]>
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; |
4ef85a9c | 45 | |
09158f00 BC |
46 | /* The name of the graph node to replace */ |
47 | char *replaces; | |
48 | /* The BDS to replace */ | |
49 | BlockDriverState *to_replace; | |
50 | /* Used to block operations on the drive-mirror-replace target */ | |
51 | Error *replace_blocker; | |
03544a6e | 52 | bool is_none_mode; |
274fccee | 53 | BlockMirrorBackingMode backing_mode; |
d06107ad | 54 | MirrorCopyMode copy_mode; |
b952b558 | 55 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 | 56 | bool synced; |
d06107ad HR |
57 | /* Set when the target is synced (dirty bitmap is clean, nothing |
58 | * in flight) and the job is running in active mode */ | |
59 | bool actively_synced; | |
d63ffd87 | 60 | bool should_complete; |
eee13dfe | 61 | int64_t granularity; |
b812f671 | 62 | size_t buf_size; |
b21c7652 | 63 | int64_t bdev_length; |
b812f671 | 64 | unsigned long *cow_bitmap; |
e4654d2d | 65 | BdrvDirtyBitmap *dirty_bitmap; |
dc162c8e | 66 | BdrvDirtyBitmapIter *dbi; |
893f7eba | 67 | uint8_t *buf; |
402a4741 PB |
68 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
69 | int buf_free_count; | |
bd48bde8 | 70 | |
49efb1f5 | 71 | uint64_t last_pause_ns; |
402a4741 | 72 | unsigned long *in_flight_bitmap; |
bd48bde8 | 73 | int in_flight; |
b436982f | 74 | int64_t bytes_in_flight; |
12aa4082 | 75 | QTAILQ_HEAD(MirrorOpList, MirrorOp) ops_in_flight; |
bd48bde8 | 76 | int ret; |
0fc9f8ea | 77 | bool unmap; |
b436982f | 78 | int target_cluster_size; |
e5b43573 | 79 | int max_iov; |
90ab48eb | 80 | bool initial_zeroing_ongoing; |
d06107ad | 81 | int in_active_write_counter; |
893f7eba PB |
82 | } MirrorBlockJob; |
83 | ||
429076e8 HR |
84 | typedef struct MirrorBDSOpaque { |
85 | MirrorBlockJob *job; | |
86 | } MirrorBDSOpaque; | |
87 | ||
12aa4082 | 88 | struct MirrorOp { |
bd48bde8 PB |
89 | MirrorBlockJob *s; |
90 | QEMUIOVector qiov; | |
b436982f EB |
91 | int64_t offset; |
92 | uint64_t bytes; | |
2e1990b2 HR |
93 | |
94 | /* The pointee is set by mirror_co_read(), mirror_co_zero(), and | |
95 | * mirror_co_discard() before yielding for the first time */ | |
96 | int64_t *bytes_handled; | |
12aa4082 | 97 | |
1181e19a | 98 | bool is_pseudo_op; |
d06107ad | 99 | bool is_active_write; |
12aa4082 HR |
100 | CoQueue waiting_requests; |
101 | ||
102 | QTAILQ_ENTRY(MirrorOp) next; | |
103 | }; | |
bd48bde8 | 104 | |
4295c5fc HR |
105 | typedef enum MirrorMethod { |
106 | MIRROR_METHOD_COPY, | |
107 | MIRROR_METHOD_ZERO, | |
108 | MIRROR_METHOD_DISCARD, | |
109 | } MirrorMethod; | |
110 | ||
b952b558 PB |
111 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
112 | int error) | |
113 | { | |
114 | s->synced = false; | |
d06107ad | 115 | s->actively_synced = false; |
b952b558 | 116 | if (read) { |
81e254dc KW |
117 | return block_job_error_action(&s->common, s->on_source_error, |
118 | true, error); | |
b952b558 | 119 | } else { |
81e254dc KW |
120 | return block_job_error_action(&s->common, s->on_target_error, |
121 | false, error); | |
b952b558 PB |
122 | } |
123 | } | |
124 | ||
1181e19a HR |
125 | static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self, |
126 | MirrorBlockJob *s, | |
127 | uint64_t offset, | |
128 | uint64_t bytes) | |
129 | { | |
130 | uint64_t self_start_chunk = offset / s->granularity; | |
131 | uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); | |
132 | uint64_t self_nb_chunks = self_end_chunk - self_start_chunk; | |
133 | ||
134 | while (find_next_bit(s->in_flight_bitmap, self_end_chunk, | |
135 | self_start_chunk) < self_end_chunk && | |
136 | s->ret >= 0) | |
137 | { | |
138 | MirrorOp *op; | |
139 | ||
140 | QTAILQ_FOREACH(op, &s->ops_in_flight, next) { | |
141 | uint64_t op_start_chunk = op->offset / s->granularity; | |
142 | uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes, | |
143 | s->granularity) - | |
144 | op_start_chunk; | |
145 | ||
146 | if (op == self) { | |
147 | continue; | |
148 | } | |
149 | ||
150 | if (ranges_overlap(self_start_chunk, self_nb_chunks, | |
151 | op_start_chunk, op_nb_chunks)) | |
152 | { | |
153 | qemu_co_queue_wait(&op->waiting_requests, NULL); | |
154 | break; | |
155 | } | |
156 | } | |
157 | } | |
158 | } | |
159 | ||
2e1990b2 | 160 | static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) |
bd48bde8 PB |
161 | { |
162 | MirrorBlockJob *s = op->s; | |
402a4741 | 163 | struct iovec *iov; |
bd48bde8 | 164 | int64_t chunk_num; |
b436982f | 165 | int i, nb_chunks; |
bd48bde8 | 166 | |
b436982f | 167 | trace_mirror_iteration_done(s, op->offset, op->bytes, ret); |
bd48bde8 PB |
168 | |
169 | s->in_flight--; | |
b436982f | 170 | s->bytes_in_flight -= op->bytes; |
402a4741 PB |
171 | iov = op->qiov.iov; |
172 | for (i = 0; i < op->qiov.niov; i++) { | |
173 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
174 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
175 | s->buf_free_count++; | |
176 | } | |
177 | ||
b436982f EB |
178 | chunk_num = op->offset / s->granularity; |
179 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
12aa4082 | 180 | |
402a4741 | 181 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
12aa4082 | 182 | QTAILQ_REMOVE(&s->ops_in_flight, op, next); |
b21c7652 HR |
183 | if (ret >= 0) { |
184 | if (s->cow_bitmap) { | |
185 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
186 | } | |
90ab48eb | 187 | if (!s->initial_zeroing_ongoing) { |
30a5c887 | 188 | job_progress_update(&s->common.job, op->bytes); |
90ab48eb | 189 | } |
bd48bde8 | 190 | } |
6df3bf8e | 191 | qemu_iovec_destroy(&op->qiov); |
7b770c72 | 192 | |
12aa4082 HR |
193 | qemu_co_queue_restart_all(&op->waiting_requests); |
194 | g_free(op); | |
bd48bde8 PB |
195 | } |
196 | ||
2e1990b2 | 197 | static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret) |
bd48bde8 | 198 | { |
bd48bde8 | 199 | MirrorBlockJob *s = op->s; |
b9e413dd PB |
200 | |
201 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 202 | if (ret < 0) { |
bd48bde8 PB |
203 | BlockErrorAction action; |
204 | ||
e0d7f73e | 205 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 206 | action = mirror_error_action(s, false, -ret); |
a589569f | 207 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
208 | s->ret = ret; |
209 | } | |
210 | } | |
211 | mirror_iteration_done(op, ret); | |
b9e413dd | 212 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
213 | } |
214 | ||
2e1990b2 | 215 | static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret) |
bd48bde8 | 216 | { |
bd48bde8 | 217 | MirrorBlockJob *s = op->s; |
b9e413dd PB |
218 | |
219 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 220 | if (ret < 0) { |
bd48bde8 PB |
221 | BlockErrorAction action; |
222 | ||
e0d7f73e | 223 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 224 | action = mirror_error_action(s, true, -ret); |
a589569f | 225 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
226 | s->ret = ret; |
227 | } | |
228 | ||
229 | mirror_iteration_done(op, ret); | |
b9e413dd | 230 | } else { |
2e1990b2 HR |
231 | ret = blk_co_pwritev(s->target, op->offset, |
232 | op->qiov.size, &op->qiov, 0); | |
233 | mirror_write_complete(op, ret); | |
bd48bde8 | 234 | } |
b9e413dd | 235 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
236 | } |
237 | ||
782d97ef EB |
238 | /* Clip bytes relative to offset to not exceed end-of-file */ |
239 | static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, | |
240 | int64_t offset, | |
241 | int64_t bytes) | |
242 | { | |
243 | return MIN(bytes, s->bdev_length - offset); | |
244 | } | |
245 | ||
782d97ef EB |
246 | /* Round offset and/or bytes to target cluster if COW is needed, and |
247 | * return the offset of the adjusted tail against original. */ | |
248 | static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, | |
ae4cc877 | 249 | uint64_t *bytes) |
893f7eba | 250 | { |
e5b43573 FZ |
251 | bool need_cow; |
252 | int ret = 0; | |
782d97ef | 253 | int64_t align_offset = *offset; |
7cfd5275 | 254 | int64_t align_bytes = *bytes; |
782d97ef | 255 | int max_bytes = s->granularity * s->max_iov; |
e5b43573 | 256 | |
782d97ef EB |
257 | need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); |
258 | need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, | |
e5b43573 FZ |
259 | s->cow_bitmap); |
260 | if (need_cow) { | |
782d97ef EB |
261 | bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, |
262 | &align_offset, &align_bytes); | |
e5b43573 | 263 | } |
3515727f | 264 | |
782d97ef EB |
265 | if (align_bytes > max_bytes) { |
266 | align_bytes = max_bytes; | |
e5b43573 | 267 | if (need_cow) { |
782d97ef | 268 | align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); |
e5b43573 | 269 | } |
8f0720ec | 270 | } |
782d97ef | 271 | /* Clipping may result in align_bytes unaligned to chunk boundary, but |
4150ae60 | 272 | * that doesn't matter because it's already the end of source image. */ |
782d97ef | 273 | align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); |
8f0720ec | 274 | |
782d97ef EB |
275 | ret = align_offset + align_bytes - (*offset + *bytes); |
276 | *offset = align_offset; | |
277 | *bytes = align_bytes; | |
e5b43573 FZ |
278 | assert(ret >= 0); |
279 | return ret; | |
280 | } | |
281 | ||
d06107ad | 282 | static inline void mirror_wait_for_any_operation(MirrorBlockJob *s, bool active) |
21cd917f | 283 | { |
12aa4082 HR |
284 | MirrorOp *op; |
285 | ||
1181e19a HR |
286 | QTAILQ_FOREACH(op, &s->ops_in_flight, next) { |
287 | /* Do not wait on pseudo ops, because it may in turn wait on | |
288 | * some other operation to start, which may in fact be the | |
289 | * caller of this function. Since there is only one pseudo op | |
290 | * at any given time, we will always find some real operation | |
291 | * to wait on. */ | |
d06107ad | 292 | if (!op->is_pseudo_op && op->is_active_write == active) { |
1181e19a HR |
293 | qemu_co_queue_wait(&op->waiting_requests, NULL); |
294 | return; | |
295 | } | |
296 | } | |
297 | abort(); | |
21cd917f FZ |
298 | } |
299 | ||
d06107ad HR |
300 | static inline void mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) |
301 | { | |
302 | /* Only non-active operations use up in-flight slots */ | |
303 | mirror_wait_for_any_operation(s, false); | |
304 | } | |
305 | ||
2e1990b2 HR |
306 | /* Perform a mirror copy operation. |
307 | * | |
308 | * *op->bytes_handled is set to the number of bytes copied after and | |
309 | * including offset, excluding any bytes copied prior to offset due | |
310 | * to alignment. This will be op->bytes if no alignment is necessary, | |
311 | * or (new_end - op->offset) if the tail is rounded up or down due to | |
312 | * alignment or buffer limit. | |
e5b43573 | 313 | */ |
2e1990b2 | 314 | static void coroutine_fn mirror_co_read(void *opaque) |
e5b43573 | 315 | { |
2e1990b2 HR |
316 | MirrorOp *op = opaque; |
317 | MirrorBlockJob *s = op->s; | |
ae4cc877 EB |
318 | int nb_chunks; |
319 | uint64_t ret; | |
ae4cc877 | 320 | uint64_t max_bytes; |
e5b43573 | 321 | |
ae4cc877 | 322 | max_bytes = s->granularity * s->max_iov; |
402a4741 | 323 | |
e5b43573 | 324 | /* We can only handle as much as buf_size at a time. */ |
2e1990b2 HR |
325 | op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes)); |
326 | assert(op->bytes); | |
327 | assert(op->bytes < BDRV_REQUEST_MAX_BYTES); | |
328 | *op->bytes_handled = op->bytes; | |
402a4741 | 329 | |
e5b43573 | 330 | if (s->cow_bitmap) { |
2e1990b2 | 331 | *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes); |
e5b43573 | 332 | } |
2e1990b2 HR |
333 | /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */ |
334 | assert(*op->bytes_handled <= UINT_MAX); | |
335 | assert(op->bytes <= s->buf_size); | |
ae4cc877 | 336 | /* The offset is granularity-aligned because: |
e5b43573 FZ |
337 | * 1) Caller passes in aligned values; |
338 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
2e1990b2 | 339 | assert(QEMU_IS_ALIGNED(op->offset, s->granularity)); |
ae4cc877 | 340 | /* The range is sector-aligned, since bdrv_getlength() rounds up. */ |
2e1990b2 HR |
341 | assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE)); |
342 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
e5b43573 FZ |
343 | |
344 | while (s->buf_free_count < nb_chunks) { | |
2e1990b2 | 345 | trace_mirror_yield_in_flight(s, op->offset, s->in_flight); |
1181e19a | 346 | mirror_wait_for_free_in_flight_slot(s); |
b812f671 PB |
347 | } |
348 | ||
402a4741 PB |
349 | /* Now make a QEMUIOVector taking enough granularity-sized chunks |
350 | * from s->buf_free. | |
351 | */ | |
352 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
353 | while (nb_chunks-- > 0) { |
354 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
2e1990b2 | 355 | size_t remaining = op->bytes - op->qiov.size; |
5a0f6fd5 | 356 | |
402a4741 PB |
357 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
358 | s->buf_free_count--; | |
5a0f6fd5 | 359 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 360 | } |
bd48bde8 | 361 | |
893f7eba | 362 | /* Copy the dirty cluster. */ |
bd48bde8 | 363 | s->in_flight++; |
2e1990b2 HR |
364 | s->bytes_in_flight += op->bytes; |
365 | trace_mirror_one_iteration(s, op->offset, op->bytes); | |
dcfb3beb | 366 | |
138f9fff HR |
367 | ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes, |
368 | &op->qiov, 0); | |
2e1990b2 | 369 | mirror_read_complete(op, ret); |
e5b43573 FZ |
370 | } |
371 | ||
2e1990b2 | 372 | static void coroutine_fn mirror_co_zero(void *opaque) |
e5b43573 | 373 | { |
2e1990b2 HR |
374 | MirrorOp *op = opaque; |
375 | int ret; | |
e5b43573 | 376 | |
2e1990b2 HR |
377 | op->s->in_flight++; |
378 | op->s->bytes_in_flight += op->bytes; | |
379 | *op->bytes_handled = op->bytes; | |
e5b43573 | 380 | |
2e1990b2 HR |
381 | ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes, |
382 | op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0); | |
383 | mirror_write_complete(op, ret); | |
384 | } | |
385 | ||
386 | static void coroutine_fn mirror_co_discard(void *opaque) | |
387 | { | |
388 | MirrorOp *op = opaque; | |
389 | int ret; | |
390 | ||
391 | op->s->in_flight++; | |
392 | op->s->bytes_in_flight += op->bytes; | |
393 | *op->bytes_handled = op->bytes; | |
394 | ||
395 | ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes); | |
396 | mirror_write_complete(op, ret); | |
e5b43573 FZ |
397 | } |
398 | ||
4295c5fc HR |
399 | static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, |
400 | unsigned bytes, MirrorMethod mirror_method) | |
401 | { | |
2e1990b2 HR |
402 | MirrorOp *op; |
403 | Coroutine *co; | |
404 | int64_t bytes_handled = -1; | |
405 | ||
406 | op = g_new(MirrorOp, 1); | |
407 | *op = (MirrorOp){ | |
408 | .s = s, | |
409 | .offset = offset, | |
410 | .bytes = bytes, | |
411 | .bytes_handled = &bytes_handled, | |
412 | }; | |
12aa4082 | 413 | qemu_co_queue_init(&op->waiting_requests); |
2e1990b2 | 414 | |
4295c5fc HR |
415 | switch (mirror_method) { |
416 | case MIRROR_METHOD_COPY: | |
2e1990b2 HR |
417 | co = qemu_coroutine_create(mirror_co_read, op); |
418 | break; | |
4295c5fc | 419 | case MIRROR_METHOD_ZERO: |
2e1990b2 HR |
420 | co = qemu_coroutine_create(mirror_co_zero, op); |
421 | break; | |
4295c5fc | 422 | case MIRROR_METHOD_DISCARD: |
2e1990b2 HR |
423 | co = qemu_coroutine_create(mirror_co_discard, op); |
424 | break; | |
4295c5fc HR |
425 | default: |
426 | abort(); | |
427 | } | |
2e1990b2 | 428 | |
12aa4082 | 429 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); |
2e1990b2 HR |
430 | qemu_coroutine_enter(co); |
431 | /* At this point, ownership of op has been moved to the coroutine | |
432 | * and the object may already be freed */ | |
433 | ||
434 | /* Assert that this value has been set */ | |
435 | assert(bytes_handled >= 0); | |
436 | ||
437 | /* Same assertion as in mirror_co_read() (and for mirror_co_read() | |
438 | * and mirror_co_discard(), bytes_handled == op->bytes, which | |
439 | * is the @bytes parameter given to this function) */ | |
440 | assert(bytes_handled <= UINT_MAX); | |
441 | return bytes_handled; | |
4295c5fc HR |
442 | } |
443 | ||
e5b43573 FZ |
444 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) |
445 | { | |
138f9fff | 446 | BlockDriverState *source = s->mirror_top_bs->backing->bs; |
1181e19a HR |
447 | MirrorOp *pseudo_op; |
448 | int64_t offset; | |
449 | uint64_t delay_ns = 0, ret = 0; | |
e5b43573 FZ |
450 | /* At least the first dirty chunk is mirrored in one iteration. */ |
451 | int nb_chunks = 1; | |
4b5004d9 | 452 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
b436982f | 453 | int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); |
e5b43573 | 454 | |
b64bd51e | 455 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
f798184c | 456 | offset = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 457 | if (offset < 0) { |
dc162c8e | 458 | bdrv_set_dirty_iter(s->dbi, 0); |
f798184c | 459 | offset = bdrv_dirty_iter_next(s->dbi); |
9a46dba7 | 460 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
fb2ef791 | 461 | assert(offset >= 0); |
e5b43573 | 462 | } |
b64bd51e | 463 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
e5b43573 | 464 | |
1181e19a | 465 | mirror_wait_on_conflicts(NULL, s, offset, 1); |
9c83625b | 466 | |
da01ff7f | 467 | job_pause_point(&s->common.job); |
565ac01f | 468 | |
e5b43573 FZ |
469 | /* Find the number of consective dirty chunks following the first dirty |
470 | * one, and wait for in flight requests in them. */ | |
b64bd51e | 471 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
fb2ef791 | 472 | while (nb_chunks * s->granularity < s->buf_size) { |
dc162c8e | 473 | int64_t next_dirty; |
fb2ef791 EB |
474 | int64_t next_offset = offset + nb_chunks * s->granularity; |
475 | int64_t next_chunk = next_offset / s->granularity; | |
476 | if (next_offset >= s->bdev_length || | |
3b5d4df0 | 477 | !bdrv_get_dirty_locked(source, s->dirty_bitmap, next_offset)) { |
e5b43573 FZ |
478 | break; |
479 | } | |
480 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 481 | break; |
e5b43573 | 482 | } |
9c83625b | 483 | |
f798184c | 484 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 485 | if (next_dirty > next_offset || next_dirty < 0) { |
f27a2742 | 486 | /* The bitmap iterator's cache is stale, refresh it */ |
715a74d8 | 487 | bdrv_set_dirty_iter(s->dbi, next_offset); |
f798184c | 488 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
f27a2742 | 489 | } |
fb2ef791 | 490 | assert(next_dirty == next_offset); |
9c83625b | 491 | nb_chunks++; |
e5b43573 FZ |
492 | } |
493 | ||
494 | /* Clear dirty bits before querying the block status, because | |
31826642 | 495 | * calling bdrv_block_status_above could yield - if some blocks are |
e5b43573 FZ |
496 | * marked dirty in this window, we need to know. |
497 | */ | |
e0d7f73e EB |
498 | bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, |
499 | nb_chunks * s->granularity); | |
b64bd51e PB |
500 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
501 | ||
1181e19a HR |
502 | /* Before claiming an area in the in-flight bitmap, we have to |
503 | * create a MirrorOp for it so that conflicting requests can wait | |
504 | * for it. mirror_perform() will create the real MirrorOps later, | |
505 | * for now we just create a pseudo operation that will wake up all | |
506 | * conflicting requests once all real operations have been | |
507 | * launched. */ | |
508 | pseudo_op = g_new(MirrorOp, 1); | |
509 | *pseudo_op = (MirrorOp){ | |
510 | .offset = offset, | |
511 | .bytes = nb_chunks * s->granularity, | |
512 | .is_pseudo_op = true, | |
513 | }; | |
514 | qemu_co_queue_init(&pseudo_op->waiting_requests); | |
515 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next); | |
516 | ||
fb2ef791 EB |
517 | bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); |
518 | while (nb_chunks > 0 && offset < s->bdev_length) { | |
31826642 | 519 | int ret; |
7cfd5275 | 520 | int64_t io_bytes; |
f3e4ce4a | 521 | int64_t io_bytes_acct; |
4295c5fc | 522 | MirrorMethod mirror_method = MIRROR_METHOD_COPY; |
e5b43573 | 523 | |
fb2ef791 | 524 | assert(!(offset % s->granularity)); |
31826642 EB |
525 | ret = bdrv_block_status_above(source, NULL, offset, |
526 | nb_chunks * s->granularity, | |
527 | &io_bytes, NULL, NULL); | |
e5b43573 | 528 | if (ret < 0) { |
fb2ef791 | 529 | io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); |
0965a41e | 530 | } else if (ret & BDRV_BLOCK_DATA) { |
fb2ef791 | 531 | io_bytes = MIN(io_bytes, max_io_bytes); |
e5b43573 FZ |
532 | } |
533 | ||
fb2ef791 EB |
534 | io_bytes -= io_bytes % s->granularity; |
535 | if (io_bytes < s->granularity) { | |
536 | io_bytes = s->granularity; | |
e5b43573 | 537 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { |
fb2ef791 | 538 | int64_t target_offset; |
7cfd5275 | 539 | int64_t target_bytes; |
fb2ef791 EB |
540 | bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes, |
541 | &target_offset, &target_bytes); | |
542 | if (target_offset == offset && | |
543 | target_bytes == io_bytes) { | |
e5b43573 FZ |
544 | mirror_method = ret & BDRV_BLOCK_ZERO ? |
545 | MIRROR_METHOD_ZERO : | |
546 | MIRROR_METHOD_DISCARD; | |
547 | } | |
548 | } | |
549 | ||
cf56a3c6 | 550 | while (s->in_flight >= MAX_IN_FLIGHT) { |
fb2ef791 | 551 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
1181e19a | 552 | mirror_wait_for_free_in_flight_slot(s); |
cf56a3c6 DL |
553 | } |
554 | ||
dbaa7b57 | 555 | if (s->ret < 0) { |
1181e19a HR |
556 | ret = 0; |
557 | goto fail; | |
dbaa7b57 VSO |
558 | } |
559 | ||
fb2ef791 | 560 | io_bytes = mirror_clip_bytes(s, offset, io_bytes); |
4295c5fc HR |
561 | io_bytes = mirror_perform(s, offset, io_bytes, mirror_method); |
562 | if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) { | |
563 | io_bytes_acct = 0; | |
564 | } else { | |
565 | io_bytes_acct = io_bytes; | |
e5b43573 | 566 | } |
fb2ef791 EB |
567 | assert(io_bytes); |
568 | offset += io_bytes; | |
569 | nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); | |
dee81d51 | 570 | delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct); |
dcfb3beb | 571 | } |
1181e19a HR |
572 | |
573 | ret = delay_ns; | |
574 | fail: | |
575 | QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next); | |
576 | qemu_co_queue_restart_all(&pseudo_op->waiting_requests); | |
577 | g_free(pseudo_op); | |
578 | ||
579 | return ret; | |
bd48bde8 | 580 | } |
b952b558 | 581 | |
402a4741 PB |
582 | static void mirror_free_init(MirrorBlockJob *s) |
583 | { | |
584 | int granularity = s->granularity; | |
585 | size_t buf_size = s->buf_size; | |
586 | uint8_t *buf = s->buf; | |
587 | ||
588 | assert(s->buf_free_count == 0); | |
589 | QSIMPLEQ_INIT(&s->buf_free); | |
590 | while (buf_size != 0) { | |
591 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
592 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
593 | s->buf_free_count++; | |
594 | buf_size -= granularity; | |
595 | buf += granularity; | |
596 | } | |
597 | } | |
598 | ||
bae8196d PB |
599 | /* This is also used for the .pause callback. There is no matching |
600 | * mirror_resume() because mirror_run() will begin iterating again | |
601 | * when the job is resumed. | |
602 | */ | |
603 | static void mirror_wait_for_all_io(MirrorBlockJob *s) | |
bd48bde8 PB |
604 | { |
605 | while (s->in_flight > 0) { | |
1181e19a | 606 | mirror_wait_for_free_in_flight_slot(s); |
bd48bde8 | 607 | } |
893f7eba PB |
608 | } |
609 | ||
5a7e7a0b SH |
610 | typedef struct { |
611 | int ret; | |
612 | } MirrorExitData; | |
613 | ||
1908a559 | 614 | static void mirror_exit(Job *job, void *opaque) |
5a7e7a0b | 615 | { |
1908a559 KW |
616 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
617 | BlockJob *bjob = &s->common; | |
5a7e7a0b | 618 | MirrorExitData *data = opaque; |
429076e8 | 619 | MirrorBDSOpaque *bs_opaque = s->mirror_top_bs->opaque; |
5a7e7a0b | 620 | AioContext *replace_aio_context = NULL; |
138f9fff | 621 | BlockDriverState *src = s->mirror_top_bs->backing->bs; |
e253f4b8 | 622 | BlockDriverState *target_bs = blk_bs(s->target); |
4ef85a9c | 623 | BlockDriverState *mirror_top_bs = s->mirror_top_bs; |
12fa4af6 | 624 | Error *local_err = NULL; |
3f09bfbc | 625 | |
2119882c PB |
626 | bdrv_release_dirty_bitmap(src, s->dirty_bitmap); |
627 | ||
3f09bfbc | 628 | /* Make sure that the source BDS doesn't go away before we called |
3d70ff53 | 629 | * job_completed(). */ |
3f09bfbc | 630 | bdrv_ref(src); |
4ef85a9c | 631 | bdrv_ref(mirror_top_bs); |
7d9fcb39 KW |
632 | bdrv_ref(target_bs); |
633 | ||
634 | /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before | |
635 | * inserting target_bs at s->to_replace, where we might not be able to get | |
63c8ef28 KW |
636 | * these permissions. |
637 | * | |
638 | * Note that blk_unref() alone doesn't necessarily drop permissions because | |
639 | * we might be running nested inside mirror_drain(), which takes an extra | |
640 | * reference, so use an explicit blk_set_perm() first. */ | |
641 | blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); | |
7d9fcb39 KW |
642 | blk_unref(s->target); |
643 | s->target = NULL; | |
4ef85a9c KW |
644 | |
645 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
646 | * required before it could become a backing file of target_bs. */ | |
647 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, | |
648 | &error_abort); | |
649 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { | |
650 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
651 | if (backing_bs(target_bs) != backing) { | |
12fa4af6 KW |
652 | bdrv_set_backing_hd(target_bs, backing, &local_err); |
653 | if (local_err) { | |
654 | error_report_err(local_err); | |
655 | data->ret = -EPERM; | |
656 | } | |
4ef85a9c KW |
657 | } |
658 | } | |
5a7e7a0b SH |
659 | |
660 | if (s->to_replace) { | |
661 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
662 | aio_context_acquire(replace_aio_context); | |
663 | } | |
664 | ||
665 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 666 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
667 | if (s->to_replace) { |
668 | to_replace = s->to_replace; | |
669 | } | |
40365552 | 670 | |
e253f4b8 KW |
671 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
672 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 673 | } |
b8804815 KW |
674 | |
675 | /* The mirror job has no requests in flight any more, but we need to | |
676 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 | 677 | bdrv_drained_begin(target_bs); |
5fe31c25 | 678 | bdrv_replace_node(to_replace, target_bs, &local_err); |
e253f4b8 | 679 | bdrv_drained_end(target_bs); |
5fe31c25 KW |
680 | if (local_err) { |
681 | error_report_err(local_err); | |
682 | data->ret = -EPERM; | |
683 | } | |
5a7e7a0b SH |
684 | } |
685 | if (s->to_replace) { | |
686 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
687 | error_free(s->replace_blocker); | |
688 | bdrv_unref(s->to_replace); | |
689 | } | |
690 | if (replace_aio_context) { | |
691 | aio_context_release(replace_aio_context); | |
692 | } | |
693 | g_free(s->replaces); | |
7d9fcb39 | 694 | bdrv_unref(target_bs); |
4ef85a9c KW |
695 | |
696 | /* Remove the mirror filter driver from the graph. Before this, get rid of | |
697 | * the blockers on the intermediate nodes so that the resulting state is | |
0bf74767 KW |
698 | * valid. Also give up permissions on mirror_top_bs->backing, which might |
699 | * block the removal. */ | |
1908a559 | 700 | block_job_remove_all_bdrv(bjob); |
c1cef672 FZ |
701 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
702 | &error_abort); | |
5fe31c25 | 703 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
4ef85a9c KW |
704 | |
705 | /* We just changed the BDS the job BB refers to (with either or both of the | |
5fe31c25 KW |
706 | * bdrv_replace_node() calls), so switch the BB back so the cleanup does |
707 | * the right thing. We don't need any permissions any more now. */ | |
1908a559 KW |
708 | blk_remove_bs(bjob->blk); |
709 | blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort); | |
710 | blk_insert_bs(bjob->blk, mirror_top_bs, &error_abort); | |
4ef85a9c | 711 | |
429076e8 | 712 | bs_opaque->job = NULL; |
1266c9b9 | 713 | job_completed(job, data->ret, NULL); |
4ef85a9c | 714 | |
5a7e7a0b | 715 | g_free(data); |
176c3699 | 716 | bdrv_drained_end(src); |
4ef85a9c | 717 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 718 | bdrv_unref(src); |
5a7e7a0b SH |
719 | } |
720 | ||
49efb1f5 DL |
721 | static void mirror_throttle(MirrorBlockJob *s) |
722 | { | |
723 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
724 | ||
18bb6928 | 725 | if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) { |
49efb1f5 | 726 | s->last_pause_ns = now; |
5d43e86e | 727 | job_sleep_ns(&s->common.job, 0); |
49efb1f5 | 728 | } else { |
da01ff7f | 729 | job_pause_point(&s->common.job); |
49efb1f5 DL |
730 | } |
731 | } | |
732 | ||
c0b363ad DL |
733 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
734 | { | |
23ca459a | 735 | int64_t offset; |
c0b363ad | 736 | BlockDriverState *base = s->base; |
138f9fff | 737 | BlockDriverState *bs = s->mirror_top_bs->backing->bs; |
c0b363ad | 738 | BlockDriverState *target_bs = blk_bs(s->target); |
23ca459a | 739 | int ret; |
51b0a488 | 740 | int64_t count; |
c0b363ad | 741 | |
b7d5062c | 742 | if (base == NULL && !bdrv_has_zero_init(target_bs)) { |
c7c2769c | 743 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
e0d7f73e | 744 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); |
c7c2769c DL |
745 | return 0; |
746 | } | |
747 | ||
90ab48eb | 748 | s->initial_zeroing_ongoing = true; |
23ca459a EB |
749 | for (offset = 0; offset < s->bdev_length; ) { |
750 | int bytes = MIN(s->bdev_length - offset, | |
751 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c7c2769c DL |
752 | |
753 | mirror_throttle(s); | |
754 | ||
daa7f2f9 | 755 | if (job_is_cancelled(&s->common.job)) { |
90ab48eb | 756 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
757 | return 0; |
758 | } | |
759 | ||
760 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
67adf4b3 EB |
761 | trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, |
762 | s->in_flight); | |
1181e19a | 763 | mirror_wait_for_free_in_flight_slot(s); |
c7c2769c DL |
764 | continue; |
765 | } | |
766 | ||
4295c5fc | 767 | mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO); |
23ca459a | 768 | offset += bytes; |
c7c2769c DL |
769 | } |
770 | ||
bae8196d | 771 | mirror_wait_for_all_io(s); |
90ab48eb | 772 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
773 | } |
774 | ||
c0b363ad | 775 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
23ca459a | 776 | for (offset = 0; offset < s->bdev_length; ) { |
c0b363ad | 777 | /* Just to make sure we are not exceeding int limit. */ |
23ca459a EB |
778 | int bytes = MIN(s->bdev_length - offset, |
779 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c0b363ad DL |
780 | |
781 | mirror_throttle(s); | |
782 | ||
daa7f2f9 | 783 | if (job_is_cancelled(&s->common.job)) { |
c0b363ad DL |
784 | return 0; |
785 | } | |
786 | ||
23ca459a | 787 | ret = bdrv_is_allocated_above(bs, base, offset, bytes, &count); |
c0b363ad DL |
788 | if (ret < 0) { |
789 | return ret; | |
790 | } | |
791 | ||
23ca459a | 792 | assert(count); |
b7d5062c | 793 | if (ret == 1) { |
23ca459a | 794 | bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); |
c0b363ad | 795 | } |
23ca459a | 796 | offset += count; |
c0b363ad DL |
797 | } |
798 | return 0; | |
799 | } | |
800 | ||
bdffb31d PB |
801 | /* Called when going out of the streaming phase to flush the bulk of the |
802 | * data to the medium, or just before completing. | |
803 | */ | |
804 | static int mirror_flush(MirrorBlockJob *s) | |
805 | { | |
806 | int ret = blk_flush(s->target); | |
807 | if (ret < 0) { | |
808 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
809 | s->ret = ret; | |
810 | } | |
811 | } | |
812 | return ret; | |
813 | } | |
814 | ||
893f7eba PB |
815 | static void coroutine_fn mirror_run(void *opaque) |
816 | { | |
817 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 818 | MirrorExitData *data; |
138f9fff | 819 | BlockDriverState *bs = s->mirror_top_bs->backing->bs; |
e253f4b8 | 820 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 821 | bool need_drain = true; |
c0b363ad | 822 | int64_t length; |
b812f671 | 823 | BlockDriverInfo bdi; |
1d33936e JC |
824 | char backing_filename[2]; /* we only need 2 characters because we are only |
825 | checking for a NULL string */ | |
893f7eba | 826 | int ret = 0; |
893f7eba | 827 | |
daa7f2f9 | 828 | if (job_is_cancelled(&s->common.job)) { |
893f7eba PB |
829 | goto immediate_exit; |
830 | } | |
831 | ||
b21c7652 HR |
832 | s->bdev_length = bdrv_getlength(bs); |
833 | if (s->bdev_length < 0) { | |
834 | ret = s->bdev_length; | |
373df5b1 | 835 | goto immediate_exit; |
becc347e KW |
836 | } |
837 | ||
838 | /* Active commit must resize the base image if its size differs from the | |
839 | * active layer. */ | |
840 | if (s->base == blk_bs(s->target)) { | |
841 | int64_t base_length; | |
842 | ||
843 | base_length = blk_getlength(s->target); | |
844 | if (base_length < 0) { | |
845 | ret = base_length; | |
846 | goto immediate_exit; | |
847 | } | |
848 | ||
849 | if (s->bdev_length > base_length) { | |
3a691c50 HR |
850 | ret = blk_truncate(s->target, s->bdev_length, PREALLOC_MODE_OFF, |
851 | NULL); | |
becc347e KW |
852 | if (ret < 0) { |
853 | goto immediate_exit; | |
854 | } | |
855 | } | |
856 | } | |
857 | ||
858 | if (s->bdev_length == 0) { | |
2e1795b5 KW |
859 | /* Transition to the READY state and wait for complete. */ |
860 | job_transition_to_ready(&s->common.job); | |
9e48b025 | 861 | s->synced = true; |
d06107ad | 862 | s->actively_synced = true; |
daa7f2f9 | 863 | while (!job_is_cancelled(&s->common.job) && !s->should_complete) { |
198c49cc | 864 | job_yield(&s->common.job); |
9e48b025 | 865 | } |
daa7f2f9 | 866 | s->common.job.cancelled = false; |
9e48b025 | 867 | goto immediate_exit; |
893f7eba PB |
868 | } |
869 | ||
b21c7652 | 870 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
871 | s->in_flight_bitmap = bitmap_new(length); |
872 | ||
b812f671 PB |
873 | /* If we have no backing file yet in the destination, we cannot let |
874 | * the destination do COW. Instead, we copy sectors around the | |
875 | * dirty data if needed. We need a bitmap to do that. | |
876 | */ | |
e253f4b8 | 877 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 878 | sizeof(backing_filename)); |
e253f4b8 | 879 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
b436982f EB |
880 | s->target_cluster_size = bdi.cluster_size; |
881 | } else { | |
882 | s->target_cluster_size = BDRV_SECTOR_SIZE; | |
e5b43573 | 883 | } |
b436982f EB |
884 | if (backing_filename[0] && !target_bs->backing && |
885 | s->granularity < s->target_cluster_size) { | |
886 | s->buf_size = MAX(s->buf_size, s->target_cluster_size); | |
e5b43573 | 887 | s->cow_bitmap = bitmap_new(length); |
b812f671 | 888 | } |
e253f4b8 | 889 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 890 | |
7504edf4 KW |
891 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
892 | if (s->buf == NULL) { | |
893 | ret = -ENOMEM; | |
894 | goto immediate_exit; | |
895 | } | |
896 | ||
402a4741 | 897 | mirror_free_init(s); |
893f7eba | 898 | |
49efb1f5 | 899 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 900 | if (!s->is_none_mode) { |
c0b363ad | 901 | ret = mirror_dirty_init(s); |
daa7f2f9 | 902 | if (ret < 0 || job_is_cancelled(&s->common.job)) { |
c0b363ad | 903 | goto immediate_exit; |
893f7eba PB |
904 | } |
905 | } | |
906 | ||
dc162c8e | 907 | assert(!s->dbi); |
715a74d8 | 908 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); |
893f7eba | 909 | for (;;) { |
cc8c9d6c | 910 | uint64_t delay_ns = 0; |
49efb1f5 | 911 | int64_t cnt, delta; |
893f7eba PB |
912 | bool should_complete; |
913 | ||
d06107ad HR |
914 | /* Do not start passive operations while there are active |
915 | * writes in progress */ | |
916 | while (s->in_active_write_counter) { | |
917 | mirror_wait_for_any_operation(s, true); | |
918 | } | |
919 | ||
bd48bde8 PB |
920 | if (s->ret < 0) { |
921 | ret = s->ret; | |
922 | goto immediate_exit; | |
923 | } | |
924 | ||
da01ff7f | 925 | job_pause_point(&s->common.job); |
565ac01f | 926 | |
20dca810 | 927 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
05df8a6a KW |
928 | /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is |
929 | * the number of bytes currently being processed; together those are | |
930 | * the current remaining operation length */ | |
30a5c887 | 931 | job_progress_set_remaining(&s->common.job, s->bytes_in_flight + cnt); |
bd48bde8 PB |
932 | |
933 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 934 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
18bb6928 KW |
935 | * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is |
936 | * an error, or when the source is clean, whichever comes first. */ | |
49efb1f5 | 937 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
18bb6928 | 938 | if (delta < BLOCK_JOB_SLICE_TIME && |
bd48bde8 | 939 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 940 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 | 941 | (cnt == 0 && s->in_flight > 0)) { |
9a46dba7 | 942 | trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); |
1181e19a | 943 | mirror_wait_for_free_in_flight_slot(s); |
bd48bde8 PB |
944 | continue; |
945 | } else if (cnt != 0) { | |
cc8c9d6c | 946 | delay_ns = mirror_iteration(s); |
893f7eba | 947 | } |
893f7eba PB |
948 | } |
949 | ||
950 | should_complete = false; | |
bd48bde8 | 951 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 952 | trace_mirror_before_flush(s); |
bdffb31d PB |
953 | if (!s->synced) { |
954 | if (mirror_flush(s) < 0) { | |
955 | /* Go check s->ret. */ | |
956 | continue; | |
b952b558 | 957 | } |
b952b558 PB |
958 | /* We're out of the streaming phase. From now on, if the job |
959 | * is cancelled we will actually complete all pending I/O and | |
960 | * report completion. This way, block-job-cancel will leave | |
961 | * the target in a consistent state. | |
962 | */ | |
2e1795b5 | 963 | job_transition_to_ready(&s->common.job); |
bdffb31d | 964 | s->synced = true; |
d06107ad HR |
965 | if (s->copy_mode != MIRROR_COPY_MODE_BACKGROUND) { |
966 | s->actively_synced = true; | |
967 | } | |
d63ffd87 | 968 | } |
bdffb31d PB |
969 | |
970 | should_complete = s->should_complete || | |
daa7f2f9 | 971 | job_is_cancelled(&s->common.job); |
bdffb31d | 972 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
893f7eba PB |
973 | } |
974 | ||
975 | if (cnt == 0 && should_complete) { | |
976 | /* The dirty bitmap is not updated while operations are pending. | |
977 | * If we're about to exit, wait for pending operations before | |
978 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
979 | * source has dirty data to copy! | |
980 | * | |
981 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
982 | * mirror_populate runs, so pause it now. Before deciding |
983 | * whether to switch to target check one last time if I/O has | |
984 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba | 985 | */ |
9a46dba7 | 986 | trace_mirror_before_drain(s, cnt); |
9a0cec66 PB |
987 | |
988 | bdrv_drained_begin(bs); | |
20dca810 | 989 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 990 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 PB |
991 | bdrv_drained_end(bs); |
992 | continue; | |
993 | } | |
994 | ||
995 | /* The two disks are in sync. Exit and report successful | |
996 | * completion. | |
997 | */ | |
998 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
daa7f2f9 | 999 | s->common.job.cancelled = false; |
9a0cec66 PB |
1000 | need_drain = false; |
1001 | break; | |
893f7eba PB |
1002 | } |
1003 | ||
1004 | ret = 0; | |
ddc4115e SH |
1005 | |
1006 | if (s->synced && !should_complete) { | |
18bb6928 KW |
1007 | delay_ns = (s->in_flight == 0 && |
1008 | cnt == 0 ? BLOCK_JOB_SLICE_TIME : 0); | |
ddc4115e | 1009 | } |
9a46dba7 | 1010 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
5d43e86e | 1011 | job_sleep_ns(&s->common.job, delay_ns); |
daa7f2f9 | 1012 | if (job_is_cancelled(&s->common.job) && |
004e95df | 1013 | (!s->synced || s->common.job.force_cancel)) |
eb36639f | 1014 | { |
b76e4458 | 1015 | break; |
893f7eba | 1016 | } |
49efb1f5 | 1017 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
1018 | } |
1019 | ||
1020 | immediate_exit: | |
bd48bde8 PB |
1021 | if (s->in_flight > 0) { |
1022 | /* We get here only if something went wrong. Either the job failed, | |
1023 | * or it was cancelled prematurely so that we do not guarantee that | |
1024 | * the target is a copy of the source. | |
1025 | */ | |
004e95df | 1026 | assert(ret < 0 || ((s->common.job.force_cancel || !s->synced) && |
daa7f2f9 | 1027 | job_is_cancelled(&s->common.job))); |
9a0cec66 | 1028 | assert(need_drain); |
bae8196d | 1029 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
1030 | } |
1031 | ||
1032 | assert(s->in_flight == 0); | |
7191bf31 | 1033 | qemu_vfree(s->buf); |
b812f671 | 1034 | g_free(s->cow_bitmap); |
402a4741 | 1035 | g_free(s->in_flight_bitmap); |
dc162c8e | 1036 | bdrv_dirty_iter_free(s->dbi); |
5a7e7a0b SH |
1037 | |
1038 | data = g_malloc(sizeof(*data)); | |
1039 | data->ret = ret; | |
9a0cec66 PB |
1040 | |
1041 | if (need_drain) { | |
1042 | bdrv_drained_begin(bs); | |
1043 | } | |
1908a559 | 1044 | job_defer_to_main_loop(&s->common.job, mirror_exit, data); |
893f7eba PB |
1045 | } |
1046 | ||
3453d972 | 1047 | static void mirror_complete(Job *job, Error **errp) |
d63ffd87 | 1048 | { |
3453d972 | 1049 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
4ef85a9c | 1050 | BlockDriverState *target; |
274fccee | 1051 | |
274fccee | 1052 | target = blk_bs(s->target); |
d63ffd87 | 1053 | |
d63ffd87 | 1054 | if (!s->synced) { |
9df229c3 | 1055 | error_setg(errp, "The active block job '%s' cannot be completed", |
3453d972 | 1056 | job->id); |
d63ffd87 PB |
1057 | return; |
1058 | } | |
1059 | ||
274fccee HR |
1060 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
1061 | int ret; | |
1062 | ||
1063 | assert(!target->backing); | |
1064 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
1065 | if (ret < 0) { | |
1066 | return; | |
1067 | } | |
1068 | } | |
1069 | ||
15d67298 | 1070 | /* block all operations on to_replace bs */ |
09158f00 | 1071 | if (s->replaces) { |
5a7e7a0b SH |
1072 | AioContext *replace_aio_context; |
1073 | ||
e12f3784 | 1074 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 1075 | if (!s->to_replace) { |
e12f3784 | 1076 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
1077 | return; |
1078 | } | |
1079 | ||
5a7e7a0b SH |
1080 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
1081 | aio_context_acquire(replace_aio_context); | |
1082 | ||
4ef85a9c KW |
1083 | /* TODO Translate this into permission system. Current definition of |
1084 | * GRAPH_MOD would require to request it for the parents; they might | |
1085 | * not even be BlockDriverStates, however, so a BdrvChild can't address | |
1086 | * them. May need redefinition of GRAPH_MOD. */ | |
09158f00 BC |
1087 | error_setg(&s->replace_blocker, |
1088 | "block device is in use by block-job-complete"); | |
1089 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
1090 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
1091 | |
1092 | aio_context_release(replace_aio_context); | |
09158f00 BC |
1093 | } |
1094 | ||
d63ffd87 | 1095 | s->should_complete = true; |
3d70ff53 | 1096 | job_enter(job); |
d63ffd87 PB |
1097 | } |
1098 | ||
da01ff7f | 1099 | static void mirror_pause(Job *job) |
565ac01f | 1100 | { |
da01ff7f | 1101 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); |
565ac01f | 1102 | |
bae8196d | 1103 | mirror_wait_for_all_io(s); |
565ac01f SH |
1104 | } |
1105 | ||
89bd0305 KW |
1106 | static bool mirror_drained_poll(BlockJob *job) |
1107 | { | |
1108 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
1109 | return !!s->in_flight; | |
1110 | } | |
1111 | ||
565ac01f SH |
1112 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) |
1113 | { | |
1114 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
1115 | ||
1116 | blk_set_aio_context(s->target, new_context); | |
1117 | } | |
1118 | ||
bae8196d PB |
1119 | static void mirror_drain(BlockJob *job) |
1120 | { | |
1121 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
1122 | ||
1123 | /* Need to keep a reference in case blk_drain triggers execution | |
1124 | * of mirror_complete... | |
1125 | */ | |
1126 | if (s->target) { | |
1127 | BlockBackend *target = s->target; | |
1128 | blk_ref(target); | |
1129 | blk_drain(target); | |
1130 | blk_unref(target); | |
1131 | } | |
1132 | } | |
1133 | ||
3fc4b10a | 1134 | static const BlockJobDriver mirror_job_driver = { |
33e9e9bd KW |
1135 | .job_driver = { |
1136 | .instance_size = sizeof(MirrorBlockJob), | |
252291ea | 1137 | .job_type = JOB_TYPE_MIRROR, |
80fa2c75 | 1138 | .free = block_job_free, |
b15de828 | 1139 | .user_resume = block_job_user_resume, |
b69f777d | 1140 | .drain = block_job_drain, |
da01ff7f KW |
1141 | .start = mirror_run, |
1142 | .pause = mirror_pause, | |
3453d972 | 1143 | .complete = mirror_complete, |
33e9e9bd | 1144 | }, |
89bd0305 | 1145 | .drained_poll = mirror_drained_poll, |
565ac01f | 1146 | .attached_aio_context = mirror_attached_aio_context, |
bae8196d | 1147 | .drain = mirror_drain, |
893f7eba PB |
1148 | }; |
1149 | ||
03544a6e | 1150 | static const BlockJobDriver commit_active_job_driver = { |
33e9e9bd KW |
1151 | .job_driver = { |
1152 | .instance_size = sizeof(MirrorBlockJob), | |
252291ea | 1153 | .job_type = JOB_TYPE_COMMIT, |
80fa2c75 | 1154 | .free = block_job_free, |
b15de828 | 1155 | .user_resume = block_job_user_resume, |
b69f777d | 1156 | .drain = block_job_drain, |
da01ff7f KW |
1157 | .start = mirror_run, |
1158 | .pause = mirror_pause, | |
3453d972 | 1159 | .complete = mirror_complete, |
33e9e9bd | 1160 | }, |
89bd0305 | 1161 | .drained_poll = mirror_drained_poll, |
565ac01f | 1162 | .attached_aio_context = mirror_attached_aio_context, |
bae8196d | 1163 | .drain = mirror_drain, |
03544a6e FZ |
1164 | }; |
1165 | ||
d06107ad HR |
1166 | static void do_sync_target_write(MirrorBlockJob *job, MirrorMethod method, |
1167 | uint64_t offset, uint64_t bytes, | |
1168 | QEMUIOVector *qiov, int flags) | |
1169 | { | |
1170 | BdrvDirtyBitmapIter *iter; | |
1171 | QEMUIOVector target_qiov; | |
1172 | uint64_t dirty_offset; | |
1173 | int dirty_bytes; | |
1174 | ||
1175 | if (qiov) { | |
1176 | qemu_iovec_init(&target_qiov, qiov->niov); | |
1177 | } | |
1178 | ||
1179 | iter = bdrv_dirty_iter_new(job->dirty_bitmap); | |
1180 | bdrv_set_dirty_iter(iter, offset); | |
1181 | ||
1182 | while (true) { | |
1183 | bool valid_area; | |
1184 | int ret; | |
1185 | ||
1186 | bdrv_dirty_bitmap_lock(job->dirty_bitmap); | |
1187 | valid_area = bdrv_dirty_iter_next_area(iter, offset + bytes, | |
1188 | &dirty_offset, &dirty_bytes); | |
1189 | if (!valid_area) { | |
1190 | bdrv_dirty_bitmap_unlock(job->dirty_bitmap); | |
1191 | break; | |
1192 | } | |
1193 | ||
1194 | bdrv_reset_dirty_bitmap_locked(job->dirty_bitmap, | |
1195 | dirty_offset, dirty_bytes); | |
1196 | bdrv_dirty_bitmap_unlock(job->dirty_bitmap); | |
1197 | ||
1198 | job_progress_increase_remaining(&job->common.job, dirty_bytes); | |
1199 | ||
1200 | assert(dirty_offset - offset <= SIZE_MAX); | |
1201 | if (qiov) { | |
1202 | qemu_iovec_reset(&target_qiov); | |
1203 | qemu_iovec_concat(&target_qiov, qiov, | |
1204 | dirty_offset - offset, dirty_bytes); | |
1205 | } | |
1206 | ||
1207 | switch (method) { | |
1208 | case MIRROR_METHOD_COPY: | |
1209 | ret = blk_co_pwritev(job->target, dirty_offset, dirty_bytes, | |
1210 | qiov ? &target_qiov : NULL, flags); | |
1211 | break; | |
1212 | ||
1213 | case MIRROR_METHOD_ZERO: | |
1214 | assert(!qiov); | |
1215 | ret = blk_co_pwrite_zeroes(job->target, dirty_offset, dirty_bytes, | |
1216 | flags); | |
1217 | break; | |
1218 | ||
1219 | case MIRROR_METHOD_DISCARD: | |
1220 | assert(!qiov); | |
1221 | ret = blk_co_pdiscard(job->target, dirty_offset, dirty_bytes); | |
1222 | break; | |
1223 | ||
1224 | default: | |
1225 | abort(); | |
1226 | } | |
1227 | ||
1228 | if (ret >= 0) { | |
1229 | job_progress_update(&job->common.job, dirty_bytes); | |
1230 | } else { | |
1231 | BlockErrorAction action; | |
1232 | ||
1233 | bdrv_set_dirty_bitmap(job->dirty_bitmap, dirty_offset, dirty_bytes); | |
1234 | job->actively_synced = false; | |
1235 | ||
1236 | action = mirror_error_action(job, false, -ret); | |
1237 | if (action == BLOCK_ERROR_ACTION_REPORT) { | |
1238 | if (!job->ret) { | |
1239 | job->ret = ret; | |
1240 | } | |
1241 | break; | |
1242 | } | |
1243 | } | |
1244 | } | |
1245 | ||
1246 | bdrv_dirty_iter_free(iter); | |
1247 | if (qiov) { | |
1248 | qemu_iovec_destroy(&target_qiov); | |
1249 | } | |
1250 | } | |
1251 | ||
1252 | static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s, | |
1253 | uint64_t offset, | |
1254 | uint64_t bytes) | |
1255 | { | |
1256 | MirrorOp *op; | |
1257 | uint64_t start_chunk = offset / s->granularity; | |
1258 | uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); | |
1259 | ||
1260 | op = g_new(MirrorOp, 1); | |
1261 | *op = (MirrorOp){ | |
1262 | .s = s, | |
1263 | .offset = offset, | |
1264 | .bytes = bytes, | |
1265 | .is_active_write = true, | |
1266 | }; | |
1267 | qemu_co_queue_init(&op->waiting_requests); | |
1268 | QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); | |
1269 | ||
1270 | s->in_active_write_counter++; | |
1271 | ||
1272 | mirror_wait_on_conflicts(op, s, offset, bytes); | |
1273 | ||
1274 | bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); | |
1275 | ||
1276 | return op; | |
1277 | } | |
1278 | ||
1279 | static void coroutine_fn active_write_settle(MirrorOp *op) | |
1280 | { | |
1281 | uint64_t start_chunk = op->offset / op->s->granularity; | |
1282 | uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes, | |
1283 | op->s->granularity); | |
1284 | ||
1285 | if (!--op->s->in_active_write_counter && op->s->actively_synced) { | |
1286 | BdrvChild *source = op->s->mirror_top_bs->backing; | |
1287 | ||
1288 | if (QLIST_FIRST(&source->bs->parents) == source && | |
1289 | QLIST_NEXT(source, next_parent) == NULL) | |
1290 | { | |
1291 | /* Assert that we are back in sync once all active write | |
1292 | * operations are settled. | |
1293 | * Note that we can only assert this if the mirror node | |
1294 | * is the source node's only parent. */ | |
1295 | assert(!bdrv_get_dirty_count(op->s->dirty_bitmap)); | |
1296 | } | |
1297 | } | |
1298 | bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); | |
1299 | QTAILQ_REMOVE(&op->s->ops_in_flight, op, next); | |
1300 | qemu_co_queue_restart_all(&op->waiting_requests); | |
1301 | g_free(op); | |
1302 | } | |
1303 | ||
4ef85a9c KW |
1304 | static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, |
1305 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1306 | { | |
1307 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1308 | } | |
1309 | ||
d06107ad HR |
1310 | static int coroutine_fn bdrv_mirror_top_do_write(BlockDriverState *bs, |
1311 | MirrorMethod method, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, | |
1312 | int flags) | |
1313 | { | |
1314 | MirrorOp *op = NULL; | |
1315 | MirrorBDSOpaque *s = bs->opaque; | |
1316 | int ret = 0; | |
1317 | bool copy_to_target; | |
1318 | ||
1319 | copy_to_target = s->job->ret >= 0 && | |
1320 | s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; | |
1321 | ||
1322 | if (copy_to_target) { | |
1323 | op = active_write_prepare(s->job, offset, bytes); | |
1324 | } | |
1325 | ||
1326 | switch (method) { | |
1327 | case MIRROR_METHOD_COPY: | |
1328 | ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1329 | break; | |
1330 | ||
1331 | case MIRROR_METHOD_ZERO: | |
1332 | ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); | |
1333 | break; | |
1334 | ||
1335 | case MIRROR_METHOD_DISCARD: | |
0b9fd3f4 | 1336 | ret = bdrv_co_pdiscard(bs->backing, offset, bytes); |
d06107ad HR |
1337 | break; |
1338 | ||
1339 | default: | |
1340 | abort(); | |
1341 | } | |
1342 | ||
1343 | if (ret < 0) { | |
1344 | goto out; | |
1345 | } | |
1346 | ||
1347 | if (copy_to_target) { | |
1348 | do_sync_target_write(s->job, method, offset, bytes, qiov, flags); | |
1349 | } | |
1350 | ||
1351 | out: | |
1352 | if (copy_to_target) { | |
1353 | active_write_settle(op); | |
1354 | } | |
1355 | return ret; | |
1356 | } | |
1357 | ||
4ef85a9c KW |
1358 | static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, |
1359 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1360 | { | |
d06107ad HR |
1361 | MirrorBDSOpaque *s = bs->opaque; |
1362 | QEMUIOVector bounce_qiov; | |
1363 | void *bounce_buf; | |
1364 | int ret = 0; | |
1365 | bool copy_to_target; | |
1366 | ||
1367 | copy_to_target = s->job->ret >= 0 && | |
1368 | s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING; | |
1369 | ||
1370 | if (copy_to_target) { | |
1371 | /* The guest might concurrently modify the data to write; but | |
1372 | * the data on source and destination must match, so we have | |
1373 | * to use a bounce buffer if we are going to write to the | |
1374 | * target now. */ | |
1375 | bounce_buf = qemu_blockalign(bs, bytes); | |
1376 | iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes); | |
1377 | ||
1378 | qemu_iovec_init(&bounce_qiov, 1); | |
1379 | qemu_iovec_add(&bounce_qiov, bounce_buf, bytes); | |
1380 | qiov = &bounce_qiov; | |
1381 | } | |
1382 | ||
1383 | ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, offset, bytes, qiov, | |
1384 | flags); | |
1385 | ||
1386 | if (copy_to_target) { | |
1387 | qemu_iovec_destroy(&bounce_qiov); | |
1388 | qemu_vfree(bounce_buf); | |
1389 | } | |
1390 | ||
1391 | return ret; | |
4ef85a9c KW |
1392 | } |
1393 | ||
1394 | static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) | |
1395 | { | |
ce960aa9 VSO |
1396 | if (bs->backing == NULL) { |
1397 | /* we can be here after failed bdrv_append in mirror_start_job */ | |
1398 | return 0; | |
1399 | } | |
4ef85a9c KW |
1400 | return bdrv_co_flush(bs->backing->bs); |
1401 | } | |
1402 | ||
4ef85a9c | 1403 | static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 1404 | int64_t offset, int bytes, BdrvRequestFlags flags) |
4ef85a9c | 1405 | { |
d06107ad HR |
1406 | return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, offset, bytes, NULL, |
1407 | flags); | |
4ef85a9c KW |
1408 | } |
1409 | ||
1410 | static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, | |
f5a5ca79 | 1411 | int64_t offset, int bytes) |
4ef85a9c | 1412 | { |
d06107ad HR |
1413 | return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, offset, bytes, |
1414 | NULL, 0); | |
4ef85a9c KW |
1415 | } |
1416 | ||
fd4a6493 KW |
1417 | static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) |
1418 | { | |
18775ff3 VSO |
1419 | if (bs->backing == NULL) { |
1420 | /* we can be here after failed bdrv_attach_child in | |
1421 | * bdrv_set_backing_hd */ | |
1422 | return; | |
1423 | } | |
fd4a6493 KW |
1424 | bdrv_refresh_filename(bs->backing->bs); |
1425 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
1426 | bs->backing->bs->filename); | |
1427 | } | |
1428 | ||
4ef85a9c KW |
1429 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, |
1430 | const BdrvChildRole *role, | |
e0995dc3 | 1431 | BlockReopenQueue *reopen_queue, |
4ef85a9c KW |
1432 | uint64_t perm, uint64_t shared, |
1433 | uint64_t *nperm, uint64_t *nshared) | |
1434 | { | |
1435 | /* Must be able to forward guest writes to the real image */ | |
1436 | *nperm = 0; | |
1437 | if (perm & BLK_PERM_WRITE) { | |
1438 | *nperm |= BLK_PERM_WRITE; | |
1439 | } | |
1440 | ||
1441 | *nshared = BLK_PERM_ALL; | |
1442 | } | |
1443 | ||
1444 | /* Dummy node that provides consistent read to its users without requiring it | |
1445 | * from its backing file and that allows writes on the backing file chain. */ | |
1446 | static BlockDriver bdrv_mirror_top = { | |
1447 | .format_name = "mirror_top", | |
1448 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1449 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1450 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1451 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1452 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
3e4d0e72 | 1453 | .bdrv_co_block_status = bdrv_co_block_status_from_backing, |
fd4a6493 | 1454 | .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, |
4ef85a9c KW |
1455 | .bdrv_child_perm = bdrv_mirror_top_child_perm, |
1456 | }; | |
1457 | ||
71aa9867 | 1458 | static void mirror_start_job(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1459 | int creation_flags, BlockDriverState *target, |
1460 | const char *replaces, int64_t speed, | |
1461 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1462 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
1463 | BlockdevOnError on_source_error, |
1464 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1465 | bool unmap, |
097310b5 | 1466 | BlockCompletionFunc *cb, |
51ccfa2d | 1467 | void *opaque, |
09158f00 | 1468 | const BlockJobDriver *driver, |
b49f7ead | 1469 | bool is_none_mode, BlockDriverState *base, |
51ccfa2d | 1470 | bool auto_complete, const char *filter_node_name, |
481debaa | 1471 | bool is_mirror, MirrorCopyMode copy_mode, |
51ccfa2d | 1472 | Error **errp) |
893f7eba PB |
1473 | { |
1474 | MirrorBlockJob *s; | |
429076e8 | 1475 | MirrorBDSOpaque *bs_opaque; |
4ef85a9c KW |
1476 | BlockDriverState *mirror_top_bs; |
1477 | bool target_graph_mod; | |
1478 | bool target_is_backing; | |
b2c2832c | 1479 | Error *local_err = NULL; |
d7086422 | 1480 | int ret; |
893f7eba | 1481 | |
eee13dfe | 1482 | if (granularity == 0) { |
341ebc2f | 1483 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1484 | } |
1485 | ||
31826642 | 1486 | assert(is_power_of_2(granularity)); |
eee13dfe | 1487 | |
48ac0a4d WC |
1488 | if (buf_size < 0) { |
1489 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
1490 | return; | |
1491 | } | |
1492 | ||
1493 | if (buf_size == 0) { | |
1494 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
1495 | } | |
5bc361b8 | 1496 | |
86fae10c KW |
1497 | if (bs == target) { |
1498 | error_setg(errp, "Can't mirror node into itself"); | |
1499 | return; | |
1500 | } | |
1501 | ||
4ef85a9c KW |
1502 | /* In the case of active commit, add dummy driver to provide consistent |
1503 | * reads on the top, while disabling it in the intermediate nodes, and make | |
1504 | * the backing chain writable. */ | |
6cdbceb1 KW |
1505 | mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, |
1506 | BDRV_O_RDWR, errp); | |
4ef85a9c KW |
1507 | if (mirror_top_bs == NULL) { |
1508 | return; | |
1509 | } | |
d3c8c674 KW |
1510 | if (!filter_node_name) { |
1511 | mirror_top_bs->implicit = true; | |
1512 | } | |
4ef85a9c | 1513 | mirror_top_bs->total_sectors = bs->total_sectors; |
228345bf HR |
1514 | mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED; |
1515 | mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED; | |
429076e8 HR |
1516 | bs_opaque = g_new0(MirrorBDSOpaque, 1); |
1517 | mirror_top_bs->opaque = bs_opaque; | |
19dd29e8 | 1518 | bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); |
4ef85a9c KW |
1519 | |
1520 | /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep | |
7a25fcd0 | 1521 | * it alive until block_job_create() succeeds even if bs has no parent. */ |
4ef85a9c KW |
1522 | bdrv_ref(mirror_top_bs); |
1523 | bdrv_drained_begin(bs); | |
b2c2832c | 1524 | bdrv_append(mirror_top_bs, bs, &local_err); |
4ef85a9c KW |
1525 | bdrv_drained_end(bs); |
1526 | ||
b2c2832c KW |
1527 | if (local_err) { |
1528 | bdrv_unref(mirror_top_bs); | |
1529 | error_propagate(errp, local_err); | |
1530 | return; | |
1531 | } | |
1532 | ||
4ef85a9c | 1533 | /* Make sure that the source is not resized while the job is running */ |
75859b94 | 1534 | s = block_job_create(job_id, driver, NULL, mirror_top_bs, |
4ef85a9c KW |
1535 | BLK_PERM_CONSISTENT_READ, |
1536 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
1537 | BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, | |
c6cc12bf | 1538 | creation_flags, cb, opaque, errp); |
893f7eba | 1539 | if (!s) { |
4ef85a9c | 1540 | goto fail; |
893f7eba | 1541 | } |
429076e8 HR |
1542 | bs_opaque->job = s; |
1543 | ||
7a25fcd0 HR |
1544 | /* The block job now has a reference to this node */ |
1545 | bdrv_unref(mirror_top_bs); | |
1546 | ||
4ef85a9c KW |
1547 | s->mirror_top_bs = mirror_top_bs; |
1548 | ||
1549 | /* No resize for the target either; while the mirror is still running, a | |
1550 | * consistent read isn't necessarily possible. We could possibly allow | |
1551 | * writes and graph modifications, though it would likely defeat the | |
1552 | * purpose of a mirror, so leave them blocked for now. | |
1553 | * | |
1554 | * In the case of active commit, things look a bit different, though, | |
1555 | * because the target is an already populated backing file in active use. | |
1556 | * We can allow anything except resize there.*/ | |
1557 | target_is_backing = bdrv_chain_contains(bs, target); | |
1558 | target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); | |
1559 | s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | | |
1560 | (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), | |
1561 | BLK_PERM_WRITE_UNCHANGED | | |
1562 | (target_is_backing ? BLK_PERM_CONSISTENT_READ | | |
1563 | BLK_PERM_WRITE | | |
1564 | BLK_PERM_GRAPH_MOD : 0)); | |
d7086422 KW |
1565 | ret = blk_insert_bs(s->target, target, errp); |
1566 | if (ret < 0) { | |
4ef85a9c | 1567 | goto fail; |
d7086422 | 1568 | } |
045a2f82 FZ |
1569 | if (is_mirror) { |
1570 | /* XXX: Mirror target could be a NBD server of target QEMU in the case | |
1571 | * of non-shared block migration. To allow migration completion, we | |
1572 | * have to allow "inactivate" of the target BB. When that happens, we | |
1573 | * know the job is drained, and the vcpus are stopped, so no write | |
1574 | * operation will be performed. Block layer already has assertions to | |
1575 | * ensure that. */ | |
1576 | blk_set_force_allow_inactivate(s->target); | |
1577 | } | |
e253f4b8 | 1578 | |
09158f00 | 1579 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1580 | s->on_source_error = on_source_error; |
1581 | s->on_target_error = on_target_error; | |
03544a6e | 1582 | s->is_none_mode = is_none_mode; |
274fccee | 1583 | s->backing_mode = backing_mode; |
481debaa | 1584 | s->copy_mode = copy_mode; |
5bc361b8 | 1585 | s->base = base; |
eee13dfe | 1586 | s->granularity = granularity; |
48ac0a4d | 1587 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1588 | s->unmap = unmap; |
b49f7ead WC |
1589 | if (auto_complete) { |
1590 | s->should_complete = true; | |
1591 | } | |
b812f671 | 1592 | |
0db6e54a | 1593 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1594 | if (!s->dirty_bitmap) { |
88f9d1b3 | 1595 | goto fail; |
b8afb520 | 1596 | } |
10f3cd15 | 1597 | |
4ef85a9c | 1598 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1599 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1600 | &error_abort); | |
1601 | ||
f3ede4b0 AG |
1602 | /* In commit_active_start() all intermediate nodes disappear, so |
1603 | * any jobs in them must be blocked */ | |
4ef85a9c | 1604 | if (target_is_backing) { |
f3ede4b0 AG |
1605 | BlockDriverState *iter; |
1606 | for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { | |
4ef85a9c KW |
1607 | /* XXX BLK_PERM_WRITE needs to be allowed so we don't block |
1608 | * ourselves at s->base (if writes are blocked for a node, they are | |
1609 | * also blocked for its backing file). The other options would be a | |
1610 | * second filter driver above s->base (== target). */ | |
1611 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, | |
1612 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, | |
1613 | errp); | |
1614 | if (ret < 0) { | |
1615 | goto fail; | |
1616 | } | |
f3ede4b0 AG |
1617 | } |
1618 | } | |
10f3cd15 | 1619 | |
12aa4082 HR |
1620 | QTAILQ_INIT(&s->ops_in_flight); |
1621 | ||
5ccac6f1 | 1622 | trace_mirror_start(bs, s, opaque); |
da01ff7f | 1623 | job_start(&s->common.job); |
4ef85a9c KW |
1624 | return; |
1625 | ||
1626 | fail: | |
1627 | if (s) { | |
7a25fcd0 HR |
1628 | /* Make sure this BDS does not go away until we have completed the graph |
1629 | * changes below */ | |
1630 | bdrv_ref(mirror_top_bs); | |
1631 | ||
4ef85a9c KW |
1632 | g_free(s->replaces); |
1633 | blk_unref(s->target); | |
429076e8 | 1634 | bs_opaque->job = NULL; |
4ad35181 | 1635 | job_early_fail(&s->common.job); |
4ef85a9c KW |
1636 | } |
1637 | ||
c1cef672 FZ |
1638 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
1639 | &error_abort); | |
5fe31c25 | 1640 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
7a25fcd0 HR |
1641 | |
1642 | bdrv_unref(mirror_top_bs); | |
893f7eba | 1643 | } |
03544a6e | 1644 | |
71aa9867 AG |
1645 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1646 | BlockDriverState *target, const char *replaces, | |
5fba6c0e | 1647 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
1648 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
1649 | BlockdevOnError on_source_error, | |
03544a6e | 1650 | BlockdevOnError on_target_error, |
481debaa HR |
1651 | bool unmap, const char *filter_node_name, |
1652 | MirrorCopyMode copy_mode, Error **errp) | |
03544a6e FZ |
1653 | { |
1654 | bool is_none_mode; | |
1655 | BlockDriverState *base; | |
1656 | ||
4b80ab2b JS |
1657 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
1658 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
1659 | return; |
1660 | } | |
03544a6e | 1661 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 1662 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
bb02b65c | 1663 | mirror_start_job(job_id, bs, JOB_DEFAULT, target, replaces, |
274fccee | 1664 | speed, granularity, buf_size, backing_mode, |
51ccfa2d | 1665 | on_source_error, on_target_error, unmap, NULL, NULL, |
6cdbceb1 | 1666 | &mirror_job_driver, is_none_mode, base, false, |
481debaa | 1667 | filter_node_name, true, copy_mode, errp); |
03544a6e FZ |
1668 | } |
1669 | ||
fd62c609 | 1670 | void commit_active_start(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1671 | BlockDriverState *base, int creation_flags, |
1672 | int64_t speed, BlockdevOnError on_error, | |
0db832f4 | 1673 | const char *filter_node_name, |
78bbd910 FZ |
1674 | BlockCompletionFunc *cb, void *opaque, |
1675 | bool auto_complete, Error **errp) | |
03544a6e | 1676 | { |
4da83585 | 1677 | int orig_base_flags; |
cc67f4d1 | 1678 | Error *local_err = NULL; |
4da83585 JC |
1679 | |
1680 | orig_base_flags = bdrv_get_flags(base); | |
1681 | ||
20a63d2c FZ |
1682 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
1683 | return; | |
1684 | } | |
4da83585 | 1685 | |
47970dfb | 1686 | mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
71aa9867 | 1687 | MIRROR_LEAVE_BACKING_CHAIN, |
51ccfa2d | 1688 | on_error, on_error, true, cb, opaque, |
6cdbceb1 | 1689 | &commit_active_job_driver, false, base, auto_complete, |
481debaa HR |
1690 | filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND, |
1691 | &local_err); | |
0fb6395c | 1692 | if (local_err) { |
cc67f4d1 | 1693 | error_propagate(errp, local_err); |
4da83585 JC |
1694 | goto error_restore_flags; |
1695 | } | |
1696 | ||
1697 | return; | |
1698 | ||
1699 | error_restore_flags: | |
1700 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1701 | * the original error */ | |
1702 | bdrv_reopen(base, orig_base_flags, NULL); | |
1703 | return; | |
03544a6e | 1704 | } |