]>
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" |
12aa4082 | 16 | #include "qemu/coroutine.h" |
1181e19a | 17 | #include "qemu/range.h" |
893f7eba | 18 | #include "trace.h" |
c87621ea | 19 | #include "block/blockjob_int.h" |
737e150e | 20 | #include "block/block_int.h" |
373340b2 | 21 | #include "sysemu/block-backend.h" |
da34e65c | 22 | #include "qapi/error.h" |
cc7a8ea7 | 23 | #include "qapi/qmp/qerror.h" |
893f7eba | 24 | #include "qemu/ratelimit.h" |
b812f671 | 25 | #include "qemu/bitmap.h" |
893f7eba | 26 | |
402a4741 | 27 | #define MAX_IN_FLIGHT 16 |
b436982f EB |
28 | #define MAX_IO_BYTES (1 << 20) /* 1 Mb */ |
29 | #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES) | |
402a4741 PB |
30 | |
31 | /* The mirroring buffer is a list of granularity-sized chunks. | |
32 | * Free chunks are organized in a list. | |
33 | */ | |
34 | typedef struct MirrorBuffer { | |
35 | QSIMPLEQ_ENTRY(MirrorBuffer) next; | |
36 | } MirrorBuffer; | |
893f7eba | 37 | |
12aa4082 HR |
38 | typedef struct MirrorOp MirrorOp; |
39 | ||
893f7eba PB |
40 | typedef struct MirrorBlockJob { |
41 | BlockJob common; | |
e253f4b8 | 42 | BlockBackend *target; |
4ef85a9c | 43 | BlockDriverState *mirror_top_bs; |
5bc361b8 | 44 | BlockDriverState *base; |
3f072a7f | 45 | BlockDriverState *base_overlay; |
4ef85a9c | 46 | |
09158f00 BC |
47 | /* The name of the graph node to replace */ |
48 | char *replaces; | |
49 | /* The BDS to replace */ | |
50 | BlockDriverState *to_replace; | |
51 | /* Used to block operations on the drive-mirror-replace target */ | |
52 | Error *replace_blocker; | |
03544a6e | 53 | bool is_none_mode; |
274fccee | 54 | BlockMirrorBackingMode backing_mode; |
cdf3bc93 HR |
55 | /* Whether the target image requires explicit zero-initialization */ |
56 | bool zero_target; | |
d06107ad | 57 | MirrorCopyMode copy_mode; |
b952b558 | 58 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 | 59 | bool synced; |
d06107ad HR |
60 | /* Set when the target is synced (dirty bitmap is clean, nothing |
61 | * in flight) and the job is running in active mode */ | |
62 | bool actively_synced; | |
d63ffd87 | 63 | bool should_complete; |
eee13dfe | 64 | int64_t granularity; |
b812f671 | 65 | size_t buf_size; |
b21c7652 | 66 | int64_t bdev_length; |
b812f671 | 67 | unsigned long *cow_bitmap; |
e4654d2d | 68 | BdrvDirtyBitmap *dirty_bitmap; |
dc162c8e | 69 | BdrvDirtyBitmapIter *dbi; |
893f7eba | 70 | uint8_t *buf; |
402a4741 PB |
71 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
72 | int buf_free_count; | |
bd48bde8 | 73 | |
49efb1f5 | 74 | uint64_t last_pause_ns; |
402a4741 | 75 | unsigned long *in_flight_bitmap; |
bd48bde8 | 76 | int in_flight; |
b436982f | 77 | int64_t bytes_in_flight; |
b58deb34 | 78 | QTAILQ_HEAD(, MirrorOp) ops_in_flight; |
bd48bde8 | 79 | int ret; |
0fc9f8ea | 80 | bool unmap; |
b436982f | 81 | int target_cluster_size; |
e5b43573 | 82 | int max_iov; |
90ab48eb | 83 | bool initial_zeroing_ongoing; |
d06107ad | 84 | int in_active_write_counter; |
737efc1e | 85 | bool prepared; |
5e771752 | 86 | bool in_drain; |
893f7eba PB |
87 | } MirrorBlockJob; |
88 | ||
429076e8 HR |
89 | typedef struct MirrorBDSOpaque { |
90 | MirrorBlockJob *job; | |
f94dc3b4 | 91 | bool stop; |
53431b90 | 92 | bool is_commit; |
429076e8 HR |
93 | } MirrorBDSOpaque; |
94 | ||
12aa4082 | 95 | struct MirrorOp { |
bd48bde8 PB |
96 | MirrorBlockJob *s; |
97 | QEMUIOVector qiov; | |
b436982f EB |
98 | int64_t offset; |
99 | uint64_t bytes; | |
2e1990b2 HR |
100 | |
101 | /* The pointee is set by mirror_co_read(), mirror_co_zero(), and | |
102 | * mirror_co_discard() before yielding for the first time */ | |
103 | int64_t *bytes_handled; | |
12aa4082 | 104 | |
1181e19a | 105 | bool is_pseudo_op; |
d06107ad | 106 | bool is_active_write; |
ce8cabbd | 107 | bool is_in_flight; |
12aa4082 | 108 | CoQueue waiting_requests; |
eed325b9 | 109 | Coroutine *co; |
d44dae1a | 110 | MirrorOp *waiting_for_op; |
12aa4082 HR |
111 | |
112 | QTAILQ_ENTRY(MirrorOp) next; | |
113 | }; | |
bd48bde8 | 114 | |
4295c5fc HR |
115 | typedef enum MirrorMethod { |
116 | MIRROR_METHOD_COPY, | |
117 | MIRROR_METHOD_ZERO, | |
118 | MIRROR_METHOD_DISCARD, | |
119 | } MirrorMethod; | |
120 | ||
b952b558 PB |
121 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
122 | int error) | |
123 | { | |
124 | s->synced = false; | |
d06107ad | 125 | s->actively_synced = false; |
b952b558 | 126 | if (read) { |
81e254dc KW |
127 | return block_job_error_action(&s->common, s->on_source_error, |
128 | true, error); | |
b952b558 | 129 | } else { |
81e254dc KW |
130 | return block_job_error_action(&s->common, s->on_target_error, |
131 | false, error); | |
b952b558 PB |
132 | } |
133 | } | |
134 | ||
1181e19a HR |
135 | static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self, |
136 | MirrorBlockJob *s, | |
137 | uint64_t offset, | |
138 | uint64_t bytes) | |
139 | { | |
140 | uint64_t self_start_chunk = offset / s->granularity; | |
141 | uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); | |
142 | uint64_t self_nb_chunks = self_end_chunk - self_start_chunk; | |
143 | ||
144 | while (find_next_bit(s->in_flight_bitmap, self_end_chunk, | |
145 | self_start_chunk) < self_end_chunk && | |
146 | s->ret >= 0) | |
147 | { | |
148 | MirrorOp *op; | |
149 | ||
150 | QTAILQ_FOREACH(op, &s->ops_in_flight, next) { | |
151 | uint64_t op_start_chunk = op->offset / s->granularity; | |
152 | uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes, | |
153 | s->granularity) - | |
154 | op_start_chunk; | |
155 | ||
156 | if (op == self) { | |
157 | continue; | |
158 | } | |
159 | ||
160 | if (ranges_overlap(self_start_chunk, self_nb_chunks, | |
161 | op_start_chunk, op_nb_chunks)) | |
162 | { | |
d44dae1a VSO |
163 | /* |
164 | * If the operation is already (indirectly) waiting for us, or | |
165 | * will wait for us as soon as it wakes up, then just go on | |
166 | * (instead of producing a deadlock in the former case). | |
167 | */ | |
168 | if (op->waiting_for_op) { | |
169 | continue; | |
170 | } | |
171 | ||
172 | self->waiting_for_op = op; | |
1181e19a | 173 | qemu_co_queue_wait(&op->waiting_requests, NULL); |
d44dae1a | 174 | self->waiting_for_op = NULL; |
1181e19a HR |
175 | break; |
176 | } | |
177 | } | |
178 | } | |
179 | } | |
180 | ||
2e1990b2 | 181 | static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) |
bd48bde8 PB |
182 | { |
183 | MirrorBlockJob *s = op->s; | |
402a4741 | 184 | struct iovec *iov; |
bd48bde8 | 185 | int64_t chunk_num; |
b436982f | 186 | int i, nb_chunks; |
bd48bde8 | 187 | |
b436982f | 188 | trace_mirror_iteration_done(s, op->offset, op->bytes, ret); |
bd48bde8 PB |
189 | |
190 | s->in_flight--; | |
b436982f | 191 | s->bytes_in_flight -= op->bytes; |
402a4741 PB |
192 | iov = op->qiov.iov; |
193 | for (i = 0; i < op->qiov.niov; i++) { | |
194 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
195 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
196 | s->buf_free_count++; | |
197 | } | |
198 | ||
b436982f EB |
199 | chunk_num = op->offset / s->granularity; |
200 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
12aa4082 | 201 | |
402a4741 | 202 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
12aa4082 | 203 | QTAILQ_REMOVE(&s->ops_in_flight, op, next); |
b21c7652 HR |
204 | if (ret >= 0) { |
205 | if (s->cow_bitmap) { | |
206 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
207 | } | |
90ab48eb | 208 | if (!s->initial_zeroing_ongoing) { |
30a5c887 | 209 | job_progress_update(&s->common.job, op->bytes); |
90ab48eb | 210 | } |
bd48bde8 | 211 | } |
6df3bf8e | 212 | qemu_iovec_destroy(&op->qiov); |
7b770c72 | 213 | |
12aa4082 HR |
214 | qemu_co_queue_restart_all(&op->waiting_requests); |
215 | g_free(op); | |
bd48bde8 PB |
216 | } |
217 | ||
2e1990b2 | 218 | static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret) |
bd48bde8 | 219 | { |
bd48bde8 | 220 | MirrorBlockJob *s = op->s; |
b9e413dd | 221 | |
bd48bde8 | 222 | if (ret < 0) { |
bd48bde8 PB |
223 | BlockErrorAction action; |
224 | ||
e0d7f73e | 225 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 226 | action = mirror_error_action(s, false, -ret); |
a589569f | 227 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
228 | s->ret = ret; |
229 | } | |
230 | } | |
d12ade57 | 231 | |
bd48bde8 PB |
232 | mirror_iteration_done(op, ret); |
233 | } | |
234 | ||
2e1990b2 | 235 | static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret) |
bd48bde8 | 236 | { |
bd48bde8 | 237 | MirrorBlockJob *s = op->s; |
b9e413dd | 238 | |
bd48bde8 | 239 | if (ret < 0) { |
bd48bde8 PB |
240 | BlockErrorAction action; |
241 | ||
e0d7f73e | 242 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 243 | action = mirror_error_action(s, true, -ret); |
a589569f | 244 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
245 | s->ret = ret; |
246 | } | |
247 | ||
248 | mirror_iteration_done(op, ret); | |
d12ade57 | 249 | return; |
bd48bde8 | 250 | } |
d12ade57 VSO |
251 | |
252 | ret = blk_co_pwritev(s->target, op->offset, op->qiov.size, &op->qiov, 0); | |
253 | mirror_write_complete(op, ret); | |
bd48bde8 PB |
254 | } |
255 | ||
782d97ef EB |
256 | /* Clip bytes relative to offset to not exceed end-of-file */ |
257 | static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, | |
258 | int64_t offset, | |
259 | int64_t bytes) | |
260 | { | |
261 | return MIN(bytes, s->bdev_length - offset); | |
262 | } | |
263 | ||
782d97ef EB |
264 | /* Round offset and/or bytes to target cluster if COW is needed, and |
265 | * return the offset of the adjusted tail against original. */ | |
266 | static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, | |
ae4cc877 | 267 | uint64_t *bytes) |
893f7eba | 268 | { |
e5b43573 FZ |
269 | bool need_cow; |
270 | int ret = 0; | |
782d97ef | 271 | int64_t align_offset = *offset; |
7cfd5275 | 272 | int64_t align_bytes = *bytes; |
782d97ef | 273 | int max_bytes = s->granularity * s->max_iov; |
e5b43573 | 274 | |
782d97ef EB |
275 | need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); |
276 | need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, | |
e5b43573 FZ |
277 | s->cow_bitmap); |
278 | if (need_cow) { | |
782d97ef EB |
279 | bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, |
280 | &align_offset, &align_bytes); | |
e5b43573 | 281 | } |
3515727f | 282 | |
782d97ef EB |
283 | if (align_bytes > max_bytes) { |
284 | align_bytes = max_bytes; | |
e5b43573 | 285 | if (need_cow) { |
782d97ef | 286 | align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); |
e5b43573 | 287 | } |
8f0720ec | 288 | } |
782d97ef | 289 | /* Clipping may result in align_bytes unaligned to chunk boundary, but |
4150ae60 | 290 | * that doesn't matter because it's already the end of source image. */ |
782d97ef | 291 | align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); |
8f0720ec | 292 | |
782d97ef EB |
293 | ret = align_offset + align_bytes - (*offset + *bytes); |
294 | *offset = align_offset; | |
295 | *bytes = align_bytes; | |
e5b43573 FZ |
296 | assert(ret >= 0); |
297 | return ret; | |
298 | } | |
299 | ||
537c3d4f | 300 | static inline void coroutine_fn |
9178f4fe | 301 | mirror_wait_for_any_operation(MirrorBlockJob *s, bool active) |
21cd917f | 302 | { |
12aa4082 HR |
303 | MirrorOp *op; |
304 | ||
1181e19a HR |
305 | QTAILQ_FOREACH(op, &s->ops_in_flight, next) { |
306 | /* Do not wait on pseudo ops, because it may in turn wait on | |
307 | * some other operation to start, which may in fact be the | |
308 | * caller of this function. Since there is only one pseudo op | |
309 | * at any given time, we will always find some real operation | |
310 | * to wait on. */ | |
ce8cabbd KW |
311 | if (!op->is_pseudo_op && op->is_in_flight && |
312 | op->is_active_write == active) | |
313 | { | |
1181e19a HR |
314 | qemu_co_queue_wait(&op->waiting_requests, NULL); |
315 | return; | |
316 | } | |
317 | } | |
318 | abort(); | |
21cd917f FZ |
319 | } |
320 | ||
537c3d4f | 321 | static inline void coroutine_fn |
9178f4fe | 322 | mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) |
d06107ad HR |
323 | { |
324 | /* Only non-active operations use up in-flight slots */ | |
9178f4fe | 325 | mirror_wait_for_any_operation(s, false); |
d06107ad HR |
326 | } |
327 | ||
2e1990b2 HR |
328 | /* Perform a mirror copy operation. |
329 | * | |
330 | * *op->bytes_handled is set to the number of bytes copied after and | |
331 | * including offset, excluding any bytes copied prior to offset due | |
332 | * to alignment. This will be op->bytes if no alignment is necessary, | |
333 | * or (new_end - op->offset) if the tail is rounded up or down due to | |
334 | * alignment or buffer limit. | |
e5b43573 | 335 | */ |
2e1990b2 | 336 | static void coroutine_fn mirror_co_read(void *opaque) |
e5b43573 | 337 | { |
2e1990b2 HR |
338 | MirrorOp *op = opaque; |
339 | MirrorBlockJob *s = op->s; | |
ae4cc877 EB |
340 | int nb_chunks; |
341 | uint64_t ret; | |
ae4cc877 | 342 | uint64_t max_bytes; |
e5b43573 | 343 | |
ae4cc877 | 344 | max_bytes = s->granularity * s->max_iov; |
402a4741 | 345 | |
e5b43573 | 346 | /* We can only handle as much as buf_size at a time. */ |
2e1990b2 HR |
347 | op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes)); |
348 | assert(op->bytes); | |
349 | assert(op->bytes < BDRV_REQUEST_MAX_BYTES); | |
350 | *op->bytes_handled = op->bytes; | |
402a4741 | 351 | |
e5b43573 | 352 | if (s->cow_bitmap) { |
2e1990b2 | 353 | *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes); |
e5b43573 | 354 | } |
2e1990b2 HR |
355 | /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */ |
356 | assert(*op->bytes_handled <= UINT_MAX); | |
357 | assert(op->bytes <= s->buf_size); | |
ae4cc877 | 358 | /* The offset is granularity-aligned because: |
e5b43573 FZ |
359 | * 1) Caller passes in aligned values; |
360 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
2e1990b2 | 361 | assert(QEMU_IS_ALIGNED(op->offset, s->granularity)); |
ae4cc877 | 362 | /* The range is sector-aligned, since bdrv_getlength() rounds up. */ |
2e1990b2 HR |
363 | assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE)); |
364 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
e5b43573 FZ |
365 | |
366 | while (s->buf_free_count < nb_chunks) { | |
2e1990b2 | 367 | trace_mirror_yield_in_flight(s, op->offset, s->in_flight); |
9178f4fe | 368 | mirror_wait_for_free_in_flight_slot(s); |
b812f671 PB |
369 | } |
370 | ||
402a4741 PB |
371 | /* Now make a QEMUIOVector taking enough granularity-sized chunks |
372 | * from s->buf_free. | |
373 | */ | |
374 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
375 | while (nb_chunks-- > 0) { |
376 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
2e1990b2 | 377 | size_t remaining = op->bytes - op->qiov.size; |
5a0f6fd5 | 378 | |
402a4741 PB |
379 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
380 | s->buf_free_count--; | |
5a0f6fd5 | 381 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 382 | } |
bd48bde8 | 383 | |
893f7eba | 384 | /* Copy the dirty cluster. */ |
bd48bde8 | 385 | s->in_flight++; |
2e1990b2 | 386 | s->bytes_in_flight += op->bytes; |
ce8cabbd | 387 | op->is_in_flight = true; |
2e1990b2 | 388 | trace_mirror_one_iteration(s, op->offset, op->bytes); |
dcfb3beb | 389 | |
138f9fff HR |
390 | ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes, |
391 | &op->qiov, 0); | |
2e1990b2 | 392 | mirror_read_complete(op, ret); |
e5b43573 FZ |
393 | } |
394 | ||
2e1990b2 | 395 | static void coroutine_fn mirror_co_zero(void *opaque) |
e5b43573 | 396 | { |
2e1990b2 HR |
397 | MirrorOp *op = opaque; |
398 | int ret; | |
e5b43573 | 399 | |
2e1990b2 HR |
400 | op->s->in_flight++; |
401 | op->s->bytes_in_flight += op->bytes; | |
402 | *op->bytes_handled = op->bytes; | |
ce8cabbd | 403 | op->is_in_flight = true; |
e5b43573 | 404 | |
2e1990b2 HR |
405 | ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes, |
406 | op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0); | |
407 | mirror_write_complete(op, ret); | |
408 | } | |
409 | ||
410 | static void coroutine_fn mirror_co_discard(void *opaque) | |
411 | { | |
412 | MirrorOp *op = opaque; | |
413 | int ret; | |
414 | ||
415 | op->s->in_flight++; | |
416 | op->s->bytes_in_flight += op->bytes; | |
417 | *op->bytes_handled = op->bytes; | |
ce8cabbd | 418 | op->is_in_flight = true; |
2e1990b2 HR |
419 | |
420 | ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes); | |
421 | mirror_write_complete(op, ret); | |
e5b43573 FZ |
422 | } |
423 | ||
4295c5fc HR |
424 | static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, |
425 | unsigned bytes, MirrorMethod mirror_method) | |
426 | { | |
2e1990b2 HR |
427 | MirrorOp *op; |
428 | Coroutine *co; | |
429 | int64_t bytes_handled = -1; | |
430 | ||
431 | op = g_new(MirrorOp, 1); | |
432 | *op = (MirrorOp){ | |
433 | .s = s, | |
434 | .offset = offset, | |
435 | .bytes = bytes, | |
436 | .bytes_handled = &bytes_handled, | |
437 | }; | |
12aa4082 | 438 | qemu_co_queue_init(&op->waiting_requests); |
2e1990b2 | 439 | |
4295c5fc HR |
440 | switch (mirror_method) { |
441 | case MIRROR_METHOD_COPY: | |
2e1990b2 HR |
442 | co = qemu_coroutine_create(mirror_co_read, op); |
443 | break; | |
4295c5fc | 444 | case MIRROR_METHOD_ZERO: |
2e1990b2 HR |
445 | co = qemu_coroutine_create(mirror_co_zero, op); |
446 | break; | |
4295c5fc | 447 | case MIRROR_METHOD_DISCARD: |
2e1990b2 HR |
448 | co = qemu_coroutine_create(mirror_co_discard, op); |
449 | break; | |
4295c5fc HR |
450 | default: |
451 | abort(); | |
452 | } | |
eed325b9 | 453 | op->co = co; |
2e1990b2 | 454 | |
12aa4082 | 455 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); |
2e1990b2 HR |
456 | qemu_coroutine_enter(co); |
457 | /* At this point, ownership of op has been moved to the coroutine | |
458 | * and the object may already be freed */ | |
459 | ||
460 | /* Assert that this value has been set */ | |
461 | assert(bytes_handled >= 0); | |
462 | ||
463 | /* Same assertion as in mirror_co_read() (and for mirror_co_read() | |
464 | * and mirror_co_discard(), bytes_handled == op->bytes, which | |
465 | * is the @bytes parameter given to this function) */ | |
466 | assert(bytes_handled <= UINT_MAX); | |
467 | return bytes_handled; | |
4295c5fc HR |
468 | } |
469 | ||
e5b43573 FZ |
470 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) |
471 | { | |
138f9fff | 472 | BlockDriverState *source = s->mirror_top_bs->backing->bs; |
1181e19a HR |
473 | MirrorOp *pseudo_op; |
474 | int64_t offset; | |
475 | uint64_t delay_ns = 0, ret = 0; | |
e5b43573 FZ |
476 | /* At least the first dirty chunk is mirrored in one iteration. */ |
477 | int nb_chunks = 1; | |
4b5004d9 | 478 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
b436982f | 479 | int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); |
e5b43573 | 480 | |
b64bd51e | 481 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
f798184c | 482 | offset = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 483 | if (offset < 0) { |
dc162c8e | 484 | bdrv_set_dirty_iter(s->dbi, 0); |
f798184c | 485 | offset = bdrv_dirty_iter_next(s->dbi); |
9a46dba7 | 486 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
fb2ef791 | 487 | assert(offset >= 0); |
e5b43573 | 488 | } |
b64bd51e | 489 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
e5b43573 | 490 | |
1181e19a | 491 | mirror_wait_on_conflicts(NULL, s, offset, 1); |
9c83625b | 492 | |
da01ff7f | 493 | job_pause_point(&s->common.job); |
565ac01f | 494 | |
e5b43573 FZ |
495 | /* Find the number of consective dirty chunks following the first dirty |
496 | * one, and wait for in flight requests in them. */ | |
b64bd51e | 497 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
fb2ef791 | 498 | while (nb_chunks * s->granularity < s->buf_size) { |
dc162c8e | 499 | int64_t next_dirty; |
fb2ef791 EB |
500 | int64_t next_offset = offset + nb_chunks * s->granularity; |
501 | int64_t next_chunk = next_offset / s->granularity; | |
502 | if (next_offset >= s->bdev_length || | |
28636b82 | 503 | !bdrv_dirty_bitmap_get_locked(s->dirty_bitmap, next_offset)) { |
e5b43573 FZ |
504 | break; |
505 | } | |
506 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 507 | break; |
e5b43573 | 508 | } |
9c83625b | 509 | |
f798184c | 510 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 511 | if (next_dirty > next_offset || next_dirty < 0) { |
f27a2742 | 512 | /* The bitmap iterator's cache is stale, refresh it */ |
715a74d8 | 513 | bdrv_set_dirty_iter(s->dbi, next_offset); |
f798184c | 514 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
f27a2742 | 515 | } |
fb2ef791 | 516 | assert(next_dirty == next_offset); |
9c83625b | 517 | nb_chunks++; |
e5b43573 FZ |
518 | } |
519 | ||
520 | /* Clear dirty bits before querying the block status, because | |
31826642 | 521 | * calling bdrv_block_status_above could yield - if some blocks are |
e5b43573 FZ |
522 | * marked dirty in this window, we need to know. |
523 | */ | |
e0d7f73e EB |
524 | bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, |
525 | nb_chunks * s->granularity); | |
b64bd51e PB |
526 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
527 | ||
1181e19a HR |
528 | /* Before claiming an area in the in-flight bitmap, we have to |
529 | * create a MirrorOp for it so that conflicting requests can wait | |
530 | * for it. mirror_perform() will create the real MirrorOps later, | |
531 | * for now we just create a pseudo operation that will wake up all | |
532 | * conflicting requests once all real operations have been | |
533 | * launched. */ | |
534 | pseudo_op = g_new(MirrorOp, 1); | |
535 | *pseudo_op = (MirrorOp){ | |
536 | .offset = offset, | |
537 | .bytes = nb_chunks * s->granularity, | |
538 | .is_pseudo_op = true, | |
539 | }; | |
540 | qemu_co_queue_init(&pseudo_op->waiting_requests); | |
541 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next); | |
542 | ||
fb2ef791 EB |
543 | bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); |
544 | while (nb_chunks > 0 && offset < s->bdev_length) { | |
31826642 | 545 | int ret; |
7cfd5275 | 546 | int64_t io_bytes; |
f3e4ce4a | 547 | int64_t io_bytes_acct; |
4295c5fc | 548 | MirrorMethod mirror_method = MIRROR_METHOD_COPY; |
e5b43573 | 549 | |
fb2ef791 | 550 | assert(!(offset % s->granularity)); |
31826642 EB |
551 | ret = bdrv_block_status_above(source, NULL, offset, |
552 | nb_chunks * s->granularity, | |
553 | &io_bytes, NULL, NULL); | |
e5b43573 | 554 | if (ret < 0) { |
fb2ef791 | 555 | io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); |
0965a41e | 556 | } else if (ret & BDRV_BLOCK_DATA) { |
fb2ef791 | 557 | io_bytes = MIN(io_bytes, max_io_bytes); |
e5b43573 FZ |
558 | } |
559 | ||
fb2ef791 EB |
560 | io_bytes -= io_bytes % s->granularity; |
561 | if (io_bytes < s->granularity) { | |
562 | io_bytes = s->granularity; | |
e5b43573 | 563 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { |
fb2ef791 | 564 | int64_t target_offset; |
7cfd5275 | 565 | int64_t target_bytes; |
fb2ef791 EB |
566 | bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes, |
567 | &target_offset, &target_bytes); | |
568 | if (target_offset == offset && | |
569 | target_bytes == io_bytes) { | |
e5b43573 FZ |
570 | mirror_method = ret & BDRV_BLOCK_ZERO ? |
571 | MIRROR_METHOD_ZERO : | |
572 | MIRROR_METHOD_DISCARD; | |
573 | } | |
574 | } | |
575 | ||
cf56a3c6 | 576 | while (s->in_flight >= MAX_IN_FLIGHT) { |
fb2ef791 | 577 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
9178f4fe | 578 | mirror_wait_for_free_in_flight_slot(s); |
cf56a3c6 DL |
579 | } |
580 | ||
dbaa7b57 | 581 | if (s->ret < 0) { |
1181e19a HR |
582 | ret = 0; |
583 | goto fail; | |
dbaa7b57 VSO |
584 | } |
585 | ||
fb2ef791 | 586 | io_bytes = mirror_clip_bytes(s, offset, io_bytes); |
4295c5fc HR |
587 | io_bytes = mirror_perform(s, offset, io_bytes, mirror_method); |
588 | if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) { | |
589 | io_bytes_acct = 0; | |
590 | } else { | |
591 | io_bytes_acct = io_bytes; | |
e5b43573 | 592 | } |
fb2ef791 EB |
593 | assert(io_bytes); |
594 | offset += io_bytes; | |
595 | nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); | |
dee81d51 | 596 | delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct); |
dcfb3beb | 597 | } |
1181e19a HR |
598 | |
599 | ret = delay_ns; | |
600 | fail: | |
601 | QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next); | |
602 | qemu_co_queue_restart_all(&pseudo_op->waiting_requests); | |
603 | g_free(pseudo_op); | |
604 | ||
605 | return ret; | |
bd48bde8 | 606 | } |
b952b558 | 607 | |
402a4741 PB |
608 | static void mirror_free_init(MirrorBlockJob *s) |
609 | { | |
610 | int granularity = s->granularity; | |
611 | size_t buf_size = s->buf_size; | |
612 | uint8_t *buf = s->buf; | |
613 | ||
614 | assert(s->buf_free_count == 0); | |
615 | QSIMPLEQ_INIT(&s->buf_free); | |
616 | while (buf_size != 0) { | |
617 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
618 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
619 | s->buf_free_count++; | |
620 | buf_size -= granularity; | |
621 | buf += granularity; | |
622 | } | |
623 | } | |
624 | ||
bae8196d PB |
625 | /* This is also used for the .pause callback. There is no matching |
626 | * mirror_resume() because mirror_run() will begin iterating again | |
627 | * when the job is resumed. | |
628 | */ | |
537c3d4f | 629 | static void coroutine_fn mirror_wait_for_all_io(MirrorBlockJob *s) |
bd48bde8 PB |
630 | { |
631 | while (s->in_flight > 0) { | |
9178f4fe | 632 | mirror_wait_for_free_in_flight_slot(s); |
bd48bde8 | 633 | } |
893f7eba PB |
634 | } |
635 | ||
737efc1e JS |
636 | /** |
637 | * mirror_exit_common: handle both abort() and prepare() cases. | |
638 | * for .prepare, returns 0 on success and -errno on failure. | |
639 | * for .abort cases, denoted by abort = true, MUST return 0. | |
640 | */ | |
641 | static int mirror_exit_common(Job *job) | |
5a7e7a0b | 642 | { |
1908a559 KW |
643 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
644 | BlockJob *bjob = &s->common; | |
f93c3add | 645 | MirrorBDSOpaque *bs_opaque; |
5a7e7a0b | 646 | AioContext *replace_aio_context = NULL; |
f93c3add HR |
647 | BlockDriverState *src; |
648 | BlockDriverState *target_bs; | |
649 | BlockDriverState *mirror_top_bs; | |
12fa4af6 | 650 | Error *local_err = NULL; |
737efc1e JS |
651 | bool abort = job->ret < 0; |
652 | int ret = 0; | |
653 | ||
654 | if (s->prepared) { | |
655 | return 0; | |
656 | } | |
657 | s->prepared = true; | |
3f09bfbc | 658 | |
f93c3add HR |
659 | mirror_top_bs = s->mirror_top_bs; |
660 | bs_opaque = mirror_top_bs->opaque; | |
661 | src = mirror_top_bs->backing->bs; | |
662 | target_bs = blk_bs(s->target); | |
663 | ||
ef53dc09 AG |
664 | if (bdrv_chain_contains(src, target_bs)) { |
665 | bdrv_unfreeze_backing_chain(mirror_top_bs, target_bs); | |
666 | } | |
667 | ||
5deb6cbd | 668 | bdrv_release_dirty_bitmap(s->dirty_bitmap); |
2119882c | 669 | |
7b508f6b JS |
670 | /* Make sure that the source BDS doesn't go away during bdrv_replace_node, |
671 | * before we can call bdrv_drained_end */ | |
3f09bfbc | 672 | bdrv_ref(src); |
4ef85a9c | 673 | bdrv_ref(mirror_top_bs); |
7d9fcb39 KW |
674 | bdrv_ref(target_bs); |
675 | ||
bb0c9409 VSO |
676 | /* |
677 | * Remove target parent that still uses BLK_PERM_WRITE/RESIZE before | |
7d9fcb39 | 678 | * inserting target_bs at s->to_replace, where we might not be able to get |
63c8ef28 | 679 | * these permissions. |
bb0c9409 | 680 | */ |
7d9fcb39 KW |
681 | blk_unref(s->target); |
682 | s->target = NULL; | |
4ef85a9c KW |
683 | |
684 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
d2da5e28 KW |
685 | * required before it could become a backing file of target_bs. Not having |
686 | * these permissions any more means that we can't allow any new requests on | |
687 | * mirror_top_bs from now on, so keep it drained. */ | |
688 | bdrv_drained_begin(mirror_top_bs); | |
f94dc3b4 HR |
689 | bs_opaque->stop = true; |
690 | bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing, | |
691 | &error_abort); | |
737efc1e | 692 | if (!abort && s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { |
4ef85a9c | 693 | BlockDriverState *backing = s->is_none_mode ? src : s->base; |
3f072a7f HR |
694 | BlockDriverState *unfiltered_target = bdrv_skip_filters(target_bs); |
695 | ||
696 | if (bdrv_cow_bs(unfiltered_target) != backing) { | |
697 | bdrv_set_backing_hd(unfiltered_target, backing, &local_err); | |
12fa4af6 KW |
698 | if (local_err) { |
699 | error_report_err(local_err); | |
66c8672d | 700 | local_err = NULL; |
7b508f6b | 701 | ret = -EPERM; |
12fa4af6 | 702 | } |
4ef85a9c | 703 | } |
c41f5b96 HR |
704 | } else if (!abort && s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
705 | assert(!bdrv_backing_chain_next(target_bs)); | |
706 | ret = bdrv_open_backing_file(bdrv_skip_filters(target_bs), NULL, | |
707 | "backing", &local_err); | |
708 | if (ret < 0) { | |
709 | error_report_err(local_err); | |
710 | local_err = NULL; | |
711 | } | |
4ef85a9c | 712 | } |
5a7e7a0b SH |
713 | |
714 | if (s->to_replace) { | |
715 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
716 | aio_context_acquire(replace_aio_context); | |
717 | } | |
718 | ||
737efc1e JS |
719 | if (s->should_complete && !abort) { |
720 | BlockDriverState *to_replace = s->to_replace ?: src; | |
1ba79388 | 721 | bool ro = bdrv_is_read_only(to_replace); |
40365552 | 722 | |
1ba79388 AG |
723 | if (ro != bdrv_is_read_only(target_bs)) { |
724 | bdrv_reopen_set_read_only(target_bs, ro, NULL); | |
5a7e7a0b | 725 | } |
b8804815 KW |
726 | |
727 | /* The mirror job has no requests in flight any more, but we need to | |
728 | * drain potential other users of the BDS before changing the graph. */ | |
5e771752 | 729 | assert(s->in_drain); |
e253f4b8 | 730 | bdrv_drained_begin(target_bs); |
6e9cc051 HR |
731 | /* |
732 | * Cannot use check_to_replace_node() here, because that would | |
733 | * check for an op blocker on @to_replace, and we have our own | |
734 | * there. | |
735 | */ | |
736 | if (bdrv_recurse_can_replace(src, to_replace)) { | |
737 | bdrv_replace_node(to_replace, target_bs, &local_err); | |
738 | } else { | |
739 | error_setg(&local_err, "Can no longer replace '%s' by '%s', " | |
740 | "because it can no longer be guaranteed that doing so " | |
741 | "would not lead to an abrupt change of visible data", | |
742 | to_replace->node_name, target_bs->node_name); | |
743 | } | |
e253f4b8 | 744 | bdrv_drained_end(target_bs); |
5fe31c25 KW |
745 | if (local_err) { |
746 | error_report_err(local_err); | |
7b508f6b | 747 | ret = -EPERM; |
5fe31c25 | 748 | } |
5a7e7a0b SH |
749 | } |
750 | if (s->to_replace) { | |
751 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
752 | error_free(s->replace_blocker); | |
753 | bdrv_unref(s->to_replace); | |
754 | } | |
755 | if (replace_aio_context) { | |
756 | aio_context_release(replace_aio_context); | |
757 | } | |
758 | g_free(s->replaces); | |
7d9fcb39 | 759 | bdrv_unref(target_bs); |
4ef85a9c | 760 | |
f94dc3b4 HR |
761 | /* |
762 | * Remove the mirror filter driver from the graph. Before this, get rid of | |
4ef85a9c | 763 | * the blockers on the intermediate nodes so that the resulting state is |
f94dc3b4 HR |
764 | * valid. |
765 | */ | |
1908a559 | 766 | block_job_remove_all_bdrv(bjob); |
3f072a7f | 767 | bdrv_replace_node(mirror_top_bs, mirror_top_bs->backing->bs, &error_abort); |
4ef85a9c KW |
768 | |
769 | /* We just changed the BDS the job BB refers to (with either or both of the | |
5fe31c25 KW |
770 | * bdrv_replace_node() calls), so switch the BB back so the cleanup does |
771 | * the right thing. We don't need any permissions any more now. */ | |
1908a559 KW |
772 | blk_remove_bs(bjob->blk); |
773 | blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort); | |
774 | blk_insert_bs(bjob->blk, mirror_top_bs, &error_abort); | |
4ef85a9c | 775 | |
429076e8 | 776 | bs_opaque->job = NULL; |
4ef85a9c | 777 | |
176c3699 | 778 | bdrv_drained_end(src); |
d2da5e28 | 779 | bdrv_drained_end(mirror_top_bs); |
5e771752 | 780 | s->in_drain = false; |
4ef85a9c | 781 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 782 | bdrv_unref(src); |
7b508f6b | 783 | |
737efc1e JS |
784 | return ret; |
785 | } | |
786 | ||
787 | static int mirror_prepare(Job *job) | |
788 | { | |
789 | return mirror_exit_common(job); | |
790 | } | |
791 | ||
792 | static void mirror_abort(Job *job) | |
793 | { | |
794 | int ret = mirror_exit_common(job); | |
795 | assert(ret == 0); | |
5a7e7a0b SH |
796 | } |
797 | ||
537c3d4f | 798 | static void coroutine_fn mirror_throttle(MirrorBlockJob *s) |
49efb1f5 DL |
799 | { |
800 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
801 | ||
18bb6928 | 802 | if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) { |
49efb1f5 | 803 | s->last_pause_ns = now; |
5d43e86e | 804 | job_sleep_ns(&s->common.job, 0); |
49efb1f5 | 805 | } else { |
da01ff7f | 806 | job_pause_point(&s->common.job); |
49efb1f5 DL |
807 | } |
808 | } | |
809 | ||
c0b363ad DL |
810 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
811 | { | |
23ca459a | 812 | int64_t offset; |
138f9fff | 813 | BlockDriverState *bs = s->mirror_top_bs->backing->bs; |
c0b363ad | 814 | BlockDriverState *target_bs = blk_bs(s->target); |
23ca459a | 815 | int ret; |
51b0a488 | 816 | int64_t count; |
c0b363ad | 817 | |
cdf3bc93 | 818 | if (s->zero_target) { |
c7c2769c | 819 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
e0d7f73e | 820 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); |
c7c2769c DL |
821 | return 0; |
822 | } | |
823 | ||
90ab48eb | 824 | s->initial_zeroing_ongoing = true; |
23ca459a EB |
825 | for (offset = 0; offset < s->bdev_length; ) { |
826 | int bytes = MIN(s->bdev_length - offset, | |
827 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c7c2769c DL |
828 | |
829 | mirror_throttle(s); | |
830 | ||
daa7f2f9 | 831 | if (job_is_cancelled(&s->common.job)) { |
90ab48eb | 832 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
833 | return 0; |
834 | } | |
835 | ||
836 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
67adf4b3 EB |
837 | trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, |
838 | s->in_flight); | |
9178f4fe | 839 | mirror_wait_for_free_in_flight_slot(s); |
c7c2769c DL |
840 | continue; |
841 | } | |
842 | ||
4295c5fc | 843 | mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO); |
23ca459a | 844 | offset += bytes; |
c7c2769c DL |
845 | } |
846 | ||
bae8196d | 847 | mirror_wait_for_all_io(s); |
90ab48eb | 848 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
849 | } |
850 | ||
c0b363ad | 851 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
23ca459a | 852 | for (offset = 0; offset < s->bdev_length; ) { |
c0b363ad | 853 | /* Just to make sure we are not exceeding int limit. */ |
23ca459a EB |
854 | int bytes = MIN(s->bdev_length - offset, |
855 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c0b363ad DL |
856 | |
857 | mirror_throttle(s); | |
858 | ||
daa7f2f9 | 859 | if (job_is_cancelled(&s->common.job)) { |
c0b363ad DL |
860 | return 0; |
861 | } | |
862 | ||
3f072a7f HR |
863 | ret = bdrv_is_allocated_above(bs, s->base_overlay, true, offset, bytes, |
864 | &count); | |
c0b363ad DL |
865 | if (ret < 0) { |
866 | return ret; | |
867 | } | |
868 | ||
23ca459a | 869 | assert(count); |
a92b1b06 | 870 | if (ret > 0) { |
23ca459a | 871 | bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); |
c0b363ad | 872 | } |
23ca459a | 873 | offset += count; |
c0b363ad DL |
874 | } |
875 | return 0; | |
876 | } | |
877 | ||
bdffb31d PB |
878 | /* Called when going out of the streaming phase to flush the bulk of the |
879 | * data to the medium, or just before completing. | |
880 | */ | |
881 | static int mirror_flush(MirrorBlockJob *s) | |
882 | { | |
883 | int ret = blk_flush(s->target); | |
884 | if (ret < 0) { | |
885 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
886 | s->ret = ret; | |
887 | } | |
888 | } | |
889 | return ret; | |
890 | } | |
891 | ||
f67432a2 | 892 | static int coroutine_fn mirror_run(Job *job, Error **errp) |
893f7eba | 893 | { |
f67432a2 | 894 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
138f9fff | 895 | BlockDriverState *bs = s->mirror_top_bs->backing->bs; |
e253f4b8 | 896 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 897 | bool need_drain = true; |
c0b363ad | 898 | int64_t length; |
e83dd680 | 899 | int64_t target_length; |
b812f671 | 900 | BlockDriverInfo bdi; |
1d33936e JC |
901 | char backing_filename[2]; /* we only need 2 characters because we are only |
902 | checking for a NULL string */ | |
893f7eba | 903 | int ret = 0; |
893f7eba | 904 | |
daa7f2f9 | 905 | if (job_is_cancelled(&s->common.job)) { |
893f7eba PB |
906 | goto immediate_exit; |
907 | } | |
908 | ||
b21c7652 HR |
909 | s->bdev_length = bdrv_getlength(bs); |
910 | if (s->bdev_length < 0) { | |
911 | ret = s->bdev_length; | |
373df5b1 | 912 | goto immediate_exit; |
becc347e KW |
913 | } |
914 | ||
e83dd680 KW |
915 | target_length = blk_getlength(s->target); |
916 | if (target_length < 0) { | |
917 | ret = target_length; | |
918 | goto immediate_exit; | |
919 | } | |
920 | ||
becc347e KW |
921 | /* Active commit must resize the base image if its size differs from the |
922 | * active layer. */ | |
923 | if (s->base == blk_bs(s->target)) { | |
e83dd680 | 924 | if (s->bdev_length > target_length) { |
c80d8b06 | 925 | ret = blk_truncate(s->target, s->bdev_length, false, |
8c6242b6 | 926 | PREALLOC_MODE_OFF, 0, NULL); |
becc347e KW |
927 | if (ret < 0) { |
928 | goto immediate_exit; | |
929 | } | |
930 | } | |
e83dd680 KW |
931 | } else if (s->bdev_length != target_length) { |
932 | error_setg(errp, "Source and target image have different sizes"); | |
933 | ret = -EINVAL; | |
934 | goto immediate_exit; | |
becc347e KW |
935 | } |
936 | ||
937 | if (s->bdev_length == 0) { | |
2e1795b5 KW |
938 | /* Transition to the READY state and wait for complete. */ |
939 | job_transition_to_ready(&s->common.job); | |
9e48b025 | 940 | s->synced = true; |
d06107ad | 941 | s->actively_synced = true; |
daa7f2f9 | 942 | while (!job_is_cancelled(&s->common.job) && !s->should_complete) { |
198c49cc | 943 | job_yield(&s->common.job); |
9e48b025 | 944 | } |
daa7f2f9 | 945 | s->common.job.cancelled = false; |
9e48b025 | 946 | goto immediate_exit; |
893f7eba PB |
947 | } |
948 | ||
b21c7652 | 949 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
950 | s->in_flight_bitmap = bitmap_new(length); |
951 | ||
b812f671 PB |
952 | /* If we have no backing file yet in the destination, we cannot let |
953 | * the destination do COW. Instead, we copy sectors around the | |
954 | * dirty data if needed. We need a bitmap to do that. | |
955 | */ | |
e253f4b8 | 956 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 957 | sizeof(backing_filename)); |
e253f4b8 | 958 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
b436982f EB |
959 | s->target_cluster_size = bdi.cluster_size; |
960 | } else { | |
961 | s->target_cluster_size = BDRV_SECTOR_SIZE; | |
e5b43573 | 962 | } |
3f072a7f | 963 | if (backing_filename[0] && !bdrv_backing_chain_next(target_bs) && |
b436982f EB |
964 | s->granularity < s->target_cluster_size) { |
965 | s->buf_size = MAX(s->buf_size, s->target_cluster_size); | |
e5b43573 | 966 | s->cow_bitmap = bitmap_new(length); |
b812f671 | 967 | } |
e253f4b8 | 968 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 969 | |
7504edf4 KW |
970 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
971 | if (s->buf == NULL) { | |
972 | ret = -ENOMEM; | |
973 | goto immediate_exit; | |
974 | } | |
975 | ||
402a4741 | 976 | mirror_free_init(s); |
893f7eba | 977 | |
49efb1f5 | 978 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 979 | if (!s->is_none_mode) { |
c0b363ad | 980 | ret = mirror_dirty_init(s); |
daa7f2f9 | 981 | if (ret < 0 || job_is_cancelled(&s->common.job)) { |
c0b363ad | 982 | goto immediate_exit; |
893f7eba PB |
983 | } |
984 | } | |
985 | ||
dc162c8e | 986 | assert(!s->dbi); |
715a74d8 | 987 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); |
893f7eba | 988 | for (;;) { |
cc8c9d6c | 989 | uint64_t delay_ns = 0; |
49efb1f5 | 990 | int64_t cnt, delta; |
893f7eba PB |
991 | bool should_complete; |
992 | ||
d06107ad HR |
993 | /* Do not start passive operations while there are active |
994 | * writes in progress */ | |
995 | while (s->in_active_write_counter) { | |
9178f4fe | 996 | mirror_wait_for_any_operation(s, true); |
d06107ad HR |
997 | } |
998 | ||
bd48bde8 PB |
999 | if (s->ret < 0) { |
1000 | ret = s->ret; | |
1001 | goto immediate_exit; | |
1002 | } | |
1003 | ||
da01ff7f | 1004 | job_pause_point(&s->common.job); |
565ac01f | 1005 | |
20dca810 | 1006 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
05df8a6a KW |
1007 | /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is |
1008 | * the number of bytes currently being processed; together those are | |
1009 | * the current remaining operation length */ | |
30a5c887 | 1010 | job_progress_set_remaining(&s->common.job, s->bytes_in_flight + cnt); |
bd48bde8 PB |
1011 | |
1012 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 1013 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
18bb6928 KW |
1014 | * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is |
1015 | * an error, or when the source is clean, whichever comes first. */ | |
49efb1f5 | 1016 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
18bb6928 | 1017 | if (delta < BLOCK_JOB_SLICE_TIME && |
bd48bde8 | 1018 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 1019 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 | 1020 | (cnt == 0 && s->in_flight > 0)) { |
9a46dba7 | 1021 | trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); |
9178f4fe | 1022 | mirror_wait_for_free_in_flight_slot(s); |
bd48bde8 PB |
1023 | continue; |
1024 | } else if (cnt != 0) { | |
cc8c9d6c | 1025 | delay_ns = mirror_iteration(s); |
893f7eba | 1026 | } |
893f7eba PB |
1027 | } |
1028 | ||
1029 | should_complete = false; | |
bd48bde8 | 1030 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 1031 | trace_mirror_before_flush(s); |
bdffb31d PB |
1032 | if (!s->synced) { |
1033 | if (mirror_flush(s) < 0) { | |
1034 | /* Go check s->ret. */ | |
1035 | continue; | |
b952b558 | 1036 | } |
b952b558 PB |
1037 | /* We're out of the streaming phase. From now on, if the job |
1038 | * is cancelled we will actually complete all pending I/O and | |
1039 | * report completion. This way, block-job-cancel will leave | |
1040 | * the target in a consistent state. | |
1041 | */ | |
2e1795b5 | 1042 | job_transition_to_ready(&s->common.job); |
bdffb31d | 1043 | s->synced = true; |
d06107ad HR |
1044 | if (s->copy_mode != MIRROR_COPY_MODE_BACKGROUND) { |
1045 | s->actively_synced = true; | |
1046 | } | |
d63ffd87 | 1047 | } |
bdffb31d PB |
1048 | |
1049 | should_complete = s->should_complete || | |
daa7f2f9 | 1050 | job_is_cancelled(&s->common.job); |
bdffb31d | 1051 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
893f7eba PB |
1052 | } |
1053 | ||
1054 | if (cnt == 0 && should_complete) { | |
1055 | /* The dirty bitmap is not updated while operations are pending. | |
1056 | * If we're about to exit, wait for pending operations before | |
1057 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
1058 | * source has dirty data to copy! | |
1059 | * | |
1060 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
1061 | * mirror_populate runs, so pause it now. Before deciding |
1062 | * whether to switch to target check one last time if I/O has | |
1063 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba | 1064 | */ |
9a46dba7 | 1065 | trace_mirror_before_drain(s, cnt); |
9a0cec66 | 1066 | |
5e771752 | 1067 | s->in_drain = true; |
9a0cec66 | 1068 | bdrv_drained_begin(bs); |
20dca810 | 1069 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 1070 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 | 1071 | bdrv_drained_end(bs); |
5e771752 | 1072 | s->in_drain = false; |
9a0cec66 PB |
1073 | continue; |
1074 | } | |
1075 | ||
1076 | /* The two disks are in sync. Exit and report successful | |
1077 | * completion. | |
1078 | */ | |
1079 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
daa7f2f9 | 1080 | s->common.job.cancelled = false; |
9a0cec66 PB |
1081 | need_drain = false; |
1082 | break; | |
893f7eba PB |
1083 | } |
1084 | ||
1085 | ret = 0; | |
ddc4115e SH |
1086 | |
1087 | if (s->synced && !should_complete) { | |
18bb6928 KW |
1088 | delay_ns = (s->in_flight == 0 && |
1089 | cnt == 0 ? BLOCK_JOB_SLICE_TIME : 0); | |
ddc4115e | 1090 | } |
9a46dba7 | 1091 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
5d43e86e | 1092 | job_sleep_ns(&s->common.job, delay_ns); |
daa7f2f9 | 1093 | if (job_is_cancelled(&s->common.job) && |
004e95df | 1094 | (!s->synced || s->common.job.force_cancel)) |
eb36639f | 1095 | { |
b76e4458 | 1096 | break; |
893f7eba | 1097 | } |
49efb1f5 | 1098 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
1099 | } |
1100 | ||
1101 | immediate_exit: | |
bd48bde8 PB |
1102 | if (s->in_flight > 0) { |
1103 | /* We get here only if something went wrong. Either the job failed, | |
1104 | * or it was cancelled prematurely so that we do not guarantee that | |
1105 | * the target is a copy of the source. | |
1106 | */ | |
004e95df | 1107 | assert(ret < 0 || ((s->common.job.force_cancel || !s->synced) && |
daa7f2f9 | 1108 | job_is_cancelled(&s->common.job))); |
9a0cec66 | 1109 | assert(need_drain); |
bae8196d | 1110 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
1111 | } |
1112 | ||
1113 | assert(s->in_flight == 0); | |
7191bf31 | 1114 | qemu_vfree(s->buf); |
b812f671 | 1115 | g_free(s->cow_bitmap); |
402a4741 | 1116 | g_free(s->in_flight_bitmap); |
dc162c8e | 1117 | bdrv_dirty_iter_free(s->dbi); |
5a7e7a0b | 1118 | |
9a0cec66 | 1119 | if (need_drain) { |
5e771752 | 1120 | s->in_drain = true; |
9a0cec66 PB |
1121 | bdrv_drained_begin(bs); |
1122 | } | |
f67432a2 | 1123 | |
f67432a2 | 1124 | return ret; |
893f7eba PB |
1125 | } |
1126 | ||
3453d972 | 1127 | static void mirror_complete(Job *job, Error **errp) |
d63ffd87 | 1128 | { |
3453d972 | 1129 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
d63ffd87 | 1130 | |
d63ffd87 | 1131 | if (!s->synced) { |
9df229c3 | 1132 | error_setg(errp, "The active block job '%s' cannot be completed", |
3453d972 | 1133 | job->id); |
d63ffd87 PB |
1134 | return; |
1135 | } | |
1136 | ||
15d67298 | 1137 | /* block all operations on to_replace bs */ |
09158f00 | 1138 | if (s->replaces) { |
5a7e7a0b SH |
1139 | AioContext *replace_aio_context; |
1140 | ||
e12f3784 | 1141 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 1142 | if (!s->to_replace) { |
e12f3784 | 1143 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
1144 | return; |
1145 | } | |
1146 | ||
5a7e7a0b SH |
1147 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
1148 | aio_context_acquire(replace_aio_context); | |
1149 | ||
4ef85a9c KW |
1150 | /* TODO Translate this into permission system. Current definition of |
1151 | * GRAPH_MOD would require to request it for the parents; they might | |
1152 | * not even be BlockDriverStates, however, so a BdrvChild can't address | |
1153 | * them. May need redefinition of GRAPH_MOD. */ | |
09158f00 BC |
1154 | error_setg(&s->replace_blocker, |
1155 | "block device is in use by block-job-complete"); | |
1156 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
1157 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
1158 | |
1159 | aio_context_release(replace_aio_context); | |
09158f00 BC |
1160 | } |
1161 | ||
d63ffd87 | 1162 | s->should_complete = true; |
00769414 HR |
1163 | |
1164 | /* If the job is paused, it will be re-entered when it is resumed */ | |
1165 | if (!job->paused) { | |
1166 | job_enter(job); | |
1167 | } | |
d63ffd87 PB |
1168 | } |
1169 | ||
537c3d4f | 1170 | static void coroutine_fn mirror_pause(Job *job) |
565ac01f | 1171 | { |
da01ff7f | 1172 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
565ac01f | 1173 | |
bae8196d | 1174 | mirror_wait_for_all_io(s); |
565ac01f SH |
1175 | } |
1176 | ||
89bd0305 KW |
1177 | static bool mirror_drained_poll(BlockJob *job) |
1178 | { | |
1179 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
5e771752 SL |
1180 | |
1181 | /* If the job isn't paused nor cancelled, we can't be sure that it won't | |
1182 | * issue more requests. We make an exception if we've reached this point | |
1183 | * from one of our own drain sections, to avoid a deadlock waiting for | |
1184 | * ourselves. | |
1185 | */ | |
1186 | if (!s->common.job.paused && !s->common.job.cancelled && !s->in_drain) { | |
1187 | return true; | |
1188 | } | |
1189 | ||
89bd0305 KW |
1190 | return !!s->in_flight; |
1191 | } | |
1192 | ||
9c785cd7 | 1193 | static void mirror_cancel(Job *job, bool force) |
521ff8b7 VSO |
1194 | { |
1195 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); | |
1196 | BlockDriverState *target = blk_bs(s->target); | |
1197 | ||
9c785cd7 VSO |
1198 | if (force || !job_is_ready(job)) { |
1199 | bdrv_cancel_in_flight(target); | |
1200 | } | |
521ff8b7 VSO |
1201 | } |
1202 | ||
3fc4b10a | 1203 | static const BlockJobDriver mirror_job_driver = { |
33e9e9bd KW |
1204 | .job_driver = { |
1205 | .instance_size = sizeof(MirrorBlockJob), | |
252291ea | 1206 | .job_type = JOB_TYPE_MIRROR, |
80fa2c75 | 1207 | .free = block_job_free, |
b15de828 | 1208 | .user_resume = block_job_user_resume, |
f67432a2 | 1209 | .run = mirror_run, |
737efc1e JS |
1210 | .prepare = mirror_prepare, |
1211 | .abort = mirror_abort, | |
da01ff7f | 1212 | .pause = mirror_pause, |
3453d972 | 1213 | .complete = mirror_complete, |
521ff8b7 | 1214 | .cancel = mirror_cancel, |
33e9e9bd | 1215 | }, |
89bd0305 | 1216 | .drained_poll = mirror_drained_poll, |
893f7eba PB |
1217 | }; |
1218 | ||
03544a6e | 1219 | static const BlockJobDriver commit_active_job_driver = { |
33e9e9bd KW |
1220 | .job_driver = { |
1221 | .instance_size = sizeof(MirrorBlockJob), | |
252291ea | 1222 | .job_type = JOB_TYPE_COMMIT, |
80fa2c75 | 1223 | .free = block_job_free, |
b15de828 | 1224 | .user_resume = block_job_user_resume, |
f67432a2 | 1225 | .run = mirror_run, |
737efc1e JS |
1226 | .prepare = mirror_prepare, |
1227 | .abort = mirror_abort, | |
da01ff7f | 1228 | .pause = mirror_pause, |
3453d972 | 1229 | .complete = mirror_complete, |
33e9e9bd | 1230 | }, |
89bd0305 | 1231 | .drained_poll = mirror_drained_poll, |
03544a6e FZ |
1232 | }; |
1233 | ||
537c3d4f SH |
1234 | static void coroutine_fn |
1235 | do_sync_target_write(MirrorBlockJob *job, MirrorMethod method, | |
1236 | uint64_t offset, uint64_t bytes, | |
1237 | QEMUIOVector *qiov, int flags) | |
d06107ad | 1238 | { |
5c511ac3 | 1239 | int ret; |
dbdf699c VSO |
1240 | size_t qiov_offset = 0; |
1241 | int64_t bitmap_offset, bitmap_end; | |
d06107ad | 1242 | |
dbdf699c VSO |
1243 | if (!QEMU_IS_ALIGNED(offset, job->granularity) && |
1244 | bdrv_dirty_bitmap_get(job->dirty_bitmap, offset)) | |
1245 | { | |
1246 | /* | |
1247 | * Dirty unaligned padding: ignore it. | |
1248 | * | |
1249 | * Reasoning: | |
1250 | * 1. If we copy it, we can't reset corresponding bit in | |
1251 | * dirty_bitmap as there may be some "dirty" bytes still not | |
1252 | * copied. | |
1253 | * 2. It's already dirty, so skipping it we don't diverge mirror | |
1254 | * progress. | |
1255 | * | |
1256 | * Note, that because of this, guest write may have no contribution | |
1257 | * into mirror converge, but that's not bad, as we have background | |
1258 | * process of mirroring. If under some bad circumstances (high guest | |
1259 | * IO load) background process starve, we will not converge anyway, | |
1260 | * even if each write will contribute, as guest is not guaranteed to | |
1261 | * rewrite the whole disk. | |
1262 | */ | |
1263 | qiov_offset = QEMU_ALIGN_UP(offset, job->granularity) - offset; | |
1264 | if (bytes <= qiov_offset) { | |
1265 | /* nothing to do after shrink */ | |
1266 | return; | |
1267 | } | |
1268 | offset += qiov_offset; | |
1269 | bytes -= qiov_offset; | |
1270 | } | |
1271 | ||
1272 | if (!QEMU_IS_ALIGNED(offset + bytes, job->granularity) && | |
1273 | bdrv_dirty_bitmap_get(job->dirty_bitmap, offset + bytes - 1)) | |
1274 | { | |
1275 | uint64_t tail = (offset + bytes) % job->granularity; | |
1276 | ||
1277 | if (bytes <= tail) { | |
1278 | /* nothing to do after shrink */ | |
1279 | return; | |
1280 | } | |
1281 | bytes -= tail; | |
1282 | } | |
1283 | ||
1284 | /* | |
1285 | * Tails are either clean or shrunk, so for bitmap resetting | |
1286 | * we safely align the range down. | |
1287 | */ | |
1288 | bitmap_offset = QEMU_ALIGN_UP(offset, job->granularity); | |
1289 | bitmap_end = QEMU_ALIGN_DOWN(offset + bytes, job->granularity); | |
1290 | if (bitmap_offset < bitmap_end) { | |
1291 | bdrv_reset_dirty_bitmap(job->dirty_bitmap, bitmap_offset, | |
1292 | bitmap_end - bitmap_offset); | |
1293 | } | |
d06107ad | 1294 | |
5c511ac3 | 1295 | job_progress_increase_remaining(&job->common.job, bytes); |
d06107ad | 1296 | |
5c511ac3 VSO |
1297 | switch (method) { |
1298 | case MIRROR_METHOD_COPY: | |
dbdf699c VSO |
1299 | ret = blk_co_pwritev_part(job->target, offset, bytes, |
1300 | qiov, qiov_offset, flags); | |
5c511ac3 | 1301 | break; |
d06107ad | 1302 | |
5c511ac3 VSO |
1303 | case MIRROR_METHOD_ZERO: |
1304 | assert(!qiov); | |
1305 | ret = blk_co_pwrite_zeroes(job->target, offset, bytes, flags); | |
1306 | break; | |
d06107ad | 1307 | |
5c511ac3 VSO |
1308 | case MIRROR_METHOD_DISCARD: |
1309 | assert(!qiov); | |
1310 | ret = blk_co_pdiscard(job->target, offset, bytes); | |
1311 | break; | |
d06107ad | 1312 | |
5c511ac3 VSO |
1313 | default: |
1314 | abort(); | |
1315 | } | |
d06107ad | 1316 | |
5c511ac3 VSO |
1317 | if (ret >= 0) { |
1318 | job_progress_update(&job->common.job, bytes); | |
1319 | } else { | |
1320 | BlockErrorAction action; | |
d06107ad | 1321 | |
dbdf699c VSO |
1322 | /* |
1323 | * We failed, so we should mark dirty the whole area, aligned up. | |
1324 | * Note that we don't care about shrunk tails if any: they were dirty | |
1325 | * at function start, and they must be still dirty, as we've locked | |
1326 | * the region for in-flight op. | |
1327 | */ | |
1328 | bitmap_offset = QEMU_ALIGN_DOWN(offset, job->granularity); | |
1329 | bitmap_end = QEMU_ALIGN_UP(offset + bytes, job->granularity); | |
1330 | bdrv_set_dirty_bitmap(job->dirty_bitmap, bitmap_offset, | |
1331 | bitmap_end - bitmap_offset); | |
5c511ac3 | 1332 | job->actively_synced = false; |
d06107ad | 1333 | |
5c511ac3 VSO |
1334 | action = mirror_error_action(job, false, -ret); |
1335 | if (action == BLOCK_ERROR_ACTION_REPORT) { | |
1336 | if (!job->ret) { | |
1337 | job->ret = ret; | |
d06107ad HR |
1338 | } |
1339 | } | |
d06107ad HR |
1340 | } |
1341 | } | |
1342 | ||
1343 | static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s, | |
1344 | uint64_t offset, | |
1345 | uint64_t bytes) | |
1346 | { | |
1347 | MirrorOp *op; | |
1348 | uint64_t start_chunk = offset / s->granularity; | |
1349 | uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); | |
1350 | ||
1351 | op = g_new(MirrorOp, 1); | |
1352 | *op = (MirrorOp){ | |
1353 | .s = s, | |
1354 | .offset = offset, | |
1355 | .bytes = bytes, | |
1356 | .is_active_write = true, | |
ce8cabbd | 1357 | .is_in_flight = true, |
ead3f1bf | 1358 | .co = qemu_coroutine_self(), |
d06107ad HR |
1359 | }; |
1360 | qemu_co_queue_init(&op->waiting_requests); | |
1361 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); | |
1362 | ||
1363 | s->in_active_write_counter++; | |
1364 | ||
1365 | mirror_wait_on_conflicts(op, s, offset, bytes); | |
1366 | ||
1367 | bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); | |
1368 | ||
1369 | return op; | |
1370 | } | |
1371 | ||
1372 | static void coroutine_fn active_write_settle(MirrorOp *op) | |
1373 | { | |
1374 | uint64_t start_chunk = op->offset / op->s->granularity; | |
1375 | uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes, | |
1376 | op->s->granularity); | |
1377 | ||
1378 | if (!--op->s->in_active_write_counter && op->s->actively_synced) { | |
1379 | BdrvChild *source = op->s->mirror_top_bs->backing; | |
1380 | ||
1381 | if (QLIST_FIRST(&source->bs->parents) == source && | |
1382 | QLIST_NEXT(source, next_parent) == NULL) | |
1383 | { | |
1384 | /* Assert that we are back in sync once all active write | |
1385 | * operations are settled. | |
1386 | * Note that we can only assert this if the mirror node | |
1387 | * is the source node's only parent. */ | |
1388 | assert(!bdrv_get_dirty_count(op->s->dirty_bitmap)); | |
1389 | } | |
1390 | } | |
1391 | bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); | |
1392 | QTAILQ_REMOVE(&op->s->ops_in_flight, op, next); | |
1393 | qemu_co_queue_restart_all(&op->waiting_requests); | |
1394 | g_free(op); | |
1395 | } | |
1396 | ||
4ef85a9c KW |
1397 | static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, |
1398 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1399 | { | |
1400 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1401 | } | |
1402 | ||
d06107ad HR |
1403 | static int coroutine_fn bdrv_mirror_top_do_write(BlockDriverState *bs, |
1404 | MirrorMethod method, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, | |
1405 | int flags) | |
1406 | { | |
1407 | MirrorOp *op = NULL; | |
1408 | MirrorBDSOpaque *s = bs->opaque; | |
1409 | int ret = 0; | |
1410 | bool copy_to_target; | |
1411 | ||
1412 | copy_to_target = s->job->ret >= 0 && | |
1413 | s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; | |
1414 | ||
1415 | if (copy_to_target) { | |
1416 | op = active_write_prepare(s->job, offset, bytes); | |
1417 | } | |
1418 | ||
1419 | switch (method) { | |
1420 | case MIRROR_METHOD_COPY: | |
1421 | ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1422 | break; | |
1423 | ||
1424 | case MIRROR_METHOD_ZERO: | |
1425 | ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); | |
1426 | break; | |
1427 | ||
1428 | case MIRROR_METHOD_DISCARD: | |
0b9fd3f4 | 1429 | ret = bdrv_co_pdiscard(bs->backing, offset, bytes); |
d06107ad HR |
1430 | break; |
1431 | ||
1432 | default: | |
1433 | abort(); | |
1434 | } | |
1435 | ||
1436 | if (ret < 0) { | |
1437 | goto out; | |
1438 | } | |
1439 | ||
1440 | if (copy_to_target) { | |
1441 | do_sync_target_write(s->job, method, offset, bytes, qiov, flags); | |
1442 | } | |
1443 | ||
1444 | out: | |
1445 | if (copy_to_target) { | |
1446 | active_write_settle(op); | |
1447 | } | |
1448 | return ret; | |
1449 | } | |
1450 | ||
4ef85a9c KW |
1451 | static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, |
1452 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1453 | { | |
d06107ad HR |
1454 | MirrorBDSOpaque *s = bs->opaque; |
1455 | QEMUIOVector bounce_qiov; | |
1456 | void *bounce_buf; | |
1457 | int ret = 0; | |
1458 | bool copy_to_target; | |
1459 | ||
1460 | copy_to_target = s->job->ret >= 0 && | |
1461 | s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; | |
1462 | ||
1463 | if (copy_to_target) { | |
1464 | /* The guest might concurrently modify the data to write; but | |
1465 | * the data on source and destination must match, so we have | |
1466 | * to use a bounce buffer if we are going to write to the | |
1467 | * target now. */ | |
1468 | bounce_buf = qemu_blockalign(bs, bytes); | |
1469 | iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes); | |
1470 | ||
1471 | qemu_iovec_init(&bounce_qiov, 1); | |
1472 | qemu_iovec_add(&bounce_qiov, bounce_buf, bytes); | |
1473 | qiov = &bounce_qiov; | |
1474 | } | |
1475 | ||
1476 | ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, offset, bytes, qiov, | |
1477 | flags); | |
1478 | ||
1479 | if (copy_to_target) { | |
1480 | qemu_iovec_destroy(&bounce_qiov); | |
1481 | qemu_vfree(bounce_buf); | |
1482 | } | |
1483 | ||
1484 | return ret; | |
4ef85a9c KW |
1485 | } |
1486 | ||
1487 | static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) | |
1488 | { | |
ce960aa9 VSO |
1489 | if (bs->backing == NULL) { |
1490 | /* we can be here after failed bdrv_append in mirror_start_job */ | |
1491 | return 0; | |
1492 | } | |
4ef85a9c KW |
1493 | return bdrv_co_flush(bs->backing->bs); |
1494 | } | |
1495 | ||
4ef85a9c | 1496 | static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 1497 | int64_t offset, int bytes, BdrvRequestFlags flags) |
4ef85a9c | 1498 | { |
d06107ad HR |
1499 | return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, offset, bytes, NULL, |
1500 | flags); | |
4ef85a9c KW |
1501 | } |
1502 | ||
1503 | static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, | |
f5a5ca79 | 1504 | int64_t offset, int bytes) |
4ef85a9c | 1505 | { |
d06107ad HR |
1506 | return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, offset, bytes, |
1507 | NULL, 0); | |
4ef85a9c KW |
1508 | } |
1509 | ||
998b3a1e | 1510 | static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs) |
fd4a6493 | 1511 | { |
18775ff3 VSO |
1512 | if (bs->backing == NULL) { |
1513 | /* we can be here after failed bdrv_attach_child in | |
1514 | * bdrv_set_backing_hd */ | |
1515 | return; | |
1516 | } | |
fd4a6493 KW |
1517 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), |
1518 | bs->backing->bs->filename); | |
1519 | } | |
1520 | ||
4ef85a9c | 1521 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, |
bf8e925e | 1522 | BdrvChildRole role, |
e0995dc3 | 1523 | BlockReopenQueue *reopen_queue, |
4ef85a9c KW |
1524 | uint64_t perm, uint64_t shared, |
1525 | uint64_t *nperm, uint64_t *nshared) | |
1526 | { | |
f94dc3b4 HR |
1527 | MirrorBDSOpaque *s = bs->opaque; |
1528 | ||
1529 | if (s->stop) { | |
1530 | /* | |
1531 | * If the job is to be stopped, we do not need to forward | |
1532 | * anything to the real image. | |
1533 | */ | |
1534 | *nperm = 0; | |
1535 | *nshared = BLK_PERM_ALL; | |
1536 | return; | |
1537 | } | |
1538 | ||
53431b90 HR |
1539 | bdrv_default_perms(bs, c, role, reopen_queue, |
1540 | perm, shared, nperm, nshared); | |
4ef85a9c | 1541 | |
53431b90 HR |
1542 | if (s->is_commit) { |
1543 | /* | |
1544 | * For commit jobs, we cannot take CONSISTENT_READ, because | |
1545 | * that permission is unshared for everything above the base | |
1546 | * node (except for filters on the base node). | |
1547 | * We also have to force-share the WRITE permission, or | |
1548 | * otherwise we would block ourselves at the base node (if | |
1549 | * writes are blocked for a node, they are also blocked for | |
1550 | * its backing file). | |
1551 | * (We could also share RESIZE, because it may be needed for | |
1552 | * the target if its size is less than the top node's; but | |
1553 | * bdrv_default_perms_for_cow() automatically shares RESIZE | |
1554 | * for backing nodes if WRITE is shared, so there is no need | |
1555 | * to do it here.) | |
1556 | */ | |
1557 | *nperm &= ~BLK_PERM_CONSISTENT_READ; | |
1558 | *nshared |= BLK_PERM_WRITE; | |
1559 | } | |
4ef85a9c KW |
1560 | } |
1561 | ||
1562 | /* Dummy node that provides consistent read to its users without requiring it | |
1563 | * from its backing file and that allows writes on the backing file chain. */ | |
1564 | static BlockDriver bdrv_mirror_top = { | |
1565 | .format_name = "mirror_top", | |
1566 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1567 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1568 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1569 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1570 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
fd4a6493 | 1571 | .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, |
4ef85a9c | 1572 | .bdrv_child_perm = bdrv_mirror_top_child_perm, |
6540fd15 HR |
1573 | |
1574 | .is_filter = true, | |
4ef85a9c KW |
1575 | }; |
1576 | ||
cc19f177 VSO |
1577 | static BlockJob *mirror_start_job( |
1578 | const char *job_id, BlockDriverState *bs, | |
47970dfb JS |
1579 | int creation_flags, BlockDriverState *target, |
1580 | const char *replaces, int64_t speed, | |
1581 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1582 | BlockMirrorBackingMode backing_mode, |
cdf3bc93 | 1583 | bool zero_target, |
09158f00 BC |
1584 | BlockdevOnError on_source_error, |
1585 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1586 | bool unmap, |
097310b5 | 1587 | BlockCompletionFunc *cb, |
51ccfa2d | 1588 | void *opaque, |
09158f00 | 1589 | const BlockJobDriver *driver, |
b49f7ead | 1590 | bool is_none_mode, BlockDriverState *base, |
51ccfa2d | 1591 | bool auto_complete, const char *filter_node_name, |
481debaa | 1592 | bool is_mirror, MirrorCopyMode copy_mode, |
51ccfa2d | 1593 | Error **errp) |
893f7eba PB |
1594 | { |
1595 | MirrorBlockJob *s; | |
429076e8 | 1596 | MirrorBDSOpaque *bs_opaque; |
4ef85a9c | 1597 | BlockDriverState *mirror_top_bs; |
4ef85a9c | 1598 | bool target_is_backing; |
3f072a7f | 1599 | uint64_t target_perms, target_shared_perms; |
d7086422 | 1600 | int ret; |
893f7eba | 1601 | |
eee13dfe | 1602 | if (granularity == 0) { |
341ebc2f | 1603 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1604 | } |
1605 | ||
31826642 | 1606 | assert(is_power_of_2(granularity)); |
eee13dfe | 1607 | |
48ac0a4d WC |
1608 | if (buf_size < 0) { |
1609 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
cc19f177 | 1610 | return NULL; |
48ac0a4d WC |
1611 | } |
1612 | ||
1613 | if (buf_size == 0) { | |
1614 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
1615 | } | |
5bc361b8 | 1616 | |
3f072a7f | 1617 | if (bdrv_skip_filters(bs) == bdrv_skip_filters(target)) { |
86fae10c | 1618 | error_setg(errp, "Can't mirror node into itself"); |
cc19f177 | 1619 | return NULL; |
86fae10c KW |
1620 | } |
1621 | ||
53431b90 HR |
1622 | target_is_backing = bdrv_chain_contains(bs, target); |
1623 | ||
4ef85a9c KW |
1624 | /* In the case of active commit, add dummy driver to provide consistent |
1625 | * reads on the top, while disabling it in the intermediate nodes, and make | |
1626 | * the backing chain writable. */ | |
6cdbceb1 KW |
1627 | mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, |
1628 | BDRV_O_RDWR, errp); | |
4ef85a9c | 1629 | if (mirror_top_bs == NULL) { |
cc19f177 | 1630 | return NULL; |
4ef85a9c | 1631 | } |
d3c8c674 KW |
1632 | if (!filter_node_name) { |
1633 | mirror_top_bs->implicit = true; | |
1634 | } | |
e5182c1c HR |
1635 | |
1636 | /* So that we can always drop this node */ | |
1637 | mirror_top_bs->never_freeze = true; | |
1638 | ||
4ef85a9c | 1639 | mirror_top_bs->total_sectors = bs->total_sectors; |
228345bf | 1640 | mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED; |
80f5c33f KW |
1641 | mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED | |
1642 | BDRV_REQ_NO_FALLBACK; | |
429076e8 HR |
1643 | bs_opaque = g_new0(MirrorBDSOpaque, 1); |
1644 | mirror_top_bs->opaque = bs_opaque; | |
4ef85a9c | 1645 | |
53431b90 HR |
1646 | bs_opaque->is_commit = target_is_backing; |
1647 | ||
4ef85a9c | 1648 | bdrv_drained_begin(bs); |
934aee14 | 1649 | ret = bdrv_append(mirror_top_bs, bs, errp); |
4ef85a9c KW |
1650 | bdrv_drained_end(bs); |
1651 | ||
934aee14 | 1652 | if (ret < 0) { |
b2c2832c | 1653 | bdrv_unref(mirror_top_bs); |
cc19f177 | 1654 | return NULL; |
b2c2832c KW |
1655 | } |
1656 | ||
4ef85a9c | 1657 | /* Make sure that the source is not resized while the job is running */ |
75859b94 | 1658 | s = block_job_create(job_id, driver, NULL, mirror_top_bs, |
4ef85a9c KW |
1659 | BLK_PERM_CONSISTENT_READ, |
1660 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
1661 | BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, | |
c6cc12bf | 1662 | creation_flags, cb, opaque, errp); |
893f7eba | 1663 | if (!s) { |
4ef85a9c | 1664 | goto fail; |
893f7eba | 1665 | } |
429076e8 HR |
1666 | bs_opaque->job = s; |
1667 | ||
7a25fcd0 HR |
1668 | /* The block job now has a reference to this node */ |
1669 | bdrv_unref(mirror_top_bs); | |
1670 | ||
4ef85a9c KW |
1671 | s->mirror_top_bs = mirror_top_bs; |
1672 | ||
1673 | /* No resize for the target either; while the mirror is still running, a | |
1674 | * consistent read isn't necessarily possible. We could possibly allow | |
1675 | * writes and graph modifications, though it would likely defeat the | |
1676 | * purpose of a mirror, so leave them blocked for now. | |
1677 | * | |
1678 | * In the case of active commit, things look a bit different, though, | |
1679 | * because the target is an already populated backing file in active use. | |
1680 | * We can allow anything except resize there.*/ | |
3f072a7f HR |
1681 | |
1682 | target_perms = BLK_PERM_WRITE; | |
1683 | target_shared_perms = BLK_PERM_WRITE_UNCHANGED; | |
1684 | ||
3f072a7f HR |
1685 | if (target_is_backing) { |
1686 | int64_t bs_size, target_size; | |
1687 | bs_size = bdrv_getlength(bs); | |
1688 | if (bs_size < 0) { | |
1689 | error_setg_errno(errp, -bs_size, | |
1690 | "Could not inquire top image size"); | |
1691 | goto fail; | |
1692 | } | |
1693 | ||
1694 | target_size = bdrv_getlength(target); | |
1695 | if (target_size < 0) { | |
1696 | error_setg_errno(errp, -target_size, | |
1697 | "Could not inquire base image size"); | |
1698 | goto fail; | |
1699 | } | |
1700 | ||
1701 | if (target_size < bs_size) { | |
1702 | target_perms |= BLK_PERM_RESIZE; | |
1703 | } | |
1704 | ||
1705 | target_shared_perms |= BLK_PERM_CONSISTENT_READ | |
1706 | | BLK_PERM_WRITE | |
1707 | | BLK_PERM_GRAPH_MOD; | |
1708 | } else if (bdrv_chain_contains(bs, bdrv_skip_filters(target))) { | |
1709 | /* | |
1710 | * We may want to allow this in the future, but it would | |
1711 | * require taking some extra care. | |
1712 | */ | |
1713 | error_setg(errp, "Cannot mirror to a filter on top of a node in the " | |
1714 | "source's backing chain"); | |
1715 | goto fail; | |
1716 | } | |
1717 | ||
1718 | if (backing_mode != MIRROR_LEAVE_BACKING_CHAIN) { | |
1719 | target_perms |= BLK_PERM_GRAPH_MOD; | |
1720 | } | |
1721 | ||
d861ab3a | 1722 | s->target = blk_new(s->common.job.aio_context, |
3f072a7f | 1723 | target_perms, target_shared_perms); |
d7086422 KW |
1724 | ret = blk_insert_bs(s->target, target, errp); |
1725 | if (ret < 0) { | |
4ef85a9c | 1726 | goto fail; |
d7086422 | 1727 | } |
045a2f82 FZ |
1728 | if (is_mirror) { |
1729 | /* XXX: Mirror target could be a NBD server of target QEMU in the case | |
1730 | * of non-shared block migration. To allow migration completion, we | |
1731 | * have to allow "inactivate" of the target BB. When that happens, we | |
1732 | * know the job is drained, and the vcpus are stopped, so no write | |
1733 | * operation will be performed. Block layer already has assertions to | |
1734 | * ensure that. */ | |
1735 | blk_set_force_allow_inactivate(s->target); | |
1736 | } | |
9ff7f0df | 1737 | blk_set_allow_aio_context_change(s->target, true); |
cf312932 | 1738 | blk_set_disable_request_queuing(s->target, true); |
e253f4b8 | 1739 | |
09158f00 | 1740 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1741 | s->on_source_error = on_source_error; |
1742 | s->on_target_error = on_target_error; | |
03544a6e | 1743 | s->is_none_mode = is_none_mode; |
274fccee | 1744 | s->backing_mode = backing_mode; |
cdf3bc93 | 1745 | s->zero_target = zero_target; |
481debaa | 1746 | s->copy_mode = copy_mode; |
5bc361b8 | 1747 | s->base = base; |
3f072a7f | 1748 | s->base_overlay = bdrv_find_overlay(bs, base); |
eee13dfe | 1749 | s->granularity = granularity; |
48ac0a4d | 1750 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1751 | s->unmap = unmap; |
b49f7ead WC |
1752 | if (auto_complete) { |
1753 | s->should_complete = true; | |
1754 | } | |
b812f671 | 1755 | |
0db6e54a | 1756 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1757 | if (!s->dirty_bitmap) { |
88f9d1b3 | 1758 | goto fail; |
b8afb520 | 1759 | } |
dbdf699c VSO |
1760 | if (s->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING) { |
1761 | bdrv_disable_dirty_bitmap(s->dirty_bitmap); | |
1762 | } | |
10f3cd15 | 1763 | |
67b24427 AG |
1764 | ret = block_job_add_bdrv(&s->common, "source", bs, 0, |
1765 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE | | |
1766 | BLK_PERM_CONSISTENT_READ, | |
1767 | errp); | |
1768 | if (ret < 0) { | |
1769 | goto fail; | |
1770 | } | |
1771 | ||
4ef85a9c | 1772 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1773 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1774 | &error_abort); | |
1775 | ||
f3ede4b0 AG |
1776 | /* In commit_active_start() all intermediate nodes disappear, so |
1777 | * any jobs in them must be blocked */ | |
4ef85a9c | 1778 | if (target_is_backing) { |
3f072a7f HR |
1779 | BlockDriverState *iter, *filtered_target; |
1780 | uint64_t iter_shared_perms; | |
1781 | ||
1782 | /* | |
1783 | * The topmost node with | |
1784 | * bdrv_skip_filters(filtered_target) == bdrv_skip_filters(target) | |
1785 | */ | |
1786 | filtered_target = bdrv_cow_bs(bdrv_find_overlay(bs, target)); | |
1787 | ||
1788 | assert(bdrv_skip_filters(filtered_target) == | |
1789 | bdrv_skip_filters(target)); | |
1790 | ||
1791 | /* | |
1792 | * XXX BLK_PERM_WRITE needs to be allowed so we don't block | |
1793 | * ourselves at s->base (if writes are blocked for a node, they are | |
1794 | * also blocked for its backing file). The other options would be a | |
1795 | * second filter driver above s->base (== target). | |
1796 | */ | |
1797 | iter_shared_perms = BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE; | |
1798 | ||
1799 | for (iter = bdrv_filter_or_cow_bs(bs); iter != target; | |
1800 | iter = bdrv_filter_or_cow_bs(iter)) | |
1801 | { | |
1802 | if (iter == filtered_target) { | |
1803 | /* | |
1804 | * From here on, all nodes are filters on the base. | |
1805 | * This allows us to share BLK_PERM_CONSISTENT_READ. | |
1806 | */ | |
1807 | iter_shared_perms |= BLK_PERM_CONSISTENT_READ; | |
1808 | } | |
1809 | ||
4ef85a9c | 1810 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, |
3f072a7f | 1811 | iter_shared_perms, errp); |
4ef85a9c KW |
1812 | if (ret < 0) { |
1813 | goto fail; | |
1814 | } | |
f3ede4b0 | 1815 | } |
ef53dc09 AG |
1816 | |
1817 | if (bdrv_freeze_backing_chain(mirror_top_bs, target, errp) < 0) { | |
1818 | goto fail; | |
1819 | } | |
f3ede4b0 | 1820 | } |
10f3cd15 | 1821 | |
12aa4082 HR |
1822 | QTAILQ_INIT(&s->ops_in_flight); |
1823 | ||
5ccac6f1 | 1824 | trace_mirror_start(bs, s, opaque); |
da01ff7f | 1825 | job_start(&s->common.job); |
cc19f177 VSO |
1826 | |
1827 | return &s->common; | |
4ef85a9c KW |
1828 | |
1829 | fail: | |
1830 | if (s) { | |
7a25fcd0 HR |
1831 | /* Make sure this BDS does not go away until we have completed the graph |
1832 | * changes below */ | |
1833 | bdrv_ref(mirror_top_bs); | |
1834 | ||
4ef85a9c KW |
1835 | g_free(s->replaces); |
1836 | blk_unref(s->target); | |
429076e8 | 1837 | bs_opaque->job = NULL; |
e917e2cb | 1838 | if (s->dirty_bitmap) { |
5deb6cbd | 1839 | bdrv_release_dirty_bitmap(s->dirty_bitmap); |
e917e2cb | 1840 | } |
4ad35181 | 1841 | job_early_fail(&s->common.job); |
4ef85a9c KW |
1842 | } |
1843 | ||
f94dc3b4 HR |
1844 | bs_opaque->stop = true; |
1845 | bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing, | |
1846 | &error_abort); | |
3f072a7f | 1847 | bdrv_replace_node(mirror_top_bs, mirror_top_bs->backing->bs, &error_abort); |
7a25fcd0 HR |
1848 | |
1849 | bdrv_unref(mirror_top_bs); | |
cc19f177 VSO |
1850 | |
1851 | return NULL; | |
893f7eba | 1852 | } |
03544a6e | 1853 | |
71aa9867 AG |
1854 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1855 | BlockDriverState *target, const char *replaces, | |
a1999b33 JS |
1856 | int creation_flags, int64_t speed, |
1857 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1858 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
cdf3bc93 | 1859 | bool zero_target, |
274fccee | 1860 | BlockdevOnError on_source_error, |
03544a6e | 1861 | BlockdevOnError on_target_error, |
481debaa HR |
1862 | bool unmap, const char *filter_node_name, |
1863 | MirrorCopyMode copy_mode, Error **errp) | |
03544a6e FZ |
1864 | { |
1865 | bool is_none_mode; | |
1866 | BlockDriverState *base; | |
1867 | ||
c8b56501 JS |
1868 | if ((mode == MIRROR_SYNC_MODE_INCREMENTAL) || |
1869 | (mode == MIRROR_SYNC_MODE_BITMAP)) { | |
1870 | error_setg(errp, "Sync mode '%s' not supported", | |
1871 | MirrorSyncMode_str(mode)); | |
d58d8453 JS |
1872 | return; |
1873 | } | |
03544a6e | 1874 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
3f072a7f | 1875 | base = mode == MIRROR_SYNC_MODE_TOP ? bdrv_backing_chain_next(bs) : NULL; |
a1999b33 | 1876 | mirror_start_job(job_id, bs, creation_flags, target, replaces, |
cdf3bc93 | 1877 | speed, granularity, buf_size, backing_mode, zero_target, |
51ccfa2d | 1878 | on_source_error, on_target_error, unmap, NULL, NULL, |
6cdbceb1 | 1879 | &mirror_job_driver, is_none_mode, base, false, |
481debaa | 1880 | filter_node_name, true, copy_mode, errp); |
03544a6e FZ |
1881 | } |
1882 | ||
cc19f177 VSO |
1883 | BlockJob *commit_active_start(const char *job_id, BlockDriverState *bs, |
1884 | BlockDriverState *base, int creation_flags, | |
1885 | int64_t speed, BlockdevOnError on_error, | |
1886 | const char *filter_node_name, | |
1887 | BlockCompletionFunc *cb, void *opaque, | |
1888 | bool auto_complete, Error **errp) | |
03544a6e | 1889 | { |
1ba79388 | 1890 | bool base_read_only; |
eb5becc1 | 1891 | BlockJob *job; |
4da83585 | 1892 | |
1ba79388 | 1893 | base_read_only = bdrv_is_read_only(base); |
4da83585 | 1894 | |
1ba79388 AG |
1895 | if (base_read_only) { |
1896 | if (bdrv_reopen_set_read_only(base, false, errp) < 0) { | |
cc19f177 | 1897 | return NULL; |
1ba79388 | 1898 | } |
20a63d2c | 1899 | } |
4da83585 | 1900 | |
eb5becc1 | 1901 | job = mirror_start_job( |
cc19f177 | 1902 | job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
cdf3bc93 | 1903 | MIRROR_LEAVE_BACKING_CHAIN, false, |
51ccfa2d | 1904 | on_error, on_error, true, cb, opaque, |
6cdbceb1 | 1905 | &commit_active_job_driver, false, base, auto_complete, |
481debaa | 1906 | filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND, |
eb5becc1 VSO |
1907 | errp); |
1908 | if (!job) { | |
4da83585 JC |
1909 | goto error_restore_flags; |
1910 | } | |
1911 | ||
eb5becc1 | 1912 | return job; |
4da83585 JC |
1913 | |
1914 | error_restore_flags: | |
1915 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1916 | * the original error */ | |
1ba79388 AG |
1917 | if (base_read_only) { |
1918 | bdrv_reopen_set_read_only(base, true, NULL); | |
1919 | } | |
cc19f177 | 1920 | return NULL; |
03544a6e | 1921 | } |