]>
Commit | Line | Data |
---|---|---|
893f7eba PB |
1 | /* |
2 | * Image mirroring | |
3 | * | |
4 | * Copyright Red Hat, Inc. 2012 | |
5 | * | |
6 | * Authors: | |
7 | * Paolo Bonzini <pbonzini@redhat.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU LGPL, version 2 or later. | |
10 | * See the COPYING.LIB file in the top-level directory. | |
11 | * | |
12 | */ | |
13 | ||
80c71a24 | 14 | #include "qemu/osdep.h" |
fd4a6493 | 15 | #include "qemu/cutils.h" |
893f7eba | 16 | #include "trace.h" |
c87621ea | 17 | #include "block/blockjob_int.h" |
737e150e | 18 | #include "block/block_int.h" |
373340b2 | 19 | #include "sysemu/block-backend.h" |
da34e65c | 20 | #include "qapi/error.h" |
cc7a8ea7 | 21 | #include "qapi/qmp/qerror.h" |
893f7eba | 22 | #include "qemu/ratelimit.h" |
b812f671 | 23 | #include "qemu/bitmap.h" |
893f7eba | 24 | |
402a4741 PB |
25 | #define SLICE_TIME 100000000ULL /* ns */ |
26 | #define MAX_IN_FLIGHT 16 | |
0965a41e VSO |
27 | #define MAX_IO_SECTORS ((1 << 20) >> BDRV_SECTOR_BITS) /* 1 Mb */ |
28 | #define DEFAULT_MIRROR_BUF_SIZE \ | |
29 | (MAX_IN_FLIGHT * MAX_IO_SECTORS * BDRV_SECTOR_SIZE) | |
402a4741 PB |
30 | |
31 | /* The mirroring buffer is a list of granularity-sized chunks. | |
32 | * Free chunks are organized in a list. | |
33 | */ | |
34 | typedef struct MirrorBuffer { | |
35 | QSIMPLEQ_ENTRY(MirrorBuffer) next; | |
36 | } MirrorBuffer; | |
893f7eba PB |
37 | |
38 | typedef struct MirrorBlockJob { | |
39 | BlockJob common; | |
40 | RateLimit limit; | |
e253f4b8 | 41 | BlockBackend *target; |
4ef85a9c KW |
42 | BlockDriverState *mirror_top_bs; |
43 | BlockDriverState *source; | |
5bc361b8 | 44 | BlockDriverState *base; |
4ef85a9c | 45 | |
09158f00 BC |
46 | /* The name of the graph node to replace */ |
47 | char *replaces; | |
48 | /* The BDS to replace */ | |
49 | BlockDriverState *to_replace; | |
50 | /* Used to block operations on the drive-mirror-replace target */ | |
51 | Error *replace_blocker; | |
03544a6e | 52 | bool is_none_mode; |
274fccee | 53 | BlockMirrorBackingMode backing_mode; |
b952b558 | 54 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 PB |
55 | bool synced; |
56 | bool should_complete; | |
eee13dfe | 57 | int64_t granularity; |
b812f671 | 58 | size_t buf_size; |
b21c7652 | 59 | int64_t bdev_length; |
b812f671 | 60 | unsigned long *cow_bitmap; |
e4654d2d | 61 | BdrvDirtyBitmap *dirty_bitmap; |
dc162c8e | 62 | BdrvDirtyBitmapIter *dbi; |
893f7eba | 63 | uint8_t *buf; |
402a4741 PB |
64 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
65 | int buf_free_count; | |
bd48bde8 | 66 | |
49efb1f5 | 67 | uint64_t last_pause_ns; |
402a4741 | 68 | unsigned long *in_flight_bitmap; |
bd48bde8 | 69 | int in_flight; |
531509ba | 70 | int64_t sectors_in_flight; |
bd48bde8 | 71 | int ret; |
0fc9f8ea | 72 | bool unmap; |
e424aff5 | 73 | bool waiting_for_io; |
e5b43573 FZ |
74 | int target_cluster_sectors; |
75 | int max_iov; | |
90ab48eb | 76 | bool initial_zeroing_ongoing; |
893f7eba PB |
77 | } MirrorBlockJob; |
78 | ||
bd48bde8 PB |
79 | typedef struct MirrorOp { |
80 | MirrorBlockJob *s; | |
81 | QEMUIOVector qiov; | |
bd48bde8 PB |
82 | int64_t sector_num; |
83 | int nb_sectors; | |
84 | } MirrorOp; | |
85 | ||
b952b558 PB |
86 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
87 | int error) | |
88 | { | |
89 | s->synced = false; | |
90 | if (read) { | |
81e254dc KW |
91 | return block_job_error_action(&s->common, s->on_source_error, |
92 | true, error); | |
b952b558 | 93 | } else { |
81e254dc KW |
94 | return block_job_error_action(&s->common, s->on_target_error, |
95 | false, error); | |
b952b558 PB |
96 | } |
97 | } | |
98 | ||
bd48bde8 PB |
99 | static void mirror_iteration_done(MirrorOp *op, int ret) |
100 | { | |
101 | MirrorBlockJob *s = op->s; | |
402a4741 | 102 | struct iovec *iov; |
bd48bde8 | 103 | int64_t chunk_num; |
402a4741 | 104 | int i, nb_chunks, sectors_per_chunk; |
bd48bde8 PB |
105 | |
106 | trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret); | |
107 | ||
108 | s->in_flight--; | |
b21c7652 | 109 | s->sectors_in_flight -= op->nb_sectors; |
402a4741 PB |
110 | iov = op->qiov.iov; |
111 | for (i = 0; i < op->qiov.niov; i++) { | |
112 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
113 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
114 | s->buf_free_count++; | |
115 | } | |
116 | ||
bd48bde8 PB |
117 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
118 | chunk_num = op->sector_num / sectors_per_chunk; | |
4150ae60 | 119 | nb_chunks = DIV_ROUND_UP(op->nb_sectors, sectors_per_chunk); |
402a4741 | 120 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
b21c7652 HR |
121 | if (ret >= 0) { |
122 | if (s->cow_bitmap) { | |
123 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
124 | } | |
90ab48eb AN |
125 | if (!s->initial_zeroing_ongoing) { |
126 | s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE; | |
127 | } | |
bd48bde8 | 128 | } |
6df3bf8e | 129 | qemu_iovec_destroy(&op->qiov); |
c84b3192 | 130 | g_free(op); |
7b770c72 | 131 | |
e424aff5 | 132 | if (s->waiting_for_io) { |
0b8b8753 | 133 | qemu_coroutine_enter(s->common.co); |
7b770c72 | 134 | } |
bd48bde8 PB |
135 | } |
136 | ||
137 | static void mirror_write_complete(void *opaque, int ret) | |
138 | { | |
139 | MirrorOp *op = opaque; | |
140 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
141 | |
142 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 143 | if (ret < 0) { |
bd48bde8 PB |
144 | BlockErrorAction action; |
145 | ||
20dca810 | 146 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 147 | action = mirror_error_action(s, false, -ret); |
a589569f | 148 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
149 | s->ret = ret; |
150 | } | |
151 | } | |
152 | mirror_iteration_done(op, ret); | |
b9e413dd | 153 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
154 | } |
155 | ||
156 | static void mirror_read_complete(void *opaque, int ret) | |
157 | { | |
158 | MirrorOp *op = opaque; | |
159 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
160 | |
161 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 162 | if (ret < 0) { |
bd48bde8 PB |
163 | BlockErrorAction action; |
164 | ||
20dca810 | 165 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 166 | action = mirror_error_action(s, true, -ret); |
a589569f | 167 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
168 | s->ret = ret; |
169 | } | |
170 | ||
171 | mirror_iteration_done(op, ret); | |
b9e413dd PB |
172 | } else { |
173 | blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov, | |
174 | 0, mirror_write_complete, op); | |
bd48bde8 | 175 | } |
b9e413dd | 176 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
177 | } |
178 | ||
4150ae60 FZ |
179 | static inline void mirror_clip_sectors(MirrorBlockJob *s, |
180 | int64_t sector_num, | |
181 | int *nb_sectors) | |
182 | { | |
183 | *nb_sectors = MIN(*nb_sectors, | |
184 | s->bdev_length / BDRV_SECTOR_SIZE - sector_num); | |
185 | } | |
186 | ||
e5b43573 FZ |
187 | /* Round sector_num and/or nb_sectors to target cluster if COW is needed, and |
188 | * return the offset of the adjusted tail sector against original. */ | |
189 | static int mirror_cow_align(MirrorBlockJob *s, | |
190 | int64_t *sector_num, | |
191 | int *nb_sectors) | |
893f7eba | 192 | { |
e5b43573 FZ |
193 | bool need_cow; |
194 | int ret = 0; | |
195 | int chunk_sectors = s->granularity >> BDRV_SECTOR_BITS; | |
196 | int64_t align_sector_num = *sector_num; | |
197 | int align_nb_sectors = *nb_sectors; | |
198 | int max_sectors = chunk_sectors * s->max_iov; | |
199 | ||
200 | need_cow = !test_bit(*sector_num / chunk_sectors, s->cow_bitmap); | |
201 | need_cow |= !test_bit((*sector_num + *nb_sectors - 1) / chunk_sectors, | |
202 | s->cow_bitmap); | |
203 | if (need_cow) { | |
244483e6 KW |
204 | bdrv_round_sectors_to_clusters(blk_bs(s->target), *sector_num, |
205 | *nb_sectors, &align_sector_num, | |
206 | &align_nb_sectors); | |
e5b43573 | 207 | } |
3515727f | 208 | |
e5b43573 FZ |
209 | if (align_nb_sectors > max_sectors) { |
210 | align_nb_sectors = max_sectors; | |
211 | if (need_cow) { | |
212 | align_nb_sectors = QEMU_ALIGN_DOWN(align_nb_sectors, | |
213 | s->target_cluster_sectors); | |
214 | } | |
8f0720ec | 215 | } |
4150ae60 FZ |
216 | /* Clipping may result in align_nb_sectors unaligned to chunk boundary, but |
217 | * that doesn't matter because it's already the end of source image. */ | |
218 | mirror_clip_sectors(s, align_sector_num, &align_nb_sectors); | |
8f0720ec | 219 | |
e5b43573 FZ |
220 | ret = align_sector_num + align_nb_sectors - (*sector_num + *nb_sectors); |
221 | *sector_num = align_sector_num; | |
222 | *nb_sectors = align_nb_sectors; | |
223 | assert(ret >= 0); | |
224 | return ret; | |
225 | } | |
226 | ||
21cd917f FZ |
227 | static inline void mirror_wait_for_io(MirrorBlockJob *s) |
228 | { | |
229 | assert(!s->waiting_for_io); | |
230 | s->waiting_for_io = true; | |
231 | qemu_coroutine_yield(); | |
232 | s->waiting_for_io = false; | |
233 | } | |
234 | ||
e5b43573 | 235 | /* Submit async read while handling COW. |
17612955 JS |
236 | * Returns: The number of sectors copied after and including sector_num, |
237 | * excluding any sectors copied prior to sector_num due to alignment. | |
238 | * This will be nb_sectors if no alignment is necessary, or | |
e5b43573 FZ |
239 | * (new_end - sector_num) if tail is rounded up or down due to |
240 | * alignment or buffer limit. | |
241 | */ | |
242 | static int mirror_do_read(MirrorBlockJob *s, int64_t sector_num, | |
243 | int nb_sectors) | |
244 | { | |
e253f4b8 | 245 | BlockBackend *source = s->common.blk; |
e5b43573 | 246 | int sectors_per_chunk, nb_chunks; |
17612955 | 247 | int ret; |
e5b43573 | 248 | MirrorOp *op; |
e4808881 | 249 | int max_sectors; |
e5b43573 | 250 | |
884fea4e | 251 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
e4808881 | 252 | max_sectors = sectors_per_chunk * s->max_iov; |
402a4741 | 253 | |
e5b43573 FZ |
254 | /* We can only handle as much as buf_size at a time. */ |
255 | nb_sectors = MIN(s->buf_size >> BDRV_SECTOR_BITS, nb_sectors); | |
e4808881 | 256 | nb_sectors = MIN(max_sectors, nb_sectors); |
e5b43573 | 257 | assert(nb_sectors); |
17612955 | 258 | ret = nb_sectors; |
402a4741 | 259 | |
e5b43573 FZ |
260 | if (s->cow_bitmap) { |
261 | ret += mirror_cow_align(s, §or_num, &nb_sectors); | |
262 | } | |
263 | assert(nb_sectors << BDRV_SECTOR_BITS <= s->buf_size); | |
264 | /* The sector range must meet granularity because: | |
265 | * 1) Caller passes in aligned values; | |
266 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
e5b43573 | 267 | assert(!(sector_num % sectors_per_chunk)); |
4150ae60 | 268 | nb_chunks = DIV_ROUND_UP(nb_sectors, sectors_per_chunk); |
e5b43573 FZ |
269 | |
270 | while (s->buf_free_count < nb_chunks) { | |
402a4741 | 271 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); |
21cd917f | 272 | mirror_wait_for_io(s); |
b812f671 PB |
273 | } |
274 | ||
bd48bde8 | 275 | /* Allocate a MirrorOp that is used as an AIO callback. */ |
c84b3192 | 276 | op = g_new(MirrorOp, 1); |
bd48bde8 | 277 | op->s = s; |
bd48bde8 PB |
278 | op->sector_num = sector_num; |
279 | op->nb_sectors = nb_sectors; | |
402a4741 PB |
280 | |
281 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
282 | * from s->buf_free. | |
283 | */ | |
284 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
285 | while (nb_chunks-- > 0) { |
286 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
e5b43573 | 287 | size_t remaining = nb_sectors * BDRV_SECTOR_SIZE - op->qiov.size; |
5a0f6fd5 | 288 | |
402a4741 PB |
289 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
290 | s->buf_free_count--; | |
5a0f6fd5 | 291 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 292 | } |
bd48bde8 | 293 | |
893f7eba | 294 | /* Copy the dirty cluster. */ |
bd48bde8 | 295 | s->in_flight++; |
b21c7652 | 296 | s->sectors_in_flight += nb_sectors; |
b812f671 | 297 | trace_mirror_one_iteration(s, sector_num, nb_sectors); |
dcfb3beb | 298 | |
73698c30 | 299 | blk_aio_preadv(source, sector_num * BDRV_SECTOR_SIZE, &op->qiov, 0, |
e5b43573 FZ |
300 | mirror_read_complete, op); |
301 | return ret; | |
302 | } | |
303 | ||
304 | static void mirror_do_zero_or_discard(MirrorBlockJob *s, | |
305 | int64_t sector_num, | |
306 | int nb_sectors, | |
307 | bool is_discard) | |
308 | { | |
309 | MirrorOp *op; | |
310 | ||
311 | /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed | |
312 | * so the freeing in mirror_iteration_done is nop. */ | |
313 | op = g_new0(MirrorOp, 1); | |
314 | op->s = s; | |
315 | op->sector_num = sector_num; | |
316 | op->nb_sectors = nb_sectors; | |
317 | ||
318 | s->in_flight++; | |
319 | s->sectors_in_flight += nb_sectors; | |
320 | if (is_discard) { | |
1c6c4bb7 EB |
321 | blk_aio_pdiscard(s->target, sector_num << BDRV_SECTOR_BITS, |
322 | op->nb_sectors << BDRV_SECTOR_BITS, | |
323 | mirror_write_complete, op); | |
e5b43573 | 324 | } else { |
e253f4b8 KW |
325 | blk_aio_pwrite_zeroes(s->target, sector_num * BDRV_SECTOR_SIZE, |
326 | op->nb_sectors * BDRV_SECTOR_SIZE, | |
dcfb3beb FZ |
327 | s->unmap ? BDRV_REQ_MAY_UNMAP : 0, |
328 | mirror_write_complete, op); | |
e5b43573 FZ |
329 | } |
330 | } | |
331 | ||
332 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) | |
333 | { | |
4ef85a9c | 334 | BlockDriverState *source = s->source; |
9c83625b | 335 | int64_t sector_num, first_chunk; |
e5b43573 FZ |
336 | uint64_t delay_ns = 0; |
337 | /* At least the first dirty chunk is mirrored in one iteration. */ | |
338 | int nb_chunks = 1; | |
339 | int64_t end = s->bdev_length / BDRV_SECTOR_SIZE; | |
340 | int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; | |
4b5004d9 | 341 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
0965a41e VSO |
342 | int max_io_sectors = MAX((s->buf_size >> BDRV_SECTOR_BITS) / MAX_IN_FLIGHT, |
343 | MAX_IO_SECTORS); | |
e5b43573 | 344 | |
dc162c8e | 345 | sector_num = bdrv_dirty_iter_next(s->dbi); |
e5b43573 | 346 | if (sector_num < 0) { |
dc162c8e FZ |
347 | bdrv_set_dirty_iter(s->dbi, 0); |
348 | sector_num = bdrv_dirty_iter_next(s->dbi); | |
e5b43573 FZ |
349 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
350 | assert(sector_num >= 0); | |
351 | } | |
352 | ||
9c83625b HR |
353 | first_chunk = sector_num / sectors_per_chunk; |
354 | while (test_bit(first_chunk, s->in_flight_bitmap)) { | |
ff04198b | 355 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); |
9c83625b HR |
356 | mirror_wait_for_io(s); |
357 | } | |
358 | ||
565ac01f SH |
359 | block_job_pause_point(&s->common); |
360 | ||
e5b43573 FZ |
361 | /* Find the number of consective dirty chunks following the first dirty |
362 | * one, and wait for in flight requests in them. */ | |
363 | while (nb_chunks * sectors_per_chunk < (s->buf_size >> BDRV_SECTOR_BITS)) { | |
dc162c8e | 364 | int64_t next_dirty; |
e5b43573 FZ |
365 | int64_t next_sector = sector_num + nb_chunks * sectors_per_chunk; |
366 | int64_t next_chunk = next_sector / sectors_per_chunk; | |
367 | if (next_sector >= end || | |
368 | !bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) { | |
369 | break; | |
370 | } | |
371 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 372 | break; |
e5b43573 | 373 | } |
9c83625b | 374 | |
dc162c8e FZ |
375 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
376 | if (next_dirty > next_sector || next_dirty < 0) { | |
f27a2742 | 377 | /* The bitmap iterator's cache is stale, refresh it */ |
dc162c8e FZ |
378 | bdrv_set_dirty_iter(s->dbi, next_sector); |
379 | next_dirty = bdrv_dirty_iter_next(s->dbi); | |
f27a2742 | 380 | } |
dc162c8e | 381 | assert(next_dirty == next_sector); |
9c83625b | 382 | nb_chunks++; |
e5b43573 FZ |
383 | } |
384 | ||
385 | /* Clear dirty bits before querying the block status, because | |
386 | * calling bdrv_get_block_status_above could yield - if some blocks are | |
387 | * marked dirty in this window, we need to know. | |
388 | */ | |
389 | bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, | |
390 | nb_chunks * sectors_per_chunk); | |
391 | bitmap_set(s->in_flight_bitmap, sector_num / sectors_per_chunk, nb_chunks); | |
392 | while (nb_chunks > 0 && sector_num < end) { | |
39c11580 | 393 | int64_t ret; |
4b5004d9 | 394 | int io_sectors, io_sectors_acct; |
e5b43573 FZ |
395 | BlockDriverState *file; |
396 | enum MirrorMethod { | |
397 | MIRROR_METHOD_COPY, | |
398 | MIRROR_METHOD_ZERO, | |
399 | MIRROR_METHOD_DISCARD | |
400 | } mirror_method = MIRROR_METHOD_COPY; | |
401 | ||
402 | assert(!(sector_num % sectors_per_chunk)); | |
403 | ret = bdrv_get_block_status_above(source, NULL, sector_num, | |
404 | nb_chunks * sectors_per_chunk, | |
405 | &io_sectors, &file); | |
406 | if (ret < 0) { | |
0965a41e VSO |
407 | io_sectors = MIN(nb_chunks * sectors_per_chunk, max_io_sectors); |
408 | } else if (ret & BDRV_BLOCK_DATA) { | |
409 | io_sectors = MIN(io_sectors, max_io_sectors); | |
e5b43573 FZ |
410 | } |
411 | ||
412 | io_sectors -= io_sectors % sectors_per_chunk; | |
413 | if (io_sectors < sectors_per_chunk) { | |
414 | io_sectors = sectors_per_chunk; | |
415 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { | |
416 | int64_t target_sector_num; | |
417 | int target_nb_sectors; | |
244483e6 KW |
418 | bdrv_round_sectors_to_clusters(blk_bs(s->target), sector_num, |
419 | io_sectors, &target_sector_num, | |
420 | &target_nb_sectors); | |
e5b43573 FZ |
421 | if (target_sector_num == sector_num && |
422 | target_nb_sectors == io_sectors) { | |
423 | mirror_method = ret & BDRV_BLOCK_ZERO ? | |
424 | MIRROR_METHOD_ZERO : | |
425 | MIRROR_METHOD_DISCARD; | |
426 | } | |
427 | } | |
428 | ||
cf56a3c6 DL |
429 | while (s->in_flight >= MAX_IN_FLIGHT) { |
430 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); | |
431 | mirror_wait_for_io(s); | |
432 | } | |
433 | ||
dbaa7b57 VSO |
434 | if (s->ret < 0) { |
435 | return 0; | |
436 | } | |
437 | ||
4150ae60 | 438 | mirror_clip_sectors(s, sector_num, &io_sectors); |
e5b43573 FZ |
439 | switch (mirror_method) { |
440 | case MIRROR_METHOD_COPY: | |
441 | io_sectors = mirror_do_read(s, sector_num, io_sectors); | |
4b5004d9 | 442 | io_sectors_acct = io_sectors; |
e5b43573 FZ |
443 | break; |
444 | case MIRROR_METHOD_ZERO: | |
e5b43573 | 445 | case MIRROR_METHOD_DISCARD: |
4b5004d9 DL |
446 | mirror_do_zero_or_discard(s, sector_num, io_sectors, |
447 | mirror_method == MIRROR_METHOD_DISCARD); | |
448 | if (write_zeroes_ok) { | |
449 | io_sectors_acct = 0; | |
450 | } else { | |
451 | io_sectors_acct = io_sectors; | |
452 | } | |
e5b43573 FZ |
453 | break; |
454 | default: | |
455 | abort(); | |
456 | } | |
457 | assert(io_sectors); | |
458 | sector_num += io_sectors; | |
4150ae60 | 459 | nb_chunks -= DIV_ROUND_UP(io_sectors, sectors_per_chunk); |
f14a39cc | 460 | if (s->common.speed) { |
4b5004d9 | 461 | delay_ns = ratelimit_calculate_delay(&s->limit, io_sectors_acct); |
f14a39cc | 462 | } |
dcfb3beb | 463 | } |
cc8c9d6c | 464 | return delay_ns; |
bd48bde8 | 465 | } |
b952b558 | 466 | |
402a4741 PB |
467 | static void mirror_free_init(MirrorBlockJob *s) |
468 | { | |
469 | int granularity = s->granularity; | |
470 | size_t buf_size = s->buf_size; | |
471 | uint8_t *buf = s->buf; | |
472 | ||
473 | assert(s->buf_free_count == 0); | |
474 | QSIMPLEQ_INIT(&s->buf_free); | |
475 | while (buf_size != 0) { | |
476 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
477 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
478 | s->buf_free_count++; | |
479 | buf_size -= granularity; | |
480 | buf += granularity; | |
481 | } | |
482 | } | |
483 | ||
bae8196d PB |
484 | /* This is also used for the .pause callback. There is no matching |
485 | * mirror_resume() because mirror_run() will begin iterating again | |
486 | * when the job is resumed. | |
487 | */ | |
488 | static void mirror_wait_for_all_io(MirrorBlockJob *s) | |
bd48bde8 PB |
489 | { |
490 | while (s->in_flight > 0) { | |
21cd917f | 491 | mirror_wait_for_io(s); |
bd48bde8 | 492 | } |
893f7eba PB |
493 | } |
494 | ||
5a7e7a0b SH |
495 | typedef struct { |
496 | int ret; | |
497 | } MirrorExitData; | |
498 | ||
499 | static void mirror_exit(BlockJob *job, void *opaque) | |
500 | { | |
501 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
502 | MirrorExitData *data = opaque; | |
503 | AioContext *replace_aio_context = NULL; | |
4ef85a9c | 504 | BlockDriverState *src = s->source; |
e253f4b8 | 505 | BlockDriverState *target_bs = blk_bs(s->target); |
4ef85a9c | 506 | BlockDriverState *mirror_top_bs = s->mirror_top_bs; |
12fa4af6 | 507 | Error *local_err = NULL; |
3f09bfbc | 508 | |
2119882c PB |
509 | bdrv_release_dirty_bitmap(src, s->dirty_bitmap); |
510 | ||
3f09bfbc KW |
511 | /* Make sure that the source BDS doesn't go away before we called |
512 | * block_job_completed(). */ | |
513 | bdrv_ref(src); | |
4ef85a9c | 514 | bdrv_ref(mirror_top_bs); |
7d9fcb39 KW |
515 | bdrv_ref(target_bs); |
516 | ||
517 | /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before | |
518 | * inserting target_bs at s->to_replace, where we might not be able to get | |
63c8ef28 KW |
519 | * these permissions. |
520 | * | |
521 | * Note that blk_unref() alone doesn't necessarily drop permissions because | |
522 | * we might be running nested inside mirror_drain(), which takes an extra | |
523 | * reference, so use an explicit blk_set_perm() first. */ | |
524 | blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); | |
7d9fcb39 KW |
525 | blk_unref(s->target); |
526 | s->target = NULL; | |
4ef85a9c KW |
527 | |
528 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
529 | * required before it could become a backing file of target_bs. */ | |
530 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, | |
531 | &error_abort); | |
532 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { | |
533 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
534 | if (backing_bs(target_bs) != backing) { | |
12fa4af6 KW |
535 | bdrv_set_backing_hd(target_bs, backing, &local_err); |
536 | if (local_err) { | |
537 | error_report_err(local_err); | |
538 | data->ret = -EPERM; | |
539 | } | |
4ef85a9c KW |
540 | } |
541 | } | |
5a7e7a0b SH |
542 | |
543 | if (s->to_replace) { | |
544 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
545 | aio_context_acquire(replace_aio_context); | |
546 | } | |
547 | ||
548 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 549 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
550 | if (s->to_replace) { |
551 | to_replace = s->to_replace; | |
552 | } | |
40365552 | 553 | |
e253f4b8 KW |
554 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
555 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 556 | } |
b8804815 KW |
557 | |
558 | /* The mirror job has no requests in flight any more, but we need to | |
559 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 | 560 | bdrv_drained_begin(target_bs); |
5fe31c25 | 561 | bdrv_replace_node(to_replace, target_bs, &local_err); |
e253f4b8 | 562 | bdrv_drained_end(target_bs); |
5fe31c25 KW |
563 | if (local_err) { |
564 | error_report_err(local_err); | |
565 | data->ret = -EPERM; | |
566 | } | |
5a7e7a0b SH |
567 | } |
568 | if (s->to_replace) { | |
569 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
570 | error_free(s->replace_blocker); | |
571 | bdrv_unref(s->to_replace); | |
572 | } | |
573 | if (replace_aio_context) { | |
574 | aio_context_release(replace_aio_context); | |
575 | } | |
576 | g_free(s->replaces); | |
7d9fcb39 | 577 | bdrv_unref(target_bs); |
4ef85a9c KW |
578 | |
579 | /* Remove the mirror filter driver from the graph. Before this, get rid of | |
580 | * the blockers on the intermediate nodes so that the resulting state is | |
0bf74767 KW |
581 | * valid. Also give up permissions on mirror_top_bs->backing, which might |
582 | * block the removal. */ | |
4ef85a9c | 583 | block_job_remove_all_bdrv(job); |
c1cef672 FZ |
584 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
585 | &error_abort); | |
5fe31c25 | 586 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
4ef85a9c KW |
587 | |
588 | /* We just changed the BDS the job BB refers to (with either or both of the | |
5fe31c25 KW |
589 | * bdrv_replace_node() calls), so switch the BB back so the cleanup does |
590 | * the right thing. We don't need any permissions any more now. */ | |
4ef85a9c KW |
591 | blk_remove_bs(job->blk); |
592 | blk_set_perm(job->blk, 0, BLK_PERM_ALL, &error_abort); | |
593 | blk_insert_bs(job->blk, mirror_top_bs, &error_abort); | |
594 | ||
5a7e7a0b | 595 | block_job_completed(&s->common, data->ret); |
4ef85a9c | 596 | |
5a7e7a0b | 597 | g_free(data); |
176c3699 | 598 | bdrv_drained_end(src); |
4ef85a9c | 599 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 600 | bdrv_unref(src); |
5a7e7a0b SH |
601 | } |
602 | ||
49efb1f5 DL |
603 | static void mirror_throttle(MirrorBlockJob *s) |
604 | { | |
605 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
606 | ||
607 | if (now - s->last_pause_ns > SLICE_TIME) { | |
608 | s->last_pause_ns = now; | |
609 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); | |
610 | } else { | |
611 | block_job_pause_point(&s->common); | |
612 | } | |
613 | } | |
614 | ||
c0b363ad DL |
615 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
616 | { | |
617 | int64_t sector_num, end; | |
618 | BlockDriverState *base = s->base; | |
4ef85a9c | 619 | BlockDriverState *bs = s->source; |
c0b363ad | 620 | BlockDriverState *target_bs = blk_bs(s->target); |
c0b363ad DL |
621 | int ret, n; |
622 | ||
623 | end = s->bdev_length / BDRV_SECTOR_SIZE; | |
624 | ||
b7d5062c | 625 | if (base == NULL && !bdrv_has_zero_init(target_bs)) { |
c7c2769c DL |
626 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
627 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, end); | |
628 | return 0; | |
629 | } | |
630 | ||
90ab48eb | 631 | s->initial_zeroing_ongoing = true; |
c7c2769c DL |
632 | for (sector_num = 0; sector_num < end; ) { |
633 | int nb_sectors = MIN(end - sector_num, | |
634 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity) >> BDRV_SECTOR_BITS); | |
635 | ||
636 | mirror_throttle(s); | |
637 | ||
638 | if (block_job_is_cancelled(&s->common)) { | |
90ab48eb | 639 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
640 | return 0; |
641 | } | |
642 | ||
643 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
67adf4b3 EB |
644 | trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, |
645 | s->in_flight); | |
c7c2769c DL |
646 | mirror_wait_for_io(s); |
647 | continue; | |
648 | } | |
649 | ||
650 | mirror_do_zero_or_discard(s, sector_num, nb_sectors, false); | |
651 | sector_num += nb_sectors; | |
652 | } | |
653 | ||
bae8196d | 654 | mirror_wait_for_all_io(s); |
90ab48eb | 655 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
656 | } |
657 | ||
c0b363ad DL |
658 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
659 | for (sector_num = 0; sector_num < end; ) { | |
660 | /* Just to make sure we are not exceeding int limit. */ | |
661 | int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, | |
662 | end - sector_num); | |
663 | ||
664 | mirror_throttle(s); | |
665 | ||
666 | if (block_job_is_cancelled(&s->common)) { | |
667 | return 0; | |
668 | } | |
669 | ||
670 | ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); | |
671 | if (ret < 0) { | |
672 | return ret; | |
673 | } | |
674 | ||
675 | assert(n > 0); | |
b7d5062c | 676 | if (ret == 1) { |
c0b363ad DL |
677 | bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); |
678 | } | |
679 | sector_num += n; | |
680 | } | |
681 | return 0; | |
682 | } | |
683 | ||
bdffb31d PB |
684 | /* Called when going out of the streaming phase to flush the bulk of the |
685 | * data to the medium, or just before completing. | |
686 | */ | |
687 | static int mirror_flush(MirrorBlockJob *s) | |
688 | { | |
689 | int ret = blk_flush(s->target); | |
690 | if (ret < 0) { | |
691 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
692 | s->ret = ret; | |
693 | } | |
694 | } | |
695 | return ret; | |
696 | } | |
697 | ||
893f7eba PB |
698 | static void coroutine_fn mirror_run(void *opaque) |
699 | { | |
700 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 701 | MirrorExitData *data; |
4ef85a9c | 702 | BlockDriverState *bs = s->source; |
e253f4b8 | 703 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 704 | bool need_drain = true; |
c0b363ad | 705 | int64_t length; |
b812f671 | 706 | BlockDriverInfo bdi; |
1d33936e JC |
707 | char backing_filename[2]; /* we only need 2 characters because we are only |
708 | checking for a NULL string */ | |
893f7eba | 709 | int ret = 0; |
e5b43573 | 710 | int target_cluster_size = BDRV_SECTOR_SIZE; |
893f7eba PB |
711 | |
712 | if (block_job_is_cancelled(&s->common)) { | |
713 | goto immediate_exit; | |
714 | } | |
715 | ||
b21c7652 HR |
716 | s->bdev_length = bdrv_getlength(bs); |
717 | if (s->bdev_length < 0) { | |
718 | ret = s->bdev_length; | |
373df5b1 | 719 | goto immediate_exit; |
becc347e KW |
720 | } |
721 | ||
722 | /* Active commit must resize the base image if its size differs from the | |
723 | * active layer. */ | |
724 | if (s->base == blk_bs(s->target)) { | |
725 | int64_t base_length; | |
726 | ||
727 | base_length = blk_getlength(s->target); | |
728 | if (base_length < 0) { | |
729 | ret = base_length; | |
730 | goto immediate_exit; | |
731 | } | |
732 | ||
733 | if (s->bdev_length > base_length) { | |
ed3d2ec9 | 734 | ret = blk_truncate(s->target, s->bdev_length, NULL); |
becc347e KW |
735 | if (ret < 0) { |
736 | goto immediate_exit; | |
737 | } | |
738 | } | |
739 | } | |
740 | ||
741 | if (s->bdev_length == 0) { | |
9e48b025 FZ |
742 | /* Report BLOCK_JOB_READY and wait for complete. */ |
743 | block_job_event_ready(&s->common); | |
744 | s->synced = true; | |
745 | while (!block_job_is_cancelled(&s->common) && !s->should_complete) { | |
746 | block_job_yield(&s->common); | |
747 | } | |
748 | s->common.cancelled = false; | |
749 | goto immediate_exit; | |
893f7eba PB |
750 | } |
751 | ||
b21c7652 | 752 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
753 | s->in_flight_bitmap = bitmap_new(length); |
754 | ||
b812f671 PB |
755 | /* If we have no backing file yet in the destination, we cannot let |
756 | * the destination do COW. Instead, we copy sectors around the | |
757 | * dirty data if needed. We need a bitmap to do that. | |
758 | */ | |
e253f4b8 | 759 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 760 | sizeof(backing_filename)); |
e253f4b8 | 761 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
e5b43573 FZ |
762 | target_cluster_size = bdi.cluster_size; |
763 | } | |
e253f4b8 | 764 | if (backing_filename[0] && !target_bs->backing |
e5b43573 FZ |
765 | && s->granularity < target_cluster_size) { |
766 | s->buf_size = MAX(s->buf_size, target_cluster_size); | |
767 | s->cow_bitmap = bitmap_new(length); | |
b812f671 | 768 | } |
e5b43573 | 769 | s->target_cluster_sectors = target_cluster_size >> BDRV_SECTOR_BITS; |
e253f4b8 | 770 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 771 | |
7504edf4 KW |
772 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
773 | if (s->buf == NULL) { | |
774 | ret = -ENOMEM; | |
775 | goto immediate_exit; | |
776 | } | |
777 | ||
402a4741 | 778 | mirror_free_init(s); |
893f7eba | 779 | |
49efb1f5 | 780 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 781 | if (!s->is_none_mode) { |
c0b363ad DL |
782 | ret = mirror_dirty_init(s); |
783 | if (ret < 0 || block_job_is_cancelled(&s->common)) { | |
784 | goto immediate_exit; | |
893f7eba PB |
785 | } |
786 | } | |
787 | ||
dc162c8e FZ |
788 | assert(!s->dbi); |
789 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap, 0); | |
893f7eba | 790 | for (;;) { |
cc8c9d6c | 791 | uint64_t delay_ns = 0; |
49efb1f5 | 792 | int64_t cnt, delta; |
893f7eba PB |
793 | bool should_complete; |
794 | ||
bd48bde8 PB |
795 | if (s->ret < 0) { |
796 | ret = s->ret; | |
797 | goto immediate_exit; | |
798 | } | |
799 | ||
565ac01f SH |
800 | block_job_pause_point(&s->common); |
801 | ||
20dca810 | 802 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
b21c7652 HR |
803 | /* s->common.offset contains the number of bytes already processed so |
804 | * far, cnt is the number of dirty sectors remaining and | |
805 | * s->sectors_in_flight is the number of sectors currently being | |
806 | * processed; together those are the current total operation length */ | |
807 | s->common.len = s->common.offset + | |
808 | (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
809 | |
810 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 811 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
bd48bde8 PB |
812 | * We do so every SLICE_TIME nanoseconds, or when there is an error, |
813 | * or when the source is clean, whichever comes first. | |
814 | */ | |
49efb1f5 DL |
815 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
816 | if (delta < SLICE_TIME && | |
bd48bde8 | 817 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 818 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 | 819 | (cnt == 0 && s->in_flight > 0)) { |
67adf4b3 | 820 | trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); |
21cd917f | 821 | mirror_wait_for_io(s); |
bd48bde8 PB |
822 | continue; |
823 | } else if (cnt != 0) { | |
cc8c9d6c | 824 | delay_ns = mirror_iteration(s); |
893f7eba | 825 | } |
893f7eba PB |
826 | } |
827 | ||
828 | should_complete = false; | |
bd48bde8 | 829 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 830 | trace_mirror_before_flush(s); |
bdffb31d PB |
831 | if (!s->synced) { |
832 | if (mirror_flush(s) < 0) { | |
833 | /* Go check s->ret. */ | |
834 | continue; | |
b952b558 | 835 | } |
b952b558 PB |
836 | /* We're out of the streaming phase. From now on, if the job |
837 | * is cancelled we will actually complete all pending I/O and | |
838 | * report completion. This way, block-job-cancel will leave | |
839 | * the target in a consistent state. | |
840 | */ | |
bdffb31d PB |
841 | block_job_event_ready(&s->common); |
842 | s->synced = true; | |
d63ffd87 | 843 | } |
bdffb31d PB |
844 | |
845 | should_complete = s->should_complete || | |
846 | block_job_is_cancelled(&s->common); | |
847 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); | |
893f7eba PB |
848 | } |
849 | ||
850 | if (cnt == 0 && should_complete) { | |
851 | /* The dirty bitmap is not updated while operations are pending. | |
852 | * If we're about to exit, wait for pending operations before | |
853 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
854 | * source has dirty data to copy! | |
855 | * | |
856 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
857 | * mirror_populate runs, so pause it now. Before deciding |
858 | * whether to switch to target check one last time if I/O has | |
859 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba PB |
860 | */ |
861 | trace_mirror_before_drain(s, cnt); | |
9a0cec66 PB |
862 | |
863 | bdrv_drained_begin(bs); | |
20dca810 | 864 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 865 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 PB |
866 | bdrv_drained_end(bs); |
867 | continue; | |
868 | } | |
869 | ||
870 | /* The two disks are in sync. Exit and report successful | |
871 | * completion. | |
872 | */ | |
873 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
874 | s->common.cancelled = false; | |
875 | need_drain = false; | |
876 | break; | |
893f7eba PB |
877 | } |
878 | ||
879 | ret = 0; | |
cc8c9d6c | 880 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
d63ffd87 | 881 | if (!s->synced) { |
7483d1e5 | 882 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
883 | if (block_job_is_cancelled(&s->common)) { |
884 | break; | |
885 | } | |
886 | } else if (!should_complete) { | |
bd48bde8 | 887 | delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); |
7483d1e5 | 888 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba | 889 | } |
49efb1f5 | 890 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
891 | } |
892 | ||
893 | immediate_exit: | |
bd48bde8 PB |
894 | if (s->in_flight > 0) { |
895 | /* We get here only if something went wrong. Either the job failed, | |
896 | * or it was cancelled prematurely so that we do not guarantee that | |
897 | * the target is a copy of the source. | |
898 | */ | |
899 | assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); | |
9a0cec66 | 900 | assert(need_drain); |
bae8196d | 901 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
902 | } |
903 | ||
904 | assert(s->in_flight == 0); | |
7191bf31 | 905 | qemu_vfree(s->buf); |
b812f671 | 906 | g_free(s->cow_bitmap); |
402a4741 | 907 | g_free(s->in_flight_bitmap); |
dc162c8e | 908 | bdrv_dirty_iter_free(s->dbi); |
5a7e7a0b SH |
909 | |
910 | data = g_malloc(sizeof(*data)); | |
911 | data->ret = ret; | |
9a0cec66 PB |
912 | |
913 | if (need_drain) { | |
914 | bdrv_drained_begin(bs); | |
915 | } | |
5a7e7a0b | 916 | block_job_defer_to_main_loop(&s->common, mirror_exit, data); |
893f7eba PB |
917 | } |
918 | ||
919 | static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
920 | { | |
921 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
922 | ||
923 | if (speed < 0) { | |
c6bd8c70 | 924 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
893f7eba PB |
925 | return; |
926 | } | |
927 | ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); | |
928 | } | |
929 | ||
d63ffd87 PB |
930 | static void mirror_complete(BlockJob *job, Error **errp) |
931 | { | |
932 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
4ef85a9c | 933 | BlockDriverState *target; |
274fccee | 934 | |
274fccee | 935 | target = blk_bs(s->target); |
d63ffd87 | 936 | |
d63ffd87 | 937 | if (!s->synced) { |
9df229c3 AG |
938 | error_setg(errp, "The active block job '%s' cannot be completed", |
939 | job->id); | |
d63ffd87 PB |
940 | return; |
941 | } | |
942 | ||
274fccee HR |
943 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
944 | int ret; | |
945 | ||
946 | assert(!target->backing); | |
947 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
948 | if (ret < 0) { | |
949 | return; | |
950 | } | |
951 | } | |
952 | ||
15d67298 | 953 | /* block all operations on to_replace bs */ |
09158f00 | 954 | if (s->replaces) { |
5a7e7a0b SH |
955 | AioContext *replace_aio_context; |
956 | ||
e12f3784 | 957 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 958 | if (!s->to_replace) { |
e12f3784 | 959 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
960 | return; |
961 | } | |
962 | ||
5a7e7a0b SH |
963 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
964 | aio_context_acquire(replace_aio_context); | |
965 | ||
4ef85a9c KW |
966 | /* TODO Translate this into permission system. Current definition of |
967 | * GRAPH_MOD would require to request it for the parents; they might | |
968 | * not even be BlockDriverStates, however, so a BdrvChild can't address | |
969 | * them. May need redefinition of GRAPH_MOD. */ | |
09158f00 BC |
970 | error_setg(&s->replace_blocker, |
971 | "block device is in use by block-job-complete"); | |
972 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
973 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
974 | |
975 | aio_context_release(replace_aio_context); | |
09158f00 BC |
976 | } |
977 | ||
d63ffd87 | 978 | s->should_complete = true; |
751ebd76 | 979 | block_job_enter(&s->common); |
d63ffd87 PB |
980 | } |
981 | ||
bae8196d | 982 | static void mirror_pause(BlockJob *job) |
565ac01f SH |
983 | { |
984 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
985 | ||
bae8196d | 986 | mirror_wait_for_all_io(s); |
565ac01f SH |
987 | } |
988 | ||
989 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) | |
990 | { | |
991 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
992 | ||
993 | blk_set_aio_context(s->target, new_context); | |
994 | } | |
995 | ||
bae8196d PB |
996 | static void mirror_drain(BlockJob *job) |
997 | { | |
998 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
999 | ||
1000 | /* Need to keep a reference in case blk_drain triggers execution | |
1001 | * of mirror_complete... | |
1002 | */ | |
1003 | if (s->target) { | |
1004 | BlockBackend *target = s->target; | |
1005 | blk_ref(target); | |
1006 | blk_drain(target); | |
1007 | blk_unref(target); | |
1008 | } | |
1009 | } | |
1010 | ||
3fc4b10a | 1011 | static const BlockJobDriver mirror_job_driver = { |
565ac01f SH |
1012 | .instance_size = sizeof(MirrorBlockJob), |
1013 | .job_type = BLOCK_JOB_TYPE_MIRROR, | |
1014 | .set_speed = mirror_set_speed, | |
a7815a76 | 1015 | .start = mirror_run, |
565ac01f SH |
1016 | .complete = mirror_complete, |
1017 | .pause = mirror_pause, | |
1018 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1019 | .drain = mirror_drain, |
893f7eba PB |
1020 | }; |
1021 | ||
03544a6e | 1022 | static const BlockJobDriver commit_active_job_driver = { |
565ac01f SH |
1023 | .instance_size = sizeof(MirrorBlockJob), |
1024 | .job_type = BLOCK_JOB_TYPE_COMMIT, | |
1025 | .set_speed = mirror_set_speed, | |
a7815a76 | 1026 | .start = mirror_run, |
565ac01f SH |
1027 | .complete = mirror_complete, |
1028 | .pause = mirror_pause, | |
1029 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1030 | .drain = mirror_drain, |
03544a6e FZ |
1031 | }; |
1032 | ||
4ef85a9c KW |
1033 | static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, |
1034 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1035 | { | |
1036 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1037 | } | |
1038 | ||
1039 | static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, | |
1040 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1041 | { | |
1042 | return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1043 | } | |
1044 | ||
1045 | static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) | |
1046 | { | |
1047 | return bdrv_co_flush(bs->backing->bs); | |
1048 | } | |
1049 | ||
1050 | static int64_t coroutine_fn bdrv_mirror_top_get_block_status( | |
1051 | BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum, | |
1052 | BlockDriverState **file) | |
1053 | { | |
1054 | *pnum = nb_sectors; | |
1055 | *file = bs->backing->bs; | |
1056 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID | BDRV_BLOCK_DATA | | |
1057 | (sector_num << BDRV_SECTOR_BITS); | |
1058 | } | |
1059 | ||
1060 | static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, | |
1061 | int64_t offset, int count, BdrvRequestFlags flags) | |
1062 | { | |
1063 | return bdrv_co_pwrite_zeroes(bs->backing, offset, count, flags); | |
1064 | } | |
1065 | ||
1066 | static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, | |
1067 | int64_t offset, int count) | |
1068 | { | |
1069 | return bdrv_co_pdiscard(bs->backing->bs, offset, count); | |
1070 | } | |
1071 | ||
fd4a6493 KW |
1072 | static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) |
1073 | { | |
1074 | bdrv_refresh_filename(bs->backing->bs); | |
1075 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
1076 | bs->backing->bs->filename); | |
1077 | } | |
1078 | ||
4ef85a9c KW |
1079 | static void bdrv_mirror_top_close(BlockDriverState *bs) |
1080 | { | |
1081 | } | |
1082 | ||
1083 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, | |
1084 | const BdrvChildRole *role, | |
1085 | uint64_t perm, uint64_t shared, | |
1086 | uint64_t *nperm, uint64_t *nshared) | |
1087 | { | |
1088 | /* Must be able to forward guest writes to the real image */ | |
1089 | *nperm = 0; | |
1090 | if (perm & BLK_PERM_WRITE) { | |
1091 | *nperm |= BLK_PERM_WRITE; | |
1092 | } | |
1093 | ||
1094 | *nshared = BLK_PERM_ALL; | |
1095 | } | |
1096 | ||
1097 | /* Dummy node that provides consistent read to its users without requiring it | |
1098 | * from its backing file and that allows writes on the backing file chain. */ | |
1099 | static BlockDriver bdrv_mirror_top = { | |
1100 | .format_name = "mirror_top", | |
1101 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1102 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1103 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1104 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1105 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
1106 | .bdrv_co_get_block_status = bdrv_mirror_top_get_block_status, | |
fd4a6493 | 1107 | .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, |
4ef85a9c KW |
1108 | .bdrv_close = bdrv_mirror_top_close, |
1109 | .bdrv_child_perm = bdrv_mirror_top_child_perm, | |
1110 | }; | |
1111 | ||
71aa9867 | 1112 | static void mirror_start_job(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1113 | int creation_flags, BlockDriverState *target, |
1114 | const char *replaces, int64_t speed, | |
1115 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1116 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
1117 | BlockdevOnError on_source_error, |
1118 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1119 | bool unmap, |
097310b5 | 1120 | BlockCompletionFunc *cb, |
51ccfa2d | 1121 | void *opaque, |
09158f00 | 1122 | const BlockJobDriver *driver, |
b49f7ead | 1123 | bool is_none_mode, BlockDriverState *base, |
51ccfa2d FZ |
1124 | bool auto_complete, const char *filter_node_name, |
1125 | Error **errp) | |
893f7eba PB |
1126 | { |
1127 | MirrorBlockJob *s; | |
4ef85a9c KW |
1128 | BlockDriverState *mirror_top_bs; |
1129 | bool target_graph_mod; | |
1130 | bool target_is_backing; | |
b2c2832c | 1131 | Error *local_err = NULL; |
d7086422 | 1132 | int ret; |
893f7eba | 1133 | |
eee13dfe | 1134 | if (granularity == 0) { |
341ebc2f | 1135 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1136 | } |
1137 | ||
1138 | assert ((granularity & (granularity - 1)) == 0); | |
1139 | ||
48ac0a4d WC |
1140 | if (buf_size < 0) { |
1141 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
1142 | return; | |
1143 | } | |
1144 | ||
1145 | if (buf_size == 0) { | |
1146 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
1147 | } | |
5bc361b8 | 1148 | |
4ef85a9c KW |
1149 | /* In the case of active commit, add dummy driver to provide consistent |
1150 | * reads on the top, while disabling it in the intermediate nodes, and make | |
1151 | * the backing chain writable. */ | |
6cdbceb1 KW |
1152 | mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, |
1153 | BDRV_O_RDWR, errp); | |
4ef85a9c KW |
1154 | if (mirror_top_bs == NULL) { |
1155 | return; | |
1156 | } | |
1157 | mirror_top_bs->total_sectors = bs->total_sectors; | |
19dd29e8 | 1158 | bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); |
4ef85a9c KW |
1159 | |
1160 | /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep | |
7a25fcd0 | 1161 | * it alive until block_job_create() succeeds even if bs has no parent. */ |
4ef85a9c KW |
1162 | bdrv_ref(mirror_top_bs); |
1163 | bdrv_drained_begin(bs); | |
b2c2832c | 1164 | bdrv_append(mirror_top_bs, bs, &local_err); |
4ef85a9c KW |
1165 | bdrv_drained_end(bs); |
1166 | ||
b2c2832c KW |
1167 | if (local_err) { |
1168 | bdrv_unref(mirror_top_bs); | |
1169 | error_propagate(errp, local_err); | |
1170 | return; | |
1171 | } | |
1172 | ||
4ef85a9c KW |
1173 | /* Make sure that the source is not resized while the job is running */ |
1174 | s = block_job_create(job_id, driver, mirror_top_bs, | |
1175 | BLK_PERM_CONSISTENT_READ, | |
1176 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
1177 | BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, | |
c6cc12bf | 1178 | creation_flags, cb, opaque, errp); |
893f7eba | 1179 | if (!s) { |
4ef85a9c | 1180 | goto fail; |
893f7eba | 1181 | } |
7a25fcd0 HR |
1182 | /* The block job now has a reference to this node */ |
1183 | bdrv_unref(mirror_top_bs); | |
1184 | ||
4ef85a9c KW |
1185 | s->source = bs; |
1186 | s->mirror_top_bs = mirror_top_bs; | |
1187 | ||
1188 | /* No resize for the target either; while the mirror is still running, a | |
1189 | * consistent read isn't necessarily possible. We could possibly allow | |
1190 | * writes and graph modifications, though it would likely defeat the | |
1191 | * purpose of a mirror, so leave them blocked for now. | |
1192 | * | |
1193 | * In the case of active commit, things look a bit different, though, | |
1194 | * because the target is an already populated backing file in active use. | |
1195 | * We can allow anything except resize there.*/ | |
1196 | target_is_backing = bdrv_chain_contains(bs, target); | |
1197 | target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); | |
1198 | s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | | |
1199 | (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), | |
1200 | BLK_PERM_WRITE_UNCHANGED | | |
1201 | (target_is_backing ? BLK_PERM_CONSISTENT_READ | | |
1202 | BLK_PERM_WRITE | | |
1203 | BLK_PERM_GRAPH_MOD : 0)); | |
d7086422 KW |
1204 | ret = blk_insert_bs(s->target, target, errp); |
1205 | if (ret < 0) { | |
4ef85a9c | 1206 | goto fail; |
d7086422 | 1207 | } |
e253f4b8 | 1208 | |
09158f00 | 1209 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1210 | s->on_source_error = on_source_error; |
1211 | s->on_target_error = on_target_error; | |
03544a6e | 1212 | s->is_none_mode = is_none_mode; |
274fccee | 1213 | s->backing_mode = backing_mode; |
5bc361b8 | 1214 | s->base = base; |
eee13dfe | 1215 | s->granularity = granularity; |
48ac0a4d | 1216 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1217 | s->unmap = unmap; |
b49f7ead WC |
1218 | if (auto_complete) { |
1219 | s->should_complete = true; | |
1220 | } | |
b812f671 | 1221 | |
0db6e54a | 1222 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1223 | if (!s->dirty_bitmap) { |
88f9d1b3 | 1224 | goto fail; |
b8afb520 | 1225 | } |
10f3cd15 | 1226 | |
4ef85a9c | 1227 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1228 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1229 | &error_abort); | |
1230 | ||
f3ede4b0 AG |
1231 | /* In commit_active_start() all intermediate nodes disappear, so |
1232 | * any jobs in them must be blocked */ | |
4ef85a9c | 1233 | if (target_is_backing) { |
f3ede4b0 AG |
1234 | BlockDriverState *iter; |
1235 | for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { | |
4ef85a9c KW |
1236 | /* XXX BLK_PERM_WRITE needs to be allowed so we don't block |
1237 | * ourselves at s->base (if writes are blocked for a node, they are | |
1238 | * also blocked for its backing file). The other options would be a | |
1239 | * second filter driver above s->base (== target). */ | |
1240 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, | |
1241 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, | |
1242 | errp); | |
1243 | if (ret < 0) { | |
1244 | goto fail; | |
1245 | } | |
f3ede4b0 AG |
1246 | } |
1247 | } | |
10f3cd15 | 1248 | |
5ccac6f1 JS |
1249 | trace_mirror_start(bs, s, opaque); |
1250 | block_job_start(&s->common); | |
4ef85a9c KW |
1251 | return; |
1252 | ||
1253 | fail: | |
1254 | if (s) { | |
7a25fcd0 HR |
1255 | /* Make sure this BDS does not go away until we have completed the graph |
1256 | * changes below */ | |
1257 | bdrv_ref(mirror_top_bs); | |
1258 | ||
4ef85a9c KW |
1259 | g_free(s->replaces); |
1260 | blk_unref(s->target); | |
05b0d8e3 | 1261 | block_job_early_fail(&s->common); |
4ef85a9c KW |
1262 | } |
1263 | ||
c1cef672 FZ |
1264 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
1265 | &error_abort); | |
5fe31c25 | 1266 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
7a25fcd0 HR |
1267 | |
1268 | bdrv_unref(mirror_top_bs); | |
893f7eba | 1269 | } |
03544a6e | 1270 | |
71aa9867 AG |
1271 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1272 | BlockDriverState *target, const char *replaces, | |
5fba6c0e | 1273 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
1274 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
1275 | BlockdevOnError on_source_error, | |
03544a6e | 1276 | BlockdevOnError on_target_error, |
6cdbceb1 | 1277 | bool unmap, const char *filter_node_name, Error **errp) |
03544a6e FZ |
1278 | { |
1279 | bool is_none_mode; | |
1280 | BlockDriverState *base; | |
1281 | ||
4b80ab2b JS |
1282 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
1283 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
1284 | return; |
1285 | } | |
03544a6e | 1286 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 1287 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
47970dfb | 1288 | mirror_start_job(job_id, bs, BLOCK_JOB_DEFAULT, target, replaces, |
274fccee | 1289 | speed, granularity, buf_size, backing_mode, |
51ccfa2d | 1290 | on_source_error, on_target_error, unmap, NULL, NULL, |
6cdbceb1 | 1291 | &mirror_job_driver, is_none_mode, base, false, |
51ccfa2d | 1292 | filter_node_name, errp); |
03544a6e FZ |
1293 | } |
1294 | ||
fd62c609 | 1295 | void commit_active_start(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1296 | BlockDriverState *base, int creation_flags, |
1297 | int64_t speed, BlockdevOnError on_error, | |
0db832f4 | 1298 | const char *filter_node_name, |
78bbd910 FZ |
1299 | BlockCompletionFunc *cb, void *opaque, |
1300 | bool auto_complete, Error **errp) | |
03544a6e | 1301 | { |
4da83585 | 1302 | int orig_base_flags; |
cc67f4d1 | 1303 | Error *local_err = NULL; |
4da83585 JC |
1304 | |
1305 | orig_base_flags = bdrv_get_flags(base); | |
1306 | ||
20a63d2c FZ |
1307 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
1308 | return; | |
1309 | } | |
4da83585 | 1310 | |
47970dfb | 1311 | mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
71aa9867 | 1312 | MIRROR_LEAVE_BACKING_CHAIN, |
51ccfa2d | 1313 | on_error, on_error, true, cb, opaque, |
6cdbceb1 | 1314 | &commit_active_job_driver, false, base, auto_complete, |
51ccfa2d | 1315 | filter_node_name, &local_err); |
0fb6395c | 1316 | if (local_err) { |
cc67f4d1 | 1317 | error_propagate(errp, local_err); |
4da83585 JC |
1318 | goto error_restore_flags; |
1319 | } | |
1320 | ||
1321 | return; | |
1322 | ||
1323 | error_restore_flags: | |
1324 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1325 | * the original error */ | |
1326 | bdrv_reopen(base, orig_base_flags, NULL); | |
1327 | return; | |
03544a6e | 1328 | } |