]>
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; |
4ef85a9c KW |
41 | BlockDriverState *mirror_top_bs; |
42 | BlockDriverState *source; | |
5bc361b8 | 43 | BlockDriverState *base; |
4ef85a9c | 44 | |
09158f00 BC |
45 | /* The name of the graph node to replace */ |
46 | char *replaces; | |
47 | /* The BDS to replace */ | |
48 | BlockDriverState *to_replace; | |
49 | /* Used to block operations on the drive-mirror-replace target */ | |
50 | Error *replace_blocker; | |
03544a6e | 51 | bool is_none_mode; |
274fccee | 52 | BlockMirrorBackingMode backing_mode; |
b952b558 | 53 | BlockdevOnError on_source_error, on_target_error; |
d63ffd87 PB |
54 | bool synced; |
55 | bool should_complete; | |
eee13dfe | 56 | int64_t granularity; |
b812f671 | 57 | size_t buf_size; |
b21c7652 | 58 | int64_t bdev_length; |
b812f671 | 59 | unsigned long *cow_bitmap; |
e4654d2d | 60 | BdrvDirtyBitmap *dirty_bitmap; |
dc162c8e | 61 | BdrvDirtyBitmapIter *dbi; |
893f7eba | 62 | uint8_t *buf; |
402a4741 PB |
63 | QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; |
64 | int buf_free_count; | |
bd48bde8 | 65 | |
49efb1f5 | 66 | uint64_t last_pause_ns; |
402a4741 | 67 | unsigned long *in_flight_bitmap; |
bd48bde8 | 68 | int in_flight; |
531509ba | 69 | int64_t sectors_in_flight; |
bd48bde8 | 70 | int ret; |
0fc9f8ea | 71 | bool unmap; |
e424aff5 | 72 | bool waiting_for_io; |
e5b43573 FZ |
73 | int target_cluster_sectors; |
74 | int max_iov; | |
90ab48eb | 75 | bool initial_zeroing_ongoing; |
893f7eba PB |
76 | } MirrorBlockJob; |
77 | ||
bd48bde8 PB |
78 | typedef struct MirrorOp { |
79 | MirrorBlockJob *s; | |
80 | QEMUIOVector qiov; | |
bd48bde8 PB |
81 | int64_t sector_num; |
82 | int nb_sectors; | |
83 | } MirrorOp; | |
84 | ||
b952b558 PB |
85 | static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, |
86 | int error) | |
87 | { | |
88 | s->synced = false; | |
89 | if (read) { | |
81e254dc KW |
90 | return block_job_error_action(&s->common, s->on_source_error, |
91 | true, error); | |
b952b558 | 92 | } else { |
81e254dc KW |
93 | return block_job_error_action(&s->common, s->on_target_error, |
94 | false, error); | |
b952b558 PB |
95 | } |
96 | } | |
97 | ||
bd48bde8 PB |
98 | static void mirror_iteration_done(MirrorOp *op, int ret) |
99 | { | |
100 | MirrorBlockJob *s = op->s; | |
402a4741 | 101 | struct iovec *iov; |
bd48bde8 | 102 | int64_t chunk_num; |
402a4741 | 103 | int i, nb_chunks, sectors_per_chunk; |
bd48bde8 PB |
104 | |
105 | trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret); | |
106 | ||
107 | s->in_flight--; | |
b21c7652 | 108 | s->sectors_in_flight -= op->nb_sectors; |
402a4741 PB |
109 | iov = op->qiov.iov; |
110 | for (i = 0; i < op->qiov.niov; i++) { | |
111 | MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; | |
112 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); | |
113 | s->buf_free_count++; | |
114 | } | |
115 | ||
bd48bde8 PB |
116 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
117 | chunk_num = op->sector_num / sectors_per_chunk; | |
4150ae60 | 118 | nb_chunks = DIV_ROUND_UP(op->nb_sectors, sectors_per_chunk); |
402a4741 | 119 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
b21c7652 HR |
120 | if (ret >= 0) { |
121 | if (s->cow_bitmap) { | |
122 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
123 | } | |
90ab48eb AN |
124 | if (!s->initial_zeroing_ongoing) { |
125 | s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE; | |
126 | } | |
bd48bde8 | 127 | } |
6df3bf8e | 128 | qemu_iovec_destroy(&op->qiov); |
c84b3192 | 129 | g_free(op); |
7b770c72 | 130 | |
e424aff5 | 131 | if (s->waiting_for_io) { |
0b8b8753 | 132 | qemu_coroutine_enter(s->common.co); |
7b770c72 | 133 | } |
bd48bde8 PB |
134 | } |
135 | ||
136 | static void mirror_write_complete(void *opaque, int ret) | |
137 | { | |
138 | MirrorOp *op = opaque; | |
139 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
140 | |
141 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 142 | if (ret < 0) { |
bd48bde8 PB |
143 | BlockErrorAction action; |
144 | ||
20dca810 | 145 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 146 | action = mirror_error_action(s, false, -ret); |
a589569f | 147 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
148 | s->ret = ret; |
149 | } | |
150 | } | |
151 | mirror_iteration_done(op, ret); | |
b9e413dd | 152 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
153 | } |
154 | ||
155 | static void mirror_read_complete(void *opaque, int ret) | |
156 | { | |
157 | MirrorOp *op = opaque; | |
158 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
159 | |
160 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 161 | if (ret < 0) { |
bd48bde8 PB |
162 | BlockErrorAction action; |
163 | ||
20dca810 | 164 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); |
bd48bde8 | 165 | action = mirror_error_action(s, true, -ret); |
a589569f | 166 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
167 | s->ret = ret; |
168 | } | |
169 | ||
170 | mirror_iteration_done(op, ret); | |
b9e413dd PB |
171 | } else { |
172 | blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov, | |
173 | 0, mirror_write_complete, op); | |
bd48bde8 | 174 | } |
b9e413dd | 175 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
176 | } |
177 | ||
4150ae60 FZ |
178 | static inline void mirror_clip_sectors(MirrorBlockJob *s, |
179 | int64_t sector_num, | |
180 | int *nb_sectors) | |
181 | { | |
182 | *nb_sectors = MIN(*nb_sectors, | |
183 | s->bdev_length / BDRV_SECTOR_SIZE - sector_num); | |
184 | } | |
185 | ||
e5b43573 FZ |
186 | /* Round sector_num and/or nb_sectors to target cluster if COW is needed, and |
187 | * return the offset of the adjusted tail sector against original. */ | |
188 | static int mirror_cow_align(MirrorBlockJob *s, | |
189 | int64_t *sector_num, | |
190 | int *nb_sectors) | |
893f7eba | 191 | { |
e5b43573 FZ |
192 | bool need_cow; |
193 | int ret = 0; | |
194 | int chunk_sectors = s->granularity >> BDRV_SECTOR_BITS; | |
195 | int64_t align_sector_num = *sector_num; | |
196 | int align_nb_sectors = *nb_sectors; | |
197 | int max_sectors = chunk_sectors * s->max_iov; | |
198 | ||
199 | need_cow = !test_bit(*sector_num / chunk_sectors, s->cow_bitmap); | |
200 | need_cow |= !test_bit((*sector_num + *nb_sectors - 1) / chunk_sectors, | |
201 | s->cow_bitmap); | |
202 | if (need_cow) { | |
244483e6 KW |
203 | bdrv_round_sectors_to_clusters(blk_bs(s->target), *sector_num, |
204 | *nb_sectors, &align_sector_num, | |
205 | &align_nb_sectors); | |
e5b43573 | 206 | } |
3515727f | 207 | |
e5b43573 FZ |
208 | if (align_nb_sectors > max_sectors) { |
209 | align_nb_sectors = max_sectors; | |
210 | if (need_cow) { | |
211 | align_nb_sectors = QEMU_ALIGN_DOWN(align_nb_sectors, | |
212 | s->target_cluster_sectors); | |
213 | } | |
8f0720ec | 214 | } |
4150ae60 FZ |
215 | /* Clipping may result in align_nb_sectors unaligned to chunk boundary, but |
216 | * that doesn't matter because it's already the end of source image. */ | |
217 | mirror_clip_sectors(s, align_sector_num, &align_nb_sectors); | |
8f0720ec | 218 | |
e5b43573 FZ |
219 | ret = align_sector_num + align_nb_sectors - (*sector_num + *nb_sectors); |
220 | *sector_num = align_sector_num; | |
221 | *nb_sectors = align_nb_sectors; | |
222 | assert(ret >= 0); | |
223 | return ret; | |
224 | } | |
225 | ||
21cd917f FZ |
226 | static inline void mirror_wait_for_io(MirrorBlockJob *s) |
227 | { | |
228 | assert(!s->waiting_for_io); | |
229 | s->waiting_for_io = true; | |
230 | qemu_coroutine_yield(); | |
231 | s->waiting_for_io = false; | |
232 | } | |
233 | ||
e5b43573 | 234 | /* Submit async read while handling COW. |
17612955 JS |
235 | * Returns: The number of sectors copied after and including sector_num, |
236 | * excluding any sectors copied prior to sector_num due to alignment. | |
237 | * This will be nb_sectors if no alignment is necessary, or | |
e5b43573 FZ |
238 | * (new_end - sector_num) if tail is rounded up or down due to |
239 | * alignment or buffer limit. | |
240 | */ | |
241 | static int mirror_do_read(MirrorBlockJob *s, int64_t sector_num, | |
242 | int nb_sectors) | |
243 | { | |
e253f4b8 | 244 | BlockBackend *source = s->common.blk; |
e5b43573 | 245 | int sectors_per_chunk, nb_chunks; |
17612955 | 246 | int ret; |
e5b43573 | 247 | MirrorOp *op; |
e4808881 | 248 | int max_sectors; |
e5b43573 | 249 | |
884fea4e | 250 | sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
e4808881 | 251 | max_sectors = sectors_per_chunk * s->max_iov; |
402a4741 | 252 | |
e5b43573 FZ |
253 | /* We can only handle as much as buf_size at a time. */ |
254 | nb_sectors = MIN(s->buf_size >> BDRV_SECTOR_BITS, nb_sectors); | |
e4808881 | 255 | nb_sectors = MIN(max_sectors, nb_sectors); |
e5b43573 | 256 | assert(nb_sectors); |
17612955 | 257 | ret = nb_sectors; |
402a4741 | 258 | |
e5b43573 FZ |
259 | if (s->cow_bitmap) { |
260 | ret += mirror_cow_align(s, §or_num, &nb_sectors); | |
261 | } | |
262 | assert(nb_sectors << BDRV_SECTOR_BITS <= s->buf_size); | |
263 | /* The sector range must meet granularity because: | |
264 | * 1) Caller passes in aligned values; | |
265 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
e5b43573 | 266 | assert(!(sector_num % sectors_per_chunk)); |
4150ae60 | 267 | nb_chunks = DIV_ROUND_UP(nb_sectors, sectors_per_chunk); |
e5b43573 FZ |
268 | |
269 | while (s->buf_free_count < nb_chunks) { | |
402a4741 | 270 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); |
21cd917f | 271 | mirror_wait_for_io(s); |
b812f671 PB |
272 | } |
273 | ||
bd48bde8 | 274 | /* Allocate a MirrorOp that is used as an AIO callback. */ |
c84b3192 | 275 | op = g_new(MirrorOp, 1); |
bd48bde8 | 276 | op->s = s; |
bd48bde8 PB |
277 | op->sector_num = sector_num; |
278 | op->nb_sectors = nb_sectors; | |
402a4741 PB |
279 | |
280 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
281 | * from s->buf_free. | |
282 | */ | |
283 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
284 | while (nb_chunks-- > 0) { |
285 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
e5b43573 | 286 | size_t remaining = nb_sectors * BDRV_SECTOR_SIZE - op->qiov.size; |
5a0f6fd5 | 287 | |
402a4741 PB |
288 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
289 | s->buf_free_count--; | |
5a0f6fd5 | 290 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 291 | } |
bd48bde8 | 292 | |
893f7eba | 293 | /* Copy the dirty cluster. */ |
bd48bde8 | 294 | s->in_flight++; |
b21c7652 | 295 | s->sectors_in_flight += nb_sectors; |
b812f671 | 296 | trace_mirror_one_iteration(s, sector_num, nb_sectors); |
dcfb3beb | 297 | |
73698c30 | 298 | blk_aio_preadv(source, sector_num * BDRV_SECTOR_SIZE, &op->qiov, 0, |
e5b43573 FZ |
299 | mirror_read_complete, op); |
300 | return ret; | |
301 | } | |
302 | ||
303 | static void mirror_do_zero_or_discard(MirrorBlockJob *s, | |
304 | int64_t sector_num, | |
305 | int nb_sectors, | |
306 | bool is_discard) | |
307 | { | |
308 | MirrorOp *op; | |
309 | ||
310 | /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed | |
311 | * so the freeing in mirror_iteration_done is nop. */ | |
312 | op = g_new0(MirrorOp, 1); | |
313 | op->s = s; | |
314 | op->sector_num = sector_num; | |
315 | op->nb_sectors = nb_sectors; | |
316 | ||
317 | s->in_flight++; | |
318 | s->sectors_in_flight += nb_sectors; | |
319 | if (is_discard) { | |
1c6c4bb7 EB |
320 | blk_aio_pdiscard(s->target, sector_num << BDRV_SECTOR_BITS, |
321 | op->nb_sectors << BDRV_SECTOR_BITS, | |
322 | mirror_write_complete, op); | |
e5b43573 | 323 | } else { |
e253f4b8 KW |
324 | blk_aio_pwrite_zeroes(s->target, sector_num * BDRV_SECTOR_SIZE, |
325 | op->nb_sectors * BDRV_SECTOR_SIZE, | |
dcfb3beb FZ |
326 | s->unmap ? BDRV_REQ_MAY_UNMAP : 0, |
327 | mirror_write_complete, op); | |
e5b43573 FZ |
328 | } |
329 | } | |
330 | ||
331 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) | |
332 | { | |
4ef85a9c | 333 | BlockDriverState *source = s->source; |
9c83625b | 334 | int64_t sector_num, first_chunk; |
e5b43573 FZ |
335 | uint64_t delay_ns = 0; |
336 | /* At least the first dirty chunk is mirrored in one iteration. */ | |
337 | int nb_chunks = 1; | |
338 | int64_t end = s->bdev_length / BDRV_SECTOR_SIZE; | |
339 | int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; | |
4b5004d9 | 340 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
0965a41e VSO |
341 | int max_io_sectors = MAX((s->buf_size >> BDRV_SECTOR_BITS) / MAX_IN_FLIGHT, |
342 | MAX_IO_SECTORS); | |
e5b43573 | 343 | |
dc162c8e | 344 | sector_num = bdrv_dirty_iter_next(s->dbi); |
e5b43573 | 345 | if (sector_num < 0) { |
dc162c8e FZ |
346 | bdrv_set_dirty_iter(s->dbi, 0); |
347 | sector_num = bdrv_dirty_iter_next(s->dbi); | |
e5b43573 FZ |
348 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
349 | assert(sector_num >= 0); | |
350 | } | |
351 | ||
9c83625b HR |
352 | first_chunk = sector_num / sectors_per_chunk; |
353 | while (test_bit(first_chunk, s->in_flight_bitmap)) { | |
ff04198b | 354 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); |
9c83625b HR |
355 | mirror_wait_for_io(s); |
356 | } | |
357 | ||
565ac01f SH |
358 | block_job_pause_point(&s->common); |
359 | ||
e5b43573 FZ |
360 | /* Find the number of consective dirty chunks following the first dirty |
361 | * one, and wait for in flight requests in them. */ | |
362 | while (nb_chunks * sectors_per_chunk < (s->buf_size >> BDRV_SECTOR_BITS)) { | |
dc162c8e | 363 | int64_t next_dirty; |
e5b43573 FZ |
364 | int64_t next_sector = sector_num + nb_chunks * sectors_per_chunk; |
365 | int64_t next_chunk = next_sector / sectors_per_chunk; | |
366 | if (next_sector >= end || | |
367 | !bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) { | |
368 | break; | |
369 | } | |
370 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 371 | break; |
e5b43573 | 372 | } |
9c83625b | 373 | |
dc162c8e FZ |
374 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
375 | if (next_dirty > next_sector || next_dirty < 0) { | |
f27a2742 | 376 | /* The bitmap iterator's cache is stale, refresh it */ |
dc162c8e FZ |
377 | bdrv_set_dirty_iter(s->dbi, next_sector); |
378 | next_dirty = bdrv_dirty_iter_next(s->dbi); | |
f27a2742 | 379 | } |
dc162c8e | 380 | assert(next_dirty == next_sector); |
9c83625b | 381 | nb_chunks++; |
e5b43573 FZ |
382 | } |
383 | ||
384 | /* Clear dirty bits before querying the block status, because | |
385 | * calling bdrv_get_block_status_above could yield - if some blocks are | |
386 | * marked dirty in this window, we need to know. | |
387 | */ | |
388 | bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, | |
389 | nb_chunks * sectors_per_chunk); | |
390 | bitmap_set(s->in_flight_bitmap, sector_num / sectors_per_chunk, nb_chunks); | |
391 | while (nb_chunks > 0 && sector_num < end) { | |
39c11580 | 392 | int64_t ret; |
4b5004d9 | 393 | int io_sectors, io_sectors_acct; |
e5b43573 FZ |
394 | BlockDriverState *file; |
395 | enum MirrorMethod { | |
396 | MIRROR_METHOD_COPY, | |
397 | MIRROR_METHOD_ZERO, | |
398 | MIRROR_METHOD_DISCARD | |
399 | } mirror_method = MIRROR_METHOD_COPY; | |
400 | ||
401 | assert(!(sector_num % sectors_per_chunk)); | |
402 | ret = bdrv_get_block_status_above(source, NULL, sector_num, | |
403 | nb_chunks * sectors_per_chunk, | |
404 | &io_sectors, &file); | |
405 | if (ret < 0) { | |
0965a41e VSO |
406 | io_sectors = MIN(nb_chunks * sectors_per_chunk, max_io_sectors); |
407 | } else if (ret & BDRV_BLOCK_DATA) { | |
408 | io_sectors = MIN(io_sectors, max_io_sectors); | |
e5b43573 FZ |
409 | } |
410 | ||
411 | io_sectors -= io_sectors % sectors_per_chunk; | |
412 | if (io_sectors < sectors_per_chunk) { | |
413 | io_sectors = sectors_per_chunk; | |
414 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { | |
415 | int64_t target_sector_num; | |
416 | int target_nb_sectors; | |
244483e6 KW |
417 | bdrv_round_sectors_to_clusters(blk_bs(s->target), sector_num, |
418 | io_sectors, &target_sector_num, | |
419 | &target_nb_sectors); | |
e5b43573 FZ |
420 | if (target_sector_num == sector_num && |
421 | target_nb_sectors == io_sectors) { | |
422 | mirror_method = ret & BDRV_BLOCK_ZERO ? | |
423 | MIRROR_METHOD_ZERO : | |
424 | MIRROR_METHOD_DISCARD; | |
425 | } | |
426 | } | |
427 | ||
cf56a3c6 DL |
428 | while (s->in_flight >= MAX_IN_FLIGHT) { |
429 | trace_mirror_yield_in_flight(s, sector_num, s->in_flight); | |
430 | mirror_wait_for_io(s); | |
431 | } | |
432 | ||
dbaa7b57 VSO |
433 | if (s->ret < 0) { |
434 | return 0; | |
435 | } | |
436 | ||
4150ae60 | 437 | mirror_clip_sectors(s, sector_num, &io_sectors); |
e5b43573 FZ |
438 | switch (mirror_method) { |
439 | case MIRROR_METHOD_COPY: | |
440 | io_sectors = mirror_do_read(s, sector_num, io_sectors); | |
4b5004d9 | 441 | io_sectors_acct = io_sectors; |
e5b43573 FZ |
442 | break; |
443 | case MIRROR_METHOD_ZERO: | |
e5b43573 | 444 | case MIRROR_METHOD_DISCARD: |
4b5004d9 DL |
445 | mirror_do_zero_or_discard(s, sector_num, io_sectors, |
446 | mirror_method == MIRROR_METHOD_DISCARD); | |
447 | if (write_zeroes_ok) { | |
448 | io_sectors_acct = 0; | |
449 | } else { | |
450 | io_sectors_acct = io_sectors; | |
451 | } | |
e5b43573 FZ |
452 | break; |
453 | default: | |
454 | abort(); | |
455 | } | |
456 | assert(io_sectors); | |
457 | sector_num += io_sectors; | |
4150ae60 | 458 | nb_chunks -= DIV_ROUND_UP(io_sectors, sectors_per_chunk); |
f14a39cc | 459 | if (s->common.speed) { |
4b5004d9 | 460 | delay_ns = ratelimit_calculate_delay(&s->limit, io_sectors_acct); |
f14a39cc | 461 | } |
dcfb3beb | 462 | } |
cc8c9d6c | 463 | return delay_ns; |
bd48bde8 | 464 | } |
b952b558 | 465 | |
402a4741 PB |
466 | static void mirror_free_init(MirrorBlockJob *s) |
467 | { | |
468 | int granularity = s->granularity; | |
469 | size_t buf_size = s->buf_size; | |
470 | uint8_t *buf = s->buf; | |
471 | ||
472 | assert(s->buf_free_count == 0); | |
473 | QSIMPLEQ_INIT(&s->buf_free); | |
474 | while (buf_size != 0) { | |
475 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
476 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
477 | s->buf_free_count++; | |
478 | buf_size -= granularity; | |
479 | buf += granularity; | |
480 | } | |
481 | } | |
482 | ||
bae8196d PB |
483 | /* This is also used for the .pause callback. There is no matching |
484 | * mirror_resume() because mirror_run() will begin iterating again | |
485 | * when the job is resumed. | |
486 | */ | |
487 | static void mirror_wait_for_all_io(MirrorBlockJob *s) | |
bd48bde8 PB |
488 | { |
489 | while (s->in_flight > 0) { | |
21cd917f | 490 | mirror_wait_for_io(s); |
bd48bde8 | 491 | } |
893f7eba PB |
492 | } |
493 | ||
5a7e7a0b SH |
494 | typedef struct { |
495 | int ret; | |
496 | } MirrorExitData; | |
497 | ||
498 | static void mirror_exit(BlockJob *job, void *opaque) | |
499 | { | |
500 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
501 | MirrorExitData *data = opaque; | |
502 | AioContext *replace_aio_context = NULL; | |
4ef85a9c | 503 | BlockDriverState *src = s->source; |
e253f4b8 | 504 | BlockDriverState *target_bs = blk_bs(s->target); |
4ef85a9c | 505 | BlockDriverState *mirror_top_bs = s->mirror_top_bs; |
3f09bfbc KW |
506 | |
507 | /* Make sure that the source BDS doesn't go away before we called | |
508 | * block_job_completed(). */ | |
509 | bdrv_ref(src); | |
4ef85a9c KW |
510 | bdrv_ref(mirror_top_bs); |
511 | ||
512 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
513 | * required before it could become a backing file of target_bs. */ | |
514 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, | |
515 | &error_abort); | |
516 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { | |
517 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
518 | if (backing_bs(target_bs) != backing) { | |
519 | bdrv_set_backing_hd(target_bs, backing); | |
520 | } | |
521 | } | |
5a7e7a0b SH |
522 | |
523 | if (s->to_replace) { | |
524 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
525 | aio_context_acquire(replace_aio_context); | |
526 | } | |
527 | ||
528 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 529 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
530 | if (s->to_replace) { |
531 | to_replace = s->to_replace; | |
532 | } | |
40365552 | 533 | |
e253f4b8 KW |
534 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
535 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 536 | } |
b8804815 KW |
537 | |
538 | /* The mirror job has no requests in flight any more, but we need to | |
539 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 KW |
540 | bdrv_drained_begin(target_bs); |
541 | bdrv_replace_in_backing_chain(to_replace, target_bs); | |
542 | bdrv_drained_end(target_bs); | |
5a7e7a0b SH |
543 | } |
544 | if (s->to_replace) { | |
545 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
546 | error_free(s->replace_blocker); | |
547 | bdrv_unref(s->to_replace); | |
548 | } | |
549 | if (replace_aio_context) { | |
550 | aio_context_release(replace_aio_context); | |
551 | } | |
552 | g_free(s->replaces); | |
e253f4b8 | 553 | blk_unref(s->target); |
bae8196d | 554 | s->target = NULL; |
4ef85a9c KW |
555 | |
556 | /* Remove the mirror filter driver from the graph. Before this, get rid of | |
557 | * the blockers on the intermediate nodes so that the resulting state is | |
558 | * valid. */ | |
559 | block_job_remove_all_bdrv(job); | |
560 | bdrv_replace_in_backing_chain(mirror_top_bs, backing_bs(mirror_top_bs)); | |
561 | ||
562 | /* We just changed the BDS the job BB refers to (with either or both of the | |
563 | * bdrv_replace_in_backing_chain() calls), so switch the BB back so the | |
564 | * cleanup does the right thing. We don't need any permissions any more | |
565 | * now. */ | |
566 | blk_remove_bs(job->blk); | |
567 | blk_set_perm(job->blk, 0, BLK_PERM_ALL, &error_abort); | |
568 | blk_insert_bs(job->blk, mirror_top_bs, &error_abort); | |
569 | ||
5a7e7a0b | 570 | block_job_completed(&s->common, data->ret); |
4ef85a9c | 571 | |
5a7e7a0b | 572 | g_free(data); |
176c3699 | 573 | bdrv_drained_end(src); |
4ef85a9c | 574 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 575 | bdrv_unref(src); |
5a7e7a0b SH |
576 | } |
577 | ||
49efb1f5 DL |
578 | static void mirror_throttle(MirrorBlockJob *s) |
579 | { | |
580 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
581 | ||
582 | if (now - s->last_pause_ns > SLICE_TIME) { | |
583 | s->last_pause_ns = now; | |
584 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); | |
585 | } else { | |
586 | block_job_pause_point(&s->common); | |
587 | } | |
588 | } | |
589 | ||
c0b363ad DL |
590 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
591 | { | |
592 | int64_t sector_num, end; | |
593 | BlockDriverState *base = s->base; | |
4ef85a9c | 594 | BlockDriverState *bs = s->source; |
c0b363ad | 595 | BlockDriverState *target_bs = blk_bs(s->target); |
c0b363ad DL |
596 | int ret, n; |
597 | ||
598 | end = s->bdev_length / BDRV_SECTOR_SIZE; | |
599 | ||
b7d5062c | 600 | if (base == NULL && !bdrv_has_zero_init(target_bs)) { |
c7c2769c DL |
601 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
602 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, end); | |
603 | return 0; | |
604 | } | |
605 | ||
90ab48eb | 606 | s->initial_zeroing_ongoing = true; |
c7c2769c DL |
607 | for (sector_num = 0; sector_num < end; ) { |
608 | int nb_sectors = MIN(end - sector_num, | |
609 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity) >> BDRV_SECTOR_BITS); | |
610 | ||
611 | mirror_throttle(s); | |
612 | ||
613 | if (block_job_is_cancelled(&s->common)) { | |
90ab48eb | 614 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
615 | return 0; |
616 | } | |
617 | ||
618 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
619 | trace_mirror_yield(s, s->in_flight, s->buf_free_count, -1); | |
620 | mirror_wait_for_io(s); | |
621 | continue; | |
622 | } | |
623 | ||
624 | mirror_do_zero_or_discard(s, sector_num, nb_sectors, false); | |
625 | sector_num += nb_sectors; | |
626 | } | |
627 | ||
bae8196d | 628 | mirror_wait_for_all_io(s); |
90ab48eb | 629 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
630 | } |
631 | ||
c0b363ad DL |
632 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
633 | for (sector_num = 0; sector_num < end; ) { | |
634 | /* Just to make sure we are not exceeding int limit. */ | |
635 | int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, | |
636 | end - sector_num); | |
637 | ||
638 | mirror_throttle(s); | |
639 | ||
640 | if (block_job_is_cancelled(&s->common)) { | |
641 | return 0; | |
642 | } | |
643 | ||
644 | ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); | |
645 | if (ret < 0) { | |
646 | return ret; | |
647 | } | |
648 | ||
649 | assert(n > 0); | |
b7d5062c | 650 | if (ret == 1) { |
c0b363ad DL |
651 | bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); |
652 | } | |
653 | sector_num += n; | |
654 | } | |
655 | return 0; | |
656 | } | |
657 | ||
bdffb31d PB |
658 | /* Called when going out of the streaming phase to flush the bulk of the |
659 | * data to the medium, or just before completing. | |
660 | */ | |
661 | static int mirror_flush(MirrorBlockJob *s) | |
662 | { | |
663 | int ret = blk_flush(s->target); | |
664 | if (ret < 0) { | |
665 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
666 | s->ret = ret; | |
667 | } | |
668 | } | |
669 | return ret; | |
670 | } | |
671 | ||
893f7eba PB |
672 | static void coroutine_fn mirror_run(void *opaque) |
673 | { | |
674 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 675 | MirrorExitData *data; |
4ef85a9c | 676 | BlockDriverState *bs = s->source; |
e253f4b8 | 677 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 678 | bool need_drain = true; |
c0b363ad | 679 | int64_t length; |
b812f671 | 680 | BlockDriverInfo bdi; |
1d33936e JC |
681 | char backing_filename[2]; /* we only need 2 characters because we are only |
682 | checking for a NULL string */ | |
893f7eba | 683 | int ret = 0; |
e5b43573 | 684 | int target_cluster_size = BDRV_SECTOR_SIZE; |
893f7eba PB |
685 | |
686 | if (block_job_is_cancelled(&s->common)) { | |
687 | goto immediate_exit; | |
688 | } | |
689 | ||
b21c7652 HR |
690 | s->bdev_length = bdrv_getlength(bs); |
691 | if (s->bdev_length < 0) { | |
692 | ret = s->bdev_length; | |
373df5b1 | 693 | goto immediate_exit; |
becc347e KW |
694 | } |
695 | ||
696 | /* Active commit must resize the base image if its size differs from the | |
697 | * active layer. */ | |
698 | if (s->base == blk_bs(s->target)) { | |
699 | int64_t base_length; | |
700 | ||
701 | base_length = blk_getlength(s->target); | |
702 | if (base_length < 0) { | |
703 | ret = base_length; | |
704 | goto immediate_exit; | |
705 | } | |
706 | ||
707 | if (s->bdev_length > base_length) { | |
708 | ret = blk_truncate(s->target, s->bdev_length); | |
709 | if (ret < 0) { | |
710 | goto immediate_exit; | |
711 | } | |
712 | } | |
713 | } | |
714 | ||
715 | if (s->bdev_length == 0) { | |
9e48b025 FZ |
716 | /* Report BLOCK_JOB_READY and wait for complete. */ |
717 | block_job_event_ready(&s->common); | |
718 | s->synced = true; | |
719 | while (!block_job_is_cancelled(&s->common) && !s->should_complete) { | |
720 | block_job_yield(&s->common); | |
721 | } | |
722 | s->common.cancelled = false; | |
723 | goto immediate_exit; | |
893f7eba PB |
724 | } |
725 | ||
b21c7652 | 726 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
727 | s->in_flight_bitmap = bitmap_new(length); |
728 | ||
b812f671 PB |
729 | /* If we have no backing file yet in the destination, we cannot let |
730 | * the destination do COW. Instead, we copy sectors around the | |
731 | * dirty data if needed. We need a bitmap to do that. | |
732 | */ | |
e253f4b8 | 733 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 734 | sizeof(backing_filename)); |
e253f4b8 | 735 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
e5b43573 FZ |
736 | target_cluster_size = bdi.cluster_size; |
737 | } | |
e253f4b8 | 738 | if (backing_filename[0] && !target_bs->backing |
e5b43573 FZ |
739 | && s->granularity < target_cluster_size) { |
740 | s->buf_size = MAX(s->buf_size, target_cluster_size); | |
741 | s->cow_bitmap = bitmap_new(length); | |
b812f671 | 742 | } |
e5b43573 | 743 | s->target_cluster_sectors = target_cluster_size >> BDRV_SECTOR_BITS; |
e253f4b8 | 744 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 745 | |
7504edf4 KW |
746 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
747 | if (s->buf == NULL) { | |
748 | ret = -ENOMEM; | |
749 | goto immediate_exit; | |
750 | } | |
751 | ||
402a4741 | 752 | mirror_free_init(s); |
893f7eba | 753 | |
49efb1f5 | 754 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 755 | if (!s->is_none_mode) { |
c0b363ad DL |
756 | ret = mirror_dirty_init(s); |
757 | if (ret < 0 || block_job_is_cancelled(&s->common)) { | |
758 | goto immediate_exit; | |
893f7eba PB |
759 | } |
760 | } | |
761 | ||
dc162c8e FZ |
762 | assert(!s->dbi); |
763 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap, 0); | |
893f7eba | 764 | for (;;) { |
cc8c9d6c | 765 | uint64_t delay_ns = 0; |
49efb1f5 | 766 | int64_t cnt, delta; |
893f7eba PB |
767 | bool should_complete; |
768 | ||
bd48bde8 PB |
769 | if (s->ret < 0) { |
770 | ret = s->ret; | |
771 | goto immediate_exit; | |
772 | } | |
773 | ||
565ac01f SH |
774 | block_job_pause_point(&s->common); |
775 | ||
20dca810 | 776 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
b21c7652 HR |
777 | /* s->common.offset contains the number of bytes already processed so |
778 | * far, cnt is the number of dirty sectors remaining and | |
779 | * s->sectors_in_flight is the number of sectors currently being | |
780 | * processed; together those are the current total operation length */ | |
781 | s->common.len = s->common.offset + | |
782 | (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE; | |
bd48bde8 PB |
783 | |
784 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 785 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
bd48bde8 PB |
786 | * We do so every SLICE_TIME nanoseconds, or when there is an error, |
787 | * or when the source is clean, whichever comes first. | |
788 | */ | |
49efb1f5 DL |
789 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
790 | if (delta < SLICE_TIME && | |
bd48bde8 | 791 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 792 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 PB |
793 | (cnt == 0 && s->in_flight > 0)) { |
794 | trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt); | |
21cd917f | 795 | mirror_wait_for_io(s); |
bd48bde8 PB |
796 | continue; |
797 | } else if (cnt != 0) { | |
cc8c9d6c | 798 | delay_ns = mirror_iteration(s); |
893f7eba | 799 | } |
893f7eba PB |
800 | } |
801 | ||
802 | should_complete = false; | |
bd48bde8 | 803 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 804 | trace_mirror_before_flush(s); |
bdffb31d PB |
805 | if (!s->synced) { |
806 | if (mirror_flush(s) < 0) { | |
807 | /* Go check s->ret. */ | |
808 | continue; | |
b952b558 | 809 | } |
b952b558 PB |
810 | /* We're out of the streaming phase. From now on, if the job |
811 | * is cancelled we will actually complete all pending I/O and | |
812 | * report completion. This way, block-job-cancel will leave | |
813 | * the target in a consistent state. | |
814 | */ | |
bdffb31d PB |
815 | block_job_event_ready(&s->common); |
816 | s->synced = true; | |
d63ffd87 | 817 | } |
bdffb31d PB |
818 | |
819 | should_complete = s->should_complete || | |
820 | block_job_is_cancelled(&s->common); | |
821 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); | |
893f7eba PB |
822 | } |
823 | ||
824 | if (cnt == 0 && should_complete) { | |
825 | /* The dirty bitmap is not updated while operations are pending. | |
826 | * If we're about to exit, wait for pending operations before | |
827 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
828 | * source has dirty data to copy! | |
829 | * | |
830 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
831 | * mirror_populate runs, so pause it now. Before deciding |
832 | * whether to switch to target check one last time if I/O has | |
833 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba PB |
834 | */ |
835 | trace_mirror_before_drain(s, cnt); | |
9a0cec66 PB |
836 | |
837 | bdrv_drained_begin(bs); | |
20dca810 | 838 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 839 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 PB |
840 | bdrv_drained_end(bs); |
841 | continue; | |
842 | } | |
843 | ||
844 | /* The two disks are in sync. Exit and report successful | |
845 | * completion. | |
846 | */ | |
847 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
848 | s->common.cancelled = false; | |
849 | need_drain = false; | |
850 | break; | |
893f7eba PB |
851 | } |
852 | ||
853 | ret = 0; | |
cc8c9d6c | 854 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
d63ffd87 | 855 | if (!s->synced) { |
7483d1e5 | 856 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
857 | if (block_job_is_cancelled(&s->common)) { |
858 | break; | |
859 | } | |
860 | } else if (!should_complete) { | |
bd48bde8 | 861 | delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); |
7483d1e5 | 862 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba | 863 | } |
49efb1f5 | 864 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
865 | } |
866 | ||
867 | immediate_exit: | |
bd48bde8 PB |
868 | if (s->in_flight > 0) { |
869 | /* We get here only if something went wrong. Either the job failed, | |
870 | * or it was cancelled prematurely so that we do not guarantee that | |
871 | * the target is a copy of the source. | |
872 | */ | |
873 | assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); | |
9a0cec66 | 874 | assert(need_drain); |
bae8196d | 875 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
876 | } |
877 | ||
878 | assert(s->in_flight == 0); | |
7191bf31 | 879 | qemu_vfree(s->buf); |
b812f671 | 880 | g_free(s->cow_bitmap); |
402a4741 | 881 | g_free(s->in_flight_bitmap); |
dc162c8e | 882 | bdrv_dirty_iter_free(s->dbi); |
e4654d2d | 883 | bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); |
5a7e7a0b SH |
884 | |
885 | data = g_malloc(sizeof(*data)); | |
886 | data->ret = ret; | |
9a0cec66 PB |
887 | |
888 | if (need_drain) { | |
889 | bdrv_drained_begin(bs); | |
890 | } | |
5a7e7a0b | 891 | block_job_defer_to_main_loop(&s->common, mirror_exit, data); |
893f7eba PB |
892 | } |
893 | ||
894 | static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
895 | { | |
896 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
897 | ||
898 | if (speed < 0) { | |
c6bd8c70 | 899 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
893f7eba PB |
900 | return; |
901 | } | |
902 | ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); | |
903 | } | |
904 | ||
d63ffd87 PB |
905 | static void mirror_complete(BlockJob *job, Error **errp) |
906 | { | |
907 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
4ef85a9c | 908 | BlockDriverState *target; |
274fccee | 909 | |
274fccee | 910 | target = blk_bs(s->target); |
d63ffd87 | 911 | |
d63ffd87 | 912 | if (!s->synced) { |
9df229c3 AG |
913 | error_setg(errp, "The active block job '%s' cannot be completed", |
914 | job->id); | |
d63ffd87 PB |
915 | return; |
916 | } | |
917 | ||
274fccee HR |
918 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
919 | int ret; | |
920 | ||
921 | assert(!target->backing); | |
922 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
923 | if (ret < 0) { | |
924 | return; | |
925 | } | |
926 | } | |
927 | ||
15d67298 | 928 | /* block all operations on to_replace bs */ |
09158f00 | 929 | if (s->replaces) { |
5a7e7a0b SH |
930 | AioContext *replace_aio_context; |
931 | ||
e12f3784 | 932 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 933 | if (!s->to_replace) { |
e12f3784 | 934 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
935 | return; |
936 | } | |
937 | ||
5a7e7a0b SH |
938 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
939 | aio_context_acquire(replace_aio_context); | |
940 | ||
4ef85a9c KW |
941 | /* TODO Translate this into permission system. Current definition of |
942 | * GRAPH_MOD would require to request it for the parents; they might | |
943 | * not even be BlockDriverStates, however, so a BdrvChild can't address | |
944 | * them. May need redefinition of GRAPH_MOD. */ | |
09158f00 BC |
945 | error_setg(&s->replace_blocker, |
946 | "block device is in use by block-job-complete"); | |
947 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
948 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
949 | |
950 | aio_context_release(replace_aio_context); | |
09158f00 BC |
951 | } |
952 | ||
d63ffd87 | 953 | s->should_complete = true; |
751ebd76 | 954 | block_job_enter(&s->common); |
d63ffd87 PB |
955 | } |
956 | ||
bae8196d | 957 | static void mirror_pause(BlockJob *job) |
565ac01f SH |
958 | { |
959 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
960 | ||
bae8196d | 961 | mirror_wait_for_all_io(s); |
565ac01f SH |
962 | } |
963 | ||
964 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) | |
965 | { | |
966 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
967 | ||
968 | blk_set_aio_context(s->target, new_context); | |
969 | } | |
970 | ||
bae8196d PB |
971 | static void mirror_drain(BlockJob *job) |
972 | { | |
973 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
974 | ||
975 | /* Need to keep a reference in case blk_drain triggers execution | |
976 | * of mirror_complete... | |
977 | */ | |
978 | if (s->target) { | |
979 | BlockBackend *target = s->target; | |
980 | blk_ref(target); | |
981 | blk_drain(target); | |
982 | blk_unref(target); | |
983 | } | |
984 | } | |
985 | ||
3fc4b10a | 986 | static const BlockJobDriver mirror_job_driver = { |
565ac01f SH |
987 | .instance_size = sizeof(MirrorBlockJob), |
988 | .job_type = BLOCK_JOB_TYPE_MIRROR, | |
989 | .set_speed = mirror_set_speed, | |
a7815a76 | 990 | .start = mirror_run, |
565ac01f SH |
991 | .complete = mirror_complete, |
992 | .pause = mirror_pause, | |
993 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 994 | .drain = mirror_drain, |
893f7eba PB |
995 | }; |
996 | ||
03544a6e | 997 | static const BlockJobDriver commit_active_job_driver = { |
565ac01f SH |
998 | .instance_size = sizeof(MirrorBlockJob), |
999 | .job_type = BLOCK_JOB_TYPE_COMMIT, | |
1000 | .set_speed = mirror_set_speed, | |
a7815a76 | 1001 | .start = mirror_run, |
565ac01f SH |
1002 | .complete = mirror_complete, |
1003 | .pause = mirror_pause, | |
1004 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1005 | .drain = mirror_drain, |
03544a6e FZ |
1006 | }; |
1007 | ||
4ef85a9c KW |
1008 | static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, |
1009 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1010 | { | |
1011 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1012 | } | |
1013 | ||
1014 | static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, | |
1015 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1016 | { | |
1017 | return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1018 | } | |
1019 | ||
1020 | static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) | |
1021 | { | |
1022 | return bdrv_co_flush(bs->backing->bs); | |
1023 | } | |
1024 | ||
1025 | static int64_t coroutine_fn bdrv_mirror_top_get_block_status( | |
1026 | BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum, | |
1027 | BlockDriverState **file) | |
1028 | { | |
1029 | *pnum = nb_sectors; | |
1030 | *file = bs->backing->bs; | |
1031 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID | BDRV_BLOCK_DATA | | |
1032 | (sector_num << BDRV_SECTOR_BITS); | |
1033 | } | |
1034 | ||
1035 | static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, | |
1036 | int64_t offset, int count, BdrvRequestFlags flags) | |
1037 | { | |
1038 | return bdrv_co_pwrite_zeroes(bs->backing, offset, count, flags); | |
1039 | } | |
1040 | ||
1041 | static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, | |
1042 | int64_t offset, int count) | |
1043 | { | |
1044 | return bdrv_co_pdiscard(bs->backing->bs, offset, count); | |
1045 | } | |
1046 | ||
1047 | static void bdrv_mirror_top_close(BlockDriverState *bs) | |
1048 | { | |
1049 | } | |
1050 | ||
1051 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, | |
1052 | const BdrvChildRole *role, | |
1053 | uint64_t perm, uint64_t shared, | |
1054 | uint64_t *nperm, uint64_t *nshared) | |
1055 | { | |
1056 | /* Must be able to forward guest writes to the real image */ | |
1057 | *nperm = 0; | |
1058 | if (perm & BLK_PERM_WRITE) { | |
1059 | *nperm |= BLK_PERM_WRITE; | |
1060 | } | |
1061 | ||
1062 | *nshared = BLK_PERM_ALL; | |
1063 | } | |
1064 | ||
1065 | /* Dummy node that provides consistent read to its users without requiring it | |
1066 | * from its backing file and that allows writes on the backing file chain. */ | |
1067 | static BlockDriver bdrv_mirror_top = { | |
1068 | .format_name = "mirror_top", | |
1069 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1070 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1071 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1072 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1073 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
1074 | .bdrv_co_get_block_status = bdrv_mirror_top_get_block_status, | |
1075 | .bdrv_close = bdrv_mirror_top_close, | |
1076 | .bdrv_child_perm = bdrv_mirror_top_child_perm, | |
1077 | }; | |
1078 | ||
71aa9867 | 1079 | static void mirror_start_job(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1080 | int creation_flags, BlockDriverState *target, |
1081 | const char *replaces, int64_t speed, | |
1082 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1083 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
1084 | BlockdevOnError on_source_error, |
1085 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1086 | bool unmap, |
097310b5 | 1087 | BlockCompletionFunc *cb, |
09158f00 BC |
1088 | void *opaque, Error **errp, |
1089 | const BlockJobDriver *driver, | |
b49f7ead | 1090 | bool is_none_mode, BlockDriverState *base, |
6cdbceb1 | 1091 | bool auto_complete, const char *filter_node_name) |
893f7eba PB |
1092 | { |
1093 | MirrorBlockJob *s; | |
4ef85a9c KW |
1094 | BlockDriverState *mirror_top_bs; |
1095 | bool target_graph_mod; | |
1096 | bool target_is_backing; | |
d7086422 | 1097 | int ret; |
893f7eba | 1098 | |
eee13dfe | 1099 | if (granularity == 0) { |
341ebc2f | 1100 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1101 | } |
1102 | ||
1103 | assert ((granularity & (granularity - 1)) == 0); | |
1104 | ||
48ac0a4d WC |
1105 | if (buf_size < 0) { |
1106 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
1107 | return; | |
1108 | } | |
1109 | ||
1110 | if (buf_size == 0) { | |
1111 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
1112 | } | |
5bc361b8 | 1113 | |
4ef85a9c KW |
1114 | /* In the case of active commit, add dummy driver to provide consistent |
1115 | * reads on the top, while disabling it in the intermediate nodes, and make | |
1116 | * the backing chain writable. */ | |
6cdbceb1 KW |
1117 | mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, |
1118 | BDRV_O_RDWR, errp); | |
4ef85a9c KW |
1119 | if (mirror_top_bs == NULL) { |
1120 | return; | |
1121 | } | |
1122 | mirror_top_bs->total_sectors = bs->total_sectors; | |
1123 | ||
1124 | /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep | |
1125 | * it alive until block_job_create() even if bs has no parent. */ | |
1126 | bdrv_ref(mirror_top_bs); | |
1127 | bdrv_drained_begin(bs); | |
1128 | bdrv_append(mirror_top_bs, bs); | |
1129 | bdrv_drained_end(bs); | |
1130 | ||
1131 | /* Make sure that the source is not resized while the job is running */ | |
1132 | s = block_job_create(job_id, driver, mirror_top_bs, | |
1133 | BLK_PERM_CONSISTENT_READ, | |
1134 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
1135 | BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, | |
c6cc12bf | 1136 | creation_flags, cb, opaque, errp); |
4ef85a9c | 1137 | bdrv_unref(mirror_top_bs); |
893f7eba | 1138 | if (!s) { |
4ef85a9c | 1139 | goto fail; |
893f7eba | 1140 | } |
4ef85a9c KW |
1141 | s->source = bs; |
1142 | s->mirror_top_bs = mirror_top_bs; | |
1143 | ||
1144 | /* No resize for the target either; while the mirror is still running, a | |
1145 | * consistent read isn't necessarily possible. We could possibly allow | |
1146 | * writes and graph modifications, though it would likely defeat the | |
1147 | * purpose of a mirror, so leave them blocked for now. | |
1148 | * | |
1149 | * In the case of active commit, things look a bit different, though, | |
1150 | * because the target is an already populated backing file in active use. | |
1151 | * We can allow anything except resize there.*/ | |
1152 | target_is_backing = bdrv_chain_contains(bs, target); | |
1153 | target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); | |
1154 | s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | | |
1155 | (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), | |
1156 | BLK_PERM_WRITE_UNCHANGED | | |
1157 | (target_is_backing ? BLK_PERM_CONSISTENT_READ | | |
1158 | BLK_PERM_WRITE | | |
1159 | BLK_PERM_GRAPH_MOD : 0)); | |
d7086422 KW |
1160 | ret = blk_insert_bs(s->target, target, errp); |
1161 | if (ret < 0) { | |
4ef85a9c | 1162 | goto fail; |
d7086422 | 1163 | } |
e253f4b8 | 1164 | |
09158f00 | 1165 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1166 | s->on_source_error = on_source_error; |
1167 | s->on_target_error = on_target_error; | |
03544a6e | 1168 | s->is_none_mode = is_none_mode; |
274fccee | 1169 | s->backing_mode = backing_mode; |
5bc361b8 | 1170 | s->base = base; |
eee13dfe | 1171 | s->granularity = granularity; |
48ac0a4d | 1172 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1173 | s->unmap = unmap; |
b49f7ead WC |
1174 | if (auto_complete) { |
1175 | s->should_complete = true; | |
1176 | } | |
b812f671 | 1177 | |
0db6e54a | 1178 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1179 | if (!s->dirty_bitmap) { |
97031164 | 1180 | g_free(s->replaces); |
e253f4b8 | 1181 | blk_unref(s->target); |
18930ba3 | 1182 | block_job_unref(&s->common); |
b8afb520 FZ |
1183 | return; |
1184 | } | |
10f3cd15 | 1185 | |
4ef85a9c | 1186 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1187 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1188 | &error_abort); | |
1189 | ||
f3ede4b0 AG |
1190 | /* In commit_active_start() all intermediate nodes disappear, so |
1191 | * any jobs in them must be blocked */ | |
4ef85a9c | 1192 | if (target_is_backing) { |
f3ede4b0 AG |
1193 | BlockDriverState *iter; |
1194 | for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { | |
4ef85a9c KW |
1195 | /* XXX BLK_PERM_WRITE needs to be allowed so we don't block |
1196 | * ourselves at s->base (if writes are blocked for a node, they are | |
1197 | * also blocked for its backing file). The other options would be a | |
1198 | * second filter driver above s->base (== target). */ | |
1199 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, | |
1200 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, | |
1201 | errp); | |
1202 | if (ret < 0) { | |
1203 | goto fail; | |
1204 | } | |
f3ede4b0 AG |
1205 | } |
1206 | } | |
10f3cd15 | 1207 | |
5ccac6f1 JS |
1208 | trace_mirror_start(bs, s, opaque); |
1209 | block_job_start(&s->common); | |
4ef85a9c KW |
1210 | return; |
1211 | ||
1212 | fail: | |
1213 | if (s) { | |
1214 | g_free(s->replaces); | |
1215 | blk_unref(s->target); | |
1216 | block_job_unref(&s->common); | |
1217 | } | |
1218 | ||
1219 | bdrv_replace_in_backing_chain(mirror_top_bs, backing_bs(mirror_top_bs)); | |
893f7eba | 1220 | } |
03544a6e | 1221 | |
71aa9867 AG |
1222 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1223 | BlockDriverState *target, const char *replaces, | |
5fba6c0e | 1224 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
1225 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
1226 | BlockdevOnError on_source_error, | |
03544a6e | 1227 | BlockdevOnError on_target_error, |
6cdbceb1 | 1228 | bool unmap, const char *filter_node_name, Error **errp) |
03544a6e FZ |
1229 | { |
1230 | bool is_none_mode; | |
1231 | BlockDriverState *base; | |
1232 | ||
4b80ab2b JS |
1233 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
1234 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
1235 | return; |
1236 | } | |
03544a6e | 1237 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 1238 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
47970dfb | 1239 | mirror_start_job(job_id, bs, BLOCK_JOB_DEFAULT, target, replaces, |
274fccee | 1240 | speed, granularity, buf_size, backing_mode, |
8254b6d9 | 1241 | on_source_error, on_target_error, unmap, NULL, NULL, errp, |
6cdbceb1 KW |
1242 | &mirror_job_driver, is_none_mode, base, false, |
1243 | filter_node_name); | |
03544a6e FZ |
1244 | } |
1245 | ||
fd62c609 | 1246 | void commit_active_start(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1247 | BlockDriverState *base, int creation_flags, |
1248 | int64_t speed, BlockdevOnError on_error, | |
1249 | BlockCompletionFunc *cb, void *opaque, Error **errp, | |
b49f7ead | 1250 | bool auto_complete) |
03544a6e | 1251 | { |
4da83585 | 1252 | int orig_base_flags; |
cc67f4d1 | 1253 | Error *local_err = NULL; |
4da83585 JC |
1254 | |
1255 | orig_base_flags = bdrv_get_flags(base); | |
1256 | ||
20a63d2c FZ |
1257 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
1258 | return; | |
1259 | } | |
4da83585 | 1260 | |
47970dfb | 1261 | mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
71aa9867 | 1262 | MIRROR_LEAVE_BACKING_CHAIN, |
6f13acf9 | 1263 | on_error, on_error, true, cb, opaque, &local_err, |
6cdbceb1 KW |
1264 | &commit_active_job_driver, false, base, auto_complete, |
1265 | NULL); | |
0fb6395c | 1266 | if (local_err) { |
cc67f4d1 | 1267 | error_propagate(errp, local_err); |
4da83585 JC |
1268 | goto error_restore_flags; |
1269 | } | |
1270 | ||
1271 | return; | |
1272 | ||
1273 | error_restore_flags: | |
1274 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1275 | * the original error */ | |
1276 | bdrv_reopen(base, orig_base_flags, NULL); | |
1277 | return; | |
03544a6e | 1278 | } |