]>
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" |
e2c1c34f | 21 | #include "block/dirty-bitmap.h" |
373340b2 | 22 | #include "sysemu/block-backend.h" |
da34e65c | 23 | #include "qapi/error.h" |
893f7eba | 24 | #include "qemu/ratelimit.h" |
b812f671 | 25 | #include "qemu/bitmap.h" |
5df022cf | 26 | #include "qemu/memalign.h" |
893f7eba | 27 | |
402a4741 | 28 | #define MAX_IN_FLIGHT 16 |
b436982f EB |
29 | #define MAX_IO_BYTES (1 << 20) /* 1 Mb */ |
30 | #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES) | |
402a4741 PB |
31 | |
32 | /* The mirroring buffer is a list of granularity-sized chunks. | |
33 | * Free chunks are organized in a list. | |
34 | */ | |
35 | typedef struct MirrorBuffer { | |
36 | QSIMPLEQ_ENTRY(MirrorBuffer) next; | |
37 | } MirrorBuffer; | |
893f7eba | 38 | |
12aa4082 HR |
39 | typedef struct MirrorOp MirrorOp; |
40 | ||
893f7eba PB |
41 | typedef struct MirrorBlockJob { |
42 | BlockJob common; | |
e253f4b8 | 43 | BlockBackend *target; |
4ef85a9c | 44 | BlockDriverState *mirror_top_bs; |
5bc361b8 | 45 | BlockDriverState *base; |
3f072a7f | 46 | BlockDriverState *base_overlay; |
4ef85a9c | 47 | |
09158f00 BC |
48 | /* The name of the graph node to replace */ |
49 | char *replaces; | |
50 | /* The BDS to replace */ | |
51 | BlockDriverState *to_replace; | |
52 | /* Used to block operations on the drive-mirror-replace target */ | |
53 | Error *replace_blocker; | |
03544a6e | 54 | bool is_none_mode; |
274fccee | 55 | BlockMirrorBackingMode backing_mode; |
cdf3bc93 HR |
56 | /* Whether the target image requires explicit zero-initialization */ |
57 | bool zero_target; | |
2d400d15 FE |
58 | /* |
59 | * To be accesssed with atomics. Written only under the BQL (required by the | |
60 | * current implementation of mirror_change()). | |
61 | */ | |
d06107ad | 62 | MirrorCopyMode copy_mode; |
b952b558 | 63 | BlockdevOnError on_source_error, on_target_error; |
76cb2f24 FE |
64 | /* |
65 | * To be accessed with atomics. | |
66 | * | |
67 | * Set when the target is synced (dirty bitmap is clean, nothing in flight) | |
68 | * and the job is running in active mode. | |
69 | */ | |
d06107ad | 70 | bool actively_synced; |
d63ffd87 | 71 | bool should_complete; |
eee13dfe | 72 | int64_t granularity; |
b812f671 | 73 | size_t buf_size; |
b21c7652 | 74 | int64_t bdev_length; |
b812f671 | 75 | unsigned long *cow_bitmap; |
e4654d2d | 76 | BdrvDirtyBitmap *dirty_bitmap; |
dc162c8e | 77 | BdrvDirtyBitmapIter *dbi; |
893f7eba | 78 | uint8_t *buf; |
402a4741 PB |
79 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
80 | int buf_free_count; | |
bd48bde8 | 81 | |
49efb1f5 | 82 | uint64_t last_pause_ns; |
402a4741 | 83 | unsigned long *in_flight_bitmap; |
1b8f7776 | 84 | unsigned in_flight; |
b436982f | 85 | int64_t bytes_in_flight; |
b58deb34 | 86 | QTAILQ_HEAD(, MirrorOp) ops_in_flight; |
bd48bde8 | 87 | int ret; |
0fc9f8ea | 88 | bool unmap; |
b436982f | 89 | int target_cluster_size; |
e5b43573 | 90 | int max_iov; |
90ab48eb | 91 | bool initial_zeroing_ongoing; |
d06107ad | 92 | int in_active_write_counter; |
d69a879b | 93 | int64_t active_write_bytes_in_flight; |
737efc1e | 94 | bool prepared; |
5e771752 | 95 | bool in_drain; |
893f7eba PB |
96 | } MirrorBlockJob; |
97 | ||
429076e8 HR |
98 | typedef struct MirrorBDSOpaque { |
99 | MirrorBlockJob *job; | |
f94dc3b4 | 100 | bool stop; |
53431b90 | 101 | bool is_commit; |
429076e8 HR |
102 | } MirrorBDSOpaque; |
103 | ||
12aa4082 | 104 | struct MirrorOp { |
bd48bde8 PB |
105 | MirrorBlockJob *s; |
106 | QEMUIOVector qiov; | |
b436982f EB |
107 | int64_t offset; |
108 | uint64_t bytes; | |
2e1990b2 HR |
109 | |
110 | /* The pointee is set by mirror_co_read(), mirror_co_zero(), and | |
111 | * mirror_co_discard() before yielding for the first time */ | |
112 | int64_t *bytes_handled; | |
12aa4082 | 113 | |
1181e19a | 114 | bool is_pseudo_op; |
d06107ad | 115 | bool is_active_write; |
ce8cabbd | 116 | bool is_in_flight; |
12aa4082 | 117 | CoQueue waiting_requests; |
eed325b9 | 118 | Coroutine *co; |
d44dae1a | 119 | MirrorOp *waiting_for_op; |
12aa4082 HR |
120 | |
121 | QTAILQ_ENTRY(MirrorOp) next; | |
122 | }; | |
bd48bde8 | 123 | |
4295c5fc HR |
124 | typedef enum MirrorMethod { |
125 | MIRROR_METHOD_COPY, | |
126 | MIRROR_METHOD_ZERO, | |
127 | MIRROR_METHOD_DISCARD, | |
128 | } MirrorMethod; | |
129 | ||
b952b558 PB |
130 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
131 | int error) | |
132 | { | |
76cb2f24 | 133 | qatomic_set(&s->actively_synced, false); |
b952b558 | 134 | if (read) { |
81e254dc KW |
135 | return block_job_error_action(&s->common, s->on_source_error, |
136 | true, error); | |
b952b558 | 137 | } else { |
81e254dc KW |
138 | return block_job_error_action(&s->common, s->on_target_error, |
139 | false, error); | |
b952b558 PB |
140 | } |
141 | } | |
142 | ||
1181e19a HR |
143 | static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self, |
144 | MirrorBlockJob *s, | |
145 | uint64_t offset, | |
146 | uint64_t bytes) | |
147 | { | |
148 | uint64_t self_start_chunk = offset / s->granularity; | |
149 | uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); | |
150 | uint64_t self_nb_chunks = self_end_chunk - self_start_chunk; | |
151 | ||
152 | while (find_next_bit(s->in_flight_bitmap, self_end_chunk, | |
153 | self_start_chunk) < self_end_chunk && | |
154 | s->ret >= 0) | |
155 | { | |
156 | MirrorOp *op; | |
157 | ||
158 | QTAILQ_FOREACH(op, &s->ops_in_flight, next) { | |
159 | uint64_t op_start_chunk = op->offset / s->granularity; | |
160 | uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes, | |
161 | s->granularity) - | |
162 | op_start_chunk; | |
163 | ||
164 | if (op == self) { | |
165 | continue; | |
166 | } | |
167 | ||
168 | if (ranges_overlap(self_start_chunk, self_nb_chunks, | |
169 | op_start_chunk, op_nb_chunks)) | |
170 | { | |
66fed30c SG |
171 | if (self) { |
172 | /* | |
173 | * If the operation is already (indirectly) waiting for us, | |
174 | * or will wait for us as soon as it wakes up, then just go | |
175 | * on (instead of producing a deadlock in the former case). | |
176 | */ | |
177 | if (op->waiting_for_op) { | |
178 | continue; | |
179 | } | |
180 | ||
181 | self->waiting_for_op = op; | |
d44dae1a VSO |
182 | } |
183 | ||
1181e19a | 184 | qemu_co_queue_wait(&op->waiting_requests, NULL); |
66fed30c SG |
185 | |
186 | if (self) { | |
187 | self->waiting_for_op = NULL; | |
188 | } | |
189 | ||
1181e19a HR |
190 | break; |
191 | } | |
192 | } | |
193 | } | |
194 | } | |
195 | ||
2e1990b2 | 196 | static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) |
bd48bde8 PB |
197 | { |
198 | MirrorBlockJob *s = op->s; | |
402a4741 | 199 | struct iovec *iov; |
bd48bde8 | 200 | int64_t chunk_num; |
b436982f | 201 | int i, nb_chunks; |
bd48bde8 | 202 | |
b436982f | 203 | trace_mirror_iteration_done(s, op->offset, op->bytes, ret); |
bd48bde8 PB |
204 | |
205 | s->in_flight--; | |
b436982f | 206 | s->bytes_in_flight -= op->bytes; |
402a4741 PB |
207 | iov = op->qiov.iov; |
208 | for (i = 0; i < op->qiov.niov; i++) { | |
209 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
210 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
211 | s->buf_free_count++; | |
212 | } | |
213 | ||
b436982f EB |
214 | chunk_num = op->offset / s->granularity; |
215 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
12aa4082 | 216 | |
402a4741 | 217 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
12aa4082 | 218 | QTAILQ_REMOVE(&s->ops_in_flight, op, next); |
b21c7652 HR |
219 | if (ret >= 0) { |
220 | if (s->cow_bitmap) { | |
221 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
222 | } | |
90ab48eb | 223 | if (!s->initial_zeroing_ongoing) { |
30a5c887 | 224 | job_progress_update(&s->common.job, op->bytes); |
90ab48eb | 225 | } |
bd48bde8 | 226 | } |
6df3bf8e | 227 | qemu_iovec_destroy(&op->qiov); |
7b770c72 | 228 | |
12aa4082 HR |
229 | qemu_co_queue_restart_all(&op->waiting_requests); |
230 | g_free(op); | |
bd48bde8 PB |
231 | } |
232 | ||
2e1990b2 | 233 | static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret) |
bd48bde8 | 234 | { |
bd48bde8 | 235 | MirrorBlockJob *s = op->s; |
b9e413dd | 236 | |
bd48bde8 | 237 | if (ret < 0) { |
bd48bde8 PB |
238 | BlockErrorAction action; |
239 | ||
e0d7f73e | 240 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 241 | action = mirror_error_action(s, false, -ret); |
a589569f | 242 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
243 | s->ret = ret; |
244 | } | |
245 | } | |
d12ade57 | 246 | |
bd48bde8 PB |
247 | mirror_iteration_done(op, ret); |
248 | } | |
249 | ||
2e1990b2 | 250 | static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret) |
bd48bde8 | 251 | { |
bd48bde8 | 252 | MirrorBlockJob *s = op->s; |
b9e413dd | 253 | |
bd48bde8 | 254 | if (ret < 0) { |
bd48bde8 PB |
255 | BlockErrorAction action; |
256 | ||
e0d7f73e | 257 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 258 | action = mirror_error_action(s, true, -ret); |
a589569f | 259 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
260 | s->ret = ret; |
261 | } | |
262 | ||
263 | mirror_iteration_done(op, ret); | |
d12ade57 | 264 | return; |
bd48bde8 | 265 | } |
d12ade57 VSO |
266 | |
267 | ret = blk_co_pwritev(s->target, op->offset, op->qiov.size, &op->qiov, 0); | |
268 | mirror_write_complete(op, ret); | |
bd48bde8 PB |
269 | } |
270 | ||
782d97ef EB |
271 | /* Clip bytes relative to offset to not exceed end-of-file */ |
272 | static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, | |
273 | int64_t offset, | |
274 | int64_t bytes) | |
275 | { | |
276 | return MIN(bytes, s->bdev_length - offset); | |
277 | } | |
278 | ||
782d97ef EB |
279 | /* Round offset and/or bytes to target cluster if COW is needed, and |
280 | * return the offset of the adjusted tail against original. */ | |
17ac39c3 PB |
281 | static int coroutine_fn mirror_cow_align(MirrorBlockJob *s, int64_t *offset, |
282 | uint64_t *bytes) | |
893f7eba | 283 | { |
e5b43573 FZ |
284 | bool need_cow; |
285 | int ret = 0; | |
782d97ef | 286 | int64_t align_offset = *offset; |
7cfd5275 | 287 | int64_t align_bytes = *bytes; |
782d97ef | 288 | int max_bytes = s->granularity * s->max_iov; |
e5b43573 | 289 | |
782d97ef EB |
290 | need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); |
291 | need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, | |
e5b43573 FZ |
292 | s->cow_bitmap); |
293 | if (need_cow) { | |
fc6b211f AD |
294 | bdrv_round_to_subclusters(blk_bs(s->target), *offset, *bytes, |
295 | &align_offset, &align_bytes); | |
e5b43573 | 296 | } |
3515727f | 297 | |
782d97ef EB |
298 | if (align_bytes > max_bytes) { |
299 | align_bytes = max_bytes; | |
e5b43573 | 300 | if (need_cow) { |
782d97ef | 301 | align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); |
e5b43573 | 302 | } |
8f0720ec | 303 | } |
782d97ef | 304 | /* Clipping may result in align_bytes unaligned to chunk boundary, but |
4150ae60 | 305 | * that doesn't matter because it's already the end of source image. */ |
782d97ef | 306 | align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); |
8f0720ec | 307 | |
782d97ef EB |
308 | ret = align_offset + align_bytes - (*offset + *bytes); |
309 | *offset = align_offset; | |
310 | *bytes = align_bytes; | |
e5b43573 FZ |
311 | assert(ret >= 0); |
312 | return ret; | |
313 | } | |
314 | ||
537c3d4f | 315 | static inline void coroutine_fn |
eb994912 | 316 | mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) |
21cd917f | 317 | { |
12aa4082 HR |
318 | MirrorOp *op; |
319 | ||
1181e19a | 320 | QTAILQ_FOREACH(op, &s->ops_in_flight, next) { |
eb994912 HR |
321 | /* |
322 | * Do not wait on pseudo ops, because it may in turn wait on | |
1181e19a HR |
323 | * some other operation to start, which may in fact be the |
324 | * caller of this function. Since there is only one pseudo op | |
325 | * at any given time, we will always find some real operation | |
eb994912 HR |
326 | * to wait on. |
327 | * Also, do not wait on active operations, because they do not | |
328 | * use up in-flight slots. | |
329 | */ | |
330 | if (!op->is_pseudo_op && op->is_in_flight && !op->is_active_write) { | |
1181e19a HR |
331 | qemu_co_queue_wait(&op->waiting_requests, NULL); |
332 | return; | |
333 | } | |
334 | } | |
335 | abort(); | |
21cd917f FZ |
336 | } |
337 | ||
2e1990b2 HR |
338 | /* Perform a mirror copy operation. |
339 | * | |
340 | * *op->bytes_handled is set to the number of bytes copied after and | |
341 | * including offset, excluding any bytes copied prior to offset due | |
342 | * to alignment. This will be op->bytes if no alignment is necessary, | |
343 | * or (new_end - op->offset) if the tail is rounded up or down due to | |
344 | * alignment or buffer limit. | |
e5b43573 | 345 | */ |
2e1990b2 | 346 | static void coroutine_fn mirror_co_read(void *opaque) |
e5b43573 | 347 | { |
2e1990b2 HR |
348 | MirrorOp *op = opaque; |
349 | MirrorBlockJob *s = op->s; | |
ae4cc877 EB |
350 | int nb_chunks; |
351 | uint64_t ret; | |
ae4cc877 | 352 | uint64_t max_bytes; |
e5b43573 | 353 | |
ae4cc877 | 354 | max_bytes = s->granularity * s->max_iov; |
402a4741 | 355 | |
e5b43573 | 356 | /* We can only handle as much as buf_size at a time. */ |
2e1990b2 HR |
357 | op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes)); |
358 | assert(op->bytes); | |
359 | assert(op->bytes < BDRV_REQUEST_MAX_BYTES); | |
360 | *op->bytes_handled = op->bytes; | |
402a4741 | 361 | |
e5b43573 | 362 | if (s->cow_bitmap) { |
2e1990b2 | 363 | *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes); |
e5b43573 | 364 | } |
2e1990b2 HR |
365 | /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */ |
366 | assert(*op->bytes_handled <= UINT_MAX); | |
367 | assert(op->bytes <= s->buf_size); | |
ae4cc877 | 368 | /* The offset is granularity-aligned because: |
e5b43573 FZ |
369 | * 1) Caller passes in aligned values; |
370 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
2e1990b2 | 371 | assert(QEMU_IS_ALIGNED(op->offset, s->granularity)); |
ae4cc877 | 372 | /* The range is sector-aligned, since bdrv_getlength() rounds up. */ |
2e1990b2 HR |
373 | assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE)); |
374 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
e5b43573 FZ |
375 | |
376 | while (s->buf_free_count < nb_chunks) { | |
2e1990b2 | 377 | trace_mirror_yield_in_flight(s, op->offset, s->in_flight); |
9178f4fe | 378 | mirror_wait_for_free_in_flight_slot(s); |
b812f671 PB |
379 | } |
380 | ||
402a4741 PB |
381 | /* Now make a QEMUIOVector taking enough granularity-sized chunks |
382 | * from s->buf_free. | |
383 | */ | |
384 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
385 | while (nb_chunks-- > 0) { |
386 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
2e1990b2 | 387 | size_t remaining = op->bytes - op->qiov.size; |
5a0f6fd5 | 388 | |
402a4741 PB |
389 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
390 | s->buf_free_count--; | |
5a0f6fd5 | 391 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 392 | } |
bd48bde8 | 393 | |
893f7eba | 394 | /* Copy the dirty cluster. */ |
bd48bde8 | 395 | s->in_flight++; |
2e1990b2 | 396 | s->bytes_in_flight += op->bytes; |
ce8cabbd | 397 | op->is_in_flight = true; |
2e1990b2 | 398 | trace_mirror_one_iteration(s, op->offset, op->bytes); |
dcfb3beb | 399 | |
b9b10c35 KW |
400 | WITH_GRAPH_RDLOCK_GUARD() { |
401 | ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes, | |
402 | &op->qiov, 0); | |
403 | } | |
2e1990b2 | 404 | mirror_read_complete(op, ret); |
e5b43573 FZ |
405 | } |
406 | ||
2e1990b2 | 407 | static void coroutine_fn mirror_co_zero(void *opaque) |
e5b43573 | 408 | { |
2e1990b2 HR |
409 | MirrorOp *op = opaque; |
410 | int ret; | |
e5b43573 | 411 | |
2e1990b2 HR |
412 | op->s->in_flight++; |
413 | op->s->bytes_in_flight += op->bytes; | |
414 | *op->bytes_handled = op->bytes; | |
ce8cabbd | 415 | op->is_in_flight = true; |
e5b43573 | 416 | |
2e1990b2 HR |
417 | ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes, |
418 | op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0); | |
419 | mirror_write_complete(op, ret); | |
420 | } | |
421 | ||
422 | static void coroutine_fn mirror_co_discard(void *opaque) | |
423 | { | |
424 | MirrorOp *op = opaque; | |
425 | int ret; | |
426 | ||
427 | op->s->in_flight++; | |
428 | op->s->bytes_in_flight += op->bytes; | |
429 | *op->bytes_handled = op->bytes; | |
ce8cabbd | 430 | op->is_in_flight = true; |
2e1990b2 HR |
431 | |
432 | ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes); | |
433 | mirror_write_complete(op, ret); | |
e5b43573 FZ |
434 | } |
435 | ||
4295c5fc HR |
436 | static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, |
437 | unsigned bytes, MirrorMethod mirror_method) | |
438 | { | |
2e1990b2 HR |
439 | MirrorOp *op; |
440 | Coroutine *co; | |
441 | int64_t bytes_handled = -1; | |
442 | ||
443 | op = g_new(MirrorOp, 1); | |
444 | *op = (MirrorOp){ | |
445 | .s = s, | |
446 | .offset = offset, | |
447 | .bytes = bytes, | |
448 | .bytes_handled = &bytes_handled, | |
449 | }; | |
12aa4082 | 450 | qemu_co_queue_init(&op->waiting_requests); |
2e1990b2 | 451 | |
4295c5fc HR |
452 | switch (mirror_method) { |
453 | case MIRROR_METHOD_COPY: | |
2e1990b2 HR |
454 | co = qemu_coroutine_create(mirror_co_read, op); |
455 | break; | |
4295c5fc | 456 | case MIRROR_METHOD_ZERO: |
2e1990b2 HR |
457 | co = qemu_coroutine_create(mirror_co_zero, op); |
458 | break; | |
4295c5fc | 459 | case MIRROR_METHOD_DISCARD: |
2e1990b2 HR |
460 | co = qemu_coroutine_create(mirror_co_discard, op); |
461 | break; | |
4295c5fc HR |
462 | default: |
463 | abort(); | |
464 | } | |
eed325b9 | 465 | op->co = co; |
2e1990b2 | 466 | |
12aa4082 | 467 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); |
2e1990b2 HR |
468 | qemu_coroutine_enter(co); |
469 | /* At this point, ownership of op has been moved to the coroutine | |
470 | * and the object may already be freed */ | |
471 | ||
472 | /* Assert that this value has been set */ | |
473 | assert(bytes_handled >= 0); | |
474 | ||
475 | /* Same assertion as in mirror_co_read() (and for mirror_co_read() | |
476 | * and mirror_co_discard(), bytes_handled == op->bytes, which | |
477 | * is the @bytes parameter given to this function) */ | |
478 | assert(bytes_handled <= UINT_MAX); | |
479 | return bytes_handled; | |
4295c5fc HR |
480 | } |
481 | ||
018e5987 | 482 | static void coroutine_fn mirror_iteration(MirrorBlockJob *s) |
e5b43573 | 483 | { |
138f9fff | 484 | BlockDriverState *source = s->mirror_top_bs->backing->bs; |
1181e19a HR |
485 | MirrorOp *pseudo_op; |
486 | int64_t offset; | |
e5b43573 FZ |
487 | /* At least the first dirty chunk is mirrored in one iteration. */ |
488 | int nb_chunks = 1; | |
4b5004d9 | 489 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
b436982f | 490 | int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); |
e5b43573 | 491 | |
b64bd51e | 492 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
f798184c | 493 | offset = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 494 | if (offset < 0) { |
dc162c8e | 495 | bdrv_set_dirty_iter(s->dbi, 0); |
f798184c | 496 | offset = bdrv_dirty_iter_next(s->dbi); |
9a46dba7 | 497 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
fb2ef791 | 498 | assert(offset >= 0); |
e5b43573 | 499 | } |
b64bd51e | 500 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
e5b43573 | 501 | |
d69a879b HR |
502 | /* |
503 | * Wait for concurrent requests to @offset. The next loop will limit the | |
504 | * copied area based on in_flight_bitmap so we only copy an area that does | |
505 | * not overlap with concurrent in-flight requests. Still, we would like to | |
506 | * copy something, so wait until there are at least no more requests to the | |
507 | * very beginning of the area. | |
508 | */ | |
1181e19a | 509 | mirror_wait_on_conflicts(NULL, s, offset, 1); |
9c83625b | 510 | |
da01ff7f | 511 | job_pause_point(&s->common.job); |
565ac01f | 512 | |
3202d8e4 | 513 | /* Find the number of consecutive dirty chunks following the first dirty |
e5b43573 | 514 | * one, and wait for in flight requests in them. */ |
b64bd51e | 515 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
fb2ef791 | 516 | while (nb_chunks * s->granularity < s->buf_size) { |
dc162c8e | 517 | int64_t next_dirty; |
fb2ef791 EB |
518 | int64_t next_offset = offset + nb_chunks * s->granularity; |
519 | int64_t next_chunk = next_offset / s->granularity; | |
520 | if (next_offset >= s->bdev_length || | |
28636b82 | 521 | !bdrv_dirty_bitmap_get_locked(s->dirty_bitmap, next_offset)) { |
e5b43573 FZ |
522 | break; |
523 | } | |
524 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 525 | break; |
e5b43573 | 526 | } |
9c83625b | 527 | |
f798184c | 528 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 529 | if (next_dirty > next_offset || next_dirty < 0) { |
f27a2742 | 530 | /* The bitmap iterator's cache is stale, refresh it */ |
715a74d8 | 531 | bdrv_set_dirty_iter(s->dbi, next_offset); |
f798184c | 532 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
f27a2742 | 533 | } |
fb2ef791 | 534 | assert(next_dirty == next_offset); |
9c83625b | 535 | nb_chunks++; |
e5b43573 FZ |
536 | } |
537 | ||
538 | /* Clear dirty bits before querying the block status, because | |
31826642 | 539 | * calling bdrv_block_status_above could yield - if some blocks are |
e5b43573 FZ |
540 | * marked dirty in this window, we need to know. |
541 | */ | |
e0d7f73e EB |
542 | bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, |
543 | nb_chunks * s->granularity); | |
b64bd51e PB |
544 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
545 | ||
1181e19a HR |
546 | /* Before claiming an area in the in-flight bitmap, we have to |
547 | * create a MirrorOp for it so that conflicting requests can wait | |
548 | * for it. mirror_perform() will create the real MirrorOps later, | |
549 | * for now we just create a pseudo operation that will wake up all | |
550 | * conflicting requests once all real operations have been | |
551 | * launched. */ | |
552 | pseudo_op = g_new(MirrorOp, 1); | |
553 | *pseudo_op = (MirrorOp){ | |
554 | .offset = offset, | |
555 | .bytes = nb_chunks * s->granularity, | |
556 | .is_pseudo_op = true, | |
557 | }; | |
558 | qemu_co_queue_init(&pseudo_op->waiting_requests); | |
559 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next); | |
560 | ||
fb2ef791 EB |
561 | bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); |
562 | while (nb_chunks > 0 && offset < s->bdev_length) { | |
31826642 | 563 | int ret; |
7cfd5275 | 564 | int64_t io_bytes; |
f3e4ce4a | 565 | int64_t io_bytes_acct; |
4295c5fc | 566 | MirrorMethod mirror_method = MIRROR_METHOD_COPY; |
e5b43573 | 567 | |
fb2ef791 | 568 | assert(!(offset % s->granularity)); |
7ff9579e | 569 | WITH_GRAPH_RDLOCK_GUARD() { |
cc323997 PB |
570 | ret = bdrv_co_block_status_above(source, NULL, offset, |
571 | nb_chunks * s->granularity, | |
572 | &io_bytes, NULL, NULL); | |
7ff9579e | 573 | } |
e5b43573 | 574 | if (ret < 0) { |
fb2ef791 | 575 | io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); |
0965a41e | 576 | } else if (ret & BDRV_BLOCK_DATA) { |
fb2ef791 | 577 | io_bytes = MIN(io_bytes, max_io_bytes); |
e5b43573 FZ |
578 | } |
579 | ||
fb2ef791 EB |
580 | io_bytes -= io_bytes % s->granularity; |
581 | if (io_bytes < s->granularity) { | |
582 | io_bytes = s->granularity; | |
e5b43573 | 583 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { |
fb2ef791 | 584 | int64_t target_offset; |
7cfd5275 | 585 | int64_t target_bytes; |
a00e70c0 | 586 | WITH_GRAPH_RDLOCK_GUARD() { |
fc6b211f AD |
587 | bdrv_round_to_subclusters(blk_bs(s->target), offset, io_bytes, |
588 | &target_offset, &target_bytes); | |
a00e70c0 | 589 | } |
fb2ef791 EB |
590 | if (target_offset == offset && |
591 | target_bytes == io_bytes) { | |
e5b43573 FZ |
592 | mirror_method = ret & BDRV_BLOCK_ZERO ? |
593 | MIRROR_METHOD_ZERO : | |
594 | MIRROR_METHOD_DISCARD; | |
595 | } | |
596 | } | |
597 | ||
cf56a3c6 | 598 | while (s->in_flight >= MAX_IN_FLIGHT) { |
fb2ef791 | 599 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
9178f4fe | 600 | mirror_wait_for_free_in_flight_slot(s); |
cf56a3c6 DL |
601 | } |
602 | ||
dbaa7b57 | 603 | if (s->ret < 0) { |
1181e19a HR |
604 | ret = 0; |
605 | goto fail; | |
dbaa7b57 VSO |
606 | } |
607 | ||
fb2ef791 | 608 | io_bytes = mirror_clip_bytes(s, offset, io_bytes); |
4295c5fc HR |
609 | io_bytes = mirror_perform(s, offset, io_bytes, mirror_method); |
610 | if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) { | |
611 | io_bytes_acct = 0; | |
612 | } else { | |
613 | io_bytes_acct = io_bytes; | |
e5b43573 | 614 | } |
fb2ef791 EB |
615 | assert(io_bytes); |
616 | offset += io_bytes; | |
617 | nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); | |
018e5987 | 618 | block_job_ratelimit_processed_bytes(&s->common, io_bytes_acct); |
dcfb3beb | 619 | } |
1181e19a | 620 | |
1181e19a HR |
621 | fail: |
622 | QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next); | |
623 | qemu_co_queue_restart_all(&pseudo_op->waiting_requests); | |
624 | g_free(pseudo_op); | |
bd48bde8 | 625 | } |
b952b558 | 626 | |
402a4741 PB |
627 | static void mirror_free_init(MirrorBlockJob *s) |
628 | { | |
629 | int granularity = s->granularity; | |
630 | size_t buf_size = s->buf_size; | |
631 | uint8_t *buf = s->buf; | |
632 | ||
633 | assert(s->buf_free_count == 0); | |
634 | QSIMPLEQ_INIT(&s->buf_free); | |
635 | while (buf_size != 0) { | |
636 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
637 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
638 | s->buf_free_count++; | |
639 | buf_size -= granularity; | |
640 | buf += granularity; | |
641 | } | |
642 | } | |
643 | ||
bae8196d PB |
644 | /* This is also used for the .pause callback. There is no matching |
645 | * mirror_resume() because mirror_run() will begin iterating again | |
646 | * when the job is resumed. | |
647 | */ | |
537c3d4f | 648 | static void coroutine_fn mirror_wait_for_all_io(MirrorBlockJob *s) |
bd48bde8 PB |
649 | { |
650 | while (s->in_flight > 0) { | |
9178f4fe | 651 | mirror_wait_for_free_in_flight_slot(s); |
bd48bde8 | 652 | } |
893f7eba PB |
653 | } |
654 | ||
737efc1e JS |
655 | /** |
656 | * mirror_exit_common: handle both abort() and prepare() cases. | |
657 | * for .prepare, returns 0 on success and -errno on failure. | |
658 | * for .abort cases, denoted by abort = true, MUST return 0. | |
659 | */ | |
660 | static int mirror_exit_common(Job *job) | |
5a7e7a0b | 661 | { |
1908a559 KW |
662 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
663 | BlockJob *bjob = &s->common; | |
f93c3add | 664 | MirrorBDSOpaque *bs_opaque; |
5a7e7a0b | 665 | AioContext *replace_aio_context = NULL; |
f93c3add HR |
666 | BlockDriverState *src; |
667 | BlockDriverState *target_bs; | |
668 | BlockDriverState *mirror_top_bs; | |
12fa4af6 | 669 | Error *local_err = NULL; |
737efc1e JS |
670 | bool abort = job->ret < 0; |
671 | int ret = 0; | |
672 | ||
2626d27f KW |
673 | GLOBAL_STATE_CODE(); |
674 | ||
737efc1e JS |
675 | if (s->prepared) { |
676 | return 0; | |
677 | } | |
678 | s->prepared = true; | |
3f09bfbc | 679 | |
2626d27f | 680 | aio_context_acquire(qemu_get_aio_context()); |
9275fc72 | 681 | bdrv_graph_rdlock_main_loop(); |
2626d27f | 682 | |
f93c3add HR |
683 | mirror_top_bs = s->mirror_top_bs; |
684 | bs_opaque = mirror_top_bs->opaque; | |
685 | src = mirror_top_bs->backing->bs; | |
686 | target_bs = blk_bs(s->target); | |
687 | ||
ef53dc09 AG |
688 | if (bdrv_chain_contains(src, target_bs)) { |
689 | bdrv_unfreeze_backing_chain(mirror_top_bs, target_bs); | |
690 | } | |
691 | ||
5deb6cbd | 692 | bdrv_release_dirty_bitmap(s->dirty_bitmap); |
2119882c | 693 | |
7b508f6b JS |
694 | /* Make sure that the source BDS doesn't go away during bdrv_replace_node, |
695 | * before we can call bdrv_drained_end */ | |
3f09bfbc | 696 | bdrv_ref(src); |
4ef85a9c | 697 | bdrv_ref(mirror_top_bs); |
7d9fcb39 KW |
698 | bdrv_ref(target_bs); |
699 | ||
9275fc72 KW |
700 | bdrv_graph_rdunlock_main_loop(); |
701 | ||
bb0c9409 VSO |
702 | /* |
703 | * Remove target parent that still uses BLK_PERM_WRITE/RESIZE before | |
7d9fcb39 | 704 | * inserting target_bs at s->to_replace, where we might not be able to get |
63c8ef28 | 705 | * these permissions. |
bb0c9409 | 706 | */ |
7d9fcb39 KW |
707 | blk_unref(s->target); |
708 | s->target = NULL; | |
4ef85a9c KW |
709 | |
710 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
d2da5e28 KW |
711 | * required before it could become a backing file of target_bs. Not having |
712 | * these permissions any more means that we can't allow any new requests on | |
713 | * mirror_top_bs from now on, so keep it drained. */ | |
714 | bdrv_drained_begin(mirror_top_bs); | |
f94dc3b4 | 715 | bs_opaque->stop = true; |
3804e3cf KW |
716 | |
717 | bdrv_graph_rdlock_main_loop(); | |
f94dc3b4 HR |
718 | bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing, |
719 | &error_abort); | |
3804e3cf | 720 | |
737efc1e | 721 | if (!abort && s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { |
4ef85a9c | 722 | BlockDriverState *backing = s->is_none_mode ? src : s->base; |
3f072a7f HR |
723 | BlockDriverState *unfiltered_target = bdrv_skip_filters(target_bs); |
724 | ||
725 | if (bdrv_cow_bs(unfiltered_target) != backing) { | |
726 | bdrv_set_backing_hd(unfiltered_target, backing, &local_err); | |
12fa4af6 KW |
727 | if (local_err) { |
728 | error_report_err(local_err); | |
66c8672d | 729 | local_err = NULL; |
7b508f6b | 730 | ret = -EPERM; |
12fa4af6 | 731 | } |
4ef85a9c | 732 | } |
c41f5b96 HR |
733 | } else if (!abort && s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
734 | assert(!bdrv_backing_chain_next(target_bs)); | |
735 | ret = bdrv_open_backing_file(bdrv_skip_filters(target_bs), NULL, | |
736 | "backing", &local_err); | |
737 | if (ret < 0) { | |
738 | error_report_err(local_err); | |
739 | local_err = NULL; | |
740 | } | |
4ef85a9c | 741 | } |
ad74751f | 742 | bdrv_graph_rdunlock_main_loop(); |
5a7e7a0b SH |
743 | |
744 | if (s->to_replace) { | |
745 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
746 | aio_context_acquire(replace_aio_context); | |
747 | } | |
748 | ||
737efc1e JS |
749 | if (s->should_complete && !abort) { |
750 | BlockDriverState *to_replace = s->to_replace ?: src; | |
1ba79388 | 751 | bool ro = bdrv_is_read_only(to_replace); |
40365552 | 752 | |
1ba79388 AG |
753 | if (ro != bdrv_is_read_only(target_bs)) { |
754 | bdrv_reopen_set_read_only(target_bs, ro, NULL); | |
5a7e7a0b | 755 | } |
b8804815 KW |
756 | |
757 | /* The mirror job has no requests in flight any more, but we need to | |
758 | * drain potential other users of the BDS before changing the graph. */ | |
5e771752 | 759 | assert(s->in_drain); |
e253f4b8 | 760 | bdrv_drained_begin(target_bs); |
6e9cc051 HR |
761 | /* |
762 | * Cannot use check_to_replace_node() here, because that would | |
763 | * check for an op blocker on @to_replace, and we have our own | |
764 | * there. | |
533c6e4e KW |
765 | * |
766 | * TODO Pull out the writer lock from bdrv_replace_node() to here | |
6e9cc051 | 767 | */ |
533c6e4e | 768 | bdrv_graph_rdlock_main_loop(); |
6e9cc051 HR |
769 | if (bdrv_recurse_can_replace(src, to_replace)) { |
770 | bdrv_replace_node(to_replace, target_bs, &local_err); | |
771 | } else { | |
772 | error_setg(&local_err, "Can no longer replace '%s' by '%s', " | |
773 | "because it can no longer be guaranteed that doing so " | |
774 | "would not lead to an abrupt change of visible data", | |
775 | to_replace->node_name, target_bs->node_name); | |
776 | } | |
533c6e4e | 777 | bdrv_graph_rdunlock_main_loop(); |
e253f4b8 | 778 | bdrv_drained_end(target_bs); |
5fe31c25 KW |
779 | if (local_err) { |
780 | error_report_err(local_err); | |
7b508f6b | 781 | ret = -EPERM; |
5fe31c25 | 782 | } |
5a7e7a0b SH |
783 | } |
784 | if (s->to_replace) { | |
785 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
786 | error_free(s->replace_blocker); | |
787 | bdrv_unref(s->to_replace); | |
788 | } | |
789 | if (replace_aio_context) { | |
790 | aio_context_release(replace_aio_context); | |
791 | } | |
792 | g_free(s->replaces); | |
7d9fcb39 | 793 | bdrv_unref(target_bs); |
4ef85a9c | 794 | |
f94dc3b4 HR |
795 | /* |
796 | * Remove the mirror filter driver from the graph. Before this, get rid of | |
4ef85a9c | 797 | * the blockers on the intermediate nodes so that the resulting state is |
f94dc3b4 HR |
798 | * valid. |
799 | */ | |
1908a559 | 800 | block_job_remove_all_bdrv(bjob); |
3f072a7f | 801 | bdrv_replace_node(mirror_top_bs, mirror_top_bs->backing->bs, &error_abort); |
4ef85a9c | 802 | |
429076e8 | 803 | bs_opaque->job = NULL; |
4ef85a9c | 804 | |
176c3699 | 805 | bdrv_drained_end(src); |
d2da5e28 | 806 | bdrv_drained_end(mirror_top_bs); |
5e771752 | 807 | s->in_drain = false; |
4ef85a9c | 808 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 809 | bdrv_unref(src); |
7b508f6b | 810 | |
2626d27f KW |
811 | aio_context_release(qemu_get_aio_context()); |
812 | ||
737efc1e JS |
813 | return ret; |
814 | } | |
815 | ||
816 | static int mirror_prepare(Job *job) | |
817 | { | |
818 | return mirror_exit_common(job); | |
819 | } | |
820 | ||
821 | static void mirror_abort(Job *job) | |
822 | { | |
823 | int ret = mirror_exit_common(job); | |
824 | assert(ret == 0); | |
5a7e7a0b SH |
825 | } |
826 | ||
537c3d4f | 827 | static void coroutine_fn mirror_throttle(MirrorBlockJob *s) |
49efb1f5 DL |
828 | { |
829 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
830 | ||
18bb6928 | 831 | if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) { |
49efb1f5 | 832 | s->last_pause_ns = now; |
5d43e86e | 833 | job_sleep_ns(&s->common.job, 0); |
49efb1f5 | 834 | } else { |
da01ff7f | 835 | job_pause_point(&s->common.job); |
49efb1f5 DL |
836 | } |
837 | } | |
838 | ||
c0b363ad DL |
839 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
840 | { | |
23ca459a | 841 | int64_t offset; |
138f9fff | 842 | BlockDriverState *bs = s->mirror_top_bs->backing->bs; |
c0b363ad | 843 | BlockDriverState *target_bs = blk_bs(s->target); |
23ca459a | 844 | int ret; |
51b0a488 | 845 | int64_t count; |
c0b363ad | 846 | |
cdf3bc93 | 847 | if (s->zero_target) { |
c7c2769c | 848 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
e0d7f73e | 849 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); |
c7c2769c DL |
850 | return 0; |
851 | } | |
852 | ||
90ab48eb | 853 | s->initial_zeroing_ongoing = true; |
23ca459a EB |
854 | for (offset = 0; offset < s->bdev_length; ) { |
855 | int bytes = MIN(s->bdev_length - offset, | |
856 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c7c2769c DL |
857 | |
858 | mirror_throttle(s); | |
859 | ||
daa7f2f9 | 860 | if (job_is_cancelled(&s->common.job)) { |
90ab48eb | 861 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
862 | return 0; |
863 | } | |
864 | ||
865 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
67adf4b3 EB |
866 | trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, |
867 | s->in_flight); | |
9178f4fe | 868 | mirror_wait_for_free_in_flight_slot(s); |
c7c2769c DL |
869 | continue; |
870 | } | |
871 | ||
4295c5fc | 872 | mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO); |
23ca459a | 873 | offset += bytes; |
c7c2769c DL |
874 | } |
875 | ||
bae8196d | 876 | mirror_wait_for_all_io(s); |
90ab48eb | 877 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
878 | } |
879 | ||
c0b363ad | 880 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
23ca459a | 881 | for (offset = 0; offset < s->bdev_length; ) { |
c0b363ad | 882 | /* Just to make sure we are not exceeding int limit. */ |
23ca459a EB |
883 | int bytes = MIN(s->bdev_length - offset, |
884 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c0b363ad DL |
885 | |
886 | mirror_throttle(s); | |
887 | ||
daa7f2f9 | 888 | if (job_is_cancelled(&s->common.job)) { |
c0b363ad DL |
889 | return 0; |
890 | } | |
891 | ||
7ff9579e | 892 | WITH_GRAPH_RDLOCK_GUARD() { |
cc323997 PB |
893 | ret = bdrv_co_is_allocated_above(bs, s->base_overlay, true, offset, |
894 | bytes, &count); | |
7ff9579e | 895 | } |
c0b363ad DL |
896 | if (ret < 0) { |
897 | return ret; | |
898 | } | |
899 | ||
23ca459a | 900 | assert(count); |
a92b1b06 | 901 | if (ret > 0) { |
23ca459a | 902 | bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); |
c0b363ad | 903 | } |
23ca459a | 904 | offset += count; |
c0b363ad DL |
905 | } |
906 | return 0; | |
907 | } | |
908 | ||
bdffb31d PB |
909 | /* Called when going out of the streaming phase to flush the bulk of the |
910 | * data to the medium, or just before completing. | |
911 | */ | |
26bef102 | 912 | static int coroutine_fn mirror_flush(MirrorBlockJob *s) |
bdffb31d | 913 | { |
26bef102 | 914 | int ret = blk_co_flush(s->target); |
bdffb31d PB |
915 | if (ret < 0) { |
916 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
917 | s->ret = ret; | |
918 | } | |
919 | } | |
920 | return ret; | |
921 | } | |
922 | ||
f67432a2 | 923 | static int coroutine_fn mirror_run(Job *job, Error **errp) |
893f7eba | 924 | { |
f67432a2 | 925 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
138f9fff | 926 | BlockDriverState *bs = s->mirror_top_bs->backing->bs; |
32125b14 | 927 | MirrorBDSOpaque *mirror_top_opaque = s->mirror_top_bs->opaque; |
e253f4b8 | 928 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 929 | bool need_drain = true; |
d59cb66d | 930 | BlockDeviceIoStatus iostatus; |
c0b363ad | 931 | int64_t length; |
e83dd680 | 932 | int64_t target_length; |
b812f671 | 933 | BlockDriverInfo bdi; |
1d33936e JC |
934 | char backing_filename[2]; /* we only need 2 characters because we are only |
935 | checking for a NULL string */ | |
893f7eba | 936 | int ret = 0; |
893f7eba | 937 | |
daa7f2f9 | 938 | if (job_is_cancelled(&s->common.job)) { |
893f7eba PB |
939 | goto immediate_exit; |
940 | } | |
941 | ||
8ab8140a | 942 | bdrv_graph_co_rdlock(); |
c86422c5 | 943 | s->bdev_length = bdrv_co_getlength(bs); |
8ab8140a KW |
944 | bdrv_graph_co_rdunlock(); |
945 | ||
b21c7652 HR |
946 | if (s->bdev_length < 0) { |
947 | ret = s->bdev_length; | |
373df5b1 | 948 | goto immediate_exit; |
becc347e KW |
949 | } |
950 | ||
c86422c5 | 951 | target_length = blk_co_getlength(s->target); |
e83dd680 KW |
952 | if (target_length < 0) { |
953 | ret = target_length; | |
954 | goto immediate_exit; | |
955 | } | |
956 | ||
becc347e KW |
957 | /* Active commit must resize the base image if its size differs from the |
958 | * active layer. */ | |
959 | if (s->base == blk_bs(s->target)) { | |
e83dd680 | 960 | if (s->bdev_length > target_length) { |
88276216 AF |
961 | ret = blk_co_truncate(s->target, s->bdev_length, false, |
962 | PREALLOC_MODE_OFF, 0, NULL); | |
becc347e KW |
963 | if (ret < 0) { |
964 | goto immediate_exit; | |
965 | } | |
966 | } | |
e83dd680 KW |
967 | } else if (s->bdev_length != target_length) { |
968 | error_setg(errp, "Source and target image have different sizes"); | |
969 | ret = -EINVAL; | |
970 | goto immediate_exit; | |
becc347e KW |
971 | } |
972 | ||
973 | if (s->bdev_length == 0) { | |
2e1795b5 KW |
974 | /* Transition to the READY state and wait for complete. */ |
975 | job_transition_to_ready(&s->common.job); | |
76cb2f24 | 976 | qatomic_set(&s->actively_synced, true); |
08b83bff | 977 | while (!job_cancel_requested(&s->common.job) && !s->should_complete) { |
198c49cc | 978 | job_yield(&s->common.job); |
9e48b025 | 979 | } |
9e48b025 | 980 | goto immediate_exit; |
893f7eba PB |
981 | } |
982 | ||
b21c7652 | 983 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
984 | s->in_flight_bitmap = bitmap_new(length); |
985 | ||
b812f671 PB |
986 | /* If we have no backing file yet in the destination, we cannot let |
987 | * the destination do COW. Instead, we copy sectors around the | |
988 | * dirty data if needed. We need a bitmap to do that. | |
989 | */ | |
e253f4b8 | 990 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 991 | sizeof(backing_filename)); |
a00e70c0 | 992 | bdrv_graph_co_rdlock(); |
3d47eb0a | 993 | if (!bdrv_co_get_info(target_bs, &bdi) && bdi.cluster_size) { |
b436982f EB |
994 | s->target_cluster_size = bdi.cluster_size; |
995 | } else { | |
996 | s->target_cluster_size = BDRV_SECTOR_SIZE; | |
e5b43573 | 997 | } |
3f072a7f | 998 | if (backing_filename[0] && !bdrv_backing_chain_next(target_bs) && |
b436982f EB |
999 | s->granularity < s->target_cluster_size) { |
1000 | s->buf_size = MAX(s->buf_size, s->target_cluster_size); | |
e5b43573 | 1001 | s->cow_bitmap = bitmap_new(length); |
b812f671 | 1002 | } |
e253f4b8 | 1003 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
ad74751f | 1004 | bdrv_graph_co_rdunlock(); |
b812f671 | 1005 | |
7504edf4 KW |
1006 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
1007 | if (s->buf == NULL) { | |
1008 | ret = -ENOMEM; | |
1009 | goto immediate_exit; | |
1010 | } | |
1011 | ||
402a4741 | 1012 | mirror_free_init(s); |
893f7eba | 1013 | |
49efb1f5 | 1014 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 1015 | if (!s->is_none_mode) { |
c0b363ad | 1016 | ret = mirror_dirty_init(s); |
daa7f2f9 | 1017 | if (ret < 0 || job_is_cancelled(&s->common.job)) { |
c0b363ad | 1018 | goto immediate_exit; |
893f7eba PB |
1019 | } |
1020 | } | |
1021 | ||
32125b14 KW |
1022 | /* |
1023 | * Only now the job is fully initialised and mirror_top_bs should start | |
1024 | * accessing it. | |
1025 | */ | |
1026 | mirror_top_opaque->job = s; | |
1027 | ||
dc162c8e | 1028 | assert(!s->dbi); |
715a74d8 | 1029 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); |
893f7eba | 1030 | for (;;) { |
49efb1f5 | 1031 | int64_t cnt, delta; |
893f7eba PB |
1032 | bool should_complete; |
1033 | ||
bd48bde8 PB |
1034 | if (s->ret < 0) { |
1035 | ret = s->ret; | |
1036 | goto immediate_exit; | |
1037 | } | |
1038 | ||
da01ff7f | 1039 | job_pause_point(&s->common.job); |
565ac01f | 1040 | |
4feeec7e HR |
1041 | if (job_is_cancelled(&s->common.job)) { |
1042 | ret = 0; | |
1043 | goto immediate_exit; | |
1044 | } | |
1045 | ||
20dca810 | 1046 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
05df8a6a KW |
1047 | /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is |
1048 | * the number of bytes currently being processed; together those are | |
1049 | * the current remaining operation length */ | |
d69a879b HR |
1050 | job_progress_set_remaining(&s->common.job, |
1051 | s->bytes_in_flight + cnt + | |
1052 | s->active_write_bytes_in_flight); | |
bd48bde8 PB |
1053 | |
1054 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 1055 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
18bb6928 KW |
1056 | * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is |
1057 | * an error, or when the source is clean, whichever comes first. */ | |
49efb1f5 | 1058 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
d59cb66d EGE |
1059 | WITH_JOB_LOCK_GUARD() { |
1060 | iostatus = s->common.iostatus; | |
1061 | } | |
18bb6928 | 1062 | if (delta < BLOCK_JOB_SLICE_TIME && |
d59cb66d | 1063 | iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 1064 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 | 1065 | (cnt == 0 && s->in_flight > 0)) { |
9a46dba7 | 1066 | trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); |
9178f4fe | 1067 | mirror_wait_for_free_in_flight_slot(s); |
bd48bde8 PB |
1068 | continue; |
1069 | } else if (cnt != 0) { | |
018e5987 | 1070 | mirror_iteration(s); |
893f7eba | 1071 | } |
893f7eba PB |
1072 | } |
1073 | ||
1074 | should_complete = false; | |
bd48bde8 | 1075 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 1076 | trace_mirror_before_flush(s); |
44716224 | 1077 | if (!job_is_ready(&s->common.job)) { |
bdffb31d PB |
1078 | if (mirror_flush(s) < 0) { |
1079 | /* Go check s->ret. */ | |
1080 | continue; | |
b952b558 | 1081 | } |
b952b558 PB |
1082 | /* We're out of the streaming phase. From now on, if the job |
1083 | * is cancelled we will actually complete all pending I/O and | |
1084 | * report completion. This way, block-job-cancel will leave | |
1085 | * the target in a consistent state. | |
1086 | */ | |
2e1795b5 | 1087 | job_transition_to_ready(&s->common.job); |
c45d0e1a | 1088 | } |
2d400d15 | 1089 | if (qatomic_read(&s->copy_mode) != MIRROR_COPY_MODE_BACKGROUND) { |
76cb2f24 | 1090 | qatomic_set(&s->actively_synced, true); |
d63ffd87 | 1091 | } |
bdffb31d PB |
1092 | |
1093 | should_complete = s->should_complete || | |
08b83bff | 1094 | job_cancel_requested(&s->common.job); |
bdffb31d | 1095 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
893f7eba PB |
1096 | } |
1097 | ||
1098 | if (cnt == 0 && should_complete) { | |
1099 | /* The dirty bitmap is not updated while operations are pending. | |
1100 | * If we're about to exit, wait for pending operations before | |
1101 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
1102 | * source has dirty data to copy! | |
1103 | * | |
1104 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
1105 | * mirror_populate runs, so pause it now. Before deciding |
1106 | * whether to switch to target check one last time if I/O has | |
1107 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba | 1108 | */ |
9a46dba7 | 1109 | trace_mirror_before_drain(s, cnt); |
9a0cec66 | 1110 | |
5e771752 | 1111 | s->in_drain = true; |
9a0cec66 | 1112 | bdrv_drained_begin(bs); |
d69a879b HR |
1113 | |
1114 | /* Must be zero because we are drained */ | |
1115 | assert(s->in_active_write_counter == 0); | |
1116 | ||
20dca810 | 1117 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 1118 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 | 1119 | bdrv_drained_end(bs); |
5e771752 | 1120 | s->in_drain = false; |
9a0cec66 PB |
1121 | continue; |
1122 | } | |
1123 | ||
1124 | /* The two disks are in sync. Exit and report successful | |
1125 | * completion. | |
1126 | */ | |
1127 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
9a0cec66 PB |
1128 | need_drain = false; |
1129 | break; | |
893f7eba PB |
1130 | } |
1131 | ||
44716224 | 1132 | if (job_is_ready(&s->common.job) && !should_complete) { |
018e5987 KW |
1133 | if (s->in_flight == 0 && cnt == 0) { |
1134 | trace_mirror_before_sleep(s, cnt, job_is_ready(&s->common.job), | |
1135 | BLOCK_JOB_SLICE_TIME); | |
1136 | job_sleep_ns(&s->common.job, BLOCK_JOB_SLICE_TIME); | |
1137 | } | |
1138 | } else { | |
1139 | block_job_ratelimit_sleep(&s->common); | |
ddc4115e | 1140 | } |
49efb1f5 | 1141 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
1142 | } |
1143 | ||
1144 | immediate_exit: | |
bd48bde8 PB |
1145 | if (s->in_flight > 0) { |
1146 | /* We get here only if something went wrong. Either the job failed, | |
1147 | * or it was cancelled prematurely so that we do not guarantee that | |
1148 | * the target is a copy of the source. | |
1149 | */ | |
08b83bff | 1150 | assert(ret < 0 || job_is_cancelled(&s->common.job)); |
9a0cec66 | 1151 | assert(need_drain); |
bae8196d | 1152 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
1153 | } |
1154 | ||
1155 | assert(s->in_flight == 0); | |
7191bf31 | 1156 | qemu_vfree(s->buf); |
b812f671 | 1157 | g_free(s->cow_bitmap); |
402a4741 | 1158 | g_free(s->in_flight_bitmap); |
dc162c8e | 1159 | bdrv_dirty_iter_free(s->dbi); |
5a7e7a0b | 1160 | |
9a0cec66 | 1161 | if (need_drain) { |
5e771752 | 1162 | s->in_drain = true; |
9a0cec66 PB |
1163 | bdrv_drained_begin(bs); |
1164 | } | |
f67432a2 | 1165 | |
f67432a2 | 1166 | return ret; |
893f7eba PB |
1167 | } |
1168 | ||
3453d972 | 1169 | static void mirror_complete(Job *job, Error **errp) |
d63ffd87 | 1170 | { |
3453d972 | 1171 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
d63ffd87 | 1172 | |
44716224 | 1173 | if (!job_is_ready(job)) { |
9df229c3 | 1174 | error_setg(errp, "The active block job '%s' cannot be completed", |
3453d972 | 1175 | job->id); |
d63ffd87 PB |
1176 | return; |
1177 | } | |
1178 | ||
15d67298 | 1179 | /* block all operations on to_replace bs */ |
09158f00 | 1180 | if (s->replaces) { |
5a7e7a0b SH |
1181 | AioContext *replace_aio_context; |
1182 | ||
e12f3784 | 1183 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 1184 | if (!s->to_replace) { |
e12f3784 | 1185 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
1186 | return; |
1187 | } | |
1188 | ||
5a7e7a0b SH |
1189 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
1190 | aio_context_acquire(replace_aio_context); | |
1191 | ||
64631f36 | 1192 | /* TODO Translate this into child freeze system. */ |
09158f00 BC |
1193 | error_setg(&s->replace_blocker, |
1194 | "block device is in use by block-job-complete"); | |
1195 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
1196 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
1197 | |
1198 | aio_context_release(replace_aio_context); | |
09158f00 BC |
1199 | } |
1200 | ||
d63ffd87 | 1201 | s->should_complete = true; |
00769414 HR |
1202 | |
1203 | /* If the job is paused, it will be re-entered when it is resumed */ | |
279ac06e EGE |
1204 | WITH_JOB_LOCK_GUARD() { |
1205 | if (!job->paused) { | |
1206 | job_enter_cond_locked(job, NULL); | |
1207 | } | |
00769414 | 1208 | } |
d63ffd87 PB |
1209 | } |
1210 | ||
537c3d4f | 1211 | static void coroutine_fn mirror_pause(Job *job) |
565ac01f | 1212 | { |
da01ff7f | 1213 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
565ac01f | 1214 | |
bae8196d | 1215 | mirror_wait_for_all_io(s); |
565ac01f SH |
1216 | } |
1217 | ||
89bd0305 KW |
1218 | static bool mirror_drained_poll(BlockJob *job) |
1219 | { | |
1220 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
5e771752 SL |
1221 | |
1222 | /* If the job isn't paused nor cancelled, we can't be sure that it won't | |
1223 | * issue more requests. We make an exception if we've reached this point | |
1224 | * from one of our own drain sections, to avoid a deadlock waiting for | |
1225 | * ourselves. | |
1226 | */ | |
279ac06e EGE |
1227 | WITH_JOB_LOCK_GUARD() { |
1228 | if (!s->common.job.paused && !job_is_cancelled_locked(&job->job) | |
1229 | && !s->in_drain) { | |
1230 | return true; | |
1231 | } | |
5e771752 SL |
1232 | } |
1233 | ||
89bd0305 KW |
1234 | return !!s->in_flight; |
1235 | } | |
1236 | ||
73895f38 | 1237 | static bool mirror_cancel(Job *job, bool force) |
521ff8b7 VSO |
1238 | { |
1239 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); | |
1240 | BlockDriverState *target = blk_bs(s->target); | |
1241 | ||
73895f38 HR |
1242 | /* |
1243 | * Before the job is READY, we treat any cancellation like a | |
1244 | * force-cancellation. | |
1245 | */ | |
1246 | force = force || !job_is_ready(job); | |
1247 | ||
1248 | if (force) { | |
9c785cd7 VSO |
1249 | bdrv_cancel_in_flight(target); |
1250 | } | |
73895f38 HR |
1251 | return force; |
1252 | } | |
1253 | ||
1254 | static bool commit_active_cancel(Job *job, bool force) | |
1255 | { | |
1256 | /* Same as above in mirror_cancel() */ | |
1257 | return force || !job_is_ready(job); | |
521ff8b7 VSO |
1258 | } |
1259 | ||
2d400d15 FE |
1260 | static void mirror_change(BlockJob *job, BlockJobChangeOptions *opts, |
1261 | Error **errp) | |
1262 | { | |
1263 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
1264 | BlockJobChangeOptionsMirror *change_opts = &opts->u.mirror; | |
1265 | MirrorCopyMode current; | |
1266 | ||
1267 | /* | |
1268 | * The implementation relies on the fact that copy_mode is only written | |
1269 | * under the BQL. Otherwise, further synchronization would be required. | |
1270 | */ | |
1271 | ||
1272 | GLOBAL_STATE_CODE(); | |
1273 | ||
1274 | if (qatomic_read(&s->copy_mode) == change_opts->copy_mode) { | |
1275 | return; | |
1276 | } | |
1277 | ||
1278 | if (change_opts->copy_mode != MIRROR_COPY_MODE_WRITE_BLOCKING) { | |
1279 | error_setg(errp, "Change to copy mode '%s' is not implemented", | |
1280 | MirrorCopyMode_str(change_opts->copy_mode)); | |
1281 | return; | |
1282 | } | |
1283 | ||
1284 | current = qatomic_cmpxchg(&s->copy_mode, MIRROR_COPY_MODE_BACKGROUND, | |
1285 | change_opts->copy_mode); | |
1286 | if (current != MIRROR_COPY_MODE_BACKGROUND) { | |
1287 | error_setg(errp, "Expected current copy mode '%s', got '%s'", | |
1288 | MirrorCopyMode_str(MIRROR_COPY_MODE_BACKGROUND), | |
1289 | MirrorCopyMode_str(current)); | |
1290 | } | |
1291 | } | |
1292 | ||
76cb2f24 FE |
1293 | static void mirror_query(BlockJob *job, BlockJobInfo *info) |
1294 | { | |
1295 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
1296 | ||
1297 | info->u.mirror = (BlockJobInfoMirror) { | |
1298 | .actively_synced = qatomic_read(&s->actively_synced), | |
1299 | }; | |
1300 | } | |
1301 | ||
3fc4b10a | 1302 | static const BlockJobDriver mirror_job_driver = { |
33e9e9bd KW |
1303 | .job_driver = { |
1304 | .instance_size = sizeof(MirrorBlockJob), | |
252291ea | 1305 | .job_type = JOB_TYPE_MIRROR, |
80fa2c75 | 1306 | .free = block_job_free, |
b15de828 | 1307 | .user_resume = block_job_user_resume, |
f67432a2 | 1308 | .run = mirror_run, |
737efc1e JS |
1309 | .prepare = mirror_prepare, |
1310 | .abort = mirror_abort, | |
da01ff7f | 1311 | .pause = mirror_pause, |
3453d972 | 1312 | .complete = mirror_complete, |
521ff8b7 | 1313 | .cancel = mirror_cancel, |
33e9e9bd | 1314 | }, |
89bd0305 | 1315 | .drained_poll = mirror_drained_poll, |
2d400d15 | 1316 | .change = mirror_change, |
76cb2f24 | 1317 | .query = mirror_query, |
893f7eba PB |
1318 | }; |
1319 | ||
03544a6e | 1320 | static const BlockJobDriver commit_active_job_driver = { |
33e9e9bd KW |
1321 | .job_driver = { |
1322 | .instance_size = sizeof(MirrorBlockJob), | |
252291ea | 1323 | .job_type = JOB_TYPE_COMMIT, |
80fa2c75 | 1324 | .free = block_job_free, |
b15de828 | 1325 | .user_resume = block_job_user_resume, |
f67432a2 | 1326 | .run = mirror_run, |
737efc1e JS |
1327 | .prepare = mirror_prepare, |
1328 | .abort = mirror_abort, | |
da01ff7f | 1329 | .pause = mirror_pause, |
3453d972 | 1330 | .complete = mirror_complete, |
73895f38 | 1331 | .cancel = commit_active_cancel, |
33e9e9bd | 1332 | }, |
89bd0305 | 1333 | .drained_poll = mirror_drained_poll, |
03544a6e FZ |
1334 | }; |
1335 | ||
537c3d4f SH |
1336 | static void coroutine_fn |
1337 | do_sync_target_write(MirrorBlockJob *job, MirrorMethod method, | |
1338 | uint64_t offset, uint64_t bytes, | |
1339 | QEMUIOVector *qiov, int flags) | |
d06107ad | 1340 | { |
5c511ac3 | 1341 | int ret; |
dbdf699c VSO |
1342 | size_t qiov_offset = 0; |
1343 | int64_t bitmap_offset, bitmap_end; | |
d06107ad | 1344 | |
dbdf699c VSO |
1345 | if (!QEMU_IS_ALIGNED(offset, job->granularity) && |
1346 | bdrv_dirty_bitmap_get(job->dirty_bitmap, offset)) | |
1347 | { | |
1348 | /* | |
1349 | * Dirty unaligned padding: ignore it. | |
1350 | * | |
1351 | * Reasoning: | |
1352 | * 1. If we copy it, we can't reset corresponding bit in | |
1353 | * dirty_bitmap as there may be some "dirty" bytes still not | |
1354 | * copied. | |
1355 | * 2. It's already dirty, so skipping it we don't diverge mirror | |
1356 | * progress. | |
1357 | * | |
1358 | * Note, that because of this, guest write may have no contribution | |
1359 | * into mirror converge, but that's not bad, as we have background | |
1360 | * process of mirroring. If under some bad circumstances (high guest | |
1361 | * IO load) background process starve, we will not converge anyway, | |
1362 | * even if each write will contribute, as guest is not guaranteed to | |
1363 | * rewrite the whole disk. | |
1364 | */ | |
1365 | qiov_offset = QEMU_ALIGN_UP(offset, job->granularity) - offset; | |
1366 | if (bytes <= qiov_offset) { | |
1367 | /* nothing to do after shrink */ | |
1368 | return; | |
1369 | } | |
1370 | offset += qiov_offset; | |
1371 | bytes -= qiov_offset; | |
1372 | } | |
1373 | ||
1374 | if (!QEMU_IS_ALIGNED(offset + bytes, job->granularity) && | |
1375 | bdrv_dirty_bitmap_get(job->dirty_bitmap, offset + bytes - 1)) | |
1376 | { | |
1377 | uint64_t tail = (offset + bytes) % job->granularity; | |
1378 | ||
1379 | if (bytes <= tail) { | |
1380 | /* nothing to do after shrink */ | |
1381 | return; | |
1382 | } | |
1383 | bytes -= tail; | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * Tails are either clean or shrunk, so for bitmap resetting | |
1388 | * we safely align the range down. | |
1389 | */ | |
1390 | bitmap_offset = QEMU_ALIGN_UP(offset, job->granularity); | |
1391 | bitmap_end = QEMU_ALIGN_DOWN(offset + bytes, job->granularity); | |
1392 | if (bitmap_offset < bitmap_end) { | |
1393 | bdrv_reset_dirty_bitmap(job->dirty_bitmap, bitmap_offset, | |
1394 | bitmap_end - bitmap_offset); | |
1395 | } | |
d06107ad | 1396 | |
5c511ac3 | 1397 | job_progress_increase_remaining(&job->common.job, bytes); |
d69a879b | 1398 | job->active_write_bytes_in_flight += bytes; |
d06107ad | 1399 | |
5c511ac3 VSO |
1400 | switch (method) { |
1401 | case MIRROR_METHOD_COPY: | |
dbdf699c VSO |
1402 | ret = blk_co_pwritev_part(job->target, offset, bytes, |
1403 | qiov, qiov_offset, flags); | |
5c511ac3 | 1404 | break; |
d06107ad | 1405 | |
5c511ac3 VSO |
1406 | case MIRROR_METHOD_ZERO: |
1407 | assert(!qiov); | |
1408 | ret = blk_co_pwrite_zeroes(job->target, offset, bytes, flags); | |
1409 | break; | |
d06107ad | 1410 | |
5c511ac3 VSO |
1411 | case MIRROR_METHOD_DISCARD: |
1412 | assert(!qiov); | |
1413 | ret = blk_co_pdiscard(job->target, offset, bytes); | |
1414 | break; | |
d06107ad | 1415 | |
5c511ac3 VSO |
1416 | default: |
1417 | abort(); | |
1418 | } | |
d06107ad | 1419 | |
d69a879b | 1420 | job->active_write_bytes_in_flight -= bytes; |
5c511ac3 VSO |
1421 | if (ret >= 0) { |
1422 | job_progress_update(&job->common.job, bytes); | |
1423 | } else { | |
1424 | BlockErrorAction action; | |
d06107ad | 1425 | |
dbdf699c VSO |
1426 | /* |
1427 | * We failed, so we should mark dirty the whole area, aligned up. | |
1428 | * Note that we don't care about shrunk tails if any: they were dirty | |
1429 | * at function start, and they must be still dirty, as we've locked | |
1430 | * the region for in-flight op. | |
1431 | */ | |
1432 | bitmap_offset = QEMU_ALIGN_DOWN(offset, job->granularity); | |
1433 | bitmap_end = QEMU_ALIGN_UP(offset + bytes, job->granularity); | |
1434 | bdrv_set_dirty_bitmap(job->dirty_bitmap, bitmap_offset, | |
1435 | bitmap_end - bitmap_offset); | |
76cb2f24 | 1436 | qatomic_set(&job->actively_synced, false); |
d06107ad | 1437 | |
5c511ac3 VSO |
1438 | action = mirror_error_action(job, false, -ret); |
1439 | if (action == BLOCK_ERROR_ACTION_REPORT) { | |
1440 | if (!job->ret) { | |
1441 | job->ret = ret; | |
d06107ad HR |
1442 | } |
1443 | } | |
d06107ad HR |
1444 | } |
1445 | } | |
1446 | ||
1447 | static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s, | |
1448 | uint64_t offset, | |
1449 | uint64_t bytes) | |
1450 | { | |
1451 | MirrorOp *op; | |
1452 | uint64_t start_chunk = offset / s->granularity; | |
1453 | uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); | |
1454 | ||
1455 | op = g_new(MirrorOp, 1); | |
1456 | *op = (MirrorOp){ | |
1457 | .s = s, | |
1458 | .offset = offset, | |
1459 | .bytes = bytes, | |
1460 | .is_active_write = true, | |
ce8cabbd | 1461 | .is_in_flight = true, |
ead3f1bf | 1462 | .co = qemu_coroutine_self(), |
d06107ad HR |
1463 | }; |
1464 | qemu_co_queue_init(&op->waiting_requests); | |
1465 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); | |
1466 | ||
1467 | s->in_active_write_counter++; | |
1468 | ||
d69a879b HR |
1469 | /* |
1470 | * Wait for concurrent requests affecting the area. If there are already | |
1471 | * running requests that are copying off now-to-be stale data in the area, | |
1472 | * we must wait for them to finish before we begin writing fresh data to the | |
1473 | * target so that the write operations appear in the correct order. | |
1474 | * Note that background requests (see mirror_iteration()) in contrast only | |
1475 | * wait for conflicting requests at the start of the dirty area, and then | |
1476 | * (based on the in_flight_bitmap) truncate the area to copy so it will not | |
1477 | * conflict with any requests beyond that. For active writes, however, we | |
1478 | * cannot truncate that area. The request from our parent must be blocked | |
1479 | * until the area is copied in full. Therefore, we must wait for the whole | |
1480 | * area to become free of concurrent requests. | |
1481 | */ | |
d06107ad HR |
1482 | mirror_wait_on_conflicts(op, s, offset, bytes); |
1483 | ||
1484 | bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); | |
1485 | ||
1486 | return op; | |
1487 | } | |
1488 | ||
9c93652d | 1489 | static void coroutine_fn GRAPH_RDLOCK active_write_settle(MirrorOp *op) |
d06107ad HR |
1490 | { |
1491 | uint64_t start_chunk = op->offset / op->s->granularity; | |
1492 | uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes, | |
1493 | op->s->granularity); | |
1494 | ||
76cb2f24 FE |
1495 | if (!--op->s->in_active_write_counter && |
1496 | qatomic_read(&op->s->actively_synced)) { | |
d06107ad HR |
1497 | BdrvChild *source = op->s->mirror_top_bs->backing; |
1498 | ||
1499 | if (QLIST_FIRST(&source->bs->parents) == source && | |
1500 | QLIST_NEXT(source, next_parent) == NULL) | |
1501 | { | |
1502 | /* Assert that we are back in sync once all active write | |
1503 | * operations are settled. | |
1504 | * Note that we can only assert this if the mirror node | |
1505 | * is the source node's only parent. */ | |
1506 | assert(!bdrv_get_dirty_count(op->s->dirty_bitmap)); | |
1507 | } | |
1508 | } | |
1509 | bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); | |
1510 | QTAILQ_REMOVE(&op->s->ops_in_flight, op, next); | |
1511 | qemu_co_queue_restart_all(&op->waiting_requests); | |
1512 | g_free(op); | |
1513 | } | |
1514 | ||
b9b10c35 KW |
1515 | static int coroutine_fn GRAPH_RDLOCK |
1516 | bdrv_mirror_top_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1517 | QEMUIOVector *qiov, BdrvRequestFlags flags) | |
4ef85a9c KW |
1518 | { |
1519 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1520 | } | |
1521 | ||
7b32ad22 FE |
1522 | static bool should_copy_to_target(MirrorBDSOpaque *s) |
1523 | { | |
1524 | return s->job && s->job->ret >= 0 && | |
1525 | !job_is_cancelled(&s->job->common.job) && | |
2d400d15 | 1526 | qatomic_read(&s->job->copy_mode) == MIRROR_COPY_MODE_WRITE_BLOCKING; |
7b32ad22 FE |
1527 | } |
1528 | ||
9a5a1c62 EGE |
1529 | static int coroutine_fn GRAPH_RDLOCK |
1530 | bdrv_mirror_top_do_write(BlockDriverState *bs, MirrorMethod method, | |
7b32ad22 FE |
1531 | bool copy_to_target, uint64_t offset, uint64_t bytes, |
1532 | QEMUIOVector *qiov, int flags) | |
d06107ad HR |
1533 | { |
1534 | MirrorOp *op = NULL; | |
1535 | MirrorBDSOpaque *s = bs->opaque; | |
1536 | int ret = 0; | |
d06107ad HR |
1537 | |
1538 | if (copy_to_target) { | |
1539 | op = active_write_prepare(s->job, offset, bytes); | |
1540 | } | |
1541 | ||
1542 | switch (method) { | |
1543 | case MIRROR_METHOD_COPY: | |
1544 | ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1545 | break; | |
1546 | ||
1547 | case MIRROR_METHOD_ZERO: | |
1548 | ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); | |
1549 | break; | |
1550 | ||
1551 | case MIRROR_METHOD_DISCARD: | |
0b9fd3f4 | 1552 | ret = bdrv_co_pdiscard(bs->backing, offset, bytes); |
d06107ad HR |
1553 | break; |
1554 | ||
1555 | default: | |
1556 | abort(); | |
1557 | } | |
1558 | ||
058cfca5 | 1559 | if (!copy_to_target && s->job && s->job->dirty_bitmap) { |
76cb2f24 | 1560 | qatomic_set(&s->job->actively_synced, false); |
058cfca5 FE |
1561 | bdrv_set_dirty_bitmap(s->job->dirty_bitmap, offset, bytes); |
1562 | } | |
1563 | ||
d06107ad HR |
1564 | if (ret < 0) { |
1565 | goto out; | |
1566 | } | |
1567 | ||
1568 | if (copy_to_target) { | |
1569 | do_sync_target_write(s->job, method, offset, bytes, qiov, flags); | |
1570 | } | |
1571 | ||
1572 | out: | |
1573 | if (copy_to_target) { | |
1574 | active_write_settle(op); | |
1575 | } | |
1576 | return ret; | |
1577 | } | |
1578 | ||
b9b10c35 KW |
1579 | static int coroutine_fn GRAPH_RDLOCK |
1580 | bdrv_mirror_top_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1581 | QEMUIOVector *qiov, BdrvRequestFlags flags) | |
4ef85a9c | 1582 | { |
d06107ad HR |
1583 | QEMUIOVector bounce_qiov; |
1584 | void *bounce_buf; | |
1585 | int ret = 0; | |
7b32ad22 | 1586 | bool copy_to_target = should_copy_to_target(bs->opaque); |
d06107ad HR |
1587 | |
1588 | if (copy_to_target) { | |
1589 | /* The guest might concurrently modify the data to write; but | |
1590 | * the data on source and destination must match, so we have | |
1591 | * to use a bounce buffer if we are going to write to the | |
1592 | * target now. */ | |
1593 | bounce_buf = qemu_blockalign(bs, bytes); | |
1594 | iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes); | |
1595 | ||
1596 | qemu_iovec_init(&bounce_qiov, 1); | |
1597 | qemu_iovec_add(&bounce_qiov, bounce_buf, bytes); | |
1598 | qiov = &bounce_qiov; | |
e8b65355 SH |
1599 | |
1600 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
d06107ad HR |
1601 | } |
1602 | ||
7b32ad22 FE |
1603 | ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, copy_to_target, |
1604 | offset, bytes, qiov, flags); | |
d06107ad HR |
1605 | |
1606 | if (copy_to_target) { | |
1607 | qemu_iovec_destroy(&bounce_qiov); | |
1608 | qemu_vfree(bounce_buf); | |
1609 | } | |
1610 | ||
1611 | return ret; | |
4ef85a9c KW |
1612 | } |
1613 | ||
88095349 | 1614 | static int coroutine_fn GRAPH_RDLOCK bdrv_mirror_top_flush(BlockDriverState *bs) |
4ef85a9c | 1615 | { |
ce960aa9 VSO |
1616 | if (bs->backing == NULL) { |
1617 | /* we can be here after failed bdrv_append in mirror_start_job */ | |
1618 | return 0; | |
1619 | } | |
4ef85a9c KW |
1620 | return bdrv_co_flush(bs->backing->bs); |
1621 | } | |
1622 | ||
abaf8b75 KW |
1623 | static int coroutine_fn GRAPH_RDLOCK |
1624 | bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, int64_t offset, | |
1625 | int64_t bytes, BdrvRequestFlags flags) | |
4ef85a9c | 1626 | { |
7b32ad22 FE |
1627 | bool copy_to_target = should_copy_to_target(bs->opaque); |
1628 | return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, copy_to_target, | |
1629 | offset, bytes, NULL, flags); | |
4ef85a9c KW |
1630 | } |
1631 | ||
9a5a1c62 EGE |
1632 | static int coroutine_fn GRAPH_RDLOCK |
1633 | bdrv_mirror_top_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
4ef85a9c | 1634 | { |
7b32ad22 FE |
1635 | bool copy_to_target = should_copy_to_target(bs->opaque); |
1636 | return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, copy_to_target, | |
1637 | offset, bytes, NULL, 0); | |
4ef85a9c KW |
1638 | } |
1639 | ||
998b3a1e | 1640 | static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs) |
fd4a6493 | 1641 | { |
18775ff3 VSO |
1642 | if (bs->backing == NULL) { |
1643 | /* we can be here after failed bdrv_attach_child in | |
1644 | * bdrv_set_backing_hd */ | |
1645 | return; | |
1646 | } | |
fd4a6493 KW |
1647 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), |
1648 | bs->backing->bs->filename); | |
1649 | } | |
1650 | ||
4ef85a9c | 1651 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, |
bf8e925e | 1652 | BdrvChildRole role, |
e0995dc3 | 1653 | BlockReopenQueue *reopen_queue, |
4ef85a9c KW |
1654 | uint64_t perm, uint64_t shared, |
1655 | uint64_t *nperm, uint64_t *nshared) | |
1656 | { | |
f94dc3b4 HR |
1657 | MirrorBDSOpaque *s = bs->opaque; |
1658 | ||
1659 | if (s->stop) { | |
1660 | /* | |
1661 | * If the job is to be stopped, we do not need to forward | |
1662 | * anything to the real image. | |
1663 | */ | |
1664 | *nperm = 0; | |
1665 | *nshared = BLK_PERM_ALL; | |
1666 | return; | |
1667 | } | |
1668 | ||
53431b90 HR |
1669 | bdrv_default_perms(bs, c, role, reopen_queue, |
1670 | perm, shared, nperm, nshared); | |
4ef85a9c | 1671 | |
53431b90 HR |
1672 | if (s->is_commit) { |
1673 | /* | |
1674 | * For commit jobs, we cannot take CONSISTENT_READ, because | |
1675 | * that permission is unshared for everything above the base | |
1676 | * node (except for filters on the base node). | |
1677 | * We also have to force-share the WRITE permission, or | |
1678 | * otherwise we would block ourselves at the base node (if | |
1679 | * writes are blocked for a node, they are also blocked for | |
1680 | * its backing file). | |
1681 | * (We could also share RESIZE, because it may be needed for | |
1682 | * the target if its size is less than the top node's; but | |
1683 | * bdrv_default_perms_for_cow() automatically shares RESIZE | |
1684 | * for backing nodes if WRITE is shared, so there is no need | |
1685 | * to do it here.) | |
1686 | */ | |
1687 | *nperm &= ~BLK_PERM_CONSISTENT_READ; | |
1688 | *nshared |= BLK_PERM_WRITE; | |
1689 | } | |
4ef85a9c KW |
1690 | } |
1691 | ||
1692 | /* Dummy node that provides consistent read to its users without requiring it | |
1693 | * from its backing file and that allows writes on the backing file chain. */ | |
1694 | static BlockDriver bdrv_mirror_top = { | |
1695 | .format_name = "mirror_top", | |
1696 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1697 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1698 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1699 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1700 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
fd4a6493 | 1701 | .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, |
4ef85a9c | 1702 | .bdrv_child_perm = bdrv_mirror_top_child_perm, |
6540fd15 HR |
1703 | |
1704 | .is_filter = true, | |
046fd84f | 1705 | .filtered_child_is_backing = true, |
4ef85a9c KW |
1706 | }; |
1707 | ||
cc19f177 VSO |
1708 | static BlockJob *mirror_start_job( |
1709 | const char *job_id, BlockDriverState *bs, | |
47970dfb JS |
1710 | int creation_flags, BlockDriverState *target, |
1711 | const char *replaces, int64_t speed, | |
1712 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1713 | BlockMirrorBackingMode backing_mode, |
cdf3bc93 | 1714 | bool zero_target, |
09158f00 BC |
1715 | BlockdevOnError on_source_error, |
1716 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1717 | bool unmap, |
097310b5 | 1718 | BlockCompletionFunc *cb, |
51ccfa2d | 1719 | void *opaque, |
09158f00 | 1720 | const BlockJobDriver *driver, |
b49f7ead | 1721 | bool is_none_mode, BlockDriverState *base, |
51ccfa2d | 1722 | bool auto_complete, const char *filter_node_name, |
481debaa | 1723 | bool is_mirror, MirrorCopyMode copy_mode, |
51ccfa2d | 1724 | Error **errp) |
893f7eba PB |
1725 | { |
1726 | MirrorBlockJob *s; | |
429076e8 | 1727 | MirrorBDSOpaque *bs_opaque; |
4ef85a9c | 1728 | BlockDriverState *mirror_top_bs; |
4ef85a9c | 1729 | bool target_is_backing; |
3f072a7f | 1730 | uint64_t target_perms, target_shared_perms; |
d7086422 | 1731 | int ret; |
893f7eba | 1732 | |
3804e3cf KW |
1733 | GLOBAL_STATE_CODE(); |
1734 | ||
eee13dfe | 1735 | if (granularity == 0) { |
341ebc2f | 1736 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1737 | } |
1738 | ||
31826642 | 1739 | assert(is_power_of_2(granularity)); |
eee13dfe | 1740 | |
48ac0a4d WC |
1741 | if (buf_size < 0) { |
1742 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
cc19f177 | 1743 | return NULL; |
48ac0a4d WC |
1744 | } |
1745 | ||
1746 | if (buf_size == 0) { | |
1747 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
1748 | } | |
5bc361b8 | 1749 | |
ad74751f | 1750 | bdrv_graph_rdlock_main_loop(); |
3f072a7f | 1751 | if (bdrv_skip_filters(bs) == bdrv_skip_filters(target)) { |
86fae10c | 1752 | error_setg(errp, "Can't mirror node into itself"); |
ad74751f | 1753 | bdrv_graph_rdunlock_main_loop(); |
cc19f177 | 1754 | return NULL; |
86fae10c KW |
1755 | } |
1756 | ||
53431b90 | 1757 | target_is_backing = bdrv_chain_contains(bs, target); |
ad74751f | 1758 | bdrv_graph_rdunlock_main_loop(); |
53431b90 | 1759 | |
4ef85a9c KW |
1760 | /* In the case of active commit, add dummy driver to provide consistent |
1761 | * reads on the top, while disabling it in the intermediate nodes, and make | |
1762 | * the backing chain writable. */ | |
6cdbceb1 KW |
1763 | mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, |
1764 | BDRV_O_RDWR, errp); | |
4ef85a9c | 1765 | if (mirror_top_bs == NULL) { |
cc19f177 | 1766 | return NULL; |
4ef85a9c | 1767 | } |
d3c8c674 KW |
1768 | if (!filter_node_name) { |
1769 | mirror_top_bs->implicit = true; | |
1770 | } | |
e5182c1c HR |
1771 | |
1772 | /* So that we can always drop this node */ | |
1773 | mirror_top_bs->never_freeze = true; | |
1774 | ||
4ef85a9c | 1775 | mirror_top_bs->total_sectors = bs->total_sectors; |
228345bf | 1776 | mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED; |
80f5c33f KW |
1777 | mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED | |
1778 | BDRV_REQ_NO_FALLBACK; | |
429076e8 HR |
1779 | bs_opaque = g_new0(MirrorBDSOpaque, 1); |
1780 | mirror_top_bs->opaque = bs_opaque; | |
4ef85a9c | 1781 | |
53431b90 HR |
1782 | bs_opaque->is_commit = target_is_backing; |
1783 | ||
4ef85a9c | 1784 | bdrv_drained_begin(bs); |
934aee14 | 1785 | ret = bdrv_append(mirror_top_bs, bs, errp); |
4ef85a9c KW |
1786 | bdrv_drained_end(bs); |
1787 | ||
934aee14 | 1788 | if (ret < 0) { |
b2c2832c | 1789 | bdrv_unref(mirror_top_bs); |
cc19f177 | 1790 | return NULL; |
b2c2832c KW |
1791 | } |
1792 | ||
4ef85a9c | 1793 | /* Make sure that the source is not resized while the job is running */ |
75859b94 | 1794 | s = block_job_create(job_id, driver, NULL, mirror_top_bs, |
4ef85a9c KW |
1795 | BLK_PERM_CONSISTENT_READ, |
1796 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
64631f36 | 1797 | BLK_PERM_WRITE, speed, |
c6cc12bf | 1798 | creation_flags, cb, opaque, errp); |
893f7eba | 1799 | if (!s) { |
4ef85a9c | 1800 | goto fail; |
893f7eba | 1801 | } |
429076e8 | 1802 | |
7a25fcd0 HR |
1803 | /* The block job now has a reference to this node */ |
1804 | bdrv_unref(mirror_top_bs); | |
1805 | ||
4ef85a9c KW |
1806 | s->mirror_top_bs = mirror_top_bs; |
1807 | ||
1808 | /* No resize for the target either; while the mirror is still running, a | |
1809 | * consistent read isn't necessarily possible. We could possibly allow | |
1810 | * writes and graph modifications, though it would likely defeat the | |
1811 | * purpose of a mirror, so leave them blocked for now. | |
1812 | * | |
1813 | * In the case of active commit, things look a bit different, though, | |
1814 | * because the target is an already populated backing file in active use. | |
1815 | * We can allow anything except resize there.*/ | |
3f072a7f HR |
1816 | |
1817 | target_perms = BLK_PERM_WRITE; | |
1818 | target_shared_perms = BLK_PERM_WRITE_UNCHANGED; | |
1819 | ||
3f072a7f HR |
1820 | if (target_is_backing) { |
1821 | int64_t bs_size, target_size; | |
1822 | bs_size = bdrv_getlength(bs); | |
1823 | if (bs_size < 0) { | |
1824 | error_setg_errno(errp, -bs_size, | |
1825 | "Could not inquire top image size"); | |
1826 | goto fail; | |
1827 | } | |
1828 | ||
1829 | target_size = bdrv_getlength(target); | |
1830 | if (target_size < 0) { | |
1831 | error_setg_errno(errp, -target_size, | |
1832 | "Could not inquire base image size"); | |
1833 | goto fail; | |
1834 | } | |
1835 | ||
1836 | if (target_size < bs_size) { | |
1837 | target_perms |= BLK_PERM_RESIZE; | |
1838 | } | |
1839 | ||
64631f36 | 1840 | target_shared_perms |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE; |
ad74751f KW |
1841 | } else { |
1842 | bdrv_graph_rdlock_main_loop(); | |
1843 | if (bdrv_chain_contains(bs, bdrv_skip_filters(target))) { | |
1844 | /* | |
1845 | * We may want to allow this in the future, but it would | |
1846 | * require taking some extra care. | |
1847 | */ | |
1848 | error_setg(errp, "Cannot mirror to a filter on top of a node in " | |
1849 | "the source's backing chain"); | |
1850 | bdrv_graph_rdunlock_main_loop(); | |
1851 | goto fail; | |
1852 | } | |
1853 | bdrv_graph_rdunlock_main_loop(); | |
3f072a7f HR |
1854 | } |
1855 | ||
d861ab3a | 1856 | s->target = blk_new(s->common.job.aio_context, |
3f072a7f | 1857 | target_perms, target_shared_perms); |
d7086422 KW |
1858 | ret = blk_insert_bs(s->target, target, errp); |
1859 | if (ret < 0) { | |
4ef85a9c | 1860 | goto fail; |
d7086422 | 1861 | } |
045a2f82 FZ |
1862 | if (is_mirror) { |
1863 | /* XXX: Mirror target could be a NBD server of target QEMU in the case | |
1864 | * of non-shared block migration. To allow migration completion, we | |
1865 | * have to allow "inactivate" of the target BB. When that happens, we | |
1866 | * know the job is drained, and the vcpus are stopped, so no write | |
1867 | * operation will be performed. Block layer already has assertions to | |
1868 | * ensure that. */ | |
1869 | blk_set_force_allow_inactivate(s->target); | |
1870 | } | |
9ff7f0df | 1871 | blk_set_allow_aio_context_change(s->target, true); |
cf312932 | 1872 | blk_set_disable_request_queuing(s->target, true); |
e253f4b8 | 1873 | |
ad74751f | 1874 | bdrv_graph_rdlock_main_loop(); |
09158f00 | 1875 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1876 | s->on_source_error = on_source_error; |
1877 | s->on_target_error = on_target_error; | |
03544a6e | 1878 | s->is_none_mode = is_none_mode; |
274fccee | 1879 | s->backing_mode = backing_mode; |
cdf3bc93 | 1880 | s->zero_target = zero_target; |
2d400d15 | 1881 | qatomic_set(&s->copy_mode, copy_mode); |
5bc361b8 | 1882 | s->base = base; |
3f072a7f | 1883 | s->base_overlay = bdrv_find_overlay(bs, base); |
eee13dfe | 1884 | s->granularity = granularity; |
48ac0a4d | 1885 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1886 | s->unmap = unmap; |
b49f7ead WC |
1887 | if (auto_complete) { |
1888 | s->should_complete = true; | |
1889 | } | |
ad74751f | 1890 | bdrv_graph_rdunlock_main_loop(); |
b812f671 | 1891 | |
058cfca5 FE |
1892 | s->dirty_bitmap = bdrv_create_dirty_bitmap(s->mirror_top_bs, granularity, |
1893 | NULL, errp); | |
b8afb520 | 1894 | if (!s->dirty_bitmap) { |
88f9d1b3 | 1895 | goto fail; |
b8afb520 | 1896 | } |
058cfca5 FE |
1897 | |
1898 | /* | |
1899 | * The dirty bitmap is set by bdrv_mirror_top_do_write() when not in active | |
1900 | * mode. | |
1901 | */ | |
1902 | bdrv_disable_dirty_bitmap(s->dirty_bitmap); | |
10f3cd15 | 1903 | |
f3bbc53d | 1904 | bdrv_graph_wrlock(bs); |
67b24427 AG |
1905 | ret = block_job_add_bdrv(&s->common, "source", bs, 0, |
1906 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE | | |
1907 | BLK_PERM_CONSISTENT_READ, | |
1908 | errp); | |
1909 | if (ret < 0) { | |
f3bbc53d | 1910 | bdrv_graph_wrunlock(); |
67b24427 AG |
1911 | goto fail; |
1912 | } | |
1913 | ||
4ef85a9c | 1914 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1915 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1916 | &error_abort); | |
1917 | ||
f3ede4b0 AG |
1918 | /* In commit_active_start() all intermediate nodes disappear, so |
1919 | * any jobs in them must be blocked */ | |
4ef85a9c | 1920 | if (target_is_backing) { |
3f072a7f HR |
1921 | BlockDriverState *iter, *filtered_target; |
1922 | uint64_t iter_shared_perms; | |
1923 | ||
1924 | /* | |
1925 | * The topmost node with | |
1926 | * bdrv_skip_filters(filtered_target) == bdrv_skip_filters(target) | |
1927 | */ | |
1928 | filtered_target = bdrv_cow_bs(bdrv_find_overlay(bs, target)); | |
1929 | ||
1930 | assert(bdrv_skip_filters(filtered_target) == | |
1931 | bdrv_skip_filters(target)); | |
1932 | ||
1933 | /* | |
1934 | * XXX BLK_PERM_WRITE needs to be allowed so we don't block | |
1935 | * ourselves at s->base (if writes are blocked for a node, they are | |
1936 | * also blocked for its backing file). The other options would be a | |
1937 | * second filter driver above s->base (== target). | |
1938 | */ | |
1939 | iter_shared_perms = BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE; | |
1940 | ||
1941 | for (iter = bdrv_filter_or_cow_bs(bs); iter != target; | |
1942 | iter = bdrv_filter_or_cow_bs(iter)) | |
1943 | { | |
1944 | if (iter == filtered_target) { | |
1945 | /* | |
1946 | * From here on, all nodes are filters on the base. | |
1947 | * This allows us to share BLK_PERM_CONSISTENT_READ. | |
1948 | */ | |
1949 | iter_shared_perms |= BLK_PERM_CONSISTENT_READ; | |
1950 | } | |
1951 | ||
4ef85a9c | 1952 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, |
3f072a7f | 1953 | iter_shared_perms, errp); |
4ef85a9c | 1954 | if (ret < 0) { |
f3bbc53d | 1955 | bdrv_graph_wrunlock(); |
4ef85a9c KW |
1956 | goto fail; |
1957 | } | |
f3ede4b0 | 1958 | } |
ef53dc09 AG |
1959 | |
1960 | if (bdrv_freeze_backing_chain(mirror_top_bs, target, errp) < 0) { | |
f3bbc53d | 1961 | bdrv_graph_wrunlock(); |
ef53dc09 AG |
1962 | goto fail; |
1963 | } | |
f3ede4b0 | 1964 | } |
f3bbc53d | 1965 | bdrv_graph_wrunlock(); |
10f3cd15 | 1966 | |
12aa4082 HR |
1967 | QTAILQ_INIT(&s->ops_in_flight); |
1968 | ||
5ccac6f1 | 1969 | trace_mirror_start(bs, s, opaque); |
da01ff7f | 1970 | job_start(&s->common.job); |
cc19f177 VSO |
1971 | |
1972 | return &s->common; | |
4ef85a9c KW |
1973 | |
1974 | fail: | |
1975 | if (s) { | |
7a25fcd0 HR |
1976 | /* Make sure this BDS does not go away until we have completed the graph |
1977 | * changes below */ | |
1978 | bdrv_ref(mirror_top_bs); | |
1979 | ||
4ef85a9c KW |
1980 | g_free(s->replaces); |
1981 | blk_unref(s->target); | |
429076e8 | 1982 | bs_opaque->job = NULL; |
e917e2cb | 1983 | if (s->dirty_bitmap) { |
5deb6cbd | 1984 | bdrv_release_dirty_bitmap(s->dirty_bitmap); |
e917e2cb | 1985 | } |
4ad35181 | 1986 | job_early_fail(&s->common.job); |
4ef85a9c KW |
1987 | } |
1988 | ||
f94dc3b4 | 1989 | bs_opaque->stop = true; |
3804e3cf | 1990 | bdrv_graph_rdlock_main_loop(); |
f94dc3b4 HR |
1991 | bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing, |
1992 | &error_abort); | |
3804e3cf | 1993 | bdrv_graph_rdunlock_main_loop(); |
3f072a7f | 1994 | bdrv_replace_node(mirror_top_bs, mirror_top_bs->backing->bs, &error_abort); |
7a25fcd0 HR |
1995 | |
1996 | bdrv_unref(mirror_top_bs); | |
cc19f177 VSO |
1997 | |
1998 | return NULL; | |
893f7eba | 1999 | } |
03544a6e | 2000 | |
71aa9867 AG |
2001 | void mirror_start(const char *job_id, BlockDriverState *bs, |
2002 | BlockDriverState *target, const char *replaces, | |
a1999b33 JS |
2003 | int creation_flags, int64_t speed, |
2004 | uint32_t granularity, int64_t buf_size, | |
274fccee | 2005 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
cdf3bc93 | 2006 | bool zero_target, |
274fccee | 2007 | BlockdevOnError on_source_error, |
03544a6e | 2008 | BlockdevOnError on_target_error, |
481debaa HR |
2009 | bool unmap, const char *filter_node_name, |
2010 | MirrorCopyMode copy_mode, Error **errp) | |
03544a6e FZ |
2011 | { |
2012 | bool is_none_mode; | |
2013 | BlockDriverState *base; | |
2014 | ||
b4ad82aa EGE |
2015 | GLOBAL_STATE_CODE(); |
2016 | ||
c8b56501 JS |
2017 | if ((mode == MIRROR_SYNC_MODE_INCREMENTAL) || |
2018 | (mode == MIRROR_SYNC_MODE_BITMAP)) { | |
2019 | error_setg(errp, "Sync mode '%s' not supported", | |
2020 | MirrorSyncMode_str(mode)); | |
d58d8453 JS |
2021 | return; |
2022 | } | |
ad74751f KW |
2023 | |
2024 | bdrv_graph_rdlock_main_loop(); | |
03544a6e | 2025 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
3f072a7f | 2026 | base = mode == MIRROR_SYNC_MODE_TOP ? bdrv_backing_chain_next(bs) : NULL; |
ad74751f KW |
2027 | bdrv_graph_rdunlock_main_loop(); |
2028 | ||
a1999b33 | 2029 | mirror_start_job(job_id, bs, creation_flags, target, replaces, |
cdf3bc93 | 2030 | speed, granularity, buf_size, backing_mode, zero_target, |
51ccfa2d | 2031 | on_source_error, on_target_error, unmap, NULL, NULL, |
6cdbceb1 | 2032 | &mirror_job_driver, is_none_mode, base, false, |
481debaa | 2033 | filter_node_name, true, copy_mode, errp); |
03544a6e FZ |
2034 | } |
2035 | ||
cc19f177 VSO |
2036 | BlockJob *commit_active_start(const char *job_id, BlockDriverState *bs, |
2037 | BlockDriverState *base, int creation_flags, | |
2038 | int64_t speed, BlockdevOnError on_error, | |
2039 | const char *filter_node_name, | |
2040 | BlockCompletionFunc *cb, void *opaque, | |
2041 | bool auto_complete, Error **errp) | |
03544a6e | 2042 | { |
1ba79388 | 2043 | bool base_read_only; |
eb5becc1 | 2044 | BlockJob *job; |
4da83585 | 2045 | |
b4ad82aa EGE |
2046 | GLOBAL_STATE_CODE(); |
2047 | ||
1ba79388 | 2048 | base_read_only = bdrv_is_read_only(base); |
4da83585 | 2049 | |
1ba79388 AG |
2050 | if (base_read_only) { |
2051 | if (bdrv_reopen_set_read_only(base, false, errp) < 0) { | |
cc19f177 | 2052 | return NULL; |
1ba79388 | 2053 | } |
20a63d2c | 2054 | } |
4da83585 | 2055 | |
eb5becc1 | 2056 | job = mirror_start_job( |
cc19f177 | 2057 | job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
cdf3bc93 | 2058 | MIRROR_LEAVE_BACKING_CHAIN, false, |
51ccfa2d | 2059 | on_error, on_error, true, cb, opaque, |
6cdbceb1 | 2060 | &commit_active_job_driver, false, base, auto_complete, |
481debaa | 2061 | filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND, |
eb5becc1 VSO |
2062 | errp); |
2063 | if (!job) { | |
4da83585 JC |
2064 | goto error_restore_flags; |
2065 | } | |
2066 | ||
eb5becc1 | 2067 | return job; |
4da83585 JC |
2068 | |
2069 | error_restore_flags: | |
2070 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
2071 | * the original error */ | |
1ba79388 AG |
2072 | if (base_read_only) { |
2073 | bdrv_reopen_set_read_only(base, true, NULL); | |
2074 | } | |
cc19f177 | 2075 | return NULL; |
03544a6e | 2076 | } |