |
314 | } else { | |
315 | assert(!(ret & BDRV_BLOCK_DATA)); | |
316 | bdrv_aio_discard(s->target, sector_num, op->nb_sectors, | |
317 | mirror_write_complete, op); | |
318 | } | |
cc8c9d6c | 319 | return delay_ns; |
bd48bde8 | 320 | } |
b952b558 | 321 | |
402a4741 PB |
322 | static void mirror_free_init(MirrorBlockJob *s) |
323 | { | |
324 | int granularity = s->granularity; | |
325 | size_t buf_size = s->buf_size; | |
326 | uint8_t *buf = s->buf; | |
327 | ||
328 | assert(s->buf_free_count == 0); | |
329 | QSIMPLEQ_INIT(&s->buf_free); | |
330 | while (buf_size != 0) { | |
331 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
332 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
333 | s->buf_free_count++; | |
334 | buf_size -= granularity; | |
335 | buf += granularity; | |
336 | } | |
337 | } | |
338 | ||
bd48bde8 PB |
339 | static void mirror_drain(MirrorBlockJob *s) |
340 | { | |
341 | while (s->in_flight > 0) { | |
e424aff5 | 342 | s->waiting_for_io = true; |
bd48bde8 | 343 | qemu_coroutine_yield(); |
e424aff5 | 344 | s->waiting_for_io = false; |
bd48bde8 | 345 | } |
893f7eba PB |
346 | } |
347 | ||
5a7e7a0b SH |
348 | typedef struct { |
349 | int ret; | |
350 | } MirrorExitData; | |
351 | ||
352 | static void mirror_exit(BlockJob *job, void *opaque) | |
353 | { | |
354 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
355 | MirrorExitData *data = opaque; | |
356 | AioContext *replace_aio_context = NULL; | |
3f09bfbc KW |
357 | BlockDriverState *src = s->common.bs; |
358 | ||
359 | /* Make sure that the source BDS doesn't go away before we called | |
360 | * block_job_completed(). */ | |
361 | bdrv_ref(src); | |
5a7e7a0b SH |
362 | |
363 | if (s->to_replace) { | |
364 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
365 | aio_context_acquire(replace_aio_context); | |
366 | } | |
367 | ||
368 | if (s->should_complete && data->ret == 0) { | |
369 | BlockDriverState *to_replace = s->common.bs; | |
370 | if (s->to_replace) { | |
371 | to_replace = s->to_replace; | |
372 | } | |
373 | if (bdrv_get_flags(s->target) != bdrv_get_flags(to_replace)) { | |
374 | bdrv_reopen(s->target, bdrv_get_flags(to_replace), NULL); | |
375 | } | |
3f09bfbc | 376 | bdrv_replace_in_backing_chain(to_replace, s->target); |
5a7e7a0b SH |
377 | } |
378 | if (s->to_replace) { | |
379 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
380 | error_free(s->replace_blocker); | |
381 | bdrv_unref(s->to_replace); | |
382 | } | |
383 | if (replace_aio_context) { | |
384 | aio_context_release(replace_aio_context); | |
385 | } | |
386 | g_free(s->replaces); | |
10f3cd15 | 387 | bdrv_op_unblock_all(s->target, s->common.blocker); |
5a7e7a0b SH |
388 | bdrv_unref(s->target); |
389 | block_job_completed(&s->common, data->ret); | |
390 | g_free(data); | |
176c3699 | 391 | bdrv_drained_end(src); |
3f09bfbc | 392 | bdrv_unref(src); |
5a7e7a0b SH |
393 | } |
394 | ||
893f7eba PB |
395 | static void coroutine_fn mirror_run(void *opaque) |
396 | { | |
397 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 398 | MirrorExitData *data; |
893f7eba | 399 | BlockDriverState *bs = s->common.bs; |
99900697 | 400 | int64_t sector_num, end, length; |
bd48bde8 | 401 | uint64_t last_pause_ns; |
b812f671 | 402 | BlockDriverInfo bdi; |
1d33936e JC |
403 | char backing_filename[2]; /* we only need 2 characters because we are only |
404 | checking for a NULL string */ | |
893f7eba PB |
405 | int ret = 0; |
406 | int n; | |
893f7eba PB |
407 | |
408 | if (block_job_is_cancelled(&s->common)) { | |
409 | goto immediate_exit; | |
410 | } | |
411 | ||
b21c7652 HR |
412 | s->bdev_length = bdrv_getlength(bs); |
413 | if (s->bdev_length < 0) { | |
414 | ret = s->bdev_length; | |
373df5b1 | 415 | goto immediate_exit; |
b21c7652 | 416 | } else if (s->bdev_length == 0) { |
9e48b025 FZ |
417 | /* Report BLOCK_JOB_READY and wait for complete. */ |
418 | block_job_event_ready(&s->common); | |
419 | s->synced = true; | |
420 | while (!block_job_is_cancelled(&s->common) && !s->should_complete) { | |
421 | block_job_yield(&s->common); | |
422 | } | |
423 | s->common.cancelled = false; | |
424 | goto immediate_exit; | |
893f7eba PB |
425 | } |
426 | ||
b21c7652 | 427 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
428 | s->in_flight_bitmap = bitmap_new(length); |
429 | ||
b812f671 PB |
430 | /* If we have no backing file yet in the destination, we cannot let |
431 | * the destination do COW. Instead, we copy sectors around the | |
432 | * dirty data if needed. We need a bitmap to do that. | |
433 | */ | |
434 | bdrv_get_backing_filename(s->target, backing_filename, | |
435 | sizeof(backing_filename)); | |
760e0063 | 436 | if (backing_filename[0] && !s->target->backing) { |
c3cc95bd FZ |
437 | ret = bdrv_get_info(s->target, &bdi); |
438 | if (ret < 0) { | |
439 | goto immediate_exit; | |
440 | } | |
eee13dfe | 441 | if (s->granularity < bdi.cluster_size) { |
08e4ed6c | 442 | s->buf_size = MAX(s->buf_size, bdi.cluster_size); |
b812f671 PB |
443 | s->cow_bitmap = bitmap_new(length); |
444 | } | |
445 | } | |
446 | ||
b21c7652 | 447 | end = s->bdev_length / BDRV_SECTOR_SIZE; |
7504edf4 KW |
448 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
449 | if (s->buf == NULL) { | |
450 | ret = -ENOMEM; | |
451 | goto immediate_exit; | |
452 | } | |
453 | ||
402a4741 | 454 | mirror_free_init(s); |
893f7eba | 455 | |
4c0cbd6f | 456 | last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 457 | if (!s->is_none_mode) { |
893f7eba | 458 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
5bc361b8 | 459 | BlockDriverState *base = s->base; |
5279efeb JC |
460 | bool mark_all_dirty = s->base == NULL && !bdrv_has_zero_init(s->target); |
461 | ||
893f7eba | 462 | for (sector_num = 0; sector_num < end; ) { |
99900697 FZ |
463 | /* Just to make sure we are not exceeding int limit. */ |
464 | int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, | |
465 | end - sector_num); | |
4c0cbd6f FZ |
466 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
467 | ||
468 | if (now - last_pause_ns > SLICE_TIME) { | |
469 | last_pause_ns = now; | |
470 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); | |
471 | } | |
472 | ||
473 | if (block_job_is_cancelled(&s->common)) { | |
474 | goto immediate_exit; | |
475 | } | |
476 | ||
99900697 | 477 | ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); |
893f7eba PB |
478 | |
479 | if (ret < 0) { | |
480 | goto immediate_exit; | |
481 | } | |
482 | ||
483 | assert(n > 0); | |
5279efeb | 484 | if (ret == 1 || mark_all_dirty) { |
20dca810 | 485 | bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); |
893f7eba | 486 | } |
99900697 | 487 | sector_num += n; |
893f7eba PB |
488 | } |
489 | } | |
490 | ||
20dca810 | 491 | bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi); |
893f7eba | 492 | for (;;) { |
cc8c9d6c | 493 | uint64_t delay_ns = 0; |
893f7eba PB |
494 | int64_t cnt; |
495 | bool should_complete; | |
496 | ||
bd48bde8 PB |
497 | if (s->ret < 0) { |
498 | ret = s->ret; | |
499 | goto immediate_exit; | |
500 | } | |
501 | ||
20dca810 | 502 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
b21c7652 HR |
503 | /* s->common.offset contains the number of bytes already processed so |
504 | * far, cnt is the number of dirty sectors remaining and | |
505 | * s->sectors_in_flight is the number of sectors currently being | |
506 | * processed; together those are the current total operation length */ | |
507 | s->common.len = s->common.offset + | |
508 | (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
509 | |
510 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 511 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
bd48bde8 PB |
512 | * We do so every SLICE_TIME nanoseconds, or when there is an error, |
513 | * or when the source is clean, whichever comes first. | |
514 | */ | |
bc72ad67 | 515 | if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME && |
bd48bde8 | 516 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
402a4741 PB |
517 | if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 || |
518 | (cnt == 0 && s->in_flight > 0)) { | |
519 | trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt); | |
e424aff5 | 520 | s->waiting_for_io = true; |
bd48bde8 | 521 | qemu_coroutine_yield(); |
e424aff5 | 522 | s->waiting_for_io = false; |
bd48bde8 PB |
523 | continue; |
524 | } else if (cnt != 0) { | |
cc8c9d6c | 525 | delay_ns = mirror_iteration(s); |
893f7eba | 526 | } |
893f7eba PB |
527 | } |
528 | ||
529 | should_complete = false; | |
bd48bde8 | 530 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba PB |
531 | trace_mirror_before_flush(s); |
532 | ret = bdrv_flush(s->target); | |
533 | if (ret < 0) { | |
a589569f WX |
534 | if (mirror_error_action(s, false, -ret) == |
535 | BLOCK_ERROR_ACTION_REPORT) { | |
b952b558 PB |
536 | goto immediate_exit; |
537 | } | |
538 | } else { | |
539 | /* We're out of the streaming phase. From now on, if the job | |
540 | * is cancelled we will actually complete all pending I/O and | |
541 | * report completion. This way, block-job-cancel will leave | |
542 | * the target in a consistent state. | |
543 | */ | |
b952b558 | 544 | if (!s->synced) { |
bcada37b | 545 | block_job_event_ready(&s->common); |
b952b558 PB |
546 | s->synced = true; |
547 | } | |
548 | ||
549 | should_complete = s->should_complete || | |
550 | block_job_is_cancelled(&s->common); | |
20dca810 | 551 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
d63ffd87 | 552 | } |
893f7eba PB |
553 | } |
554 | ||
555 | if (cnt == 0 && should_complete) { | |
556 | /* The dirty bitmap is not updated while operations are pending. | |
557 | * If we're about to exit, wait for pending operations before | |
558 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
559 | * source has dirty data to copy! | |
560 | * | |
561 | * Note that I/O can be submitted by the guest while | |
562 | * mirror_populate runs. | |
563 | */ | |
564 | trace_mirror_before_drain(s, cnt); | |
5a7e7a0b | 565 | bdrv_drain(bs); |
20dca810 | 566 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
893f7eba PB |
567 | } |
568 | ||
569 | ret = 0; | |
cc8c9d6c | 570 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
d63ffd87 | 571 | if (!s->synced) { |
7483d1e5 | 572 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
573 | if (block_job_is_cancelled(&s->common)) { |
574 | break; | |
575 | } | |
576 | } else if (!should_complete) { | |
bd48bde8 | 577 | delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); |
7483d1e5 | 578 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
579 | } else if (cnt == 0) { |
580 | /* The two disks are in sync. Exit and report successful | |
581 | * completion. | |
582 | */ | |
583 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
584 | s->common.cancelled = false; | |
585 | break; | |
586 | } | |
bc72ad67 | 587 | last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
588 | } |
589 | ||
590 | immediate_exit: | |
bd48bde8 PB |
591 | if (s->in_flight > 0) { |
592 | /* We get here only if something went wrong. Either the job failed, | |
593 | * or it was cancelled prematurely so that we do not guarantee that | |
594 | * the target is a copy of the source. | |
595 | */ | |
596 | assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); | |
597 | mirror_drain(s); | |
598 | } | |
599 | ||
600 | assert(s->in_flight == 0); | |
7191bf31 | 601 | qemu_vfree(s->buf); |
b812f671 | 602 | g_free(s->cow_bitmap); |
402a4741 | 603 | g_free(s->in_flight_bitmap); |
e4654d2d | 604 | bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); |
373340b2 HR |
605 | if (s->target->blk) { |
606 | blk_iostatus_disable(s->target->blk); | |
607 | } | |
5a7e7a0b SH |
608 | |
609 | data = g_malloc(sizeof(*data)); | |
610 | data->ret = ret; | |
176c3699 FZ |
611 | /* Before we switch to target in mirror_exit, make sure data doesn't |
612 | * change. */ | |
613 | bdrv_drained_begin(s->common.bs); | |
5a7e7a0b | 614 | block_job_defer_to_main_loop(&s->common, mirror_exit, data); |
893f7eba PB |
615 | } |
616 | ||
617 | static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
618 | { | |
619 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
620 | ||
621 | if (speed < 0) { | |
c6bd8c70 | 622 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
893f7eba PB |
623 | return; |
624 | } | |
625 | ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); | |
626 | } | |
627 | ||
b952b558 PB |
628 | static void mirror_iostatus_reset(BlockJob *job) |
629 | { | |
630 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
631 | ||
373340b2 HR |
632 | if (s->target->blk) { |
633 | blk_iostatus_reset(s->target->blk); | |
634 | } | |
b952b558 PB |
635 | } |
636 | ||
d63ffd87 PB |
637 | static void mirror_complete(BlockJob *job, Error **errp) |
638 | { | |
639 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
34b5d2c6 | 640 | Error *local_err = NULL; |
d63ffd87 PB |
641 | int ret; |
642 | ||
34b5d2c6 | 643 | ret = bdrv_open_backing_file(s->target, NULL, &local_err); |
d63ffd87 | 644 | if (ret < 0) { |
34b5d2c6 | 645 | error_propagate(errp, local_err); |
d63ffd87 PB |
646 | return; |
647 | } | |
648 | if (!s->synced) { | |
8ccb9569 | 649 | error_setg(errp, QERR_BLOCK_JOB_NOT_READY, job->id); |
d63ffd87 PB |
650 | return; |
651 | } | |
652 | ||
09158f00 BC |
653 | /* check the target bs is not blocked and block all operations on it */ |
654 | if (s->replaces) { | |
5a7e7a0b SH |
655 | AioContext *replace_aio_context; |
656 | ||
e12f3784 | 657 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 658 | if (!s->to_replace) { |
e12f3784 | 659 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
660 | return; |
661 | } | |
662 | ||
5a7e7a0b SH |
663 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
664 | aio_context_acquire(replace_aio_context); | |
665 | ||
09158f00 BC |
666 | error_setg(&s->replace_blocker, |
667 | "block device is in use by block-job-complete"); | |
668 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
669 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
670 | |
671 | aio_context_release(replace_aio_context); | |
09158f00 BC |
672 | } |
673 | ||
d63ffd87 | 674 | s->should_complete = true; |
751ebd76 | 675 | block_job_enter(&s->common); |
d63ffd87 PB |
676 | } |
677 | ||
3fc4b10a | 678 | static const BlockJobDriver mirror_job_driver = { |
893f7eba | 679 | .instance_size = sizeof(MirrorBlockJob), |
79e14bf7 | 680 | .job_type = BLOCK_JOB_TYPE_MIRROR, |
893f7eba | 681 | .set_speed = mirror_set_speed, |
b952b558 | 682 | .iostatus_reset= mirror_iostatus_reset, |
d63ffd87 | 683 | .complete = mirror_complete, |
893f7eba PB |
684 | }; |
685 | ||
03544a6e FZ |
686 | static const BlockJobDriver commit_active_job_driver = { |
687 | .instance_size = sizeof(MirrorBlockJob), | |
688 | .job_type = BLOCK_JOB_TYPE_COMMIT, | |
689 | .set_speed = mirror_set_speed, | |
690 | .iostatus_reset | |
691 | = mirror_iostatus_reset, | |
692 | .complete = mirror_complete, | |
693 | }; | |
694 | ||
695 | static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target, | |
09158f00 | 696 | const char *replaces, |
5fba6c0e | 697 | int64_t speed, uint32_t granularity, |
09158f00 BC |
698 | int64_t buf_size, |
699 | BlockdevOnError on_source_error, | |
700 | BlockdevOnError on_target_error, | |
0fc9f8ea | 701 | bool unmap, |
097310b5 | 702 | BlockCompletionFunc *cb, |
09158f00 BC |
703 | void *opaque, Error **errp, |
704 | const BlockJobDriver *driver, | |
705 | bool is_none_mode, BlockDriverState *base) | |
893f7eba PB |
706 | { |
707 | MirrorBlockJob *s; | |
708 | ||
eee13dfe | 709 | if (granularity == 0) { |
341ebc2f | 710 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
711 | } |
712 | ||
713 | assert ((granularity & (granularity - 1)) == 0); | |
714 | ||
b952b558 PB |
715 | if ((on_source_error == BLOCKDEV_ON_ERROR_STOP || |
716 | on_source_error == BLOCKDEV_ON_ERROR_ENOSPC) && | |
373340b2 | 717 | (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) { |
c6bd8c70 | 718 | error_setg(errp, QERR_INVALID_PARAMETER, "on-source-error"); |
b952b558 PB |
719 | return; |
720 | } | |
721 | ||
48ac0a4d WC |
722 | if (buf_size < 0) { |
723 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
724 | return; | |
725 | } | |
726 | ||
727 | if (buf_size == 0) { | |
728 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
729 | } | |
5bc361b8 | 730 | |
03544a6e | 731 | s = block_job_create(driver, bs, speed, cb, opaque, errp); |
893f7eba PB |
732 | if (!s) { |
733 | return; | |
734 | } | |
735 | ||
09158f00 | 736 | s->replaces = g_strdup(replaces); |
b952b558 PB |
737 | s->on_source_error = on_source_error; |
738 | s->on_target_error = on_target_error; | |
893f7eba | 739 | s->target = target; |
03544a6e | 740 | s->is_none_mode = is_none_mode; |
5bc361b8 | 741 | s->base = base; |
eee13dfe | 742 | s->granularity = granularity; |
48ac0a4d | 743 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 744 | s->unmap = unmap; |
b812f671 | 745 | |
0db6e54a | 746 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 747 | if (!s->dirty_bitmap) { |
97031164 | 748 | g_free(s->replaces); |
18930ba3 | 749 | block_job_unref(&s->common); |
b8afb520 FZ |
750 | return; |
751 | } | |
10f3cd15 AG |
752 | |
753 | bdrv_op_block_all(s->target, s->common.blocker); | |
754 | ||
893f7eba | 755 | bdrv_set_enable_write_cache(s->target, true); |
373340b2 HR |
756 | if (s->target->blk) { |
757 | blk_set_on_error(s->target->blk, on_target_error, on_target_error); | |
758 | blk_iostatus_enable(s->target->blk); | |
759 | } | |
893f7eba PB |
760 | s->common.co = qemu_coroutine_create(mirror_run); |
761 | trace_mirror_start(bs, s, s->common.co, opaque); | |
762 | qemu_coroutine_enter(s->common.co, s); | |
763 | } | |
03544a6e FZ |
764 | |
765 | void mirror_start(BlockDriverState *bs, BlockDriverState *target, | |
09158f00 | 766 | const char *replaces, |
5fba6c0e | 767 | int64_t speed, uint32_t granularity, int64_t buf_size, |
03544a6e FZ |
768 | MirrorSyncMode mode, BlockdevOnError on_source_error, |
769 | BlockdevOnError on_target_error, | |
0fc9f8ea | 770 | bool unmap, |
097310b5 | 771 | BlockCompletionFunc *cb, |
03544a6e FZ |
772 | void *opaque, Error **errp) |
773 | { | |
774 | bool is_none_mode; | |
775 | BlockDriverState *base; | |
776 | ||
4b80ab2b JS |
777 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
778 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
779 | return; |
780 | } | |
03544a6e | 781 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 782 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
09158f00 BC |
783 | mirror_start_job(bs, target, replaces, |
784 | speed, granularity, buf_size, | |
0fc9f8ea | 785 | on_source_error, on_target_error, unmap, cb, opaque, errp, |
03544a6e FZ |
786 | &mirror_job_driver, is_none_mode, base); |
787 | } | |
788 | ||
789 | void commit_active_start(BlockDriverState *bs, BlockDriverState *base, | |
790 | int64_t speed, | |
791 | BlockdevOnError on_error, | |
097310b5 | 792 | BlockCompletionFunc *cb, |
03544a6e FZ |
793 | void *opaque, Error **errp) |
794 | { | |
4da83585 JC |
795 | int64_t length, base_length; |
796 | int orig_base_flags; | |
39a611a3 | 797 | int ret; |
cc67f4d1 | 798 | Error *local_err = NULL; |
4da83585 JC |
799 | |
800 | orig_base_flags = bdrv_get_flags(base); | |
801 | ||
20a63d2c FZ |
802 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
803 | return; | |
804 | } | |
4da83585 JC |
805 | |
806 | length = bdrv_getlength(bs); | |
807 | if (length < 0) { | |
39a611a3 JC |
808 | error_setg_errno(errp, -length, |
809 | "Unable to determine length of %s", bs->filename); | |
4da83585 JC |
810 | goto error_restore_flags; |
811 | } | |
812 | ||
813 | base_length = bdrv_getlength(base); | |
814 | if (base_length < 0) { | |
39a611a3 JC |
815 | error_setg_errno(errp, -base_length, |
816 | "Unable to determine length of %s", base->filename); | |
4da83585 JC |
817 | goto error_restore_flags; |
818 | } | |
819 | ||
820 | if (length > base_length) { | |
39a611a3 JC |
821 | ret = bdrv_truncate(base, length); |
822 | if (ret < 0) { | |
823 | error_setg_errno(errp, -ret, | |
824 | "Top image %s is larger than base image %s, and " | |
4da83585 JC |
825 | "resize of base image failed", |
826 | bs->filename, base->filename); | |
827 | goto error_restore_flags; | |
828 | } | |
829 | } | |
830 | ||
20a63d2c | 831 | bdrv_ref(base); |
09158f00 | 832 | mirror_start_job(bs, base, NULL, speed, 0, 0, |
0fc9f8ea | 833 | on_error, on_error, false, cb, opaque, &local_err, |
03544a6e | 834 | &commit_active_job_driver, false, base); |
0fb6395c | 835 | if (local_err) { |
cc67f4d1 | 836 | error_propagate(errp, local_err); |
4da83585 JC |
837 | goto error_restore_flags; |
838 | } | |
839 | ||
840 | return; | |
841 | ||
842 | error_restore_flags: | |
843 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
844 | * the original error */ | |
845 | bdrv_reopen(base, orig_base_flags, NULL); | |
846 | return; | |
03544a6e | 847 | } |