]>
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 | ||
80c71a24 | 14 | #include "qemu/osdep.h" |
893f7eba | 15 | #include "trace.h" |
737e150e PB |
16 | #include "block/blockjob.h" |
17 | #include "block/block_int.h" | |
373340b2 | 18 | #include "sysemu/block-backend.h" |
da34e65c | 19 | #include "qapi/error.h" |
cc7a8ea7 | 20 | #include "qapi/qmp/qerror.h" |
893f7eba | 21 | #include "qemu/ratelimit.h" |
b812f671 | 22 | #include "qemu/bitmap.h" |
893f7eba | 23 | |
402a4741 PB |
24 | #define SLICE_TIME 100000000ULL /* ns */ |
25 | #define MAX_IN_FLIGHT 16 | |
48ac0a4d | 26 | #define DEFAULT_MIRROR_BUF_SIZE (10 << 20) |
402a4741 PB |
27 | |
28 | /* The mirroring buffer is a list of granularity-sized chunks. | |
29 | * Free chunks are organized in a list. | |
30 | */ | |
31 | typedef struct MirrorBuffer { | |
32 | QSIMPLEQ_ENTRY(MirrorBuffer) next; | |
33 | } MirrorBuffer; | |
893f7eba PB |
34 | |
35 | typedef struct MirrorBlockJob { | |
36 | BlockJob common; | |
37 | RateLimit limit; | |
e253f4b8 | 38 | BlockBackend *target; |
5bc361b8 | 39 | BlockDriverState *base; |
09158f00 BC |
40 | /* The name of the graph node to replace */ |
41 | char *replaces; | |
42 | /* The BDS to replace */ | |
43 | BlockDriverState *to_replace; | |
44 | /* Used to block operations on the drive-mirror-replace target */ | |
45 | Error *replace_blocker; | |
03544a6e | 46 | bool is_none_mode; |
274fccee | 47 | BlockMirrorBackingMode backing_mode; |
b952b558 | 48 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 PB |
49 | bool synced; |
50 | bool should_complete; | |
eee13dfe | 51 | int64_t granularity; |
b812f671 | 52 | size_t buf_size; |
b21c7652 | 53 | int64_t bdev_length; |
b812f671 | 54 | unsigned long *cow_bitmap; |
e4654d2d | 55 | BdrvDirtyBitmap *dirty_bitmap; |
8f0720ec | 56 | HBitmapIter hbi; |
893f7eba | 57 | uint8_t *buf; |
402a4741 PB |
58 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
59 | int buf_free_count; | |
bd48bde8 | 60 | |
402a4741 | 61 | unsigned long *in_flight_bitmap; |
bd48bde8 | 62 | int in_flight; |
b21c7652 | 63 | int sectors_in_flight; |
bd48bde8 | 64 | int ret; |
0fc9f8ea | 65 | bool unmap; |
e424aff5 | 66 | bool waiting_for_io; |
e5b43573 FZ |
67 | int target_cluster_sectors; |
68 | int max_iov; | |
893f7eba PB |
69 | } MirrorBlockJob; |
70 | ||
bd48bde8 PB |
71 | typedef struct MirrorOp { |
72 | MirrorBlockJob *s; | |
73 | QEMUIOVector qiov; | |
bd48bde8 PB |
74 | int64_t sector_num; |
75 | int nb_sectors; | |
76 | } MirrorOp; | |
77 | ||
b952b558 PB |
78 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
79 | int error) | |
80 | { | |
81 | s->synced = false; | |
82 | if (read) { | |
81e254dc KW |
83 | return block_job_error_action(&s->common, s->on_source_error, |
84 | true, error); | |
b952b558 | 85 | } else { |
81e254dc KW |
86 | return block_job_error_action(&s->common, s->on_target_error, |
87 | false, error); | |
b952b558 PB |
88 | } |
89 | } | |
90 | ||
bd48bde8 PB |
91 | static void mirror_iteration_done(MirrorOp *op, int ret) |
92 | { | |
93 | MirrorBlockJob *s = op->s; | |
402a4741 | 94 | struct iovec *iov; |
bd48bde8 | 95 | int64_t chunk_num; |
402a4741 | 96 | int i, nb_chunks, sectors_per_chunk; |
bd48bde8 PB |
97 | |
98 | trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret); | |
99 | ||
100 | s->in_flight--; | |
b21c7652 | 101 | s->sectors_in_flight -= op->nb_sectors; |
402a4741 PB |
102 | iov = op->qiov.iov; |
103 | for (i = 0; i < op->qiov.niov; i++) { | |
104 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
105 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
106 | s->buf_free_count++; | |
107 | } | |
108 | ||
bd48bde8 PB |
109 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
110 | chunk_num = op->sector_num / sectors_per_chunk; | |
4150ae60 | 111 | nb_chunks = DIV_ROUND_UP(op->nb_sectors, sectors_per_chunk); |
402a4741 | 112 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
b21c7652 HR |
113 | if (ret >= 0) { |
114 | if (s->cow_bitmap) { | |
115 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
116 | } | |
117 | s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
118 | } |
119 | ||
6df3bf8e | 120 | qemu_iovec_destroy(&op->qiov); |
c84b3192 | 121 | g_free(op); |
7b770c72 | 122 | |
e424aff5 | 123 | if (s->waiting_for_io) { |
7b770c72 SH |
124 | qemu_coroutine_enter(s->common.co, NULL); |
125 | } | |
bd48bde8 PB |
126 | } |
127 | ||
128 | static void mirror_write_complete(void *opaque, int ret) | |
129 | { | |
130 | MirrorOp *op = opaque; | |
131 | MirrorBlockJob *s = op->s; | |
132 | if (ret < 0) { | |
bd48bde8 PB |
133 | BlockErrorAction action; |
134 | ||
20dca810 | 135 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 136 | action = mirror_error_action(s, false, -ret); |
a589569f | 137 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
138 | s->ret = ret; |
139 | } | |
140 | } | |
141 | mirror_iteration_done(op, ret); | |
142 | } | |
143 | ||
144 | static void mirror_read_complete(void *opaque, int ret) | |
145 | { | |
146 | MirrorOp *op = opaque; | |
147 | MirrorBlockJob *s = op->s; | |
148 | if (ret < 0) { | |
bd48bde8 PB |
149 | BlockErrorAction action; |
150 | ||
20dca810 | 151 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 152 | action = mirror_error_action(s, true, -ret); |
a589569f | 153 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
154 | s->ret = ret; |
155 | } | |
156 | ||
157 | mirror_iteration_done(op, ret); | |
158 | return; | |
159 | } | |
e253f4b8 | 160 | blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov, |
73698c30 | 161 | 0, mirror_write_complete, op); |
bd48bde8 PB |
162 | } |
163 | ||
4150ae60 FZ |
164 | static inline void mirror_clip_sectors(MirrorBlockJob *s, |
165 | int64_t sector_num, | |
166 | int *nb_sectors) | |
167 | { | |
168 | *nb_sectors = MIN(*nb_sectors, | |
169 | s->bdev_length / BDRV_SECTOR_SIZE - sector_num); | |
170 | } | |
171 | ||
e5b43573 FZ |
172 | /* Round sector_num and/or nb_sectors to target cluster if COW is needed, and |
173 | * return the offset of the adjusted tail sector against original. */ | |
174 | static int mirror_cow_align(MirrorBlockJob *s, | |
175 | int64_t *sector_num, | |
176 | int *nb_sectors) | |
893f7eba | 177 | { |
e5b43573 FZ |
178 | bool need_cow; |
179 | int ret = 0; | |
180 | int chunk_sectors = s->granularity >> BDRV_SECTOR_BITS; | |
181 | int64_t align_sector_num = *sector_num; | |
182 | int align_nb_sectors = *nb_sectors; | |
183 | int max_sectors = chunk_sectors * s->max_iov; | |
184 | ||
185 | need_cow = !test_bit(*sector_num / chunk_sectors, s->cow_bitmap); | |
186 | need_cow |= !test_bit((*sector_num + *nb_sectors - 1) / chunk_sectors, | |
187 | s->cow_bitmap); | |
188 | if (need_cow) { | |
244483e6 KW |
189 | bdrv_round_sectors_to_clusters(blk_bs(s->target), *sector_num, |
190 | *nb_sectors, &align_sector_num, | |
191 | &align_nb_sectors); | |
e5b43573 | 192 | } |
3515727f | 193 | |
e5b43573 FZ |
194 | if (align_nb_sectors > max_sectors) { |
195 | align_nb_sectors = max_sectors; | |
196 | if (need_cow) { | |
197 | align_nb_sectors = QEMU_ALIGN_DOWN(align_nb_sectors, | |
198 | s->target_cluster_sectors); | |
199 | } | |
8f0720ec | 200 | } |
4150ae60 FZ |
201 | /* Clipping may result in align_nb_sectors unaligned to chunk boundary, but |
202 | * that doesn't matter because it's already the end of source image. */ | |
203 | mirror_clip_sectors(s, align_sector_num, &align_nb_sectors); | |
8f0720ec | 204 | |
e5b43573 FZ |
205 | ret = align_sector_num + align_nb_sectors - (*sector_num + *nb_sectors); |
206 | *sector_num = align_sector_num; | |
207 | *nb_sectors = align_nb_sectors; | |
208 | assert(ret >= 0); | |
209 | return ret; | |
210 | } | |
211 | ||
21cd917f FZ |
212 | static inline void mirror_wait_for_io(MirrorBlockJob *s) |
213 | { | |
214 | assert(!s->waiting_for_io); | |
215 | s->waiting_for_io = true; | |
216 | qemu_coroutine_yield(); | |
217 | s->waiting_for_io = false; | |
218 | } | |
219 | ||
e5b43573 FZ |
220 | /* Submit async read while handling COW. |
221 | * Returns: nb_sectors if no alignment is necessary, or | |
222 | * (new_end - sector_num) if tail is rounded up or down due to | |
223 | * alignment or buffer limit. | |
224 | */ | |
225 | static int mirror_do_read(MirrorBlockJob *s, int64_t sector_num, | |
226 | int nb_sectors) | |
227 | { | |
e253f4b8 | 228 | BlockBackend *source = s->common.blk; |
e5b43573 FZ |
229 | int sectors_per_chunk, nb_chunks; |
230 | int ret = nb_sectors; | |
231 | MirrorOp *op; | |
232 | ||
884fea4e | 233 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
402a4741 | 234 | |
e5b43573 FZ |
235 | /* We can only handle as much as buf_size at a time. */ |
236 | nb_sectors = MIN(s->buf_size >> BDRV_SECTOR_BITS, nb_sectors); | |
237 | assert(nb_sectors); | |
402a4741 | 238 | |
e5b43573 FZ |
239 | if (s->cow_bitmap) { |
240 | ret += mirror_cow_align(s, §or_num, &nb_sectors); | |
241 | } | |
242 | assert(nb_sectors << BDRV_SECTOR_BITS <= s->buf_size); | |
243 | /* The sector range must meet granularity because: | |
244 | * 1) Caller passes in aligned values; | |
245 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
e5b43573 | 246 | assert(!(sector_num % sectors_per_chunk)); |
4150ae60 | 247 | nb_chunks = DIV_ROUND_UP(nb_sectors, sectors_per_chunk); |
e5b43573 FZ |
248 | |
249 | while (s->buf_free_count < nb_chunks) { | |
402a4741 | 250 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); |
21cd917f | 251 | mirror_wait_for_io(s); |
b812f671 PB |
252 | } |
253 | ||
bd48bde8 | 254 | /* Allocate a MirrorOp that is used as an AIO callback. */ |
c84b3192 | 255 | op = g_new(MirrorOp, 1); |
bd48bde8 | 256 | op->s = s; |
bd48bde8 PB |
257 | op->sector_num = sector_num; |
258 | op->nb_sectors = nb_sectors; | |
402a4741 PB |
259 | |
260 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
261 | * from s->buf_free. | |
262 | */ | |
263 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
264 | while (nb_chunks-- > 0) { |
265 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
e5b43573 | 266 | size_t remaining = nb_sectors * BDRV_SECTOR_SIZE - op->qiov.size; |
5a0f6fd5 | 267 | |
402a4741 PB |
268 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
269 | s->buf_free_count--; | |
5a0f6fd5 | 270 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 271 | } |
bd48bde8 | 272 | |
893f7eba | 273 | /* Copy the dirty cluster. */ |
bd48bde8 | 274 | s->in_flight++; |
b21c7652 | 275 | s->sectors_in_flight += nb_sectors; |
b812f671 | 276 | trace_mirror_one_iteration(s, sector_num, nb_sectors); |
dcfb3beb | 277 | |
73698c30 | 278 | blk_aio_preadv(source, sector_num * BDRV_SECTOR_SIZE, &op->qiov, 0, |
e5b43573 FZ |
279 | mirror_read_complete, op); |
280 | return ret; | |
281 | } | |
282 | ||
283 | static void mirror_do_zero_or_discard(MirrorBlockJob *s, | |
284 | int64_t sector_num, | |
285 | int nb_sectors, | |
286 | bool is_discard) | |
287 | { | |
288 | MirrorOp *op; | |
289 | ||
290 | /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed | |
291 | * so the freeing in mirror_iteration_done is nop. */ | |
292 | op = g_new0(MirrorOp, 1); | |
293 | op->s = s; | |
294 | op->sector_num = sector_num; | |
295 | op->nb_sectors = nb_sectors; | |
296 | ||
297 | s->in_flight++; | |
298 | s->sectors_in_flight += nb_sectors; | |
299 | if (is_discard) { | |
e253f4b8 KW |
300 | blk_aio_discard(s->target, sector_num, op->nb_sectors, |
301 | mirror_write_complete, op); | |
e5b43573 | 302 | } else { |
e253f4b8 KW |
303 | blk_aio_pwrite_zeroes(s->target, sector_num * BDRV_SECTOR_SIZE, |
304 | op->nb_sectors * BDRV_SECTOR_SIZE, | |
dcfb3beb FZ |
305 | s->unmap ? BDRV_REQ_MAY_UNMAP : 0, |
306 | mirror_write_complete, op); | |
e5b43573 FZ |
307 | } |
308 | } | |
309 | ||
310 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) | |
311 | { | |
e253f4b8 | 312 | BlockDriverState *source = blk_bs(s->common.blk); |
9c83625b | 313 | int64_t sector_num, first_chunk; |
e5b43573 FZ |
314 | uint64_t delay_ns = 0; |
315 | /* At least the first dirty chunk is mirrored in one iteration. */ | |
316 | int nb_chunks = 1; | |
317 | int64_t end = s->bdev_length / BDRV_SECTOR_SIZE; | |
318 | int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; | |
319 | ||
320 | sector_num = hbitmap_iter_next(&s->hbi); | |
321 | if (sector_num < 0) { | |
322 | bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi); | |
323 | sector_num = hbitmap_iter_next(&s->hbi); | |
324 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); | |
325 | assert(sector_num >= 0); | |
326 | } | |
327 | ||
9c83625b HR |
328 | first_chunk = sector_num / sectors_per_chunk; |
329 | while (test_bit(first_chunk, s->in_flight_bitmap)) { | |
330 | trace_mirror_yield_in_flight(s, first_chunk, s->in_flight); | |
331 | mirror_wait_for_io(s); | |
332 | } | |
333 | ||
565ac01f SH |
334 | block_job_pause_point(&s->common); |
335 | ||
e5b43573 FZ |
336 | /* Find the number of consective dirty chunks following the first dirty |
337 | * one, and wait for in flight requests in them. */ | |
338 | while (nb_chunks * sectors_per_chunk < (s->buf_size >> BDRV_SECTOR_BITS)) { | |
339 | int64_t hbitmap_next; | |
340 | int64_t next_sector = sector_num + nb_chunks * sectors_per_chunk; | |
341 | int64_t next_chunk = next_sector / sectors_per_chunk; | |
342 | if (next_sector >= end || | |
343 | !bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) { | |
344 | break; | |
345 | } | |
346 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 347 | break; |
e5b43573 | 348 | } |
9c83625b HR |
349 | |
350 | hbitmap_next = hbitmap_iter_next(&s->hbi); | |
f27a2742 HR |
351 | if (hbitmap_next > next_sector || hbitmap_next < 0) { |
352 | /* The bitmap iterator's cache is stale, refresh it */ | |
353 | bdrv_set_dirty_iter(&s->hbi, next_sector); | |
354 | hbitmap_next = hbitmap_iter_next(&s->hbi); | |
355 | } | |
9c83625b HR |
356 | assert(hbitmap_next == next_sector); |
357 | nb_chunks++; | |
e5b43573 FZ |
358 | } |
359 | ||
360 | /* Clear dirty bits before querying the block status, because | |
361 | * calling bdrv_get_block_status_above could yield - if some blocks are | |
362 | * marked dirty in this window, we need to know. | |
363 | */ | |
364 | bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, | |
365 | nb_chunks * sectors_per_chunk); | |
366 | bitmap_set(s->in_flight_bitmap, sector_num / sectors_per_chunk, nb_chunks); | |
367 | while (nb_chunks > 0 && sector_num < end) { | |
368 | int ret; | |
369 | int io_sectors; | |
370 | BlockDriverState *file; | |
371 | enum MirrorMethod { | |
372 | MIRROR_METHOD_COPY, | |
373 | MIRROR_METHOD_ZERO, | |
374 | MIRROR_METHOD_DISCARD | |
375 | } mirror_method = MIRROR_METHOD_COPY; | |
376 | ||
377 | assert(!(sector_num % sectors_per_chunk)); | |
378 | ret = bdrv_get_block_status_above(source, NULL, sector_num, | |
379 | nb_chunks * sectors_per_chunk, | |
380 | &io_sectors, &file); | |
381 | if (ret < 0) { | |
382 | io_sectors = nb_chunks * sectors_per_chunk; | |
383 | } | |
384 | ||
385 | io_sectors -= io_sectors % sectors_per_chunk; | |
386 | if (io_sectors < sectors_per_chunk) { | |
387 | io_sectors = sectors_per_chunk; | |
388 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { | |
389 | int64_t target_sector_num; | |
390 | int target_nb_sectors; | |
244483e6 KW |
391 | bdrv_round_sectors_to_clusters(blk_bs(s->target), sector_num, |
392 | io_sectors, &target_sector_num, | |
393 | &target_nb_sectors); | |
e5b43573 FZ |
394 | if (target_sector_num == sector_num && |
395 | target_nb_sectors == io_sectors) { | |
396 | mirror_method = ret & BDRV_BLOCK_ZERO ? | |
397 | MIRROR_METHOD_ZERO : | |
398 | MIRROR_METHOD_DISCARD; | |
399 | } | |
400 | } | |
401 | ||
4150ae60 | 402 | mirror_clip_sectors(s, sector_num, &io_sectors); |
e5b43573 FZ |
403 | switch (mirror_method) { |
404 | case MIRROR_METHOD_COPY: | |
405 | io_sectors = mirror_do_read(s, sector_num, io_sectors); | |
406 | break; | |
407 | case MIRROR_METHOD_ZERO: | |
408 | mirror_do_zero_or_discard(s, sector_num, io_sectors, false); | |
409 | break; | |
410 | case MIRROR_METHOD_DISCARD: | |
411 | mirror_do_zero_or_discard(s, sector_num, io_sectors, true); | |
412 | break; | |
413 | default: | |
414 | abort(); | |
415 | } | |
416 | assert(io_sectors); | |
417 | sector_num += io_sectors; | |
4150ae60 | 418 | nb_chunks -= DIV_ROUND_UP(io_sectors, sectors_per_chunk); |
e5b43573 | 419 | delay_ns += ratelimit_calculate_delay(&s->limit, io_sectors); |
dcfb3beb | 420 | } |
cc8c9d6c | 421 | return delay_ns; |
bd48bde8 | 422 | } |
b952b558 | 423 | |
402a4741 PB |
424 | static void mirror_free_init(MirrorBlockJob *s) |
425 | { | |
426 | int granularity = s->granularity; | |
427 | size_t buf_size = s->buf_size; | |
428 | uint8_t *buf = s->buf; | |
429 | ||
430 | assert(s->buf_free_count == 0); | |
431 | QSIMPLEQ_INIT(&s->buf_free); | |
432 | while (buf_size != 0) { | |
433 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
434 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
435 | s->buf_free_count++; | |
436 | buf_size -= granularity; | |
437 | buf += granularity; | |
438 | } | |
439 | } | |
440 | ||
bd48bde8 PB |
441 | static void mirror_drain(MirrorBlockJob *s) |
442 | { | |
443 | while (s->in_flight > 0) { | |
21cd917f | 444 | mirror_wait_for_io(s); |
bd48bde8 | 445 | } |
893f7eba PB |
446 | } |
447 | ||
5a7e7a0b SH |
448 | typedef struct { |
449 | int ret; | |
450 | } MirrorExitData; | |
451 | ||
452 | static void mirror_exit(BlockJob *job, void *opaque) | |
453 | { | |
454 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
455 | MirrorExitData *data = opaque; | |
456 | AioContext *replace_aio_context = NULL; | |
e253f4b8 KW |
457 | BlockDriverState *src = blk_bs(s->common.blk); |
458 | BlockDriverState *target_bs = blk_bs(s->target); | |
3f09bfbc KW |
459 | |
460 | /* Make sure that the source BDS doesn't go away before we called | |
461 | * block_job_completed(). */ | |
462 | bdrv_ref(src); | |
5a7e7a0b SH |
463 | |
464 | if (s->to_replace) { | |
465 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
466 | aio_context_acquire(replace_aio_context); | |
467 | } | |
468 | ||
469 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 470 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
471 | if (s->to_replace) { |
472 | to_replace = s->to_replace; | |
473 | } | |
40365552 | 474 | |
e253f4b8 KW |
475 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
476 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 477 | } |
b8804815 KW |
478 | |
479 | /* The mirror job has no requests in flight any more, but we need to | |
480 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 KW |
481 | bdrv_drained_begin(target_bs); |
482 | bdrv_replace_in_backing_chain(to_replace, target_bs); | |
483 | bdrv_drained_end(target_bs); | |
b8804815 | 484 | |
b6d2e599 KW |
485 | /* We just changed the BDS the job BB refers to */ |
486 | blk_remove_bs(job->blk); | |
487 | blk_insert_bs(job->blk, src); | |
5a7e7a0b SH |
488 | } |
489 | if (s->to_replace) { | |
490 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
491 | error_free(s->replace_blocker); | |
492 | bdrv_unref(s->to_replace); | |
493 | } | |
494 | if (replace_aio_context) { | |
495 | aio_context_release(replace_aio_context); | |
496 | } | |
497 | g_free(s->replaces); | |
e253f4b8 KW |
498 | bdrv_op_unblock_all(target_bs, s->common.blocker); |
499 | blk_unref(s->target); | |
5a7e7a0b SH |
500 | block_job_completed(&s->common, data->ret); |
501 | g_free(data); | |
176c3699 | 502 | bdrv_drained_end(src); |
ab27c3b5 FZ |
503 | if (qemu_get_aio_context() == bdrv_get_aio_context(src)) { |
504 | aio_enable_external(iohandler_get_aio_context()); | |
505 | } | |
3f09bfbc | 506 | bdrv_unref(src); |
5a7e7a0b SH |
507 | } |
508 | ||
893f7eba PB |
509 | static void coroutine_fn mirror_run(void *opaque) |
510 | { | |
511 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 512 | MirrorExitData *data; |
e253f4b8 KW |
513 | BlockDriverState *bs = blk_bs(s->common.blk); |
514 | BlockDriverState *target_bs = blk_bs(s->target); | |
99900697 | 515 | int64_t sector_num, end, length; |
bd48bde8 | 516 | uint64_t last_pause_ns; |
b812f671 | 517 | BlockDriverInfo bdi; |
1d33936e JC |
518 | char backing_filename[2]; /* we only need 2 characters because we are only |
519 | checking for a NULL string */ | |
893f7eba PB |
520 | int ret = 0; |
521 | int n; | |
e5b43573 | 522 | int target_cluster_size = BDRV_SECTOR_SIZE; |
893f7eba PB |
523 | |
524 | if (block_job_is_cancelled(&s->common)) { | |
525 | goto immediate_exit; | |
526 | } | |
527 | ||
b21c7652 HR |
528 | s->bdev_length = bdrv_getlength(bs); |
529 | if (s->bdev_length < 0) { | |
530 | ret = s->bdev_length; | |
373df5b1 | 531 | goto immediate_exit; |
b21c7652 | 532 | } else if (s->bdev_length == 0) { |
9e48b025 FZ |
533 | /* Report BLOCK_JOB_READY and wait for complete. */ |
534 | block_job_event_ready(&s->common); | |
535 | s->synced = true; | |
536 | while (!block_job_is_cancelled(&s->common) && !s->should_complete) { | |
537 | block_job_yield(&s->common); | |
538 | } | |
539 | s->common.cancelled = false; | |
540 | goto immediate_exit; | |
893f7eba PB |
541 | } |
542 | ||
b21c7652 | 543 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
544 | s->in_flight_bitmap = bitmap_new(length); |
545 | ||
b812f671 PB |
546 | /* If we have no backing file yet in the destination, we cannot let |
547 | * the destination do COW. Instead, we copy sectors around the | |
548 | * dirty data if needed. We need a bitmap to do that. | |
549 | */ | |
e253f4b8 | 550 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 551 | sizeof(backing_filename)); |
e253f4b8 | 552 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
e5b43573 FZ |
553 | target_cluster_size = bdi.cluster_size; |
554 | } | |
e253f4b8 | 555 | if (backing_filename[0] && !target_bs->backing |
e5b43573 FZ |
556 | && s->granularity < target_cluster_size) { |
557 | s->buf_size = MAX(s->buf_size, target_cluster_size); | |
558 | s->cow_bitmap = bitmap_new(length); | |
b812f671 | 559 | } |
e5b43573 | 560 | s->target_cluster_sectors = target_cluster_size >> BDRV_SECTOR_BITS; |
e253f4b8 | 561 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 562 | |
b21c7652 | 563 | end = s->bdev_length / BDRV_SECTOR_SIZE; |
7504edf4 KW |
564 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
565 | if (s->buf == NULL) { | |
566 | ret = -ENOMEM; | |
567 | goto immediate_exit; | |
568 | } | |
569 | ||
402a4741 | 570 | mirror_free_init(s); |
893f7eba | 571 | |
4c0cbd6f | 572 | last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 573 | if (!s->is_none_mode) { |
893f7eba | 574 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
5bc361b8 | 575 | BlockDriverState *base = s->base; |
e253f4b8 | 576 | bool mark_all_dirty = s->base == NULL && !bdrv_has_zero_init(target_bs); |
5279efeb | 577 | |
893f7eba | 578 | for (sector_num = 0; sector_num < end; ) { |
99900697 FZ |
579 | /* Just to make sure we are not exceeding int limit. */ |
580 | int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, | |
581 | end - sector_num); | |
4c0cbd6f FZ |
582 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
583 | ||
584 | if (now - last_pause_ns > SLICE_TIME) { | |
585 | last_pause_ns = now; | |
586 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); | |
565ac01f SH |
587 | } else { |
588 | block_job_pause_point(&s->common); | |
4c0cbd6f FZ |
589 | } |
590 | ||
591 | if (block_job_is_cancelled(&s->common)) { | |
592 | goto immediate_exit; | |
593 | } | |
594 | ||
99900697 | 595 | ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); |
893f7eba PB |
596 | |
597 | if (ret < 0) { | |
598 | goto immediate_exit; | |
599 | } | |
600 | ||
601 | assert(n > 0); | |
5279efeb | 602 | if (ret == 1 || mark_all_dirty) { |
20dca810 | 603 | bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); |
893f7eba | 604 | } |
99900697 | 605 | sector_num += n; |
893f7eba PB |
606 | } |
607 | } | |
608 | ||
20dca810 | 609 | bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi); |
893f7eba | 610 | for (;;) { |
cc8c9d6c | 611 | uint64_t delay_ns = 0; |
893f7eba PB |
612 | int64_t cnt; |
613 | bool should_complete; | |
614 | ||
bd48bde8 PB |
615 | if (s->ret < 0) { |
616 | ret = s->ret; | |
617 | goto immediate_exit; | |
618 | } | |
619 | ||
565ac01f SH |
620 | block_job_pause_point(&s->common); |
621 | ||
20dca810 | 622 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
b21c7652 HR |
623 | /* s->common.offset contains the number of bytes already processed so |
624 | * far, cnt is the number of dirty sectors remaining and | |
625 | * s->sectors_in_flight is the number of sectors currently being | |
626 | * processed; together those are the current total operation length */ | |
627 | s->common.len = s->common.offset + | |
628 | (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
629 | |
630 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 631 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
bd48bde8 PB |
632 | * We do so every SLICE_TIME nanoseconds, or when there is an error, |
633 | * or when the source is clean, whichever comes first. | |
634 | */ | |
bc72ad67 | 635 | if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME && |
bd48bde8 | 636 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
402a4741 PB |
637 | if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 || |
638 | (cnt == 0 && s->in_flight > 0)) { | |
639 | trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt); | |
21cd917f | 640 | mirror_wait_for_io(s); |
bd48bde8 PB |
641 | continue; |
642 | } else if (cnt != 0) { | |
cc8c9d6c | 643 | delay_ns = mirror_iteration(s); |
893f7eba | 644 | } |
893f7eba PB |
645 | } |
646 | ||
647 | should_complete = false; | |
bd48bde8 | 648 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 649 | trace_mirror_before_flush(s); |
e253f4b8 | 650 | ret = blk_flush(s->target); |
893f7eba | 651 | if (ret < 0) { |
a589569f WX |
652 | if (mirror_error_action(s, false, -ret) == |
653 | BLOCK_ERROR_ACTION_REPORT) { | |
b952b558 PB |
654 | goto immediate_exit; |
655 | } | |
656 | } else { | |
657 | /* We're out of the streaming phase. From now on, if the job | |
658 | * is cancelled we will actually complete all pending I/O and | |
659 | * report completion. This way, block-job-cancel will leave | |
660 | * the target in a consistent state. | |
661 | */ | |
b952b558 | 662 | if (!s->synced) { |
bcada37b | 663 | block_job_event_ready(&s->common); |
b952b558 PB |
664 | s->synced = true; |
665 | } | |
666 | ||
667 | should_complete = s->should_complete || | |
668 | block_job_is_cancelled(&s->common); | |
20dca810 | 669 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
d63ffd87 | 670 | } |
893f7eba PB |
671 | } |
672 | ||
673 | if (cnt == 0 && should_complete) { | |
674 | /* The dirty bitmap is not updated while operations are pending. | |
675 | * If we're about to exit, wait for pending operations before | |
676 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
677 | * source has dirty data to copy! | |
678 | * | |
679 | * Note that I/O can be submitted by the guest while | |
680 | * mirror_populate runs. | |
681 | */ | |
682 | trace_mirror_before_drain(s, cnt); | |
39bf92dd | 683 | bdrv_co_drain(bs); |
20dca810 | 684 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
893f7eba PB |
685 | } |
686 | ||
687 | ret = 0; | |
cc8c9d6c | 688 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
d63ffd87 | 689 | if (!s->synced) { |
7483d1e5 | 690 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
691 | if (block_job_is_cancelled(&s->common)) { |
692 | break; | |
693 | } | |
694 | } else if (!should_complete) { | |
bd48bde8 | 695 | delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); |
7483d1e5 | 696 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
697 | } else if (cnt == 0) { |
698 | /* The two disks are in sync. Exit and report successful | |
699 | * completion. | |
700 | */ | |
701 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
702 | s->common.cancelled = false; | |
703 | break; | |
704 | } | |
bc72ad67 | 705 | last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
706 | } |
707 | ||
708 | immediate_exit: | |
bd48bde8 PB |
709 | if (s->in_flight > 0) { |
710 | /* We get here only if something went wrong. Either the job failed, | |
711 | * or it was cancelled prematurely so that we do not guarantee that | |
712 | * the target is a copy of the source. | |
713 | */ | |
714 | assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); | |
715 | mirror_drain(s); | |
716 | } | |
717 | ||
718 | assert(s->in_flight == 0); | |
7191bf31 | 719 | qemu_vfree(s->buf); |
b812f671 | 720 | g_free(s->cow_bitmap); |
402a4741 | 721 | g_free(s->in_flight_bitmap); |
e4654d2d | 722 | bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); |
5a7e7a0b SH |
723 | |
724 | data = g_malloc(sizeof(*data)); | |
725 | data->ret = ret; | |
176c3699 FZ |
726 | /* Before we switch to target in mirror_exit, make sure data doesn't |
727 | * change. */ | |
e253f4b8 | 728 | bdrv_drained_begin(bs); |
ab27c3b5 FZ |
729 | if (qemu_get_aio_context() == bdrv_get_aio_context(bs)) { |
730 | /* FIXME: virtio host notifiers run on iohandler_ctx, therefore the | |
731 | * above bdrv_drained_end isn't enough to quiesce it. This is ugly, we | |
732 | * need a block layer API change to achieve this. */ | |
733 | aio_disable_external(iohandler_get_aio_context()); | |
734 | } | |
5a7e7a0b | 735 | block_job_defer_to_main_loop(&s->common, mirror_exit, data); |
893f7eba PB |
736 | } |
737 | ||
738 | static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
739 | { | |
740 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
741 | ||
742 | if (speed < 0) { | |
c6bd8c70 | 743 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
893f7eba PB |
744 | return; |
745 | } | |
746 | ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); | |
747 | } | |
748 | ||
d63ffd87 PB |
749 | static void mirror_complete(BlockJob *job, Error **errp) |
750 | { | |
751 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
274fccee HR |
752 | BlockDriverState *src, *target; |
753 | ||
754 | src = blk_bs(job->blk); | |
755 | target = blk_bs(s->target); | |
d63ffd87 | 756 | |
d63ffd87 | 757 | if (!s->synced) { |
8ccb9569 | 758 | error_setg(errp, QERR_BLOCK_JOB_NOT_READY, job->id); |
d63ffd87 PB |
759 | return; |
760 | } | |
761 | ||
274fccee HR |
762 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
763 | int ret; | |
764 | ||
765 | assert(!target->backing); | |
766 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
767 | if (ret < 0) { | |
768 | return; | |
769 | } | |
770 | } | |
771 | ||
09158f00 BC |
772 | /* check the target bs is not blocked and block all operations on it */ |
773 | if (s->replaces) { | |
5a7e7a0b SH |
774 | AioContext *replace_aio_context; |
775 | ||
e12f3784 | 776 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 777 | if (!s->to_replace) { |
e12f3784 | 778 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
779 | return; |
780 | } | |
781 | ||
5a7e7a0b SH |
782 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
783 | aio_context_acquire(replace_aio_context); | |
784 | ||
09158f00 BC |
785 | error_setg(&s->replace_blocker, |
786 | "block device is in use by block-job-complete"); | |
787 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
788 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
789 | |
790 | aio_context_release(replace_aio_context); | |
09158f00 BC |
791 | } |
792 | ||
274fccee HR |
793 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { |
794 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
795 | if (backing_bs(target) != backing) { | |
796 | bdrv_set_backing_hd(target, backing); | |
797 | } | |
798 | } | |
799 | ||
d63ffd87 | 800 | s->should_complete = true; |
751ebd76 | 801 | block_job_enter(&s->common); |
d63ffd87 PB |
802 | } |
803 | ||
565ac01f SH |
804 | /* There is no matching mirror_resume() because mirror_run() will begin |
805 | * iterating again when the job is resumed. | |
806 | */ | |
807 | static void coroutine_fn mirror_pause(BlockJob *job) | |
808 | { | |
809 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
810 | ||
811 | mirror_drain(s); | |
812 | } | |
813 | ||
814 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) | |
815 | { | |
816 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
817 | ||
818 | blk_set_aio_context(s->target, new_context); | |
819 | } | |
820 | ||
3fc4b10a | 821 | static const BlockJobDriver mirror_job_driver = { |
565ac01f SH |
822 | .instance_size = sizeof(MirrorBlockJob), |
823 | .job_type = BLOCK_JOB_TYPE_MIRROR, | |
824 | .set_speed = mirror_set_speed, | |
825 | .complete = mirror_complete, | |
826 | .pause = mirror_pause, | |
827 | .attached_aio_context = mirror_attached_aio_context, | |
893f7eba PB |
828 | }; |
829 | ||
03544a6e | 830 | static const BlockJobDriver commit_active_job_driver = { |
565ac01f SH |
831 | .instance_size = sizeof(MirrorBlockJob), |
832 | .job_type = BLOCK_JOB_TYPE_COMMIT, | |
833 | .set_speed = mirror_set_speed, | |
834 | .complete = mirror_complete, | |
835 | .pause = mirror_pause, | |
836 | .attached_aio_context = mirror_attached_aio_context, | |
03544a6e FZ |
837 | }; |
838 | ||
839 | static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target, | |
09158f00 | 840 | const char *replaces, |
5fba6c0e | 841 | int64_t speed, uint32_t granularity, |
09158f00 | 842 | int64_t buf_size, |
274fccee | 843 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
844 | BlockdevOnError on_source_error, |
845 | BlockdevOnError on_target_error, | |
0fc9f8ea | 846 | bool unmap, |
097310b5 | 847 | BlockCompletionFunc *cb, |
09158f00 BC |
848 | void *opaque, Error **errp, |
849 | const BlockJobDriver *driver, | |
850 | bool is_none_mode, BlockDriverState *base) | |
893f7eba PB |
851 | { |
852 | MirrorBlockJob *s; | |
853 | ||
eee13dfe | 854 | if (granularity == 0) { |
341ebc2f | 855 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
856 | } |
857 | ||
858 | assert ((granularity & (granularity - 1)) == 0); | |
859 | ||
48ac0a4d WC |
860 | if (buf_size < 0) { |
861 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
862 | return; | |
863 | } | |
864 | ||
865 | if (buf_size == 0) { | |
866 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
867 | } | |
5bc361b8 | 868 | |
03544a6e | 869 | s = block_job_create(driver, bs, speed, cb, opaque, errp); |
893f7eba PB |
870 | if (!s) { |
871 | return; | |
872 | } | |
873 | ||
e253f4b8 KW |
874 | s->target = blk_new(); |
875 | blk_insert_bs(s->target, target); | |
876 | ||
09158f00 | 877 | s->replaces = g_strdup(replaces); |
b952b558 PB |
878 | s->on_source_error = on_source_error; |
879 | s->on_target_error = on_target_error; | |
03544a6e | 880 | s->is_none_mode = is_none_mode; |
274fccee | 881 | s->backing_mode = backing_mode; |
5bc361b8 | 882 | s->base = base; |
eee13dfe | 883 | s->granularity = granularity; |
48ac0a4d | 884 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 885 | s->unmap = unmap; |
b812f671 | 886 | |
0db6e54a | 887 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 888 | if (!s->dirty_bitmap) { |
97031164 | 889 | g_free(s->replaces); |
e253f4b8 | 890 | blk_unref(s->target); |
18930ba3 | 891 | block_job_unref(&s->common); |
b8afb520 FZ |
892 | return; |
893 | } | |
10f3cd15 | 894 | |
e253f4b8 | 895 | bdrv_op_block_all(target, s->common.blocker); |
10f3cd15 | 896 | |
893f7eba PB |
897 | s->common.co = qemu_coroutine_create(mirror_run); |
898 | trace_mirror_start(bs, s, s->common.co, opaque); | |
899 | qemu_coroutine_enter(s->common.co, s); | |
900 | } | |
03544a6e FZ |
901 | |
902 | void mirror_start(BlockDriverState *bs, BlockDriverState *target, | |
09158f00 | 903 | const char *replaces, |
5fba6c0e | 904 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
905 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
906 | BlockdevOnError on_source_error, | |
03544a6e | 907 | BlockdevOnError on_target_error, |
0fc9f8ea | 908 | bool unmap, |
097310b5 | 909 | BlockCompletionFunc *cb, |
03544a6e FZ |
910 | void *opaque, Error **errp) |
911 | { | |
912 | bool is_none_mode; | |
913 | BlockDriverState *base; | |
914 | ||
4b80ab2b JS |
915 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
916 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
917 | return; |
918 | } | |
03544a6e | 919 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 920 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
09158f00 | 921 | mirror_start_job(bs, target, replaces, |
274fccee | 922 | speed, granularity, buf_size, backing_mode, |
0fc9f8ea | 923 | on_source_error, on_target_error, unmap, cb, opaque, errp, |
03544a6e FZ |
924 | &mirror_job_driver, is_none_mode, base); |
925 | } | |
926 | ||
927 | void commit_active_start(BlockDriverState *bs, BlockDriverState *base, | |
928 | int64_t speed, | |
929 | BlockdevOnError on_error, | |
097310b5 | 930 | BlockCompletionFunc *cb, |
03544a6e FZ |
931 | void *opaque, Error **errp) |
932 | { | |
4da83585 JC |
933 | int64_t length, base_length; |
934 | int orig_base_flags; | |
39a611a3 | 935 | int ret; |
cc67f4d1 | 936 | Error *local_err = NULL; |
4da83585 JC |
937 | |
938 | orig_base_flags = bdrv_get_flags(base); | |
939 | ||
20a63d2c FZ |
940 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
941 | return; | |
942 | } | |
4da83585 JC |
943 | |
944 | length = bdrv_getlength(bs); | |
945 | if (length < 0) { | |
39a611a3 JC |
946 | error_setg_errno(errp, -length, |
947 | "Unable to determine length of %s", bs->filename); | |
4da83585 JC |
948 | goto error_restore_flags; |
949 | } | |
950 | ||
951 | base_length = bdrv_getlength(base); | |
952 | if (base_length < 0) { | |
39a611a3 JC |
953 | error_setg_errno(errp, -base_length, |
954 | "Unable to determine length of %s", base->filename); | |
4da83585 JC |
955 | goto error_restore_flags; |
956 | } | |
957 | ||
958 | if (length > base_length) { | |
39a611a3 JC |
959 | ret = bdrv_truncate(base, length); |
960 | if (ret < 0) { | |
961 | error_setg_errno(errp, -ret, | |
962 | "Top image %s is larger than base image %s, and " | |
4da83585 JC |
963 | "resize of base image failed", |
964 | bs->filename, base->filename); | |
965 | goto error_restore_flags; | |
966 | } | |
967 | } | |
968 | ||
274fccee | 969 | mirror_start_job(bs, base, NULL, speed, 0, 0, MIRROR_LEAVE_BACKING_CHAIN, |
0fc9f8ea | 970 | on_error, on_error, false, cb, opaque, &local_err, |
03544a6e | 971 | &commit_active_job_driver, false, base); |
0fb6395c | 972 | if (local_err) { |
cc67f4d1 | 973 | error_propagate(errp, local_err); |
4da83585 JC |
974 | goto error_restore_flags; |
975 | } | |
976 | ||
977 | return; | |
978 | ||
979 | error_restore_flags: | |
980 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
981 | * the original error */ | |
982 | bdrv_reopen(base, orig_base_flags, NULL); | |
983 | return; | |
03544a6e | 984 | } |