]>
Commit | Line | Data |
---|---|---|
61007b31 SH |
1 | /* |
2 | * Block layer I/O functions | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
24 | ||
80c71a24 | 25 | #include "qemu/osdep.h" |
61007b31 | 26 | #include "trace.h" |
7f0e9da6 | 27 | #include "sysemu/block-backend.h" |
7719f3c9 | 28 | #include "block/aio-wait.h" |
61007b31 | 29 | #include "block/blockjob.h" |
f321dcb5 | 30 | #include "block/blockjob_int.h" |
61007b31 | 31 | #include "block/block_int.h" |
21c2283e | 32 | #include "block/coroutines.h" |
e2c1c34f | 33 | #include "block/dirty-bitmap.h" |
94783301 | 34 | #include "block/write-threshold.h" |
f348b6d1 | 35 | #include "qemu/cutils.h" |
5df022cf | 36 | #include "qemu/memalign.h" |
da34e65c | 37 | #include "qapi/error.h" |
d49b6836 | 38 | #include "qemu/error-report.h" |
db725815 | 39 | #include "qemu/main-loop.h" |
c8aa7895 | 40 | #include "sysemu/replay.h" |
61007b31 | 41 | |
cb2e2878 EB |
42 | /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */ |
43 | #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS) | |
44 | ||
7f8f03ef | 45 | static void bdrv_parent_cb_resize(BlockDriverState *bs); |
d05aa8bb | 46 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
5ae07b14 | 47 | int64_t offset, int64_t bytes, BdrvRequestFlags flags); |
61007b31 | 48 | |
a82a3bd1 | 49 | static void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore) |
61007b31 | 50 | { |
02d21300 | 51 | BdrvChild *c, *next; |
27ccdd52 | 52 | |
02d21300 | 53 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { |
a82a3bd1 | 54 | if (c == ignore) { |
0152bf40 KW |
55 | continue; |
56 | } | |
606ed756 | 57 | bdrv_parent_drained_begin_single(c); |
ce0f1412 PB |
58 | } |
59 | } | |
61007b31 | 60 | |
2f65df6e | 61 | void bdrv_parent_drained_end_single(BdrvChild *c) |
804db8ea | 62 | { |
ab613350 | 63 | GLOBAL_STATE_CODE(); |
2f65df6e | 64 | |
57e05be3 KW |
65 | assert(c->quiesced_parent); |
66 | c->quiesced_parent = false; | |
67 | ||
bd86fb99 | 68 | if (c->klass->drained_end) { |
2f65df6e | 69 | c->klass->drained_end(c); |
804db8ea HR |
70 | } |
71 | } | |
72 | ||
a82a3bd1 | 73 | static void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore) |
ce0f1412 | 74 | { |
61ad631c | 75 | BdrvChild *c; |
27ccdd52 | 76 | |
61ad631c | 77 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
a82a3bd1 | 78 | if (c == ignore) { |
0152bf40 KW |
79 | continue; |
80 | } | |
2f65df6e | 81 | bdrv_parent_drained_end_single(c); |
27ccdd52 | 82 | } |
61007b31 SH |
83 | } |
84 | ||
23987471 | 85 | bool bdrv_parent_drained_poll_single(BdrvChild *c) |
4be6a6d1 | 86 | { |
bd86fb99 HR |
87 | if (c->klass->drained_poll) { |
88 | return c->klass->drained_poll(c); | |
4be6a6d1 KW |
89 | } |
90 | return false; | |
91 | } | |
92 | ||
6cd5c9d7 KW |
93 | static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore, |
94 | bool ignore_bds_parents) | |
89bd0305 KW |
95 | { |
96 | BdrvChild *c, *next; | |
97 | bool busy = false; | |
98 | ||
99 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { | |
bd86fb99 | 100 | if (c == ignore || (ignore_bds_parents && c->klass->parent_is_bds)) { |
89bd0305 KW |
101 | continue; |
102 | } | |
4be6a6d1 | 103 | busy |= bdrv_parent_drained_poll_single(c); |
89bd0305 KW |
104 | } |
105 | ||
106 | return busy; | |
107 | } | |
108 | ||
606ed756 | 109 | void bdrv_parent_drained_begin_single(BdrvChild *c) |
4be6a6d1 | 110 | { |
ab613350 | 111 | GLOBAL_STATE_CODE(); |
57e05be3 KW |
112 | |
113 | assert(!c->quiesced_parent); | |
114 | c->quiesced_parent = true; | |
115 | ||
bd86fb99 HR |
116 | if (c->klass->drained_begin) { |
117 | c->klass->drained_begin(c); | |
4be6a6d1 | 118 | } |
4be6a6d1 KW |
119 | } |
120 | ||
d9e0dfa2 EB |
121 | static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) |
122 | { | |
9f460c64 AO |
123 | dst->pdiscard_alignment = MAX(dst->pdiscard_alignment, |
124 | src->pdiscard_alignment); | |
d9e0dfa2 EB |
125 | dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); |
126 | dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer); | |
24b36e98 PB |
127 | dst->max_hw_transfer = MIN_NON_ZERO(dst->max_hw_transfer, |
128 | src->max_hw_transfer); | |
d9e0dfa2 EB |
129 | dst->opt_mem_alignment = MAX(dst->opt_mem_alignment, |
130 | src->opt_mem_alignment); | |
131 | dst->min_mem_alignment = MAX(dst->min_mem_alignment, | |
132 | src->min_mem_alignment); | |
133 | dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov); | |
cc071629 | 134 | dst->max_hw_iov = MIN_NON_ZERO(dst->max_hw_iov, src->max_hw_iov); |
d9e0dfa2 EB |
135 | } |
136 | ||
1e4c797c VSO |
137 | typedef struct BdrvRefreshLimitsState { |
138 | BlockDriverState *bs; | |
139 | BlockLimits old_bl; | |
140 | } BdrvRefreshLimitsState; | |
141 | ||
142 | static void bdrv_refresh_limits_abort(void *opaque) | |
143 | { | |
144 | BdrvRefreshLimitsState *s = opaque; | |
145 | ||
146 | s->bs->bl = s->old_bl; | |
147 | } | |
148 | ||
149 | static TransactionActionDrv bdrv_refresh_limits_drv = { | |
150 | .abort = bdrv_refresh_limits_abort, | |
151 | .clean = g_free, | |
152 | }; | |
153 | ||
154 | /* @tran is allowed to be NULL, in this case no rollback is possible. */ | |
155 | void bdrv_refresh_limits(BlockDriverState *bs, Transaction *tran, Error **errp) | |
61007b31 | 156 | { |
33985614 | 157 | ERRP_GUARD(); |
61007b31 | 158 | BlockDriver *drv = bs->drv; |
66b129ac HR |
159 | BdrvChild *c; |
160 | bool have_limits; | |
61007b31 | 161 | |
f791bf7f EGE |
162 | GLOBAL_STATE_CODE(); |
163 | ||
1e4c797c VSO |
164 | if (tran) { |
165 | BdrvRefreshLimitsState *s = g_new(BdrvRefreshLimitsState, 1); | |
166 | *s = (BdrvRefreshLimitsState) { | |
167 | .bs = bs, | |
168 | .old_bl = bs->bl, | |
169 | }; | |
170 | tran_add(tran, &bdrv_refresh_limits_drv, s); | |
171 | } | |
172 | ||
61007b31 SH |
173 | memset(&bs->bl, 0, sizeof(bs->bl)); |
174 | ||
175 | if (!drv) { | |
176 | return; | |
177 | } | |
178 | ||
79ba8c98 | 179 | /* Default alignment based on whether driver has byte interface */ |
e31f6864 | 180 | bs->bl.request_alignment = (drv->bdrv_co_preadv || |
ac850bf0 VSO |
181 | drv->bdrv_aio_preadv || |
182 | drv->bdrv_co_preadv_part) ? 1 : 512; | |
79ba8c98 | 183 | |
61007b31 | 184 | /* Take some limits from the children as a default */ |
66b129ac HR |
185 | have_limits = false; |
186 | QLIST_FOREACH(c, &bs->children, next) { | |
187 | if (c->role & (BDRV_CHILD_DATA | BDRV_CHILD_FILTERED | BDRV_CHILD_COW)) | |
188 | { | |
66b129ac HR |
189 | bdrv_merge_limits(&bs->bl, &c->bs->bl); |
190 | have_limits = true; | |
61007b31 | 191 | } |
160a29e2 PB |
192 | |
193 | if (c->role & BDRV_CHILD_FILTERED) { | |
194 | bs->bl.has_variable_length |= c->bs->bl.has_variable_length; | |
195 | } | |
66b129ac HR |
196 | } |
197 | ||
198 | if (!have_limits) { | |
4196d2f0 | 199 | bs->bl.min_mem_alignment = 512; |
8e3b0cbb | 200 | bs->bl.opt_mem_alignment = qemu_real_host_page_size(); |
bd44feb7 SH |
201 | |
202 | /* Safe default since most protocols use readv()/writev()/etc */ | |
203 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
204 | } |
205 | ||
61007b31 SH |
206 | /* Then let the driver override it */ |
207 | if (drv->bdrv_refresh_limits) { | |
208 | drv->bdrv_refresh_limits(bs, errp); | |
8b117001 VSO |
209 | if (*errp) { |
210 | return; | |
211 | } | |
212 | } | |
213 | ||
214 | if (bs->bl.request_alignment > BDRV_MAX_ALIGNMENT) { | |
215 | error_setg(errp, "Driver requires too large request alignment"); | |
61007b31 SH |
216 | } |
217 | } | |
218 | ||
219 | /** | |
220 | * The copy-on-read flag is actually a reference count so multiple users may | |
221 | * use the feature without worrying about clobbering its previous state. | |
222 | * Copy-on-read stays enabled until all users have called to disable it. | |
223 | */ | |
224 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
225 | { | |
384a48fb | 226 | IO_CODE(); |
d73415a3 | 227 | qatomic_inc(&bs->copy_on_read); |
61007b31 SH |
228 | } |
229 | ||
230 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
231 | { | |
d73415a3 | 232 | int old = qatomic_fetch_dec(&bs->copy_on_read); |
384a48fb | 233 | IO_CODE(); |
d3faa13e | 234 | assert(old >= 1); |
61007b31 SH |
235 | } |
236 | ||
61124f03 PB |
237 | typedef struct { |
238 | Coroutine *co; | |
239 | BlockDriverState *bs; | |
240 | bool done; | |
481cad48 | 241 | bool begin; |
fe4f0614 | 242 | bool poll; |
0152bf40 | 243 | BdrvChild *parent; |
61124f03 PB |
244 | } BdrvCoDrainData; |
245 | ||
1cc8e54a | 246 | /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ |
299403ae KW |
247 | bool bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent, |
248 | bool ignore_bds_parents) | |
89bd0305 | 249 | { |
ab613350 | 250 | GLOBAL_STATE_CODE(); |
fe4f0614 | 251 | |
6cd5c9d7 | 252 | if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) { |
89bd0305 KW |
253 | return true; |
254 | } | |
255 | ||
d73415a3 | 256 | if (qatomic_read(&bs->in_flight)) { |
fe4f0614 KW |
257 | return true; |
258 | } | |
259 | ||
fe4f0614 | 260 | return false; |
89bd0305 KW |
261 | } |
262 | ||
299403ae | 263 | static bool bdrv_drain_poll_top_level(BlockDriverState *bs, |
89bd0305 | 264 | BdrvChild *ignore_parent) |
1cc8e54a | 265 | { |
299403ae | 266 | return bdrv_drain_poll(bs, ignore_parent, false); |
1cc8e54a KW |
267 | } |
268 | ||
299403ae | 269 | static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent, |
a82a3bd1 KW |
270 | bool poll); |
271 | static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent); | |
0152bf40 | 272 | |
a77fd4bb FZ |
273 | static void bdrv_co_drain_bh_cb(void *opaque) |
274 | { | |
275 | BdrvCoDrainData *data = opaque; | |
276 | Coroutine *co = data->co; | |
99723548 | 277 | BlockDriverState *bs = data->bs; |
a77fd4bb | 278 | |
c8ca33d0 | 279 | if (bs) { |
aa1361d5 | 280 | AioContext *ctx = bdrv_get_aio_context(bs); |
960d5fb3 | 281 | aio_context_acquire(ctx); |
c8ca33d0 KW |
282 | bdrv_dec_in_flight(bs); |
283 | if (data->begin) { | |
a82a3bd1 | 284 | bdrv_do_drained_begin(bs, data->parent, data->poll); |
c8ca33d0 | 285 | } else { |
e037c09c | 286 | assert(!data->poll); |
a82a3bd1 | 287 | bdrv_do_drained_end(bs, data->parent); |
c8ca33d0 | 288 | } |
960d5fb3 | 289 | aio_context_release(ctx); |
481cad48 | 290 | } else { |
c8ca33d0 KW |
291 | assert(data->begin); |
292 | bdrv_drain_all_begin(); | |
481cad48 MP |
293 | } |
294 | ||
a77fd4bb | 295 | data->done = true; |
1919631e | 296 | aio_co_wake(co); |
a77fd4bb FZ |
297 | } |
298 | ||
481cad48 | 299 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, |
299403ae | 300 | bool begin, |
6cd5c9d7 | 301 | BdrvChild *parent, |
2f65df6e | 302 | bool poll) |
a77fd4bb FZ |
303 | { |
304 | BdrvCoDrainData data; | |
960d5fb3 KW |
305 | Coroutine *self = qemu_coroutine_self(); |
306 | AioContext *ctx = bdrv_get_aio_context(bs); | |
307 | AioContext *co_ctx = qemu_coroutine_get_aio_context(self); | |
a77fd4bb FZ |
308 | |
309 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
c40a2545 | 310 | * other coroutines run if they were queued by aio_co_enter(). */ |
a77fd4bb FZ |
311 | |
312 | assert(qemu_in_coroutine()); | |
313 | data = (BdrvCoDrainData) { | |
960d5fb3 | 314 | .co = self, |
a77fd4bb FZ |
315 | .bs = bs, |
316 | .done = false, | |
481cad48 | 317 | .begin = begin, |
0152bf40 | 318 | .parent = parent, |
fe4f0614 | 319 | .poll = poll, |
a77fd4bb | 320 | }; |
8e1da77e | 321 | |
c8ca33d0 KW |
322 | if (bs) { |
323 | bdrv_inc_in_flight(bs); | |
324 | } | |
960d5fb3 KW |
325 | |
326 | /* | |
327 | * Temporarily drop the lock across yield or we would get deadlocks. | |
328 | * bdrv_co_drain_bh_cb() reaquires the lock as needed. | |
329 | * | |
330 | * When we yield below, the lock for the current context will be | |
331 | * released, so if this is actually the lock that protects bs, don't drop | |
332 | * it a second time. | |
333 | */ | |
334 | if (ctx != co_ctx) { | |
335 | aio_context_release(ctx); | |
336 | } | |
ab613350 SH |
337 | replay_bh_schedule_oneshot_event(qemu_get_aio_context(), |
338 | bdrv_co_drain_bh_cb, &data); | |
a77fd4bb FZ |
339 | |
340 | qemu_coroutine_yield(); | |
341 | /* If we are resumed from some other event (such as an aio completion or a | |
342 | * timer callback), it is a bug in the caller that should be fixed. */ | |
343 | assert(data.done); | |
960d5fb3 | 344 | |
3202d8e4 | 345 | /* Reacquire the AioContext of bs if we dropped it */ |
960d5fb3 KW |
346 | if (ctx != co_ctx) { |
347 | aio_context_acquire(ctx); | |
348 | } | |
a77fd4bb FZ |
349 | } |
350 | ||
05c272ff KW |
351 | static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent, |
352 | bool poll) | |
6820643f | 353 | { |
384a48fb | 354 | IO_OR_GS_CODE(); |
05c272ff KW |
355 | |
356 | if (qemu_in_coroutine()) { | |
357 | bdrv_co_yield_to_drain(bs, true, parent, poll); | |
358 | return; | |
359 | } | |
d42cf288 | 360 | |
ab613350 SH |
361 | GLOBAL_STATE_CODE(); |
362 | ||
60369b86 | 363 | /* Stop things in parent-to-child order */ |
d73415a3 | 364 | if (qatomic_fetch_inc(&bs->quiesce_counter) == 0) { |
a82a3bd1 | 365 | bdrv_parent_drained_begin(bs, parent); |
57e05be3 KW |
366 | if (bs->drv && bs->drv->bdrv_drain_begin) { |
367 | bs->drv->bdrv_drain_begin(bs); | |
368 | } | |
c7bc05f7 | 369 | } |
d30b8e64 | 370 | |
fe4f0614 KW |
371 | /* |
372 | * Wait for drained requests to finish. | |
373 | * | |
374 | * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The | |
375 | * call is needed so things in this AioContext can make progress even | |
376 | * though we don't return to the main AioContext loop - this automatically | |
377 | * includes other nodes in the same AioContext and therefore all child | |
378 | * nodes. | |
379 | */ | |
380 | if (poll) { | |
299403ae | 381 | BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, parent)); |
fe4f0614 | 382 | } |
6820643f KW |
383 | } |
384 | ||
05c272ff KW |
385 | void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, BdrvChild *parent) |
386 | { | |
387 | bdrv_do_drained_begin(bs, parent, false); | |
388 | } | |
389 | ||
e2dbca03 PB |
390 | void coroutine_mixed_fn |
391 | bdrv_drained_begin(BlockDriverState *bs) | |
0152bf40 | 392 | { |
384a48fb | 393 | IO_OR_GS_CODE(); |
a82a3bd1 | 394 | bdrv_do_drained_begin(bs, NULL, true); |
0152bf40 KW |
395 | } |
396 | ||
e037c09c HR |
397 | /** |
398 | * This function does not poll, nor must any of its recursively called | |
2f65df6e | 399 | * functions. |
e037c09c | 400 | */ |
a82a3bd1 | 401 | static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent) |
6820643f | 402 | { |
0f115168 KW |
403 | int old_quiesce_counter; |
404 | ||
ab613350 SH |
405 | IO_OR_GS_CODE(); |
406 | ||
481cad48 | 407 | if (qemu_in_coroutine()) { |
a82a3bd1 | 408 | bdrv_co_yield_to_drain(bs, false, parent, false); |
481cad48 MP |
409 | return; |
410 | } | |
6820643f | 411 | assert(bs->quiesce_counter > 0); |
ab613350 | 412 | GLOBAL_STATE_CODE(); |
6820643f | 413 | |
60369b86 | 414 | /* Re-enable things in child-to-parent order */ |
d73415a3 | 415 | old_quiesce_counter = qatomic_fetch_dec(&bs->quiesce_counter); |
0f115168 | 416 | if (old_quiesce_counter == 1) { |
57e05be3 KW |
417 | if (bs->drv && bs->drv->bdrv_drain_end) { |
418 | bs->drv->bdrv_drain_end(bs); | |
419 | } | |
a82a3bd1 | 420 | bdrv_parent_drained_end(bs, parent); |
0f115168 | 421 | } |
6820643f KW |
422 | } |
423 | ||
0152bf40 KW |
424 | void bdrv_drained_end(BlockDriverState *bs) |
425 | { | |
384a48fb | 426 | IO_OR_GS_CODE(); |
a82a3bd1 | 427 | bdrv_do_drained_end(bs, NULL); |
d736f119 KW |
428 | } |
429 | ||
b6e84c97 PB |
430 | void bdrv_drain(BlockDriverState *bs) |
431 | { | |
384a48fb | 432 | IO_OR_GS_CODE(); |
6820643f KW |
433 | bdrv_drained_begin(bs); |
434 | bdrv_drained_end(bs); | |
61007b31 SH |
435 | } |
436 | ||
c13ad59f KW |
437 | static void bdrv_drain_assert_idle(BlockDriverState *bs) |
438 | { | |
439 | BdrvChild *child, *next; | |
440 | ||
d73415a3 | 441 | assert(qatomic_read(&bs->in_flight) == 0); |
c13ad59f KW |
442 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
443 | bdrv_drain_assert_idle(child->bs); | |
444 | } | |
445 | } | |
446 | ||
0f12264e KW |
447 | unsigned int bdrv_drain_all_count = 0; |
448 | ||
449 | static bool bdrv_drain_all_poll(void) | |
450 | { | |
451 | BlockDriverState *bs = NULL; | |
452 | bool result = false; | |
f791bf7f | 453 | GLOBAL_STATE_CODE(); |
0f12264e | 454 | |
0f12264e KW |
455 | /* bdrv_drain_poll() can't make changes to the graph and we are holding the |
456 | * main AioContext lock, so iterating bdrv_next_all_states() is safe. */ | |
457 | while ((bs = bdrv_next_all_states(bs))) { | |
458 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
459 | aio_context_acquire(aio_context); | |
299403ae | 460 | result |= bdrv_drain_poll(bs, NULL, true); |
0f12264e KW |
461 | aio_context_release(aio_context); |
462 | } | |
463 | ||
464 | return result; | |
465 | } | |
466 | ||
61007b31 SH |
467 | /* |
468 | * Wait for pending requests to complete across all BlockDriverStates | |
469 | * | |
470 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
471 | * after calling this function. | |
c0778f66 AG |
472 | * |
473 | * This pauses all block jobs and disables external clients. It must | |
474 | * be paired with bdrv_drain_all_end(). | |
475 | * | |
476 | * NOTE: no new block jobs or BlockDriverStates can be created between | |
477 | * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls. | |
61007b31 | 478 | */ |
da0bd744 | 479 | void bdrv_drain_all_begin_nopoll(void) |
61007b31 | 480 | { |
0f12264e | 481 | BlockDriverState *bs = NULL; |
f791bf7f | 482 | GLOBAL_STATE_CODE(); |
61007b31 | 483 | |
c8aa7895 PD |
484 | /* |
485 | * bdrv queue is managed by record/replay, | |
486 | * waiting for finishing the I/O requests may | |
487 | * be infinite | |
488 | */ | |
489 | if (replay_events_enabled()) { | |
490 | return; | |
491 | } | |
492 | ||
0f12264e KW |
493 | /* AIO_WAIT_WHILE() with a NULL context can only be called from the main |
494 | * loop AioContext, so make sure we're in the main context. */ | |
9a7e86c8 | 495 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
496 | assert(bdrv_drain_all_count < INT_MAX); |
497 | bdrv_drain_all_count++; | |
9a7e86c8 | 498 | |
0f12264e KW |
499 | /* Quiesce all nodes, without polling in-flight requests yet. The graph |
500 | * cannot change during this loop. */ | |
501 | while ((bs = bdrv_next_all_states(bs))) { | |
61007b31 SH |
502 | AioContext *aio_context = bdrv_get_aio_context(bs); |
503 | ||
504 | aio_context_acquire(aio_context); | |
a82a3bd1 | 505 | bdrv_do_drained_begin(bs, NULL, false); |
61007b31 SH |
506 | aio_context_release(aio_context); |
507 | } | |
da0bd744 KW |
508 | } |
509 | ||
e2dbca03 | 510 | void coroutine_mixed_fn bdrv_drain_all_begin(void) |
da0bd744 KW |
511 | { |
512 | BlockDriverState *bs = NULL; | |
513 | ||
514 | if (qemu_in_coroutine()) { | |
515 | bdrv_co_yield_to_drain(NULL, true, NULL, true); | |
516 | return; | |
517 | } | |
518 | ||
63945789 PM |
519 | /* |
520 | * bdrv queue is managed by record/replay, | |
521 | * waiting for finishing the I/O requests may | |
522 | * be infinite | |
523 | */ | |
524 | if (replay_events_enabled()) { | |
525 | return; | |
526 | } | |
527 | ||
da0bd744 | 528 | bdrv_drain_all_begin_nopoll(); |
61007b31 | 529 | |
0f12264e | 530 | /* Now poll the in-flight requests */ |
263d5e12 | 531 | AIO_WAIT_WHILE_UNLOCKED(NULL, bdrv_drain_all_poll()); |
0f12264e KW |
532 | |
533 | while ((bs = bdrv_next_all_states(bs))) { | |
c13ad59f | 534 | bdrv_drain_assert_idle(bs); |
61007b31 | 535 | } |
c0778f66 AG |
536 | } |
537 | ||
1a6d3bd2 GK |
538 | void bdrv_drain_all_end_quiesce(BlockDriverState *bs) |
539 | { | |
b4ad82aa | 540 | GLOBAL_STATE_CODE(); |
1a6d3bd2 GK |
541 | |
542 | g_assert(bs->quiesce_counter > 0); | |
543 | g_assert(!bs->refcnt); | |
544 | ||
545 | while (bs->quiesce_counter) { | |
a82a3bd1 | 546 | bdrv_do_drained_end(bs, NULL); |
1a6d3bd2 | 547 | } |
1a6d3bd2 GK |
548 | } |
549 | ||
c0778f66 AG |
550 | void bdrv_drain_all_end(void) |
551 | { | |
0f12264e | 552 | BlockDriverState *bs = NULL; |
f791bf7f | 553 | GLOBAL_STATE_CODE(); |
c0778f66 | 554 | |
c8aa7895 PD |
555 | /* |
556 | * bdrv queue is managed by record/replay, | |
557 | * waiting for finishing the I/O requests may | |
558 | * be endless | |
559 | */ | |
560 | if (replay_events_enabled()) { | |
561 | return; | |
562 | } | |
563 | ||
0f12264e | 564 | while ((bs = bdrv_next_all_states(bs))) { |
61007b31 SH |
565 | AioContext *aio_context = bdrv_get_aio_context(bs); |
566 | ||
567 | aio_context_acquire(aio_context); | |
a82a3bd1 | 568 | bdrv_do_drained_end(bs, NULL); |
61007b31 SH |
569 | aio_context_release(aio_context); |
570 | } | |
0f12264e | 571 | |
e037c09c | 572 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
573 | assert(bdrv_drain_all_count > 0); |
574 | bdrv_drain_all_count--; | |
61007b31 SH |
575 | } |
576 | ||
c0778f66 AG |
577 | void bdrv_drain_all(void) |
578 | { | |
f791bf7f | 579 | GLOBAL_STATE_CODE(); |
c0778f66 AG |
580 | bdrv_drain_all_begin(); |
581 | bdrv_drain_all_end(); | |
582 | } | |
583 | ||
61007b31 SH |
584 | /** |
585 | * Remove an active request from the tracked requests list | |
586 | * | |
587 | * This function should be called when a tracked request is completing. | |
588 | */ | |
f0d43b1e | 589 | static void coroutine_fn tracked_request_end(BdrvTrackedRequest *req) |
61007b31 SH |
590 | { |
591 | if (req->serialising) { | |
d73415a3 | 592 | qatomic_dec(&req->bs->serialising_in_flight); |
61007b31 SH |
593 | } |
594 | ||
fa9185fc | 595 | qemu_mutex_lock(&req->bs->reqs_lock); |
61007b31 | 596 | QLIST_REMOVE(req, list); |
fa9185fc | 597 | qemu_mutex_unlock(&req->bs->reqs_lock); |
3480ce69 SH |
598 | |
599 | /* | |
600 | * At this point qemu_co_queue_wait(&req->wait_queue, ...) won't be called | |
601 | * anymore because the request has been removed from the list, so it's safe | |
602 | * to restart the queue outside reqs_lock to minimize the critical section. | |
603 | */ | |
61007b31 SH |
604 | qemu_co_queue_restart_all(&req->wait_queue); |
605 | } | |
606 | ||
607 | /** | |
608 | * Add an active request to the tracked requests list | |
609 | */ | |
881a4c55 PB |
610 | static void coroutine_fn tracked_request_begin(BdrvTrackedRequest *req, |
611 | BlockDriverState *bs, | |
612 | int64_t offset, | |
613 | int64_t bytes, | |
614 | enum BdrvTrackedRequestType type) | |
61007b31 | 615 | { |
80247264 | 616 | bdrv_check_request(offset, bytes, &error_abort); |
22931a15 | 617 | |
61007b31 SH |
618 | *req = (BdrvTrackedRequest){ |
619 | .bs = bs, | |
620 | .offset = offset, | |
621 | .bytes = bytes, | |
ebde595c | 622 | .type = type, |
61007b31 SH |
623 | .co = qemu_coroutine_self(), |
624 | .serialising = false, | |
625 | .overlap_offset = offset, | |
626 | .overlap_bytes = bytes, | |
627 | }; | |
628 | ||
629 | qemu_co_queue_init(&req->wait_queue); | |
630 | ||
fa9185fc | 631 | qemu_mutex_lock(&bs->reqs_lock); |
61007b31 | 632 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
fa9185fc | 633 | qemu_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
634 | } |
635 | ||
3ba0e1a0 | 636 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, |
80247264 | 637 | int64_t offset, int64_t bytes) |
3ba0e1a0 | 638 | { |
80247264 EB |
639 | bdrv_check_request(offset, bytes, &error_abort); |
640 | ||
3ba0e1a0 PB |
641 | /* aaaa bbbb */ |
642 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
643 | return false; | |
644 | } | |
645 | /* bbbb aaaa */ | |
646 | if (req->overlap_offset >= offset + bytes) { | |
647 | return false; | |
648 | } | |
649 | return true; | |
650 | } | |
651 | ||
3183937f | 652 | /* Called with self->bs->reqs_lock held */ |
881a4c55 | 653 | static coroutine_fn BdrvTrackedRequest * |
3183937f VSO |
654 | bdrv_find_conflicting_request(BdrvTrackedRequest *self) |
655 | { | |
656 | BdrvTrackedRequest *req; | |
657 | ||
658 | QLIST_FOREACH(req, &self->bs->tracked_requests, list) { | |
659 | if (req == self || (!req->serialising && !self->serialising)) { | |
660 | continue; | |
661 | } | |
662 | if (tracked_request_overlaps(req, self->overlap_offset, | |
663 | self->overlap_bytes)) | |
664 | { | |
665 | /* | |
666 | * Hitting this means there was a reentrant request, for | |
667 | * example, a block driver issuing nested requests. This must | |
668 | * never happen since it means deadlock. | |
669 | */ | |
670 | assert(qemu_coroutine_self() != req->co); | |
671 | ||
672 | /* | |
673 | * If the request is already (indirectly) waiting for us, or | |
674 | * will wait for us as soon as it wakes up, then just go on | |
675 | * (instead of producing a deadlock in the former case). | |
676 | */ | |
677 | if (!req->waiting_for) { | |
678 | return req; | |
679 | } | |
680 | } | |
681 | } | |
682 | ||
683 | return NULL; | |
684 | } | |
685 | ||
ec1c8868 | 686 | /* Called with self->bs->reqs_lock held */ |
131498f7 | 687 | static void coroutine_fn |
ec1c8868 | 688 | bdrv_wait_serialising_requests_locked(BdrvTrackedRequest *self) |
3ba0e1a0 PB |
689 | { |
690 | BdrvTrackedRequest *req; | |
3ba0e1a0 | 691 | |
3183937f VSO |
692 | while ((req = bdrv_find_conflicting_request(self))) { |
693 | self->waiting_for = req; | |
ec1c8868 | 694 | qemu_co_queue_wait(&req->wait_queue, &self->bs->reqs_lock); |
3183937f | 695 | self->waiting_for = NULL; |
3183937f | 696 | } |
3ba0e1a0 PB |
697 | } |
698 | ||
8ac5aab2 VSO |
699 | /* Called with req->bs->reqs_lock held */ |
700 | static void tracked_request_set_serialising(BdrvTrackedRequest *req, | |
701 | uint64_t align) | |
61007b31 SH |
702 | { |
703 | int64_t overlap_offset = req->offset & ~(align - 1); | |
80247264 EB |
704 | int64_t overlap_bytes = |
705 | ROUND_UP(req->offset + req->bytes, align) - overlap_offset; | |
706 | ||
707 | bdrv_check_request(req->offset, req->bytes, &error_abort); | |
61007b31 SH |
708 | |
709 | if (!req->serialising) { | |
d73415a3 | 710 | qatomic_inc(&req->bs->serialising_in_flight); |
61007b31 SH |
711 | req->serialising = true; |
712 | } | |
713 | ||
714 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
715 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
09d2f948 VSO |
716 | } |
717 | ||
c28107e9 HR |
718 | /** |
719 | * Return the tracked request on @bs for the current coroutine, or | |
720 | * NULL if there is none. | |
721 | */ | |
722 | BdrvTrackedRequest *coroutine_fn bdrv_co_get_self_request(BlockDriverState *bs) | |
723 | { | |
724 | BdrvTrackedRequest *req; | |
725 | Coroutine *self = qemu_coroutine_self(); | |
967d7905 | 726 | IO_CODE(); |
c28107e9 HR |
727 | |
728 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
729 | if (req->co == self) { | |
730 | return req; | |
731 | } | |
732 | } | |
733 | ||
734 | return NULL; | |
735 | } | |
736 | ||
244483e6 | 737 | /** |
fc6b211f | 738 | * Round a region to subcluster (if supported) or cluster boundaries |
244483e6 | 739 | */ |
a00e70c0 | 740 | void coroutine_fn GRAPH_RDLOCK |
fc6b211f AD |
741 | bdrv_round_to_subclusters(BlockDriverState *bs, int64_t offset, int64_t bytes, |
742 | int64_t *align_offset, int64_t *align_bytes) | |
244483e6 KW |
743 | { |
744 | BlockDriverInfo bdi; | |
384a48fb | 745 | IO_CODE(); |
fc6b211f AD |
746 | if (bdrv_co_get_info(bs, &bdi) < 0 || bdi.subcluster_size == 0) { |
747 | *align_offset = offset; | |
748 | *align_bytes = bytes; | |
244483e6 | 749 | } else { |
fc6b211f AD |
750 | int64_t c = bdi.subcluster_size; |
751 | *align_offset = QEMU_ALIGN_DOWN(offset, c); | |
752 | *align_bytes = QEMU_ALIGN_UP(offset - *align_offset + bytes, c); | |
244483e6 KW |
753 | } |
754 | } | |
755 | ||
a00e70c0 | 756 | static int coroutine_fn GRAPH_RDLOCK bdrv_get_cluster_size(BlockDriverState *bs) |
61007b31 SH |
757 | { |
758 | BlockDriverInfo bdi; | |
759 | int ret; | |
760 | ||
3d47eb0a | 761 | ret = bdrv_co_get_info(bs, &bdi); |
61007b31 | 762 | if (ret < 0 || bdi.cluster_size == 0) { |
a5b8dd2c | 763 | return bs->bl.request_alignment; |
61007b31 SH |
764 | } else { |
765 | return bdi.cluster_size; | |
766 | } | |
767 | } | |
768 | ||
99723548 PB |
769 | void bdrv_inc_in_flight(BlockDriverState *bs) |
770 | { | |
967d7905 | 771 | IO_CODE(); |
d73415a3 | 772 | qatomic_inc(&bs->in_flight); |
99723548 PB |
773 | } |
774 | ||
c9d1a561 PB |
775 | void bdrv_wakeup(BlockDriverState *bs) |
776 | { | |
967d7905 | 777 | IO_CODE(); |
cfe29d82 | 778 | aio_wait_kick(); |
c9d1a561 PB |
779 | } |
780 | ||
99723548 PB |
781 | void bdrv_dec_in_flight(BlockDriverState *bs) |
782 | { | |
967d7905 | 783 | IO_CODE(); |
d73415a3 | 784 | qatomic_dec(&bs->in_flight); |
c9d1a561 | 785 | bdrv_wakeup(bs); |
99723548 PB |
786 | } |
787 | ||
131498f7 DL |
788 | static void coroutine_fn |
789 | bdrv_wait_serialising_requests(BdrvTrackedRequest *self) | |
61007b31 SH |
790 | { |
791 | BlockDriverState *bs = self->bs; | |
61007b31 | 792 | |
d73415a3 | 793 | if (!qatomic_read(&bs->serialising_in_flight)) { |
131498f7 | 794 | return; |
61007b31 SH |
795 | } |
796 | ||
fa9185fc | 797 | qemu_mutex_lock(&bs->reqs_lock); |
131498f7 | 798 | bdrv_wait_serialising_requests_locked(self); |
fa9185fc | 799 | qemu_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
800 | } |
801 | ||
131498f7 | 802 | void coroutine_fn bdrv_make_request_serialising(BdrvTrackedRequest *req, |
8ac5aab2 VSO |
803 | uint64_t align) |
804 | { | |
967d7905 | 805 | IO_CODE(); |
8ac5aab2 | 806 | |
fa9185fc | 807 | qemu_mutex_lock(&req->bs->reqs_lock); |
8ac5aab2 VSO |
808 | |
809 | tracked_request_set_serialising(req, align); | |
131498f7 | 810 | bdrv_wait_serialising_requests_locked(req); |
8ac5aab2 | 811 | |
fa9185fc | 812 | qemu_mutex_unlock(&req->bs->reqs_lock); |
8ac5aab2 VSO |
813 | } |
814 | ||
558902cc VSO |
815 | int bdrv_check_qiov_request(int64_t offset, int64_t bytes, |
816 | QEMUIOVector *qiov, size_t qiov_offset, | |
817 | Error **errp) | |
61007b31 | 818 | { |
63f4ad11 VSO |
819 | /* |
820 | * Check generic offset/bytes correctness | |
821 | */ | |
822 | ||
69b55e03 VSO |
823 | if (offset < 0) { |
824 | error_setg(errp, "offset is negative: %" PRIi64, offset); | |
825 | return -EIO; | |
826 | } | |
827 | ||
828 | if (bytes < 0) { | |
829 | error_setg(errp, "bytes is negative: %" PRIi64, bytes); | |
61007b31 SH |
830 | return -EIO; |
831 | } | |
832 | ||
8b117001 | 833 | if (bytes > BDRV_MAX_LENGTH) { |
69b55e03 VSO |
834 | error_setg(errp, "bytes(%" PRIi64 ") exceeds maximum(%" PRIi64 ")", |
835 | bytes, BDRV_MAX_LENGTH); | |
836 | return -EIO; | |
837 | } | |
838 | ||
839 | if (offset > BDRV_MAX_LENGTH) { | |
840 | error_setg(errp, "offset(%" PRIi64 ") exceeds maximum(%" PRIi64 ")", | |
841 | offset, BDRV_MAX_LENGTH); | |
8b117001 VSO |
842 | return -EIO; |
843 | } | |
844 | ||
845 | if (offset > BDRV_MAX_LENGTH - bytes) { | |
69b55e03 VSO |
846 | error_setg(errp, "sum of offset(%" PRIi64 ") and bytes(%" PRIi64 ") " |
847 | "exceeds maximum(%" PRIi64 ")", offset, bytes, | |
848 | BDRV_MAX_LENGTH); | |
8b117001 VSO |
849 | return -EIO; |
850 | } | |
851 | ||
63f4ad11 VSO |
852 | if (!qiov) { |
853 | return 0; | |
854 | } | |
855 | ||
856 | /* | |
857 | * Check qiov and qiov_offset | |
858 | */ | |
859 | ||
860 | if (qiov_offset > qiov->size) { | |
861 | error_setg(errp, "qiov_offset(%zu) overflow io vector size(%zu)", | |
862 | qiov_offset, qiov->size); | |
863 | return -EIO; | |
864 | } | |
865 | ||
866 | if (bytes > qiov->size - qiov_offset) { | |
867 | error_setg(errp, "bytes(%" PRIi64 ") + qiov_offset(%zu) overflow io " | |
868 | "vector size(%zu)", bytes, qiov_offset, qiov->size); | |
869 | return -EIO; | |
870 | } | |
871 | ||
8b117001 VSO |
872 | return 0; |
873 | } | |
874 | ||
63f4ad11 VSO |
875 | int bdrv_check_request(int64_t offset, int64_t bytes, Error **errp) |
876 | { | |
877 | return bdrv_check_qiov_request(offset, bytes, NULL, 0, errp); | |
878 | } | |
879 | ||
880 | static int bdrv_check_request32(int64_t offset, int64_t bytes, | |
881 | QEMUIOVector *qiov, size_t qiov_offset) | |
8b117001 | 882 | { |
63f4ad11 | 883 | int ret = bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, NULL); |
8b117001 VSO |
884 | if (ret < 0) { |
885 | return ret; | |
886 | } | |
887 | ||
888 | if (bytes > BDRV_REQUEST_MAX_BYTES) { | |
61007b31 SH |
889 | return -EIO; |
890 | } | |
891 | ||
892 | return 0; | |
893 | } | |
894 | ||
61007b31 | 895 | /* |
74021bc4 | 896 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
897 | * The operation is sped up by checking the block status and only writing |
898 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 899 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 900 | * BDRV_REQ_FUA). |
61007b31 | 901 | * |
f4649069 | 902 | * Returns < 0 on error, 0 on success. For error codes see bdrv_pwrite(). |
61007b31 | 903 | */ |
720ff280 | 904 | int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags) |
61007b31 | 905 | { |
237d78f8 EB |
906 | int ret; |
907 | int64_t target_size, bytes, offset = 0; | |
720ff280 | 908 | BlockDriverState *bs = child->bs; |
384a48fb | 909 | IO_CODE(); |
61007b31 | 910 | |
7286d610 EB |
911 | target_size = bdrv_getlength(bs); |
912 | if (target_size < 0) { | |
913 | return target_size; | |
61007b31 SH |
914 | } |
915 | ||
916 | for (;;) { | |
7286d610 EB |
917 | bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES); |
918 | if (bytes <= 0) { | |
61007b31 SH |
919 | return 0; |
920 | } | |
237d78f8 | 921 | ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL); |
61007b31 | 922 | if (ret < 0) { |
61007b31 SH |
923 | return ret; |
924 | } | |
925 | if (ret & BDRV_BLOCK_ZERO) { | |
237d78f8 | 926 | offset += bytes; |
61007b31 SH |
927 | continue; |
928 | } | |
237d78f8 | 929 | ret = bdrv_pwrite_zeroes(child, offset, bytes, flags); |
61007b31 | 930 | if (ret < 0) { |
61007b31 SH |
931 | return ret; |
932 | } | |
237d78f8 | 933 | offset += bytes; |
61007b31 SH |
934 | } |
935 | } | |
936 | ||
61007b31 SH |
937 | /* |
938 | * Writes to the file and ensures that no writes are reordered across this | |
939 | * request (acts as a barrier) | |
940 | * | |
941 | * Returns 0 on success, -errno in error cases. | |
942 | */ | |
e97190a4 AF |
943 | int coroutine_fn bdrv_co_pwrite_sync(BdrvChild *child, int64_t offset, |
944 | int64_t bytes, const void *buf, | |
945 | BdrvRequestFlags flags) | |
61007b31 SH |
946 | { |
947 | int ret; | |
384a48fb | 948 | IO_CODE(); |
b24a4c41 | 949 | assert_bdrv_graph_readable(); |
88095349 | 950 | |
e97190a4 | 951 | ret = bdrv_co_pwrite(child, offset, bytes, buf, flags); |
61007b31 SH |
952 | if (ret < 0) { |
953 | return ret; | |
954 | } | |
955 | ||
e97190a4 | 956 | ret = bdrv_co_flush(child->bs); |
855a6a93 KW |
957 | if (ret < 0) { |
958 | return ret; | |
61007b31 SH |
959 | } |
960 | ||
961 | return 0; | |
962 | } | |
963 | ||
08844473 KW |
964 | typedef struct CoroutineIOCompletion { |
965 | Coroutine *coroutine; | |
966 | int ret; | |
967 | } CoroutineIOCompletion; | |
968 | ||
969 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
970 | { | |
971 | CoroutineIOCompletion *co = opaque; | |
972 | ||
973 | co->ret = ret; | |
b9e413dd | 974 | aio_co_wake(co->coroutine); |
08844473 KW |
975 | } |
976 | ||
7b1fb72e KW |
977 | static int coroutine_fn GRAPH_RDLOCK |
978 | bdrv_driver_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
979 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
166fe960 KW |
980 | { |
981 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
982 | int64_t sector_num; |
983 | unsigned int nb_sectors; | |
ac850bf0 VSO |
984 | QEMUIOVector local_qiov; |
985 | int ret; | |
b9b10c35 | 986 | assert_bdrv_graph_readable(); |
3fb06697 | 987 | |
17abcbee | 988 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
e8b65355 | 989 | assert(!(flags & ~bs->supported_read_flags)); |
fa166538 | 990 | |
d470ad42 HR |
991 | if (!drv) { |
992 | return -ENOMEDIUM; | |
993 | } | |
994 | ||
ac850bf0 VSO |
995 | if (drv->bdrv_co_preadv_part) { |
996 | return drv->bdrv_co_preadv_part(bs, offset, bytes, qiov, qiov_offset, | |
997 | flags); | |
998 | } | |
999 | ||
1000 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1001 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1002 | qiov = &local_qiov; | |
1003 | } | |
1004 | ||
3fb06697 | 1005 | if (drv->bdrv_co_preadv) { |
ac850bf0 VSO |
1006 | ret = drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); |
1007 | goto out; | |
3fb06697 KW |
1008 | } |
1009 | ||
edfab6a0 | 1010 | if (drv->bdrv_aio_preadv) { |
08844473 KW |
1011 | BlockAIOCB *acb; |
1012 | CoroutineIOCompletion co = { | |
1013 | .coroutine = qemu_coroutine_self(), | |
1014 | }; | |
1015 | ||
edfab6a0 EB |
1016 | acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags, |
1017 | bdrv_co_io_em_complete, &co); | |
08844473 | 1018 | if (acb == NULL) { |
ac850bf0 VSO |
1019 | ret = -EIO; |
1020 | goto out; | |
08844473 KW |
1021 | } else { |
1022 | qemu_coroutine_yield(); | |
ac850bf0 VSO |
1023 | ret = co.ret; |
1024 | goto out; | |
08844473 KW |
1025 | } |
1026 | } | |
edfab6a0 EB |
1027 | |
1028 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1029 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1030 | ||
1bbbf32d NS |
1031 | assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)); |
1032 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
41ae31e3 | 1033 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 EB |
1034 | assert(drv->bdrv_co_readv); |
1035 | ||
ac850bf0 VSO |
1036 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); |
1037 | ||
1038 | out: | |
1039 | if (qiov == &local_qiov) { | |
1040 | qemu_iovec_destroy(&local_qiov); | |
1041 | } | |
1042 | ||
1043 | return ret; | |
166fe960 KW |
1044 | } |
1045 | ||
7b1fb72e KW |
1046 | static int coroutine_fn GRAPH_RDLOCK |
1047 | bdrv_driver_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1048 | QEMUIOVector *qiov, size_t qiov_offset, | |
1049 | BdrvRequestFlags flags) | |
78a07294 KW |
1050 | { |
1051 | BlockDriver *drv = bs->drv; | |
e8b65355 | 1052 | bool emulate_fua = false; |
3fb06697 KW |
1053 | int64_t sector_num; |
1054 | unsigned int nb_sectors; | |
ac850bf0 | 1055 | QEMUIOVector local_qiov; |
78a07294 | 1056 | int ret; |
b9b10c35 | 1057 | assert_bdrv_graph_readable(); |
78a07294 | 1058 | |
17abcbee | 1059 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
fa166538 | 1060 | |
d470ad42 HR |
1061 | if (!drv) { |
1062 | return -ENOMEDIUM; | |
1063 | } | |
1064 | ||
e8b65355 SH |
1065 | if ((flags & BDRV_REQ_FUA) && |
1066 | (~bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1067 | flags &= ~BDRV_REQ_FUA; | |
1068 | emulate_fua = true; | |
1069 | } | |
1070 | ||
1071 | flags &= bs->supported_write_flags; | |
1072 | ||
ac850bf0 VSO |
1073 | if (drv->bdrv_co_pwritev_part) { |
1074 | ret = drv->bdrv_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, | |
e8b65355 | 1075 | flags); |
ac850bf0 VSO |
1076 | goto emulate_flags; |
1077 | } | |
1078 | ||
1079 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1080 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1081 | qiov = &local_qiov; | |
1082 | } | |
1083 | ||
3fb06697 | 1084 | if (drv->bdrv_co_pwritev) { |
e8b65355 | 1085 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, flags); |
3fb06697 KW |
1086 | goto emulate_flags; |
1087 | } | |
1088 | ||
edfab6a0 | 1089 | if (drv->bdrv_aio_pwritev) { |
08844473 KW |
1090 | BlockAIOCB *acb; |
1091 | CoroutineIOCompletion co = { | |
1092 | .coroutine = qemu_coroutine_self(), | |
1093 | }; | |
1094 | ||
e8b65355 | 1095 | acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov, flags, |
edfab6a0 | 1096 | bdrv_co_io_em_complete, &co); |
08844473 | 1097 | if (acb == NULL) { |
3fb06697 | 1098 | ret = -EIO; |
08844473 KW |
1099 | } else { |
1100 | qemu_coroutine_yield(); | |
3fb06697 | 1101 | ret = co.ret; |
08844473 | 1102 | } |
edfab6a0 EB |
1103 | goto emulate_flags; |
1104 | } | |
1105 | ||
1106 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1107 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1108 | ||
1bbbf32d NS |
1109 | assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)); |
1110 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
41ae31e3 | 1111 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 | 1112 | |
e18a58b4 | 1113 | assert(drv->bdrv_co_writev); |
e8b65355 | 1114 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov, flags); |
78a07294 | 1115 | |
3fb06697 | 1116 | emulate_flags: |
e8b65355 | 1117 | if (ret == 0 && emulate_fua) { |
78a07294 KW |
1118 | ret = bdrv_co_flush(bs); |
1119 | } | |
1120 | ||
ac850bf0 VSO |
1121 | if (qiov == &local_qiov) { |
1122 | qemu_iovec_destroy(&local_qiov); | |
1123 | } | |
1124 | ||
78a07294 KW |
1125 | return ret; |
1126 | } | |
1127 | ||
7b1fb72e | 1128 | static int coroutine_fn GRAPH_RDLOCK |
17abcbee VSO |
1129 | bdrv_driver_pwritev_compressed(BlockDriverState *bs, int64_t offset, |
1130 | int64_t bytes, QEMUIOVector *qiov, | |
ac850bf0 | 1131 | size_t qiov_offset) |
29a298af PB |
1132 | { |
1133 | BlockDriver *drv = bs->drv; | |
ac850bf0 VSO |
1134 | QEMUIOVector local_qiov; |
1135 | int ret; | |
b9b10c35 | 1136 | assert_bdrv_graph_readable(); |
29a298af | 1137 | |
17abcbee VSO |
1138 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1139 | ||
d470ad42 HR |
1140 | if (!drv) { |
1141 | return -ENOMEDIUM; | |
1142 | } | |
1143 | ||
ac850bf0 | 1144 | if (!block_driver_can_compress(drv)) { |
29a298af PB |
1145 | return -ENOTSUP; |
1146 | } | |
1147 | ||
ac850bf0 VSO |
1148 | if (drv->bdrv_co_pwritev_compressed_part) { |
1149 | return drv->bdrv_co_pwritev_compressed_part(bs, offset, bytes, | |
1150 | qiov, qiov_offset); | |
1151 | } | |
1152 | ||
1153 | if (qiov_offset == 0) { | |
1154 | return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov); | |
1155 | } | |
1156 | ||
1157 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1158 | ret = drv->bdrv_co_pwritev_compressed(bs, offset, bytes, &local_qiov); | |
1159 | qemu_iovec_destroy(&local_qiov); | |
1160 | ||
1161 | return ret; | |
29a298af PB |
1162 | } |
1163 | ||
7b1fb72e KW |
1164 | static int coroutine_fn GRAPH_RDLOCK |
1165 | bdrv_co_do_copy_on_readv(BdrvChild *child, int64_t offset, int64_t bytes, | |
1166 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
61007b31 | 1167 | { |
85c97ca7 KW |
1168 | BlockDriverState *bs = child->bs; |
1169 | ||
61007b31 SH |
1170 | /* Perform I/O through a temporary buffer so that users who scribble over |
1171 | * their read buffer while the operation is in progress do not end up | |
1172 | * modifying the image file. This is critical for zero-copy guest I/O | |
1173 | * where anything might happen inside guest memory. | |
1174 | */ | |
2275cc90 | 1175 | void *bounce_buffer = NULL; |
61007b31 SH |
1176 | |
1177 | BlockDriver *drv = bs->drv; | |
fc6b211f AD |
1178 | int64_t align_offset; |
1179 | int64_t align_bytes; | |
9df5afbd | 1180 | int64_t skip_bytes; |
61007b31 | 1181 | int ret; |
cb2e2878 EB |
1182 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
1183 | BDRV_REQUEST_MAX_BYTES); | |
9df5afbd | 1184 | int64_t progress = 0; |
8644476e | 1185 | bool skip_write; |
61007b31 | 1186 | |
9df5afbd VSO |
1187 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1188 | ||
d470ad42 HR |
1189 | if (!drv) { |
1190 | return -ENOMEDIUM; | |
1191 | } | |
1192 | ||
8644476e HR |
1193 | /* |
1194 | * Do not write anything when the BDS is inactive. That is not | |
1195 | * allowed, and it would not help. | |
1196 | */ | |
1197 | skip_write = (bs->open_flags & BDRV_O_INACTIVE); | |
1198 | ||
1bf03e66 KW |
1199 | /* FIXME We cannot require callers to have write permissions when all they |
1200 | * are doing is a read request. If we did things right, write permissions | |
1201 | * would be obtained anyway, but internally by the copy-on-read code. As | |
765d9df9 | 1202 | * long as it is implemented here rather than in a separate filter driver, |
1bf03e66 KW |
1203 | * the copy-on-read code doesn't have its own BdrvChild, however, for which |
1204 | * it could request permissions. Therefore we have to bypass the permission | |
1205 | * system for the moment. */ | |
1206 | // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
afa4b293 | 1207 | |
61007b31 | 1208 | /* Cover entire cluster so no additional backing file I/O is required when |
cb2e2878 EB |
1209 | * allocating cluster in the image file. Note that this value may exceed |
1210 | * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which | |
1211 | * is one reason we loop rather than doing it all at once. | |
61007b31 | 1212 | */ |
fc6b211f AD |
1213 | bdrv_round_to_subclusters(bs, offset, bytes, &align_offset, &align_bytes); |
1214 | skip_bytes = offset - align_offset; | |
61007b31 | 1215 | |
244483e6 | 1216 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
fc6b211f | 1217 | align_offset, align_bytes); |
61007b31 | 1218 | |
fc6b211f | 1219 | while (align_bytes) { |
cb2e2878 | 1220 | int64_t pnum; |
61007b31 | 1221 | |
8644476e HR |
1222 | if (skip_write) { |
1223 | ret = 1; /* "already allocated", so nothing will be copied */ | |
fc6b211f | 1224 | pnum = MIN(align_bytes, max_transfer); |
8644476e | 1225 | } else { |
fc6b211f AD |
1226 | ret = bdrv_is_allocated(bs, align_offset, |
1227 | MIN(align_bytes, max_transfer), &pnum); | |
8644476e HR |
1228 | if (ret < 0) { |
1229 | /* | |
1230 | * Safe to treat errors in querying allocation as if | |
1231 | * unallocated; we'll probably fail again soon on the | |
1232 | * read, but at least that will set a decent errno. | |
1233 | */ | |
fc6b211f | 1234 | pnum = MIN(align_bytes, max_transfer); |
8644476e | 1235 | } |
61007b31 | 1236 | |
8644476e HR |
1237 | /* Stop at EOF if the image ends in the middle of the cluster */ |
1238 | if (ret == 0 && pnum == 0) { | |
1239 | assert(progress >= bytes); | |
1240 | break; | |
1241 | } | |
b0ddcbbb | 1242 | |
8644476e HR |
1243 | assert(skip_bytes < pnum); |
1244 | } | |
61007b31 | 1245 | |
cb2e2878 | 1246 | if (ret <= 0) { |
1143ec5e VSO |
1247 | QEMUIOVector local_qiov; |
1248 | ||
cb2e2878 | 1249 | /* Must copy-on-read; use the bounce buffer */ |
0d93ed08 | 1250 | pnum = MIN(pnum, MAX_BOUNCE_BUFFER); |
2275cc90 | 1251 | if (!bounce_buffer) { |
fc6b211f | 1252 | int64_t max_we_need = MAX(pnum, align_bytes - pnum); |
2275cc90 VSO |
1253 | int64_t max_allowed = MIN(max_transfer, MAX_BOUNCE_BUFFER); |
1254 | int64_t bounce_buffer_len = MIN(max_we_need, max_allowed); | |
1255 | ||
1256 | bounce_buffer = qemu_try_blockalign(bs, bounce_buffer_len); | |
1257 | if (!bounce_buffer) { | |
1258 | ret = -ENOMEM; | |
1259 | goto err; | |
1260 | } | |
1261 | } | |
0d93ed08 | 1262 | qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum); |
61007b31 | 1263 | |
fc6b211f | 1264 | ret = bdrv_driver_preadv(bs, align_offset, pnum, |
ac850bf0 | 1265 | &local_qiov, 0, 0); |
cb2e2878 EB |
1266 | if (ret < 0) { |
1267 | goto err; | |
1268 | } | |
1269 | ||
c834dc05 | 1270 | bdrv_co_debug_event(bs, BLKDBG_COR_WRITE); |
cb2e2878 EB |
1271 | if (drv->bdrv_co_pwrite_zeroes && |
1272 | buffer_is_zero(bounce_buffer, pnum)) { | |
1273 | /* FIXME: Should we (perhaps conditionally) be setting | |
1274 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
1275 | * that still correctly reads as zero? */ | |
fc6b211f | 1276 | ret = bdrv_co_do_pwrite_zeroes(bs, align_offset, pnum, |
7adcf59f | 1277 | BDRV_REQ_WRITE_UNCHANGED); |
cb2e2878 EB |
1278 | } else { |
1279 | /* This does not change the data on the disk, it is not | |
1280 | * necessary to flush even in cache=writethrough mode. | |
1281 | */ | |
fc6b211f | 1282 | ret = bdrv_driver_pwritev(bs, align_offset, pnum, |
ac850bf0 | 1283 | &local_qiov, 0, |
7adcf59f | 1284 | BDRV_REQ_WRITE_UNCHANGED); |
cb2e2878 EB |
1285 | } |
1286 | ||
1287 | if (ret < 0) { | |
1288 | /* It might be okay to ignore write errors for guest | |
1289 | * requests. If this is a deliberate copy-on-read | |
1290 | * then we don't want to ignore the error. Simply | |
1291 | * report it in all cases. | |
1292 | */ | |
1293 | goto err; | |
1294 | } | |
1295 | ||
3299e5ec | 1296 | if (!(flags & BDRV_REQ_PREFETCH)) { |
1143ec5e VSO |
1297 | qemu_iovec_from_buf(qiov, qiov_offset + progress, |
1298 | bounce_buffer + skip_bytes, | |
4ab78b19 | 1299 | MIN(pnum - skip_bytes, bytes - progress)); |
3299e5ec VSO |
1300 | } |
1301 | } else if (!(flags & BDRV_REQ_PREFETCH)) { | |
cb2e2878 | 1302 | /* Read directly into the destination */ |
1143ec5e VSO |
1303 | ret = bdrv_driver_preadv(bs, offset + progress, |
1304 | MIN(pnum - skip_bytes, bytes - progress), | |
1305 | qiov, qiov_offset + progress, 0); | |
cb2e2878 EB |
1306 | if (ret < 0) { |
1307 | goto err; | |
1308 | } | |
1309 | } | |
1310 | ||
fc6b211f AD |
1311 | align_offset += pnum; |
1312 | align_bytes -= pnum; | |
cb2e2878 EB |
1313 | progress += pnum - skip_bytes; |
1314 | skip_bytes = 0; | |
1315 | } | |
1316 | ret = 0; | |
61007b31 SH |
1317 | |
1318 | err: | |
1319 | qemu_vfree(bounce_buffer); | |
1320 | return ret; | |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * Forwards an already correctly aligned request to the BlockDriver. This | |
1a62d0ac EB |
1325 | * handles copy on read, zeroing after EOF, and fragmentation of large |
1326 | * reads; any other features must be implemented by the caller. | |
61007b31 | 1327 | */ |
7b1fb72e KW |
1328 | static int coroutine_fn GRAPH_RDLOCK |
1329 | bdrv_aligned_preadv(BdrvChild *child, BdrvTrackedRequest *req, | |
1330 | int64_t offset, int64_t bytes, int64_t align, | |
1331 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
61007b31 | 1332 | { |
85c97ca7 | 1333 | BlockDriverState *bs = child->bs; |
c9d20029 | 1334 | int64_t total_bytes, max_bytes; |
1a62d0ac | 1335 | int ret = 0; |
8b0c5d76 | 1336 | int64_t bytes_remaining = bytes; |
1a62d0ac | 1337 | int max_transfer; |
61007b31 | 1338 | |
8b0c5d76 | 1339 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
49c07526 KW |
1340 | assert(is_power_of_2(align)); |
1341 | assert((offset & (align - 1)) == 0); | |
1342 | assert((bytes & (align - 1)) == 0); | |
abb06c5a | 1343 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
1a62d0ac EB |
1344 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1345 | align); | |
a604fa2b | 1346 | |
e8b65355 SH |
1347 | /* |
1348 | * TODO: We would need a per-BDS .supported_read_flags and | |
a604fa2b EB |
1349 | * potential fallback support, if we ever implement any read flags |
1350 | * to pass through to drivers. For now, there aren't any | |
e8b65355 SH |
1351 | * passthrough flags except the BDRV_REQ_REGISTERED_BUF optimization hint. |
1352 | */ | |
1353 | assert(!(flags & ~(BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH | | |
1354 | BDRV_REQ_REGISTERED_BUF))); | |
61007b31 SH |
1355 | |
1356 | /* Handle Copy on Read and associated serialisation */ | |
1357 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
1358 | /* If we touch the same cluster it counts as an overlap. This | |
1359 | * guarantees that allocating writes will be serialized and not race | |
1360 | * with each other for the same cluster. For example, in copy-on-read | |
1361 | * it ensures that the CoR read and write operations are atomic and | |
1362 | * guest writes cannot interleave between them. */ | |
8ac5aab2 | 1363 | bdrv_make_request_serialising(req, bdrv_get_cluster_size(bs)); |
18fbd0de PB |
1364 | } else { |
1365 | bdrv_wait_serialising_requests(req); | |
61007b31 SH |
1366 | } |
1367 | ||
61007b31 | 1368 | if (flags & BDRV_REQ_COPY_ON_READ) { |
d6a644bb | 1369 | int64_t pnum; |
61007b31 | 1370 | |
897dd0ec AS |
1371 | /* The flag BDRV_REQ_COPY_ON_READ has reached its addressee */ |
1372 | flags &= ~BDRV_REQ_COPY_ON_READ; | |
1373 | ||
88e63df2 | 1374 | ret = bdrv_is_allocated(bs, offset, bytes, &pnum); |
61007b31 SH |
1375 | if (ret < 0) { |
1376 | goto out; | |
1377 | } | |
1378 | ||
88e63df2 | 1379 | if (!ret || pnum != bytes) { |
65cd4424 VSO |
1380 | ret = bdrv_co_do_copy_on_readv(child, offset, bytes, |
1381 | qiov, qiov_offset, flags); | |
3299e5ec VSO |
1382 | goto out; |
1383 | } else if (flags & BDRV_REQ_PREFETCH) { | |
61007b31 SH |
1384 | goto out; |
1385 | } | |
1386 | } | |
1387 | ||
1a62d0ac | 1388 | /* Forward the request to the BlockDriver, possibly fragmenting it */ |
0af02bd1 | 1389 | total_bytes = bdrv_co_getlength(bs); |
c9d20029 KW |
1390 | if (total_bytes < 0) { |
1391 | ret = total_bytes; | |
1392 | goto out; | |
1393 | } | |
61007b31 | 1394 | |
e8b65355 | 1395 | assert(!(flags & ~(bs->supported_read_flags | BDRV_REQ_REGISTERED_BUF))); |
897dd0ec | 1396 | |
c9d20029 | 1397 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
1a62d0ac | 1398 | if (bytes <= max_bytes && bytes <= max_transfer) { |
897dd0ec | 1399 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, qiov_offset, flags); |
1a62d0ac EB |
1400 | goto out; |
1401 | } | |
61007b31 | 1402 | |
1a62d0ac | 1403 | while (bytes_remaining) { |
8b0c5d76 | 1404 | int64_t num; |
61007b31 | 1405 | |
1a62d0ac | 1406 | if (max_bytes) { |
1a62d0ac EB |
1407 | num = MIN(bytes_remaining, MIN(max_bytes, max_transfer)); |
1408 | assert(num); | |
61007b31 | 1409 | |
1a62d0ac | 1410 | ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining, |
134b7dec | 1411 | num, qiov, |
897dd0ec AS |
1412 | qiov_offset + bytes - bytes_remaining, |
1413 | flags); | |
1a62d0ac | 1414 | max_bytes -= num; |
1a62d0ac EB |
1415 | } else { |
1416 | num = bytes_remaining; | |
134b7dec HR |
1417 | ret = qemu_iovec_memset(qiov, qiov_offset + bytes - bytes_remaining, |
1418 | 0, bytes_remaining); | |
1a62d0ac EB |
1419 | } |
1420 | if (ret < 0) { | |
1421 | goto out; | |
1422 | } | |
1423 | bytes_remaining -= num; | |
61007b31 SH |
1424 | } |
1425 | ||
1426 | out: | |
1a62d0ac | 1427 | return ret < 0 ? ret : 0; |
61007b31 SH |
1428 | } |
1429 | ||
61007b31 | 1430 | /* |
7a3f542f VSO |
1431 | * Request padding |
1432 | * | |
1433 | * |<---- align ----->| |<----- align ---->| | |
1434 | * |<- head ->|<------------- bytes ------------->|<-- tail -->| | |
1435 | * | | | | | | | |
1436 | * -*----------$-------*-------- ... --------*-----$------------*--- | |
1437 | * | | | | | | | |
1438 | * | offset | | end | | |
1439 | * ALIGN_DOWN(offset) ALIGN_UP(offset) ALIGN_DOWN(end) ALIGN_UP(end) | |
1440 | * [buf ... ) [tail_buf ) | |
1441 | * | |
1442 | * @buf is an aligned allocation needed to store @head and @tail paddings. @head | |
1443 | * is placed at the beginning of @buf and @tail at the @end. | |
1444 | * | |
1445 | * @tail_buf is a pointer to sub-buffer, corresponding to align-sized chunk | |
1446 | * around tail, if tail exists. | |
1447 | * | |
1448 | * @merge_reads is true for small requests, | |
1449 | * if @buf_len == @head + bytes + @tail. In this case it is possible that both | |
1450 | * head and tail exist but @buf_len == align and @tail_buf == @buf. | |
18743311 HC |
1451 | * |
1452 | * @write is true for write requests, false for read requests. | |
1453 | * | |
1454 | * If padding makes the vector too long (exceeding IOV_MAX), then we need to | |
1455 | * merge existing vector elements into a single one. @collapse_bounce_buf acts | |
1456 | * as the bounce buffer in such cases. @pre_collapse_qiov has the pre-collapse | |
1457 | * I/O vector elements so for read requests, the data can be copied back after | |
1458 | * the read is done. | |
7a3f542f VSO |
1459 | */ |
1460 | typedef struct BdrvRequestPadding { | |
1461 | uint8_t *buf; | |
1462 | size_t buf_len; | |
1463 | uint8_t *tail_buf; | |
1464 | size_t head; | |
1465 | size_t tail; | |
1466 | bool merge_reads; | |
18743311 | 1467 | bool write; |
7a3f542f | 1468 | QEMUIOVector local_qiov; |
18743311 HC |
1469 | |
1470 | uint8_t *collapse_bounce_buf; | |
1471 | size_t collapse_len; | |
1472 | QEMUIOVector pre_collapse_qiov; | |
7a3f542f VSO |
1473 | } BdrvRequestPadding; |
1474 | ||
1475 | static bool bdrv_init_padding(BlockDriverState *bs, | |
1476 | int64_t offset, int64_t bytes, | |
18743311 | 1477 | bool write, |
7a3f542f VSO |
1478 | BdrvRequestPadding *pad) |
1479 | { | |
a56ed80c VSO |
1480 | int64_t align = bs->bl.request_alignment; |
1481 | int64_t sum; | |
1482 | ||
1483 | bdrv_check_request(offset, bytes, &error_abort); | |
1484 | assert(align <= INT_MAX); /* documented in block/block_int.h */ | |
1485 | assert(align <= SIZE_MAX / 2); /* so we can allocate the buffer */ | |
7a3f542f VSO |
1486 | |
1487 | memset(pad, 0, sizeof(*pad)); | |
1488 | ||
1489 | pad->head = offset & (align - 1); | |
1490 | pad->tail = ((offset + bytes) & (align - 1)); | |
1491 | if (pad->tail) { | |
1492 | pad->tail = align - pad->tail; | |
1493 | } | |
1494 | ||
ac9d00bf | 1495 | if (!pad->head && !pad->tail) { |
7a3f542f VSO |
1496 | return false; |
1497 | } | |
1498 | ||
ac9d00bf VSO |
1499 | assert(bytes); /* Nothing good in aligning zero-length requests */ |
1500 | ||
7a3f542f VSO |
1501 | sum = pad->head + bytes + pad->tail; |
1502 | pad->buf_len = (sum > align && pad->head && pad->tail) ? 2 * align : align; | |
1503 | pad->buf = qemu_blockalign(bs, pad->buf_len); | |
1504 | pad->merge_reads = sum == pad->buf_len; | |
1505 | if (pad->tail) { | |
1506 | pad->tail_buf = pad->buf + pad->buf_len - align; | |
1507 | } | |
1508 | ||
18743311 HC |
1509 | pad->write = write; |
1510 | ||
7a3f542f VSO |
1511 | return true; |
1512 | } | |
1513 | ||
7b1fb72e KW |
1514 | static int coroutine_fn GRAPH_RDLOCK |
1515 | bdrv_padding_rmw_read(BdrvChild *child, BdrvTrackedRequest *req, | |
1516 | BdrvRequestPadding *pad, bool zero_middle) | |
7a3f542f VSO |
1517 | { |
1518 | QEMUIOVector local_qiov; | |
1519 | BlockDriverState *bs = child->bs; | |
1520 | uint64_t align = bs->bl.request_alignment; | |
1521 | int ret; | |
1522 | ||
1523 | assert(req->serialising && pad->buf); | |
1524 | ||
1525 | if (pad->head || pad->merge_reads) { | |
8b0c5d76 | 1526 | int64_t bytes = pad->merge_reads ? pad->buf_len : align; |
7a3f542f VSO |
1527 | |
1528 | qemu_iovec_init_buf(&local_qiov, pad->buf, bytes); | |
1529 | ||
1530 | if (pad->head) { | |
c834dc05 | 1531 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
7a3f542f VSO |
1532 | } |
1533 | if (pad->merge_reads && pad->tail) { | |
c834dc05 | 1534 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
7a3f542f VSO |
1535 | } |
1536 | ret = bdrv_aligned_preadv(child, req, req->overlap_offset, bytes, | |
65cd4424 | 1537 | align, &local_qiov, 0, 0); |
7a3f542f VSO |
1538 | if (ret < 0) { |
1539 | return ret; | |
1540 | } | |
1541 | if (pad->head) { | |
c834dc05 | 1542 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
7a3f542f VSO |
1543 | } |
1544 | if (pad->merge_reads && pad->tail) { | |
c834dc05 | 1545 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
7a3f542f VSO |
1546 | } |
1547 | ||
1548 | if (pad->merge_reads) { | |
1549 | goto zero_mem; | |
1550 | } | |
1551 | } | |
1552 | ||
1553 | if (pad->tail) { | |
1554 | qemu_iovec_init_buf(&local_qiov, pad->tail_buf, align); | |
1555 | ||
c834dc05 | 1556 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
7a3f542f VSO |
1557 | ret = bdrv_aligned_preadv( |
1558 | child, req, | |
1559 | req->overlap_offset + req->overlap_bytes - align, | |
65cd4424 | 1560 | align, align, &local_qiov, 0, 0); |
7a3f542f VSO |
1561 | if (ret < 0) { |
1562 | return ret; | |
1563 | } | |
c834dc05 | 1564 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
7a3f542f VSO |
1565 | } |
1566 | ||
1567 | zero_mem: | |
1568 | if (zero_middle) { | |
1569 | memset(pad->buf + pad->head, 0, pad->buf_len - pad->head - pad->tail); | |
1570 | } | |
1571 | ||
1572 | return 0; | |
1573 | } | |
1574 | ||
18743311 HC |
1575 | /** |
1576 | * Free *pad's associated buffers, and perform any necessary finalization steps. | |
1577 | */ | |
1578 | static void bdrv_padding_finalize(BdrvRequestPadding *pad) | |
7a3f542f | 1579 | { |
18743311 HC |
1580 | if (pad->collapse_bounce_buf) { |
1581 | if (!pad->write) { | |
1582 | /* | |
1583 | * If padding required elements in the vector to be collapsed into a | |
1584 | * bounce buffer, copy the bounce buffer content back | |
1585 | */ | |
1586 | qemu_iovec_from_buf(&pad->pre_collapse_qiov, 0, | |
1587 | pad->collapse_bounce_buf, pad->collapse_len); | |
1588 | } | |
1589 | qemu_vfree(pad->collapse_bounce_buf); | |
1590 | qemu_iovec_destroy(&pad->pre_collapse_qiov); | |
1591 | } | |
7a3f542f VSO |
1592 | if (pad->buf) { |
1593 | qemu_vfree(pad->buf); | |
1594 | qemu_iovec_destroy(&pad->local_qiov); | |
1595 | } | |
98ca4549 | 1596 | memset(pad, 0, sizeof(*pad)); |
7a3f542f VSO |
1597 | } |
1598 | ||
18743311 HC |
1599 | /* |
1600 | * Create pad->local_qiov by wrapping @iov in the padding head and tail, while | |
1601 | * ensuring that the resulting vector will not exceed IOV_MAX elements. | |
1602 | * | |
1603 | * To ensure this, when necessary, the first two or three elements of @iov are | |
1604 | * merged into pad->collapse_bounce_buf and replaced by a reference to that | |
1605 | * bounce buffer in pad->local_qiov. | |
1606 | * | |
1607 | * After performing a read request, the data from the bounce buffer must be | |
1608 | * copied back into pad->pre_collapse_qiov (e.g. by bdrv_padding_finalize()). | |
1609 | */ | |
1610 | static int bdrv_create_padded_qiov(BlockDriverState *bs, | |
1611 | BdrvRequestPadding *pad, | |
1612 | struct iovec *iov, int niov, | |
1613 | size_t iov_offset, size_t bytes) | |
1614 | { | |
1615 | int padded_niov, surplus_count, collapse_count; | |
1616 | ||
1617 | /* Assert this invariant */ | |
1618 | assert(niov <= IOV_MAX); | |
1619 | ||
1620 | /* | |
1621 | * Cannot pad if resulting length would exceed SIZE_MAX. Returning an error | |
1622 | * to the guest is not ideal, but there is little else we can do. At least | |
1623 | * this will practically never happen on 64-bit systems. | |
1624 | */ | |
1625 | if (SIZE_MAX - pad->head < bytes || | |
1626 | SIZE_MAX - pad->head - bytes < pad->tail) | |
1627 | { | |
1628 | return -EINVAL; | |
1629 | } | |
1630 | ||
1631 | /* Length of the resulting IOV if we just concatenated everything */ | |
1632 | padded_niov = !!pad->head + niov + !!pad->tail; | |
1633 | ||
1634 | qemu_iovec_init(&pad->local_qiov, MIN(padded_niov, IOV_MAX)); | |
1635 | ||
1636 | if (pad->head) { | |
1637 | qemu_iovec_add(&pad->local_qiov, pad->buf, pad->head); | |
1638 | } | |
1639 | ||
1640 | /* | |
1641 | * If padded_niov > IOV_MAX, we cannot just concatenate everything. | |
1642 | * Instead, merge the first two or three elements of @iov to reduce the | |
1643 | * number of vector elements as necessary. | |
1644 | */ | |
1645 | if (padded_niov > IOV_MAX) { | |
1646 | /* | |
1647 | * Only head and tail can have lead to the number of entries exceeding | |
1648 | * IOV_MAX, so we can exceed it by the head and tail at most. We need | |
1649 | * to reduce the number of elements by `surplus_count`, so we merge that | |
1650 | * many elements plus one into one element. | |
1651 | */ | |
1652 | surplus_count = padded_niov - IOV_MAX; | |
1653 | assert(surplus_count <= !!pad->head + !!pad->tail); | |
1654 | collapse_count = surplus_count + 1; | |
1655 | ||
1656 | /* | |
1657 | * Move the elements to collapse into `pad->pre_collapse_qiov`, then | |
1658 | * advance `iov` (and associated variables) by those elements. | |
1659 | */ | |
1660 | qemu_iovec_init(&pad->pre_collapse_qiov, collapse_count); | |
1661 | qemu_iovec_concat_iov(&pad->pre_collapse_qiov, iov, | |
1662 | collapse_count, iov_offset, SIZE_MAX); | |
1663 | iov += collapse_count; | |
1664 | iov_offset = 0; | |
1665 | niov -= collapse_count; | |
1666 | bytes -= pad->pre_collapse_qiov.size; | |
1667 | ||
1668 | /* | |
1669 | * Construct the bounce buffer to match the length of the to-collapse | |
1670 | * vector elements, and for write requests, initialize it with the data | |
1671 | * from those elements. Then add it to `pad->local_qiov`. | |
1672 | */ | |
1673 | pad->collapse_len = pad->pre_collapse_qiov.size; | |
1674 | pad->collapse_bounce_buf = qemu_blockalign(bs, pad->collapse_len); | |
1675 | if (pad->write) { | |
1676 | qemu_iovec_to_buf(&pad->pre_collapse_qiov, 0, | |
1677 | pad->collapse_bounce_buf, pad->collapse_len); | |
1678 | } | |
1679 | qemu_iovec_add(&pad->local_qiov, | |
1680 | pad->collapse_bounce_buf, pad->collapse_len); | |
1681 | } | |
1682 | ||
1683 | qemu_iovec_concat_iov(&pad->local_qiov, iov, niov, iov_offset, bytes); | |
1684 | ||
1685 | if (pad->tail) { | |
1686 | qemu_iovec_add(&pad->local_qiov, | |
1687 | pad->buf + pad->buf_len - pad->tail, pad->tail); | |
1688 | } | |
1689 | ||
1690 | assert(pad->local_qiov.niov == MIN(padded_niov, IOV_MAX)); | |
1691 | return 0; | |
1692 | } | |
1693 | ||
7a3f542f VSO |
1694 | /* |
1695 | * bdrv_pad_request | |
1696 | * | |
1697 | * Exchange request parameters with padded request if needed. Don't include RMW | |
1698 | * read of padding, bdrv_padding_rmw_read() should be called separately if | |
1699 | * needed. | |
1700 | * | |
18743311 HC |
1701 | * @write is true for write requests, false for read requests. |
1702 | * | |
98ca4549 VSO |
1703 | * Request parameters (@qiov, &qiov_offset, &offset, &bytes) are in-out: |
1704 | * - on function start they represent original request | |
1705 | * - on failure or when padding is not needed they are unchanged | |
1706 | * - on success when padding is needed they represent padded request | |
61007b31 | 1707 | */ |
98ca4549 VSO |
1708 | static int bdrv_pad_request(BlockDriverState *bs, |
1709 | QEMUIOVector **qiov, size_t *qiov_offset, | |
37e9403e | 1710 | int64_t *offset, int64_t *bytes, |
18743311 | 1711 | bool write, |
e8b65355 SH |
1712 | BdrvRequestPadding *pad, bool *padded, |
1713 | BdrvRequestFlags *flags) | |
7a3f542f | 1714 | { |
4c002cef | 1715 | int ret; |
18743311 HC |
1716 | struct iovec *sliced_iov; |
1717 | int sliced_niov; | |
1718 | size_t sliced_head, sliced_tail; | |
4c002cef | 1719 | |
ef256751 HC |
1720 | /* Should have been checked by the caller already */ |
1721 | ret = bdrv_check_request32(*offset, *bytes, *qiov, *qiov_offset); | |
1722 | if (ret < 0) { | |
1723 | return ret; | |
1724 | } | |
37e9403e | 1725 | |
18743311 | 1726 | if (!bdrv_init_padding(bs, *offset, *bytes, write, pad)) { |
98ca4549 VSO |
1727 | if (padded) { |
1728 | *padded = false; | |
1729 | } | |
1730 | return 0; | |
7a3f542f VSO |
1731 | } |
1732 | ||
18743311 HC |
1733 | sliced_iov = qemu_iovec_slice(*qiov, *qiov_offset, *bytes, |
1734 | &sliced_head, &sliced_tail, | |
1735 | &sliced_niov); | |
1736 | ||
ef256751 | 1737 | /* Guaranteed by bdrv_check_request32() */ |
18743311 HC |
1738 | assert(*bytes <= SIZE_MAX); |
1739 | ret = bdrv_create_padded_qiov(bs, pad, sliced_iov, sliced_niov, | |
1740 | sliced_head, *bytes); | |
98ca4549 | 1741 | if (ret < 0) { |
18743311 | 1742 | bdrv_padding_finalize(pad); |
98ca4549 VSO |
1743 | return ret; |
1744 | } | |
7a3f542f VSO |
1745 | *bytes += pad->head + pad->tail; |
1746 | *offset -= pad->head; | |
1747 | *qiov = &pad->local_qiov; | |
1acc3466 | 1748 | *qiov_offset = 0; |
98ca4549 VSO |
1749 | if (padded) { |
1750 | *padded = true; | |
1751 | } | |
e8b65355 SH |
1752 | if (flags) { |
1753 | /* Can't use optimization hint with bounce buffer */ | |
1754 | *flags &= ~BDRV_REQ_REGISTERED_BUF; | |
1755 | } | |
7a3f542f | 1756 | |
98ca4549 | 1757 | return 0; |
7a3f542f VSO |
1758 | } |
1759 | ||
a03ef88f | 1760 | int coroutine_fn bdrv_co_preadv(BdrvChild *child, |
e9e52efd | 1761 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, |
61007b31 | 1762 | BdrvRequestFlags flags) |
1acc3466 | 1763 | { |
967d7905 | 1764 | IO_CODE(); |
1acc3466 VSO |
1765 | return bdrv_co_preadv_part(child, offset, bytes, qiov, 0, flags); |
1766 | } | |
1767 | ||
1768 | int coroutine_fn bdrv_co_preadv_part(BdrvChild *child, | |
37e9403e | 1769 | int64_t offset, int64_t bytes, |
1acc3466 VSO |
1770 | QEMUIOVector *qiov, size_t qiov_offset, |
1771 | BdrvRequestFlags flags) | |
61007b31 | 1772 | { |
a03ef88f | 1773 | BlockDriverState *bs = child->bs; |
61007b31 | 1774 | BdrvTrackedRequest req; |
7a3f542f | 1775 | BdrvRequestPadding pad; |
61007b31 | 1776 | int ret; |
967d7905 | 1777 | IO_CODE(); |
61007b31 | 1778 | |
37e9403e | 1779 | trace_bdrv_co_preadv_part(bs, offset, bytes, flags); |
61007b31 | 1780 | |
1e97be91 | 1781 | if (!bdrv_co_is_inserted(bs)) { |
f4dad307 VSO |
1782 | return -ENOMEDIUM; |
1783 | } | |
1784 | ||
63f4ad11 | 1785 | ret = bdrv_check_request32(offset, bytes, qiov, qiov_offset); |
61007b31 SH |
1786 | if (ret < 0) { |
1787 | return ret; | |
1788 | } | |
1789 | ||
ac9d00bf VSO |
1790 | if (bytes == 0 && !QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)) { |
1791 | /* | |
1792 | * Aligning zero request is nonsense. Even if driver has special meaning | |
1793 | * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass | |
1794 | * it to driver due to request_alignment. | |
1795 | * | |
1796 | * Still, no reason to return an error if someone do unaligned | |
1797 | * zero-length read occasionally. | |
1798 | */ | |
1799 | return 0; | |
1800 | } | |
1801 | ||
99723548 PB |
1802 | bdrv_inc_in_flight(bs); |
1803 | ||
9568b511 | 1804 | /* Don't do copy-on-read if we read data before write operation */ |
d73415a3 | 1805 | if (qatomic_read(&bs->copy_on_read)) { |
61007b31 SH |
1806 | flags |= BDRV_REQ_COPY_ON_READ; |
1807 | } | |
1808 | ||
18743311 HC |
1809 | ret = bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, false, |
1810 | &pad, NULL, &flags); | |
98ca4549 | 1811 | if (ret < 0) { |
87ab8802 | 1812 | goto fail; |
98ca4549 | 1813 | } |
61007b31 | 1814 | |
ebde595c | 1815 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
7a3f542f VSO |
1816 | ret = bdrv_aligned_preadv(child, &req, offset, bytes, |
1817 | bs->bl.request_alignment, | |
1acc3466 | 1818 | qiov, qiov_offset, flags); |
61007b31 | 1819 | tracked_request_end(&req); |
18743311 | 1820 | bdrv_padding_finalize(&pad); |
61007b31 | 1821 | |
87ab8802 KW |
1822 | fail: |
1823 | bdrv_dec_in_flight(bs); | |
1824 | ||
61007b31 SH |
1825 | return ret; |
1826 | } | |
1827 | ||
eeb47775 KW |
1828 | static int coroutine_fn GRAPH_RDLOCK |
1829 | bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1830 | BdrvRequestFlags flags) | |
61007b31 SH |
1831 | { |
1832 | BlockDriver *drv = bs->drv; | |
1833 | QEMUIOVector qiov; | |
0d93ed08 | 1834 | void *buf = NULL; |
61007b31 | 1835 | int ret = 0; |
465fe887 | 1836 | bool need_flush = false; |
443668ca DL |
1837 | int head = 0; |
1838 | int tail = 0; | |
61007b31 | 1839 | |
2aaa3f9b VSO |
1840 | int64_t max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, |
1841 | INT64_MAX); | |
a5b8dd2c EB |
1842 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1843 | bs->bl.request_alignment); | |
cb2e2878 | 1844 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER); |
d05aa8bb | 1845 | |
abaf8b75 | 1846 | assert_bdrv_graph_readable(); |
5ae07b14 VSO |
1847 | bdrv_check_request(offset, bytes, &error_abort); |
1848 | ||
d470ad42 HR |
1849 | if (!drv) { |
1850 | return -ENOMEDIUM; | |
1851 | } | |
1852 | ||
fe0480d6 KW |
1853 | if ((flags & ~bs->supported_zero_flags) & BDRV_REQ_NO_FALLBACK) { |
1854 | return -ENOTSUP; | |
1855 | } | |
1856 | ||
e8b65355 SH |
1857 | /* By definition there is no user buffer so this flag doesn't make sense */ |
1858 | if (flags & BDRV_REQ_REGISTERED_BUF) { | |
1859 | return -EINVAL; | |
1860 | } | |
1861 | ||
0bc329fb HR |
1862 | /* Invalidate the cached block-status data range if this write overlaps */ |
1863 | bdrv_bsc_invalidate_range(bs, offset, bytes); | |
1864 | ||
b8d0a980 EB |
1865 | assert(alignment % bs->bl.request_alignment == 0); |
1866 | head = offset % alignment; | |
f5a5ca79 | 1867 | tail = (offset + bytes) % alignment; |
b8d0a980 EB |
1868 | max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment); |
1869 | assert(max_write_zeroes >= bs->bl.request_alignment); | |
61007b31 | 1870 | |
f5a5ca79 | 1871 | while (bytes > 0 && !ret) { |
5ae07b14 | 1872 | int64_t num = bytes; |
61007b31 SH |
1873 | |
1874 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1875 | * to be aligned, and that unaligned requests do not cross cluster |
1876 | * boundaries. | |
61007b31 | 1877 | */ |
443668ca | 1878 | if (head) { |
b2f95fee EB |
1879 | /* Make a small request up to the first aligned sector. For |
1880 | * convenience, limit this request to max_transfer even if | |
1881 | * we don't need to fall back to writes. */ | |
f5a5ca79 | 1882 | num = MIN(MIN(bytes, max_transfer), alignment - head); |
b2f95fee EB |
1883 | head = (head + num) % alignment; |
1884 | assert(num < max_write_zeroes); | |
d05aa8bb | 1885 | } else if (tail && num > alignment) { |
443668ca DL |
1886 | /* Shorten the request to the last aligned sector. */ |
1887 | num -= tail; | |
61007b31 SH |
1888 | } |
1889 | ||
1890 | /* limit request size */ | |
1891 | if (num > max_write_zeroes) { | |
1892 | num = max_write_zeroes; | |
1893 | } | |
1894 | ||
1895 | ret = -ENOTSUP; | |
1896 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1897 | if (drv->bdrv_co_pwrite_zeroes) { |
1898 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1899 | flags & bs->supported_zero_flags); | |
1900 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1901 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1902 | need_flush = true; | |
1903 | } | |
465fe887 EB |
1904 | } else { |
1905 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1906 | } |
1907 | ||
294682cc | 1908 | if (ret == -ENOTSUP && !(flags & BDRV_REQ_NO_FALLBACK)) { |
61007b31 | 1909 | /* Fall back to bounce buffer if write zeroes is unsupported */ |
465fe887 EB |
1910 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1911 | ||
1912 | if ((flags & BDRV_REQ_FUA) && | |
1913 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1914 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1915 | * flush on each chunk; use just one at the end */ | |
1916 | write_flags &= ~BDRV_REQ_FUA; | |
1917 | need_flush = true; | |
1918 | } | |
5def6b80 | 1919 | num = MIN(num, max_transfer); |
0d93ed08 VSO |
1920 | if (buf == NULL) { |
1921 | buf = qemu_try_blockalign0(bs, num); | |
1922 | if (buf == NULL) { | |
61007b31 SH |
1923 | ret = -ENOMEM; |
1924 | goto fail; | |
1925 | } | |
61007b31 | 1926 | } |
0d93ed08 | 1927 | qemu_iovec_init_buf(&qiov, buf, num); |
61007b31 | 1928 | |
ac850bf0 | 1929 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, 0, write_flags); |
61007b31 SH |
1930 | |
1931 | /* Keep bounce buffer around if it is big enough for all | |
1932 | * all future requests. | |
1933 | */ | |
5def6b80 | 1934 | if (num < max_transfer) { |
0d93ed08 VSO |
1935 | qemu_vfree(buf); |
1936 | buf = NULL; | |
61007b31 SH |
1937 | } |
1938 | } | |
1939 | ||
d05aa8bb | 1940 | offset += num; |
f5a5ca79 | 1941 | bytes -= num; |
61007b31 SH |
1942 | } |
1943 | ||
1944 | fail: | |
465fe887 EB |
1945 | if (ret == 0 && need_flush) { |
1946 | ret = bdrv_co_flush(bs); | |
1947 | } | |
0d93ed08 | 1948 | qemu_vfree(buf); |
61007b31 SH |
1949 | return ret; |
1950 | } | |
1951 | ||
a00e70c0 | 1952 | static inline int coroutine_fn GRAPH_RDLOCK |
fcfd9ade | 1953 | bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, int64_t bytes, |
85fe2479 FZ |
1954 | BdrvTrackedRequest *req, int flags) |
1955 | { | |
1956 | BlockDriverState *bs = child->bs; | |
fcfd9ade VSO |
1957 | |
1958 | bdrv_check_request(offset, bytes, &error_abort); | |
85fe2479 | 1959 | |
307261b2 | 1960 | if (bdrv_is_read_only(bs)) { |
85fe2479 FZ |
1961 | return -EPERM; |
1962 | } | |
1963 | ||
85fe2479 FZ |
1964 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
1965 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); | |
1966 | assert(!(flags & ~BDRV_REQ_MASK)); | |
d1a764d1 | 1967 | assert(!((flags & BDRV_REQ_NO_WAIT) && !(flags & BDRV_REQ_SERIALISING))); |
85fe2479 FZ |
1968 | |
1969 | if (flags & BDRV_REQ_SERIALISING) { | |
d1a764d1 VSO |
1970 | QEMU_LOCK_GUARD(&bs->reqs_lock); |
1971 | ||
1972 | tracked_request_set_serialising(req, bdrv_get_cluster_size(bs)); | |
1973 | ||
1974 | if ((flags & BDRV_REQ_NO_WAIT) && bdrv_find_conflicting_request(req)) { | |
1975 | return -EBUSY; | |
1976 | } | |
1977 | ||
1978 | bdrv_wait_serialising_requests_locked(req); | |
18fbd0de PB |
1979 | } else { |
1980 | bdrv_wait_serialising_requests(req); | |
85fe2479 FZ |
1981 | } |
1982 | ||
85fe2479 FZ |
1983 | assert(req->overlap_offset <= offset); |
1984 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
fcfd9ade VSO |
1985 | assert(offset + bytes <= bs->total_sectors * BDRV_SECTOR_SIZE || |
1986 | child->perm & BLK_PERM_RESIZE); | |
85fe2479 | 1987 | |
cd47d792 FZ |
1988 | switch (req->type) { |
1989 | case BDRV_TRACKED_WRITE: | |
1990 | case BDRV_TRACKED_DISCARD: | |
1991 | if (flags & BDRV_REQ_WRITE_UNCHANGED) { | |
1992 | assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
1993 | } else { | |
1994 | assert(child->perm & BLK_PERM_WRITE); | |
1995 | } | |
94783301 VSO |
1996 | bdrv_write_threshold_check_write(bs, offset, bytes); |
1997 | return 0; | |
cd47d792 FZ |
1998 | case BDRV_TRACKED_TRUNCATE: |
1999 | assert(child->perm & BLK_PERM_RESIZE); | |
2000 | return 0; | |
2001 | default: | |
2002 | abort(); | |
85fe2479 | 2003 | } |
85fe2479 FZ |
2004 | } |
2005 | ||
2006 | static inline void coroutine_fn | |
fcfd9ade | 2007 | bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, int64_t bytes, |
85fe2479 FZ |
2008 | BdrvTrackedRequest *req, int ret) |
2009 | { | |
2010 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
2011 | BlockDriverState *bs = child->bs; | |
2012 | ||
fcfd9ade VSO |
2013 | bdrv_check_request(offset, bytes, &error_abort); |
2014 | ||
d73415a3 | 2015 | qatomic_inc(&bs->write_gen); |
85fe2479 | 2016 | |
00695c27 FZ |
2017 | /* |
2018 | * Discard cannot extend the image, but in error handling cases, such as | |
2019 | * when reverting a qcow2 cluster allocation, the discarded range can pass | |
2020 | * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD | |
2021 | * here. Instead, just skip it, since semantically a discard request | |
2022 | * beyond EOF cannot expand the image anyway. | |
2023 | */ | |
7f8f03ef | 2024 | if (ret == 0 && |
cd47d792 FZ |
2025 | (req->type == BDRV_TRACKED_TRUNCATE || |
2026 | end_sector > bs->total_sectors) && | |
2027 | req->type != BDRV_TRACKED_DISCARD) { | |
7f8f03ef FZ |
2028 | bs->total_sectors = end_sector; |
2029 | bdrv_parent_cb_resize(bs); | |
2030 | bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS); | |
85fe2479 | 2031 | } |
00695c27 FZ |
2032 | if (req->bytes) { |
2033 | switch (req->type) { | |
2034 | case BDRV_TRACKED_WRITE: | |
2035 | stat64_max(&bs->wr_highest_offset, offset + bytes); | |
2036 | /* fall through, to set dirty bits */ | |
2037 | case BDRV_TRACKED_DISCARD: | |
2038 | bdrv_set_dirty(bs, offset, bytes); | |
2039 | break; | |
2040 | default: | |
2041 | break; | |
2042 | } | |
2043 | } | |
85fe2479 FZ |
2044 | } |
2045 | ||
61007b31 | 2046 | /* |
04ed95f4 EB |
2047 | * Forwards an already correctly aligned write request to the BlockDriver, |
2048 | * after possibly fragmenting it. | |
61007b31 | 2049 | */ |
7b1fb72e KW |
2050 | static int coroutine_fn GRAPH_RDLOCK |
2051 | bdrv_aligned_pwritev(BdrvChild *child, BdrvTrackedRequest *req, | |
2052 | int64_t offset, int64_t bytes, int64_t align, | |
2053 | QEMUIOVector *qiov, size_t qiov_offset, | |
2054 | BdrvRequestFlags flags) | |
61007b31 | 2055 | { |
85c97ca7 | 2056 | BlockDriverState *bs = child->bs; |
61007b31 | 2057 | BlockDriver *drv = bs->drv; |
61007b31 SH |
2058 | int ret; |
2059 | ||
fcfd9ade | 2060 | int64_t bytes_remaining = bytes; |
04ed95f4 | 2061 | int max_transfer; |
61007b31 | 2062 | |
fcfd9ade VSO |
2063 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
2064 | ||
d470ad42 HR |
2065 | if (!drv) { |
2066 | return -ENOMEDIUM; | |
2067 | } | |
2068 | ||
d6883bc9 VSO |
2069 | if (bdrv_has_readonly_bitmaps(bs)) { |
2070 | return -EPERM; | |
2071 | } | |
2072 | ||
cff86b38 EB |
2073 | assert(is_power_of_2(align)); |
2074 | assert((offset & (align - 1)) == 0); | |
2075 | assert((bytes & (align - 1)) == 0); | |
04ed95f4 EB |
2076 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
2077 | align); | |
61007b31 | 2078 | |
85fe2479 | 2079 | ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags); |
61007b31 SH |
2080 | |
2081 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 2082 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
28c4da28 | 2083 | qemu_iovec_is_zero(qiov, qiov_offset, bytes)) { |
61007b31 SH |
2084 | flags |= BDRV_REQ_ZERO_WRITE; |
2085 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
2086 | flags |= BDRV_REQ_MAY_UNMAP; | |
2087 | } | |
3c586715 SH |
2088 | |
2089 | /* Can't use optimization hint with bufferless zero write */ | |
2090 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
61007b31 SH |
2091 | } |
2092 | ||
2093 | if (ret < 0) { | |
2094 | /* Do nothing, write notifier decided to fail this request */ | |
2095 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
c834dc05 | 2096 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 2097 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
3ea1a091 | 2098 | } else if (flags & BDRV_REQ_WRITE_COMPRESSED) { |
28c4da28 VSO |
2099 | ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, |
2100 | qiov, qiov_offset); | |
04ed95f4 | 2101 | } else if (bytes <= max_transfer) { |
c834dc05 | 2102 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV); |
28c4da28 | 2103 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, qiov_offset, flags); |
04ed95f4 | 2104 | } else { |
c834dc05 | 2105 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV); |
04ed95f4 EB |
2106 | while (bytes_remaining) { |
2107 | int num = MIN(bytes_remaining, max_transfer); | |
04ed95f4 EB |
2108 | int local_flags = flags; |
2109 | ||
2110 | assert(num); | |
2111 | if (num < bytes_remaining && (flags & BDRV_REQ_FUA) && | |
2112 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
2113 | /* If FUA is going to be emulated by flush, we only | |
2114 | * need to flush on the last iteration */ | |
2115 | local_flags &= ~BDRV_REQ_FUA; | |
2116 | } | |
04ed95f4 EB |
2117 | |
2118 | ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining, | |
134b7dec HR |
2119 | num, qiov, |
2120 | qiov_offset + bytes - bytes_remaining, | |
28c4da28 | 2121 | local_flags); |
04ed95f4 EB |
2122 | if (ret < 0) { |
2123 | break; | |
2124 | } | |
2125 | bytes_remaining -= num; | |
2126 | } | |
61007b31 | 2127 | } |
c834dc05 | 2128 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 2129 | |
61007b31 | 2130 | if (ret >= 0) { |
04ed95f4 | 2131 | ret = 0; |
61007b31 | 2132 | } |
85fe2479 | 2133 | bdrv_co_write_req_finish(child, offset, bytes, req, ret); |
61007b31 SH |
2134 | |
2135 | return ret; | |
2136 | } | |
2137 | ||
7b1fb72e KW |
2138 | static int coroutine_fn GRAPH_RDLOCK |
2139 | bdrv_co_do_zero_pwritev(BdrvChild *child, int64_t offset, int64_t bytes, | |
2140 | BdrvRequestFlags flags, BdrvTrackedRequest *req) | |
9eeb6dd1 | 2141 | { |
85c97ca7 | 2142 | BlockDriverState *bs = child->bs; |
9eeb6dd1 | 2143 | QEMUIOVector local_qiov; |
a5b8dd2c | 2144 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 | 2145 | int ret = 0; |
7a3f542f VSO |
2146 | bool padding; |
2147 | BdrvRequestPadding pad; | |
9eeb6dd1 | 2148 | |
e8b65355 SH |
2149 | /* This flag doesn't make sense for padding or zero writes */ |
2150 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
2151 | ||
18743311 | 2152 | padding = bdrv_init_padding(bs, offset, bytes, true, &pad); |
7a3f542f | 2153 | if (padding) { |
45e62b46 | 2154 | assert(!(flags & BDRV_REQ_NO_WAIT)); |
8ac5aab2 | 2155 | bdrv_make_request_serialising(req, align); |
9eeb6dd1 | 2156 | |
7a3f542f VSO |
2157 | bdrv_padding_rmw_read(child, req, &pad, true); |
2158 | ||
2159 | if (pad.head || pad.merge_reads) { | |
2160 | int64_t aligned_offset = offset & ~(align - 1); | |
2161 | int64_t write_bytes = pad.merge_reads ? pad.buf_len : align; | |
2162 | ||
2163 | qemu_iovec_init_buf(&local_qiov, pad.buf, write_bytes); | |
2164 | ret = bdrv_aligned_pwritev(child, req, aligned_offset, write_bytes, | |
28c4da28 | 2165 | align, &local_qiov, 0, |
7a3f542f VSO |
2166 | flags & ~BDRV_REQ_ZERO_WRITE); |
2167 | if (ret < 0 || pad.merge_reads) { | |
2168 | /* Error or all work is done */ | |
2169 | goto out; | |
2170 | } | |
2171 | offset += write_bytes - pad.head; | |
2172 | bytes -= write_bytes - pad.head; | |
9eeb6dd1 | 2173 | } |
9eeb6dd1 FZ |
2174 | } |
2175 | ||
2176 | assert(!bytes || (offset & (align - 1)) == 0); | |
2177 | if (bytes >= align) { | |
2178 | /* Write the aligned part in the middle. */ | |
fcfd9ade | 2179 | int64_t aligned_bytes = bytes & ~(align - 1); |
85c97ca7 | 2180 | ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align, |
28c4da28 | 2181 | NULL, 0, flags); |
9eeb6dd1 | 2182 | if (ret < 0) { |
7a3f542f | 2183 | goto out; |
9eeb6dd1 FZ |
2184 | } |
2185 | bytes -= aligned_bytes; | |
2186 | offset += aligned_bytes; | |
2187 | } | |
2188 | ||
2189 | assert(!bytes || (offset & (align - 1)) == 0); | |
2190 | if (bytes) { | |
7a3f542f | 2191 | assert(align == pad.tail + bytes); |
9eeb6dd1 | 2192 | |
7a3f542f | 2193 | qemu_iovec_init_buf(&local_qiov, pad.tail_buf, align); |
85c97ca7 | 2194 | ret = bdrv_aligned_pwritev(child, req, offset, align, align, |
28c4da28 VSO |
2195 | &local_qiov, 0, |
2196 | flags & ~BDRV_REQ_ZERO_WRITE); | |
9eeb6dd1 | 2197 | } |
9eeb6dd1 | 2198 | |
7a3f542f | 2199 | out: |
18743311 | 2200 | bdrv_padding_finalize(&pad); |
7a3f542f VSO |
2201 | |
2202 | return ret; | |
9eeb6dd1 FZ |
2203 | } |
2204 | ||
61007b31 SH |
2205 | /* |
2206 | * Handle a write request in coroutine context | |
2207 | */ | |
a03ef88f | 2208 | int coroutine_fn bdrv_co_pwritev(BdrvChild *child, |
e9e52efd | 2209 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, |
61007b31 | 2210 | BdrvRequestFlags flags) |
1acc3466 | 2211 | { |
967d7905 | 2212 | IO_CODE(); |
1acc3466 VSO |
2213 | return bdrv_co_pwritev_part(child, offset, bytes, qiov, 0, flags); |
2214 | } | |
2215 | ||
2216 | int coroutine_fn bdrv_co_pwritev_part(BdrvChild *child, | |
37e9403e | 2217 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, size_t qiov_offset, |
1acc3466 | 2218 | BdrvRequestFlags flags) |
61007b31 | 2219 | { |
a03ef88f | 2220 | BlockDriverState *bs = child->bs; |
61007b31 | 2221 | BdrvTrackedRequest req; |
a5b8dd2c | 2222 | uint64_t align = bs->bl.request_alignment; |
7a3f542f | 2223 | BdrvRequestPadding pad; |
61007b31 | 2224 | int ret; |
f0deecff | 2225 | bool padded = false; |
967d7905 | 2226 | IO_CODE(); |
61007b31 | 2227 | |
37e9403e | 2228 | trace_bdrv_co_pwritev_part(child->bs, offset, bytes, flags); |
f42cf447 | 2229 | |
1e97be91 | 2230 | if (!bdrv_co_is_inserted(bs)) { |
61007b31 SH |
2231 | return -ENOMEDIUM; |
2232 | } | |
61007b31 | 2233 | |
2aaa3f9b VSO |
2234 | if (flags & BDRV_REQ_ZERO_WRITE) { |
2235 | ret = bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, NULL); | |
2236 | } else { | |
2237 | ret = bdrv_check_request32(offset, bytes, qiov, qiov_offset); | |
2238 | } | |
61007b31 SH |
2239 | if (ret < 0) { |
2240 | return ret; | |
2241 | } | |
2242 | ||
f2208fdc AG |
2243 | /* If the request is misaligned then we can't make it efficient */ |
2244 | if ((flags & BDRV_REQ_NO_FALLBACK) && | |
2245 | !QEMU_IS_ALIGNED(offset | bytes, align)) | |
2246 | { | |
2247 | return -ENOTSUP; | |
2248 | } | |
2249 | ||
ac9d00bf VSO |
2250 | if (bytes == 0 && !QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)) { |
2251 | /* | |
2252 | * Aligning zero request is nonsense. Even if driver has special meaning | |
2253 | * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass | |
2254 | * it to driver due to request_alignment. | |
2255 | * | |
2256 | * Still, no reason to return an error if someone do unaligned | |
2257 | * zero-length write occasionally. | |
2258 | */ | |
2259 | return 0; | |
2260 | } | |
2261 | ||
f0deecff VSO |
2262 | if (!(flags & BDRV_REQ_ZERO_WRITE)) { |
2263 | /* | |
2264 | * Pad request for following read-modify-write cycle. | |
2265 | * bdrv_co_do_zero_pwritev() does aligning by itself, so, we do | |
2266 | * alignment only if there is no ZERO flag. | |
2267 | */ | |
18743311 HC |
2268 | ret = bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, true, |
2269 | &pad, &padded, &flags); | |
98ca4549 VSO |
2270 | if (ret < 0) { |
2271 | return ret; | |
2272 | } | |
f0deecff VSO |
2273 | } |
2274 | ||
99723548 | 2275 | bdrv_inc_in_flight(bs); |
ebde595c | 2276 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 2277 | |
18a59f03 | 2278 | if (flags & BDRV_REQ_ZERO_WRITE) { |
f0deecff | 2279 | assert(!padded); |
85c97ca7 | 2280 | ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req); |
9eeb6dd1 FZ |
2281 | goto out; |
2282 | } | |
2283 | ||
f0deecff VSO |
2284 | if (padded) { |
2285 | /* | |
2286 | * Request was unaligned to request_alignment and therefore | |
2287 | * padded. We are going to do read-modify-write, and must | |
2288 | * serialize the request to prevent interactions of the | |
2289 | * widened region with other transactions. | |
2290 | */ | |
45e62b46 | 2291 | assert(!(flags & BDRV_REQ_NO_WAIT)); |
8ac5aab2 | 2292 | bdrv_make_request_serialising(&req, align); |
7a3f542f | 2293 | bdrv_padding_rmw_read(child, &req, &pad, false); |
61007b31 SH |
2294 | } |
2295 | ||
85c97ca7 | 2296 | ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align, |
1acc3466 | 2297 | qiov, qiov_offset, flags); |
61007b31 | 2298 | |
18743311 | 2299 | bdrv_padding_finalize(&pad); |
61007b31 | 2300 | |
9eeb6dd1 FZ |
2301 | out: |
2302 | tracked_request_end(&req); | |
99723548 | 2303 | bdrv_dec_in_flight(bs); |
7a3f542f | 2304 | |
61007b31 SH |
2305 | return ret; |
2306 | } | |
2307 | ||
a03ef88f | 2308 | int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, |
e9e52efd | 2309 | int64_t bytes, BdrvRequestFlags flags) |
61007b31 | 2310 | { |
384a48fb | 2311 | IO_CODE(); |
f5a5ca79 | 2312 | trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags); |
abaf8b75 | 2313 | assert_bdrv_graph_readable(); |
61007b31 | 2314 | |
a03ef88f | 2315 | if (!(child->bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
2316 | flags &= ~BDRV_REQ_MAY_UNMAP; |
2317 | } | |
61007b31 | 2318 | |
f5a5ca79 | 2319 | return bdrv_co_pwritev(child, offset, bytes, NULL, |
74021bc4 | 2320 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
2321 | } |
2322 | ||
4085f5c7 JS |
2323 | /* |
2324 | * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not. | |
2325 | */ | |
2326 | int bdrv_flush_all(void) | |
2327 | { | |
2328 | BdrvNextIterator it; | |
2329 | BlockDriverState *bs = NULL; | |
2330 | int result = 0; | |
2331 | ||
f791bf7f EGE |
2332 | GLOBAL_STATE_CODE(); |
2333 | ||
c8aa7895 PD |
2334 | /* |
2335 | * bdrv queue is managed by record/replay, | |
2336 | * creating new flush request for stopping | |
2337 | * the VM may break the determinism | |
2338 | */ | |
2339 | if (replay_events_enabled()) { | |
2340 | return result; | |
2341 | } | |
2342 | ||
4085f5c7 JS |
2343 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
2344 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2345 | int ret; | |
2346 | ||
2347 | aio_context_acquire(aio_context); | |
2348 | ret = bdrv_flush(bs); | |
2349 | if (ret < 0 && !result) { | |
2350 | result = ret; | |
2351 | } | |
2352 | aio_context_release(aio_context); | |
2353 | } | |
2354 | ||
2355 | return result; | |
2356 | } | |
2357 | ||
61007b31 SH |
2358 | /* |
2359 | * Returns the allocation status of the specified sectors. | |
2360 | * Drivers not implementing the functionality are assumed to not support | |
2361 | * backing files, hence all their sectors are reported as allocated. | |
2362 | * | |
86a3d5c6 EB |
2363 | * If 'want_zero' is true, the caller is querying for mapping |
2364 | * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and | |
2365 | * _ZERO where possible; otherwise, the result favors larger 'pnum', | |
2366 | * with a focus on accurate BDRV_BLOCK_ALLOCATED. | |
c9ce8c4d | 2367 | * |
2e8bc787 | 2368 | * If 'offset' is beyond the end of the disk image the return value is |
fb0d8654 | 2369 | * BDRV_BLOCK_EOF and 'pnum' is set to 0. |
61007b31 | 2370 | * |
2e8bc787 | 2371 | * 'bytes' is the max value 'pnum' should be set to. If bytes goes |
fb0d8654 EB |
2372 | * beyond the end of the disk image it will be clamped; if 'pnum' is set to |
2373 | * the end of the image, then the returned value will include BDRV_BLOCK_EOF. | |
67a0fd2a | 2374 | * |
2e8bc787 EB |
2375 | * 'pnum' is set to the number of bytes (including and immediately |
2376 | * following the specified offset) that are easily known to be in the | |
2377 | * same allocated/unallocated state. Note that a second call starting | |
2378 | * at the original offset plus returned pnum may have the same status. | |
2379 | * The returned value is non-zero on success except at end-of-file. | |
2380 | * | |
2381 | * Returns negative errno on failure. Otherwise, if the | |
2382 | * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are | |
2383 | * set to the host mapping and BDS corresponding to the guest offset. | |
61007b31 | 2384 | */ |
7ff9579e KW |
2385 | static int coroutine_fn GRAPH_RDLOCK |
2386 | bdrv_co_block_status(BlockDriverState *bs, bool want_zero, | |
2387 | int64_t offset, int64_t bytes, | |
2388 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
2e8bc787 EB |
2389 | { |
2390 | int64_t total_size; | |
2391 | int64_t n; /* bytes */ | |
efa6e2ed | 2392 | int ret; |
2e8bc787 | 2393 | int64_t local_map = 0; |
298a1665 | 2394 | BlockDriverState *local_file = NULL; |
efa6e2ed EB |
2395 | int64_t aligned_offset, aligned_bytes; |
2396 | uint32_t align; | |
549ec0d9 | 2397 | bool has_filtered_child; |
61007b31 | 2398 | |
298a1665 | 2399 | assert(pnum); |
7ff9579e | 2400 | assert_bdrv_graph_readable(); |
298a1665 | 2401 | *pnum = 0; |
0af02bd1 | 2402 | total_size = bdrv_co_getlength(bs); |
2e8bc787 EB |
2403 | if (total_size < 0) { |
2404 | ret = total_size; | |
298a1665 | 2405 | goto early_out; |
61007b31 SH |
2406 | } |
2407 | ||
2e8bc787 | 2408 | if (offset >= total_size) { |
298a1665 EB |
2409 | ret = BDRV_BLOCK_EOF; |
2410 | goto early_out; | |
61007b31 | 2411 | } |
2e8bc787 | 2412 | if (!bytes) { |
298a1665 EB |
2413 | ret = 0; |
2414 | goto early_out; | |
9cdcfd9f | 2415 | } |
61007b31 | 2416 | |
2e8bc787 EB |
2417 | n = total_size - offset; |
2418 | if (n < bytes) { | |
2419 | bytes = n; | |
61007b31 SH |
2420 | } |
2421 | ||
0af02bd1 | 2422 | /* Must be non-NULL or bdrv_co_getlength() would have failed */ |
d470ad42 | 2423 | assert(bs->drv); |
549ec0d9 HR |
2424 | has_filtered_child = bdrv_filter_child(bs); |
2425 | if (!bs->drv->bdrv_co_block_status && !has_filtered_child) { | |
2e8bc787 | 2426 | *pnum = bytes; |
61007b31 | 2427 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; |
2e8bc787 | 2428 | if (offset + bytes == total_size) { |
fb0d8654 EB |
2429 | ret |= BDRV_BLOCK_EOF; |
2430 | } | |
61007b31 | 2431 | if (bs->drv->protocol_name) { |
2e8bc787 EB |
2432 | ret |= BDRV_BLOCK_OFFSET_VALID; |
2433 | local_map = offset; | |
298a1665 | 2434 | local_file = bs; |
61007b31 | 2435 | } |
298a1665 | 2436 | goto early_out; |
61007b31 SH |
2437 | } |
2438 | ||
99723548 | 2439 | bdrv_inc_in_flight(bs); |
efa6e2ed EB |
2440 | |
2441 | /* Round out to request_alignment boundaries */ | |
86a3d5c6 | 2442 | align = bs->bl.request_alignment; |
efa6e2ed EB |
2443 | aligned_offset = QEMU_ALIGN_DOWN(offset, align); |
2444 | aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset; | |
2445 | ||
549ec0d9 | 2446 | if (bs->drv->bdrv_co_block_status) { |
0bc329fb HR |
2447 | /* |
2448 | * Use the block-status cache only for protocol nodes: Format | |
2449 | * drivers are generally quick to inquire the status, but protocol | |
2450 | * drivers often need to get information from outside of qemu, so | |
2451 | * we do not have control over the actual implementation. There | |
2452 | * have been cases where inquiring the status took an unreasonably | |
2453 | * long time, and we can do nothing in qemu to fix it. | |
2454 | * This is especially problematic for images with large data areas, | |
2455 | * because finding the few holes in them and giving them special | |
2456 | * treatment does not gain much performance. Therefore, we try to | |
2457 | * cache the last-identified data region. | |
2458 | * | |
2459 | * Second, limiting ourselves to protocol nodes allows us to assume | |
2460 | * the block status for data regions to be DATA | OFFSET_VALID, and | |
2461 | * that the host offset is the same as the guest offset. | |
2462 | * | |
2463 | * Note that it is possible that external writers zero parts of | |
2464 | * the cached regions without the cache being invalidated, and so | |
2465 | * we may report zeroes as data. This is not catastrophic, | |
2466 | * however, because reporting zeroes as data is fine. | |
2467 | */ | |
2468 | if (QLIST_EMPTY(&bs->children) && | |
2469 | bdrv_bsc_is_data(bs, aligned_offset, pnum)) | |
2470 | { | |
2471 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
2472 | local_file = bs; | |
2473 | local_map = aligned_offset; | |
2474 | } else { | |
2475 | ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset, | |
2476 | aligned_bytes, pnum, &local_map, | |
2477 | &local_file); | |
2478 | ||
2479 | /* | |
2480 | * Note that checking QLIST_EMPTY(&bs->children) is also done when | |
2481 | * the cache is queried above. Technically, we do not need to check | |
2482 | * it here; the worst that can happen is that we fill the cache for | |
2483 | * non-protocol nodes, and then it is never used. However, filling | |
2484 | * the cache requires an RCU update, so double check here to avoid | |
2485 | * such an update if possible. | |
113b727c HR |
2486 | * |
2487 | * Check want_zero, because we only want to update the cache when we | |
2488 | * have accurate information about what is zero and what is data. | |
0bc329fb | 2489 | */ |
113b727c HR |
2490 | if (want_zero && |
2491 | ret == (BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID) && | |
0bc329fb HR |
2492 | QLIST_EMPTY(&bs->children)) |
2493 | { | |
2494 | /* | |
2495 | * When a protocol driver reports BLOCK_OFFSET_VALID, the | |
2496 | * returned local_map value must be the same as the offset we | |
2497 | * have passed (aligned_offset), and local_bs must be the node | |
2498 | * itself. | |
2499 | * Assert this, because we follow this rule when reading from | |
2500 | * the cache (see the `local_file = bs` and | |
2501 | * `local_map = aligned_offset` assignments above), and the | |
2502 | * result the cache delivers must be the same as the driver | |
2503 | * would deliver. | |
2504 | */ | |
2505 | assert(local_file == bs); | |
2506 | assert(local_map == aligned_offset); | |
2507 | bdrv_bsc_fill(bs, aligned_offset, *pnum); | |
2508 | } | |
2509 | } | |
549ec0d9 HR |
2510 | } else { |
2511 | /* Default code for filters */ | |
2512 | ||
2513 | local_file = bdrv_filter_bs(bs); | |
2514 | assert(local_file); | |
2515 | ||
2516 | *pnum = aligned_bytes; | |
2517 | local_map = aligned_offset; | |
2518 | ret = BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; | |
2519 | } | |
636cb512 EB |
2520 | if (ret < 0) { |
2521 | *pnum = 0; | |
2522 | goto out; | |
efa6e2ed EB |
2523 | } |
2524 | ||
2e8bc787 | 2525 | /* |
636cb512 | 2526 | * The driver's result must be a non-zero multiple of request_alignment. |
efa6e2ed | 2527 | * Clamp pnum and adjust map to original request. |
2e8bc787 | 2528 | */ |
636cb512 EB |
2529 | assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) && |
2530 | align > offset - aligned_offset); | |
69f47505 VSO |
2531 | if (ret & BDRV_BLOCK_RECURSE) { |
2532 | assert(ret & BDRV_BLOCK_DATA); | |
2533 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
2534 | assert(!(ret & BDRV_BLOCK_ZERO)); | |
2535 | } | |
2536 | ||
efa6e2ed EB |
2537 | *pnum -= offset - aligned_offset; |
2538 | if (*pnum > bytes) { | |
2539 | *pnum = bytes; | |
61007b31 | 2540 | } |
2e8bc787 | 2541 | if (ret & BDRV_BLOCK_OFFSET_VALID) { |
efa6e2ed | 2542 | local_map += offset - aligned_offset; |
2e8bc787 | 2543 | } |
61007b31 SH |
2544 | |
2545 | if (ret & BDRV_BLOCK_RAW) { | |
298a1665 | 2546 | assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file); |
2e8bc787 EB |
2547 | ret = bdrv_co_block_status(local_file, want_zero, local_map, |
2548 | *pnum, pnum, &local_map, &local_file); | |
99723548 | 2549 | goto out; |
61007b31 SH |
2550 | } |
2551 | ||
2552 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
2553 | ret |= BDRV_BLOCK_ALLOCATED; | |
d40f4a56 | 2554 | } else if (bs->drv->supports_backing) { |
cb850315 HR |
2555 | BlockDriverState *cow_bs = bdrv_cow_bs(bs); |
2556 | ||
d40f4a56 AG |
2557 | if (!cow_bs) { |
2558 | ret |= BDRV_BLOCK_ZERO; | |
2559 | } else if (want_zero) { | |
0af02bd1 | 2560 | int64_t size2 = bdrv_co_getlength(cow_bs); |
c9ce8c4d | 2561 | |
2e8bc787 | 2562 | if (size2 >= 0 && offset >= size2) { |
61007b31 SH |
2563 | ret |= BDRV_BLOCK_ZERO; |
2564 | } | |
2565 | } | |
2566 | } | |
2567 | ||
69f47505 VSO |
2568 | if (want_zero && ret & BDRV_BLOCK_RECURSE && |
2569 | local_file && local_file != bs && | |
61007b31 SH |
2570 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
2571 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
2e8bc787 EB |
2572 | int64_t file_pnum; |
2573 | int ret2; | |
61007b31 | 2574 | |
2e8bc787 EB |
2575 | ret2 = bdrv_co_block_status(local_file, want_zero, local_map, |
2576 | *pnum, &file_pnum, NULL, NULL); | |
61007b31 SH |
2577 | if (ret2 >= 0) { |
2578 | /* Ignore errors. This is just providing extra information, it | |
2579 | * is useful but not necessary. | |
2580 | */ | |
c61e684e EB |
2581 | if (ret2 & BDRV_BLOCK_EOF && |
2582 | (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) { | |
2583 | /* | |
2584 | * It is valid for the format block driver to read | |
2585 | * beyond the end of the underlying file's current | |
2586 | * size; such areas read as zero. | |
2587 | */ | |
61007b31 SH |
2588 | ret |= BDRV_BLOCK_ZERO; |
2589 | } else { | |
2590 | /* Limit request to the range reported by the protocol driver */ | |
2591 | *pnum = file_pnum; | |
2592 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
2593 | } | |
2594 | } | |
2595 | } | |
2596 | ||
99723548 PB |
2597 | out: |
2598 | bdrv_dec_in_flight(bs); | |
2e8bc787 | 2599 | if (ret >= 0 && offset + *pnum == total_size) { |
fb0d8654 EB |
2600 | ret |= BDRV_BLOCK_EOF; |
2601 | } | |
298a1665 EB |
2602 | early_out: |
2603 | if (file) { | |
2604 | *file = local_file; | |
2605 | } | |
2e8bc787 EB |
2606 | if (map) { |
2607 | *map = local_map; | |
2608 | } | |
61007b31 SH |
2609 | return ret; |
2610 | } | |
2611 | ||
21c2283e | 2612 | int coroutine_fn |
f9e694cb VSO |
2613 | bdrv_co_common_block_status_above(BlockDriverState *bs, |
2614 | BlockDriverState *base, | |
3555a432 | 2615 | bool include_base, |
f9e694cb VSO |
2616 | bool want_zero, |
2617 | int64_t offset, | |
2618 | int64_t bytes, | |
2619 | int64_t *pnum, | |
2620 | int64_t *map, | |
a92b1b06 EB |
2621 | BlockDriverState **file, |
2622 | int *depth) | |
ba3f0e25 | 2623 | { |
67c095c8 | 2624 | int ret; |
ba3f0e25 | 2625 | BlockDriverState *p; |
67c095c8 | 2626 | int64_t eof = 0; |
a92b1b06 | 2627 | int dummy; |
1581a70d | 2628 | IO_CODE(); |
ba3f0e25 | 2629 | |
3555a432 | 2630 | assert(!include_base || base); /* Can't include NULL base */ |
7ff9579e | 2631 | assert_bdrv_graph_readable(); |
67c095c8 | 2632 | |
a92b1b06 EB |
2633 | if (!depth) { |
2634 | depth = &dummy; | |
2635 | } | |
2636 | *depth = 0; | |
2637 | ||
624f27bb VSO |
2638 | if (!include_base && bs == base) { |
2639 | *pnum = bytes; | |
2640 | return 0; | |
2641 | } | |
2642 | ||
67c095c8 | 2643 | ret = bdrv_co_block_status(bs, want_zero, offset, bytes, pnum, map, file); |
a92b1b06 | 2644 | ++*depth; |
3555a432 | 2645 | if (ret < 0 || *pnum == 0 || ret & BDRV_BLOCK_ALLOCATED || bs == base) { |
67c095c8 VSO |
2646 | return ret; |
2647 | } | |
2648 | ||
2649 | if (ret & BDRV_BLOCK_EOF) { | |
2650 | eof = offset + *pnum; | |
2651 | } | |
2652 | ||
2653 | assert(*pnum <= bytes); | |
2654 | bytes = *pnum; | |
2655 | ||
3555a432 | 2656 | for (p = bdrv_filter_or_cow_bs(bs); include_base || p != base; |
67c095c8 VSO |
2657 | p = bdrv_filter_or_cow_bs(p)) |
2658 | { | |
5b648c67 EB |
2659 | ret = bdrv_co_block_status(p, want_zero, offset, bytes, pnum, map, |
2660 | file); | |
a92b1b06 | 2661 | ++*depth; |
c61e684e | 2662 | if (ret < 0) { |
67c095c8 | 2663 | return ret; |
c61e684e | 2664 | } |
67c095c8 | 2665 | if (*pnum == 0) { |
c61e684e | 2666 | /* |
67c095c8 VSO |
2667 | * The top layer deferred to this layer, and because this layer is |
2668 | * short, any zeroes that we synthesize beyond EOF behave as if they | |
2669 | * were allocated at this layer. | |
2670 | * | |
2671 | * We don't include BDRV_BLOCK_EOF into ret, as upper layer may be | |
2672 | * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see | |
2673 | * below. | |
c61e684e | 2674 | */ |
67c095c8 | 2675 | assert(ret & BDRV_BLOCK_EOF); |
5b648c67 | 2676 | *pnum = bytes; |
67c095c8 VSO |
2677 | if (file) { |
2678 | *file = p; | |
2679 | } | |
2680 | ret = BDRV_BLOCK_ZERO | BDRV_BLOCK_ALLOCATED; | |
2681 | break; | |
c61e684e | 2682 | } |
67c095c8 VSO |
2683 | if (ret & BDRV_BLOCK_ALLOCATED) { |
2684 | /* | |
2685 | * We've found the node and the status, we must break. | |
2686 | * | |
2687 | * Drop BDRV_BLOCK_EOF, as it's not for upper layer, which may be | |
2688 | * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see | |
2689 | * below. | |
2690 | */ | |
2691 | ret &= ~BDRV_BLOCK_EOF; | |
ba3f0e25 FZ |
2692 | break; |
2693 | } | |
67c095c8 | 2694 | |
3555a432 VSO |
2695 | if (p == base) { |
2696 | assert(include_base); | |
2697 | break; | |
2698 | } | |
2699 | ||
67c095c8 VSO |
2700 | /* |
2701 | * OK, [offset, offset + *pnum) region is unallocated on this layer, | |
2702 | * let's continue the diving. | |
2703 | */ | |
2704 | assert(*pnum <= bytes); | |
2705 | bytes = *pnum; | |
ba3f0e25 | 2706 | } |
67c095c8 VSO |
2707 | |
2708 | if (offset + *pnum == eof) { | |
2709 | ret |= BDRV_BLOCK_EOF; | |
2710 | } | |
2711 | ||
ba3f0e25 FZ |
2712 | return ret; |
2713 | } | |
2714 | ||
7b52a921 EGE |
2715 | int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs, |
2716 | BlockDriverState *base, | |
2717 | int64_t offset, int64_t bytes, | |
2718 | int64_t *pnum, int64_t *map, | |
2719 | BlockDriverState **file) | |
2720 | { | |
2721 | IO_CODE(); | |
2722 | return bdrv_co_common_block_status_above(bs, base, false, true, offset, | |
2723 | bytes, pnum, map, file, NULL); | |
2724 | } | |
2725 | ||
31826642 EB |
2726 | int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base, |
2727 | int64_t offset, int64_t bytes, int64_t *pnum, | |
2728 | int64_t *map, BlockDriverState **file) | |
c9ce8c4d | 2729 | { |
384a48fb | 2730 | IO_CODE(); |
3555a432 | 2731 | return bdrv_common_block_status_above(bs, base, false, true, offset, bytes, |
a92b1b06 | 2732 | pnum, map, file, NULL); |
c9ce8c4d EB |
2733 | } |
2734 | ||
237d78f8 EB |
2735 | int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2736 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
ba3f0e25 | 2737 | { |
384a48fb | 2738 | IO_CODE(); |
cb850315 | 2739 | return bdrv_block_status_above(bs, bdrv_filter_or_cow_bs(bs), |
31826642 | 2740 | offset, bytes, pnum, map, file); |
ba3f0e25 FZ |
2741 | } |
2742 | ||
46cd1e8a AG |
2743 | /* |
2744 | * Check @bs (and its backing chain) to see if the range defined | |
2745 | * by @offset and @bytes is known to read as zeroes. | |
2746 | * Return 1 if that is the case, 0 otherwise and -errno on error. | |
2747 | * This test is meant to be fast rather than accurate so returning 0 | |
2748 | * does not guarantee non-zero data. | |
2749 | */ | |
2750 | int coroutine_fn bdrv_co_is_zero_fast(BlockDriverState *bs, int64_t offset, | |
2751 | int64_t bytes) | |
2752 | { | |
2753 | int ret; | |
2754 | int64_t pnum = bytes; | |
384a48fb | 2755 | IO_CODE(); |
46cd1e8a AG |
2756 | |
2757 | if (!bytes) { | |
2758 | return 1; | |
2759 | } | |
2760 | ||
ce47ff20 AF |
2761 | ret = bdrv_co_common_block_status_above(bs, NULL, false, false, offset, |
2762 | bytes, &pnum, NULL, NULL, NULL); | |
46cd1e8a AG |
2763 | |
2764 | if (ret < 0) { | |
2765 | return ret; | |
2766 | } | |
2767 | ||
2768 | return (pnum == bytes) && (ret & BDRV_BLOCK_ZERO); | |
2769 | } | |
2770 | ||
7b52a921 EGE |
2771 | int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t offset, |
2772 | int64_t bytes, int64_t *pnum) | |
2773 | { | |
2774 | int ret; | |
2775 | int64_t dummy; | |
2776 | IO_CODE(); | |
2777 | ||
2778 | ret = bdrv_co_common_block_status_above(bs, bs, true, false, offset, | |
2779 | bytes, pnum ? pnum : &dummy, NULL, | |
2780 | NULL, NULL); | |
2781 | if (ret < 0) { | |
2782 | return ret; | |
2783 | } | |
2784 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2785 | } | |
2786 | ||
7c85803c AF |
2787 | int bdrv_is_allocated(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2788 | int64_t *pnum) | |
61007b31 | 2789 | { |
7ddb99b9 EB |
2790 | int ret; |
2791 | int64_t dummy; | |
384a48fb | 2792 | IO_CODE(); |
d6a644bb | 2793 | |
3555a432 VSO |
2794 | ret = bdrv_common_block_status_above(bs, bs, true, false, offset, |
2795 | bytes, pnum ? pnum : &dummy, NULL, | |
a92b1b06 | 2796 | NULL, NULL); |
61007b31 SH |
2797 | if (ret < 0) { |
2798 | return ret; | |
2799 | } | |
2800 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2801 | } | |
2802 | ||
7b52a921 EGE |
2803 | /* See bdrv_is_allocated_above for documentation */ |
2804 | int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *top, | |
2805 | BlockDriverState *base, | |
2806 | bool include_base, int64_t offset, | |
2807 | int64_t bytes, int64_t *pnum) | |
2808 | { | |
2809 | int depth; | |
2810 | int ret; | |
2811 | IO_CODE(); | |
2812 | ||
2813 | ret = bdrv_co_common_block_status_above(top, base, include_base, false, | |
2814 | offset, bytes, pnum, NULL, NULL, | |
2815 | &depth); | |
2816 | if (ret < 0) { | |
2817 | return ret; | |
2818 | } | |
2819 | ||
2820 | if (ret & BDRV_BLOCK_ALLOCATED) { | |
2821 | return depth; | |
2822 | } | |
2823 | return 0; | |
2824 | } | |
2825 | ||
61007b31 SH |
2826 | /* |
2827 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
2828 | * | |
a92b1b06 EB |
2829 | * Return a positive depth if (a prefix of) the given range is allocated |
2830 | * in any image between BASE and TOP (BASE is only included if include_base | |
2831 | * is set). Depth 1 is TOP, 2 is the first backing layer, and so forth. | |
170d3bd3 AS |
2832 | * BASE can be NULL to check if the given offset is allocated in any |
2833 | * image of the chain. Return 0 otherwise, or negative errno on | |
2834 | * failure. | |
61007b31 | 2835 | * |
51b0a488 EB |
2836 | * 'pnum' is set to the number of bytes (including and immediately |
2837 | * following the specified offset) that are known to be in the same | |
2838 | * allocated/unallocated state. Note that a subsequent call starting | |
2839 | * at 'offset + *pnum' may return the same allocation status (in other | |
2840 | * words, the result is not necessarily the maximum possible range); | |
2841 | * but 'pnum' will only be 0 when end of file is reached. | |
61007b31 SH |
2842 | */ |
2843 | int bdrv_is_allocated_above(BlockDriverState *top, | |
2844 | BlockDriverState *base, | |
170d3bd3 AS |
2845 | bool include_base, int64_t offset, |
2846 | int64_t bytes, int64_t *pnum) | |
61007b31 | 2847 | { |
a92b1b06 | 2848 | int depth; |
7b52a921 | 2849 | int ret; |
384a48fb | 2850 | IO_CODE(); |
7b52a921 EGE |
2851 | |
2852 | ret = bdrv_common_block_status_above(top, base, include_base, false, | |
2853 | offset, bytes, pnum, NULL, NULL, | |
2854 | &depth); | |
7e7e5100 VSO |
2855 | if (ret < 0) { |
2856 | return ret; | |
61007b31 SH |
2857 | } |
2858 | ||
a92b1b06 EB |
2859 | if (ret & BDRV_BLOCK_ALLOCATED) { |
2860 | return depth; | |
2861 | } | |
2862 | return 0; | |
61007b31 SH |
2863 | } |
2864 | ||
21c2283e | 2865 | int coroutine_fn |
b33b354f | 2866 | bdrv_co_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) |
1a8ae822 KW |
2867 | { |
2868 | BlockDriver *drv = bs->drv; | |
c4db2e25 | 2869 | BlockDriverState *child_bs = bdrv_primary_bs(bs); |
b984b296 | 2870 | int ret; |
1581a70d | 2871 | IO_CODE(); |
1b3ff9fe | 2872 | assert_bdrv_graph_readable(); |
b984b296 VSO |
2873 | |
2874 | ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); | |
2875 | if (ret < 0) { | |
2876 | return ret; | |
2877 | } | |
dc88a467 | 2878 | |
b33b354f VSO |
2879 | if (!drv) { |
2880 | return -ENOMEDIUM; | |
2881 | } | |
2882 | ||
dc88a467 | 2883 | bdrv_inc_in_flight(bs); |
1a8ae822 | 2884 | |
ca5e2ad9 EGE |
2885 | if (drv->bdrv_co_load_vmstate) { |
2886 | ret = drv->bdrv_co_load_vmstate(bs, qiov, pos); | |
c4db2e25 | 2887 | } else if (child_bs) { |
b33b354f | 2888 | ret = bdrv_co_readv_vmstate(child_bs, qiov, pos); |
b984b296 VSO |
2889 | } else { |
2890 | ret = -ENOTSUP; | |
1a8ae822 KW |
2891 | } |
2892 | ||
dc88a467 | 2893 | bdrv_dec_in_flight(bs); |
b33b354f | 2894 | |
dc88a467 | 2895 | return ret; |
1a8ae822 KW |
2896 | } |
2897 | ||
b33b354f VSO |
2898 | int coroutine_fn |
2899 | bdrv_co_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 2900 | { |
b33b354f VSO |
2901 | BlockDriver *drv = bs->drv; |
2902 | BlockDriverState *child_bs = bdrv_primary_bs(bs); | |
b984b296 | 2903 | int ret; |
1581a70d | 2904 | IO_CODE(); |
1b3ff9fe | 2905 | assert_bdrv_graph_readable(); |
b984b296 VSO |
2906 | |
2907 | ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); | |
2908 | if (ret < 0) { | |
2909 | return ret; | |
2910 | } | |
61007b31 | 2911 | |
b33b354f VSO |
2912 | if (!drv) { |
2913 | return -ENOMEDIUM; | |
b433d942 KW |
2914 | } |
2915 | ||
b33b354f | 2916 | bdrv_inc_in_flight(bs); |
61007b31 | 2917 | |
ca5e2ad9 EGE |
2918 | if (drv->bdrv_co_save_vmstate) { |
2919 | ret = drv->bdrv_co_save_vmstate(bs, qiov, pos); | |
b33b354f VSO |
2920 | } else if (child_bs) { |
2921 | ret = bdrv_co_writev_vmstate(child_bs, qiov, pos); | |
b984b296 VSO |
2922 | } else { |
2923 | ret = -ENOTSUP; | |
b33b354f VSO |
2924 | } |
2925 | ||
2926 | bdrv_dec_in_flight(bs); | |
2927 | ||
2928 | return ret; | |
61007b31 SH |
2929 | } |
2930 | ||
b33b354f | 2931 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
61007b31 | 2932 | int64_t pos, int size) |
5ddda0b8 | 2933 | { |
0d93ed08 | 2934 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b33b354f | 2935 | int ret = bdrv_writev_vmstate(bs, &qiov, pos); |
384a48fb | 2936 | IO_CODE(); |
b433d942 | 2937 | |
b33b354f | 2938 | return ret < 0 ? ret : size; |
5ddda0b8 KW |
2939 | } |
2940 | ||
b33b354f VSO |
2941 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, |
2942 | int64_t pos, int size) | |
61007b31 | 2943 | { |
b33b354f VSO |
2944 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
2945 | int ret = bdrv_readv_vmstate(bs, &qiov, pos); | |
384a48fb | 2946 | IO_CODE(); |
b33b354f VSO |
2947 | |
2948 | return ret < 0 ? ret : size; | |
61007b31 SH |
2949 | } |
2950 | ||
2951 | /**************************************************************/ | |
2952 | /* async I/Os */ | |
2953 | ||
652b0dd8 SH |
2954 | /** |
2955 | * Synchronously cancels an acb. Must be called with the BQL held and the acb | |
2956 | * must be processed with the BQL held too (IOThreads are not allowed). | |
2957 | * | |
2958 | * Use bdrv_aio_cancel_async() instead when possible. | |
2959 | */ | |
61007b31 SH |
2960 | void bdrv_aio_cancel(BlockAIOCB *acb) |
2961 | { | |
652b0dd8 | 2962 | GLOBAL_STATE_CODE(); |
61007b31 SH |
2963 | qemu_aio_ref(acb); |
2964 | bdrv_aio_cancel_async(acb); | |
652b0dd8 | 2965 | AIO_WAIT_WHILE_UNLOCKED(NULL, acb->refcnt > 1); |
61007b31 SH |
2966 | qemu_aio_unref(acb); |
2967 | } | |
2968 | ||
2969 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2970 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2971 | * In either case the completion callback must be called. */ | |
2972 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2973 | { | |
384a48fb | 2974 | IO_CODE(); |
61007b31 SH |
2975 | if (acb->aiocb_info->cancel_async) { |
2976 | acb->aiocb_info->cancel_async(acb); | |
2977 | } | |
2978 | } | |
2979 | ||
61007b31 SH |
2980 | /**************************************************************/ |
2981 | /* Coroutine block device emulation */ | |
2982 | ||
61007b31 SH |
2983 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) |
2984 | { | |
883833e2 HR |
2985 | BdrvChild *primary_child = bdrv_primary_child(bs); |
2986 | BdrvChild *child; | |
49ca6259 FZ |
2987 | int current_gen; |
2988 | int ret = 0; | |
384a48fb | 2989 | IO_CODE(); |
49ca6259 | 2990 | |
88095349 | 2991 | assert_bdrv_graph_readable(); |
49ca6259 | 2992 | bdrv_inc_in_flight(bs); |
61007b31 | 2993 | |
1e97be91 | 2994 | if (!bdrv_co_is_inserted(bs) || bdrv_is_read_only(bs) || |
1b6bc94d | 2995 | bdrv_is_sg(bs)) { |
49ca6259 | 2996 | goto early_exit; |
61007b31 SH |
2997 | } |
2998 | ||
fa9185fc | 2999 | qemu_mutex_lock(&bs->reqs_lock); |
d73415a3 | 3000 | current_gen = qatomic_read(&bs->write_gen); |
3ff2f67a EY |
3001 | |
3002 | /* Wait until any previous flushes are completed */ | |
99723548 | 3003 | while (bs->active_flush_req) { |
3783fa3d | 3004 | qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock); |
3ff2f67a EY |
3005 | } |
3006 | ||
3783fa3d | 3007 | /* Flushes reach this point in nondecreasing current_gen order. */ |
99723548 | 3008 | bs->active_flush_req = true; |
fa9185fc | 3009 | qemu_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 3010 | |
c32b82af PD |
3011 | /* Write back all layers by calling one driver function */ |
3012 | if (bs->drv->bdrv_co_flush) { | |
3013 | ret = bs->drv->bdrv_co_flush(bs); | |
3014 | goto out; | |
3015 | } | |
3016 | ||
61007b31 | 3017 | /* Write back cached data to the OS even with cache=unsafe */ |
17362398 | 3018 | BLKDBG_CO_EVENT(primary_child, BLKDBG_FLUSH_TO_OS); |
61007b31 SH |
3019 | if (bs->drv->bdrv_co_flush_to_os) { |
3020 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
3021 | if (ret < 0) { | |
cdb5e315 | 3022 | goto out; |
61007b31 SH |
3023 | } |
3024 | } | |
3025 | ||
3026 | /* But don't actually force it to the disk with cache=unsafe */ | |
3027 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
883833e2 | 3028 | goto flush_children; |
61007b31 SH |
3029 | } |
3030 | ||
3ff2f67a EY |
3031 | /* Check if we really need to flush anything */ |
3032 | if (bs->flushed_gen == current_gen) { | |
883833e2 | 3033 | goto flush_children; |
3ff2f67a EY |
3034 | } |
3035 | ||
17362398 | 3036 | BLKDBG_CO_EVENT(primary_child, BLKDBG_FLUSH_TO_DISK); |
d470ad42 HR |
3037 | if (!bs->drv) { |
3038 | /* bs->drv->bdrv_co_flush() might have ejected the BDS | |
3039 | * (even in case of apparent success) */ | |
3040 | ret = -ENOMEDIUM; | |
3041 | goto out; | |
3042 | } | |
61007b31 SH |
3043 | if (bs->drv->bdrv_co_flush_to_disk) { |
3044 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
3045 | } else if (bs->drv->bdrv_aio_flush) { | |
3046 | BlockAIOCB *acb; | |
3047 | CoroutineIOCompletion co = { | |
3048 | .coroutine = qemu_coroutine_self(), | |
3049 | }; | |
3050 | ||
3051 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
3052 | if (acb == NULL) { | |
3053 | ret = -EIO; | |
3054 | } else { | |
3055 | qemu_coroutine_yield(); | |
3056 | ret = co.ret; | |
3057 | } | |
3058 | } else { | |
3059 | /* | |
3060 | * Some block drivers always operate in either writethrough or unsafe | |
3061 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
3062 | * know how the server works (because the behaviour is hardcoded or | |
3063 | * depends on server-side configuration), so we can't ensure that | |
3064 | * everything is safe on disk. Returning an error doesn't work because | |
3065 | * that would break guests even if the server operates in writethrough | |
3066 | * mode. | |
3067 | * | |
3068 | * Let's hope the user knows what he's doing. | |
3069 | */ | |
3070 | ret = 0; | |
3071 | } | |
3ff2f67a | 3072 | |
61007b31 | 3073 | if (ret < 0) { |
cdb5e315 | 3074 | goto out; |
61007b31 SH |
3075 | } |
3076 | ||
3077 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
3078 | * in the case of cache=unsafe, so there are no useless flushes. | |
3079 | */ | |
883833e2 HR |
3080 | flush_children: |
3081 | ret = 0; | |
3082 | QLIST_FOREACH(child, &bs->children, next) { | |
3083 | if (child->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
3084 | int this_child_ret = bdrv_co_flush(child->bs); | |
3085 | if (!ret) { | |
3086 | ret = this_child_ret; | |
3087 | } | |
3088 | } | |
3089 | } | |
3090 | ||
cdb5e315 | 3091 | out: |
3ff2f67a | 3092 | /* Notify any pending flushes that we have completed */ |
e6af1e08 KW |
3093 | if (ret == 0) { |
3094 | bs->flushed_gen = current_gen; | |
3095 | } | |
3783fa3d | 3096 | |
fa9185fc | 3097 | qemu_mutex_lock(&bs->reqs_lock); |
99723548 | 3098 | bs->active_flush_req = false; |
156af3ac DL |
3099 | /* Return value is ignored - it's ok if wait queue is empty */ |
3100 | qemu_co_queue_next(&bs->flush_queue); | |
fa9185fc | 3101 | qemu_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 3102 | |
49ca6259 | 3103 | early_exit: |
99723548 | 3104 | bdrv_dec_in_flight(bs); |
cdb5e315 | 3105 | return ret; |
61007b31 SH |
3106 | } |
3107 | ||
d93e5726 VSO |
3108 | int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, |
3109 | int64_t bytes) | |
61007b31 | 3110 | { |
b1066c87 | 3111 | BdrvTrackedRequest req; |
39af49c0 VSO |
3112 | int ret; |
3113 | int64_t max_pdiscard; | |
3482b9bc | 3114 | int head, tail, align; |
0b9fd3f4 | 3115 | BlockDriverState *bs = child->bs; |
384a48fb | 3116 | IO_CODE(); |
9a5a1c62 | 3117 | assert_bdrv_graph_readable(); |
61007b31 | 3118 | |
1e97be91 | 3119 | if (!bs || !bs->drv || !bdrv_co_is_inserted(bs)) { |
61007b31 SH |
3120 | return -ENOMEDIUM; |
3121 | } | |
3122 | ||
d6883bc9 VSO |
3123 | if (bdrv_has_readonly_bitmaps(bs)) { |
3124 | return -EPERM; | |
3125 | } | |
3126 | ||
69b55e03 | 3127 | ret = bdrv_check_request(offset, bytes, NULL); |
8b117001 VSO |
3128 | if (ret < 0) { |
3129 | return ret; | |
61007b31 SH |
3130 | } |
3131 | ||
61007b31 SH |
3132 | /* Do nothing if disabled. */ |
3133 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
3134 | return 0; | |
3135 | } | |
3136 | ||
02aefe43 | 3137 | if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) { |
61007b31 SH |
3138 | return 0; |
3139 | } | |
3140 | ||
0bc329fb HR |
3141 | /* Invalidate the cached block-status data range if this discard overlaps */ |
3142 | bdrv_bsc_invalidate_range(bs, offset, bytes); | |
3143 | ||
3482b9bc EB |
3144 | /* Discard is advisory, but some devices track and coalesce |
3145 | * unaligned requests, so we must pass everything down rather than | |
3146 | * round here. Still, most devices will just silently ignore | |
3147 | * unaligned requests (by returning -ENOTSUP), so we must fragment | |
3148 | * the request accordingly. */ | |
02aefe43 | 3149 | align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment); |
b8d0a980 EB |
3150 | assert(align % bs->bl.request_alignment == 0); |
3151 | head = offset % align; | |
f5a5ca79 | 3152 | tail = (offset + bytes) % align; |
9f1963b3 | 3153 | |
99723548 | 3154 | bdrv_inc_in_flight(bs); |
f5a5ca79 | 3155 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD); |
50824995 | 3156 | |
00695c27 | 3157 | ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0); |
ec050f77 DL |
3158 | if (ret < 0) { |
3159 | goto out; | |
3160 | } | |
3161 | ||
6a8f3dbb | 3162 | max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT64_MAX), |
9f1963b3 | 3163 | align); |
3482b9bc | 3164 | assert(max_pdiscard >= bs->bl.request_alignment); |
61007b31 | 3165 | |
f5a5ca79 | 3166 | while (bytes > 0) { |
d93e5726 | 3167 | int64_t num = bytes; |
3482b9bc EB |
3168 | |
3169 | if (head) { | |
3170 | /* Make small requests to get to alignment boundaries. */ | |
f5a5ca79 | 3171 | num = MIN(bytes, align - head); |
3482b9bc EB |
3172 | if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) { |
3173 | num %= bs->bl.request_alignment; | |
3174 | } | |
3175 | head = (head + num) % align; | |
3176 | assert(num < max_pdiscard); | |
3177 | } else if (tail) { | |
3178 | if (num > align) { | |
3179 | /* Shorten the request to the last aligned cluster. */ | |
3180 | num -= tail; | |
3181 | } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) && | |
3182 | tail > bs->bl.request_alignment) { | |
3183 | tail %= bs->bl.request_alignment; | |
3184 | num -= tail; | |
3185 | } | |
3186 | } | |
3187 | /* limit request size */ | |
3188 | if (num > max_pdiscard) { | |
3189 | num = max_pdiscard; | |
3190 | } | |
61007b31 | 3191 | |
d470ad42 HR |
3192 | if (!bs->drv) { |
3193 | ret = -ENOMEDIUM; | |
3194 | goto out; | |
3195 | } | |
47a5486d EB |
3196 | if (bs->drv->bdrv_co_pdiscard) { |
3197 | ret = bs->drv->bdrv_co_pdiscard(bs, offset, num); | |
61007b31 SH |
3198 | } else { |
3199 | BlockAIOCB *acb; | |
3200 | CoroutineIOCompletion co = { | |
3201 | .coroutine = qemu_coroutine_self(), | |
3202 | }; | |
3203 | ||
4da444a0 EB |
3204 | acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num, |
3205 | bdrv_co_io_em_complete, &co); | |
61007b31 | 3206 | if (acb == NULL) { |
b1066c87 FZ |
3207 | ret = -EIO; |
3208 | goto out; | |
61007b31 SH |
3209 | } else { |
3210 | qemu_coroutine_yield(); | |
3211 | ret = co.ret; | |
3212 | } | |
3213 | } | |
3214 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 3215 | goto out; |
61007b31 SH |
3216 | } |
3217 | ||
9f1963b3 | 3218 | offset += num; |
f5a5ca79 | 3219 | bytes -= num; |
61007b31 | 3220 | } |
b1066c87 FZ |
3221 | ret = 0; |
3222 | out: | |
00695c27 | 3223 | bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret); |
b1066c87 | 3224 | tracked_request_end(&req); |
99723548 | 3225 | bdrv_dec_in_flight(bs); |
b1066c87 | 3226 | return ret; |
61007b31 SH |
3227 | } |
3228 | ||
881a4c55 | 3229 | int coroutine_fn bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
3230 | { |
3231 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
3232 | CoroutineIOCompletion co = { |
3233 | .coroutine = qemu_coroutine_self(), | |
3234 | }; | |
3235 | BlockAIOCB *acb; | |
384a48fb | 3236 | IO_CODE(); |
26c518ab | 3237 | assert_bdrv_graph_readable(); |
61007b31 | 3238 | |
99723548 | 3239 | bdrv_inc_in_flight(bs); |
16a389dc | 3240 | if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) { |
5c5ae76a FZ |
3241 | co.ret = -ENOTSUP; |
3242 | goto out; | |
3243 | } | |
3244 | ||
16a389dc KW |
3245 | if (drv->bdrv_co_ioctl) { |
3246 | co.ret = drv->bdrv_co_ioctl(bs, req, buf); | |
3247 | } else { | |
3248 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
3249 | if (!acb) { | |
3250 | co.ret = -ENOTSUP; | |
3251 | goto out; | |
3252 | } | |
3253 | qemu_coroutine_yield(); | |
5c5ae76a | 3254 | } |
5c5ae76a | 3255 | out: |
99723548 | 3256 | bdrv_dec_in_flight(bs); |
5c5ae76a FZ |
3257 | return co.ret; |
3258 | } | |
3259 | ||
6d43eaa3 SL |
3260 | int coroutine_fn bdrv_co_zone_report(BlockDriverState *bs, int64_t offset, |
3261 | unsigned int *nr_zones, | |
3262 | BlockZoneDescriptor *zones) | |
3263 | { | |
3264 | BlockDriver *drv = bs->drv; | |
3265 | CoroutineIOCompletion co = { | |
3266 | .coroutine = qemu_coroutine_self(), | |
3267 | }; | |
3268 | IO_CODE(); | |
3269 | ||
3270 | bdrv_inc_in_flight(bs); | |
3271 | if (!drv || !drv->bdrv_co_zone_report || bs->bl.zoned == BLK_Z_NONE) { | |
3272 | co.ret = -ENOTSUP; | |
3273 | goto out; | |
3274 | } | |
3275 | co.ret = drv->bdrv_co_zone_report(bs, offset, nr_zones, zones); | |
3276 | out: | |
3277 | bdrv_dec_in_flight(bs); | |
3278 | return co.ret; | |
3279 | } | |
3280 | ||
3281 | int coroutine_fn bdrv_co_zone_mgmt(BlockDriverState *bs, BlockZoneOp op, | |
3282 | int64_t offset, int64_t len) | |
3283 | { | |
3284 | BlockDriver *drv = bs->drv; | |
3285 | CoroutineIOCompletion co = { | |
3286 | .coroutine = qemu_coroutine_self(), | |
3287 | }; | |
3288 | IO_CODE(); | |
3289 | ||
3290 | bdrv_inc_in_flight(bs); | |
3291 | if (!drv || !drv->bdrv_co_zone_mgmt || bs->bl.zoned == BLK_Z_NONE) { | |
3292 | co.ret = -ENOTSUP; | |
3293 | goto out; | |
3294 | } | |
3295 | co.ret = drv->bdrv_co_zone_mgmt(bs, op, offset, len); | |
3296 | out: | |
3297 | bdrv_dec_in_flight(bs); | |
3298 | return co.ret; | |
3299 | } | |
3300 | ||
4751d09a SL |
3301 | int coroutine_fn bdrv_co_zone_append(BlockDriverState *bs, int64_t *offset, |
3302 | QEMUIOVector *qiov, | |
3303 | BdrvRequestFlags flags) | |
3304 | { | |
3305 | int ret; | |
3306 | BlockDriver *drv = bs->drv; | |
3307 | CoroutineIOCompletion co = { | |
3308 | .coroutine = qemu_coroutine_self(), | |
3309 | }; | |
3310 | IO_CODE(); | |
3311 | ||
3312 | ret = bdrv_check_qiov_request(*offset, qiov->size, qiov, 0, NULL); | |
3313 | if (ret < 0) { | |
3314 | return ret; | |
3315 | } | |
3316 | ||
3317 | bdrv_inc_in_flight(bs); | |
3318 | if (!drv || !drv->bdrv_co_zone_append || bs->bl.zoned == BLK_Z_NONE) { | |
3319 | co.ret = -ENOTSUP; | |
3320 | goto out; | |
3321 | } | |
3322 | co.ret = drv->bdrv_co_zone_append(bs, offset, qiov, flags); | |
3323 | out: | |
3324 | bdrv_dec_in_flight(bs); | |
3325 | return co.ret; | |
3326 | } | |
3327 | ||
61007b31 SH |
3328 | void *qemu_blockalign(BlockDriverState *bs, size_t size) |
3329 | { | |
384a48fb | 3330 | IO_CODE(); |
61007b31 SH |
3331 | return qemu_memalign(bdrv_opt_mem_align(bs), size); |
3332 | } | |
3333 | ||
3334 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
3335 | { | |
384a48fb | 3336 | IO_CODE(); |
61007b31 SH |
3337 | return memset(qemu_blockalign(bs, size), 0, size); |
3338 | } | |
3339 | ||
3340 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
3341 | { | |
3342 | size_t align = bdrv_opt_mem_align(bs); | |
384a48fb | 3343 | IO_CODE(); |
61007b31 SH |
3344 | |
3345 | /* Ensure that NULL is never returned on success */ | |
3346 | assert(align > 0); | |
3347 | if (size == 0) { | |
3348 | size = align; | |
3349 | } | |
3350 | ||
3351 | return qemu_try_memalign(align, size); | |
3352 | } | |
3353 | ||
3354 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
3355 | { | |
3356 | void *mem = qemu_try_blockalign(bs, size); | |
384a48fb | 3357 | IO_CODE(); |
61007b31 SH |
3358 | |
3359 | if (mem) { | |
3360 | memset(mem, 0, size); | |
3361 | } | |
3362 | ||
3363 | return mem; | |
3364 | } | |
3365 | ||
f4ec04ba | 3366 | /* Helper that undoes bdrv_register_buf() when it fails partway through */ |
d9249c25 KW |
3367 | static void GRAPH_RDLOCK |
3368 | bdrv_register_buf_rollback(BlockDriverState *bs, void *host, size_t size, | |
3369 | BdrvChild *final_child) | |
f4ec04ba SH |
3370 | { |
3371 | BdrvChild *child; | |
3372 | ||
d9249c25 KW |
3373 | GLOBAL_STATE_CODE(); |
3374 | assert_bdrv_graph_readable(); | |
3375 | ||
f4ec04ba SH |
3376 | QLIST_FOREACH(child, &bs->children, next) { |
3377 | if (child == final_child) { | |
3378 | break; | |
3379 | } | |
3380 | ||
3381 | bdrv_unregister_buf(child->bs, host, size); | |
3382 | } | |
3383 | ||
3384 | if (bs->drv && bs->drv->bdrv_unregister_buf) { | |
3385 | bs->drv->bdrv_unregister_buf(bs, host, size); | |
3386 | } | |
3387 | } | |
3388 | ||
3389 | bool bdrv_register_buf(BlockDriverState *bs, void *host, size_t size, | |
3390 | Error **errp) | |
23d0ba93 FZ |
3391 | { |
3392 | BdrvChild *child; | |
3393 | ||
f791bf7f | 3394 | GLOBAL_STATE_CODE(); |
d9249c25 KW |
3395 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
3396 | ||
23d0ba93 | 3397 | if (bs->drv && bs->drv->bdrv_register_buf) { |
f4ec04ba SH |
3398 | if (!bs->drv->bdrv_register_buf(bs, host, size, errp)) { |
3399 | return false; | |
3400 | } | |
23d0ba93 FZ |
3401 | } |
3402 | QLIST_FOREACH(child, &bs->children, next) { | |
f4ec04ba SH |
3403 | if (!bdrv_register_buf(child->bs, host, size, errp)) { |
3404 | bdrv_register_buf_rollback(bs, host, size, child); | |
3405 | return false; | |
3406 | } | |
23d0ba93 | 3407 | } |
f4ec04ba | 3408 | return true; |
23d0ba93 FZ |
3409 | } |
3410 | ||
4f384011 | 3411 | void bdrv_unregister_buf(BlockDriverState *bs, void *host, size_t size) |
23d0ba93 FZ |
3412 | { |
3413 | BdrvChild *child; | |
3414 | ||
f791bf7f | 3415 | GLOBAL_STATE_CODE(); |
d9249c25 KW |
3416 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
3417 | ||
23d0ba93 | 3418 | if (bs->drv && bs->drv->bdrv_unregister_buf) { |
4f384011 | 3419 | bs->drv->bdrv_unregister_buf(bs, host, size); |
23d0ba93 FZ |
3420 | } |
3421 | QLIST_FOREACH(child, &bs->children, next) { | |
4f384011 | 3422 | bdrv_unregister_buf(child->bs, host, size); |
23d0ba93 FZ |
3423 | } |
3424 | } | |
fcc67678 | 3425 | |
abaf8b75 | 3426 | static int coroutine_fn GRAPH_RDLOCK bdrv_co_copy_range_internal( |
a5215b8f VSO |
3427 | BdrvChild *src, int64_t src_offset, BdrvChild *dst, |
3428 | int64_t dst_offset, int64_t bytes, | |
67b51fb9 VSO |
3429 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags, |
3430 | bool recurse_src) | |
fcc67678 | 3431 | { |
999658a0 | 3432 | BdrvTrackedRequest req; |
fcc67678 | 3433 | int ret; |
742bf09b | 3434 | assert_bdrv_graph_readable(); |
fcc67678 | 3435 | |
fe0480d6 KW |
3436 | /* TODO We can support BDRV_REQ_NO_FALLBACK here */ |
3437 | assert(!(read_flags & BDRV_REQ_NO_FALLBACK)); | |
3438 | assert(!(write_flags & BDRV_REQ_NO_FALLBACK)); | |
45e62b46 VSO |
3439 | assert(!(read_flags & BDRV_REQ_NO_WAIT)); |
3440 | assert(!(write_flags & BDRV_REQ_NO_WAIT)); | |
fe0480d6 | 3441 | |
1e97be91 | 3442 | if (!dst || !dst->bs || !bdrv_co_is_inserted(dst->bs)) { |
fcc67678 FZ |
3443 | return -ENOMEDIUM; |
3444 | } | |
63f4ad11 | 3445 | ret = bdrv_check_request32(dst_offset, bytes, NULL, 0); |
fcc67678 FZ |
3446 | if (ret) { |
3447 | return ret; | |
3448 | } | |
67b51fb9 VSO |
3449 | if (write_flags & BDRV_REQ_ZERO_WRITE) { |
3450 | return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags); | |
fcc67678 FZ |
3451 | } |
3452 | ||
1e97be91 | 3453 | if (!src || !src->bs || !bdrv_co_is_inserted(src->bs)) { |
d4d3e5a0 FZ |
3454 | return -ENOMEDIUM; |
3455 | } | |
63f4ad11 | 3456 | ret = bdrv_check_request32(src_offset, bytes, NULL, 0); |
d4d3e5a0 FZ |
3457 | if (ret) { |
3458 | return ret; | |
3459 | } | |
3460 | ||
fcc67678 FZ |
3461 | if (!src->bs->drv->bdrv_co_copy_range_from |
3462 | || !dst->bs->drv->bdrv_co_copy_range_to | |
3463 | || src->bs->encrypted || dst->bs->encrypted) { | |
3464 | return -ENOTSUP; | |
3465 | } | |
37aec7d7 | 3466 | |
fcc67678 | 3467 | if (recurse_src) { |
999658a0 VSO |
3468 | bdrv_inc_in_flight(src->bs); |
3469 | tracked_request_begin(&req, src->bs, src_offset, bytes, | |
3470 | BDRV_TRACKED_READ); | |
3471 | ||
09d2f948 VSO |
3472 | /* BDRV_REQ_SERIALISING is only for write operation */ |
3473 | assert(!(read_flags & BDRV_REQ_SERIALISING)); | |
c53cb427 | 3474 | bdrv_wait_serialising_requests(&req); |
999658a0 | 3475 | |
37aec7d7 FZ |
3476 | ret = src->bs->drv->bdrv_co_copy_range_from(src->bs, |
3477 | src, src_offset, | |
3478 | dst, dst_offset, | |
67b51fb9 VSO |
3479 | bytes, |
3480 | read_flags, write_flags); | |
999658a0 VSO |
3481 | |
3482 | tracked_request_end(&req); | |
3483 | bdrv_dec_in_flight(src->bs); | |
fcc67678 | 3484 | } else { |
999658a0 VSO |
3485 | bdrv_inc_in_flight(dst->bs); |
3486 | tracked_request_begin(&req, dst->bs, dst_offset, bytes, | |
3487 | BDRV_TRACKED_WRITE); | |
0eb1e891 FZ |
3488 | ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req, |
3489 | write_flags); | |
3490 | if (!ret) { | |
3491 | ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs, | |
3492 | src, src_offset, | |
3493 | dst, dst_offset, | |
3494 | bytes, | |
3495 | read_flags, write_flags); | |
3496 | } | |
3497 | bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret); | |
999658a0 VSO |
3498 | tracked_request_end(&req); |
3499 | bdrv_dec_in_flight(dst->bs); | |
fcc67678 | 3500 | } |
999658a0 | 3501 | |
37aec7d7 | 3502 | return ret; |
fcc67678 FZ |
3503 | } |
3504 | ||
3505 | /* Copy range from @src to @dst. | |
3506 | * | |
3507 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3508 | * semantics. */ | |
a5215b8f VSO |
3509 | int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, int64_t src_offset, |
3510 | BdrvChild *dst, int64_t dst_offset, | |
3511 | int64_t bytes, | |
67b51fb9 VSO |
3512 | BdrvRequestFlags read_flags, |
3513 | BdrvRequestFlags write_flags) | |
fcc67678 | 3514 | { |
967d7905 | 3515 | IO_CODE(); |
742bf09b | 3516 | assert_bdrv_graph_readable(); |
ecc983a5 FZ |
3517 | trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes, |
3518 | read_flags, write_flags); | |
fcc67678 | 3519 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3520 | bytes, read_flags, write_flags, true); |
fcc67678 FZ |
3521 | } |
3522 | ||
3523 | /* Copy range from @src to @dst. | |
3524 | * | |
3525 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3526 | * semantics. */ | |
a5215b8f VSO |
3527 | int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, int64_t src_offset, |
3528 | BdrvChild *dst, int64_t dst_offset, | |
3529 | int64_t bytes, | |
67b51fb9 VSO |
3530 | BdrvRequestFlags read_flags, |
3531 | BdrvRequestFlags write_flags) | |
fcc67678 | 3532 | { |
967d7905 | 3533 | IO_CODE(); |
742bf09b | 3534 | assert_bdrv_graph_readable(); |
ecc983a5 FZ |
3535 | trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3536 | read_flags, write_flags); | |
fcc67678 | 3537 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3538 | bytes, read_flags, write_flags, false); |
fcc67678 FZ |
3539 | } |
3540 | ||
a5215b8f VSO |
3541 | int coroutine_fn bdrv_co_copy_range(BdrvChild *src, int64_t src_offset, |
3542 | BdrvChild *dst, int64_t dst_offset, | |
3543 | int64_t bytes, BdrvRequestFlags read_flags, | |
67b51fb9 | 3544 | BdrvRequestFlags write_flags) |
fcc67678 | 3545 | { |
384a48fb | 3546 | IO_CODE(); |
742bf09b EGE |
3547 | assert_bdrv_graph_readable(); |
3548 | ||
37aec7d7 FZ |
3549 | return bdrv_co_copy_range_from(src, src_offset, |
3550 | dst, dst_offset, | |
67b51fb9 | 3551 | bytes, read_flags, write_flags); |
fcc67678 | 3552 | } |
3d9f2d2a KW |
3553 | |
3554 | static void bdrv_parent_cb_resize(BlockDriverState *bs) | |
3555 | { | |
3556 | BdrvChild *c; | |
3557 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
3558 | if (c->klass->resize) { |
3559 | c->klass->resize(c); | |
3d9f2d2a KW |
3560 | } |
3561 | } | |
3562 | } | |
3563 | ||
3564 | /** | |
3565 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
c80d8b06 HR |
3566 | * |
3567 | * If 'exact' is true, the file must be resized to exactly the given | |
3568 | * 'offset'. Otherwise, it is sufficient for the node to be at least | |
3569 | * 'offset' bytes in length. | |
3d9f2d2a | 3570 | */ |
c80d8b06 | 3571 | int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, |
7b8e4857 KW |
3572 | PreallocMode prealloc, BdrvRequestFlags flags, |
3573 | Error **errp) | |
3d9f2d2a KW |
3574 | { |
3575 | BlockDriverState *bs = child->bs; | |
23b93525 | 3576 | BdrvChild *filtered, *backing; |
3d9f2d2a | 3577 | BlockDriver *drv = bs->drv; |
1bc5f09f KW |
3578 | BdrvTrackedRequest req; |
3579 | int64_t old_size, new_bytes; | |
3d9f2d2a | 3580 | int ret; |
384a48fb | 3581 | IO_CODE(); |
c2b8e315 | 3582 | assert_bdrv_graph_readable(); |
3d9f2d2a KW |
3583 | |
3584 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ | |
3585 | if (!drv) { | |
3586 | error_setg(errp, "No medium inserted"); | |
3587 | return -ENOMEDIUM; | |
3588 | } | |
3589 | if (offset < 0) { | |
3590 | error_setg(errp, "Image size cannot be negative"); | |
3591 | return -EINVAL; | |
3592 | } | |
3593 | ||
69b55e03 | 3594 | ret = bdrv_check_request(offset, 0, errp); |
8b117001 | 3595 | if (ret < 0) { |
8b117001 VSO |
3596 | return ret; |
3597 | } | |
3598 | ||
0af02bd1 | 3599 | old_size = bdrv_co_getlength(bs); |
1bc5f09f KW |
3600 | if (old_size < 0) { |
3601 | error_setg_errno(errp, -old_size, "Failed to get old image size"); | |
3602 | return old_size; | |
3603 | } | |
3604 | ||
97efa869 EB |
3605 | if (bdrv_is_read_only(bs)) { |
3606 | error_setg(errp, "Image is read-only"); | |
3607 | return -EACCES; | |
3608 | } | |
3609 | ||
1bc5f09f KW |
3610 | if (offset > old_size) { |
3611 | new_bytes = offset - old_size; | |
3612 | } else { | |
3613 | new_bytes = 0; | |
3614 | } | |
3615 | ||
3d9f2d2a | 3616 | bdrv_inc_in_flight(bs); |
5416a11e FZ |
3617 | tracked_request_begin(&req, bs, offset - new_bytes, new_bytes, |
3618 | BDRV_TRACKED_TRUNCATE); | |
1bc5f09f KW |
3619 | |
3620 | /* If we are growing the image and potentially using preallocation for the | |
3621 | * new area, we need to make sure that no write requests are made to it | |
3622 | * concurrently or they might be overwritten by preallocation. */ | |
3623 | if (new_bytes) { | |
8ac5aab2 | 3624 | bdrv_make_request_serialising(&req, 1); |
cd47d792 | 3625 | } |
cd47d792 FZ |
3626 | ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req, |
3627 | 0); | |
3628 | if (ret < 0) { | |
3629 | error_setg_errno(errp, -ret, | |
3630 | "Failed to prepare request for truncation"); | |
3631 | goto out; | |
1bc5f09f | 3632 | } |
3d9f2d2a | 3633 | |
93393e69 | 3634 | filtered = bdrv_filter_child(bs); |
23b93525 | 3635 | backing = bdrv_cow_child(bs); |
93393e69 | 3636 | |
955c7d66 KW |
3637 | /* |
3638 | * If the image has a backing file that is large enough that it would | |
3639 | * provide data for the new area, we cannot leave it unallocated because | |
3640 | * then the backing file content would become visible. Instead, zero-fill | |
3641 | * the new area. | |
3642 | * | |
3643 | * Note that if the image has a backing file, but was opened without the | |
3644 | * backing file, taking care of keeping things consistent with that backing | |
3645 | * file is the user's responsibility. | |
3646 | */ | |
23b93525 | 3647 | if (new_bytes && backing) { |
955c7d66 KW |
3648 | int64_t backing_len; |
3649 | ||
bd53086e | 3650 | backing_len = bdrv_co_getlength(backing->bs); |
955c7d66 KW |
3651 | if (backing_len < 0) { |
3652 | ret = backing_len; | |
3653 | error_setg_errno(errp, -ret, "Could not get backing file size"); | |
3654 | goto out; | |
3655 | } | |
3656 | ||
3657 | if (backing_len > old_size) { | |
3658 | flags |= BDRV_REQ_ZERO_WRITE; | |
3659 | } | |
3660 | } | |
3661 | ||
6b7e8f8b | 3662 | if (drv->bdrv_co_truncate) { |
92b92799 KW |
3663 | if (flags & ~bs->supported_truncate_flags) { |
3664 | error_setg(errp, "Block driver does not support requested flags"); | |
3665 | ret = -ENOTSUP; | |
3666 | goto out; | |
3667 | } | |
3668 | ret = drv->bdrv_co_truncate(bs, offset, exact, prealloc, flags, errp); | |
93393e69 HR |
3669 | } else if (filtered) { |
3670 | ret = bdrv_co_truncate(filtered, offset, exact, prealloc, flags, errp); | |
6b7e8f8b | 3671 | } else { |
3d9f2d2a KW |
3672 | error_setg(errp, "Image format driver does not support resize"); |
3673 | ret = -ENOTSUP; | |
3674 | goto out; | |
3675 | } | |
3d9f2d2a KW |
3676 | if (ret < 0) { |
3677 | goto out; | |
3678 | } | |
6b7e8f8b | 3679 | |
bd53086e | 3680 | ret = bdrv_co_refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); |
3d9f2d2a KW |
3681 | if (ret < 0) { |
3682 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
3683 | } else { | |
3684 | offset = bs->total_sectors * BDRV_SECTOR_SIZE; | |
3685 | } | |
c057960c EGE |
3686 | /* |
3687 | * It's possible that truncation succeeded but bdrv_refresh_total_sectors | |
cd47d792 | 3688 | * failed, but the latter doesn't affect how we should finish the request. |
c057960c EGE |
3689 | * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. |
3690 | */ | |
cd47d792 | 3691 | bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0); |
3d9f2d2a KW |
3692 | |
3693 | out: | |
1bc5f09f | 3694 | tracked_request_end(&req); |
3d9f2d2a | 3695 | bdrv_dec_in_flight(bs); |
1bc5f09f | 3696 | |
3d9f2d2a KW |
3697 | return ret; |
3698 | } | |
bd54669a VSO |
3699 | |
3700 | void bdrv_cancel_in_flight(BlockDriverState *bs) | |
3701 | { | |
f791bf7f | 3702 | GLOBAL_STATE_CODE(); |
bd54669a VSO |
3703 | if (!bs || !bs->drv) { |
3704 | return; | |
3705 | } | |
3706 | ||
3707 | if (bs->drv->bdrv_cancel_in_flight) { | |
3708 | bs->drv->bdrv_cancel_in_flight(bs); | |
3709 | } | |
3710 | } | |
ce14f3b4 VSO |
3711 | |
3712 | int coroutine_fn | |
3713 | bdrv_co_preadv_snapshot(BdrvChild *child, int64_t offset, int64_t bytes, | |
3714 | QEMUIOVector *qiov, size_t qiov_offset) | |
3715 | { | |
3716 | BlockDriverState *bs = child->bs; | |
3717 | BlockDriver *drv = bs->drv; | |
3718 | int ret; | |
3719 | IO_CODE(); | |
7b9e8b22 | 3720 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3721 | |
3722 | if (!drv) { | |
3723 | return -ENOMEDIUM; | |
3724 | } | |
3725 | ||
3726 | if (!drv->bdrv_co_preadv_snapshot) { | |
3727 | return -ENOTSUP; | |
3728 | } | |
3729 | ||
3730 | bdrv_inc_in_flight(bs); | |
3731 | ret = drv->bdrv_co_preadv_snapshot(bs, offset, bytes, qiov, qiov_offset); | |
3732 | bdrv_dec_in_flight(bs); | |
3733 | ||
3734 | return ret; | |
3735 | } | |
3736 | ||
3737 | int coroutine_fn | |
3738 | bdrv_co_snapshot_block_status(BlockDriverState *bs, | |
3739 | bool want_zero, int64_t offset, int64_t bytes, | |
3740 | int64_t *pnum, int64_t *map, | |
3741 | BlockDriverState **file) | |
3742 | { | |
3743 | BlockDriver *drv = bs->drv; | |
3744 | int ret; | |
3745 | IO_CODE(); | |
7b9e8b22 | 3746 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3747 | |
3748 | if (!drv) { | |
3749 | return -ENOMEDIUM; | |
3750 | } | |
3751 | ||
3752 | if (!drv->bdrv_co_snapshot_block_status) { | |
3753 | return -ENOTSUP; | |
3754 | } | |
3755 | ||
3756 | bdrv_inc_in_flight(bs); | |
3757 | ret = drv->bdrv_co_snapshot_block_status(bs, want_zero, offset, bytes, | |
3758 | pnum, map, file); | |
3759 | bdrv_dec_in_flight(bs); | |
3760 | ||
3761 | return ret; | |
3762 | } | |
3763 | ||
3764 | int coroutine_fn | |
3765 | bdrv_co_pdiscard_snapshot(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
3766 | { | |
3767 | BlockDriver *drv = bs->drv; | |
3768 | int ret; | |
3769 | IO_CODE(); | |
9a5a1c62 | 3770 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3771 | |
3772 | if (!drv) { | |
3773 | return -ENOMEDIUM; | |
3774 | } | |
3775 | ||
3776 | if (!drv->bdrv_co_pdiscard_snapshot) { | |
3777 | return -ENOTSUP; | |
3778 | } | |
3779 | ||
3780 | bdrv_inc_in_flight(bs); | |
3781 | ret = drv->bdrv_co_pdiscard_snapshot(bs, offset, bytes); | |
3782 | bdrv_dec_in_flight(bs); | |
3783 | ||
3784 | return ret; | |
3785 | } |