]> git.proxmox.com Git - mirror_qemu.git/blame - block/mirror.c
block/snapshot: stress that we fallback to primary child
[mirror_qemu.git] / block / mirror.c
CommitLineData
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"
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 */
35typedef struct MirrorBuffer {
36 QSIMPLEQ_ENTRY(MirrorBuffer) next;
37} MirrorBuffer;
893f7eba 38
12aa4082
HR
39typedef struct MirrorOp MirrorOp;
40
893f7eba
PB
41typedef 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;
d06107ad 58 MirrorCopyMode copy_mode;
b952b558 59 BlockdevOnError on_source_error, on_target_error;
d06107ad
HR
60 /* Set when the target is synced (dirty bitmap is clean, nothing
61 * in flight) and the job is running in active mode */
62 bool actively_synced;
d63ffd87 63 bool should_complete;
eee13dfe 64 int64_t granularity;
b812f671 65 size_t buf_size;
b21c7652 66 int64_t bdev_length;
b812f671 67 unsigned long *cow_bitmap;
e4654d2d 68 BdrvDirtyBitmap *dirty_bitmap;
dc162c8e 69 BdrvDirtyBitmapIter *dbi;
893f7eba 70 uint8_t *buf;
402a4741
PB
71 QSIMPLEQ_HEAD(, MirrorBuffer) buf_free;
72 int buf_free_count;
bd48bde8 73
49efb1f5 74 uint64_t last_pause_ns;
402a4741 75 unsigned long *in_flight_bitmap;
1b8f7776 76 unsigned in_flight;
b436982f 77 int64_t bytes_in_flight;
b58deb34 78 QTAILQ_HEAD(, MirrorOp) ops_in_flight;
bd48bde8 79 int ret;
0fc9f8ea 80 bool unmap;
b436982f 81 int target_cluster_size;
e5b43573 82 int max_iov;
90ab48eb 83 bool initial_zeroing_ongoing;
d06107ad 84 int in_active_write_counter;
737efc1e 85 bool prepared;
5e771752 86 bool in_drain;
893f7eba
PB
87} MirrorBlockJob;
88
429076e8
HR
89typedef struct MirrorBDSOpaque {
90 MirrorBlockJob *job;
f94dc3b4 91 bool stop;
53431b90 92 bool is_commit;
429076e8
HR
93} MirrorBDSOpaque;
94
12aa4082 95struct MirrorOp {
bd48bde8
PB
96 MirrorBlockJob *s;
97 QEMUIOVector qiov;
b436982f
EB
98 int64_t offset;
99 uint64_t bytes;
2e1990b2
HR
100
101 /* The pointee is set by mirror_co_read(), mirror_co_zero(), and
102 * mirror_co_discard() before yielding for the first time */
103 int64_t *bytes_handled;
12aa4082 104
1181e19a 105 bool is_pseudo_op;
d06107ad 106 bool is_active_write;
ce8cabbd 107 bool is_in_flight;
12aa4082 108 CoQueue waiting_requests;
eed325b9 109 Coroutine *co;
d44dae1a 110 MirrorOp *waiting_for_op;
12aa4082
HR
111
112 QTAILQ_ENTRY(MirrorOp) next;
113};
bd48bde8 114
4295c5fc
HR
115typedef enum MirrorMethod {
116 MIRROR_METHOD_COPY,
117 MIRROR_METHOD_ZERO,
118 MIRROR_METHOD_DISCARD,
119} MirrorMethod;
120
b952b558
PB
121static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read,
122 int error)
123{
d06107ad 124 s->actively_synced = false;
b952b558 125 if (read) {
81e254dc
KW
126 return block_job_error_action(&s->common, s->on_source_error,
127 true, error);
b952b558 128 } else {
81e254dc
KW
129 return block_job_error_action(&s->common, s->on_target_error,
130 false, error);
b952b558
PB
131 }
132}
133
1181e19a
HR
134static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self,
135 MirrorBlockJob *s,
136 uint64_t offset,
137 uint64_t bytes)
138{
139 uint64_t self_start_chunk = offset / s->granularity;
140 uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity);
141 uint64_t self_nb_chunks = self_end_chunk - self_start_chunk;
142
143 while (find_next_bit(s->in_flight_bitmap, self_end_chunk,
144 self_start_chunk) < self_end_chunk &&
145 s->ret >= 0)
146 {
147 MirrorOp *op;
148
149 QTAILQ_FOREACH(op, &s->ops_in_flight, next) {
150 uint64_t op_start_chunk = op->offset / s->granularity;
151 uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes,
152 s->granularity) -
153 op_start_chunk;
154
155 if (op == self) {
156 continue;
157 }
158
159 if (ranges_overlap(self_start_chunk, self_nb_chunks,
160 op_start_chunk, op_nb_chunks))
161 {
66fed30c
SG
162 if (self) {
163 /*
164 * If the operation is already (indirectly) waiting for us,
165 * or will wait for us as soon as it wakes up, then just go
166 * on (instead of producing a deadlock in the former case).
167 */
168 if (op->waiting_for_op) {
169 continue;
170 }
171
172 self->waiting_for_op = op;
d44dae1a
VSO
173 }
174
1181e19a 175 qemu_co_queue_wait(&op->waiting_requests, NULL);
66fed30c
SG
176
177 if (self) {
178 self->waiting_for_op = NULL;
179 }
180
1181e19a
HR
181 break;
182 }
183 }
184 }
185}
186
2e1990b2 187static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret)
bd48bde8
PB
188{
189 MirrorBlockJob *s = op->s;
402a4741 190 struct iovec *iov;
bd48bde8 191 int64_t chunk_num;
b436982f 192 int i, nb_chunks;
bd48bde8 193
b436982f 194 trace_mirror_iteration_done(s, op->offset, op->bytes, ret);
bd48bde8
PB
195
196 s->in_flight--;
b436982f 197 s->bytes_in_flight -= op->bytes;
402a4741
PB
198 iov = op->qiov.iov;
199 for (i = 0; i < op->qiov.niov; i++) {
200 MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base;
201 QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next);
202 s->buf_free_count++;
203 }
204
b436982f
EB
205 chunk_num = op->offset / s->granularity;
206 nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity);
12aa4082 207
402a4741 208 bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks);
12aa4082 209 QTAILQ_REMOVE(&s->ops_in_flight, op, next);
b21c7652
HR
210 if (ret >= 0) {
211 if (s->cow_bitmap) {
212 bitmap_set(s->cow_bitmap, chunk_num, nb_chunks);
213 }
90ab48eb 214 if (!s->initial_zeroing_ongoing) {
30a5c887 215 job_progress_update(&s->common.job, op->bytes);
90ab48eb 216 }
bd48bde8 217 }
6df3bf8e 218 qemu_iovec_destroy(&op->qiov);
7b770c72 219
12aa4082
HR
220 qemu_co_queue_restart_all(&op->waiting_requests);
221 g_free(op);
bd48bde8
PB
222}
223
2e1990b2 224static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret)
bd48bde8 225{
bd48bde8 226 MirrorBlockJob *s = op->s;
b9e413dd 227
bd48bde8 228 if (ret < 0) {
bd48bde8
PB
229 BlockErrorAction action;
230
e0d7f73e 231 bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes);
bd48bde8 232 action = mirror_error_action(s, false, -ret);
a589569f 233 if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
bd48bde8
PB
234 s->ret = ret;
235 }
236 }
d12ade57 237
bd48bde8
PB
238 mirror_iteration_done(op, ret);
239}
240
2e1990b2 241static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret)
bd48bde8 242{
bd48bde8 243 MirrorBlockJob *s = op->s;
b9e413dd 244
bd48bde8 245 if (ret < 0) {
bd48bde8
PB
246 BlockErrorAction action;
247
e0d7f73e 248 bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes);
bd48bde8 249 action = mirror_error_action(s, true, -ret);
a589569f 250 if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) {
bd48bde8
PB
251 s->ret = ret;
252 }
253
254 mirror_iteration_done(op, ret);
d12ade57 255 return;
bd48bde8 256 }
d12ade57
VSO
257
258 ret = blk_co_pwritev(s->target, op->offset, op->qiov.size, &op->qiov, 0);
259 mirror_write_complete(op, ret);
bd48bde8
PB
260}
261
782d97ef
EB
262/* Clip bytes relative to offset to not exceed end-of-file */
263static inline int64_t mirror_clip_bytes(MirrorBlockJob *s,
264 int64_t offset,
265 int64_t bytes)
266{
267 return MIN(bytes, s->bdev_length - offset);
268}
269
782d97ef
EB
270/* Round offset and/or bytes to target cluster if COW is needed, and
271 * return the offset of the adjusted tail against original. */
272static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset,
ae4cc877 273 uint64_t *bytes)
893f7eba 274{
e5b43573
FZ
275 bool need_cow;
276 int ret = 0;
782d97ef 277 int64_t align_offset = *offset;
7cfd5275 278 int64_t align_bytes = *bytes;
782d97ef 279 int max_bytes = s->granularity * s->max_iov;
e5b43573 280
782d97ef
EB
281 need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap);
282 need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity,
e5b43573
FZ
283 s->cow_bitmap);
284 if (need_cow) {
782d97ef
EB
285 bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes,
286 &align_offset, &align_bytes);
e5b43573 287 }
3515727f 288
782d97ef
EB
289 if (align_bytes > max_bytes) {
290 align_bytes = max_bytes;
e5b43573 291 if (need_cow) {
782d97ef 292 align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size);
e5b43573 293 }
8f0720ec 294 }
782d97ef 295 /* Clipping may result in align_bytes unaligned to chunk boundary, but
4150ae60 296 * that doesn't matter because it's already the end of source image. */
782d97ef 297 align_bytes = mirror_clip_bytes(s, align_offset, align_bytes);
8f0720ec 298
782d97ef
EB
299 ret = align_offset + align_bytes - (*offset + *bytes);
300 *offset = align_offset;
301 *bytes = align_bytes;
e5b43573
FZ
302 assert(ret >= 0);
303 return ret;
304}
305
537c3d4f 306static inline void coroutine_fn
9178f4fe 307mirror_wait_for_any_operation(MirrorBlockJob *s, bool active)
21cd917f 308{
12aa4082
HR
309 MirrorOp *op;
310
1181e19a
HR
311 QTAILQ_FOREACH(op, &s->ops_in_flight, next) {
312 /* Do not wait on pseudo ops, because it may in turn wait on
313 * some other operation to start, which may in fact be the
314 * caller of this function. Since there is only one pseudo op
315 * at any given time, we will always find some real operation
316 * to wait on. */
ce8cabbd
KW
317 if (!op->is_pseudo_op && op->is_in_flight &&
318 op->is_active_write == active)
319 {
1181e19a
HR
320 qemu_co_queue_wait(&op->waiting_requests, NULL);
321 return;
322 }
323 }
324 abort();
21cd917f
FZ
325}
326
537c3d4f 327static inline void coroutine_fn
9178f4fe 328mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s)
d06107ad
HR
329{
330 /* Only non-active operations use up in-flight slots */
9178f4fe 331 mirror_wait_for_any_operation(s, false);
d06107ad
HR
332}
333
2e1990b2
HR
334/* Perform a mirror copy operation.
335 *
336 * *op->bytes_handled is set to the number of bytes copied after and
337 * including offset, excluding any bytes copied prior to offset due
338 * to alignment. This will be op->bytes if no alignment is necessary,
339 * or (new_end - op->offset) if the tail is rounded up or down due to
340 * alignment or buffer limit.
e5b43573 341 */
2e1990b2 342static void coroutine_fn mirror_co_read(void *opaque)
e5b43573 343{
2e1990b2
HR
344 MirrorOp *op = opaque;
345 MirrorBlockJob *s = op->s;
ae4cc877
EB
346 int nb_chunks;
347 uint64_t ret;
ae4cc877 348 uint64_t max_bytes;
e5b43573 349
ae4cc877 350 max_bytes = s->granularity * s->max_iov;
402a4741 351
e5b43573 352 /* We can only handle as much as buf_size at a time. */
2e1990b2
HR
353 op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes));
354 assert(op->bytes);
355 assert(op->bytes < BDRV_REQUEST_MAX_BYTES);
356 *op->bytes_handled = op->bytes;
402a4741 357
e5b43573 358 if (s->cow_bitmap) {
2e1990b2 359 *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes);
e5b43573 360 }
2e1990b2
HR
361 /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */
362 assert(*op->bytes_handled <= UINT_MAX);
363 assert(op->bytes <= s->buf_size);
ae4cc877 364 /* The offset is granularity-aligned because:
e5b43573
FZ
365 * 1) Caller passes in aligned values;
366 * 2) mirror_cow_align is used only when target cluster is larger. */
2e1990b2 367 assert(QEMU_IS_ALIGNED(op->offset, s->granularity));
ae4cc877 368 /* The range is sector-aligned, since bdrv_getlength() rounds up. */
2e1990b2
HR
369 assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE));
370 nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity);
e5b43573
FZ
371
372 while (s->buf_free_count < nb_chunks) {
2e1990b2 373 trace_mirror_yield_in_flight(s, op->offset, s->in_flight);
9178f4fe 374 mirror_wait_for_free_in_flight_slot(s);
b812f671
PB
375 }
376
402a4741
PB
377 /* Now make a QEMUIOVector taking enough granularity-sized chunks
378 * from s->buf_free.
379 */
380 qemu_iovec_init(&op->qiov, nb_chunks);
402a4741
PB
381 while (nb_chunks-- > 0) {
382 MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
2e1990b2 383 size_t remaining = op->bytes - op->qiov.size;
5a0f6fd5 384
402a4741
PB
385 QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
386 s->buf_free_count--;
5a0f6fd5 387 qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining));
402a4741 388 }
bd48bde8 389
893f7eba 390 /* Copy the dirty cluster. */
bd48bde8 391 s->in_flight++;
2e1990b2 392 s->bytes_in_flight += op->bytes;
ce8cabbd 393 op->is_in_flight = true;
2e1990b2 394 trace_mirror_one_iteration(s, op->offset, op->bytes);
dcfb3beb 395
138f9fff
HR
396 ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes,
397 &op->qiov, 0);
2e1990b2 398 mirror_read_complete(op, ret);
e5b43573
FZ
399}
400
2e1990b2 401static void coroutine_fn mirror_co_zero(void *opaque)
e5b43573 402{
2e1990b2
HR
403 MirrorOp *op = opaque;
404 int ret;
e5b43573 405
2e1990b2
HR
406 op->s->in_flight++;
407 op->s->bytes_in_flight += op->bytes;
408 *op->bytes_handled = op->bytes;
ce8cabbd 409 op->is_in_flight = true;
e5b43573 410
2e1990b2
HR
411 ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes,
412 op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0);
413 mirror_write_complete(op, ret);
414}
415
416static void coroutine_fn mirror_co_discard(void *opaque)
417{
418 MirrorOp *op = opaque;
419 int ret;
420
421 op->s->in_flight++;
422 op->s->bytes_in_flight += op->bytes;
423 *op->bytes_handled = op->bytes;
ce8cabbd 424 op->is_in_flight = true;
2e1990b2
HR
425
426 ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes);
427 mirror_write_complete(op, ret);
e5b43573
FZ
428}
429
4295c5fc
HR
430static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset,
431 unsigned bytes, MirrorMethod mirror_method)
432{
2e1990b2
HR
433 MirrorOp *op;
434 Coroutine *co;
435 int64_t bytes_handled = -1;
436
437 op = g_new(MirrorOp, 1);
438 *op = (MirrorOp){
439 .s = s,
440 .offset = offset,
441 .bytes = bytes,
442 .bytes_handled = &bytes_handled,
443 };
12aa4082 444 qemu_co_queue_init(&op->waiting_requests);
2e1990b2 445
4295c5fc
HR
446 switch (mirror_method) {
447 case MIRROR_METHOD_COPY:
2e1990b2
HR
448 co = qemu_coroutine_create(mirror_co_read, op);
449 break;
4295c5fc 450 case MIRROR_METHOD_ZERO:
2e1990b2
HR
451 co = qemu_coroutine_create(mirror_co_zero, op);
452 break;
4295c5fc 453 case MIRROR_METHOD_DISCARD:
2e1990b2
HR
454 co = qemu_coroutine_create(mirror_co_discard, op);
455 break;
4295c5fc
HR
456 default:
457 abort();
458 }
eed325b9 459 op->co = co;
2e1990b2 460
12aa4082 461 QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next);
2e1990b2
HR
462 qemu_coroutine_enter(co);
463 /* At this point, ownership of op has been moved to the coroutine
464 * and the object may already be freed */
465
466 /* Assert that this value has been set */
467 assert(bytes_handled >= 0);
468
469 /* Same assertion as in mirror_co_read() (and for mirror_co_read()
470 * and mirror_co_discard(), bytes_handled == op->bytes, which
471 * is the @bytes parameter given to this function) */
472 assert(bytes_handled <= UINT_MAX);
473 return bytes_handled;
4295c5fc
HR
474}
475
e5b43573
FZ
476static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)
477{
138f9fff 478 BlockDriverState *source = s->mirror_top_bs->backing->bs;
1181e19a
HR
479 MirrorOp *pseudo_op;
480 int64_t offset;
481 uint64_t delay_ns = 0, ret = 0;
e5b43573
FZ
482 /* At least the first dirty chunk is mirrored in one iteration. */
483 int nb_chunks = 1;
4b5004d9 484 bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target));
b436982f 485 int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES);
e5b43573 486
b64bd51e 487 bdrv_dirty_bitmap_lock(s->dirty_bitmap);
f798184c 488 offset = bdrv_dirty_iter_next(s->dbi);
fb2ef791 489 if (offset < 0) {
dc162c8e 490 bdrv_set_dirty_iter(s->dbi, 0);
f798184c 491 offset = bdrv_dirty_iter_next(s->dbi);
9a46dba7 492 trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap));
fb2ef791 493 assert(offset >= 0);
e5b43573 494 }
b64bd51e 495 bdrv_dirty_bitmap_unlock(s->dirty_bitmap);
e5b43573 496
1181e19a 497 mirror_wait_on_conflicts(NULL, s, offset, 1);
9c83625b 498
da01ff7f 499 job_pause_point(&s->common.job);
565ac01f 500
e5b43573
FZ
501 /* Find the number of consective dirty chunks following the first dirty
502 * one, and wait for in flight requests in them. */
b64bd51e 503 bdrv_dirty_bitmap_lock(s->dirty_bitmap);
fb2ef791 504 while (nb_chunks * s->granularity < s->buf_size) {
dc162c8e 505 int64_t next_dirty;
fb2ef791
EB
506 int64_t next_offset = offset + nb_chunks * s->granularity;
507 int64_t next_chunk = next_offset / s->granularity;
508 if (next_offset >= s->bdev_length ||
28636b82 509 !bdrv_dirty_bitmap_get_locked(s->dirty_bitmap, next_offset)) {
e5b43573
FZ
510 break;
511 }
512 if (test_bit(next_chunk, s->in_flight_bitmap)) {
9c83625b 513 break;
e5b43573 514 }
9c83625b 515
f798184c 516 next_dirty = bdrv_dirty_iter_next(s->dbi);
fb2ef791 517 if (next_dirty > next_offset || next_dirty < 0) {
f27a2742 518 /* The bitmap iterator's cache is stale, refresh it */
715a74d8 519 bdrv_set_dirty_iter(s->dbi, next_offset);
f798184c 520 next_dirty = bdrv_dirty_iter_next(s->dbi);
f27a2742 521 }
fb2ef791 522 assert(next_dirty == next_offset);
9c83625b 523 nb_chunks++;
e5b43573
FZ
524 }
525
526 /* Clear dirty bits before querying the block status, because
31826642 527 * calling bdrv_block_status_above could yield - if some blocks are
e5b43573
FZ
528 * marked dirty in this window, we need to know.
529 */
e0d7f73e
EB
530 bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset,
531 nb_chunks * s->granularity);
b64bd51e
PB
532 bdrv_dirty_bitmap_unlock(s->dirty_bitmap);
533
1181e19a
HR
534 /* Before claiming an area in the in-flight bitmap, we have to
535 * create a MirrorOp for it so that conflicting requests can wait
536 * for it. mirror_perform() will create the real MirrorOps later,
537 * for now we just create a pseudo operation that will wake up all
538 * conflicting requests once all real operations have been
539 * launched. */
540 pseudo_op = g_new(MirrorOp, 1);
541 *pseudo_op = (MirrorOp){
542 .offset = offset,
543 .bytes = nb_chunks * s->granularity,
544 .is_pseudo_op = true,
545 };
546 qemu_co_queue_init(&pseudo_op->waiting_requests);
547 QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next);
548
fb2ef791
EB
549 bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks);
550 while (nb_chunks > 0 && offset < s->bdev_length) {
31826642 551 int ret;
7cfd5275 552 int64_t io_bytes;
f3e4ce4a 553 int64_t io_bytes_acct;
4295c5fc 554 MirrorMethod mirror_method = MIRROR_METHOD_COPY;
e5b43573 555
fb2ef791 556 assert(!(offset % s->granularity));
31826642
EB
557 ret = bdrv_block_status_above(source, NULL, offset,
558 nb_chunks * s->granularity,
559 &io_bytes, NULL, NULL);
e5b43573 560 if (ret < 0) {
fb2ef791 561 io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes);
0965a41e 562 } else if (ret & BDRV_BLOCK_DATA) {
fb2ef791 563 io_bytes = MIN(io_bytes, max_io_bytes);
e5b43573
FZ
564 }
565
fb2ef791
EB
566 io_bytes -= io_bytes % s->granularity;
567 if (io_bytes < s->granularity) {
568 io_bytes = s->granularity;
e5b43573 569 } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) {
fb2ef791 570 int64_t target_offset;
7cfd5275 571 int64_t target_bytes;
fb2ef791
EB
572 bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes,
573 &target_offset, &target_bytes);
574 if (target_offset == offset &&
575 target_bytes == io_bytes) {
e5b43573
FZ
576 mirror_method = ret & BDRV_BLOCK_ZERO ?
577 MIRROR_METHOD_ZERO :
578 MIRROR_METHOD_DISCARD;
579 }
580 }
581
cf56a3c6 582 while (s->in_flight >= MAX_IN_FLIGHT) {
fb2ef791 583 trace_mirror_yield_in_flight(s, offset, s->in_flight);
9178f4fe 584 mirror_wait_for_free_in_flight_slot(s);
cf56a3c6
DL
585 }
586
dbaa7b57 587 if (s->ret < 0) {
1181e19a
HR
588 ret = 0;
589 goto fail;
dbaa7b57
VSO
590 }
591
fb2ef791 592 io_bytes = mirror_clip_bytes(s, offset, io_bytes);
4295c5fc
HR
593 io_bytes = mirror_perform(s, offset, io_bytes, mirror_method);
594 if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) {
595 io_bytes_acct = 0;
596 } else {
597 io_bytes_acct = io_bytes;
e5b43573 598 }
fb2ef791
EB
599 assert(io_bytes);
600 offset += io_bytes;
601 nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity);
dee81d51 602 delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct);
dcfb3beb 603 }
1181e19a
HR
604
605 ret = delay_ns;
606fail:
607 QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next);
608 qemu_co_queue_restart_all(&pseudo_op->waiting_requests);
609 g_free(pseudo_op);
610
611 return ret;
bd48bde8 612}
b952b558 613
402a4741
PB
614static void mirror_free_init(MirrorBlockJob *s)
615{
616 int granularity = s->granularity;
617 size_t buf_size = s->buf_size;
618 uint8_t *buf = s->buf;
619
620 assert(s->buf_free_count == 0);
621 QSIMPLEQ_INIT(&s->buf_free);
622 while (buf_size != 0) {
623 MirrorBuffer *cur = (MirrorBuffer *)buf;
624 QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next);
625 s->buf_free_count++;
626 buf_size -= granularity;
627 buf += granularity;
628 }
629}
630
bae8196d
PB
631/* This is also used for the .pause callback. There is no matching
632 * mirror_resume() because mirror_run() will begin iterating again
633 * when the job is resumed.
634 */
537c3d4f 635static void coroutine_fn mirror_wait_for_all_io(MirrorBlockJob *s)
bd48bde8
PB
636{
637 while (s->in_flight > 0) {
9178f4fe 638 mirror_wait_for_free_in_flight_slot(s);
bd48bde8 639 }
893f7eba
PB
640}
641
737efc1e
JS
642/**
643 * mirror_exit_common: handle both abort() and prepare() cases.
644 * for .prepare, returns 0 on success and -errno on failure.
645 * for .abort cases, denoted by abort = true, MUST return 0.
646 */
647static int mirror_exit_common(Job *job)
5a7e7a0b 648{
1908a559
KW
649 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
650 BlockJob *bjob = &s->common;
f93c3add 651 MirrorBDSOpaque *bs_opaque;
5a7e7a0b 652 AioContext *replace_aio_context = NULL;
f93c3add
HR
653 BlockDriverState *src;
654 BlockDriverState *target_bs;
655 BlockDriverState *mirror_top_bs;
12fa4af6 656 Error *local_err = NULL;
737efc1e
JS
657 bool abort = job->ret < 0;
658 int ret = 0;
659
660 if (s->prepared) {
661 return 0;
662 }
663 s->prepared = true;
3f09bfbc 664
f93c3add
HR
665 mirror_top_bs = s->mirror_top_bs;
666 bs_opaque = mirror_top_bs->opaque;
667 src = mirror_top_bs->backing->bs;
668 target_bs = blk_bs(s->target);
669
ef53dc09
AG
670 if (bdrv_chain_contains(src, target_bs)) {
671 bdrv_unfreeze_backing_chain(mirror_top_bs, target_bs);
672 }
673
5deb6cbd 674 bdrv_release_dirty_bitmap(s->dirty_bitmap);
2119882c 675
7b508f6b
JS
676 /* Make sure that the source BDS doesn't go away during bdrv_replace_node,
677 * before we can call bdrv_drained_end */
3f09bfbc 678 bdrv_ref(src);
4ef85a9c 679 bdrv_ref(mirror_top_bs);
7d9fcb39
KW
680 bdrv_ref(target_bs);
681
bb0c9409
VSO
682 /*
683 * Remove target parent that still uses BLK_PERM_WRITE/RESIZE before
7d9fcb39 684 * inserting target_bs at s->to_replace, where we might not be able to get
63c8ef28 685 * these permissions.
bb0c9409 686 */
7d9fcb39
KW
687 blk_unref(s->target);
688 s->target = NULL;
4ef85a9c
KW
689
690 /* We don't access the source any more. Dropping any WRITE/RESIZE is
d2da5e28
KW
691 * required before it could become a backing file of target_bs. Not having
692 * these permissions any more means that we can't allow any new requests on
693 * mirror_top_bs from now on, so keep it drained. */
694 bdrv_drained_begin(mirror_top_bs);
f94dc3b4
HR
695 bs_opaque->stop = true;
696 bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing,
697 &error_abort);
737efc1e 698 if (!abort && s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) {
4ef85a9c 699 BlockDriverState *backing = s->is_none_mode ? src : s->base;
3f072a7f
HR
700 BlockDriverState *unfiltered_target = bdrv_skip_filters(target_bs);
701
702 if (bdrv_cow_bs(unfiltered_target) != backing) {
703 bdrv_set_backing_hd(unfiltered_target, backing, &local_err);
12fa4af6
KW
704 if (local_err) {
705 error_report_err(local_err);
66c8672d 706 local_err = NULL;
7b508f6b 707 ret = -EPERM;
12fa4af6 708 }
4ef85a9c 709 }
c41f5b96
HR
710 } else if (!abort && s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) {
711 assert(!bdrv_backing_chain_next(target_bs));
712 ret = bdrv_open_backing_file(bdrv_skip_filters(target_bs), NULL,
713 "backing", &local_err);
714 if (ret < 0) {
715 error_report_err(local_err);
716 local_err = NULL;
717 }
4ef85a9c 718 }
5a7e7a0b
SH
719
720 if (s->to_replace) {
721 replace_aio_context = bdrv_get_aio_context(s->to_replace);
722 aio_context_acquire(replace_aio_context);
723 }
724
737efc1e
JS
725 if (s->should_complete && !abort) {
726 BlockDriverState *to_replace = s->to_replace ?: src;
1ba79388 727 bool ro = bdrv_is_read_only(to_replace);
40365552 728
1ba79388
AG
729 if (ro != bdrv_is_read_only(target_bs)) {
730 bdrv_reopen_set_read_only(target_bs, ro, NULL);
5a7e7a0b 731 }
b8804815
KW
732
733 /* The mirror job has no requests in flight any more, but we need to
734 * drain potential other users of the BDS before changing the graph. */
5e771752 735 assert(s->in_drain);
e253f4b8 736 bdrv_drained_begin(target_bs);
6e9cc051
HR
737 /*
738 * Cannot use check_to_replace_node() here, because that would
739 * check for an op blocker on @to_replace, and we have our own
740 * there.
741 */
742 if (bdrv_recurse_can_replace(src, to_replace)) {
743 bdrv_replace_node(to_replace, target_bs, &local_err);
744 } else {
745 error_setg(&local_err, "Can no longer replace '%s' by '%s', "
746 "because it can no longer be guaranteed that doing so "
747 "would not lead to an abrupt change of visible data",
748 to_replace->node_name, target_bs->node_name);
749 }
e253f4b8 750 bdrv_drained_end(target_bs);
5fe31c25
KW
751 if (local_err) {
752 error_report_err(local_err);
7b508f6b 753 ret = -EPERM;
5fe31c25 754 }
5a7e7a0b
SH
755 }
756 if (s->to_replace) {
757 bdrv_op_unblock_all(s->to_replace, s->replace_blocker);
758 error_free(s->replace_blocker);
759 bdrv_unref(s->to_replace);
760 }
761 if (replace_aio_context) {
762 aio_context_release(replace_aio_context);
763 }
764 g_free(s->replaces);
7d9fcb39 765 bdrv_unref(target_bs);
4ef85a9c 766
f94dc3b4
HR
767 /*
768 * Remove the mirror filter driver from the graph. Before this, get rid of
4ef85a9c 769 * the blockers on the intermediate nodes so that the resulting state is
f94dc3b4
HR
770 * valid.
771 */
1908a559 772 block_job_remove_all_bdrv(bjob);
3f072a7f 773 bdrv_replace_node(mirror_top_bs, mirror_top_bs->backing->bs, &error_abort);
4ef85a9c 774
429076e8 775 bs_opaque->job = NULL;
4ef85a9c 776
176c3699 777 bdrv_drained_end(src);
d2da5e28 778 bdrv_drained_end(mirror_top_bs);
5e771752 779 s->in_drain = false;
4ef85a9c 780 bdrv_unref(mirror_top_bs);
3f09bfbc 781 bdrv_unref(src);
7b508f6b 782
737efc1e
JS
783 return ret;
784}
785
786static int mirror_prepare(Job *job)
787{
788 return mirror_exit_common(job);
789}
790
791static void mirror_abort(Job *job)
792{
793 int ret = mirror_exit_common(job);
794 assert(ret == 0);
5a7e7a0b
SH
795}
796
537c3d4f 797static void coroutine_fn mirror_throttle(MirrorBlockJob *s)
49efb1f5
DL
798{
799 int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
800
18bb6928 801 if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) {
49efb1f5 802 s->last_pause_ns = now;
5d43e86e 803 job_sleep_ns(&s->common.job, 0);
49efb1f5 804 } else {
da01ff7f 805 job_pause_point(&s->common.job);
49efb1f5
DL
806 }
807}
808
c0b363ad
DL
809static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s)
810{
23ca459a 811 int64_t offset;
138f9fff 812 BlockDriverState *bs = s->mirror_top_bs->backing->bs;
c0b363ad 813 BlockDriverState *target_bs = blk_bs(s->target);
23ca459a 814 int ret;
51b0a488 815 int64_t count;
c0b363ad 816
cdf3bc93 817 if (s->zero_target) {
c7c2769c 818 if (!bdrv_can_write_zeroes_with_unmap(target_bs)) {
e0d7f73e 819 bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length);
c7c2769c
DL
820 return 0;
821 }
822
90ab48eb 823 s->initial_zeroing_ongoing = true;
23ca459a
EB
824 for (offset = 0; offset < s->bdev_length; ) {
825 int bytes = MIN(s->bdev_length - offset,
826 QEMU_ALIGN_DOWN(INT_MAX, s->granularity));
c7c2769c
DL
827
828 mirror_throttle(s);
829
daa7f2f9 830 if (job_is_cancelled(&s->common.job)) {
90ab48eb 831 s->initial_zeroing_ongoing = false;
c7c2769c
DL
832 return 0;
833 }
834
835 if (s->in_flight >= MAX_IN_FLIGHT) {
67adf4b3
EB
836 trace_mirror_yield(s, UINT64_MAX, s->buf_free_count,
837 s->in_flight);
9178f4fe 838 mirror_wait_for_free_in_flight_slot(s);
c7c2769c
DL
839 continue;
840 }
841
4295c5fc 842 mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO);
23ca459a 843 offset += bytes;
c7c2769c
DL
844 }
845
bae8196d 846 mirror_wait_for_all_io(s);
90ab48eb 847 s->initial_zeroing_ongoing = false;
b7d5062c
DL
848 }
849
c0b363ad 850 /* First part, loop on the sectors and initialize the dirty bitmap. */
23ca459a 851 for (offset = 0; offset < s->bdev_length; ) {
c0b363ad 852 /* Just to make sure we are not exceeding int limit. */
23ca459a
EB
853 int bytes = MIN(s->bdev_length - offset,
854 QEMU_ALIGN_DOWN(INT_MAX, s->granularity));
c0b363ad
DL
855
856 mirror_throttle(s);
857
daa7f2f9 858 if (job_is_cancelled(&s->common.job)) {
c0b363ad
DL
859 return 0;
860 }
861
3f072a7f
HR
862 ret = bdrv_is_allocated_above(bs, s->base_overlay, true, offset, bytes,
863 &count);
c0b363ad
DL
864 if (ret < 0) {
865 return ret;
866 }
867
23ca459a 868 assert(count);
a92b1b06 869 if (ret > 0) {
23ca459a 870 bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count);
c0b363ad 871 }
23ca459a 872 offset += count;
c0b363ad
DL
873 }
874 return 0;
875}
876
bdffb31d
PB
877/* Called when going out of the streaming phase to flush the bulk of the
878 * data to the medium, or just before completing.
879 */
880static int mirror_flush(MirrorBlockJob *s)
881{
882 int ret = blk_flush(s->target);
883 if (ret < 0) {
884 if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) {
885 s->ret = ret;
886 }
887 }
888 return ret;
889}
890
f67432a2 891static int coroutine_fn mirror_run(Job *job, Error **errp)
893f7eba 892{
f67432a2 893 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
138f9fff 894 BlockDriverState *bs = s->mirror_top_bs->backing->bs;
e253f4b8 895 BlockDriverState *target_bs = blk_bs(s->target);
9a0cec66 896 bool need_drain = true;
d59cb66d 897 BlockDeviceIoStatus iostatus;
c0b363ad 898 int64_t length;
e83dd680 899 int64_t target_length;
b812f671 900 BlockDriverInfo bdi;
1d33936e
JC
901 char backing_filename[2]; /* we only need 2 characters because we are only
902 checking for a NULL string */
893f7eba 903 int ret = 0;
893f7eba 904
daa7f2f9 905 if (job_is_cancelled(&s->common.job)) {
893f7eba
PB
906 goto immediate_exit;
907 }
908
b21c7652
HR
909 s->bdev_length = bdrv_getlength(bs);
910 if (s->bdev_length < 0) {
911 ret = s->bdev_length;
373df5b1 912 goto immediate_exit;
becc347e
KW
913 }
914
e83dd680
KW
915 target_length = blk_getlength(s->target);
916 if (target_length < 0) {
917 ret = target_length;
918 goto immediate_exit;
919 }
920
becc347e
KW
921 /* Active commit must resize the base image if its size differs from the
922 * active layer. */
923 if (s->base == blk_bs(s->target)) {
e83dd680 924 if (s->bdev_length > target_length) {
c80d8b06 925 ret = blk_truncate(s->target, s->bdev_length, false,
8c6242b6 926 PREALLOC_MODE_OFF, 0, NULL);
becc347e
KW
927 if (ret < 0) {
928 goto immediate_exit;
929 }
930 }
e83dd680
KW
931 } else if (s->bdev_length != target_length) {
932 error_setg(errp, "Source and target image have different sizes");
933 ret = -EINVAL;
934 goto immediate_exit;
becc347e
KW
935 }
936
937 if (s->bdev_length == 0) {
2e1795b5
KW
938 /* Transition to the READY state and wait for complete. */
939 job_transition_to_ready(&s->common.job);
d06107ad 940 s->actively_synced = true;
08b83bff 941 while (!job_cancel_requested(&s->common.job) && !s->should_complete) {
198c49cc 942 job_yield(&s->common.job);
9e48b025 943 }
9e48b025 944 goto immediate_exit;
893f7eba
PB
945 }
946
b21c7652 947 length = DIV_ROUND_UP(s->bdev_length, s->granularity);
402a4741
PB
948 s->in_flight_bitmap = bitmap_new(length);
949
b812f671
PB
950 /* If we have no backing file yet in the destination, we cannot let
951 * the destination do COW. Instead, we copy sectors around the
952 * dirty data if needed. We need a bitmap to do that.
953 */
e253f4b8 954 bdrv_get_backing_filename(target_bs, backing_filename,
b812f671 955 sizeof(backing_filename));
e253f4b8 956 if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) {
b436982f
EB
957 s->target_cluster_size = bdi.cluster_size;
958 } else {
959 s->target_cluster_size = BDRV_SECTOR_SIZE;
e5b43573 960 }
3f072a7f 961 if (backing_filename[0] && !bdrv_backing_chain_next(target_bs) &&
b436982f
EB
962 s->granularity < s->target_cluster_size) {
963 s->buf_size = MAX(s->buf_size, s->target_cluster_size);
e5b43573 964 s->cow_bitmap = bitmap_new(length);
b812f671 965 }
e253f4b8 966 s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov);
b812f671 967
7504edf4
KW
968 s->buf = qemu_try_blockalign(bs, s->buf_size);
969 if (s->buf == NULL) {
970 ret = -ENOMEM;
971 goto immediate_exit;
972 }
973
402a4741 974 mirror_free_init(s);
893f7eba 975
49efb1f5 976 s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
03544a6e 977 if (!s->is_none_mode) {
c0b363ad 978 ret = mirror_dirty_init(s);
daa7f2f9 979 if (ret < 0 || job_is_cancelled(&s->common.job)) {
c0b363ad 980 goto immediate_exit;
893f7eba
PB
981 }
982 }
983
dc162c8e 984 assert(!s->dbi);
715a74d8 985 s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap);
893f7eba 986 for (;;) {
cc8c9d6c 987 uint64_t delay_ns = 0;
49efb1f5 988 int64_t cnt, delta;
893f7eba
PB
989 bool should_complete;
990
d06107ad
HR
991 /* Do not start passive operations while there are active
992 * writes in progress */
993 while (s->in_active_write_counter) {
9178f4fe 994 mirror_wait_for_any_operation(s, true);
d06107ad
HR
995 }
996
bd48bde8
PB
997 if (s->ret < 0) {
998 ret = s->ret;
999 goto immediate_exit;
1000 }
1001
da01ff7f 1002 job_pause_point(&s->common.job);
565ac01f 1003
4feeec7e
HR
1004 if (job_is_cancelled(&s->common.job)) {
1005 ret = 0;
1006 goto immediate_exit;
1007 }
1008
20dca810 1009 cnt = bdrv_get_dirty_count(s->dirty_bitmap);
05df8a6a
KW
1010 /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is
1011 * the number of bytes currently being processed; together those are
1012 * the current remaining operation length */
30a5c887 1013 job_progress_set_remaining(&s->common.job, s->bytes_in_flight + cnt);
bd48bde8
PB
1014
1015 /* Note that even when no rate limit is applied we need to yield
a7282330 1016 * periodically with no pending I/O so that bdrv_drain_all() returns.
18bb6928
KW
1017 * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is
1018 * an error, or when the source is clean, whichever comes first. */
49efb1f5 1019 delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns;
d59cb66d
EGE
1020 WITH_JOB_LOCK_GUARD() {
1021 iostatus = s->common.iostatus;
1022 }
18bb6928 1023 if (delta < BLOCK_JOB_SLICE_TIME &&
d59cb66d 1024 iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
cf56a3c6 1025 if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 ||
402a4741 1026 (cnt == 0 && s->in_flight > 0)) {
9a46dba7 1027 trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight);
9178f4fe 1028 mirror_wait_for_free_in_flight_slot(s);
bd48bde8
PB
1029 continue;
1030 } else if (cnt != 0) {
cc8c9d6c 1031 delay_ns = mirror_iteration(s);
893f7eba 1032 }
893f7eba
PB
1033 }
1034
1035 should_complete = false;
bd48bde8 1036 if (s->in_flight == 0 && cnt == 0) {
893f7eba 1037 trace_mirror_before_flush(s);
44716224 1038 if (!job_is_ready(&s->common.job)) {
bdffb31d
PB
1039 if (mirror_flush(s) < 0) {
1040 /* Go check s->ret. */
1041 continue;
b952b558 1042 }
b952b558
PB
1043 /* We're out of the streaming phase. From now on, if the job
1044 * is cancelled we will actually complete all pending I/O and
1045 * report completion. This way, block-job-cancel will leave
1046 * the target in a consistent state.
1047 */
2e1795b5 1048 job_transition_to_ready(&s->common.job);
d06107ad
HR
1049 if (s->copy_mode != MIRROR_COPY_MODE_BACKGROUND) {
1050 s->actively_synced = true;
1051 }
d63ffd87 1052 }
bdffb31d
PB
1053
1054 should_complete = s->should_complete ||
08b83bff 1055 job_cancel_requested(&s->common.job);
bdffb31d 1056 cnt = bdrv_get_dirty_count(s->dirty_bitmap);
893f7eba
PB
1057 }
1058
1059 if (cnt == 0 && should_complete) {
1060 /* The dirty bitmap is not updated while operations are pending.
1061 * If we're about to exit, wait for pending operations before
1062 * calling bdrv_get_dirty_count(bs), or we may exit while the
1063 * source has dirty data to copy!
1064 *
1065 * Note that I/O can be submitted by the guest while
9a0cec66
PB
1066 * mirror_populate runs, so pause it now. Before deciding
1067 * whether to switch to target check one last time if I/O has
1068 * come in the meanwhile, and if not flush the data to disk.
893f7eba 1069 */
9a46dba7 1070 trace_mirror_before_drain(s, cnt);
9a0cec66 1071
5e771752 1072 s->in_drain = true;
9a0cec66 1073 bdrv_drained_begin(bs);
20dca810 1074 cnt = bdrv_get_dirty_count(s->dirty_bitmap);
bdffb31d 1075 if (cnt > 0 || mirror_flush(s) < 0) {
9a0cec66 1076 bdrv_drained_end(bs);
5e771752 1077 s->in_drain = false;
9a0cec66
PB
1078 continue;
1079 }
1080
1081 /* The two disks are in sync. Exit and report successful
1082 * completion.
1083 */
1084 assert(QLIST_EMPTY(&bs->tracked_requests));
9a0cec66
PB
1085 need_drain = false;
1086 break;
893f7eba
PB
1087 }
1088
44716224 1089 if (job_is_ready(&s->common.job) && !should_complete) {
18bb6928
KW
1090 delay_ns = (s->in_flight == 0 &&
1091 cnt == 0 ? BLOCK_JOB_SLICE_TIME : 0);
ddc4115e 1092 }
44716224
HR
1093 trace_mirror_before_sleep(s, cnt, job_is_ready(&s->common.job),
1094 delay_ns);
5d43e86e 1095 job_sleep_ns(&s->common.job, delay_ns);
49efb1f5 1096 s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
893f7eba
PB
1097 }
1098
1099immediate_exit:
bd48bde8
PB
1100 if (s->in_flight > 0) {
1101 /* We get here only if something went wrong. Either the job failed,
1102 * or it was cancelled prematurely so that we do not guarantee that
1103 * the target is a copy of the source.
1104 */
08b83bff 1105 assert(ret < 0 || job_is_cancelled(&s->common.job));
9a0cec66 1106 assert(need_drain);
bae8196d 1107 mirror_wait_for_all_io(s);
bd48bde8
PB
1108 }
1109
1110 assert(s->in_flight == 0);
7191bf31 1111 qemu_vfree(s->buf);
b812f671 1112 g_free(s->cow_bitmap);
402a4741 1113 g_free(s->in_flight_bitmap);
dc162c8e 1114 bdrv_dirty_iter_free(s->dbi);
5a7e7a0b 1115
9a0cec66 1116 if (need_drain) {
5e771752 1117 s->in_drain = true;
9a0cec66
PB
1118 bdrv_drained_begin(bs);
1119 }
f67432a2 1120
f67432a2 1121 return ret;
893f7eba
PB
1122}
1123
3453d972 1124static void mirror_complete(Job *job, Error **errp)
d63ffd87 1125{
3453d972 1126 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
d63ffd87 1127
44716224 1128 if (!job_is_ready(job)) {
9df229c3 1129 error_setg(errp, "The active block job '%s' cannot be completed",
3453d972 1130 job->id);
d63ffd87
PB
1131 return;
1132 }
1133
15d67298 1134 /* block all operations on to_replace bs */
09158f00 1135 if (s->replaces) {
5a7e7a0b
SH
1136 AioContext *replace_aio_context;
1137
e12f3784 1138 s->to_replace = bdrv_find_node(s->replaces);
09158f00 1139 if (!s->to_replace) {
e12f3784 1140 error_setg(errp, "Node name '%s' not found", s->replaces);
09158f00
BC
1141 return;
1142 }
1143
5a7e7a0b
SH
1144 replace_aio_context = bdrv_get_aio_context(s->to_replace);
1145 aio_context_acquire(replace_aio_context);
1146
64631f36 1147 /* TODO Translate this into child freeze system. */
09158f00
BC
1148 error_setg(&s->replace_blocker,
1149 "block device is in use by block-job-complete");
1150 bdrv_op_block_all(s->to_replace, s->replace_blocker);
1151 bdrv_ref(s->to_replace);
5a7e7a0b
SH
1152
1153 aio_context_release(replace_aio_context);
09158f00
BC
1154 }
1155
d63ffd87 1156 s->should_complete = true;
00769414
HR
1157
1158 /* If the job is paused, it will be re-entered when it is resumed */
279ac06e
EGE
1159 WITH_JOB_LOCK_GUARD() {
1160 if (!job->paused) {
1161 job_enter_cond_locked(job, NULL);
1162 }
00769414 1163 }
d63ffd87
PB
1164}
1165
537c3d4f 1166static void coroutine_fn mirror_pause(Job *job)
565ac01f 1167{
da01ff7f 1168 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
565ac01f 1169
bae8196d 1170 mirror_wait_for_all_io(s);
565ac01f
SH
1171}
1172
89bd0305
KW
1173static bool mirror_drained_poll(BlockJob *job)
1174{
1175 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
5e771752
SL
1176
1177 /* If the job isn't paused nor cancelled, we can't be sure that it won't
1178 * issue more requests. We make an exception if we've reached this point
1179 * from one of our own drain sections, to avoid a deadlock waiting for
1180 * ourselves.
1181 */
279ac06e
EGE
1182 WITH_JOB_LOCK_GUARD() {
1183 if (!s->common.job.paused && !job_is_cancelled_locked(&job->job)
1184 && !s->in_drain) {
1185 return true;
1186 }
5e771752
SL
1187 }
1188
89bd0305
KW
1189 return !!s->in_flight;
1190}
1191
73895f38 1192static bool mirror_cancel(Job *job, bool force)
521ff8b7
VSO
1193{
1194 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job);
1195 BlockDriverState *target = blk_bs(s->target);
1196
73895f38
HR
1197 /*
1198 * Before the job is READY, we treat any cancellation like a
1199 * force-cancellation.
1200 */
1201 force = force || !job_is_ready(job);
1202
1203 if (force) {
9c785cd7
VSO
1204 bdrv_cancel_in_flight(target);
1205 }
73895f38
HR
1206 return force;
1207}
1208
1209static bool commit_active_cancel(Job *job, bool force)
1210{
1211 /* Same as above in mirror_cancel() */
1212 return force || !job_is_ready(job);
521ff8b7
VSO
1213}
1214
3fc4b10a 1215static const BlockJobDriver mirror_job_driver = {
33e9e9bd
KW
1216 .job_driver = {
1217 .instance_size = sizeof(MirrorBlockJob),
252291ea 1218 .job_type = JOB_TYPE_MIRROR,
80fa2c75 1219 .free = block_job_free,
b15de828 1220 .user_resume = block_job_user_resume,
f67432a2 1221 .run = mirror_run,
737efc1e
JS
1222 .prepare = mirror_prepare,
1223 .abort = mirror_abort,
da01ff7f 1224 .pause = mirror_pause,
3453d972 1225 .complete = mirror_complete,
521ff8b7 1226 .cancel = mirror_cancel,
33e9e9bd 1227 },
89bd0305 1228 .drained_poll = mirror_drained_poll,
893f7eba
PB
1229};
1230
03544a6e 1231static const BlockJobDriver commit_active_job_driver = {
33e9e9bd
KW
1232 .job_driver = {
1233 .instance_size = sizeof(MirrorBlockJob),
252291ea 1234 .job_type = JOB_TYPE_COMMIT,
80fa2c75 1235 .free = block_job_free,
b15de828 1236 .user_resume = block_job_user_resume,
f67432a2 1237 .run = mirror_run,
737efc1e
JS
1238 .prepare = mirror_prepare,
1239 .abort = mirror_abort,
da01ff7f 1240 .pause = mirror_pause,
3453d972 1241 .complete = mirror_complete,
73895f38 1242 .cancel = commit_active_cancel,
33e9e9bd 1243 },
89bd0305 1244 .drained_poll = mirror_drained_poll,
03544a6e
FZ
1245};
1246
537c3d4f
SH
1247static void coroutine_fn
1248do_sync_target_write(MirrorBlockJob *job, MirrorMethod method,
1249 uint64_t offset, uint64_t bytes,
1250 QEMUIOVector *qiov, int flags)
d06107ad 1251{
5c511ac3 1252 int ret;
dbdf699c
VSO
1253 size_t qiov_offset = 0;
1254 int64_t bitmap_offset, bitmap_end;
d06107ad 1255
dbdf699c
VSO
1256 if (!QEMU_IS_ALIGNED(offset, job->granularity) &&
1257 bdrv_dirty_bitmap_get(job->dirty_bitmap, offset))
1258 {
1259 /*
1260 * Dirty unaligned padding: ignore it.
1261 *
1262 * Reasoning:
1263 * 1. If we copy it, we can't reset corresponding bit in
1264 * dirty_bitmap as there may be some "dirty" bytes still not
1265 * copied.
1266 * 2. It's already dirty, so skipping it we don't diverge mirror
1267 * progress.
1268 *
1269 * Note, that because of this, guest write may have no contribution
1270 * into mirror converge, but that's not bad, as we have background
1271 * process of mirroring. If under some bad circumstances (high guest
1272 * IO load) background process starve, we will not converge anyway,
1273 * even if each write will contribute, as guest is not guaranteed to
1274 * rewrite the whole disk.
1275 */
1276 qiov_offset = QEMU_ALIGN_UP(offset, job->granularity) - offset;
1277 if (bytes <= qiov_offset) {
1278 /* nothing to do after shrink */
1279 return;
1280 }
1281 offset += qiov_offset;
1282 bytes -= qiov_offset;
1283 }
1284
1285 if (!QEMU_IS_ALIGNED(offset + bytes, job->granularity) &&
1286 bdrv_dirty_bitmap_get(job->dirty_bitmap, offset + bytes - 1))
1287 {
1288 uint64_t tail = (offset + bytes) % job->granularity;
1289
1290 if (bytes <= tail) {
1291 /* nothing to do after shrink */
1292 return;
1293 }
1294 bytes -= tail;
1295 }
1296
1297 /*
1298 * Tails are either clean or shrunk, so for bitmap resetting
1299 * we safely align the range down.
1300 */
1301 bitmap_offset = QEMU_ALIGN_UP(offset, job->granularity);
1302 bitmap_end = QEMU_ALIGN_DOWN(offset + bytes, job->granularity);
1303 if (bitmap_offset < bitmap_end) {
1304 bdrv_reset_dirty_bitmap(job->dirty_bitmap, bitmap_offset,
1305 bitmap_end - bitmap_offset);
1306 }
d06107ad 1307
5c511ac3 1308 job_progress_increase_remaining(&job->common.job, bytes);
d06107ad 1309
5c511ac3
VSO
1310 switch (method) {
1311 case MIRROR_METHOD_COPY:
dbdf699c
VSO
1312 ret = blk_co_pwritev_part(job->target, offset, bytes,
1313 qiov, qiov_offset, flags);
5c511ac3 1314 break;
d06107ad 1315
5c511ac3
VSO
1316 case MIRROR_METHOD_ZERO:
1317 assert(!qiov);
1318 ret = blk_co_pwrite_zeroes(job->target, offset, bytes, flags);
1319 break;
d06107ad 1320
5c511ac3
VSO
1321 case MIRROR_METHOD_DISCARD:
1322 assert(!qiov);
1323 ret = blk_co_pdiscard(job->target, offset, bytes);
1324 break;
d06107ad 1325
5c511ac3
VSO
1326 default:
1327 abort();
1328 }
d06107ad 1329
5c511ac3
VSO
1330 if (ret >= 0) {
1331 job_progress_update(&job->common.job, bytes);
1332 } else {
1333 BlockErrorAction action;
d06107ad 1334
dbdf699c
VSO
1335 /*
1336 * We failed, so we should mark dirty the whole area, aligned up.
1337 * Note that we don't care about shrunk tails if any: they were dirty
1338 * at function start, and they must be still dirty, as we've locked
1339 * the region for in-flight op.
1340 */
1341 bitmap_offset = QEMU_ALIGN_DOWN(offset, job->granularity);
1342 bitmap_end = QEMU_ALIGN_UP(offset + bytes, job->granularity);
1343 bdrv_set_dirty_bitmap(job->dirty_bitmap, bitmap_offset,
1344 bitmap_end - bitmap_offset);
5c511ac3 1345 job->actively_synced = false;
d06107ad 1346
5c511ac3
VSO
1347 action = mirror_error_action(job, false, -ret);
1348 if (action == BLOCK_ERROR_ACTION_REPORT) {
1349 if (!job->ret) {
1350 job->ret = ret;
d06107ad
HR
1351 }
1352 }
d06107ad
HR
1353 }
1354}
1355
1356static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s,
1357 uint64_t offset,
1358 uint64_t bytes)
1359{
1360 MirrorOp *op;
1361 uint64_t start_chunk = offset / s->granularity;
1362 uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity);
1363
1364 op = g_new(MirrorOp, 1);
1365 *op = (MirrorOp){
1366 .s = s,
1367 .offset = offset,
1368 .bytes = bytes,
1369 .is_active_write = true,
ce8cabbd 1370 .is_in_flight = true,
ead3f1bf 1371 .co = qemu_coroutine_self(),
d06107ad
HR
1372 };
1373 qemu_co_queue_init(&op->waiting_requests);
1374 QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next);
1375
1376 s->in_active_write_counter++;
1377
1378 mirror_wait_on_conflicts(op, s, offset, bytes);
1379
1380 bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk);
1381
1382 return op;
1383}
1384
1385static void coroutine_fn active_write_settle(MirrorOp *op)
1386{
1387 uint64_t start_chunk = op->offset / op->s->granularity;
1388 uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes,
1389 op->s->granularity);
1390
1391 if (!--op->s->in_active_write_counter && op->s->actively_synced) {
1392 BdrvChild *source = op->s->mirror_top_bs->backing;
1393
1394 if (QLIST_FIRST(&source->bs->parents) == source &&
1395 QLIST_NEXT(source, next_parent) == NULL)
1396 {
1397 /* Assert that we are back in sync once all active write
1398 * operations are settled.
1399 * Note that we can only assert this if the mirror node
1400 * is the source node's only parent. */
1401 assert(!bdrv_get_dirty_count(op->s->dirty_bitmap));
1402 }
1403 }
1404 bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk);
1405 QTAILQ_REMOVE(&op->s->ops_in_flight, op, next);
1406 qemu_co_queue_restart_all(&op->waiting_requests);
1407 g_free(op);
1408}
1409
4ef85a9c 1410static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs,
f7ef38dd 1411 int64_t offset, int64_t bytes, QEMUIOVector *qiov, BdrvRequestFlags flags)
4ef85a9c
KW
1412{
1413 return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags);
1414}
1415
d06107ad
HR
1416static int coroutine_fn bdrv_mirror_top_do_write(BlockDriverState *bs,
1417 MirrorMethod method, uint64_t offset, uint64_t bytes, QEMUIOVector *qiov,
1418 int flags)
1419{
1420 MirrorOp *op = NULL;
1421 MirrorBDSOpaque *s = bs->opaque;
1422 int ret = 0;
1423 bool copy_to_target;
1424
1425 copy_to_target = s->job->ret >= 0 &&
9b230ef9 1426 !job_is_cancelled(&s->job->common.job) &&
d06107ad
HR
1427 s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING;
1428
1429 if (copy_to_target) {
1430 op = active_write_prepare(s->job, offset, bytes);
1431 }
1432
1433 switch (method) {
1434 case MIRROR_METHOD_COPY:
1435 ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags);
1436 break;
1437
1438 case MIRROR_METHOD_ZERO:
1439 ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags);
1440 break;
1441
1442 case MIRROR_METHOD_DISCARD:
0b9fd3f4 1443 ret = bdrv_co_pdiscard(bs->backing, offset, bytes);
d06107ad
HR
1444 break;
1445
1446 default:
1447 abort();
1448 }
1449
1450 if (ret < 0) {
1451 goto out;
1452 }
1453
1454 if (copy_to_target) {
1455 do_sync_target_write(s->job, method, offset, bytes, qiov, flags);
1456 }
1457
1458out:
1459 if (copy_to_target) {
1460 active_write_settle(op);
1461 }
1462 return ret;
1463}
1464
4ef85a9c 1465static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs,
e75abeda 1466 int64_t offset, int64_t bytes, QEMUIOVector *qiov, BdrvRequestFlags flags)
4ef85a9c 1467{
d06107ad
HR
1468 MirrorBDSOpaque *s = bs->opaque;
1469 QEMUIOVector bounce_qiov;
1470 void *bounce_buf;
1471 int ret = 0;
1472 bool copy_to_target;
1473
1474 copy_to_target = s->job->ret >= 0 &&
9b230ef9 1475 !job_is_cancelled(&s->job->common.job) &&
d06107ad
HR
1476 s->job->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING;
1477
1478 if (copy_to_target) {
1479 /* The guest might concurrently modify the data to write; but
1480 * the data on source and destination must match, so we have
1481 * to use a bounce buffer if we are going to write to the
1482 * target now. */
1483 bounce_buf = qemu_blockalign(bs, bytes);
1484 iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes);
1485
1486 qemu_iovec_init(&bounce_qiov, 1);
1487 qemu_iovec_add(&bounce_qiov, bounce_buf, bytes);
1488 qiov = &bounce_qiov;
1489 }
1490
1491 ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, offset, bytes, qiov,
1492 flags);
1493
1494 if (copy_to_target) {
1495 qemu_iovec_destroy(&bounce_qiov);
1496 qemu_vfree(bounce_buf);
1497 }
1498
1499 return ret;
4ef85a9c
KW
1500}
1501
1502static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs)
1503{
ce960aa9
VSO
1504 if (bs->backing == NULL) {
1505 /* we can be here after failed bdrv_append in mirror_start_job */
1506 return 0;
1507 }
4ef85a9c
KW
1508 return bdrv_co_flush(bs->backing->bs);
1509}
1510
4ef85a9c 1511static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs,
f34b2bcf 1512 int64_t offset, int64_t bytes, BdrvRequestFlags flags)
4ef85a9c 1513{
d06107ad
HR
1514 return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, offset, bytes, NULL,
1515 flags);
4ef85a9c
KW
1516}
1517
1518static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs,
0c802287 1519 int64_t offset, int64_t bytes)
4ef85a9c 1520{
d06107ad
HR
1521 return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, offset, bytes,
1522 NULL, 0);
4ef85a9c
KW
1523}
1524
998b3a1e 1525static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs)
fd4a6493 1526{
18775ff3
VSO
1527 if (bs->backing == NULL) {
1528 /* we can be here after failed bdrv_attach_child in
1529 * bdrv_set_backing_hd */
1530 return;
1531 }
fd4a6493
KW
1532 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
1533 bs->backing->bs->filename);
1534}
1535
4ef85a9c 1536static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c,
bf8e925e 1537 BdrvChildRole role,
e0995dc3 1538 BlockReopenQueue *reopen_queue,
4ef85a9c
KW
1539 uint64_t perm, uint64_t shared,
1540 uint64_t *nperm, uint64_t *nshared)
1541{
f94dc3b4
HR
1542 MirrorBDSOpaque *s = bs->opaque;
1543
1544 if (s->stop) {
1545 /*
1546 * If the job is to be stopped, we do not need to forward
1547 * anything to the real image.
1548 */
1549 *nperm = 0;
1550 *nshared = BLK_PERM_ALL;
1551 return;
1552 }
1553
53431b90
HR
1554 bdrv_default_perms(bs, c, role, reopen_queue,
1555 perm, shared, nperm, nshared);
4ef85a9c 1556
53431b90
HR
1557 if (s->is_commit) {
1558 /*
1559 * For commit jobs, we cannot take CONSISTENT_READ, because
1560 * that permission is unshared for everything above the base
1561 * node (except for filters on the base node).
1562 * We also have to force-share the WRITE permission, or
1563 * otherwise we would block ourselves at the base node (if
1564 * writes are blocked for a node, they are also blocked for
1565 * its backing file).
1566 * (We could also share RESIZE, because it may be needed for
1567 * the target if its size is less than the top node's; but
1568 * bdrv_default_perms_for_cow() automatically shares RESIZE
1569 * for backing nodes if WRITE is shared, so there is no need
1570 * to do it here.)
1571 */
1572 *nperm &= ~BLK_PERM_CONSISTENT_READ;
1573 *nshared |= BLK_PERM_WRITE;
1574 }
4ef85a9c
KW
1575}
1576
1577/* Dummy node that provides consistent read to its users without requiring it
1578 * from its backing file and that allows writes on the backing file chain. */
1579static BlockDriver bdrv_mirror_top = {
1580 .format_name = "mirror_top",
1581 .bdrv_co_preadv = bdrv_mirror_top_preadv,
1582 .bdrv_co_pwritev = bdrv_mirror_top_pwritev,
1583 .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes,
1584 .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard,
1585 .bdrv_co_flush = bdrv_mirror_top_flush,
fd4a6493 1586 .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename,
4ef85a9c 1587 .bdrv_child_perm = bdrv_mirror_top_child_perm,
6540fd15
HR
1588
1589 .is_filter = true,
046fd84f 1590 .filtered_child_is_backing = true,
4ef85a9c
KW
1591};
1592
cc19f177
VSO
1593static BlockJob *mirror_start_job(
1594 const char *job_id, BlockDriverState *bs,
47970dfb
JS
1595 int creation_flags, BlockDriverState *target,
1596 const char *replaces, int64_t speed,
1597 uint32_t granularity, int64_t buf_size,
274fccee 1598 BlockMirrorBackingMode backing_mode,
cdf3bc93 1599 bool zero_target,
09158f00
BC
1600 BlockdevOnError on_source_error,
1601 BlockdevOnError on_target_error,
0fc9f8ea 1602 bool unmap,
097310b5 1603 BlockCompletionFunc *cb,
51ccfa2d 1604 void *opaque,
09158f00 1605 const BlockJobDriver *driver,
b49f7ead 1606 bool is_none_mode, BlockDriverState *base,
51ccfa2d 1607 bool auto_complete, const char *filter_node_name,
481debaa 1608 bool is_mirror, MirrorCopyMode copy_mode,
51ccfa2d 1609 Error **errp)
893f7eba
PB
1610{
1611 MirrorBlockJob *s;
429076e8 1612 MirrorBDSOpaque *bs_opaque;
4ef85a9c 1613 BlockDriverState *mirror_top_bs;
4ef85a9c 1614 bool target_is_backing;
3f072a7f 1615 uint64_t target_perms, target_shared_perms;
d7086422 1616 int ret;
893f7eba 1617
eee13dfe 1618 if (granularity == 0) {
341ebc2f 1619 granularity = bdrv_get_default_bitmap_granularity(target);
eee13dfe
PB
1620 }
1621
31826642 1622 assert(is_power_of_2(granularity));
eee13dfe 1623
48ac0a4d
WC
1624 if (buf_size < 0) {
1625 error_setg(errp, "Invalid parameter 'buf-size'");
cc19f177 1626 return NULL;
48ac0a4d
WC
1627 }
1628
1629 if (buf_size == 0) {
1630 buf_size = DEFAULT_MIRROR_BUF_SIZE;
1631 }
5bc361b8 1632
3f072a7f 1633 if (bdrv_skip_filters(bs) == bdrv_skip_filters(target)) {
86fae10c 1634 error_setg(errp, "Can't mirror node into itself");
cc19f177 1635 return NULL;
86fae10c
KW
1636 }
1637
53431b90
HR
1638 target_is_backing = bdrv_chain_contains(bs, target);
1639
4ef85a9c
KW
1640 /* In the case of active commit, add dummy driver to provide consistent
1641 * reads on the top, while disabling it in the intermediate nodes, and make
1642 * the backing chain writable. */
6cdbceb1
KW
1643 mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name,
1644 BDRV_O_RDWR, errp);
4ef85a9c 1645 if (mirror_top_bs == NULL) {
cc19f177 1646 return NULL;
4ef85a9c 1647 }
d3c8c674
KW
1648 if (!filter_node_name) {
1649 mirror_top_bs->implicit = true;
1650 }
e5182c1c
HR
1651
1652 /* So that we can always drop this node */
1653 mirror_top_bs->never_freeze = true;
1654
4ef85a9c 1655 mirror_top_bs->total_sectors = bs->total_sectors;
228345bf 1656 mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED;
80f5c33f
KW
1657 mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
1658 BDRV_REQ_NO_FALLBACK;
429076e8
HR
1659 bs_opaque = g_new0(MirrorBDSOpaque, 1);
1660 mirror_top_bs->opaque = bs_opaque;
4ef85a9c 1661
53431b90
HR
1662 bs_opaque->is_commit = target_is_backing;
1663
4ef85a9c 1664 bdrv_drained_begin(bs);
934aee14 1665 ret = bdrv_append(mirror_top_bs, bs, errp);
4ef85a9c
KW
1666 bdrv_drained_end(bs);
1667
934aee14 1668 if (ret < 0) {
b2c2832c 1669 bdrv_unref(mirror_top_bs);
cc19f177 1670 return NULL;
b2c2832c
KW
1671 }
1672
4ef85a9c 1673 /* Make sure that the source is not resized while the job is running */
75859b94 1674 s = block_job_create(job_id, driver, NULL, mirror_top_bs,
4ef85a9c
KW
1675 BLK_PERM_CONSISTENT_READ,
1676 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
64631f36 1677 BLK_PERM_WRITE, speed,
c6cc12bf 1678 creation_flags, cb, opaque, errp);
893f7eba 1679 if (!s) {
4ef85a9c 1680 goto fail;
893f7eba 1681 }
429076e8
HR
1682 bs_opaque->job = s;
1683
7a25fcd0
HR
1684 /* The block job now has a reference to this node */
1685 bdrv_unref(mirror_top_bs);
1686
4ef85a9c
KW
1687 s->mirror_top_bs = mirror_top_bs;
1688
1689 /* No resize for the target either; while the mirror is still running, a
1690 * consistent read isn't necessarily possible. We could possibly allow
1691 * writes and graph modifications, though it would likely defeat the
1692 * purpose of a mirror, so leave them blocked for now.
1693 *
1694 * In the case of active commit, things look a bit different, though,
1695 * because the target is an already populated backing file in active use.
1696 * We can allow anything except resize there.*/
3f072a7f
HR
1697
1698 target_perms = BLK_PERM_WRITE;
1699 target_shared_perms = BLK_PERM_WRITE_UNCHANGED;
1700
3f072a7f
HR
1701 if (target_is_backing) {
1702 int64_t bs_size, target_size;
1703 bs_size = bdrv_getlength(bs);
1704 if (bs_size < 0) {
1705 error_setg_errno(errp, -bs_size,
1706 "Could not inquire top image size");
1707 goto fail;
1708 }
1709
1710 target_size = bdrv_getlength(target);
1711 if (target_size < 0) {
1712 error_setg_errno(errp, -target_size,
1713 "Could not inquire base image size");
1714 goto fail;
1715 }
1716
1717 if (target_size < bs_size) {
1718 target_perms |= BLK_PERM_RESIZE;
1719 }
1720
64631f36 1721 target_shared_perms |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE;
3f072a7f
HR
1722 } else if (bdrv_chain_contains(bs, bdrv_skip_filters(target))) {
1723 /*
1724 * We may want to allow this in the future, but it would
1725 * require taking some extra care.
1726 */
1727 error_setg(errp, "Cannot mirror to a filter on top of a node in the "
1728 "source's backing chain");
1729 goto fail;
1730 }
1731
d861ab3a 1732 s->target = blk_new(s->common.job.aio_context,
3f072a7f 1733 target_perms, target_shared_perms);
d7086422
KW
1734 ret = blk_insert_bs(s->target, target, errp);
1735 if (ret < 0) {
4ef85a9c 1736 goto fail;
d7086422 1737 }
045a2f82
FZ
1738 if (is_mirror) {
1739 /* XXX: Mirror target could be a NBD server of target QEMU in the case
1740 * of non-shared block migration. To allow migration completion, we
1741 * have to allow "inactivate" of the target BB. When that happens, we
1742 * know the job is drained, and the vcpus are stopped, so no write
1743 * operation will be performed. Block layer already has assertions to
1744 * ensure that. */
1745 blk_set_force_allow_inactivate(s->target);
1746 }
9ff7f0df 1747 blk_set_allow_aio_context_change(s->target, true);
cf312932 1748 blk_set_disable_request_queuing(s->target, true);
e253f4b8 1749
09158f00 1750 s->replaces = g_strdup(replaces);
b952b558
PB
1751 s->on_source_error = on_source_error;
1752 s->on_target_error = on_target_error;
03544a6e 1753 s->is_none_mode = is_none_mode;
274fccee 1754 s->backing_mode = backing_mode;
cdf3bc93 1755 s->zero_target = zero_target;
481debaa 1756 s->copy_mode = copy_mode;
5bc361b8 1757 s->base = base;
3f072a7f 1758 s->base_overlay = bdrv_find_overlay(bs, base);
eee13dfe 1759 s->granularity = granularity;
48ac0a4d 1760 s->buf_size = ROUND_UP(buf_size, granularity);
0fc9f8ea 1761 s->unmap = unmap;
b49f7ead
WC
1762 if (auto_complete) {
1763 s->should_complete = true;
1764 }
b812f671 1765
0db6e54a 1766 s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp);
b8afb520 1767 if (!s->dirty_bitmap) {
88f9d1b3 1768 goto fail;
b8afb520 1769 }
dbdf699c
VSO
1770 if (s->copy_mode == MIRROR_COPY_MODE_WRITE_BLOCKING) {
1771 bdrv_disable_dirty_bitmap(s->dirty_bitmap);
1772 }
10f3cd15 1773
67b24427
AG
1774 ret = block_job_add_bdrv(&s->common, "source", bs, 0,
1775 BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE |
1776 BLK_PERM_CONSISTENT_READ,
1777 errp);
1778 if (ret < 0) {
1779 goto fail;
1780 }
1781
4ef85a9c 1782 /* Required permissions are already taken with blk_new() */
76d554e2
KW
1783 block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL,
1784 &error_abort);
1785
f3ede4b0
AG
1786 /* In commit_active_start() all intermediate nodes disappear, so
1787 * any jobs in them must be blocked */
4ef85a9c 1788 if (target_is_backing) {
3f072a7f
HR
1789 BlockDriverState *iter, *filtered_target;
1790 uint64_t iter_shared_perms;
1791
1792 /*
1793 * The topmost node with
1794 * bdrv_skip_filters(filtered_target) == bdrv_skip_filters(target)
1795 */
1796 filtered_target = bdrv_cow_bs(bdrv_find_overlay(bs, target));
1797
1798 assert(bdrv_skip_filters(filtered_target) ==
1799 bdrv_skip_filters(target));
1800
1801 /*
1802 * XXX BLK_PERM_WRITE needs to be allowed so we don't block
1803 * ourselves at s->base (if writes are blocked for a node, they are
1804 * also blocked for its backing file). The other options would be a
1805 * second filter driver above s->base (== target).
1806 */
1807 iter_shared_perms = BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE;
1808
1809 for (iter = bdrv_filter_or_cow_bs(bs); iter != target;
1810 iter = bdrv_filter_or_cow_bs(iter))
1811 {
1812 if (iter == filtered_target) {
1813 /*
1814 * From here on, all nodes are filters on the base.
1815 * This allows us to share BLK_PERM_CONSISTENT_READ.
1816 */
1817 iter_shared_perms |= BLK_PERM_CONSISTENT_READ;
1818 }
1819
4ef85a9c 1820 ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
3f072a7f 1821 iter_shared_perms, errp);
4ef85a9c
KW
1822 if (ret < 0) {
1823 goto fail;
1824 }
f3ede4b0 1825 }
ef53dc09
AG
1826
1827 if (bdrv_freeze_backing_chain(mirror_top_bs, target, errp) < 0) {
1828 goto fail;
1829 }
f3ede4b0 1830 }
10f3cd15 1831
12aa4082
HR
1832 QTAILQ_INIT(&s->ops_in_flight);
1833
5ccac6f1 1834 trace_mirror_start(bs, s, opaque);
da01ff7f 1835 job_start(&s->common.job);
cc19f177
VSO
1836
1837 return &s->common;
4ef85a9c
KW
1838
1839fail:
1840 if (s) {
7a25fcd0
HR
1841 /* Make sure this BDS does not go away until we have completed the graph
1842 * changes below */
1843 bdrv_ref(mirror_top_bs);
1844
4ef85a9c
KW
1845 g_free(s->replaces);
1846 blk_unref(s->target);
429076e8 1847 bs_opaque->job = NULL;
e917e2cb 1848 if (s->dirty_bitmap) {
5deb6cbd 1849 bdrv_release_dirty_bitmap(s->dirty_bitmap);
e917e2cb 1850 }
4ad35181 1851 job_early_fail(&s->common.job);
4ef85a9c
KW
1852 }
1853
f94dc3b4
HR
1854 bs_opaque->stop = true;
1855 bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing,
1856 &error_abort);
3f072a7f 1857 bdrv_replace_node(mirror_top_bs, mirror_top_bs->backing->bs, &error_abort);
7a25fcd0
HR
1858
1859 bdrv_unref(mirror_top_bs);
cc19f177
VSO
1860
1861 return NULL;
893f7eba 1862}
03544a6e 1863
71aa9867
AG
1864void mirror_start(const char *job_id, BlockDriverState *bs,
1865 BlockDriverState *target, const char *replaces,
a1999b33
JS
1866 int creation_flags, int64_t speed,
1867 uint32_t granularity, int64_t buf_size,
274fccee 1868 MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
cdf3bc93 1869 bool zero_target,
274fccee 1870 BlockdevOnError on_source_error,
03544a6e 1871 BlockdevOnError on_target_error,
481debaa
HR
1872 bool unmap, const char *filter_node_name,
1873 MirrorCopyMode copy_mode, Error **errp)
03544a6e
FZ
1874{
1875 bool is_none_mode;
1876 BlockDriverState *base;
1877
b4ad82aa
EGE
1878 GLOBAL_STATE_CODE();
1879
c8b56501
JS
1880 if ((mode == MIRROR_SYNC_MODE_INCREMENTAL) ||
1881 (mode == MIRROR_SYNC_MODE_BITMAP)) {
1882 error_setg(errp, "Sync mode '%s' not supported",
1883 MirrorSyncMode_str(mode));
d58d8453
JS
1884 return;
1885 }
03544a6e 1886 is_none_mode = mode == MIRROR_SYNC_MODE_NONE;
3f072a7f 1887 base = mode == MIRROR_SYNC_MODE_TOP ? bdrv_backing_chain_next(bs) : NULL;
a1999b33 1888 mirror_start_job(job_id, bs, creation_flags, target, replaces,
cdf3bc93 1889 speed, granularity, buf_size, backing_mode, zero_target,
51ccfa2d 1890 on_source_error, on_target_error, unmap, NULL, NULL,
6cdbceb1 1891 &mirror_job_driver, is_none_mode, base, false,
481debaa 1892 filter_node_name, true, copy_mode, errp);
03544a6e
FZ
1893}
1894
cc19f177
VSO
1895BlockJob *commit_active_start(const char *job_id, BlockDriverState *bs,
1896 BlockDriverState *base, int creation_flags,
1897 int64_t speed, BlockdevOnError on_error,
1898 const char *filter_node_name,
1899 BlockCompletionFunc *cb, void *opaque,
1900 bool auto_complete, Error **errp)
03544a6e 1901{
1ba79388 1902 bool base_read_only;
eb5becc1 1903 BlockJob *job;
4da83585 1904
b4ad82aa
EGE
1905 GLOBAL_STATE_CODE();
1906
1ba79388 1907 base_read_only = bdrv_is_read_only(base);
4da83585 1908
1ba79388
AG
1909 if (base_read_only) {
1910 if (bdrv_reopen_set_read_only(base, false, errp) < 0) {
cc19f177 1911 return NULL;
1ba79388 1912 }
20a63d2c 1913 }
4da83585 1914
eb5becc1 1915 job = mirror_start_job(
cc19f177 1916 job_id, bs, creation_flags, base, NULL, speed, 0, 0,
cdf3bc93 1917 MIRROR_LEAVE_BACKING_CHAIN, false,
51ccfa2d 1918 on_error, on_error, true, cb, opaque,
6cdbceb1 1919 &commit_active_job_driver, false, base, auto_complete,
481debaa 1920 filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND,
eb5becc1
VSO
1921 errp);
1922 if (!job) {
4da83585
JC
1923 goto error_restore_flags;
1924 }
1925
eb5becc1 1926 return job;
4da83585
JC
1927
1928error_restore_flags:
1929 /* ignore error and errp for bdrv_reopen, because we want to propagate
1930 * the original error */
1ba79388
AG
1931 if (base_read_only) {
1932 bdrv_reopen_set_read_only(base, true, NULL);
1933 }
cc19f177 1934 return NULL;
03544a6e 1935}