]>
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 | 192 | int64_t align_offset = *offset; |
7cfd5275 | 193 | int64_t align_bytes = *bytes; |
782d97ef | 194 | int max_bytes = s->granularity * s->max_iov; |
e5b43573 | 195 | |
782d97ef EB |
196 | need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); |
197 | need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, | |
e5b43573 FZ |
198 | s->cow_bitmap); |
199 | if (need_cow) { | |
782d97ef EB |
200 | bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, |
201 | &align_offset, &align_bytes); | |
e5b43573 | 202 | } |
3515727f | 203 | |
782d97ef EB |
204 | if (align_bytes > max_bytes) { |
205 | align_bytes = max_bytes; | |
e5b43573 | 206 | if (need_cow) { |
782d97ef | 207 | align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); |
e5b43573 | 208 | } |
8f0720ec | 209 | } |
782d97ef | 210 | /* Clipping may result in align_bytes unaligned to chunk boundary, but |
4150ae60 | 211 | * that doesn't matter because it's already the end of source image. */ |
782d97ef | 212 | align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); |
8f0720ec | 213 | |
782d97ef EB |
214 | ret = align_offset + align_bytes - (*offset + *bytes); |
215 | *offset = align_offset; | |
216 | *bytes = align_bytes; | |
e5b43573 FZ |
217 | assert(ret >= 0); |
218 | return ret; | |
219 | } | |
220 | ||
21cd917f FZ |
221 | static inline void mirror_wait_for_io(MirrorBlockJob *s) |
222 | { | |
223 | assert(!s->waiting_for_io); | |
224 | s->waiting_for_io = true; | |
225 | qemu_coroutine_yield(); | |
226 | s->waiting_for_io = false; | |
227 | } | |
228 | ||
e5b43573 | 229 | /* Submit async read while handling COW. |
ae4cc877 EB |
230 | * Returns: The number of bytes copied after and including offset, |
231 | * excluding any bytes copied prior to offset due to alignment. | |
232 | * This will be @bytes if no alignment is necessary, or | |
233 | * (new_end - offset) if tail is rounded up or down due to | |
e5b43573 FZ |
234 | * alignment or buffer limit. |
235 | */ | |
ae4cc877 EB |
236 | static uint64_t mirror_do_read(MirrorBlockJob *s, int64_t offset, |
237 | uint64_t bytes) | |
e5b43573 | 238 | { |
e253f4b8 | 239 | BlockBackend *source = s->common.blk; |
ae4cc877 EB |
240 | int nb_chunks; |
241 | uint64_t ret; | |
e5b43573 | 242 | MirrorOp *op; |
ae4cc877 | 243 | uint64_t max_bytes; |
e5b43573 | 244 | |
ae4cc877 | 245 | max_bytes = s->granularity * s->max_iov; |
402a4741 | 246 | |
e5b43573 | 247 | /* We can only handle as much as buf_size at a time. */ |
ae4cc877 EB |
248 | bytes = MIN(s->buf_size, MIN(max_bytes, bytes)); |
249 | assert(bytes); | |
250 | assert(bytes < BDRV_REQUEST_MAX_BYTES); | |
251 | ret = bytes; | |
402a4741 | 252 | |
e5b43573 | 253 | if (s->cow_bitmap) { |
ae4cc877 | 254 | ret += mirror_cow_align(s, &offset, &bytes); |
e5b43573 | 255 | } |
ae4cc877 EB |
256 | assert(bytes <= s->buf_size); |
257 | /* The offset is granularity-aligned because: | |
e5b43573 FZ |
258 | * 1) Caller passes in aligned values; |
259 | * 2) mirror_cow_align is used only when target cluster is larger. */ | |
ae4cc877 EB |
260 | assert(QEMU_IS_ALIGNED(offset, s->granularity)); |
261 | /* The range is sector-aligned, since bdrv_getlength() rounds up. */ | |
262 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
263 | nb_chunks = DIV_ROUND_UP(bytes, s->granularity); | |
e5b43573 FZ |
264 | |
265 | while (s->buf_free_count < nb_chunks) { | |
ae4cc877 | 266 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
21cd917f | 267 | mirror_wait_for_io(s); |
b812f671 PB |
268 | } |
269 | ||
bd48bde8 | 270 | /* Allocate a MirrorOp that is used as an AIO callback. */ |
c84b3192 | 271 | op = g_new(MirrorOp, 1); |
bd48bde8 | 272 | op->s = s; |
ae4cc877 EB |
273 | op->offset = offset; |
274 | op->bytes = bytes; | |
402a4741 PB |
275 | |
276 | /* Now make a QEMUIOVector taking enough granularity-sized chunks | |
277 | * from s->buf_free. | |
278 | */ | |
279 | qemu_iovec_init(&op->qiov, nb_chunks); | |
402a4741 PB |
280 | while (nb_chunks-- > 0) { |
281 | MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); | |
ae4cc877 | 282 | size_t remaining = bytes - op->qiov.size; |
5a0f6fd5 | 283 | |
402a4741 PB |
284 | QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); |
285 | s->buf_free_count--; | |
5a0f6fd5 | 286 | qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); |
402a4741 | 287 | } |
bd48bde8 | 288 | |
893f7eba | 289 | /* Copy the dirty cluster. */ |
bd48bde8 | 290 | s->in_flight++; |
ae4cc877 EB |
291 | s->bytes_in_flight += bytes; |
292 | trace_mirror_one_iteration(s, offset, bytes); | |
dcfb3beb | 293 | |
ae4cc877 | 294 | blk_aio_preadv(source, offset, &op->qiov, 0, mirror_read_complete, op); |
e5b43573 FZ |
295 | return ret; |
296 | } | |
297 | ||
298 | static void mirror_do_zero_or_discard(MirrorBlockJob *s, | |
e6f24193 EB |
299 | int64_t offset, |
300 | uint64_t bytes, | |
e5b43573 FZ |
301 | bool is_discard) |
302 | { | |
303 | MirrorOp *op; | |
304 | ||
305 | /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed | |
306 | * so the freeing in mirror_iteration_done is nop. */ | |
307 | op = g_new0(MirrorOp, 1); | |
308 | op->s = s; | |
e6f24193 EB |
309 | op->offset = offset; |
310 | op->bytes = bytes; | |
e5b43573 FZ |
311 | |
312 | s->in_flight++; | |
e6f24193 | 313 | s->bytes_in_flight += bytes; |
e5b43573 | 314 | if (is_discard) { |
e6f24193 | 315 | blk_aio_pdiscard(s->target, offset, |
b436982f | 316 | op->bytes, mirror_write_complete, op); |
e5b43573 | 317 | } else { |
e6f24193 | 318 | blk_aio_pwrite_zeroes(s->target, offset, |
b436982f | 319 | op->bytes, s->unmap ? BDRV_REQ_MAY_UNMAP : 0, |
dcfb3beb | 320 | mirror_write_complete, op); |
e5b43573 FZ |
321 | } |
322 | } | |
323 | ||
324 | static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) | |
325 | { | |
4ef85a9c | 326 | BlockDriverState *source = s->source; |
fb2ef791 | 327 | int64_t offset, first_chunk; |
e5b43573 FZ |
328 | uint64_t delay_ns = 0; |
329 | /* At least the first dirty chunk is mirrored in one iteration. */ | |
330 | int nb_chunks = 1; | |
4b5004d9 | 331 | bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); |
b436982f | 332 | int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); |
e5b43573 | 333 | |
b64bd51e | 334 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
f798184c | 335 | offset = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 336 | if (offset < 0) { |
dc162c8e | 337 | bdrv_set_dirty_iter(s->dbi, 0); |
f798184c | 338 | offset = bdrv_dirty_iter_next(s->dbi); |
9a46dba7 | 339 | trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); |
fb2ef791 | 340 | assert(offset >= 0); |
e5b43573 | 341 | } |
b64bd51e | 342 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
e5b43573 | 343 | |
fb2ef791 | 344 | first_chunk = offset / s->granularity; |
9c83625b | 345 | while (test_bit(first_chunk, s->in_flight_bitmap)) { |
fb2ef791 | 346 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
9c83625b HR |
347 | mirror_wait_for_io(s); |
348 | } | |
349 | ||
565ac01f SH |
350 | block_job_pause_point(&s->common); |
351 | ||
e5b43573 FZ |
352 | /* Find the number of consective dirty chunks following the first dirty |
353 | * one, and wait for in flight requests in them. */ | |
b64bd51e | 354 | bdrv_dirty_bitmap_lock(s->dirty_bitmap); |
fb2ef791 | 355 | while (nb_chunks * s->granularity < s->buf_size) { |
dc162c8e | 356 | int64_t next_dirty; |
fb2ef791 EB |
357 | int64_t next_offset = offset + nb_chunks * s->granularity; |
358 | int64_t next_chunk = next_offset / s->granularity; | |
359 | if (next_offset >= s->bdev_length || | |
3b5d4df0 | 360 | !bdrv_get_dirty_locked(source, s->dirty_bitmap, next_offset)) { |
e5b43573 FZ |
361 | break; |
362 | } | |
363 | if (test_bit(next_chunk, s->in_flight_bitmap)) { | |
9c83625b | 364 | break; |
e5b43573 | 365 | } |
9c83625b | 366 | |
f798184c | 367 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
fb2ef791 | 368 | if (next_dirty > next_offset || next_dirty < 0) { |
f27a2742 | 369 | /* The bitmap iterator's cache is stale, refresh it */ |
715a74d8 | 370 | bdrv_set_dirty_iter(s->dbi, next_offset); |
f798184c | 371 | next_dirty = bdrv_dirty_iter_next(s->dbi); |
f27a2742 | 372 | } |
fb2ef791 | 373 | assert(next_dirty == next_offset); |
9c83625b | 374 | nb_chunks++; |
e5b43573 FZ |
375 | } |
376 | ||
377 | /* Clear dirty bits before querying the block status, because | |
31826642 | 378 | * calling bdrv_block_status_above could yield - if some blocks are |
e5b43573 FZ |
379 | * marked dirty in this window, we need to know. |
380 | */ | |
e0d7f73e EB |
381 | bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, |
382 | nb_chunks * s->granularity); | |
b64bd51e PB |
383 | bdrv_dirty_bitmap_unlock(s->dirty_bitmap); |
384 | ||
fb2ef791 EB |
385 | bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); |
386 | while (nb_chunks > 0 && offset < s->bdev_length) { | |
31826642 | 387 | int ret; |
7cfd5275 | 388 | int64_t io_bytes; |
f3e4ce4a | 389 | int64_t io_bytes_acct; |
e5b43573 FZ |
390 | enum MirrorMethod { |
391 | MIRROR_METHOD_COPY, | |
392 | MIRROR_METHOD_ZERO, | |
393 | MIRROR_METHOD_DISCARD | |
394 | } mirror_method = MIRROR_METHOD_COPY; | |
395 | ||
fb2ef791 | 396 | assert(!(offset % s->granularity)); |
31826642 EB |
397 | ret = bdrv_block_status_above(source, NULL, offset, |
398 | nb_chunks * s->granularity, | |
399 | &io_bytes, NULL, NULL); | |
e5b43573 | 400 | if (ret < 0) { |
fb2ef791 | 401 | io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); |
0965a41e | 402 | } else if (ret & BDRV_BLOCK_DATA) { |
fb2ef791 | 403 | io_bytes = MIN(io_bytes, max_io_bytes); |
e5b43573 FZ |
404 | } |
405 | ||
fb2ef791 EB |
406 | io_bytes -= io_bytes % s->granularity; |
407 | if (io_bytes < s->granularity) { | |
408 | io_bytes = s->granularity; | |
e5b43573 | 409 | } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { |
fb2ef791 | 410 | int64_t target_offset; |
7cfd5275 | 411 | int64_t target_bytes; |
fb2ef791 EB |
412 | bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes, |
413 | &target_offset, &target_bytes); | |
414 | if (target_offset == offset && | |
415 | target_bytes == io_bytes) { | |
e5b43573 FZ |
416 | mirror_method = ret & BDRV_BLOCK_ZERO ? |
417 | MIRROR_METHOD_ZERO : | |
418 | MIRROR_METHOD_DISCARD; | |
419 | } | |
420 | } | |
421 | ||
cf56a3c6 | 422 | while (s->in_flight >= MAX_IN_FLIGHT) { |
fb2ef791 | 423 | trace_mirror_yield_in_flight(s, offset, s->in_flight); |
cf56a3c6 DL |
424 | mirror_wait_for_io(s); |
425 | } | |
426 | ||
dbaa7b57 VSO |
427 | if (s->ret < 0) { |
428 | return 0; | |
429 | } | |
430 | ||
fb2ef791 | 431 | io_bytes = mirror_clip_bytes(s, offset, io_bytes); |
e5b43573 FZ |
432 | switch (mirror_method) { |
433 | case MIRROR_METHOD_COPY: | |
fb2ef791 | 434 | io_bytes = io_bytes_acct = mirror_do_read(s, offset, io_bytes); |
e5b43573 FZ |
435 | break; |
436 | case MIRROR_METHOD_ZERO: | |
e5b43573 | 437 | case MIRROR_METHOD_DISCARD: |
fb2ef791 | 438 | mirror_do_zero_or_discard(s, offset, io_bytes, |
4b5004d9 DL |
439 | mirror_method == MIRROR_METHOD_DISCARD); |
440 | if (write_zeroes_ok) { | |
f3e4ce4a | 441 | io_bytes_acct = 0; |
4b5004d9 | 442 | } else { |
fb2ef791 | 443 | io_bytes_acct = io_bytes; |
4b5004d9 | 444 | } |
e5b43573 FZ |
445 | break; |
446 | default: | |
447 | abort(); | |
448 | } | |
fb2ef791 EB |
449 | assert(io_bytes); |
450 | offset += io_bytes; | |
451 | nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); | |
f14a39cc | 452 | if (s->common.speed) { |
f3e4ce4a | 453 | delay_ns = ratelimit_calculate_delay(&s->limit, io_bytes_acct); |
f14a39cc | 454 | } |
dcfb3beb | 455 | } |
cc8c9d6c | 456 | return delay_ns; |
bd48bde8 | 457 | } |
b952b558 | 458 | |
402a4741 PB |
459 | static void mirror_free_init(MirrorBlockJob *s) |
460 | { | |
461 | int granularity = s->granularity; | |
462 | size_t buf_size = s->buf_size; | |
463 | uint8_t *buf = s->buf; | |
464 | ||
465 | assert(s->buf_free_count == 0); | |
466 | QSIMPLEQ_INIT(&s->buf_free); | |
467 | while (buf_size != 0) { | |
468 | MirrorBuffer *cur = (MirrorBuffer *)buf; | |
469 | QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); | |
470 | s->buf_free_count++; | |
471 | buf_size -= granularity; | |
472 | buf += granularity; | |
473 | } | |
474 | } | |
475 | ||
bae8196d PB |
476 | /* This is also used for the .pause callback. There is no matching |
477 | * mirror_resume() because mirror_run() will begin iterating again | |
478 | * when the job is resumed. | |
479 | */ | |
480 | static void mirror_wait_for_all_io(MirrorBlockJob *s) | |
bd48bde8 PB |
481 | { |
482 | while (s->in_flight > 0) { | |
21cd917f | 483 | mirror_wait_for_io(s); |
bd48bde8 | 484 | } |
893f7eba PB |
485 | } |
486 | ||
5a7e7a0b SH |
487 | typedef struct { |
488 | int ret; | |
489 | } MirrorExitData; | |
490 | ||
491 | static void mirror_exit(BlockJob *job, void *opaque) | |
492 | { | |
493 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
494 | MirrorExitData *data = opaque; | |
495 | AioContext *replace_aio_context = NULL; | |
4ef85a9c | 496 | BlockDriverState *src = s->source; |
e253f4b8 | 497 | BlockDriverState *target_bs = blk_bs(s->target); |
4ef85a9c | 498 | BlockDriverState *mirror_top_bs = s->mirror_top_bs; |
12fa4af6 | 499 | Error *local_err = NULL; |
3f09bfbc | 500 | |
2119882c PB |
501 | bdrv_release_dirty_bitmap(src, s->dirty_bitmap); |
502 | ||
3f09bfbc KW |
503 | /* Make sure that the source BDS doesn't go away before we called |
504 | * block_job_completed(). */ | |
505 | bdrv_ref(src); | |
4ef85a9c | 506 | bdrv_ref(mirror_top_bs); |
7d9fcb39 KW |
507 | bdrv_ref(target_bs); |
508 | ||
509 | /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before | |
510 | * inserting target_bs at s->to_replace, where we might not be able to get | |
63c8ef28 KW |
511 | * these permissions. |
512 | * | |
513 | * Note that blk_unref() alone doesn't necessarily drop permissions because | |
514 | * we might be running nested inside mirror_drain(), which takes an extra | |
515 | * reference, so use an explicit blk_set_perm() first. */ | |
516 | blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); | |
7d9fcb39 KW |
517 | blk_unref(s->target); |
518 | s->target = NULL; | |
4ef85a9c KW |
519 | |
520 | /* We don't access the source any more. Dropping any WRITE/RESIZE is | |
521 | * required before it could become a backing file of target_bs. */ | |
522 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, | |
523 | &error_abort); | |
524 | if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { | |
525 | BlockDriverState *backing = s->is_none_mode ? src : s->base; | |
526 | if (backing_bs(target_bs) != backing) { | |
12fa4af6 KW |
527 | bdrv_set_backing_hd(target_bs, backing, &local_err); |
528 | if (local_err) { | |
529 | error_report_err(local_err); | |
530 | data->ret = -EPERM; | |
531 | } | |
4ef85a9c KW |
532 | } |
533 | } | |
5a7e7a0b SH |
534 | |
535 | if (s->to_replace) { | |
536 | replace_aio_context = bdrv_get_aio_context(s->to_replace); | |
537 | aio_context_acquire(replace_aio_context); | |
538 | } | |
539 | ||
540 | if (s->should_complete && data->ret == 0) { | |
e253f4b8 | 541 | BlockDriverState *to_replace = src; |
5a7e7a0b SH |
542 | if (s->to_replace) { |
543 | to_replace = s->to_replace; | |
544 | } | |
40365552 | 545 | |
e253f4b8 KW |
546 | if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { |
547 | bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); | |
5a7e7a0b | 548 | } |
b8804815 KW |
549 | |
550 | /* The mirror job has no requests in flight any more, but we need to | |
551 | * drain potential other users of the BDS before changing the graph. */ | |
e253f4b8 | 552 | bdrv_drained_begin(target_bs); |
5fe31c25 | 553 | bdrv_replace_node(to_replace, target_bs, &local_err); |
e253f4b8 | 554 | bdrv_drained_end(target_bs); |
5fe31c25 KW |
555 | if (local_err) { |
556 | error_report_err(local_err); | |
557 | data->ret = -EPERM; | |
558 | } | |
5a7e7a0b SH |
559 | } |
560 | if (s->to_replace) { | |
561 | bdrv_op_unblock_all(s->to_replace, s->replace_blocker); | |
562 | error_free(s->replace_blocker); | |
563 | bdrv_unref(s->to_replace); | |
564 | } | |
565 | if (replace_aio_context) { | |
566 | aio_context_release(replace_aio_context); | |
567 | } | |
568 | g_free(s->replaces); | |
7d9fcb39 | 569 | bdrv_unref(target_bs); |
4ef85a9c KW |
570 | |
571 | /* Remove the mirror filter driver from the graph. Before this, get rid of | |
572 | * the blockers on the intermediate nodes so that the resulting state is | |
0bf74767 KW |
573 | * valid. Also give up permissions on mirror_top_bs->backing, which might |
574 | * block the removal. */ | |
4ef85a9c | 575 | block_job_remove_all_bdrv(job); |
c1cef672 FZ |
576 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
577 | &error_abort); | |
5fe31c25 | 578 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
4ef85a9c KW |
579 | |
580 | /* We just changed the BDS the job BB refers to (with either or both of the | |
5fe31c25 KW |
581 | * bdrv_replace_node() calls), so switch the BB back so the cleanup does |
582 | * the right thing. We don't need any permissions any more now. */ | |
4ef85a9c KW |
583 | blk_remove_bs(job->blk); |
584 | blk_set_perm(job->blk, 0, BLK_PERM_ALL, &error_abort); | |
585 | blk_insert_bs(job->blk, mirror_top_bs, &error_abort); | |
586 | ||
5a7e7a0b | 587 | block_job_completed(&s->common, data->ret); |
4ef85a9c | 588 | |
5a7e7a0b | 589 | g_free(data); |
176c3699 | 590 | bdrv_drained_end(src); |
4ef85a9c | 591 | bdrv_unref(mirror_top_bs); |
3f09bfbc | 592 | bdrv_unref(src); |
5a7e7a0b SH |
593 | } |
594 | ||
49efb1f5 DL |
595 | static void mirror_throttle(MirrorBlockJob *s) |
596 | { | |
597 | int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); | |
598 | ||
599 | if (now - s->last_pause_ns > SLICE_TIME) { | |
600 | s->last_pause_ns = now; | |
5bf1d5a7 | 601 | block_job_sleep_ns(&s->common, 0); |
49efb1f5 DL |
602 | } else { |
603 | block_job_pause_point(&s->common); | |
604 | } | |
605 | } | |
606 | ||
c0b363ad DL |
607 | static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) |
608 | { | |
23ca459a | 609 | int64_t offset; |
c0b363ad | 610 | BlockDriverState *base = s->base; |
4ef85a9c | 611 | BlockDriverState *bs = s->source; |
c0b363ad | 612 | BlockDriverState *target_bs = blk_bs(s->target); |
23ca459a | 613 | int ret; |
51b0a488 | 614 | int64_t count; |
c0b363ad | 615 | |
b7d5062c | 616 | if (base == NULL && !bdrv_has_zero_init(target_bs)) { |
c7c2769c | 617 | if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { |
e0d7f73e | 618 | bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); |
c7c2769c DL |
619 | return 0; |
620 | } | |
621 | ||
90ab48eb | 622 | s->initial_zeroing_ongoing = true; |
23ca459a EB |
623 | for (offset = 0; offset < s->bdev_length; ) { |
624 | int bytes = MIN(s->bdev_length - offset, | |
625 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c7c2769c DL |
626 | |
627 | mirror_throttle(s); | |
628 | ||
629 | if (block_job_is_cancelled(&s->common)) { | |
90ab48eb | 630 | s->initial_zeroing_ongoing = false; |
c7c2769c DL |
631 | return 0; |
632 | } | |
633 | ||
634 | if (s->in_flight >= MAX_IN_FLIGHT) { | |
67adf4b3 EB |
635 | trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, |
636 | s->in_flight); | |
c7c2769c DL |
637 | mirror_wait_for_io(s); |
638 | continue; | |
639 | } | |
640 | ||
23ca459a EB |
641 | mirror_do_zero_or_discard(s, offset, bytes, false); |
642 | offset += bytes; | |
c7c2769c DL |
643 | } |
644 | ||
bae8196d | 645 | mirror_wait_for_all_io(s); |
90ab48eb | 646 | s->initial_zeroing_ongoing = false; |
b7d5062c DL |
647 | } |
648 | ||
c0b363ad | 649 | /* First part, loop on the sectors and initialize the dirty bitmap. */ |
23ca459a | 650 | for (offset = 0; offset < s->bdev_length; ) { |
c0b363ad | 651 | /* Just to make sure we are not exceeding int limit. */ |
23ca459a EB |
652 | int bytes = MIN(s->bdev_length - offset, |
653 | QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); | |
c0b363ad DL |
654 | |
655 | mirror_throttle(s); | |
656 | ||
657 | if (block_job_is_cancelled(&s->common)) { | |
658 | return 0; | |
659 | } | |
660 | ||
23ca459a | 661 | ret = bdrv_is_allocated_above(bs, base, offset, bytes, &count); |
c0b363ad DL |
662 | if (ret < 0) { |
663 | return ret; | |
664 | } | |
665 | ||
23ca459a | 666 | assert(count); |
b7d5062c | 667 | if (ret == 1) { |
23ca459a | 668 | bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); |
c0b363ad | 669 | } |
23ca459a | 670 | offset += count; |
c0b363ad DL |
671 | } |
672 | return 0; | |
673 | } | |
674 | ||
bdffb31d PB |
675 | /* Called when going out of the streaming phase to flush the bulk of the |
676 | * data to the medium, or just before completing. | |
677 | */ | |
678 | static int mirror_flush(MirrorBlockJob *s) | |
679 | { | |
680 | int ret = blk_flush(s->target); | |
681 | if (ret < 0) { | |
682 | if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { | |
683 | s->ret = ret; | |
684 | } | |
685 | } | |
686 | return ret; | |
687 | } | |
688 | ||
893f7eba PB |
689 | static void coroutine_fn mirror_run(void *opaque) |
690 | { | |
691 | MirrorBlockJob *s = opaque; | |
5a7e7a0b | 692 | MirrorExitData *data; |
4ef85a9c | 693 | BlockDriverState *bs = s->source; |
e253f4b8 | 694 | BlockDriverState *target_bs = blk_bs(s->target); |
9a0cec66 | 695 | bool need_drain = true; |
c0b363ad | 696 | int64_t length; |
b812f671 | 697 | BlockDriverInfo bdi; |
1d33936e JC |
698 | char backing_filename[2]; /* we only need 2 characters because we are only |
699 | checking for a NULL string */ | |
893f7eba | 700 | int ret = 0; |
893f7eba PB |
701 | |
702 | if (block_job_is_cancelled(&s->common)) { | |
703 | goto immediate_exit; | |
704 | } | |
705 | ||
b21c7652 HR |
706 | s->bdev_length = bdrv_getlength(bs); |
707 | if (s->bdev_length < 0) { | |
708 | ret = s->bdev_length; | |
373df5b1 | 709 | goto immediate_exit; |
becc347e KW |
710 | } |
711 | ||
712 | /* Active commit must resize the base image if its size differs from the | |
713 | * active layer. */ | |
714 | if (s->base == blk_bs(s->target)) { | |
715 | int64_t base_length; | |
716 | ||
717 | base_length = blk_getlength(s->target); | |
718 | if (base_length < 0) { | |
719 | ret = base_length; | |
720 | goto immediate_exit; | |
721 | } | |
722 | ||
723 | if (s->bdev_length > base_length) { | |
3a691c50 HR |
724 | ret = blk_truncate(s->target, s->bdev_length, PREALLOC_MODE_OFF, |
725 | NULL); | |
becc347e KW |
726 | if (ret < 0) { |
727 | goto immediate_exit; | |
728 | } | |
729 | } | |
730 | } | |
731 | ||
732 | if (s->bdev_length == 0) { | |
9e48b025 FZ |
733 | /* Report BLOCK_JOB_READY and wait for complete. */ |
734 | block_job_event_ready(&s->common); | |
735 | s->synced = true; | |
736 | while (!block_job_is_cancelled(&s->common) && !s->should_complete) { | |
737 | block_job_yield(&s->common); | |
738 | } | |
739 | s->common.cancelled = false; | |
740 | goto immediate_exit; | |
893f7eba PB |
741 | } |
742 | ||
b21c7652 | 743 | length = DIV_ROUND_UP(s->bdev_length, s->granularity); |
402a4741 PB |
744 | s->in_flight_bitmap = bitmap_new(length); |
745 | ||
b812f671 PB |
746 | /* If we have no backing file yet in the destination, we cannot let |
747 | * the destination do COW. Instead, we copy sectors around the | |
748 | * dirty data if needed. We need a bitmap to do that. | |
749 | */ | |
e253f4b8 | 750 | bdrv_get_backing_filename(target_bs, backing_filename, |
b812f671 | 751 | sizeof(backing_filename)); |
e253f4b8 | 752 | if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { |
b436982f EB |
753 | s->target_cluster_size = bdi.cluster_size; |
754 | } else { | |
755 | s->target_cluster_size = BDRV_SECTOR_SIZE; | |
e5b43573 | 756 | } |
b436982f EB |
757 | if (backing_filename[0] && !target_bs->backing && |
758 | s->granularity < s->target_cluster_size) { | |
759 | s->buf_size = MAX(s->buf_size, s->target_cluster_size); | |
e5b43573 | 760 | s->cow_bitmap = bitmap_new(length); |
b812f671 | 761 | } |
e253f4b8 | 762 | s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); |
b812f671 | 763 | |
7504edf4 KW |
764 | s->buf = qemu_try_blockalign(bs, s->buf_size); |
765 | if (s->buf == NULL) { | |
766 | ret = -ENOMEM; | |
767 | goto immediate_exit; | |
768 | } | |
769 | ||
402a4741 | 770 | mirror_free_init(s); |
893f7eba | 771 | |
49efb1f5 | 772 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
03544a6e | 773 | if (!s->is_none_mode) { |
c0b363ad DL |
774 | ret = mirror_dirty_init(s); |
775 | if (ret < 0 || block_job_is_cancelled(&s->common)) { | |
776 | goto immediate_exit; | |
893f7eba PB |
777 | } |
778 | } | |
779 | ||
dc162c8e | 780 | assert(!s->dbi); |
715a74d8 | 781 | s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); |
893f7eba | 782 | for (;;) { |
cc8c9d6c | 783 | uint64_t delay_ns = 0; |
49efb1f5 | 784 | int64_t cnt, delta; |
893f7eba PB |
785 | bool should_complete; |
786 | ||
bd48bde8 PB |
787 | if (s->ret < 0) { |
788 | ret = s->ret; | |
789 | goto immediate_exit; | |
790 | } | |
791 | ||
565ac01f SH |
792 | block_job_pause_point(&s->common); |
793 | ||
20dca810 | 794 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
b21c7652 | 795 | /* s->common.offset contains the number of bytes already processed so |
9a46dba7 | 796 | * far, cnt is the number of dirty bytes remaining and |
b436982f | 797 | * s->bytes_in_flight is the number of bytes currently being |
b21c7652 | 798 | * processed; together those are the current total operation length */ |
9a46dba7 | 799 | s->common.len = s->common.offset + s->bytes_in_flight + cnt; |
bd48bde8 PB |
800 | |
801 | /* Note that even when no rate limit is applied we need to yield | |
a7282330 | 802 | * periodically with no pending I/O so that bdrv_drain_all() returns. |
bd48bde8 PB |
803 | * We do so every SLICE_TIME nanoseconds, or when there is an error, |
804 | * or when the source is clean, whichever comes first. | |
805 | */ | |
49efb1f5 DL |
806 | delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; |
807 | if (delta < SLICE_TIME && | |
bd48bde8 | 808 | s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { |
cf56a3c6 | 809 | if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || |
402a4741 | 810 | (cnt == 0 && s->in_flight > 0)) { |
9a46dba7 | 811 | trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); |
21cd917f | 812 | mirror_wait_for_io(s); |
bd48bde8 PB |
813 | continue; |
814 | } else if (cnt != 0) { | |
cc8c9d6c | 815 | delay_ns = mirror_iteration(s); |
893f7eba | 816 | } |
893f7eba PB |
817 | } |
818 | ||
819 | should_complete = false; | |
bd48bde8 | 820 | if (s->in_flight == 0 && cnt == 0) { |
893f7eba | 821 | trace_mirror_before_flush(s); |
bdffb31d PB |
822 | if (!s->synced) { |
823 | if (mirror_flush(s) < 0) { | |
824 | /* Go check s->ret. */ | |
825 | continue; | |
b952b558 | 826 | } |
b952b558 PB |
827 | /* We're out of the streaming phase. From now on, if the job |
828 | * is cancelled we will actually complete all pending I/O and | |
829 | * report completion. This way, block-job-cancel will leave | |
830 | * the target in a consistent state. | |
831 | */ | |
bdffb31d PB |
832 | block_job_event_ready(&s->common); |
833 | s->synced = true; | |
d63ffd87 | 834 | } |
bdffb31d PB |
835 | |
836 | should_complete = s->should_complete || | |
837 | block_job_is_cancelled(&s->common); | |
838 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); | |
893f7eba PB |
839 | } |
840 | ||
841 | if (cnt == 0 && should_complete) { | |
842 | /* The dirty bitmap is not updated while operations are pending. | |
843 | * If we're about to exit, wait for pending operations before | |
844 | * calling bdrv_get_dirty_count(bs), or we may exit while the | |
845 | * source has dirty data to copy! | |
846 | * | |
847 | * Note that I/O can be submitted by the guest while | |
9a0cec66 PB |
848 | * mirror_populate runs, so pause it now. Before deciding |
849 | * whether to switch to target check one last time if I/O has | |
850 | * come in the meanwhile, and if not flush the data to disk. | |
893f7eba | 851 | */ |
9a46dba7 | 852 | trace_mirror_before_drain(s, cnt); |
9a0cec66 PB |
853 | |
854 | bdrv_drained_begin(bs); | |
20dca810 | 855 | cnt = bdrv_get_dirty_count(s->dirty_bitmap); |
bdffb31d | 856 | if (cnt > 0 || mirror_flush(s) < 0) { |
9a0cec66 PB |
857 | bdrv_drained_end(bs); |
858 | continue; | |
859 | } | |
860 | ||
861 | /* The two disks are in sync. Exit and report successful | |
862 | * completion. | |
863 | */ | |
864 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
865 | s->common.cancelled = false; | |
866 | need_drain = false; | |
867 | break; | |
893f7eba PB |
868 | } |
869 | ||
870 | ret = 0; | |
9a46dba7 | 871 | trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); |
b76e4458 LL |
872 | if (block_job_is_cancelled(&s->common) && s->common.force) { |
873 | break; | |
893f7eba | 874 | } else if (!should_complete) { |
bd48bde8 | 875 | delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); |
5bf1d5a7 | 876 | block_job_sleep_ns(&s->common, delay_ns); |
893f7eba | 877 | } |
49efb1f5 | 878 | s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); |
893f7eba PB |
879 | } |
880 | ||
881 | immediate_exit: | |
bd48bde8 PB |
882 | if (s->in_flight > 0) { |
883 | /* We get here only if something went wrong. Either the job failed, | |
884 | * or it was cancelled prematurely so that we do not guarantee that | |
885 | * the target is a copy of the source. | |
886 | */ | |
b76e4458 LL |
887 | assert(ret < 0 || ((s->common.force || !s->synced) && |
888 | block_job_is_cancelled(&s->common))); | |
9a0cec66 | 889 | assert(need_drain); |
bae8196d | 890 | mirror_wait_for_all_io(s); |
bd48bde8 PB |
891 | } |
892 | ||
893 | assert(s->in_flight == 0); | |
7191bf31 | 894 | qemu_vfree(s->buf); |
b812f671 | 895 | g_free(s->cow_bitmap); |
402a4741 | 896 | g_free(s->in_flight_bitmap); |
dc162c8e | 897 | bdrv_dirty_iter_free(s->dbi); |
5a7e7a0b SH |
898 | |
899 | data = g_malloc(sizeof(*data)); | |
900 | data->ret = ret; | |
9a0cec66 PB |
901 | |
902 | if (need_drain) { | |
903 | bdrv_drained_begin(bs); | |
904 | } | |
5a7e7a0b | 905 | block_job_defer_to_main_loop(&s->common, mirror_exit, data); |
893f7eba PB |
906 | } |
907 | ||
908 | static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
909 | { | |
910 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
911 | ||
912 | if (speed < 0) { | |
c6bd8c70 | 913 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); |
893f7eba PB |
914 | return; |
915 | } | |
f3e4ce4a | 916 | ratelimit_set_speed(&s->limit, speed, SLICE_TIME); |
893f7eba PB |
917 | } |
918 | ||
d63ffd87 PB |
919 | static void mirror_complete(BlockJob *job, Error **errp) |
920 | { | |
921 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
4ef85a9c | 922 | BlockDriverState *target; |
274fccee | 923 | |
274fccee | 924 | target = blk_bs(s->target); |
d63ffd87 | 925 | |
d63ffd87 | 926 | if (!s->synced) { |
9df229c3 AG |
927 | error_setg(errp, "The active block job '%s' cannot be completed", |
928 | job->id); | |
d63ffd87 PB |
929 | return; |
930 | } | |
931 | ||
274fccee HR |
932 | if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { |
933 | int ret; | |
934 | ||
935 | assert(!target->backing); | |
936 | ret = bdrv_open_backing_file(target, NULL, "backing", errp); | |
937 | if (ret < 0) { | |
938 | return; | |
939 | } | |
940 | } | |
941 | ||
15d67298 | 942 | /* block all operations on to_replace bs */ |
09158f00 | 943 | if (s->replaces) { |
5a7e7a0b SH |
944 | AioContext *replace_aio_context; |
945 | ||
e12f3784 | 946 | s->to_replace = bdrv_find_node(s->replaces); |
09158f00 | 947 | if (!s->to_replace) { |
e12f3784 | 948 | error_setg(errp, "Node name '%s' not found", s->replaces); |
09158f00 BC |
949 | return; |
950 | } | |
951 | ||
5a7e7a0b SH |
952 | replace_aio_context = bdrv_get_aio_context(s->to_replace); |
953 | aio_context_acquire(replace_aio_context); | |
954 | ||
4ef85a9c KW |
955 | /* TODO Translate this into permission system. Current definition of |
956 | * GRAPH_MOD would require to request it for the parents; they might | |
957 | * not even be BlockDriverStates, however, so a BdrvChild can't address | |
958 | * them. May need redefinition of GRAPH_MOD. */ | |
09158f00 BC |
959 | error_setg(&s->replace_blocker, |
960 | "block device is in use by block-job-complete"); | |
961 | bdrv_op_block_all(s->to_replace, s->replace_blocker); | |
962 | bdrv_ref(s->to_replace); | |
5a7e7a0b SH |
963 | |
964 | aio_context_release(replace_aio_context); | |
09158f00 BC |
965 | } |
966 | ||
d63ffd87 | 967 | s->should_complete = true; |
751ebd76 | 968 | block_job_enter(&s->common); |
d63ffd87 PB |
969 | } |
970 | ||
bae8196d | 971 | static void mirror_pause(BlockJob *job) |
565ac01f SH |
972 | { |
973 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
974 | ||
bae8196d | 975 | mirror_wait_for_all_io(s); |
565ac01f SH |
976 | } |
977 | ||
978 | static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) | |
979 | { | |
980 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
981 | ||
982 | blk_set_aio_context(s->target, new_context); | |
983 | } | |
984 | ||
bae8196d PB |
985 | static void mirror_drain(BlockJob *job) |
986 | { | |
987 | MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); | |
988 | ||
989 | /* Need to keep a reference in case blk_drain triggers execution | |
990 | * of mirror_complete... | |
991 | */ | |
992 | if (s->target) { | |
993 | BlockBackend *target = s->target; | |
994 | blk_ref(target); | |
995 | blk_drain(target); | |
996 | blk_unref(target); | |
997 | } | |
998 | } | |
999 | ||
3fc4b10a | 1000 | static const BlockJobDriver mirror_job_driver = { |
565ac01f SH |
1001 | .instance_size = sizeof(MirrorBlockJob), |
1002 | .job_type = BLOCK_JOB_TYPE_MIRROR, | |
1003 | .set_speed = mirror_set_speed, | |
a7815a76 | 1004 | .start = mirror_run, |
565ac01f SH |
1005 | .complete = mirror_complete, |
1006 | .pause = mirror_pause, | |
1007 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1008 | .drain = mirror_drain, |
893f7eba PB |
1009 | }; |
1010 | ||
03544a6e | 1011 | static const BlockJobDriver commit_active_job_driver = { |
565ac01f SH |
1012 | .instance_size = sizeof(MirrorBlockJob), |
1013 | .job_type = BLOCK_JOB_TYPE_COMMIT, | |
1014 | .set_speed = mirror_set_speed, | |
a7815a76 | 1015 | .start = mirror_run, |
565ac01f SH |
1016 | .complete = mirror_complete, |
1017 | .pause = mirror_pause, | |
1018 | .attached_aio_context = mirror_attached_aio_context, | |
bae8196d | 1019 | .drain = mirror_drain, |
03544a6e FZ |
1020 | }; |
1021 | ||
4ef85a9c KW |
1022 | static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, |
1023 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1024 | { | |
1025 | return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); | |
1026 | } | |
1027 | ||
1028 | static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, | |
1029 | uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) | |
1030 | { | |
1031 | return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); | |
1032 | } | |
1033 | ||
1034 | static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) | |
1035 | { | |
ce960aa9 VSO |
1036 | if (bs->backing == NULL) { |
1037 | /* we can be here after failed bdrv_append in mirror_start_job */ | |
1038 | return 0; | |
1039 | } | |
4ef85a9c KW |
1040 | return bdrv_co_flush(bs->backing->bs); |
1041 | } | |
1042 | ||
4ef85a9c | 1043 | static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 1044 | int64_t offset, int bytes, BdrvRequestFlags flags) |
4ef85a9c | 1045 | { |
f5a5ca79 | 1046 | return bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); |
4ef85a9c KW |
1047 | } |
1048 | ||
1049 | static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, | |
f5a5ca79 | 1050 | int64_t offset, int bytes) |
4ef85a9c | 1051 | { |
f5a5ca79 | 1052 | return bdrv_co_pdiscard(bs->backing->bs, offset, bytes); |
4ef85a9c KW |
1053 | } |
1054 | ||
fd4a6493 KW |
1055 | static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) |
1056 | { | |
18775ff3 VSO |
1057 | if (bs->backing == NULL) { |
1058 | /* we can be here after failed bdrv_attach_child in | |
1059 | * bdrv_set_backing_hd */ | |
1060 | return; | |
1061 | } | |
fd4a6493 KW |
1062 | bdrv_refresh_filename(bs->backing->bs); |
1063 | pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), | |
1064 | bs->backing->bs->filename); | |
1065 | } | |
1066 | ||
4ef85a9c KW |
1067 | static void bdrv_mirror_top_close(BlockDriverState *bs) |
1068 | { | |
1069 | } | |
1070 | ||
1071 | static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, | |
1072 | const BdrvChildRole *role, | |
e0995dc3 | 1073 | BlockReopenQueue *reopen_queue, |
4ef85a9c KW |
1074 | uint64_t perm, uint64_t shared, |
1075 | uint64_t *nperm, uint64_t *nshared) | |
1076 | { | |
1077 | /* Must be able to forward guest writes to the real image */ | |
1078 | *nperm = 0; | |
1079 | if (perm & BLK_PERM_WRITE) { | |
1080 | *nperm |= BLK_PERM_WRITE; | |
1081 | } | |
1082 | ||
1083 | *nshared = BLK_PERM_ALL; | |
1084 | } | |
1085 | ||
1086 | /* Dummy node that provides consistent read to its users without requiring it | |
1087 | * from its backing file and that allows writes on the backing file chain. */ | |
1088 | static BlockDriver bdrv_mirror_top = { | |
1089 | .format_name = "mirror_top", | |
1090 | .bdrv_co_preadv = bdrv_mirror_top_preadv, | |
1091 | .bdrv_co_pwritev = bdrv_mirror_top_pwritev, | |
1092 | .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, | |
1093 | .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, | |
1094 | .bdrv_co_flush = bdrv_mirror_top_flush, | |
3e4d0e72 | 1095 | .bdrv_co_block_status = bdrv_co_block_status_from_backing, |
fd4a6493 | 1096 | .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, |
4ef85a9c KW |
1097 | .bdrv_close = bdrv_mirror_top_close, |
1098 | .bdrv_child_perm = bdrv_mirror_top_child_perm, | |
1099 | }; | |
1100 | ||
71aa9867 | 1101 | static void mirror_start_job(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1102 | int creation_flags, BlockDriverState *target, |
1103 | const char *replaces, int64_t speed, | |
1104 | uint32_t granularity, int64_t buf_size, | |
274fccee | 1105 | BlockMirrorBackingMode backing_mode, |
09158f00 BC |
1106 | BlockdevOnError on_source_error, |
1107 | BlockdevOnError on_target_error, | |
0fc9f8ea | 1108 | bool unmap, |
097310b5 | 1109 | BlockCompletionFunc *cb, |
51ccfa2d | 1110 | void *opaque, |
09158f00 | 1111 | const BlockJobDriver *driver, |
b49f7ead | 1112 | bool is_none_mode, BlockDriverState *base, |
51ccfa2d | 1113 | bool auto_complete, const char *filter_node_name, |
045a2f82 | 1114 | bool is_mirror, |
51ccfa2d | 1115 | Error **errp) |
893f7eba PB |
1116 | { |
1117 | MirrorBlockJob *s; | |
4ef85a9c KW |
1118 | BlockDriverState *mirror_top_bs; |
1119 | bool target_graph_mod; | |
1120 | bool target_is_backing; | |
b2c2832c | 1121 | Error *local_err = NULL; |
d7086422 | 1122 | int ret; |
893f7eba | 1123 | |
eee13dfe | 1124 | if (granularity == 0) { |
341ebc2f | 1125 | granularity = bdrv_get_default_bitmap_granularity(target); |
eee13dfe PB |
1126 | } |
1127 | ||
31826642 | 1128 | assert(is_power_of_2(granularity)); |
eee13dfe | 1129 | |
48ac0a4d WC |
1130 | if (buf_size < 0) { |
1131 | error_setg(errp, "Invalid parameter 'buf-size'"); | |
1132 | return; | |
1133 | } | |
1134 | ||
1135 | if (buf_size == 0) { | |
1136 | buf_size = DEFAULT_MIRROR_BUF_SIZE; | |
1137 | } | |
5bc361b8 | 1138 | |
4ef85a9c KW |
1139 | /* In the case of active commit, add dummy driver to provide consistent |
1140 | * reads on the top, while disabling it in the intermediate nodes, and make | |
1141 | * the backing chain writable. */ | |
6cdbceb1 KW |
1142 | mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, |
1143 | BDRV_O_RDWR, errp); | |
4ef85a9c KW |
1144 | if (mirror_top_bs == NULL) { |
1145 | return; | |
1146 | } | |
d3c8c674 KW |
1147 | if (!filter_node_name) { |
1148 | mirror_top_bs->implicit = true; | |
1149 | } | |
4ef85a9c | 1150 | mirror_top_bs->total_sectors = bs->total_sectors; |
19dd29e8 | 1151 | bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); |
4ef85a9c KW |
1152 | |
1153 | /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep | |
7a25fcd0 | 1154 | * it alive until block_job_create() succeeds even if bs has no parent. */ |
4ef85a9c KW |
1155 | bdrv_ref(mirror_top_bs); |
1156 | bdrv_drained_begin(bs); | |
b2c2832c | 1157 | bdrv_append(mirror_top_bs, bs, &local_err); |
4ef85a9c KW |
1158 | bdrv_drained_end(bs); |
1159 | ||
b2c2832c KW |
1160 | if (local_err) { |
1161 | bdrv_unref(mirror_top_bs); | |
1162 | error_propagate(errp, local_err); | |
1163 | return; | |
1164 | } | |
1165 | ||
4ef85a9c | 1166 | /* Make sure that the source is not resized while the job is running */ |
75859b94 | 1167 | s = block_job_create(job_id, driver, NULL, mirror_top_bs, |
4ef85a9c KW |
1168 | BLK_PERM_CONSISTENT_READ, |
1169 | BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | | |
1170 | BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, | |
c6cc12bf | 1171 | creation_flags, cb, opaque, errp); |
893f7eba | 1172 | if (!s) { |
4ef85a9c | 1173 | goto fail; |
893f7eba | 1174 | } |
7a25fcd0 HR |
1175 | /* The block job now has a reference to this node */ |
1176 | bdrv_unref(mirror_top_bs); | |
1177 | ||
4ef85a9c KW |
1178 | s->source = bs; |
1179 | s->mirror_top_bs = mirror_top_bs; | |
1180 | ||
1181 | /* No resize for the target either; while the mirror is still running, a | |
1182 | * consistent read isn't necessarily possible. We could possibly allow | |
1183 | * writes and graph modifications, though it would likely defeat the | |
1184 | * purpose of a mirror, so leave them blocked for now. | |
1185 | * | |
1186 | * In the case of active commit, things look a bit different, though, | |
1187 | * because the target is an already populated backing file in active use. | |
1188 | * We can allow anything except resize there.*/ | |
1189 | target_is_backing = bdrv_chain_contains(bs, target); | |
1190 | target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); | |
1191 | s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | | |
1192 | (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), | |
1193 | BLK_PERM_WRITE_UNCHANGED | | |
1194 | (target_is_backing ? BLK_PERM_CONSISTENT_READ | | |
1195 | BLK_PERM_WRITE | | |
1196 | BLK_PERM_GRAPH_MOD : 0)); | |
d7086422 KW |
1197 | ret = blk_insert_bs(s->target, target, errp); |
1198 | if (ret < 0) { | |
4ef85a9c | 1199 | goto fail; |
d7086422 | 1200 | } |
045a2f82 FZ |
1201 | if (is_mirror) { |
1202 | /* XXX: Mirror target could be a NBD server of target QEMU in the case | |
1203 | * of non-shared block migration. To allow migration completion, we | |
1204 | * have to allow "inactivate" of the target BB. When that happens, we | |
1205 | * know the job is drained, and the vcpus are stopped, so no write | |
1206 | * operation will be performed. Block layer already has assertions to | |
1207 | * ensure that. */ | |
1208 | blk_set_force_allow_inactivate(s->target); | |
1209 | } | |
e253f4b8 | 1210 | |
09158f00 | 1211 | s->replaces = g_strdup(replaces); |
b952b558 PB |
1212 | s->on_source_error = on_source_error; |
1213 | s->on_target_error = on_target_error; | |
03544a6e | 1214 | s->is_none_mode = is_none_mode; |
274fccee | 1215 | s->backing_mode = backing_mode; |
5bc361b8 | 1216 | s->base = base; |
eee13dfe | 1217 | s->granularity = granularity; |
48ac0a4d | 1218 | s->buf_size = ROUND_UP(buf_size, granularity); |
0fc9f8ea | 1219 | s->unmap = unmap; |
b49f7ead WC |
1220 | if (auto_complete) { |
1221 | s->should_complete = true; | |
1222 | } | |
b812f671 | 1223 | |
0db6e54a | 1224 | s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); |
b8afb520 | 1225 | if (!s->dirty_bitmap) { |
88f9d1b3 | 1226 | goto fail; |
b8afb520 | 1227 | } |
10f3cd15 | 1228 | |
4ef85a9c | 1229 | /* Required permissions are already taken with blk_new() */ |
76d554e2 KW |
1230 | block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, |
1231 | &error_abort); | |
1232 | ||
f3ede4b0 AG |
1233 | /* In commit_active_start() all intermediate nodes disappear, so |
1234 | * any jobs in them must be blocked */ | |
4ef85a9c | 1235 | if (target_is_backing) { |
f3ede4b0 AG |
1236 | BlockDriverState *iter; |
1237 | for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { | |
4ef85a9c KW |
1238 | /* XXX BLK_PERM_WRITE needs to be allowed so we don't block |
1239 | * ourselves at s->base (if writes are blocked for a node, they are | |
1240 | * also blocked for its backing file). The other options would be a | |
1241 | * second filter driver above s->base (== target). */ | |
1242 | ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, | |
1243 | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, | |
1244 | errp); | |
1245 | if (ret < 0) { | |
1246 | goto fail; | |
1247 | } | |
f3ede4b0 AG |
1248 | } |
1249 | } | |
10f3cd15 | 1250 | |
5ccac6f1 JS |
1251 | trace_mirror_start(bs, s, opaque); |
1252 | block_job_start(&s->common); | |
4ef85a9c KW |
1253 | return; |
1254 | ||
1255 | fail: | |
1256 | if (s) { | |
7a25fcd0 HR |
1257 | /* Make sure this BDS does not go away until we have completed the graph |
1258 | * changes below */ | |
1259 | bdrv_ref(mirror_top_bs); | |
1260 | ||
4ef85a9c KW |
1261 | g_free(s->replaces); |
1262 | blk_unref(s->target); | |
05b0d8e3 | 1263 | block_job_early_fail(&s->common); |
4ef85a9c KW |
1264 | } |
1265 | ||
c1cef672 FZ |
1266 | bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, |
1267 | &error_abort); | |
5fe31c25 | 1268 | bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); |
7a25fcd0 HR |
1269 | |
1270 | bdrv_unref(mirror_top_bs); | |
893f7eba | 1271 | } |
03544a6e | 1272 | |
71aa9867 AG |
1273 | void mirror_start(const char *job_id, BlockDriverState *bs, |
1274 | BlockDriverState *target, const char *replaces, | |
5fba6c0e | 1275 | int64_t speed, uint32_t granularity, int64_t buf_size, |
274fccee HR |
1276 | MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, |
1277 | BlockdevOnError on_source_error, | |
03544a6e | 1278 | BlockdevOnError on_target_error, |
6cdbceb1 | 1279 | bool unmap, const char *filter_node_name, Error **errp) |
03544a6e FZ |
1280 | { |
1281 | bool is_none_mode; | |
1282 | BlockDriverState *base; | |
1283 | ||
4b80ab2b JS |
1284 | if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { |
1285 | error_setg(errp, "Sync mode 'incremental' not supported"); | |
d58d8453 JS |
1286 | return; |
1287 | } | |
03544a6e | 1288 | is_none_mode = mode == MIRROR_SYNC_MODE_NONE; |
760e0063 | 1289 | base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; |
47970dfb | 1290 | mirror_start_job(job_id, bs, BLOCK_JOB_DEFAULT, target, replaces, |
274fccee | 1291 | speed, granularity, buf_size, backing_mode, |
51ccfa2d | 1292 | on_source_error, on_target_error, unmap, NULL, NULL, |
6cdbceb1 | 1293 | &mirror_job_driver, is_none_mode, base, false, |
045a2f82 | 1294 | filter_node_name, true, errp); |
03544a6e FZ |
1295 | } |
1296 | ||
fd62c609 | 1297 | void commit_active_start(const char *job_id, BlockDriverState *bs, |
47970dfb JS |
1298 | BlockDriverState *base, int creation_flags, |
1299 | int64_t speed, BlockdevOnError on_error, | |
0db832f4 | 1300 | const char *filter_node_name, |
78bbd910 FZ |
1301 | BlockCompletionFunc *cb, void *opaque, |
1302 | bool auto_complete, Error **errp) | |
03544a6e | 1303 | { |
4da83585 | 1304 | int orig_base_flags; |
cc67f4d1 | 1305 | Error *local_err = NULL; |
4da83585 JC |
1306 | |
1307 | orig_base_flags = bdrv_get_flags(base); | |
1308 | ||
20a63d2c FZ |
1309 | if (bdrv_reopen(base, bs->open_flags, errp)) { |
1310 | return; | |
1311 | } | |
4da83585 | 1312 | |
47970dfb | 1313 | mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, |
71aa9867 | 1314 | MIRROR_LEAVE_BACKING_CHAIN, |
51ccfa2d | 1315 | on_error, on_error, true, cb, opaque, |
6cdbceb1 | 1316 | &commit_active_job_driver, false, base, auto_complete, |
045a2f82 | 1317 | filter_node_name, false, &local_err); |
0fb6395c | 1318 | if (local_err) { |
cc67f4d1 | 1319 | error_propagate(errp, local_err); |
4da83585 JC |
1320 | goto error_restore_flags; |
1321 | } | |
1322 | ||
1323 | return; | |
1324 | ||
1325 | error_restore_flags: | |
1326 | /* ignore error and errp for bdrv_reopen, because we want to propagate | |
1327 | * the original error */ | |
1328 | bdrv_reopen(base, orig_base_flags, NULL); | |
1329 | return; | |
03544a6e | 1330 | } |