]>
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" |
fd4a6493 | 15 | #include "qemu/cutils.h" |
893f7eba | 16 | #include "trace.h" |
c87621ea | 17 | #include "block/blockjob_int.h" |
737e150e | 18 | #include "block/block_int.h" |
373340b2 | 19 | #include "sysemu/block-backend.h" |
da34e65c | 20 | #include "qapi/error.h" |
cc7a8ea7 | 21 | #include "qapi/qmp/qerror.h" |
893f7eba | 22 | #include "qemu/ratelimit.h" |
b812f671 | 23 | #include "qemu/bitmap.h" |
893f7eba | 24 | |
402a4741 | 25 | #define MAX_IN_FLIGHT 16 |
b436982f EB |
26 | #define MAX_IO_BYTES (1 << 20) /* 1 Mb */ |
27 | #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES) | |
402a4741 PB |
28 | |
29 | /* The mirroring buffer is a list of granularity-sized chunks. | |
30 | * Free chunks are organized in a list. | |
31 | */ | |
32 | typedef struct MirrorBuffer { | |
33 | QSIMPLEQ_ENTRY(MirrorBuffer) next; | |
34 | } MirrorBuffer; | |
893f7eba PB |
35 | |
36 | typedef struct MirrorBlockJob { | |
37 | BlockJob common; | |
e253f4b8 | 38 | BlockBackend *target; |
4ef85a9c KW |
39 | BlockDriverState *mirror_top_bs; |
40 | BlockDriverState *source; | |
5bc361b8 | 41 | BlockDriverState *base; |
4ef85a9c | 42 | |
09158f00 BC |
43 | /* The name of the graph node to replace */ |
44 | char *replaces; | |
45 | /* The BDS to replace */ | |
46 | BlockDriverState *to_replace; | |
47 | /* Used to block operations on the drive-mirror-replace target */ | |
48 | Error *replace_blocker; | |
03544a6e | 49 | bool is_none_mode; |
274fccee | 50 | BlockMirrorBackingMode backing_mode; |
b952b558 | 51 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 PB |
52 | bool synced; |
53 | bool should_complete; | |
eee13dfe | 54 | int64_t granularity; |
b812f671 | 55 | size_t buf_size; |
b21c7652 | 56 | int64_t bdev_length; |
b812f671 | 57 | unsigned long *cow_bitmap; |
e4654d2d | 58 | BdrvDirtyBitmap *dirty_bitmap; |
dc162c8e | 59 | BdrvDirtyBitmapIter *dbi; |
893f7eba | 60 | uint8_t *buf; |
402a4741 PB |
61 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
62 | int buf_free_count; | |
bd48bde8 | 63 | |
49efb1f5 | 64 | uint64_t last_pause_ns; |
402a4741 | 65 | unsigned long *in_flight_bitmap; |
bd48bde8 | 66 | int in_flight; |
b436982f | 67 | int64_t bytes_in_flight; |
bd48bde8 | 68 | int ret; |
0fc9f8ea | 69 | bool unmap; |
e424aff5 | 70 | bool waiting_for_io; |
b436982f | 71 | int target_cluster_size; |
e5b43573 | 72 | int max_iov; |
90ab48eb | 73 | bool initial_zeroing_ongoing; |
893f7eba PB |
74 | } MirrorBlockJob; |
75 | ||
bd48bde8 PB |
76 | typedef struct MirrorOp { |
77 | MirrorBlockJob *s; | |
78 | QEMUIOVector qiov; | |
b436982f EB |
79 | int64_t offset; |
80 | uint64_t bytes; | |
bd48bde8 PB |
81 | } MirrorOp; |
82 | ||
4295c5fc HR |
83 | typedef enum MirrorMethod { |
84 | MIRROR_METHOD_COPY, | |
85 | MIRROR_METHOD_ZERO, | |
86 | MIRROR_METHOD_DISCARD, | |
87 | } MirrorMethod; | |
88 | ||
b952b558 PB |
89 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
90 | int error) | |
91 | { | |
92 | s->synced = false; | |
93 | if (read) { | |
81e254dc KW |
94 | return block_job_error_action(&s->common, s->on_source_error, |
95 | true, error); | |
b952b558 | 96 | } else { |
81e254dc KW |
97 | return block_job_error_action(&s->common, s->on_target_error, |
98 | false, error); | |
b952b558 PB |
99 | } |
100 | } | |
101 | ||
bd48bde8 PB |
102 | static void mirror_iteration_done(MirrorOp *op, int ret) |
103 | { | |
104 | MirrorBlockJob *s = op->s; | |
402a4741 | 105 | struct iovec *iov; |
bd48bde8 | 106 | int64_t chunk_num; |
b436982f | 107 | int i, nb_chunks; |
bd48bde8 | 108 | |
b436982f | 109 | trace_mirror_iteration_done(s, op->offset, op->bytes, ret); |
bd48bde8 PB |
110 | |
111 | s->in_flight--; | |
b436982f | 112 | s->bytes_in_flight -= op->bytes; |
402a4741 PB |
113 | iov = op->qiov.iov; |
114 | for (i = 0; i < op->qiov.niov; i++) { | |
115 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
116 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
117 | s->buf_free_count++; | |
118 | } | |
119 | ||
b436982f EB |
120 | chunk_num = op->offset / s->granularity; |
121 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
402a4741 | 122 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
b21c7652 HR |
123 | if (ret >= 0) { |
124 | if (s->cow_bitmap) { | |
125 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
126 | } | |
90ab48eb | 127 | if (!s->initial_zeroing_ongoing) { |
30a5c887 | 128 | job_progress_update(&s->common.job, op->bytes); |
90ab48eb | 129 | } |
bd48bde8 | 130 | } |
6df3bf8e | 131 | qemu_iovec_destroy(&op->qiov); |
c84b3192 | 132 | g_free(op); |
7b770c72 | 133 | |
e424aff5 | 134 | if (s->waiting_for_io) { |
da01ff7f | 135 | qemu_coroutine_enter(s->common.job.co); |
7b770c72 | 136 | } |
bd48bde8 PB |
137 | } |
138 | ||
139 | static void mirror_write_complete(void *opaque, int ret) | |
140 | { | |
141 | MirrorOp *op = opaque; | |
142 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
143 | |
144 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 145 | if (ret < 0) { |
bd48bde8 PB |
146 | BlockErrorAction action; |
147 | ||
e0d7f73e | 148 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 149 | action = mirror_error_action(s, false, -ret); |
a589569f | 150 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
151 | s->ret = ret; |
152 | } | |
153 | } | |
154 | mirror_iteration_done(op, ret); | |
b9e413dd | 155 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
156 | } |
157 | ||
158 | static void mirror_read_complete(void *opaque, int ret) | |
159 | { | |
160 | MirrorOp *op = opaque; | |
161 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
162 | |
163 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 164 | if (ret < 0) { |
bd48bde8 PB |
165 | BlockErrorAction action; |
166 | ||
e0d7f73e | 167 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 168 | action = mirror_error_action(s, true, -ret); |
a589569f | 169 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
170 | s->ret = ret; |
171 | } | |
172 | ||
173 | mirror_iteration_done(op, ret); | |
b9e413dd | 174 | } else { |
b436982f | 175 | blk_aio_pwritev(s->target, op->offset, &op->qiov, |
b9e413dd | 176 | 0, mirror_write_complete, op); |
bd48bde8 | 177 | } |
b9e413dd | 178 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
179 | } |
180 | ||
782d97ef EB |
181 | /* Clip bytes relative to offset to not exceed end-of-file */ |
182 | static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, | |
183 | int64_t offset, | |
184 | int64_t bytes) | |
185 | { | |
186 | return MIN(bytes, s->bdev_length - offset); | |
187 | } | |
188 | ||
782d97ef EB |
189 | /* Round offset and/or bytes to target cluster if COW is needed, and |
190 | * return the offset of the adjusted tail against original. */ | |
191 | static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, | |
ae4cc877 | 192 | uint64_t *bytes) |
893f7eba | 193 | { |
e5b43573 FZ |
194 | bool need_cow; |
195 | int ret = 0; | |
782d97ef | 196 | int64_t align_offset = *offset; |
7cfd5275 | 197 | int64_t align_bytes = *bytes; |
782d97ef | 198 | int max_bytes = s->granularity * s->max_iov; |
e5b43573 | 199 | |
782d97ef EB |
200 | need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); |
201 | need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, | |
e5b43573 FZ |
202 | s->cow_bitmap); |
203 | if (need_cow) { | |
782d97ef EB |
204 | bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, |
205 | &align_offset, &align_bytes); | |
e5b43573 | 206 | } |
3515727f | 207 | |
782d97ef EB |
208 | if (align_bytes > max_bytes) { |
209 | align_bytes = max_bytes; | |
e5b43573 | 210 | if (need_cow) { |
782d97ef | 211 | align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); |
e5b43573 | 212 | } |
8f0720ec | 213 | } |
782d97ef | 214 | /* Clipping may result in align_bytes unaligned to chunk boundary, but |
4150ae60 | 215 | * that doesn't matter because it's already the end of source image. */ |
782d97ef | 216 | align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); |
8f0720ec | 217 | |
782d97ef EB |
218 | ret = align_offset + align_bytes - (*offset + *bytes); |
219 | *offset = align_offset; | |
220 | *bytes = align_bytes; | |
e5b43573 FZ |
221 | assert(ret >= 0); |
222 | return ret; | |
223 | } | |
224 | ||
21cd917f FZ |
225 | static inline void mirror_wait_for_io(MirrorBlockJob *s) |
226 | { | |
227 | assert(!s->waiting_for_io); | |
228 | s->waiting_for_io = true; | |
229 | qemu_coroutine_yield(); | |
230 | s->waiting_for_io = false; | |
231 | } | |
232 | ||
e5b43573 | 233 | /* Submit async read while handling COW. |
ae4cc877 EB |
234 | * Returns: The number of bytes copied after and including offset, |
235 | * excluding any bytes copied prior to offset due to alignment. | |
236 | * This will be @bytes if no alignment is necessary, or | |
237 | * (new_end - offset) if tail is rounded up or down due to | |
e5b43573 FZ |
238 | * alignment or buffer limit. |
239 | */ | |
ae4cc877 EB |
240 | static uint64_t mirror_do_read(MirrorBlockJob *s, int64_t offset, |
241 | uint64_t bytes) | |
e5b43573 | 242 | { |
e253f4b8 | 243 | BlockBackend *source = s->common.blk; |
ae4cc877 EB |
244 | int nb_chunks; |
245 | uint64_t ret; | |
e5b43573 | 246 | MirrorOp *op; |
ae4cc877 | 247 | uint64_t max_bytes; |
e5b43573 | 248 | |
ae4cc877 | 249 | max_bytes = s->granularity * s->max_iov; |
402a4741 | 250 | |
e5b43573 | 251 | /* We can only handle as much as buf_size at a time. */ |
ae4cc877 EB |
252 | bytes = MIN(s->buf_size, MIN(max_bytes, bytes)); |
253 | assert(bytes); | |
254 | assert(bytes < BDRV_REQUEST_MAX_BYTES); | |
255 | ret = bytes; | |
402a4741 | 256 | |
e5b43573 | 257 | if (s->cow_bitmap) { |
ae4cc877 | 258 | ret += mirror_cow_align(s, &offset, &bytes); |
e5b43573 | 259 | } |
ae4cc877 EB |
260 | assert(bytes <= s->buf_size); |
261 | /* The offset is granularity-aligned because: | |
e5b43573 FZ |
262 | * 1) Caller passes in aligned values; |
263 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
ae4cc877 EB |
264 | assert(QEMU_IS_ALIGNED(offset, s->granularity)); |
265 | /* The range is sector-aligned, since bdrv_getlength() rounds up. */ | |
266 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
267 | nb_chunks = DIV_ROUND_UP(bytes, s->granularity); | |
e5b43573 FZ |
268 | |
269 | while (s->buf_free_count < nb_chunks) { | |
ae4cc877 | 270 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
21cd917f | 271 | mirror_wait_for_io(s); |
b812f671 PB |
272 | } |
273 | ||
bd48bde8 | 274 | /* Allocate a MirrorOp that is used as an AIO callback. */ |
c84b3192 | 275 | op = g_new(MirrorOp, 1); |
bd48bde8 | 276 | op->s = s; |
ae4cc877 EB |
277 | op->offset = offset; |
278 | op->bytes = bytes; | |
402a4741 PB |
279 | |
280 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
281 | * from s->buf_free. | |
282 | */ | |
283 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
284 | while (nb_chunks-- > 0) { |
285 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
ae4cc877 | 286 | size_t remaining = bytes - op->qiov.size; |
5a0f6fd5 | 287 | |
402a4741 PB |
288 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
289 | s->buf_free_count--; | |
5a0f6fd5 | 290 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 291 | } |
bd48bde8 | 292 | |
893f7eba | 293 | /* Copy the dirty cluster. */ |
bd48bde8 | 294 | s->in_flight++; |
ae4cc877 EB |
295 | s->bytes_in_flight += bytes; |
296 | trace_mirror_one_iteration(s, offset, bytes); | |
dcfb3beb | 297 | |
ae4cc877 | 298 | blk_aio_preadv(source, offset, &op->qiov, 0, mirror_read_complete, op); |
e5b43573 FZ |
299 | return ret; |
300 | } | |
301 | ||
302 | static void mirror_do_zero_or_discard(MirrorBlockJob *s, | |
e6f24193 EB |
303 | int64_t offset, |
304 | uint64_t bytes, | |
e5b43573 FZ |
305 | bool is_discard) |
306 | { | |
307 | MirrorOp *op; | |
308 | ||
309 | /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed | |
310 | * so the freeing in mirror_iteration_done is nop. */ | |
311 | op = g_new0(MirrorOp, 1); | |
312 | op->s = s; | |
e6f24193 EB |
313 | op->offset = offset; |
314 | op->bytes = bytes; | |
e5b43573 FZ |
315 | |
316 | s->in_flight++; | |
e6f24193 | 317 | s->bytes_in_flight += bytes; |
e5b43573 | 318 | if (is_discard) { |
e6f24193 | 319 | blk_aio_pdiscard(s->target, offset, |
b436982f | 320 | op->bytes, mirror_write_complete, op); |
e5b43573 | 321 | } else { |
e6f24193 | 322 | blk_aio_pwrite_zeroes(s->target, offset, |
b436982f | 323 | op->bytes, s->unmap ? BDRV_REQ_MAY_UNMAP : 0, |
dcfb3beb | 324 | mirror_write_complete, op); |
e5b43573 FZ |
325 | } |
326 | } | |
327 | ||
4295c5fc HR |
328 | static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, |
329 | unsigned bytes, MirrorMethod mirror_method) | |
330 | { | |
331 | switch (mirror_method) { | |
332 | case MIRROR_METHOD_COPY: | |
333 | return mirror_do_read(s, offset, bytes); | |
334 | case MIRROR_METHOD_ZERO: | |
335 | case MIRROR_METHOD_DISCARD: | |
336 | mirror_do_zero_or_discard(s, offset, bytes, | |
337 | mirror_method == MIRROR_METHOD_DISCARD); | |
338 | return bytes; | |
339 | default: | |
340 | abort(); | |
341 | } | |
342 | } | |
343 | ||
e5b43573 FZ |
344 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) |
345 | { | |
4ef85a9c | 346 | BlockDriverState *source = s->source; |
fb2ef791 | 347 | int64_t offset, first_chunk; |
e5b43573 FZ |
348 | uint64_t delay_ns = 0; |
349 | /* At least the first dirty chunk is mirrored in one iteration. */ | |
350 | int nb_chunks = 1; | |
4b5004d9 | 351 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
b436982f | 352 | int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); |
e5b43573 | 353 | |
b64bd51e | 354 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
f798184c | 355 | offset = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 356 | if (offset < 0) { |
dc162c8e | 357 | bdrv_set_dirty_iter(s->dbi, 0); |
f798184c | 358 | offset = bdrv_dirty_iter_next(s->dbi); |
9a46dba7 | 359 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
fb2ef791 | 360 | assert(offset >= 0); |
e5b43573 | 361 | } |
b64bd51e | 362 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
e5b43573 | 363 | |
fb2ef791 | 364 | first_chunk = offset / s->granularity; |
9c83625b | 365 | while (test_bit(first_chunk, s->in_flight_bitmap)) { |
fb2ef791 | 366 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
9c83625b HR |
367 | mirror_wait_for_io(s); |
368 | } | |
369 | ||
da01ff7f | 370 | job_pause_point(&s->common.job); |
565ac01f | 371 | |
e5b43573 FZ |
372 | /* Find the number of consective dirty chunks following the first dirty |
373 | * one, and wait for in flight requests in them. */ | |
b64bd51e | 374 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
fb2ef791 | 375 | while (nb_chunks * s->granularity < s->buf_size) { |
dc162c8e | 376 | int64_t next_dirty; |
fb2ef791 EB |
377 | int64_t next_offset = offset + nb_chunks * s->granularity; |
378 | int64_t next_chunk = next_offset / s->granularity; | |
379 | if (next_offset >= s->bdev_length || | |
3b5d4df0 | 380 | !bdrv_get_dirty_locked(source, s->dirty_bitmap, next_offset)) { |
e5b43573 FZ |
381 | break; |
382 | } | |
383 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 384 | break; |
e5b43573 | 385 | } |
9c83625b | 386 | |
f798184c | 387 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 388 | if (next_dirty > next_offset || next_dirty < 0) { |
f27a2742 | 389 | /* The bitmap iterator's cache is stale, refresh it */ |
715a74d8 | 390 | bdrv_set_dirty_iter(s->dbi, next_offset); |
f798184c | 391 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
f27a2742 | 392 | } |
fb2ef791 | 393 | assert(next_dirty == next_offset); |
9c83625b | 394 | nb_chunks++; |
e5b43573 FZ |
395 | } |
396 | ||
397 | /* Clear dirty bits before querying the block status, because | |
31826642 | 398 | * calling bdrv_block_status_above could yield - if some blocks are |
e5b43573 FZ |
399 | * marked dirty in this window, we need to know. |
400 | */ | |
e0d7f73e EB |
401 | bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, |
402 | nb_chunks * s->granularity); | |
b64bd51e PB |
403 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
404 | ||
fb2ef791 EB |
405 | bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); |
406 | while (nb_chunks > 0 && offset < s->bdev_length) { | |
31826642 | 407 | int ret; |
7cfd5275 | 408 | int64_t io_bytes; |
f3e4ce4a | 409 | int64_t io_bytes_acct; |
4295c5fc | 410 | MirrorMethod mirror_method = MIRROR_METHOD_COPY; |
e5b43573 | 411 | |
fb2ef791 | 412 | assert(!(offset % s->granularity)); |
31826642 EB |
413 | ret = bdrv_block_status_above(source, NULL, offset, |
414 | nb_chunks * s->granularity, | |
415 | &io_bytes, NULL, NULL); | |
e5b43573 | 416 | if (ret < 0) { |
fb2ef791 | 417 | io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); |
0965a41e | 418 | } else if (ret & BDRV_BLOCK_DATA) { |
fb2ef791 | 419 | io_bytes = MIN(io_bytes, max_io_bytes); |
e5b43573 FZ |
420 | } |
421 | ||
fb2ef791 EB |
422 | io_bytes -= io_bytes % s->granularity; |
423 | if (io_bytes < s->granularity) { | |
424 | io_bytes = s->granularity; | |
e5b43573 | 425 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { |
fb2ef791 | 426 | int64_t target_offset; |
7cfd5275 | 427 | int64_t target_bytes; |
fb2ef791 EB |
428 | bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes, |
429 | &target_offset, &target_bytes); | |
430 | if (target_offset == offset && | |
431 | target_bytes == io_bytes) { | |
e5b43573 FZ |
432 | mirror_method = ret & BDRV_BLOCK_ZERO ? |
433 | MIRROR_METHOD_ZERO : | |
434 | MIRROR_METHOD_DISCARD; | |
435 | } | |
436 | } | |
437 | ||
cf56a3c6 | 438 | while (s->in_flight >= MAX_IN_FLIGHT) { |
fb2ef791 | 439 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
cf56a3c6 DL |
440 | mirror_wait_for_io(s); |
441 | } | |
442 | ||
dbaa7b57 VSO |
443 | if (s->ret < 0) { |
444 | return 0; | |
445 | } | |
446 | ||
fb2ef791 | 447 | io_bytes = mirror_clip_bytes(s, offset, io_bytes); |
4295c5fc HR |
448 | io_bytes = mirror_perform(s, offset, io_bytes, mirror_method); |
449 | if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) { | |
450 | io_bytes_acct = 0; | |
451 | } else { | |
452 | io_bytes_acct = io_bytes; | |
e5b43573 | 453 | } |
fb2ef791 EB |
454 | assert(io_bytes); |
455 | offset += io_bytes; | |
456 | nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); | |
dee81d51 | 457 | delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct); |
dcfb3beb | 458 | } |
cc8c9d6c | 459 | return delay_ns; |
bd48bde8 | 460 | } |
b952b558 | 461 | |
402a4741 PB |
462 | static void mirror_free_init(MirrorBlockJob *s) |
463 | { | |
464 | int granularity = s->granularity; | |
465 | size_t buf_size = s->buf_size; | |
466 | uint8_t *buf = s->buf; | |
467 | ||
468 | assert(s->buf_free_count == 0); | |
469 | QSIMPLEQ_INIT(&s->buf_free); | |
470 | while (buf_size != 0) { | |
471 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
472 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
473 | s->buf_free_count++; | |
474 | buf_size -= granularity; | |
475 | buf += granularity; | |
476 | } | |
477 | } | |
478 | ||
bae8196d PB |
479 | /* This is also used for the .pause callback. There is no matching |
480 | * mirror_resume() because mirror_run() will begin iterating again | |
481 | * when the job is resumed. | |
482 | */ | |
483 | static void mirror_wait_for_all_io(MirrorBlockJob *s) | |
bd48bde8 PB |
484 | { |
485 | while (s->in_flight > 0) { | |
21cd917f | 486 | mirror_wait_for_io(s); |
bd48bde8 | 487 | } |
893f7eba PB |
488 | } |
489 | ||
5a7e7a0b SH |
490 | typedef struct { |
491 | int ret; | |
492 | } MirrorExitData; | |
493 | ||
1908a559 | 494 | static void mirror_exit(Job *job, void *opaque) |
5a7e7a0b | 495 | { |
1908a559 KW |
496 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
497 | BlockJob *bjob = &s->common; | |
5a7e7a0b SH |
498 | MirrorExitData *data = opaque; |
499 | AioContext *replace_aio_context = NULL; | |
4ef85a9c | 500 | BlockDriverState *src = s->source; |
e253f4b8 | 501 | BlockDriverState *target_bs = blk_bs(s->target); |
4ef85a9c | 502 | BlockDriverState *mirror_top_bs = s->mirror_top_bs; |
12fa4af6 | 503 | Error *local_err = NULL; |
3f09bfbc | 504 | |
2119882c PB |
505 | bdrv_release_dirty_bitmap(src, s->dirty_bitmap); |
506 | ||
3f09bfbc | 507 | /* Make sure that the source BDS doesn't go away before we called |
3d70ff53 | 508 | * job_completed(). */ |
3f09bfbc | 509 | bdrv_ref(src); |
4ef85a9c | 510 | bdrv_ref(mirror_top_bs); |
7d9fcb39 KW |
511 | bdrv_ref(target_bs); |
512 | ||
513 | /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before | |
514 | * inserting target_bs at s->to_replace, where we might not be able to get | |
63c8ef28 KW |
515 | * these permissions. |
516 | * | |
517 | * Note that blk_unref() alone doesn't necessarily drop permissions because | |
518 | * we might be running nested inside mirror_drain(), which takes an extra | |
519 | * reference, so use an explicit blk_set_perm() first. */ | |
520 | blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); | |
7d9fcb39 KW |
521 | blk_unref(s->target); |
522 | s->target = NULL; | |
4ef85a9c KW |
523 | |
524 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
525 | * required before it could become a backing file of target_bs. */ | |
526 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, | |
527 | &error_abort); | |
528 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { | |
529 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
530 | if (backing_bs(target_bs) != backing) { | |
12fa4af6 KW |
531 | bdrv_set_backing_hd(target_bs, backing, &local_err); |
532 | if (local_err) { | |
533 | error_report_err(local_err); | |
534 | data->ret = -EPERM; | |
535 | } | |
4ef85a9c KW |
536 | } |
537 | } | |
5a7e7a0b SH |
538 | |
539 | if (s->to_replace) { | |
540 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
541 | aio_context_acquire(replace_aio_context); | |
542 | } | |
543 | ||
544 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 545 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
546 | if (s->to_replace) { |
547 | to_replace = s->to_replace; | |
548 | } | |
40365552 | 549 | |
e253f4b8 KW |
550 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
551 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 552 | } |
b8804815 KW |
553 | |
554 | /* The mirror job has no requests in flight any more, but we need to | |
555 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 | 556 | bdrv_drained_begin(target_bs); |
5fe31c25 | 557 | bdrv_replace_node(to_replace, target_bs, &local_err); |
e253f4b8 | 558 | bdrv_drained_end(target_bs); |
5fe31c25 KW |
559 | if (local_err) { |
560 | error_report_err(local_err); | |
561 | data->ret = -EPERM; | |
562 | } | |
5a7e7a0b SH |
563 | } |
564 | if (s->to_replace) { | |
565 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
566 | error_free(s->replace_blocker); | |
567 | bdrv_unref(s->to_replace); | |
568 | } | |
569 | if (replace_aio_context) { | |
570 | aio_context_release(replace_aio_context); | |
571 | } | |
572 | g_free(s->replaces); | |
7d9fcb39 | 573 | bdrv_unref(target_bs); |
4ef85a9c KW |
574 | |
575 | /* Remove the mirror filter driver from the graph. Before this, get rid of | |
576 | * the blockers on the intermediate nodes so that the resulting state is | |
0bf74767 KW |
577 | * valid. Also give up permissions on mirror_top_bs->backing, which might |
578 | * block the removal. */ | |
1908a559 | 579 | block_job_remove_all_bdrv(bjob); |
c1cef672 FZ |
580 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
581 | &error_abort); | |
5fe31c25 | 582 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
4ef85a9c KW |
583 | |
584 | /* We just changed the BDS the job BB refers to (with either or both of the | |
5fe31c25 KW |
585 | * bdrv_replace_node() calls), so switch the BB back so the cleanup does |
586 | * the right thing. We don't need any permissions any more now. */ | |
1908a559 KW |
587 | blk_remove_bs(bjob->blk); |
588 | blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort); | |
589 | blk_insert_bs(bjob->blk, mirror_top_bs, &error_abort); | |
4ef85a9c | 590 | |
1266c9b9 | 591 | job_completed(job, data->ret, NULL); |
4ef85a9c | 592 | |
5a7e7a0b | 593 | g_free(data); |
176c3699 | 594 | bdrv_drained_end(src); |
4ef85a9c | 595 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 596 | bdrv_unref(src); |
5a7e7a0b SH |
597 | } |
598 | ||
49efb1f5 DL |
599 | static void mirror_throttle(MirrorBlockJob *s) |
600 | { | |
601 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
602 | ||
18bb6928 | 603 | if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) { |
49efb1f5 | 604 | s->last_pause_ns = now; |
5d43e86e | 605 | job_sleep_ns(&s->common.job, 0); |
49efb1f5 | 606 | } else { |
da01ff7f | 607 | job_pause_point(&s->common.job); |
49efb1f5 DL |
608 | } |
609 | } | |
610 | ||
c0b363ad DL |
611 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
612 | { | |
23ca459a | 613 | int64_t offset; |
c0b363ad | 614 | BlockDriverState *base = s->base; |
4ef85a9c | 615 | BlockDriverState *bs = s->source; |
c0b363ad | 616 | BlockDriverState *target_bs = blk_bs(s->target); |
23ca459a | 617 | int ret; |
51b0a488 | 618 | int64_t count; |
c0b363ad | 619 | |
b7d5062c | 620 | if (base == NULL && !bdrv_has_zero_init(target_bs)) { |
c7c2769c | 621 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
e0d7f73e | 622 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); |
c7c2769c DL |
623 | return 0; |
624 | } | |
625 | ||
90ab48eb | 626 | s->initial_zeroing_ongoing = true; |
23ca459a EB |
627 | for (offset = 0; offset < s->bdev_length; ) { |
628 | int bytes = MIN(s->bdev_length - offset, | |
629 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c7c2769c DL |
630 | |
631 | mirror_throttle(s); | |
632 | ||
daa7f2f9 | 633 | if (job_is_cancelled(&s->common.job)) { |
90ab48eb | 634 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
635 | return 0; |
636 | } | |
637 | ||
638 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
67adf4b3 EB |
639 | trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, |
640 | s->in_flight); | |
c7c2769c DL |
641 | mirror_wait_for_io(s); |
642 | continue; | |
643 | } | |
644 | ||
4295c5fc | 645 | mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO); |
23ca459a | 646 | offset += bytes; |
c7c2769c DL |
647 | } |
648 | ||
bae8196d | 649 | mirror_wait_for_all_io(s); |
90ab48eb | 650 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
651 | } |
652 | ||
c0b363ad | 653 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
23ca459a | 654 | for (offset = 0; offset < s->bdev_length; ) { |
c0b363ad | 655 | /* Just to make sure we are not exceeding int limit. */ |
23ca459a EB |
656 | int bytes = MIN(s->bdev_length - offset, |
657 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c0b363ad DL |
658 | |
659 | mirror_throttle(s); | |
660 | ||
daa7f2f9 | 661 | if (job_is_cancelled(&s->common.job)) { |
c0b363ad DL |
662 | return 0; |
663 | } | |
664 | ||
23ca459a | 665 | ret = bdrv_is_allocated_above(bs, base, offset, bytes, &count); |
c0b363ad DL |
666 | if (ret < 0) { |
667 | return ret; | |
668 | } | |
669 | ||
23ca459a | 670 | assert(count); |
b7d5062c | 671 | if (ret == 1) { |
23ca459a | 672 | bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); |
c0b363ad | 673 | } |
23ca459a | 674 | offset += count; |
c0b363ad DL |
675 | } |
676 | return 0; | |
677 | } | |
678 | ||
bdffb31d PB |
679 | /* Called when going out of the streaming phase to flush the bulk of the |
680 | * data to the medium, or just before completing. | |
681 | */ | |
682 | static int mirror_flush(MirrorBlockJob *s) | |
683 | { | |
684 | int ret = blk_flush(s->target); | |
685 | if (ret < 0) { | |
686 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
687 | s->ret = ret; | |
688 | } | |
689 | } | |
690 | return ret; | |
691 | } | |
692 | ||
893f7eba PB |
693 | static void coroutine_fn mirror_run(void *opaque) |
694 | { | |
695 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 696 | MirrorExitData *data; |
4ef85a9c | 697 | BlockDriverState *bs = s->source; |
e253f4b8 | 698 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 699 | bool need_drain = true; |
c0b363ad | 700 | int64_t length; |
b812f671 | 701 | BlockDriverInfo bdi; |
1d33936e JC |
702 | char backing_filename[2]; /* we only need 2 characters because we are only |
703 | checking for a NULL string */ | |
893f7eba | 704 | int ret = 0; |
893f7eba | 705 | |
daa7f2f9 | 706 | if (job_is_cancelled(&s->common.job)) { |
893f7eba PB |
707 | goto immediate_exit; |
708 | } | |
709 | ||
b21c7652 HR |
710 | s->bdev_length = bdrv_getlength(bs); |
711 | if (s->bdev_length < 0) { | |
712 | ret = s->bdev_length; | |
373df5b1 | 713 | goto immediate_exit; |
becc347e KW |
714 | } |
715 | ||
716 | /* Active commit must resize the base image if its size differs from the | |
717 | * active layer. */ | |
718 | if (s->base == blk_bs(s->target)) { | |
719 | int64_t base_length; | |
720 | ||
721 | base_length = blk_getlength(s->target); | |
722 | if (base_length < 0) { | |
723 | ret = base_length; | |
724 | goto immediate_exit; | |
725 | } | |
726 | ||
727 | if (s->bdev_length > base_length) { | |
3a691c50 HR |
728 | ret = blk_truncate(s->target, s->bdev_length, PREALLOC_MODE_OFF, |
729 | NULL); | |
becc347e KW |
730 | if (ret < 0) { |
731 | goto immediate_exit; | |
732 | } | |
733 | } | |
734 | } | |
735 | ||
736 | if (s->bdev_length == 0) { | |
2e1795b5 KW |
737 | /* Transition to the READY state and wait for complete. */ |
738 | job_transition_to_ready(&s->common.job); | |
9e48b025 | 739 | s->synced = true; |
daa7f2f9 | 740 | while (!job_is_cancelled(&s->common.job) && !s->should_complete) { |
198c49cc | 741 | job_yield(&s->common.job); |
9e48b025 | 742 | } |
daa7f2f9 | 743 | s->common.job.cancelled = false; |
9e48b025 | 744 | goto immediate_exit; |
893f7eba PB |
745 | } |
746 | ||
b21c7652 | 747 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
748 | s->in_flight_bitmap = bitmap_new(length); |
749 | ||
b812f671 PB |
750 | /* If we have no backing file yet in the destination, we cannot let |
751 | * the destination do COW. Instead, we copy sectors around the | |
752 | * dirty data if needed. We need a bitmap to do that. | |
753 | */ | |
e253f4b8 | 754 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 755 | sizeof(backing_filename)); |
e253f4b8 | 756 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
b436982f EB |
757 | s->target_cluster_size = bdi.cluster_size; |
758 | } else { | |
759 | s->target_cluster_size = BDRV_SECTOR_SIZE; | |
e5b43573 | 760 | } |
b436982f EB |
761 | if (backing_filename[0] && !target_bs->backing && |
762 | s->granularity < s->target_cluster_size) { | |
763 | s->buf_size = MAX(s->buf_size, s->target_cluster_size); | |
e5b43573 | 764 | s->cow_bitmap = bitmap_new(length); |
b812f671 | 765 | } |
e253f4b8 | 766 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 767 | |
7504edf4 KW |
768 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
769 | if (s->buf == NULL) { | |
770 | ret = -ENOMEM; | |
771 | goto immediate_exit; | |
772 | } | |
773 | ||
402a4741 | 774 | mirror_free_init(s); |
893f7eba | 775 | |
49efb1f5 | 776 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 777 | if (!s->is_none_mode) { |
c0b363ad | 778 | ret = mirror_dirty_init(s); |
daa7f2f9 | 779 | if (ret < 0 || job_is_cancelled(&s->common.job)) { |
c0b363ad | 780 | goto immediate_exit; |
893f7eba PB |
781 | } |
782 | } | |
783 | ||
dc162c8e | 784 | assert(!s->dbi); |
715a74d8 | 785 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); |
893f7eba | 786 | for (;;) { |
cc8c9d6c | 787 | uint64_t delay_ns = 0; |
49efb1f5 | 788 | int64_t cnt, delta; |
893f7eba PB |
789 | bool should_complete; |
790 | ||
bd48bde8 PB |
791 | if (s->ret < 0) { |
792 | ret = s->ret; | |
793 | goto immediate_exit; | |
794 | } | |
795 | ||
da01ff7f | 796 | job_pause_point(&s->common.job); |
565ac01f | 797 | |
20dca810 | 798 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
05df8a6a KW |
799 | /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is |
800 | * the number of bytes currently being processed; together those are | |
801 | * the current remaining operation length */ | |
30a5c887 | 802 | job_progress_set_remaining(&s->common.job, s->bytes_in_flight + cnt); |
bd48bde8 PB |
803 | |
804 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 805 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
18bb6928 KW |
806 | * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is |
807 | * an error, or when the source is clean, whichever comes first. */ | |
49efb1f5 | 808 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
18bb6928 | 809 | if (delta < BLOCK_JOB_SLICE_TIME && |
bd48bde8 | 810 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 811 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 | 812 | (cnt == 0 && s->in_flight > 0)) { |
9a46dba7 | 813 | trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); |
21cd917f | 814 | mirror_wait_for_io(s); |
bd48bde8 PB |
815 | continue; |
816 | } else if (cnt != 0) { | |
cc8c9d6c | 817 | delay_ns = mirror_iteration(s); |
893f7eba | 818 | } |
893f7eba PB |
819 | } |
820 | ||
821 | should_complete = false; | |
bd48bde8 | 822 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 823 | trace_mirror_before_flush(s); |
bdffb31d PB |
824 | if (!s->synced) { |
825 | if (mirror_flush(s) < 0) { | |
826 | /* Go check s->ret. */ | |
827 | continue; | |
b952b558 | 828 | } |
b952b558 PB |
829 | /* We're out of the streaming phase. From now on, if the job |
830 | * is cancelled we will actually complete all pending I/O and | |
831 | * report completion. This way, block-job-cancel will leave | |
832 | * the target in a consistent state. | |
833 | */ | |
2e1795b5 | 834 | job_transition_to_ready(&s->common.job); |
bdffb31d | 835 | s->synced = true; |
d63ffd87 | 836 | } |
bdffb31d PB |
837 | |
838 | should_complete = s->should_complete || | |
daa7f2f9 | 839 | job_is_cancelled(&s->common.job); |
bdffb31d | 840 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
893f7eba PB |
841 | } |
842 | ||
843 | if (cnt == 0 && should_complete) { | |
844 | /* The dirty bitmap is not updated while operations are pending. | |
845 | * If we're about to exit, wait for pending operations before | |
846 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
847 | * source has dirty data to copy! | |
848 | * | |
849 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
850 | * mirror_populate runs, so pause it now. Before deciding |
851 | * whether to switch to target check one last time if I/O has | |
852 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba | 853 | */ |
9a46dba7 | 854 | trace_mirror_before_drain(s, cnt); |
9a0cec66 PB |
855 | |
856 | bdrv_drained_begin(bs); | |
20dca810 | 857 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 858 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 PB |
859 | bdrv_drained_end(bs); |
860 | continue; | |
861 | } | |
862 | ||
863 | /* The two disks are in sync. Exit and report successful | |
864 | * completion. | |
865 | */ | |
866 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
daa7f2f9 | 867 | s->common.job.cancelled = false; |
9a0cec66 PB |
868 | need_drain = false; |
869 | break; | |
893f7eba PB |
870 | } |
871 | ||
872 | ret = 0; | |
ddc4115e SH |
873 | |
874 | if (s->synced && !should_complete) { | |
18bb6928 KW |
875 | delay_ns = (s->in_flight == 0 && |
876 | cnt == 0 ? BLOCK_JOB_SLICE_TIME : 0); | |
ddc4115e | 877 | } |
9a46dba7 | 878 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
5d43e86e | 879 | job_sleep_ns(&s->common.job, delay_ns); |
daa7f2f9 | 880 | if (job_is_cancelled(&s->common.job) && |
004e95df | 881 | (!s->synced || s->common.job.force_cancel)) |
eb36639f | 882 | { |
b76e4458 | 883 | break; |
893f7eba | 884 | } |
49efb1f5 | 885 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
886 | } |
887 | ||
888 | immediate_exit: | |
bd48bde8 PB |
889 | if (s->in_flight > 0) { |
890 | /* We get here only if something went wrong. Either the job failed, | |
891 | * or it was cancelled prematurely so that we do not guarantee that | |
892 | * the target is a copy of the source. | |
893 | */ | |
004e95df | 894 | assert(ret < 0 || ((s->common.job.force_cancel || !s->synced) && |
daa7f2f9 | 895 | job_is_cancelled(&s->common.job))); |
9a0cec66 | 896 | assert(need_drain); |
bae8196d | 897 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
898 | } |
899 | ||
900 | assert(s->in_flight == 0); | |
7191bf31 | 901 | qemu_vfree(s->buf); |
b812f671 | 902 | g_free(s->cow_bitmap); |
402a4741 | 903 | g_free(s->in_flight_bitmap); |
dc162c8e | 904 | bdrv_dirty_iter_free(s->dbi); |
5a7e7a0b SH |
905 | |
906 | data = g_malloc(sizeof(*data)); | |
907 | data->ret = ret; | |
9a0cec66 PB |
908 | |
909 | if (need_drain) { | |
910 | bdrv_drained_begin(bs); | |
911 | } | |
1908a559 | 912 | job_defer_to_main_loop(&s->common.job, mirror_exit, data); |
893f7eba PB |
913 | } |
914 | ||
3453d972 | 915 | static void mirror_complete(Job *job, Error **errp) |
d63ffd87 | 916 | { |
3453d972 | 917 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
4ef85a9c | 918 | BlockDriverState *target; |
274fccee | 919 | |
274fccee | 920 | target = blk_bs(s->target); |
d63ffd87 | 921 | |
d63ffd87 | 922 | if (!s->synced) { |
9df229c3 | 923 | error_setg(errp, "The active block job '%s' cannot be completed", |
3453d972 | 924 | job->id); |
d63ffd87 PB |
925 | return; |
926 | } | |
927 | ||
274fccee HR |
928 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
929 | int ret; | |
930 | ||
931 | assert(!target->backing); | |
932 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
933 | if (ret < 0) { | |
934 | return; | |
935 | } | |
936 | } | |
937 | ||
15d67298 | 938 | /* block all operations on to_replace bs */ |
09158f00 | 939 | if (s->replaces) { |
5a7e7a0b SH |
940 | AioContext *replace_aio_context; |
941 | ||
e12f3784 | 942 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 943 | if (!s->to_replace) { |
e12f3784 | 944 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
945 | return; |
946 | } | |
947 | ||
5a7e7a0b SH |
948 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
949 | aio_context_acquire(replace_aio_context); | |
950 | ||
4ef85a9c KW |
951 | /* TODO Translate this into permission system. Current definition of |
952 | * GRAPH_MOD would require to request it for the parents; they might | |
953 | * not even be BlockDriverStates, however, so a BdrvChild can't address | |
954 | * them. May need redefinition of GRAPH_MOD. */ | |
09158f00 BC |
955 | error_setg(&s->replace_blocker, |
956 | "block device is in use by block-job-complete"); | |
957 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
958 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
959 | |
960 | aio_context_release(replace_aio_context); | |
09158f00 BC |
961 | } |
962 | ||
d63ffd87 | 963 | s->should_complete = true; |
3d70ff53 | 964 | job_enter(job); |
d63ffd87 PB |
965 | } |
966 | ||
da01ff7f | 967 | static void mirror_pause(Job *job) |
565ac01f | 968 | { |
da01ff7f | 969 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
565ac01f | 970 | |
bae8196d | 971 | mirror_wait_for_all_io(s); |
565ac01f SH |
972 | } |
973 | ||
89bd0305 KW |
974 | static bool mirror_drained_poll(BlockJob *job) |
975 | { | |
976 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
977 | return !!s->in_flight; | |
978 | } | |
979 | ||
565ac01f SH |
980 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) |
981 | { | |
982 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
983 | ||
984 | blk_set_aio_context(s->target, new_context); | |
985 | } | |
986 | ||
bae8196d PB |
987 | static void mirror_drain(BlockJob *job) |
988 | { | |
989 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
990 | ||
991 | /* Need to keep a reference in case blk_drain triggers execution | |
992 | * of mirror_complete... | |
993 | */ | |
994 | if (s->target) { | |
995 | BlockBackend *target = s->target; | |
996 | blk_ref(target); | |
997 | blk_drain(target); | |
998 | blk_unref(target); | |
999 | } | |
1000 | } | |
1001 | ||
3fc4b10a | 1002 | static const BlockJobDriver mirror_job_driver = { |
33e9e9bd KW |
1003 | .job_driver = { |
1004 | .instance_size = sizeof(MirrorBlockJob), | |
252291ea | 1005 | .job_type = JOB_TYPE_MIRROR, |
80fa2c75 | 1006 | .free = block_job_free, |
b15de828 | 1007 | .user_resume = block_job_user_resume, |
b69f777d | 1008 | .drain = block_job_drain, |
da01ff7f KW |
1009 | .start = mirror_run, |
1010 | .pause = mirror_pause, | |
3453d972 | 1011 | .complete = mirror_complete, |
33e9e9bd | 1012 | }, |
89bd0305 | 1013 | .drained_poll = mirror_drained_poll, |
565ac01f | 1014 | .attached_aio_context = mirror_attached_aio_context, |
bae8196d | 1015 | .drain = mirror_drain, |
893f7eba PB |
1016 | }; |
1017 | ||
03544a6e | 1018 | static const BlockJobDriver commit_active_job_driver = { |
33e9e9bd KW |
1019 | .job_driver = { |
1020 | .instance_size = sizeof(MirrorBlockJob), | |
252291ea | 1021 | .job_type = JOB_TYPE_COMMIT, |
80fa2c75 | 1022 | .free = block_job_free, |
b15de828 | 1023 | .user_resume = block_job_user_resume, |
b69f777d | 1024 | .drain = block_job_drain, |
da01ff7f KW |
1025 | .start = mirror_run, |
1026 | .pause = mirror_pause, | |
3453d972 | 1027 | .complete = mirror_complete, |
33e9e9bd | 1028 | }, |
89bd0305 | 1029 | .drained_poll = mirror_drained_poll, |
565ac01f | 1030 | .attached_aio_context = mirror_attached_aio_context, |
bae8196d | 1031 | .drain = mirror_drain, |
03544a6e FZ |
1032 | }; |
1033 | ||
4ef85a9c KW |
1034 | static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, |
1035 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1036 | { | |
1037 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1038 | } | |
1039 | ||
1040 | static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, | |
1041 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1042 | { | |
1043 | return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1044 | } | |
1045 | ||
1046 | static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) | |
1047 | { | |
ce960aa9 VSO |
1048 | if (bs->backing == NULL) { |
1049 | /* we can be here after failed bdrv_append in mirror_start_job */ | |
1050 | return 0; | |
1051 | } | |
4ef85a9c KW |
1052 | return bdrv_co_flush(bs->backing->bs); |
1053 | } | |
1054 | ||
4ef85a9c | 1055 | static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 1056 | int64_t offset, int bytes, BdrvRequestFlags flags) |
4ef85a9c | 1057 | { |
f5a5ca79 | 1058 | return bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); |
4ef85a9c KW |
1059 | } |
1060 | ||
1061 | static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, | |
f5a5ca79 | 1062 | int64_t offset, int bytes) |
4ef85a9c | 1063 | { |
f5a5ca79 | 1064 | return bdrv_co_pdiscard(bs->backing->bs, offset, bytes); |
4ef85a9c KW |
1065 | } |
1066 | ||
fd4a6493 KW |
1067 | static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) |
1068 | { | |
18775ff3 VSO |
1069 | if (bs->backing == NULL) { |
1070 | /* we can be here after failed bdrv_attach_child in | |
1071 | * bdrv_set_backing_hd */ | |
1072 | return; | |
1073 | } | |
fd4a6493 KW |
1074 | bdrv_refresh_filename(bs->backing->bs); |
1075 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
1076 | bs->backing->bs->filename); | |
1077 | } | |
1078 | ||
4ef85a9c KW |
1079 | static void bdrv_mirror_top_close(BlockDriverState *bs) |
1080 | { | |
1081 | } | |
1082 | ||
1083 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, | |
1084 | const BdrvChildRole *role, | |
e0995dc3 | 1085 | BlockReopenQueue *reopen_queue, |
4ef85a9c KW |
1086 | uint64_t perm, uint64_t shared, |
1087 | uint64_t *nperm, uint64_t *nshared) | |
1088 | { | |
1089 | /* Must be able to forward guest writes to the real image */ | |
1090 | *nperm = 0; | |
1091 | if (perm & BLK_PERM_WRITE) { | |
1092 | *nperm |= BLK_PERM_WRITE; | |
1093 | } | |
1094 | ||
1095 | *nshared = BLK_PERM_ALL; | |
1096 | } | |
1097 | ||
1098 | /* Dummy node that provides consistent read to its users without requiring it | |
1099 | * from its backing file and that allows writes on the backing file chain. */ | |
1100 | static BlockDriver bdrv_mirror_top = { | |
1101 | .format_name = "mirror_top", | |
1102 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1103 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1104 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1105 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1106 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
3e4d0e72 | 1107 | .bdrv_co_block_status = bdrv_co_block_status_from_backing, |
fd4a6493 | 1108 | .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, |
4ef85a9c KW |
1109 | .bdrv_close = bdrv_mirror_top_close, |
1110 | .bdrv_child_perm = bdrv_mirror_top_child_perm, | |
1111 | }; | |
1112 | ||
71aa9867 | 1113 | static void mirror_start_job(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1114 | int creation_flags, BlockDriverState *target, |
1115 | const char *replaces, int64_t speed, | |
1116 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1117 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
1118 | BlockdevOnError on_source_error, |
1119 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1120 | bool unmap, |
097310b5 | 1121 | BlockCompletionFunc *cb, |
51ccfa2d | 1122 | void *opaque, |
09158f00 | 1123 | const BlockJobDriver *driver, |
b49f7ead | 1124 | bool is_none_mode, BlockDriverState *base, |
51ccfa2d | 1125 | bool auto_complete, const char *filter_node_name, |
045a2f82 | 1126 | bool is_mirror, |
51ccfa2d | 1127 | Error **errp) |
893f7eba PB |
1128 | { |
1129 | MirrorBlockJob *s; | |
4ef85a9c KW |
1130 | BlockDriverState *mirror_top_bs; |
1131 | bool target_graph_mod; | |
1132 | bool target_is_backing; | |
b2c2832c | 1133 | Error *local_err = NULL; |
d7086422 | 1134 | int ret; |
893f7eba | 1135 | |
eee13dfe | 1136 | if (granularity == 0) { |
341ebc2f | 1137 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1138 | } |
1139 | ||
31826642 | 1140 | assert(is_power_of_2(granularity)); |
eee13dfe | 1141 | |
48ac0a4d WC |
1142 | if (buf_size < 0) { |
1143 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
1144 | return; | |
1145 | } | |
1146 | ||
1147 | if (buf_size == 0) { | |
1148 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
1149 | } | |
5bc361b8 | 1150 | |
4ef85a9c KW |
1151 | /* In the case of active commit, add dummy driver to provide consistent |
1152 | * reads on the top, while disabling it in the intermediate nodes, and make | |
1153 | * the backing chain writable. */ | |
6cdbceb1 KW |
1154 | mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, |
1155 | BDRV_O_RDWR, errp); | |
4ef85a9c KW |
1156 | if (mirror_top_bs == NULL) { |
1157 | return; | |
1158 | } | |
d3c8c674 KW |
1159 | if (!filter_node_name) { |
1160 | mirror_top_bs->implicit = true; | |
1161 | } | |
4ef85a9c | 1162 | mirror_top_bs->total_sectors = bs->total_sectors; |
228345bf HR |
1163 | mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED; |
1164 | mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED; | |
19dd29e8 | 1165 | bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); |
4ef85a9c KW |
1166 | |
1167 | /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep | |
7a25fcd0 | 1168 | * it alive until block_job_create() succeeds even if bs has no parent. */ |
4ef85a9c KW |
1169 | bdrv_ref(mirror_top_bs); |
1170 | bdrv_drained_begin(bs); | |
b2c2832c | 1171 | bdrv_append(mirror_top_bs, bs, &local_err); |
4ef85a9c KW |
1172 | bdrv_drained_end(bs); |
1173 | ||
b2c2832c KW |
1174 | if (local_err) { |
1175 | bdrv_unref(mirror_top_bs); | |
1176 | error_propagate(errp, local_err); | |
1177 | return; | |
1178 | } | |
1179 | ||
4ef85a9c | 1180 | /* Make sure that the source is not resized while the job is running */ |
75859b94 | 1181 | s = block_job_create(job_id, driver, NULL, mirror_top_bs, |
4ef85a9c KW |
1182 | BLK_PERM_CONSISTENT_READ, |
1183 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
1184 | BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, | |
c6cc12bf | 1185 | creation_flags, cb, opaque, errp); |
893f7eba | 1186 | if (!s) { |
4ef85a9c | 1187 | goto fail; |
893f7eba | 1188 | } |
7a25fcd0 HR |
1189 | /* The block job now has a reference to this node */ |
1190 | bdrv_unref(mirror_top_bs); | |
1191 | ||
4ef85a9c KW |
1192 | s->source = bs; |
1193 | s->mirror_top_bs = mirror_top_bs; | |
1194 | ||
1195 | /* No resize for the target either; while the mirror is still running, a | |
1196 | * consistent read isn't necessarily possible. We could possibly allow | |
1197 | * writes and graph modifications, though it would likely defeat the | |
1198 | * purpose of a mirror, so leave them blocked for now. | |
1199 | * | |
1200 | * In the case of active commit, things look a bit different, though, | |
1201 | * because the target is an already populated backing file in active use. | |
1202 | * We can allow anything except resize there.*/ | |
1203 | target_is_backing = bdrv_chain_contains(bs, target); | |
1204 | target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); | |
1205 | s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | | |
1206 | (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), | |
1207 | BLK_PERM_WRITE_UNCHANGED | | |
1208 | (target_is_backing ? BLK_PERM_CONSISTENT_READ | | |
1209 | BLK_PERM_WRITE | | |
1210 | BLK_PERM_GRAPH_MOD : 0)); | |
d7086422 KW |
1211 | ret = blk_insert_bs(s->target, target, errp); |
1212 | if (ret < 0) { | |
4ef85a9c | 1213 | goto fail; |
d7086422 | 1214 | } |
045a2f82 FZ |
1215 | if (is_mirror) { |
1216 | /* XXX: Mirror target could be a NBD server of target QEMU in the case | |
1217 | * of non-shared block migration. To allow migration completion, we | |
1218 | * have to allow "inactivate" of the target BB. When that happens, we | |
1219 | * know the job is drained, and the vcpus are stopped, so no write | |
1220 | * operation will be performed. Block layer already has assertions to | |
1221 | * ensure that. */ | |
1222 | blk_set_force_allow_inactivate(s->target); | |
1223 | } | |
e253f4b8 | 1224 | |
09158f00 | 1225 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1226 | s->on_source_error = on_source_error; |
1227 | s->on_target_error = on_target_error; | |
03544a6e | 1228 | s->is_none_mode = is_none_mode; |
274fccee | 1229 | s->backing_mode = backing_mode; |
5bc361b8 | 1230 | s->base = base; |
eee13dfe | 1231 | s->granularity = granularity; |
48ac0a4d | 1232 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1233 | s->unmap = unmap; |
b49f7ead WC |
1234 | if (auto_complete) { |
1235 | s->should_complete = true; | |
1236 | } | |
b812f671 | 1237 | |
0db6e54a | 1238 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1239 | if (!s->dirty_bitmap) { |
88f9d1b3 | 1240 | goto fail; |
b8afb520 | 1241 | } |
10f3cd15 | 1242 | |
4ef85a9c | 1243 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1244 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1245 | &error_abort); | |
1246 | ||
f3ede4b0 AG |
1247 | /* In commit_active_start() all intermediate nodes disappear, so |
1248 | * any jobs in them must be blocked */ | |
4ef85a9c | 1249 | if (target_is_backing) { |
f3ede4b0 AG |
1250 | BlockDriverState *iter; |
1251 | for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { | |
4ef85a9c KW |
1252 | /* XXX BLK_PERM_WRITE needs to be allowed so we don't block |
1253 | * ourselves at s->base (if writes are blocked for a node, they are | |
1254 | * also blocked for its backing file). The other options would be a | |
1255 | * second filter driver above s->base (== target). */ | |
1256 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, | |
1257 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, | |
1258 | errp); | |
1259 | if (ret < 0) { | |
1260 | goto fail; | |
1261 | } | |
f3ede4b0 AG |
1262 | } |
1263 | } | |
10f3cd15 | 1264 | |
5ccac6f1 | 1265 | trace_mirror_start(bs, s, opaque); |
da01ff7f | 1266 | job_start(&s->common.job); |
4ef85a9c KW |
1267 | return; |
1268 | ||
1269 | fail: | |
1270 | if (s) { | |
7a25fcd0 HR |
1271 | /* Make sure this BDS does not go away until we have completed the graph |
1272 | * changes below */ | |
1273 | bdrv_ref(mirror_top_bs); | |
1274 | ||
4ef85a9c KW |
1275 | g_free(s->replaces); |
1276 | blk_unref(s->target); | |
4ad35181 | 1277 | job_early_fail(&s->common.job); |
4ef85a9c KW |
1278 | } |
1279 | ||
c1cef672 FZ |
1280 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
1281 | &error_abort); | |
5fe31c25 | 1282 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
7a25fcd0 HR |
1283 | |
1284 | bdrv_unref(mirror_top_bs); | |
893f7eba | 1285 | } |
03544a6e | 1286 | |
71aa9867 AG |
1287 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1288 | BlockDriverState *target, const char *replaces, | |
5fba6c0e | 1289 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
1290 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
1291 | BlockdevOnError on_source_error, | |
03544a6e | 1292 | BlockdevOnError on_target_error, |
6cdbceb1 | 1293 | bool unmap, const char *filter_node_name, Error **errp) |
03544a6e FZ |
1294 | { |
1295 | bool is_none_mode; | |
1296 | BlockDriverState *base; | |
1297 | ||
4b80ab2b JS |
1298 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
1299 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
1300 | return; |
1301 | } | |
03544a6e | 1302 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 1303 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
bb02b65c | 1304 | mirror_start_job(job_id, bs, JOB_DEFAULT, target, replaces, |
274fccee | 1305 | speed, granularity, buf_size, backing_mode, |
51ccfa2d | 1306 | on_source_error, on_target_error, unmap, NULL, NULL, |
6cdbceb1 | 1307 | &mirror_job_driver, is_none_mode, base, false, |
045a2f82 | 1308 | filter_node_name, true, errp); |
03544a6e FZ |
1309 | } |
1310 | ||
fd62c609 | 1311 | void commit_active_start(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1312 | BlockDriverState *base, int creation_flags, |
1313 | int64_t speed, BlockdevOnError on_error, | |
0db832f4 | 1314 | const char *filter_node_name, |
78bbd910 FZ |
1315 | BlockCompletionFunc *cb, void *opaque, |
1316 | bool auto_complete, Error **errp) | |
03544a6e | 1317 | { |
4da83585 | 1318 | int orig_base_flags; |
cc67f4d1 | 1319 | Error *local_err = NULL; |
4da83585 JC |
1320 | |
1321 | orig_base_flags = bdrv_get_flags(base); | |
1322 | ||
20a63d2c FZ |
1323 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
1324 | return; | |
1325 | } | |
4da83585 | 1326 | |
47970dfb | 1327 | mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
71aa9867 | 1328 | MIRROR_LEAVE_BACKING_CHAIN, |
51ccfa2d | 1329 | on_error, on_error, true, cb, opaque, |
6cdbceb1 | 1330 | &commit_active_job_driver, false, base, auto_complete, |
045a2f82 | 1331 | filter_node_name, false, &local_err); |
0fb6395c | 1332 | if (local_err) { |
cc67f4d1 | 1333 | error_propagate(errp, local_err); |
4da83585 JC |
1334 | goto error_restore_flags; |
1335 | } | |
1336 | ||
1337 | return; | |
1338 | ||
1339 | error_restore_flags: | |
1340 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1341 | * the original error */ | |
1342 | bdrv_reopen(base, orig_base_flags, NULL); | |
1343 | return; | |
03544a6e | 1344 | } |