]>
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" |
893f7eba | 15 | #include "trace.h" |
c87621ea | 16 | #include "block/blockjob_int.h" |
737e150e | 17 | #include "block/block_int.h" |
373340b2 | 18 | #include "sysemu/block-backend.h" |
da34e65c | 19 | #include "qapi/error.h" |
cc7a8ea7 | 20 | #include "qapi/qmp/qerror.h" |
893f7eba | 21 | #include "qemu/ratelimit.h" |
b812f671 | 22 | #include "qemu/bitmap.h" |
893f7eba | 23 | |
402a4741 PB |
24 | #define SLICE_TIME 100000000ULL /* ns */ |
25 | #define MAX_IN_FLIGHT 16 | |
0965a41e VSO |
26 | #define MAX_IO_SECTORS ((1 << 20) >> BDRV_SECTOR_BITS) /* 1 Mb */ |
27 | #define DEFAULT_MIRROR_BUF_SIZE \ | |
28 | (MAX_IN_FLIGHT * MAX_IO_SECTORS * BDRV_SECTOR_SIZE) | |
402a4741 PB |
29 | |
30 | /* The mirroring buffer is a list of granularity-sized chunks. | |
31 | * Free chunks are organized in a list. | |
32 | */ | |
33 | typedef struct MirrorBuffer { | |
34 | QSIMPLEQ_ENTRY(MirrorBuffer) next; | |
35 | } MirrorBuffer; | |
893f7eba PB |
36 | |
37 | typedef struct MirrorBlockJob { | |
38 | BlockJob common; | |
39 | RateLimit limit; | |
e253f4b8 | 40 | BlockBackend *target; |
5bc361b8 | 41 | BlockDriverState *base; |
09158f00 BC |
42 | /* The name of the graph node to replace */ |
43 | char *replaces; | |
44 | /* The BDS to replace */ | |
45 | BlockDriverState *to_replace; | |
46 | /* Used to block operations on the drive-mirror-replace target */ | |
47 | Error *replace_blocker; | |
03544a6e | 48 | bool is_none_mode; |
274fccee | 49 | BlockMirrorBackingMode backing_mode; |
b952b558 | 50 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 PB |
51 | bool synced; |
52 | bool should_complete; | |
eee13dfe | 53 | int64_t granularity; |
b812f671 | 54 | size_t buf_size; |
b21c7652 | 55 | int64_t bdev_length; |
b812f671 | 56 | unsigned long *cow_bitmap; |
e4654d2d | 57 | BdrvDirtyBitmap *dirty_bitmap; |
dc162c8e | 58 | BdrvDirtyBitmapIter *dbi; |
893f7eba | 59 | uint8_t *buf; |
402a4741 PB |
60 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
61 | int buf_free_count; | |
bd48bde8 | 62 | |
49efb1f5 | 63 | uint64_t last_pause_ns; |
402a4741 | 64 | unsigned long *in_flight_bitmap; |
bd48bde8 | 65 | int in_flight; |
531509ba | 66 | int64_t sectors_in_flight; |
bd48bde8 | 67 | int ret; |
0fc9f8ea | 68 | bool unmap; |
e424aff5 | 69 | bool waiting_for_io; |
e5b43573 FZ |
70 | int target_cluster_sectors; |
71 | int max_iov; | |
893f7eba PB |
72 | } MirrorBlockJob; |
73 | ||
bd48bde8 PB |
74 | typedef struct MirrorOp { |
75 | MirrorBlockJob *s; | |
76 | QEMUIOVector qiov; | |
bd48bde8 PB |
77 | int64_t sector_num; |
78 | int nb_sectors; | |
79 | } MirrorOp; | |
80 | ||
b952b558 PB |
81 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
82 | int error) | |
83 | { | |
84 | s->synced = false; | |
85 | if (read) { | |
81e254dc KW |
86 | return block_job_error_action(&s->common, s->on_source_error, |
87 | true, error); | |
b952b558 | 88 | } else { |
81e254dc KW |
89 | return block_job_error_action(&s->common, s->on_target_error, |
90 | false, error); | |
b952b558 PB |
91 | } |
92 | } | |
93 | ||
bd48bde8 PB |
94 | static void mirror_iteration_done(MirrorOp *op, int ret) |
95 | { | |
96 | MirrorBlockJob *s = op->s; | |
402a4741 | 97 | struct iovec *iov; |
bd48bde8 | 98 | int64_t chunk_num; |
402a4741 | 99 | int i, nb_chunks, sectors_per_chunk; |
bd48bde8 PB |
100 | |
101 | trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret); | |
102 | ||
103 | s->in_flight--; | |
b21c7652 | 104 | s->sectors_in_flight -= op->nb_sectors; |
402a4741 PB |
105 | iov = op->qiov.iov; |
106 | for (i = 0; i < op->qiov.niov; i++) { | |
107 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
108 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
109 | s->buf_free_count++; | |
110 | } | |
111 | ||
bd48bde8 PB |
112 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
113 | chunk_num = op->sector_num / sectors_per_chunk; | |
4150ae60 | 114 | nb_chunks = DIV_ROUND_UP(op->nb_sectors, sectors_per_chunk); |
402a4741 | 115 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
b21c7652 HR |
116 | if (ret >= 0) { |
117 | if (s->cow_bitmap) { | |
118 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
119 | } | |
120 | s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
121 | } |
122 | ||
6df3bf8e | 123 | qemu_iovec_destroy(&op->qiov); |
c84b3192 | 124 | g_free(op); |
7b770c72 | 125 | |
e424aff5 | 126 | if (s->waiting_for_io) { |
0b8b8753 | 127 | qemu_coroutine_enter(s->common.co); |
7b770c72 | 128 | } |
bd48bde8 PB |
129 | } |
130 | ||
131 | static void mirror_write_complete(void *opaque, int ret) | |
132 | { | |
133 | MirrorOp *op = opaque; | |
134 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
135 | |
136 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 137 | if (ret < 0) { |
bd48bde8 PB |
138 | BlockErrorAction action; |
139 | ||
20dca810 | 140 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 141 | action = mirror_error_action(s, false, -ret); |
a589569f | 142 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
143 | s->ret = ret; |
144 | } | |
145 | } | |
146 | mirror_iteration_done(op, ret); | |
b9e413dd | 147 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
148 | } |
149 | ||
150 | static void mirror_read_complete(void *opaque, int ret) | |
151 | { | |
152 | MirrorOp *op = opaque; | |
153 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
154 | |
155 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 156 | if (ret < 0) { |
bd48bde8 PB |
157 | BlockErrorAction action; |
158 | ||
20dca810 | 159 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 160 | action = mirror_error_action(s, true, -ret); |
a589569f | 161 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
162 | s->ret = ret; |
163 | } | |
164 | ||
165 | mirror_iteration_done(op, ret); | |
b9e413dd PB |
166 | } else { |
167 | blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov, | |
168 | 0, mirror_write_complete, op); | |
bd48bde8 | 169 | } |
b9e413dd | 170 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
171 | } |
172 | ||
4150ae60 FZ |
173 | static inline void mirror_clip_sectors(MirrorBlockJob *s, |
174 | int64_t sector_num, | |
175 | int *nb_sectors) | |
176 | { | |
177 | *nb_sectors = MIN(*nb_sectors, | |
178 | s->bdev_length / BDRV_SECTOR_SIZE - sector_num); | |
179 | } | |
180 | ||
e5b43573 FZ |
181 | /* Round sector_num and/or nb_sectors to target cluster if COW is needed, and |
182 | * return the offset of the adjusted tail sector against original. */ | |
183 | static int mirror_cow_align(MirrorBlockJob *s, | |
184 | int64_t *sector_num, | |
185 | int *nb_sectors) | |
893f7eba | 186 | { |
e5b43573 FZ |
187 | bool need_cow; |
188 | int ret = 0; | |
189 | int chunk_sectors = s->granularity >> BDRV_SECTOR_BITS; | |
190 | int64_t align_sector_num = *sector_num; | |
191 | int align_nb_sectors = *nb_sectors; | |
192 | int max_sectors = chunk_sectors * s->max_iov; | |
193 | ||
194 | need_cow = !test_bit(*sector_num / chunk_sectors, s->cow_bitmap); | |
195 | need_cow |= !test_bit((*sector_num + *nb_sectors - 1) / chunk_sectors, | |
196 | s->cow_bitmap); | |
197 | if (need_cow) { | |
244483e6 KW |
198 | bdrv_round_sectors_to_clusters(blk_bs(s->target), *sector_num, |
199 | *nb_sectors, &align_sector_num, | |
200 | &align_nb_sectors); | |
e5b43573 | 201 | } |
3515727f | 202 | |
e5b43573 FZ |
203 | if (align_nb_sectors > max_sectors) { |
204 | align_nb_sectors = max_sectors; | |
205 | if (need_cow) { | |
206 | align_nb_sectors = QEMU_ALIGN_DOWN(align_nb_sectors, | |
207 | s->target_cluster_sectors); | |
208 | } | |
8f0720ec | 209 | } |
4150ae60 FZ |
210 | /* Clipping may result in align_nb_sectors unaligned to chunk boundary, but |
211 | * that doesn't matter because it's already the end of source image. */ | |
212 | mirror_clip_sectors(s, align_sector_num, &align_nb_sectors); | |
8f0720ec | 213 | |
e5b43573 FZ |
214 | ret = align_sector_num + align_nb_sectors - (*sector_num + *nb_sectors); |
215 | *sector_num = align_sector_num; | |
216 | *nb_sectors = align_nb_sectors; | |
217 | assert(ret >= 0); | |
218 | return ret; | |
219 | } | |
220 | ||
21cd917f FZ |
221 | static inline void mirror_wait_for_io(MirrorBlockJob *s) |
222 | { | |
223 | assert(!s->waiting_for_io); | |
224 | s->waiting_for_io = true; | |
225 | qemu_coroutine_yield(); | |
226 | s->waiting_for_io = false; | |
227 | } | |
228 | ||
e5b43573 | 229 | /* Submit async read while handling COW. |
17612955 JS |
230 | * Returns: The number of sectors copied after and including sector_num, |
231 | * excluding any sectors copied prior to sector_num due to alignment. | |
232 | * This will be nb_sectors if no alignment is necessary, or | |
e5b43573 FZ |
233 | * (new_end - sector_num) if tail is rounded up or down due to |
234 | * alignment or buffer limit. | |
235 | */ | |
236 | static int mirror_do_read(MirrorBlockJob *s, int64_t sector_num, | |
237 | int nb_sectors) | |
238 | { | |
e253f4b8 | 239 | BlockBackend *source = s->common.blk; |
e5b43573 | 240 | int sectors_per_chunk, nb_chunks; |
17612955 | 241 | int ret; |
e5b43573 | 242 | MirrorOp *op; |
e4808881 | 243 | int max_sectors; |
e5b43573 | 244 | |
884fea4e | 245 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
e4808881 | 246 | max_sectors = sectors_per_chunk * s->max_iov; |
402a4741 | 247 | |
e5b43573 FZ |
248 | /* We can only handle as much as buf_size at a time. */ |
249 | nb_sectors = MIN(s->buf_size >> BDRV_SECTOR_BITS, nb_sectors); | |
e4808881 | 250 | nb_sectors = MIN(max_sectors, nb_sectors); |
e5b43573 | 251 | assert(nb_sectors); |
17612955 | 252 | ret = nb_sectors; |
402a4741 | 253 | |
e5b43573 FZ |
254 | if (s->cow_bitmap) { |
255 | ret += mirror_cow_align(s, §or_num, &nb_sectors); | |
256 | } | |
257 | assert(nb_sectors << BDRV_SECTOR_BITS <= s->buf_size); | |
258 | /* The sector range must meet granularity because: | |
259 | * 1) Caller passes in aligned values; | |
260 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
e5b43573 | 261 | assert(!(sector_num % sectors_per_chunk)); |
4150ae60 | 262 | nb_chunks = DIV_ROUND_UP(nb_sectors, sectors_per_chunk); |
e5b43573 FZ |
263 | |
264 | while (s->buf_free_count < nb_chunks) { | |
402a4741 | 265 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); |
21cd917f | 266 | mirror_wait_for_io(s); |
b812f671 PB |
267 | } |
268 | ||
bd48bde8 | 269 | /* Allocate a MirrorOp that is used as an AIO callback. */ |
c84b3192 | 270 | op = g_new(MirrorOp, 1); |
bd48bde8 | 271 | op->s = s; |
bd48bde8 PB |
272 | op->sector_num = sector_num; |
273 | op->nb_sectors = nb_sectors; | |
402a4741 PB |
274 | |
275 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
276 | * from s->buf_free. | |
277 | */ | |
278 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
279 | while (nb_chunks-- > 0) { |
280 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
e5b43573 | 281 | size_t remaining = nb_sectors * BDRV_SECTOR_SIZE - op->qiov.size; |
5a0f6fd5 | 282 | |
402a4741 PB |
283 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
284 | s->buf_free_count--; | |
5a0f6fd5 | 285 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 286 | } |
bd48bde8 | 287 | |
893f7eba | 288 | /* Copy the dirty cluster. */ |
bd48bde8 | 289 | s->in_flight++; |
b21c7652 | 290 | s->sectors_in_flight += nb_sectors; |
b812f671 | 291 | trace_mirror_one_iteration(s, sector_num, nb_sectors); |
dcfb3beb | 292 | |
73698c30 | 293 | blk_aio_preadv(source, sector_num * BDRV_SECTOR_SIZE, &op->qiov, 0, |
e5b43573 FZ |
294 | mirror_read_complete, op); |
295 | return ret; | |
296 | } | |
297 | ||
298 | static void mirror_do_zero_or_discard(MirrorBlockJob *s, | |
299 | int64_t sector_num, | |
300 | int nb_sectors, | |
301 | bool is_discard) | |
302 | { | |
303 | MirrorOp *op; | |
304 | ||
305 | /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed | |
306 | * so the freeing in mirror_iteration_done is nop. */ | |
307 | op = g_new0(MirrorOp, 1); | |
308 | op->s = s; | |
309 | op->sector_num = sector_num; | |
310 | op->nb_sectors = nb_sectors; | |
311 | ||
312 | s->in_flight++; | |
313 | s->sectors_in_flight += nb_sectors; | |
314 | if (is_discard) { | |
1c6c4bb7 EB |
315 | blk_aio_pdiscard(s->target, sector_num << BDRV_SECTOR_BITS, |
316 | op->nb_sectors << BDRV_SECTOR_BITS, | |
317 | mirror_write_complete, op); | |
e5b43573 | 318 | } else { |
e253f4b8 KW |
319 | blk_aio_pwrite_zeroes(s->target, sector_num * BDRV_SECTOR_SIZE, |
320 | op->nb_sectors * BDRV_SECTOR_SIZE, | |
dcfb3beb FZ |
321 | s->unmap ? BDRV_REQ_MAY_UNMAP : 0, |
322 | mirror_write_complete, op); | |
e5b43573 FZ |
323 | } |
324 | } | |
325 | ||
326 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) | |
327 | { | |
e253f4b8 | 328 | BlockDriverState *source = blk_bs(s->common.blk); |
9c83625b | 329 | int64_t sector_num, first_chunk; |
e5b43573 FZ |
330 | uint64_t delay_ns = 0; |
331 | /* At least the first dirty chunk is mirrored in one iteration. */ | |
332 | int nb_chunks = 1; | |
333 | int64_t end = s->bdev_length / BDRV_SECTOR_SIZE; | |
334 | int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; | |
4b5004d9 | 335 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
0965a41e VSO |
336 | int max_io_sectors = MAX((s->buf_size >> BDRV_SECTOR_BITS) / MAX_IN_FLIGHT, |
337 | MAX_IO_SECTORS); | |
e5b43573 | 338 | |
dc162c8e | 339 | sector_num = bdrv_dirty_iter_next(s->dbi); |
e5b43573 | 340 | if (sector_num < 0) { |
dc162c8e FZ |
341 | bdrv_set_dirty_iter(s->dbi, 0); |
342 | sector_num = bdrv_dirty_iter_next(s->dbi); | |
e5b43573 FZ |
343 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
344 | assert(sector_num >= 0); | |
345 | } | |
346 | ||
9c83625b HR |
347 | first_chunk = sector_num / sectors_per_chunk; |
348 | while (test_bit(first_chunk, s->in_flight_bitmap)) { | |
ff04198b | 349 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); |
9c83625b HR |
350 | mirror_wait_for_io(s); |
351 | } | |
352 | ||
565ac01f SH |
353 | block_job_pause_point(&s->common); |
354 | ||
e5b43573 FZ |
355 | /* Find the number of consective dirty chunks following the first dirty |
356 | * one, and wait for in flight requests in them. */ | |
357 | while (nb_chunks * sectors_per_chunk < (s->buf_size >> BDRV_SECTOR_BITS)) { | |
dc162c8e | 358 | int64_t next_dirty; |
e5b43573 FZ |
359 | int64_t next_sector = sector_num + nb_chunks * sectors_per_chunk; |
360 | int64_t next_chunk = next_sector / sectors_per_chunk; | |
361 | if (next_sector >= end || | |
362 | !bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) { | |
363 | break; | |
364 | } | |
365 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 366 | break; |
e5b43573 | 367 | } |
9c83625b | 368 | |
dc162c8e FZ |
369 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
370 | if (next_dirty > next_sector || next_dirty < 0) { | |
f27a2742 | 371 | /* The bitmap iterator's cache is stale, refresh it */ |
dc162c8e FZ |
372 | bdrv_set_dirty_iter(s->dbi, next_sector); |
373 | next_dirty = bdrv_dirty_iter_next(s->dbi); | |
f27a2742 | 374 | } |
dc162c8e | 375 | assert(next_dirty == next_sector); |
9c83625b | 376 | nb_chunks++; |
e5b43573 FZ |
377 | } |
378 | ||
379 | /* Clear dirty bits before querying the block status, because | |
380 | * calling bdrv_get_block_status_above could yield - if some blocks are | |
381 | * marked dirty in this window, we need to know. | |
382 | */ | |
383 | bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, | |
384 | nb_chunks * sectors_per_chunk); | |
385 | bitmap_set(s->in_flight_bitmap, sector_num / sectors_per_chunk, nb_chunks); | |
386 | while (nb_chunks > 0 && sector_num < end) { | |
387 | int ret; | |
4b5004d9 | 388 | int io_sectors, io_sectors_acct; |
e5b43573 FZ |
389 | BlockDriverState *file; |
390 | enum MirrorMethod { | |
391 | MIRROR_METHOD_COPY, | |
392 | MIRROR_METHOD_ZERO, | |
393 | MIRROR_METHOD_DISCARD | |
394 | } mirror_method = MIRROR_METHOD_COPY; | |
395 | ||
396 | assert(!(sector_num % sectors_per_chunk)); | |
397 | ret = bdrv_get_block_status_above(source, NULL, sector_num, | |
398 | nb_chunks * sectors_per_chunk, | |
399 | &io_sectors, &file); | |
400 | if (ret < 0) { | |
0965a41e VSO |
401 | io_sectors = MIN(nb_chunks * sectors_per_chunk, max_io_sectors); |
402 | } else if (ret & BDRV_BLOCK_DATA) { | |
403 | io_sectors = MIN(io_sectors, max_io_sectors); | |
e5b43573 FZ |
404 | } |
405 | ||
406 | io_sectors -= io_sectors % sectors_per_chunk; | |
407 | if (io_sectors < sectors_per_chunk) { | |
408 | io_sectors = sectors_per_chunk; | |
409 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { | |
410 | int64_t target_sector_num; | |
411 | int target_nb_sectors; | |
244483e6 KW |
412 | bdrv_round_sectors_to_clusters(blk_bs(s->target), sector_num, |
413 | io_sectors, &target_sector_num, | |
414 | &target_nb_sectors); | |
e5b43573 FZ |
415 | if (target_sector_num == sector_num && |
416 | target_nb_sectors == io_sectors) { | |
417 | mirror_method = ret & BDRV_BLOCK_ZERO ? | |
418 | MIRROR_METHOD_ZERO : | |
419 | MIRROR_METHOD_DISCARD; | |
420 | } | |
421 | } | |
422 | ||
cf56a3c6 DL |
423 | while (s->in_flight >= MAX_IN_FLIGHT) { |
424 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); | |
425 | mirror_wait_for_io(s); | |
426 | } | |
427 | ||
dbaa7b57 VSO |
428 | if (s->ret < 0) { |
429 | return 0; | |
430 | } | |
431 | ||
4150ae60 | 432 | mirror_clip_sectors(s, sector_num, &io_sectors); |
e5b43573 FZ |
433 | switch (mirror_method) { |
434 | case MIRROR_METHOD_COPY: | |
435 | io_sectors = mirror_do_read(s, sector_num, io_sectors); | |
4b5004d9 | 436 | io_sectors_acct = io_sectors; |
e5b43573 FZ |
437 | break; |
438 | case MIRROR_METHOD_ZERO: | |
e5b43573 | 439 | case MIRROR_METHOD_DISCARD: |
4b5004d9 DL |
440 | mirror_do_zero_or_discard(s, sector_num, io_sectors, |
441 | mirror_method == MIRROR_METHOD_DISCARD); | |
442 | if (write_zeroes_ok) { | |
443 | io_sectors_acct = 0; | |
444 | } else { | |
445 | io_sectors_acct = io_sectors; | |
446 | } | |
e5b43573 FZ |
447 | break; |
448 | default: | |
449 | abort(); | |
450 | } | |
451 | assert(io_sectors); | |
452 | sector_num += io_sectors; | |
4150ae60 | 453 | nb_chunks -= DIV_ROUND_UP(io_sectors, sectors_per_chunk); |
f14a39cc | 454 | if (s->common.speed) { |
4b5004d9 | 455 | delay_ns = ratelimit_calculate_delay(&s->limit, io_sectors_acct); |
f14a39cc | 456 | } |
dcfb3beb | 457 | } |
cc8c9d6c | 458 | return delay_ns; |
bd48bde8 | 459 | } |
b952b558 | 460 | |
402a4741 PB |
461 | static void mirror_free_init(MirrorBlockJob *s) |
462 | { | |
463 | int granularity = s->granularity; | |
464 | size_t buf_size = s->buf_size; | |
465 | uint8_t *buf = s->buf; | |
466 | ||
467 | assert(s->buf_free_count == 0); | |
468 | QSIMPLEQ_INIT(&s->buf_free); | |
469 | while (buf_size != 0) { | |
470 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
471 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
472 | s->buf_free_count++; | |
473 | buf_size -= granularity; | |
474 | buf += granularity; | |
475 | } | |
476 | } | |
477 | ||
bae8196d PB |
478 | /* This is also used for the .pause callback. There is no matching |
479 | * mirror_resume() because mirror_run() will begin iterating again | |
480 | * when the job is resumed. | |
481 | */ | |
482 | static void mirror_wait_for_all_io(MirrorBlockJob *s) | |
bd48bde8 PB |
483 | { |
484 | while (s->in_flight > 0) { | |
21cd917f | 485 | mirror_wait_for_io(s); |
bd48bde8 | 486 | } |
893f7eba PB |
487 | } |
488 | ||
5a7e7a0b SH |
489 | typedef struct { |
490 | int ret; | |
491 | } MirrorExitData; | |
492 | ||
493 | static void mirror_exit(BlockJob *job, void *opaque) | |
494 | { | |
495 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
496 | MirrorExitData *data = opaque; | |
497 | AioContext *replace_aio_context = NULL; | |
e253f4b8 KW |
498 | BlockDriverState *src = blk_bs(s->common.blk); |
499 | BlockDriverState *target_bs = blk_bs(s->target); | |
3f09bfbc KW |
500 | |
501 | /* Make sure that the source BDS doesn't go away before we called | |
502 | * block_job_completed(). */ | |
503 | bdrv_ref(src); | |
5a7e7a0b SH |
504 | |
505 | if (s->to_replace) { | |
506 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
507 | aio_context_acquire(replace_aio_context); | |
508 | } | |
509 | ||
510 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 511 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
512 | if (s->to_replace) { |
513 | to_replace = s->to_replace; | |
514 | } | |
40365552 | 515 | |
e253f4b8 KW |
516 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
517 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 518 | } |
b8804815 KW |
519 | |
520 | /* The mirror job has no requests in flight any more, but we need to | |
521 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 KW |
522 | bdrv_drained_begin(target_bs); |
523 | bdrv_replace_in_backing_chain(to_replace, target_bs); | |
524 | bdrv_drained_end(target_bs); | |
b8804815 | 525 | |
b6d2e599 KW |
526 | /* We just changed the BDS the job BB refers to */ |
527 | blk_remove_bs(job->blk); | |
528 | blk_insert_bs(job->blk, src); | |
5a7e7a0b SH |
529 | } |
530 | if (s->to_replace) { | |
531 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
532 | error_free(s->replace_blocker); | |
533 | bdrv_unref(s->to_replace); | |
534 | } | |
535 | if (replace_aio_context) { | |
536 | aio_context_release(replace_aio_context); | |
537 | } | |
538 | g_free(s->replaces); | |
e253f4b8 | 539 | blk_unref(s->target); |
bae8196d | 540 | s->target = NULL; |
5a7e7a0b SH |
541 | block_job_completed(&s->common, data->ret); |
542 | g_free(data); | |
176c3699 | 543 | bdrv_drained_end(src); |
3f09bfbc | 544 | bdrv_unref(src); |
5a7e7a0b SH |
545 | } |
546 | ||
49efb1f5 DL |
547 | static void mirror_throttle(MirrorBlockJob *s) |
548 | { | |
549 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
550 | ||
551 | if (now - s->last_pause_ns > SLICE_TIME) { | |
552 | s->last_pause_ns = now; | |
553 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); | |
554 | } else { | |
555 | block_job_pause_point(&s->common); | |
556 | } | |
557 | } | |
558 | ||
c0b363ad DL |
559 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
560 | { | |
561 | int64_t sector_num, end; | |
562 | BlockDriverState *base = s->base; | |
563 | BlockDriverState *bs = blk_bs(s->common.blk); | |
564 | BlockDriverState *target_bs = blk_bs(s->target); | |
c0b363ad DL |
565 | int ret, n; |
566 | ||
567 | end = s->bdev_length / BDRV_SECTOR_SIZE; | |
568 | ||
b7d5062c | 569 | if (base == NULL && !bdrv_has_zero_init(target_bs)) { |
c7c2769c DL |
570 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
571 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, end); | |
572 | return 0; | |
573 | } | |
574 | ||
575 | for (sector_num = 0; sector_num < end; ) { | |
576 | int nb_sectors = MIN(end - sector_num, | |
577 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity) >> BDRV_SECTOR_BITS); | |
578 | ||
579 | mirror_throttle(s); | |
580 | ||
581 | if (block_job_is_cancelled(&s->common)) { | |
582 | return 0; | |
583 | } | |
584 | ||
585 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
586 | trace_mirror_yield(s, s->in_flight, s->buf_free_count, -1); | |
587 | mirror_wait_for_io(s); | |
588 | continue; | |
589 | } | |
590 | ||
591 | mirror_do_zero_or_discard(s, sector_num, nb_sectors, false); | |
592 | sector_num += nb_sectors; | |
593 | } | |
594 | ||
bae8196d | 595 | mirror_wait_for_all_io(s); |
b7d5062c DL |
596 | } |
597 | ||
c0b363ad DL |
598 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
599 | for (sector_num = 0; sector_num < end; ) { | |
600 | /* Just to make sure we are not exceeding int limit. */ | |
601 | int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, | |
602 | end - sector_num); | |
603 | ||
604 | mirror_throttle(s); | |
605 | ||
606 | if (block_job_is_cancelled(&s->common)) { | |
607 | return 0; | |
608 | } | |
609 | ||
610 | ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); | |
611 | if (ret < 0) { | |
612 | return ret; | |
613 | } | |
614 | ||
615 | assert(n > 0); | |
b7d5062c | 616 | if (ret == 1) { |
c0b363ad DL |
617 | bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); |
618 | } | |
619 | sector_num += n; | |
620 | } | |
621 | return 0; | |
622 | } | |
623 | ||
bdffb31d PB |
624 | /* Called when going out of the streaming phase to flush the bulk of the |
625 | * data to the medium, or just before completing. | |
626 | */ | |
627 | static int mirror_flush(MirrorBlockJob *s) | |
628 | { | |
629 | int ret = blk_flush(s->target); | |
630 | if (ret < 0) { | |
631 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
632 | s->ret = ret; | |
633 | } | |
634 | } | |
635 | return ret; | |
636 | } | |
637 | ||
893f7eba PB |
638 | static void coroutine_fn mirror_run(void *opaque) |
639 | { | |
640 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 641 | MirrorExitData *data; |
e253f4b8 KW |
642 | BlockDriverState *bs = blk_bs(s->common.blk); |
643 | BlockDriverState *target_bs = blk_bs(s->target); | |
9a0cec66 | 644 | bool need_drain = true; |
c0b363ad | 645 | int64_t length; |
b812f671 | 646 | BlockDriverInfo bdi; |
1d33936e JC |
647 | char backing_filename[2]; /* we only need 2 characters because we are only |
648 | checking for a NULL string */ | |
893f7eba | 649 | int ret = 0; |
e5b43573 | 650 | int target_cluster_size = BDRV_SECTOR_SIZE; |
893f7eba PB |
651 | |
652 | if (block_job_is_cancelled(&s->common)) { | |
653 | goto immediate_exit; | |
654 | } | |
655 | ||
b21c7652 HR |
656 | s->bdev_length = bdrv_getlength(bs); |
657 | if (s->bdev_length < 0) { | |
658 | ret = s->bdev_length; | |
373df5b1 | 659 | goto immediate_exit; |
b21c7652 | 660 | } else if (s->bdev_length == 0) { |
9e48b025 FZ |
661 | /* Report BLOCK_JOB_READY and wait for complete. */ |
662 | block_job_event_ready(&s->common); | |
663 | s->synced = true; | |
664 | while (!block_job_is_cancelled(&s->common) && !s->should_complete) { | |
665 | block_job_yield(&s->common); | |
666 | } | |
667 | s->common.cancelled = false; | |
668 | goto immediate_exit; | |
893f7eba PB |
669 | } |
670 | ||
b21c7652 | 671 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
672 | s->in_flight_bitmap = bitmap_new(length); |
673 | ||
b812f671 PB |
674 | /* If we have no backing file yet in the destination, we cannot let |
675 | * the destination do COW. Instead, we copy sectors around the | |
676 | * dirty data if needed. We need a bitmap to do that. | |
677 | */ | |
e253f4b8 | 678 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 679 | sizeof(backing_filename)); |
e253f4b8 | 680 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
e5b43573 FZ |
681 | target_cluster_size = bdi.cluster_size; |
682 | } | |
e253f4b8 | 683 | if (backing_filename[0] && !target_bs->backing |
e5b43573 FZ |
684 | && s->granularity < target_cluster_size) { |
685 | s->buf_size = MAX(s->buf_size, target_cluster_size); | |
686 | s->cow_bitmap = bitmap_new(length); | |
b812f671 | 687 | } |
e5b43573 | 688 | s->target_cluster_sectors = target_cluster_size >> BDRV_SECTOR_BITS; |
e253f4b8 | 689 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 690 | |
7504edf4 KW |
691 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
692 | if (s->buf == NULL) { | |
693 | ret = -ENOMEM; | |
694 | goto immediate_exit; | |
695 | } | |
696 | ||
402a4741 | 697 | mirror_free_init(s); |
893f7eba | 698 | |
49efb1f5 | 699 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 700 | if (!s->is_none_mode) { |
c0b363ad DL |
701 | ret = mirror_dirty_init(s); |
702 | if (ret < 0 || block_job_is_cancelled(&s->common)) { | |
703 | goto immediate_exit; | |
893f7eba PB |
704 | } |
705 | } | |
706 | ||
dc162c8e FZ |
707 | assert(!s->dbi); |
708 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap, 0); | |
893f7eba | 709 | for (;;) { |
cc8c9d6c | 710 | uint64_t delay_ns = 0; |
49efb1f5 | 711 | int64_t cnt, delta; |
893f7eba PB |
712 | bool should_complete; |
713 | ||
bd48bde8 PB |
714 | if (s->ret < 0) { |
715 | ret = s->ret; | |
716 | goto immediate_exit; | |
717 | } | |
718 | ||
565ac01f SH |
719 | block_job_pause_point(&s->common); |
720 | ||
20dca810 | 721 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
b21c7652 HR |
722 | /* s->common.offset contains the number of bytes already processed so |
723 | * far, cnt is the number of dirty sectors remaining and | |
724 | * s->sectors_in_flight is the number of sectors currently being | |
725 | * processed; together those are the current total operation length */ | |
726 | s->common.len = s->common.offset + | |
727 | (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
728 | |
729 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 730 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
bd48bde8 PB |
731 | * We do so every SLICE_TIME nanoseconds, or when there is an error, |
732 | * or when the source is clean, whichever comes first. | |
733 | */ | |
49efb1f5 DL |
734 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
735 | if (delta < SLICE_TIME && | |
bd48bde8 | 736 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 737 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 PB |
738 | (cnt == 0 && s->in_flight > 0)) { |
739 | trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt); | |
21cd917f | 740 | mirror_wait_for_io(s); |
bd48bde8 PB |
741 | continue; |
742 | } else if (cnt != 0) { | |
cc8c9d6c | 743 | delay_ns = mirror_iteration(s); |
893f7eba | 744 | } |
893f7eba PB |
745 | } |
746 | ||
747 | should_complete = false; | |
bd48bde8 | 748 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 749 | trace_mirror_before_flush(s); |
bdffb31d PB |
750 | if (!s->synced) { |
751 | if (mirror_flush(s) < 0) { | |
752 | /* Go check s->ret. */ | |
753 | continue; | |
b952b558 | 754 | } |
b952b558 PB |
755 | /* We're out of the streaming phase. From now on, if the job |
756 | * is cancelled we will actually complete all pending I/O and | |
757 | * report completion. This way, block-job-cancel will leave | |
758 | * the target in a consistent state. | |
759 | */ | |
bdffb31d PB |
760 | block_job_event_ready(&s->common); |
761 | s->synced = true; | |
d63ffd87 | 762 | } |
bdffb31d PB |
763 | |
764 | should_complete = s->should_complete || | |
765 | block_job_is_cancelled(&s->common); | |
766 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); | |
893f7eba PB |
767 | } |
768 | ||
769 | if (cnt == 0 && should_complete) { | |
770 | /* The dirty bitmap is not updated while operations are pending. | |
771 | * If we're about to exit, wait for pending operations before | |
772 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
773 | * source has dirty data to copy! | |
774 | * | |
775 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
776 | * mirror_populate runs, so pause it now. Before deciding |
777 | * whether to switch to target check one last time if I/O has | |
778 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba PB |
779 | */ |
780 | trace_mirror_before_drain(s, cnt); | |
9a0cec66 PB |
781 | |
782 | bdrv_drained_begin(bs); | |
20dca810 | 783 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 784 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 PB |
785 | bdrv_drained_end(bs); |
786 | continue; | |
787 | } | |
788 | ||
789 | /* The two disks are in sync. Exit and report successful | |
790 | * completion. | |
791 | */ | |
792 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
793 | s->common.cancelled = false; | |
794 | need_drain = false; | |
795 | break; | |
893f7eba PB |
796 | } |
797 | ||
798 | ret = 0; | |
cc8c9d6c | 799 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
d63ffd87 | 800 | if (!s->synced) { |
7483d1e5 | 801 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
802 | if (block_job_is_cancelled(&s->common)) { |
803 | break; | |
804 | } | |
805 | } else if (!should_complete) { | |
bd48bde8 | 806 | delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); |
7483d1e5 | 807 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba | 808 | } |
49efb1f5 | 809 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
810 | } |
811 | ||
812 | immediate_exit: | |
bd48bde8 PB |
813 | if (s->in_flight > 0) { |
814 | /* We get here only if something went wrong. Either the job failed, | |
815 | * or it was cancelled prematurely so that we do not guarantee that | |
816 | * the target is a copy of the source. | |
817 | */ | |
818 | assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); | |
9a0cec66 | 819 | assert(need_drain); |
bae8196d | 820 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
821 | } |
822 | ||
823 | assert(s->in_flight == 0); | |
7191bf31 | 824 | qemu_vfree(s->buf); |
b812f671 | 825 | g_free(s->cow_bitmap); |
402a4741 | 826 | g_free(s->in_flight_bitmap); |
dc162c8e | 827 | bdrv_dirty_iter_free(s->dbi); |
e4654d2d | 828 | bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); |
5a7e7a0b SH |
829 | |
830 | data = g_malloc(sizeof(*data)); | |
831 | data->ret = ret; | |
9a0cec66 PB |
832 | |
833 | if (need_drain) { | |
834 | bdrv_drained_begin(bs); | |
835 | } | |
5a7e7a0b | 836 | block_job_defer_to_main_loop(&s->common, mirror_exit, data); |
893f7eba PB |
837 | } |
838 | ||
839 | static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
840 | { | |
841 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
842 | ||
843 | if (speed < 0) { | |
c6bd8c70 | 844 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
893f7eba PB |
845 | return; |
846 | } | |
847 | ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); | |
848 | } | |
849 | ||
d63ffd87 PB |
850 | static void mirror_complete(BlockJob *job, Error **errp) |
851 | { | |
852 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
274fccee HR |
853 | BlockDriverState *src, *target; |
854 | ||
855 | src = blk_bs(job->blk); | |
856 | target = blk_bs(s->target); | |
d63ffd87 | 857 | |
d63ffd87 | 858 | if (!s->synced) { |
9df229c3 AG |
859 | error_setg(errp, "The active block job '%s' cannot be completed", |
860 | job->id); | |
d63ffd87 PB |
861 | return; |
862 | } | |
863 | ||
274fccee HR |
864 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
865 | int ret; | |
866 | ||
867 | assert(!target->backing); | |
868 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
869 | if (ret < 0) { | |
870 | return; | |
871 | } | |
872 | } | |
873 | ||
15d67298 | 874 | /* block all operations on to_replace bs */ |
09158f00 | 875 | if (s->replaces) { |
5a7e7a0b SH |
876 | AioContext *replace_aio_context; |
877 | ||
e12f3784 | 878 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 879 | if (!s->to_replace) { |
e12f3784 | 880 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
881 | return; |
882 | } | |
883 | ||
5a7e7a0b SH |
884 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
885 | aio_context_acquire(replace_aio_context); | |
886 | ||
09158f00 BC |
887 | error_setg(&s->replace_blocker, |
888 | "block device is in use by block-job-complete"); | |
889 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
890 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
891 | |
892 | aio_context_release(replace_aio_context); | |
09158f00 BC |
893 | } |
894 | ||
274fccee HR |
895 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { |
896 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
897 | if (backing_bs(target) != backing) { | |
898 | bdrv_set_backing_hd(target, backing); | |
899 | } | |
900 | } | |
901 | ||
d63ffd87 | 902 | s->should_complete = true; |
751ebd76 | 903 | block_job_enter(&s->common); |
d63ffd87 PB |
904 | } |
905 | ||
bae8196d | 906 | static void mirror_pause(BlockJob *job) |
565ac01f SH |
907 | { |
908 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
909 | ||
bae8196d | 910 | mirror_wait_for_all_io(s); |
565ac01f SH |
911 | } |
912 | ||
913 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) | |
914 | { | |
915 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
916 | ||
917 | blk_set_aio_context(s->target, new_context); | |
918 | } | |
919 | ||
bae8196d PB |
920 | static void mirror_drain(BlockJob *job) |
921 | { | |
922 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
923 | ||
924 | /* Need to keep a reference in case blk_drain triggers execution | |
925 | * of mirror_complete... | |
926 | */ | |
927 | if (s->target) { | |
928 | BlockBackend *target = s->target; | |
929 | blk_ref(target); | |
930 | blk_drain(target); | |
931 | blk_unref(target); | |
932 | } | |
933 | } | |
934 | ||
3fc4b10a | 935 | static const BlockJobDriver mirror_job_driver = { |
565ac01f SH |
936 | .instance_size = sizeof(MirrorBlockJob), |
937 | .job_type = BLOCK_JOB_TYPE_MIRROR, | |
938 | .set_speed = mirror_set_speed, | |
a7815a76 | 939 | .start = mirror_run, |
565ac01f SH |
940 | .complete = mirror_complete, |
941 | .pause = mirror_pause, | |
942 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 943 | .drain = mirror_drain, |
893f7eba PB |
944 | }; |
945 | ||
03544a6e | 946 | static const BlockJobDriver commit_active_job_driver = { |
565ac01f SH |
947 | .instance_size = sizeof(MirrorBlockJob), |
948 | .job_type = BLOCK_JOB_TYPE_COMMIT, | |
949 | .set_speed = mirror_set_speed, | |
a7815a76 | 950 | .start = mirror_run, |
565ac01f SH |
951 | .complete = mirror_complete, |
952 | .pause = mirror_pause, | |
953 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 954 | .drain = mirror_drain, |
03544a6e FZ |
955 | }; |
956 | ||
71aa9867 | 957 | static void mirror_start_job(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
958 | int creation_flags, BlockDriverState *target, |
959 | const char *replaces, int64_t speed, | |
960 | uint32_t granularity, int64_t buf_size, | |
274fccee | 961 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
962 | BlockdevOnError on_source_error, |
963 | BlockdevOnError on_target_error, | |
0fc9f8ea | 964 | bool unmap, |
097310b5 | 965 | BlockCompletionFunc *cb, |
09158f00 BC |
966 | void *opaque, Error **errp, |
967 | const BlockJobDriver *driver, | |
b49f7ead WC |
968 | bool is_none_mode, BlockDriverState *base, |
969 | bool auto_complete) | |
893f7eba PB |
970 | { |
971 | MirrorBlockJob *s; | |
972 | ||
eee13dfe | 973 | if (granularity == 0) { |
341ebc2f | 974 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
975 | } |
976 | ||
977 | assert ((granularity & (granularity - 1)) == 0); | |
978 | ||
48ac0a4d WC |
979 | if (buf_size < 0) { |
980 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
981 | return; | |
982 | } | |
983 | ||
984 | if (buf_size == 0) { | |
985 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
986 | } | |
5bc361b8 | 987 | |
47970dfb JS |
988 | s = block_job_create(job_id, driver, bs, speed, creation_flags, |
989 | cb, opaque, errp); | |
893f7eba PB |
990 | if (!s) { |
991 | return; | |
992 | } | |
993 | ||
e253f4b8 KW |
994 | s->target = blk_new(); |
995 | blk_insert_bs(s->target, target); | |
996 | ||
09158f00 | 997 | s->replaces = g_strdup(replaces); |
b952b558 PB |
998 | s->on_source_error = on_source_error; |
999 | s->on_target_error = on_target_error; | |
03544a6e | 1000 | s->is_none_mode = is_none_mode; |
274fccee | 1001 | s->backing_mode = backing_mode; |
5bc361b8 | 1002 | s->base = base; |
eee13dfe | 1003 | s->granularity = granularity; |
48ac0a4d | 1004 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1005 | s->unmap = unmap; |
b49f7ead WC |
1006 | if (auto_complete) { |
1007 | s->should_complete = true; | |
1008 | } | |
b812f671 | 1009 | |
0db6e54a | 1010 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1011 | if (!s->dirty_bitmap) { |
97031164 | 1012 | g_free(s->replaces); |
e253f4b8 | 1013 | blk_unref(s->target); |
18930ba3 | 1014 | block_job_unref(&s->common); |
b8afb520 FZ |
1015 | return; |
1016 | } | |
10f3cd15 | 1017 | |
cee3c6b5 | 1018 | block_job_add_bdrv(&s->common, target); |
f3ede4b0 AG |
1019 | /* In commit_active_start() all intermediate nodes disappear, so |
1020 | * any jobs in them must be blocked */ | |
1021 | if (bdrv_chain_contains(bs, target)) { | |
1022 | BlockDriverState *iter; | |
1023 | for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { | |
1024 | block_job_add_bdrv(&s->common, iter); | |
1025 | } | |
1026 | } | |
10f3cd15 | 1027 | |
5ccac6f1 JS |
1028 | trace_mirror_start(bs, s, opaque); |
1029 | block_job_start(&s->common); | |
893f7eba | 1030 | } |
03544a6e | 1031 | |
71aa9867 AG |
1032 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1033 | BlockDriverState *target, const char *replaces, | |
5fba6c0e | 1034 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
1035 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
1036 | BlockdevOnError on_source_error, | |
03544a6e | 1037 | BlockdevOnError on_target_error, |
8254b6d9 | 1038 | bool unmap, Error **errp) |
03544a6e FZ |
1039 | { |
1040 | bool is_none_mode; | |
1041 | BlockDriverState *base; | |
1042 | ||
4b80ab2b JS |
1043 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
1044 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
1045 | return; |
1046 | } | |
03544a6e | 1047 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 1048 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
47970dfb | 1049 | mirror_start_job(job_id, bs, BLOCK_JOB_DEFAULT, target, replaces, |
274fccee | 1050 | speed, granularity, buf_size, backing_mode, |
8254b6d9 | 1051 | on_source_error, on_target_error, unmap, NULL, NULL, errp, |
b49f7ead | 1052 | &mirror_job_driver, is_none_mode, base, false); |
03544a6e FZ |
1053 | } |
1054 | ||
fd62c609 | 1055 | void commit_active_start(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1056 | BlockDriverState *base, int creation_flags, |
1057 | int64_t speed, BlockdevOnError on_error, | |
1058 | BlockCompletionFunc *cb, void *opaque, Error **errp, | |
b49f7ead | 1059 | bool auto_complete) |
03544a6e | 1060 | { |
4da83585 JC |
1061 | int64_t length, base_length; |
1062 | int orig_base_flags; | |
39a611a3 | 1063 | int ret; |
cc67f4d1 | 1064 | Error *local_err = NULL; |
4da83585 JC |
1065 | |
1066 | orig_base_flags = bdrv_get_flags(base); | |
1067 | ||
20a63d2c FZ |
1068 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
1069 | return; | |
1070 | } | |
4da83585 JC |
1071 | |
1072 | length = bdrv_getlength(bs); | |
1073 | if (length < 0) { | |
39a611a3 JC |
1074 | error_setg_errno(errp, -length, |
1075 | "Unable to determine length of %s", bs->filename); | |
4da83585 JC |
1076 | goto error_restore_flags; |
1077 | } | |
1078 | ||
1079 | base_length = bdrv_getlength(base); | |
1080 | if (base_length < 0) { | |
39a611a3 JC |
1081 | error_setg_errno(errp, -base_length, |
1082 | "Unable to determine length of %s", base->filename); | |
4da83585 JC |
1083 | goto error_restore_flags; |
1084 | } | |
1085 | ||
1086 | if (length > base_length) { | |
39a611a3 JC |
1087 | ret = bdrv_truncate(base, length); |
1088 | if (ret < 0) { | |
1089 | error_setg_errno(errp, -ret, | |
1090 | "Top image %s is larger than base image %s, and " | |
4da83585 JC |
1091 | "resize of base image failed", |
1092 | bs->filename, base->filename); | |
1093 | goto error_restore_flags; | |
1094 | } | |
1095 | } | |
1096 | ||
47970dfb | 1097 | mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
71aa9867 | 1098 | MIRROR_LEAVE_BACKING_CHAIN, |
6f13acf9 | 1099 | on_error, on_error, true, cb, opaque, &local_err, |
b49f7ead | 1100 | &commit_active_job_driver, false, base, auto_complete); |
0fb6395c | 1101 | if (local_err) { |
cc67f4d1 | 1102 | error_propagate(errp, local_err); |
4da83585 JC |
1103 | goto error_restore_flags; |
1104 | } | |
1105 | ||
1106 | return; | |
1107 | ||
1108 | error_restore_flags: | |
1109 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1110 | * the original error */ | |
1111 | bdrv_reopen(base, orig_base_flags, NULL); | |
1112 | return; | |
03544a6e | 1113 | } |