]>
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 | |
b436982f EB |
27 | #define MAX_IO_BYTES (1 << 20) /* 1 Mb */ |
28 | #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES) | |
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; |
b436982f | 69 | int64_t bytes_in_flight; |
bd48bde8 | 70 | int ret; |
0fc9f8ea | 71 | bool unmap; |
e424aff5 | 72 | bool waiting_for_io; |
b436982f | 73 | int target_cluster_size; |
e5b43573 | 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; | |
b436982f EB |
81 | int64_t offset; |
82 | uint64_t bytes; | |
bd48bde8 PB |
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; |
b436982f | 103 | int i, nb_chunks; |
bd48bde8 | 104 | |
b436982f | 105 | trace_mirror_iteration_done(s, op->offset, op->bytes, ret); |
bd48bde8 PB |
106 | |
107 | s->in_flight--; | |
b436982f | 108 | s->bytes_in_flight -= op->bytes; |
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 | ||
b436982f EB |
116 | chunk_num = op->offset / s->granularity; |
117 | nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); | |
402a4741 | 118 | bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); |
b21c7652 HR |
119 | if (ret >= 0) { |
120 | if (s->cow_bitmap) { | |
121 | bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); | |
122 | } | |
90ab48eb | 123 | if (!s->initial_zeroing_ongoing) { |
b436982f | 124 | s->common.offset += op->bytes; |
90ab48eb | 125 | } |
bd48bde8 | 126 | } |
6df3bf8e | 127 | qemu_iovec_destroy(&op->qiov); |
c84b3192 | 128 | g_free(op); |
7b770c72 | 129 | |
e424aff5 | 130 | if (s->waiting_for_io) { |
0b8b8753 | 131 | qemu_coroutine_enter(s->common.co); |
7b770c72 | 132 | } |
bd48bde8 PB |
133 | } |
134 | ||
135 | static void mirror_write_complete(void *opaque, int ret) | |
136 | { | |
137 | MirrorOp *op = opaque; | |
138 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
139 | |
140 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 141 | if (ret < 0) { |
bd48bde8 PB |
142 | BlockErrorAction action; |
143 | ||
e0d7f73e | 144 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 145 | action = mirror_error_action(s, false, -ret); |
a589569f | 146 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
147 | s->ret = ret; |
148 | } | |
149 | } | |
150 | mirror_iteration_done(op, ret); | |
b9e413dd | 151 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
152 | } |
153 | ||
154 | static void mirror_read_complete(void *opaque, int ret) | |
155 | { | |
156 | MirrorOp *op = opaque; | |
157 | MirrorBlockJob *s = op->s; | |
b9e413dd PB |
158 | |
159 | aio_context_acquire(blk_get_aio_context(s->common.blk)); | |
bd48bde8 | 160 | if (ret < 0) { |
bd48bde8 PB |
161 | BlockErrorAction action; |
162 | ||
e0d7f73e | 163 | bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); |
bd48bde8 | 164 | action = mirror_error_action(s, true, -ret); |
a589569f | 165 | if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { |
bd48bde8 PB |
166 | s->ret = ret; |
167 | } | |
168 | ||
169 | mirror_iteration_done(op, ret); | |
b9e413dd | 170 | } else { |
b436982f | 171 | blk_aio_pwritev(s->target, op->offset, &op->qiov, |
b9e413dd | 172 | 0, mirror_write_complete, op); |
bd48bde8 | 173 | } |
b9e413dd | 174 | aio_context_release(blk_get_aio_context(s->common.blk)); |
bd48bde8 PB |
175 | } |
176 | ||
782d97ef EB |
177 | /* Clip bytes relative to offset to not exceed end-of-file */ |
178 | static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, | |
179 | int64_t offset, | |
180 | int64_t bytes) | |
181 | { | |
182 | return MIN(bytes, s->bdev_length - offset); | |
183 | } | |
184 | ||
782d97ef EB |
185 | /* Round offset and/or bytes to target cluster if COW is needed, and |
186 | * return the offset of the adjusted tail against original. */ | |
187 | static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, | |
ae4cc877 | 188 | uint64_t *bytes) |
893f7eba | 189 | { |
e5b43573 FZ |
190 | bool need_cow; |
191 | int ret = 0; | |
782d97ef EB |
192 | int64_t align_offset = *offset; |
193 | unsigned int align_bytes = *bytes; | |
194 | int max_bytes = s->granularity * s->max_iov; | |
e5b43573 | 195 | |
ae4cc877 | 196 | assert(*bytes < INT_MAX); |
782d97ef EB |
197 | need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); |
198 | need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, | |
e5b43573 FZ |
199 | s->cow_bitmap); |
200 | if (need_cow) { | |
782d97ef EB |
201 | bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, |
202 | &align_offset, &align_bytes); | |
e5b43573 | 203 | } |
3515727f | 204 | |
782d97ef EB |
205 | if (align_bytes > max_bytes) { |
206 | align_bytes = max_bytes; | |
e5b43573 | 207 | if (need_cow) { |
782d97ef | 208 | align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); |
e5b43573 | 209 | } |
8f0720ec | 210 | } |
782d97ef | 211 | /* Clipping may result in align_bytes unaligned to chunk boundary, but |
4150ae60 | 212 | * that doesn't matter because it's already the end of source image. */ |
782d97ef | 213 | align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); |
8f0720ec | 214 | |
782d97ef EB |
215 | ret = align_offset + align_bytes - (*offset + *bytes); |
216 | *offset = align_offset; | |
217 | *bytes = align_bytes; | |
e5b43573 FZ |
218 | assert(ret >= 0); |
219 | return ret; | |
220 | } | |
221 | ||
21cd917f FZ |
222 | static inline void mirror_wait_for_io(MirrorBlockJob *s) |
223 | { | |
224 | assert(!s->waiting_for_io); | |
225 | s->waiting_for_io = true; | |
226 | qemu_coroutine_yield(); | |
227 | s->waiting_for_io = false; | |
228 | } | |
229 | ||
e5b43573 | 230 | /* Submit async read while handling COW. |
ae4cc877 EB |
231 | * Returns: The number of bytes copied after and including offset, |
232 | * excluding any bytes copied prior to offset due to alignment. | |
233 | * This will be @bytes if no alignment is necessary, or | |
234 | * (new_end - offset) if tail is rounded up or down due to | |
e5b43573 FZ |
235 | * alignment or buffer limit. |
236 | */ | |
ae4cc877 EB |
237 | static uint64_t mirror_do_read(MirrorBlockJob *s, int64_t offset, |
238 | uint64_t bytes) | |
e5b43573 | 239 | { |
e253f4b8 | 240 | BlockBackend *source = s->common.blk; |
ae4cc877 EB |
241 | int nb_chunks; |
242 | uint64_t ret; | |
e5b43573 | 243 | MirrorOp *op; |
ae4cc877 | 244 | uint64_t max_bytes; |
e5b43573 | 245 | |
ae4cc877 | 246 | max_bytes = s->granularity * s->max_iov; |
402a4741 | 247 | |
e5b43573 | 248 | /* We can only handle as much as buf_size at a time. */ |
ae4cc877 EB |
249 | bytes = MIN(s->buf_size, MIN(max_bytes, bytes)); |
250 | assert(bytes); | |
251 | assert(bytes < BDRV_REQUEST_MAX_BYTES); | |
252 | ret = bytes; | |
402a4741 | 253 | |
e5b43573 | 254 | if (s->cow_bitmap) { |
ae4cc877 | 255 | ret += mirror_cow_align(s, &offset, &bytes); |
e5b43573 | 256 | } |
ae4cc877 EB |
257 | assert(bytes <= s->buf_size); |
258 | /* The offset is granularity-aligned because: | |
e5b43573 FZ |
259 | * 1) Caller passes in aligned values; |
260 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
ae4cc877 EB |
261 | assert(QEMU_IS_ALIGNED(offset, s->granularity)); |
262 | /* The range is sector-aligned, since bdrv_getlength() rounds up. */ | |
263 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
264 | nb_chunks = DIV_ROUND_UP(bytes, s->granularity); | |
e5b43573 FZ |
265 | |
266 | while (s->buf_free_count < nb_chunks) { | |
ae4cc877 | 267 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
21cd917f | 268 | mirror_wait_for_io(s); |
b812f671 PB |
269 | } |
270 | ||
bd48bde8 | 271 | /* Allocate a MirrorOp that is used as an AIO callback. */ |
c84b3192 | 272 | op = g_new(MirrorOp, 1); |
bd48bde8 | 273 | op->s = s; |
ae4cc877 EB |
274 | op->offset = offset; |
275 | op->bytes = bytes; | |
402a4741 PB |
276 | |
277 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
278 | * from s->buf_free. | |
279 | */ | |
280 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
281 | while (nb_chunks-- > 0) { |
282 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
ae4cc877 | 283 | size_t remaining = bytes - op->qiov.size; |
5a0f6fd5 | 284 | |
402a4741 PB |
285 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
286 | s->buf_free_count--; | |
5a0f6fd5 | 287 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 288 | } |
bd48bde8 | 289 | |
893f7eba | 290 | /* Copy the dirty cluster. */ |
bd48bde8 | 291 | s->in_flight++; |
ae4cc877 EB |
292 | s->bytes_in_flight += bytes; |
293 | trace_mirror_one_iteration(s, offset, bytes); | |
dcfb3beb | 294 | |
ae4cc877 | 295 | blk_aio_preadv(source, offset, &op->qiov, 0, mirror_read_complete, op); |
e5b43573 FZ |
296 | return ret; |
297 | } | |
298 | ||
299 | static void mirror_do_zero_or_discard(MirrorBlockJob *s, | |
e6f24193 EB |
300 | int64_t offset, |
301 | uint64_t bytes, | |
e5b43573 FZ |
302 | bool is_discard) |
303 | { | |
304 | MirrorOp *op; | |
305 | ||
306 | /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed | |
307 | * so the freeing in mirror_iteration_done is nop. */ | |
308 | op = g_new0(MirrorOp, 1); | |
309 | op->s = s; | |
e6f24193 EB |
310 | op->offset = offset; |
311 | op->bytes = bytes; | |
e5b43573 FZ |
312 | |
313 | s->in_flight++; | |
e6f24193 | 314 | s->bytes_in_flight += bytes; |
e5b43573 | 315 | if (is_discard) { |
e6f24193 | 316 | blk_aio_pdiscard(s->target, offset, |
b436982f | 317 | op->bytes, mirror_write_complete, op); |
e5b43573 | 318 | } else { |
e6f24193 | 319 | blk_aio_pwrite_zeroes(s->target, offset, |
b436982f | 320 | op->bytes, s->unmap ? BDRV_REQ_MAY_UNMAP : 0, |
dcfb3beb | 321 | mirror_write_complete, op); |
e5b43573 FZ |
322 | } |
323 | } | |
324 | ||
325 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) | |
326 | { | |
4ef85a9c | 327 | BlockDriverState *source = s->source; |
fb2ef791 | 328 | int64_t offset, first_chunk; |
e5b43573 FZ |
329 | uint64_t delay_ns = 0; |
330 | /* At least the first dirty chunk is mirrored in one iteration. */ | |
331 | int nb_chunks = 1; | |
e5b43573 | 332 | int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; |
4b5004d9 | 333 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
b436982f | 334 | int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); |
e5b43573 | 335 | |
b64bd51e | 336 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
f798184c | 337 | offset = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 338 | if (offset < 0) { |
dc162c8e | 339 | bdrv_set_dirty_iter(s->dbi, 0); |
f798184c | 340 | offset = bdrv_dirty_iter_next(s->dbi); |
9a46dba7 | 341 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
fb2ef791 | 342 | assert(offset >= 0); |
e5b43573 | 343 | } |
b64bd51e | 344 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
e5b43573 | 345 | |
fb2ef791 | 346 | first_chunk = offset / s->granularity; |
9c83625b | 347 | while (test_bit(first_chunk, s->in_flight_bitmap)) { |
fb2ef791 | 348 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
9c83625b HR |
349 | mirror_wait_for_io(s); |
350 | } | |
351 | ||
565ac01f SH |
352 | block_job_pause_point(&s->common); |
353 | ||
e5b43573 FZ |
354 | /* Find the number of consective dirty chunks following the first dirty |
355 | * one, and wait for in flight requests in them. */ | |
b64bd51e | 356 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
fb2ef791 | 357 | while (nb_chunks * s->granularity < s->buf_size) { |
dc162c8e | 358 | int64_t next_dirty; |
fb2ef791 EB |
359 | int64_t next_offset = offset + nb_chunks * s->granularity; |
360 | int64_t next_chunk = next_offset / s->granularity; | |
361 | if (next_offset >= s->bdev_length || | |
3b5d4df0 | 362 | !bdrv_get_dirty_locked(source, s->dirty_bitmap, next_offset)) { |
e5b43573 FZ |
363 | break; |
364 | } | |
365 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 366 | break; |
e5b43573 | 367 | } |
9c83625b | 368 | |
f798184c | 369 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 370 | if (next_dirty > next_offset || next_dirty < 0) { |
f27a2742 | 371 | /* The bitmap iterator's cache is stale, refresh it */ |
715a74d8 | 372 | bdrv_set_dirty_iter(s->dbi, next_offset); |
f798184c | 373 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
f27a2742 | 374 | } |
fb2ef791 | 375 | assert(next_dirty == next_offset); |
9c83625b | 376 | nb_chunks++; |
e5b43573 FZ |
377 | } |
378 | ||
379 | /* Clear dirty bits before querying the block status, because | |
380 | * calling bdrv_get_block_status_above could yield - if some blocks are | |
381 | * marked dirty in this window, we need to know. | |
382 | */ | |
e0d7f73e EB |
383 | bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, |
384 | nb_chunks * s->granularity); | |
b64bd51e PB |
385 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
386 | ||
fb2ef791 EB |
387 | bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); |
388 | while (nb_chunks > 0 && offset < s->bdev_length) { | |
39c11580 | 389 | int64_t ret; |
f3e4ce4a | 390 | int io_sectors; |
fb2ef791 | 391 | unsigned int io_bytes; |
f3e4ce4a | 392 | int64_t io_bytes_acct; |
e5b43573 FZ |
393 | BlockDriverState *file; |
394 | enum MirrorMethod { | |
395 | MIRROR_METHOD_COPY, | |
396 | MIRROR_METHOD_ZERO, | |
397 | MIRROR_METHOD_DISCARD | |
398 | } mirror_method = MIRROR_METHOD_COPY; | |
399 | ||
fb2ef791 EB |
400 | assert(!(offset % s->granularity)); |
401 | ret = bdrv_get_block_status_above(source, NULL, | |
402 | offset >> BDRV_SECTOR_BITS, | |
e5b43573 FZ |
403 | nb_chunks * sectors_per_chunk, |
404 | &io_sectors, &file); | |
fb2ef791 | 405 | io_bytes = io_sectors * BDRV_SECTOR_SIZE; |
e5b43573 | 406 | if (ret < 0) { |
fb2ef791 | 407 | io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); |
0965a41e | 408 | } else if (ret & BDRV_BLOCK_DATA) { |
fb2ef791 | 409 | io_bytes = MIN(io_bytes, max_io_bytes); |
e5b43573 FZ |
410 | } |
411 | ||
fb2ef791 EB |
412 | io_bytes -= io_bytes % s->granularity; |
413 | if (io_bytes < s->granularity) { | |
414 | io_bytes = s->granularity; | |
e5b43573 | 415 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { |
fb2ef791 EB |
416 | int64_t target_offset; |
417 | unsigned int target_bytes; | |
418 | bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes, | |
419 | &target_offset, &target_bytes); | |
420 | if (target_offset == offset && | |
421 | target_bytes == io_bytes) { | |
e5b43573 FZ |
422 | mirror_method = ret & BDRV_BLOCK_ZERO ? |
423 | MIRROR_METHOD_ZERO : | |
424 | MIRROR_METHOD_DISCARD; | |
425 | } | |
426 | } | |
427 | ||
cf56a3c6 | 428 | while (s->in_flight >= MAX_IN_FLIGHT) { |
fb2ef791 | 429 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
cf56a3c6 DL |
430 | mirror_wait_for_io(s); |
431 | } | |
432 | ||
dbaa7b57 VSO |
433 | if (s->ret < 0) { |
434 | return 0; | |
435 | } | |
436 | ||
fb2ef791 | 437 | io_bytes = mirror_clip_bytes(s, offset, io_bytes); |
e5b43573 FZ |
438 | switch (mirror_method) { |
439 | case MIRROR_METHOD_COPY: | |
fb2ef791 | 440 | io_bytes = io_bytes_acct = mirror_do_read(s, offset, io_bytes); |
e5b43573 FZ |
441 | break; |
442 | case MIRROR_METHOD_ZERO: | |
e5b43573 | 443 | case MIRROR_METHOD_DISCARD: |
fb2ef791 | 444 | mirror_do_zero_or_discard(s, offset, io_bytes, |
4b5004d9 DL |
445 | mirror_method == MIRROR_METHOD_DISCARD); |
446 | if (write_zeroes_ok) { | |
f3e4ce4a | 447 | io_bytes_acct = 0; |
4b5004d9 | 448 | } else { |
fb2ef791 | 449 | io_bytes_acct = io_bytes; |
4b5004d9 | 450 | } |
e5b43573 FZ |
451 | break; |
452 | default: | |
453 | abort(); | |
454 | } | |
fb2ef791 EB |
455 | assert(io_bytes); |
456 | offset += io_bytes; | |
457 | nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); | |
f14a39cc | 458 | if (s->common.speed) { |
f3e4ce4a | 459 | delay_ns = ratelimit_calculate_delay(&s->limit, io_bytes_acct); |
f14a39cc | 460 | } |
dcfb3beb | 461 | } |
cc8c9d6c | 462 | return delay_ns; |
bd48bde8 | 463 | } |
b952b558 | 464 | |
402a4741 PB |
465 | static void mirror_free_init(MirrorBlockJob *s) |
466 | { | |
467 | int granularity = s->granularity; | |
468 | size_t buf_size = s->buf_size; | |
469 | uint8_t *buf = s->buf; | |
470 | ||
471 | assert(s->buf_free_count == 0); | |
472 | QSIMPLEQ_INIT(&s->buf_free); | |
473 | while (buf_size != 0) { | |
474 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
475 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
476 | s->buf_free_count++; | |
477 | buf_size -= granularity; | |
478 | buf += granularity; | |
479 | } | |
480 | } | |
481 | ||
bae8196d PB |
482 | /* This is also used for the .pause callback. There is no matching |
483 | * mirror_resume() because mirror_run() will begin iterating again | |
484 | * when the job is resumed. | |
485 | */ | |
486 | static void mirror_wait_for_all_io(MirrorBlockJob *s) | |
bd48bde8 PB |
487 | { |
488 | while (s->in_flight > 0) { | |
21cd917f | 489 | mirror_wait_for_io(s); |
bd48bde8 | 490 | } |
893f7eba PB |
491 | } |
492 | ||
5a7e7a0b SH |
493 | typedef struct { |
494 | int ret; | |
495 | } MirrorExitData; | |
496 | ||
497 | static void mirror_exit(BlockJob *job, void *opaque) | |
498 | { | |
499 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
500 | MirrorExitData *data = opaque; | |
501 | AioContext *replace_aio_context = NULL; | |
4ef85a9c | 502 | BlockDriverState *src = s->source; |
e253f4b8 | 503 | BlockDriverState *target_bs = blk_bs(s->target); |
4ef85a9c | 504 | BlockDriverState *mirror_top_bs = s->mirror_top_bs; |
12fa4af6 | 505 | Error *local_err = NULL; |
3f09bfbc | 506 | |
2119882c PB |
507 | bdrv_release_dirty_bitmap(src, s->dirty_bitmap); |
508 | ||
3f09bfbc KW |
509 | /* Make sure that the source BDS doesn't go away before we called |
510 | * block_job_completed(). */ | |
511 | bdrv_ref(src); | |
4ef85a9c | 512 | bdrv_ref(mirror_top_bs); |
7d9fcb39 KW |
513 | bdrv_ref(target_bs); |
514 | ||
515 | /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before | |
516 | * inserting target_bs at s->to_replace, where we might not be able to get | |
63c8ef28 KW |
517 | * these permissions. |
518 | * | |
519 | * Note that blk_unref() alone doesn't necessarily drop permissions because | |
520 | * we might be running nested inside mirror_drain(), which takes an extra | |
521 | * reference, so use an explicit blk_set_perm() first. */ | |
522 | blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); | |
7d9fcb39 KW |
523 | blk_unref(s->target); |
524 | s->target = NULL; | |
4ef85a9c KW |
525 | |
526 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
527 | * required before it could become a backing file of target_bs. */ | |
528 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, | |
529 | &error_abort); | |
530 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { | |
531 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
532 | if (backing_bs(target_bs) != backing) { | |
12fa4af6 KW |
533 | bdrv_set_backing_hd(target_bs, backing, &local_err); |
534 | if (local_err) { | |
535 | error_report_err(local_err); | |
536 | data->ret = -EPERM; | |
537 | } | |
4ef85a9c KW |
538 | } |
539 | } | |
5a7e7a0b SH |
540 | |
541 | if (s->to_replace) { | |
542 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
543 | aio_context_acquire(replace_aio_context); | |
544 | } | |
545 | ||
546 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 547 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
548 | if (s->to_replace) { |
549 | to_replace = s->to_replace; | |
550 | } | |
40365552 | 551 | |
e253f4b8 KW |
552 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
553 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 554 | } |
b8804815 KW |
555 | |
556 | /* The mirror job has no requests in flight any more, but we need to | |
557 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 | 558 | bdrv_drained_begin(target_bs); |
5fe31c25 | 559 | bdrv_replace_node(to_replace, target_bs, &local_err); |
e253f4b8 | 560 | bdrv_drained_end(target_bs); |
5fe31c25 KW |
561 | if (local_err) { |
562 | error_report_err(local_err); | |
563 | data->ret = -EPERM; | |
564 | } | |
5a7e7a0b SH |
565 | } |
566 | if (s->to_replace) { | |
567 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
568 | error_free(s->replace_blocker); | |
569 | bdrv_unref(s->to_replace); | |
570 | } | |
571 | if (replace_aio_context) { | |
572 | aio_context_release(replace_aio_context); | |
573 | } | |
574 | g_free(s->replaces); | |
7d9fcb39 | 575 | bdrv_unref(target_bs); |
4ef85a9c KW |
576 | |
577 | /* Remove the mirror filter driver from the graph. Before this, get rid of | |
578 | * the blockers on the intermediate nodes so that the resulting state is | |
0bf74767 KW |
579 | * valid. Also give up permissions on mirror_top_bs->backing, which might |
580 | * block the removal. */ | |
4ef85a9c | 581 | block_job_remove_all_bdrv(job); |
c1cef672 FZ |
582 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
583 | &error_abort); | |
5fe31c25 | 584 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
4ef85a9c KW |
585 | |
586 | /* We just changed the BDS the job BB refers to (with either or both of the | |
5fe31c25 KW |
587 | * bdrv_replace_node() calls), so switch the BB back so the cleanup does |
588 | * the right thing. We don't need any permissions any more now. */ | |
4ef85a9c KW |
589 | blk_remove_bs(job->blk); |
590 | blk_set_perm(job->blk, 0, BLK_PERM_ALL, &error_abort); | |
591 | blk_insert_bs(job->blk, mirror_top_bs, &error_abort); | |
592 | ||
5a7e7a0b | 593 | block_job_completed(&s->common, data->ret); |
4ef85a9c | 594 | |
5a7e7a0b | 595 | g_free(data); |
176c3699 | 596 | bdrv_drained_end(src); |
4ef85a9c | 597 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 598 | bdrv_unref(src); |
5a7e7a0b SH |
599 | } |
600 | ||
49efb1f5 DL |
601 | static void mirror_throttle(MirrorBlockJob *s) |
602 | { | |
603 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
604 | ||
605 | if (now - s->last_pause_ns > SLICE_TIME) { | |
606 | s->last_pause_ns = now; | |
607 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); | |
608 | } else { | |
609 | block_job_pause_point(&s->common); | |
610 | } | |
611 | } | |
612 | ||
c0b363ad DL |
613 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
614 | { | |
23ca459a | 615 | int64_t offset; |
c0b363ad | 616 | BlockDriverState *base = s->base; |
4ef85a9c | 617 | BlockDriverState *bs = s->source; |
c0b363ad | 618 | BlockDriverState *target_bs = blk_bs(s->target); |
23ca459a | 619 | int ret; |
51b0a488 | 620 | int64_t count; |
c0b363ad | 621 | |
b7d5062c | 622 | if (base == NULL && !bdrv_has_zero_init(target_bs)) { |
c7c2769c | 623 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
e0d7f73e | 624 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); |
c7c2769c DL |
625 | return 0; |
626 | } | |
627 | ||
90ab48eb | 628 | s->initial_zeroing_ongoing = true; |
23ca459a EB |
629 | for (offset = 0; offset < s->bdev_length; ) { |
630 | int bytes = MIN(s->bdev_length - offset, | |
631 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c7c2769c DL |
632 | |
633 | mirror_throttle(s); | |
634 | ||
635 | if (block_job_is_cancelled(&s->common)) { | |
90ab48eb | 636 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
637 | return 0; |
638 | } | |
639 | ||
640 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
67adf4b3 EB |
641 | trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, |
642 | s->in_flight); | |
c7c2769c DL |
643 | mirror_wait_for_io(s); |
644 | continue; | |
645 | } | |
646 | ||
23ca459a EB |
647 | mirror_do_zero_or_discard(s, offset, bytes, false); |
648 | offset += bytes; | |
c7c2769c DL |
649 | } |
650 | ||
bae8196d | 651 | mirror_wait_for_all_io(s); |
90ab48eb | 652 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
653 | } |
654 | ||
c0b363ad | 655 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
23ca459a | 656 | for (offset = 0; offset < s->bdev_length; ) { |
c0b363ad | 657 | /* Just to make sure we are not exceeding int limit. */ |
23ca459a EB |
658 | int bytes = MIN(s->bdev_length - offset, |
659 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c0b363ad DL |
660 | |
661 | mirror_throttle(s); | |
662 | ||
663 | if (block_job_is_cancelled(&s->common)) { | |
664 | return 0; | |
665 | } | |
666 | ||
23ca459a | 667 | ret = bdrv_is_allocated_above(bs, base, offset, bytes, &count); |
c0b363ad DL |
668 | if (ret < 0) { |
669 | return ret; | |
670 | } | |
671 | ||
23ca459a | 672 | assert(count); |
b7d5062c | 673 | if (ret == 1) { |
23ca459a | 674 | bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); |
c0b363ad | 675 | } |
23ca459a | 676 | offset += count; |
c0b363ad DL |
677 | } |
678 | return 0; | |
679 | } | |
680 | ||
bdffb31d PB |
681 | /* Called when going out of the streaming phase to flush the bulk of the |
682 | * data to the medium, or just before completing. | |
683 | */ | |
684 | static int mirror_flush(MirrorBlockJob *s) | |
685 | { | |
686 | int ret = blk_flush(s->target); | |
687 | if (ret < 0) { | |
688 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
689 | s->ret = ret; | |
690 | } | |
691 | } | |
692 | return ret; | |
693 | } | |
694 | ||
893f7eba PB |
695 | static void coroutine_fn mirror_run(void *opaque) |
696 | { | |
697 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 698 | MirrorExitData *data; |
4ef85a9c | 699 | BlockDriverState *bs = s->source; |
e253f4b8 | 700 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 701 | bool need_drain = true; |
c0b363ad | 702 | int64_t length; |
b812f671 | 703 | BlockDriverInfo bdi; |
1d33936e JC |
704 | char backing_filename[2]; /* we only need 2 characters because we are only |
705 | checking for a NULL string */ | |
893f7eba | 706 | int ret = 0; |
893f7eba PB |
707 | |
708 | if (block_job_is_cancelled(&s->common)) { | |
709 | goto immediate_exit; | |
710 | } | |
711 | ||
b21c7652 HR |
712 | s->bdev_length = bdrv_getlength(bs); |
713 | if (s->bdev_length < 0) { | |
714 | ret = s->bdev_length; | |
373df5b1 | 715 | goto immediate_exit; |
becc347e KW |
716 | } |
717 | ||
718 | /* Active commit must resize the base image if its size differs from the | |
719 | * active layer. */ | |
720 | if (s->base == blk_bs(s->target)) { | |
721 | int64_t base_length; | |
722 | ||
723 | base_length = blk_getlength(s->target); | |
724 | if (base_length < 0) { | |
725 | ret = base_length; | |
726 | goto immediate_exit; | |
727 | } | |
728 | ||
729 | if (s->bdev_length > base_length) { | |
3a691c50 HR |
730 | ret = blk_truncate(s->target, s->bdev_length, PREALLOC_MODE_OFF, |
731 | NULL); | |
becc347e KW |
732 | if (ret < 0) { |
733 | goto immediate_exit; | |
734 | } | |
735 | } | |
736 | } | |
737 | ||
738 | if (s->bdev_length == 0) { | |
9e48b025 FZ |
739 | /* Report BLOCK_JOB_READY and wait for complete. */ |
740 | block_job_event_ready(&s->common); | |
741 | s->synced = true; | |
742 | while (!block_job_is_cancelled(&s->common) && !s->should_complete) { | |
743 | block_job_yield(&s->common); | |
744 | } | |
745 | s->common.cancelled = false; | |
746 | goto immediate_exit; | |
893f7eba PB |
747 | } |
748 | ||
b21c7652 | 749 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
750 | s->in_flight_bitmap = bitmap_new(length); |
751 | ||
b812f671 PB |
752 | /* If we have no backing file yet in the destination, we cannot let |
753 | * the destination do COW. Instead, we copy sectors around the | |
754 | * dirty data if needed. We need a bitmap to do that. | |
755 | */ | |
e253f4b8 | 756 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 757 | sizeof(backing_filename)); |
e253f4b8 | 758 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
b436982f EB |
759 | s->target_cluster_size = bdi.cluster_size; |
760 | } else { | |
761 | s->target_cluster_size = BDRV_SECTOR_SIZE; | |
e5b43573 | 762 | } |
b436982f EB |
763 | if (backing_filename[0] && !target_bs->backing && |
764 | s->granularity < s->target_cluster_size) { | |
765 | s->buf_size = MAX(s->buf_size, s->target_cluster_size); | |
e5b43573 | 766 | s->cow_bitmap = bitmap_new(length); |
b812f671 | 767 | } |
e253f4b8 | 768 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 769 | |
7504edf4 KW |
770 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
771 | if (s->buf == NULL) { | |
772 | ret = -ENOMEM; | |
773 | goto immediate_exit; | |
774 | } | |
775 | ||
402a4741 | 776 | mirror_free_init(s); |
893f7eba | 777 | |
49efb1f5 | 778 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 779 | if (!s->is_none_mode) { |
c0b363ad DL |
780 | ret = mirror_dirty_init(s); |
781 | if (ret < 0 || block_job_is_cancelled(&s->common)) { | |
782 | goto immediate_exit; | |
893f7eba PB |
783 | } |
784 | } | |
785 | ||
dc162c8e | 786 | assert(!s->dbi); |
715a74d8 | 787 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); |
893f7eba | 788 | for (;;) { |
cc8c9d6c | 789 | uint64_t delay_ns = 0; |
49efb1f5 | 790 | int64_t cnt, delta; |
893f7eba PB |
791 | bool should_complete; |
792 | ||
bd48bde8 PB |
793 | if (s->ret < 0) { |
794 | ret = s->ret; | |
795 | goto immediate_exit; | |
796 | } | |
797 | ||
565ac01f SH |
798 | block_job_pause_point(&s->common); |
799 | ||
20dca810 | 800 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
b21c7652 | 801 | /* s->common.offset contains the number of bytes already processed so |
9a46dba7 | 802 | * far, cnt is the number of dirty bytes remaining and |
b436982f | 803 | * s->bytes_in_flight is the number of bytes currently being |
b21c7652 | 804 | * processed; together those are the current total operation length */ |
9a46dba7 | 805 | s->common.len = s->common.offset + s->bytes_in_flight + cnt; |
bd48bde8 PB |
806 | |
807 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 808 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
bd48bde8 PB |
809 | * We do so every SLICE_TIME nanoseconds, or when there is an error, |
810 | * or when the source is clean, whichever comes first. | |
811 | */ | |
49efb1f5 DL |
812 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
813 | if (delta < SLICE_TIME && | |
bd48bde8 | 814 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 815 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 | 816 | (cnt == 0 && s->in_flight > 0)) { |
9a46dba7 | 817 | trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); |
21cd917f | 818 | mirror_wait_for_io(s); |
bd48bde8 PB |
819 | continue; |
820 | } else if (cnt != 0) { | |
cc8c9d6c | 821 | delay_ns = mirror_iteration(s); |
893f7eba | 822 | } |
893f7eba PB |
823 | } |
824 | ||
825 | should_complete = false; | |
bd48bde8 | 826 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 827 | trace_mirror_before_flush(s); |
bdffb31d PB |
828 | if (!s->synced) { |
829 | if (mirror_flush(s) < 0) { | |
830 | /* Go check s->ret. */ | |
831 | continue; | |
b952b558 | 832 | } |
b952b558 PB |
833 | /* We're out of the streaming phase. From now on, if the job |
834 | * is cancelled we will actually complete all pending I/O and | |
835 | * report completion. This way, block-job-cancel will leave | |
836 | * the target in a consistent state. | |
837 | */ | |
bdffb31d PB |
838 | block_job_event_ready(&s->common); |
839 | s->synced = true; | |
d63ffd87 | 840 | } |
bdffb31d PB |
841 | |
842 | should_complete = s->should_complete || | |
843 | block_job_is_cancelled(&s->common); | |
844 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); | |
893f7eba PB |
845 | } |
846 | ||
847 | if (cnt == 0 && should_complete) { | |
848 | /* The dirty bitmap is not updated while operations are pending. | |
849 | * If we're about to exit, wait for pending operations before | |
850 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
851 | * source has dirty data to copy! | |
852 | * | |
853 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
854 | * mirror_populate runs, so pause it now. Before deciding |
855 | * whether to switch to target check one last time if I/O has | |
856 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba | 857 | */ |
9a46dba7 | 858 | trace_mirror_before_drain(s, cnt); |
9a0cec66 PB |
859 | |
860 | bdrv_drained_begin(bs); | |
20dca810 | 861 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 862 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 PB |
863 | bdrv_drained_end(bs); |
864 | continue; | |
865 | } | |
866 | ||
867 | /* The two disks are in sync. Exit and report successful | |
868 | * completion. | |
869 | */ | |
870 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
871 | s->common.cancelled = false; | |
872 | need_drain = false; | |
873 | break; | |
893f7eba PB |
874 | } |
875 | ||
876 | ret = 0; | |
9a46dba7 | 877 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
d63ffd87 | 878 | if (!s->synced) { |
7483d1e5 | 879 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba PB |
880 | if (block_job_is_cancelled(&s->common)) { |
881 | break; | |
882 | } | |
883 | } else if (!should_complete) { | |
bd48bde8 | 884 | delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); |
7483d1e5 | 885 | block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); |
893f7eba | 886 | } |
49efb1f5 | 887 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
888 | } |
889 | ||
890 | immediate_exit: | |
bd48bde8 PB |
891 | if (s->in_flight > 0) { |
892 | /* We get here only if something went wrong. Either the job failed, | |
893 | * or it was cancelled prematurely so that we do not guarantee that | |
894 | * the target is a copy of the source. | |
895 | */ | |
896 | assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); | |
9a0cec66 | 897 | assert(need_drain); |
bae8196d | 898 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
899 | } |
900 | ||
901 | assert(s->in_flight == 0); | |
7191bf31 | 902 | qemu_vfree(s->buf); |
b812f671 | 903 | g_free(s->cow_bitmap); |
402a4741 | 904 | g_free(s->in_flight_bitmap); |
dc162c8e | 905 | bdrv_dirty_iter_free(s->dbi); |
5a7e7a0b SH |
906 | |
907 | data = g_malloc(sizeof(*data)); | |
908 | data->ret = ret; | |
9a0cec66 PB |
909 | |
910 | if (need_drain) { | |
911 | bdrv_drained_begin(bs); | |
912 | } | |
5a7e7a0b | 913 | block_job_defer_to_main_loop(&s->common, mirror_exit, data); |
893f7eba PB |
914 | } |
915 | ||
916 | static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
917 | { | |
918 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
919 | ||
920 | if (speed < 0) { | |
c6bd8c70 | 921 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
893f7eba PB |
922 | return; |
923 | } | |
f3e4ce4a | 924 | ratelimit_set_speed(&s->limit, speed, SLICE_TIME); |
893f7eba PB |
925 | } |
926 | ||
d63ffd87 PB |
927 | static void mirror_complete(BlockJob *job, Error **errp) |
928 | { | |
929 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
4ef85a9c | 930 | BlockDriverState *target; |
274fccee | 931 | |
274fccee | 932 | target = blk_bs(s->target); |
d63ffd87 | 933 | |
d63ffd87 | 934 | if (!s->synced) { |
9df229c3 AG |
935 | error_setg(errp, "The active block job '%s' cannot be completed", |
936 | job->id); | |
d63ffd87 PB |
937 | return; |
938 | } | |
939 | ||
274fccee HR |
940 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
941 | int ret; | |
942 | ||
943 | assert(!target->backing); | |
944 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
945 | if (ret < 0) { | |
946 | return; | |
947 | } | |
948 | } | |
949 | ||
15d67298 | 950 | /* block all operations on to_replace bs */ |
09158f00 | 951 | if (s->replaces) { |
5a7e7a0b SH |
952 | AioContext *replace_aio_context; |
953 | ||
e12f3784 | 954 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 955 | if (!s->to_replace) { |
e12f3784 | 956 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
957 | return; |
958 | } | |
959 | ||
5a7e7a0b SH |
960 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
961 | aio_context_acquire(replace_aio_context); | |
962 | ||
4ef85a9c KW |
963 | /* TODO Translate this into permission system. Current definition of |
964 | * GRAPH_MOD would require to request it for the parents; they might | |
965 | * not even be BlockDriverStates, however, so a BdrvChild can't address | |
966 | * them. May need redefinition of GRAPH_MOD. */ | |
09158f00 BC |
967 | error_setg(&s->replace_blocker, |
968 | "block device is in use by block-job-complete"); | |
969 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
970 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
971 | |
972 | aio_context_release(replace_aio_context); | |
09158f00 BC |
973 | } |
974 | ||
d63ffd87 | 975 | s->should_complete = true; |
751ebd76 | 976 | block_job_enter(&s->common); |
d63ffd87 PB |
977 | } |
978 | ||
bae8196d | 979 | static void mirror_pause(BlockJob *job) |
565ac01f SH |
980 | { |
981 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
982 | ||
bae8196d | 983 | mirror_wait_for_all_io(s); |
565ac01f SH |
984 | } |
985 | ||
986 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) | |
987 | { | |
988 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
989 | ||
990 | blk_set_aio_context(s->target, new_context); | |
991 | } | |
992 | ||
bae8196d PB |
993 | static void mirror_drain(BlockJob *job) |
994 | { | |
995 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
996 | ||
997 | /* Need to keep a reference in case blk_drain triggers execution | |
998 | * of mirror_complete... | |
999 | */ | |
1000 | if (s->target) { | |
1001 | BlockBackend *target = s->target; | |
1002 | blk_ref(target); | |
1003 | blk_drain(target); | |
1004 | blk_unref(target); | |
1005 | } | |
1006 | } | |
1007 | ||
3fc4b10a | 1008 | static const BlockJobDriver mirror_job_driver = { |
565ac01f SH |
1009 | .instance_size = sizeof(MirrorBlockJob), |
1010 | .job_type = BLOCK_JOB_TYPE_MIRROR, | |
1011 | .set_speed = mirror_set_speed, | |
a7815a76 | 1012 | .start = mirror_run, |
565ac01f SH |
1013 | .complete = mirror_complete, |
1014 | .pause = mirror_pause, | |
1015 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1016 | .drain = mirror_drain, |
893f7eba PB |
1017 | }; |
1018 | ||
03544a6e | 1019 | static const BlockJobDriver commit_active_job_driver = { |
565ac01f SH |
1020 | .instance_size = sizeof(MirrorBlockJob), |
1021 | .job_type = BLOCK_JOB_TYPE_COMMIT, | |
1022 | .set_speed = mirror_set_speed, | |
a7815a76 | 1023 | .start = mirror_run, |
565ac01f SH |
1024 | .complete = mirror_complete, |
1025 | .pause = mirror_pause, | |
1026 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1027 | .drain = mirror_drain, |
03544a6e FZ |
1028 | }; |
1029 | ||
4ef85a9c KW |
1030 | static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, |
1031 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1032 | { | |
1033 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1034 | } | |
1035 | ||
1036 | static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, | |
1037 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1038 | { | |
1039 | return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1040 | } | |
1041 | ||
1042 | static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) | |
1043 | { | |
ce960aa9 VSO |
1044 | if (bs->backing == NULL) { |
1045 | /* we can be here after failed bdrv_append in mirror_start_job */ | |
1046 | return 0; | |
1047 | } | |
4ef85a9c KW |
1048 | return bdrv_co_flush(bs->backing->bs); |
1049 | } | |
1050 | ||
4ef85a9c | 1051 | static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 1052 | int64_t offset, int bytes, BdrvRequestFlags flags) |
4ef85a9c | 1053 | { |
f5a5ca79 | 1054 | return bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); |
4ef85a9c KW |
1055 | } |
1056 | ||
1057 | static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, | |
f5a5ca79 | 1058 | int64_t offset, int bytes) |
4ef85a9c | 1059 | { |
f5a5ca79 | 1060 | return bdrv_co_pdiscard(bs->backing->bs, offset, bytes); |
4ef85a9c KW |
1061 | } |
1062 | ||
fd4a6493 KW |
1063 | static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) |
1064 | { | |
18775ff3 VSO |
1065 | if (bs->backing == NULL) { |
1066 | /* we can be here after failed bdrv_attach_child in | |
1067 | * bdrv_set_backing_hd */ | |
1068 | return; | |
1069 | } | |
fd4a6493 KW |
1070 | bdrv_refresh_filename(bs->backing->bs); |
1071 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
1072 | bs->backing->bs->filename); | |
1073 | } | |
1074 | ||
4ef85a9c KW |
1075 | static void bdrv_mirror_top_close(BlockDriverState *bs) |
1076 | { | |
1077 | } | |
1078 | ||
1079 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, | |
1080 | const BdrvChildRole *role, | |
e0995dc3 | 1081 | BlockReopenQueue *reopen_queue, |
4ef85a9c KW |
1082 | uint64_t perm, uint64_t shared, |
1083 | uint64_t *nperm, uint64_t *nshared) | |
1084 | { | |
1085 | /* Must be able to forward guest writes to the real image */ | |
1086 | *nperm = 0; | |
1087 | if (perm & BLK_PERM_WRITE) { | |
1088 | *nperm |= BLK_PERM_WRITE; | |
1089 | } | |
1090 | ||
1091 | *nshared = BLK_PERM_ALL; | |
1092 | } | |
1093 | ||
1094 | /* Dummy node that provides consistent read to its users without requiring it | |
1095 | * from its backing file and that allows writes on the backing file chain. */ | |
1096 | static BlockDriver bdrv_mirror_top = { | |
1097 | .format_name = "mirror_top", | |
1098 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1099 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1100 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1101 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1102 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
f7cc69b3 | 1103 | .bdrv_co_get_block_status = bdrv_co_get_block_status_from_backing, |
fd4a6493 | 1104 | .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, |
4ef85a9c KW |
1105 | .bdrv_close = bdrv_mirror_top_close, |
1106 | .bdrv_child_perm = bdrv_mirror_top_child_perm, | |
1107 | }; | |
1108 | ||
71aa9867 | 1109 | static void mirror_start_job(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1110 | int creation_flags, BlockDriverState *target, |
1111 | const char *replaces, int64_t speed, | |
1112 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1113 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
1114 | BlockdevOnError on_source_error, |
1115 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1116 | bool unmap, |
097310b5 | 1117 | BlockCompletionFunc *cb, |
51ccfa2d | 1118 | void *opaque, |
09158f00 | 1119 | const BlockJobDriver *driver, |
b49f7ead | 1120 | bool is_none_mode, BlockDriverState *base, |
51ccfa2d | 1121 | bool auto_complete, const char *filter_node_name, |
045a2f82 | 1122 | bool is_mirror, |
51ccfa2d | 1123 | Error **errp) |
893f7eba PB |
1124 | { |
1125 | MirrorBlockJob *s; | |
4ef85a9c KW |
1126 | BlockDriverState *mirror_top_bs; |
1127 | bool target_graph_mod; | |
1128 | bool target_is_backing; | |
b2c2832c | 1129 | Error *local_err = NULL; |
d7086422 | 1130 | int ret; |
893f7eba | 1131 | |
eee13dfe | 1132 | if (granularity == 0) { |
341ebc2f | 1133 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1134 | } |
1135 | ||
1136 | assert ((granularity & (granularity - 1)) == 0); | |
b436982f EB |
1137 | /* Granularity must be large enough for sector-based dirty bitmap */ |
1138 | assert(granularity >= BDRV_SECTOR_SIZE); | |
eee13dfe | 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 | } | |
d3c8c674 KW |
1157 | if (!filter_node_name) { |
1158 | mirror_top_bs->implicit = true; | |
1159 | } | |
4ef85a9c | 1160 | mirror_top_bs->total_sectors = bs->total_sectors; |
19dd29e8 | 1161 | bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); |
4ef85a9c KW |
1162 | |
1163 | /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep | |
7a25fcd0 | 1164 | * it alive until block_job_create() succeeds even if bs has no parent. */ |
4ef85a9c KW |
1165 | bdrv_ref(mirror_top_bs); |
1166 | bdrv_drained_begin(bs); | |
b2c2832c | 1167 | bdrv_append(mirror_top_bs, bs, &local_err); |
4ef85a9c KW |
1168 | bdrv_drained_end(bs); |
1169 | ||
b2c2832c KW |
1170 | if (local_err) { |
1171 | bdrv_unref(mirror_top_bs); | |
1172 | error_propagate(errp, local_err); | |
1173 | return; | |
1174 | } | |
1175 | ||
4ef85a9c KW |
1176 | /* Make sure that the source is not resized while the job is running */ |
1177 | s = block_job_create(job_id, driver, mirror_top_bs, | |
1178 | BLK_PERM_CONSISTENT_READ, | |
1179 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
1180 | BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, | |
c6cc12bf | 1181 | creation_flags, cb, opaque, errp); |
893f7eba | 1182 | if (!s) { |
4ef85a9c | 1183 | goto fail; |
893f7eba | 1184 | } |
7a25fcd0 HR |
1185 | /* The block job now has a reference to this node */ |
1186 | bdrv_unref(mirror_top_bs); | |
1187 | ||
4ef85a9c KW |
1188 | s->source = bs; |
1189 | s->mirror_top_bs = mirror_top_bs; | |
1190 | ||
1191 | /* No resize for the target either; while the mirror is still running, a | |
1192 | * consistent read isn't necessarily possible. We could possibly allow | |
1193 | * writes and graph modifications, though it would likely defeat the | |
1194 | * purpose of a mirror, so leave them blocked for now. | |
1195 | * | |
1196 | * In the case of active commit, things look a bit different, though, | |
1197 | * because the target is an already populated backing file in active use. | |
1198 | * We can allow anything except resize there.*/ | |
1199 | target_is_backing = bdrv_chain_contains(bs, target); | |
1200 | target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); | |
1201 | s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | | |
1202 | (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), | |
1203 | BLK_PERM_WRITE_UNCHANGED | | |
1204 | (target_is_backing ? BLK_PERM_CONSISTENT_READ | | |
1205 | BLK_PERM_WRITE | | |
1206 | BLK_PERM_GRAPH_MOD : 0)); | |
d7086422 KW |
1207 | ret = blk_insert_bs(s->target, target, errp); |
1208 | if (ret < 0) { | |
4ef85a9c | 1209 | goto fail; |
d7086422 | 1210 | } |
045a2f82 FZ |
1211 | if (is_mirror) { |
1212 | /* XXX: Mirror target could be a NBD server of target QEMU in the case | |
1213 | * of non-shared block migration. To allow migration completion, we | |
1214 | * have to allow "inactivate" of the target BB. When that happens, we | |
1215 | * know the job is drained, and the vcpus are stopped, so no write | |
1216 | * operation will be performed. Block layer already has assertions to | |
1217 | * ensure that. */ | |
1218 | blk_set_force_allow_inactivate(s->target); | |
1219 | } | |
e253f4b8 | 1220 | |
09158f00 | 1221 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1222 | s->on_source_error = on_source_error; |
1223 | s->on_target_error = on_target_error; | |
03544a6e | 1224 | s->is_none_mode = is_none_mode; |
274fccee | 1225 | s->backing_mode = backing_mode; |
5bc361b8 | 1226 | s->base = base; |
eee13dfe | 1227 | s->granularity = granularity; |
48ac0a4d | 1228 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1229 | s->unmap = unmap; |
b49f7ead WC |
1230 | if (auto_complete) { |
1231 | s->should_complete = true; | |
1232 | } | |
b812f671 | 1233 | |
0db6e54a | 1234 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1235 | if (!s->dirty_bitmap) { |
88f9d1b3 | 1236 | goto fail; |
b8afb520 | 1237 | } |
10f3cd15 | 1238 | |
4ef85a9c | 1239 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1240 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1241 | &error_abort); | |
1242 | ||
f3ede4b0 AG |
1243 | /* In commit_active_start() all intermediate nodes disappear, so |
1244 | * any jobs in them must be blocked */ | |
4ef85a9c | 1245 | if (target_is_backing) { |
f3ede4b0 AG |
1246 | BlockDriverState *iter; |
1247 | for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { | |
4ef85a9c KW |
1248 | /* XXX BLK_PERM_WRITE needs to be allowed so we don't block |
1249 | * ourselves at s->base (if writes are blocked for a node, they are | |
1250 | * also blocked for its backing file). The other options would be a | |
1251 | * second filter driver above s->base (== target). */ | |
1252 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, | |
1253 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, | |
1254 | errp); | |
1255 | if (ret < 0) { | |
1256 | goto fail; | |
1257 | } | |
f3ede4b0 AG |
1258 | } |
1259 | } | |
10f3cd15 | 1260 | |
5ccac6f1 JS |
1261 | trace_mirror_start(bs, s, opaque); |
1262 | block_job_start(&s->common); | |
4ef85a9c KW |
1263 | return; |
1264 | ||
1265 | fail: | |
1266 | if (s) { | |
7a25fcd0 HR |
1267 | /* Make sure this BDS does not go away until we have completed the graph |
1268 | * changes below */ | |
1269 | bdrv_ref(mirror_top_bs); | |
1270 | ||
4ef85a9c KW |
1271 | g_free(s->replaces); |
1272 | blk_unref(s->target); | |
05b0d8e3 | 1273 | block_job_early_fail(&s->common); |
4ef85a9c KW |
1274 | } |
1275 | ||
c1cef672 FZ |
1276 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
1277 | &error_abort); | |
5fe31c25 | 1278 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
7a25fcd0 HR |
1279 | |
1280 | bdrv_unref(mirror_top_bs); | |
893f7eba | 1281 | } |
03544a6e | 1282 | |
71aa9867 AG |
1283 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1284 | BlockDriverState *target, const char *replaces, | |
5fba6c0e | 1285 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
1286 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
1287 | BlockdevOnError on_source_error, | |
03544a6e | 1288 | BlockdevOnError on_target_error, |
6cdbceb1 | 1289 | bool unmap, const char *filter_node_name, Error **errp) |
03544a6e FZ |
1290 | { |
1291 | bool is_none_mode; | |
1292 | BlockDriverState *base; | |
1293 | ||
4b80ab2b JS |
1294 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
1295 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
1296 | return; |
1297 | } | |
03544a6e | 1298 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 1299 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
47970dfb | 1300 | mirror_start_job(job_id, bs, BLOCK_JOB_DEFAULT, target, replaces, |
274fccee | 1301 | speed, granularity, buf_size, backing_mode, |
51ccfa2d | 1302 | on_source_error, on_target_error, unmap, NULL, NULL, |
6cdbceb1 | 1303 | &mirror_job_driver, is_none_mode, base, false, |
045a2f82 | 1304 | filter_node_name, true, errp); |
03544a6e FZ |
1305 | } |
1306 | ||
fd62c609 | 1307 | void commit_active_start(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1308 | BlockDriverState *base, int creation_flags, |
1309 | int64_t speed, BlockdevOnError on_error, | |
0db832f4 | 1310 | const char *filter_node_name, |
78bbd910 FZ |
1311 | BlockCompletionFunc *cb, void *opaque, |
1312 | bool auto_complete, Error **errp) | |
03544a6e | 1313 | { |
4da83585 | 1314 | int orig_base_flags; |
cc67f4d1 | 1315 | Error *local_err = NULL; |
4da83585 JC |
1316 | |
1317 | orig_base_flags = bdrv_get_flags(base); | |
1318 | ||
20a63d2c FZ |
1319 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
1320 | return; | |
1321 | } | |
4da83585 | 1322 | |
47970dfb | 1323 | mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
71aa9867 | 1324 | MIRROR_LEAVE_BACKING_CHAIN, |
51ccfa2d | 1325 | on_error, on_error, true, cb, opaque, |
6cdbceb1 | 1326 | &commit_active_job_driver, false, base, auto_complete, |
045a2f82 | 1327 | filter_node_name, false, &local_err); |
0fb6395c | 1328 | if (local_err) { |
cc67f4d1 | 1329 | error_propagate(errp, local_err); |
4da83585 JC |
1330 | goto error_restore_flags; |
1331 | } | |
1332 | ||
1333 | return; | |
1334 | ||
1335 | error_restore_flags: | |
1336 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1337 | * the original error */ | |
1338 | bdrv_reopen(base, orig_base_flags, NULL); | |
1339 | return; | |
03544a6e | 1340 | } |