]>
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 | |
d05ab380 EGE |
49 | static void GRAPH_RDLOCK |
50 | bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore) | |
61007b31 | 51 | { |
02d21300 | 52 | BdrvChild *c, *next; |
d05ab380 EGE |
53 | IO_OR_GS_CODE(); |
54 | assert_bdrv_graph_readable(); | |
27ccdd52 | 55 | |
02d21300 | 56 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { |
a82a3bd1 | 57 | if (c == ignore) { |
0152bf40 KW |
58 | continue; |
59 | } | |
606ed756 | 60 | bdrv_parent_drained_begin_single(c); |
ce0f1412 PB |
61 | } |
62 | } | |
61007b31 | 63 | |
2f65df6e | 64 | void bdrv_parent_drained_end_single(BdrvChild *c) |
804db8ea | 65 | { |
ab613350 | 66 | GLOBAL_STATE_CODE(); |
2f65df6e | 67 | |
57e05be3 KW |
68 | assert(c->quiesced_parent); |
69 | c->quiesced_parent = false; | |
70 | ||
bd86fb99 | 71 | if (c->klass->drained_end) { |
2f65df6e | 72 | c->klass->drained_end(c); |
804db8ea HR |
73 | } |
74 | } | |
75 | ||
d05ab380 EGE |
76 | static void GRAPH_RDLOCK |
77 | bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore) | |
ce0f1412 | 78 | { |
61ad631c | 79 | BdrvChild *c; |
d05ab380 EGE |
80 | IO_OR_GS_CODE(); |
81 | assert_bdrv_graph_readable(); | |
27ccdd52 | 82 | |
61ad631c | 83 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
a82a3bd1 | 84 | if (c == ignore) { |
0152bf40 KW |
85 | continue; |
86 | } | |
2f65df6e | 87 | bdrv_parent_drained_end_single(c); |
27ccdd52 | 88 | } |
61007b31 SH |
89 | } |
90 | ||
23987471 | 91 | bool bdrv_parent_drained_poll_single(BdrvChild *c) |
4be6a6d1 | 92 | { |
d05ab380 EGE |
93 | IO_OR_GS_CODE(); |
94 | ||
bd86fb99 HR |
95 | if (c->klass->drained_poll) { |
96 | return c->klass->drained_poll(c); | |
4be6a6d1 KW |
97 | } |
98 | return false; | |
99 | } | |
100 | ||
d05ab380 EGE |
101 | static bool GRAPH_RDLOCK |
102 | bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore, | |
103 | bool ignore_bds_parents) | |
89bd0305 KW |
104 | { |
105 | BdrvChild *c, *next; | |
106 | bool busy = false; | |
d05ab380 EGE |
107 | IO_OR_GS_CODE(); |
108 | assert_bdrv_graph_readable(); | |
89bd0305 KW |
109 | |
110 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { | |
bd86fb99 | 111 | if (c == ignore || (ignore_bds_parents && c->klass->parent_is_bds)) { |
89bd0305 KW |
112 | continue; |
113 | } | |
4be6a6d1 | 114 | busy |= bdrv_parent_drained_poll_single(c); |
89bd0305 KW |
115 | } |
116 | ||
117 | return busy; | |
118 | } | |
119 | ||
606ed756 | 120 | void bdrv_parent_drained_begin_single(BdrvChild *c) |
4be6a6d1 | 121 | { |
ab613350 | 122 | GLOBAL_STATE_CODE(); |
57e05be3 KW |
123 | |
124 | assert(!c->quiesced_parent); | |
125 | c->quiesced_parent = true; | |
126 | ||
bd86fb99 | 127 | if (c->klass->drained_begin) { |
d05ab380 | 128 | /* called with rdlock taken, but it doesn't really need it. */ |
bd86fb99 | 129 | c->klass->drained_begin(c); |
4be6a6d1 | 130 | } |
4be6a6d1 KW |
131 | } |
132 | ||
d9e0dfa2 EB |
133 | static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) |
134 | { | |
9f460c64 AO |
135 | dst->pdiscard_alignment = MAX(dst->pdiscard_alignment, |
136 | src->pdiscard_alignment); | |
d9e0dfa2 EB |
137 | dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); |
138 | dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer); | |
24b36e98 PB |
139 | dst->max_hw_transfer = MIN_NON_ZERO(dst->max_hw_transfer, |
140 | src->max_hw_transfer); | |
d9e0dfa2 EB |
141 | dst->opt_mem_alignment = MAX(dst->opt_mem_alignment, |
142 | src->opt_mem_alignment); | |
143 | dst->min_mem_alignment = MAX(dst->min_mem_alignment, | |
144 | src->min_mem_alignment); | |
145 | dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov); | |
cc071629 | 146 | dst->max_hw_iov = MIN_NON_ZERO(dst->max_hw_iov, src->max_hw_iov); |
d9e0dfa2 EB |
147 | } |
148 | ||
1e4c797c VSO |
149 | typedef struct BdrvRefreshLimitsState { |
150 | BlockDriverState *bs; | |
151 | BlockLimits old_bl; | |
152 | } BdrvRefreshLimitsState; | |
153 | ||
154 | static void bdrv_refresh_limits_abort(void *opaque) | |
155 | { | |
156 | BdrvRefreshLimitsState *s = opaque; | |
157 | ||
158 | s->bs->bl = s->old_bl; | |
159 | } | |
160 | ||
161 | static TransactionActionDrv bdrv_refresh_limits_drv = { | |
162 | .abort = bdrv_refresh_limits_abort, | |
163 | .clean = g_free, | |
164 | }; | |
165 | ||
166 | /* @tran is allowed to be NULL, in this case no rollback is possible. */ | |
167 | void bdrv_refresh_limits(BlockDriverState *bs, Transaction *tran, Error **errp) | |
61007b31 | 168 | { |
33985614 | 169 | ERRP_GUARD(); |
61007b31 | 170 | BlockDriver *drv = bs->drv; |
66b129ac HR |
171 | BdrvChild *c; |
172 | bool have_limits; | |
61007b31 | 173 | |
f791bf7f EGE |
174 | GLOBAL_STATE_CODE(); |
175 | ||
1e4c797c VSO |
176 | if (tran) { |
177 | BdrvRefreshLimitsState *s = g_new(BdrvRefreshLimitsState, 1); | |
178 | *s = (BdrvRefreshLimitsState) { | |
179 | .bs = bs, | |
180 | .old_bl = bs->bl, | |
181 | }; | |
182 | tran_add(tran, &bdrv_refresh_limits_drv, s); | |
183 | } | |
184 | ||
61007b31 SH |
185 | memset(&bs->bl, 0, sizeof(bs->bl)); |
186 | ||
187 | if (!drv) { | |
188 | return; | |
189 | } | |
190 | ||
79ba8c98 | 191 | /* Default alignment based on whether driver has byte interface */ |
e31f6864 | 192 | bs->bl.request_alignment = (drv->bdrv_co_preadv || |
ac850bf0 VSO |
193 | drv->bdrv_aio_preadv || |
194 | drv->bdrv_co_preadv_part) ? 1 : 512; | |
79ba8c98 | 195 | |
61007b31 | 196 | /* Take some limits from the children as a default */ |
66b129ac HR |
197 | have_limits = false; |
198 | QLIST_FOREACH(c, &bs->children, next) { | |
199 | if (c->role & (BDRV_CHILD_DATA | BDRV_CHILD_FILTERED | BDRV_CHILD_COW)) | |
200 | { | |
66b129ac HR |
201 | bdrv_merge_limits(&bs->bl, &c->bs->bl); |
202 | have_limits = true; | |
61007b31 | 203 | } |
160a29e2 PB |
204 | |
205 | if (c->role & BDRV_CHILD_FILTERED) { | |
206 | bs->bl.has_variable_length |= c->bs->bl.has_variable_length; | |
207 | } | |
66b129ac HR |
208 | } |
209 | ||
210 | if (!have_limits) { | |
4196d2f0 | 211 | bs->bl.min_mem_alignment = 512; |
8e3b0cbb | 212 | bs->bl.opt_mem_alignment = qemu_real_host_page_size(); |
bd44feb7 SH |
213 | |
214 | /* Safe default since most protocols use readv()/writev()/etc */ | |
215 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
216 | } |
217 | ||
61007b31 SH |
218 | /* Then let the driver override it */ |
219 | if (drv->bdrv_refresh_limits) { | |
220 | drv->bdrv_refresh_limits(bs, errp); | |
8b117001 VSO |
221 | if (*errp) { |
222 | return; | |
223 | } | |
224 | } | |
225 | ||
226 | if (bs->bl.request_alignment > BDRV_MAX_ALIGNMENT) { | |
227 | error_setg(errp, "Driver requires too large request alignment"); | |
61007b31 SH |
228 | } |
229 | } | |
230 | ||
231 | /** | |
232 | * The copy-on-read flag is actually a reference count so multiple users may | |
233 | * use the feature without worrying about clobbering its previous state. | |
234 | * Copy-on-read stays enabled until all users have called to disable it. | |
235 | */ | |
236 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
237 | { | |
384a48fb | 238 | IO_CODE(); |
d73415a3 | 239 | qatomic_inc(&bs->copy_on_read); |
61007b31 SH |
240 | } |
241 | ||
242 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
243 | { | |
d73415a3 | 244 | int old = qatomic_fetch_dec(&bs->copy_on_read); |
384a48fb | 245 | IO_CODE(); |
d3faa13e | 246 | assert(old >= 1); |
61007b31 SH |
247 | } |
248 | ||
61124f03 PB |
249 | typedef struct { |
250 | Coroutine *co; | |
251 | BlockDriverState *bs; | |
252 | bool done; | |
481cad48 | 253 | bool begin; |
fe4f0614 | 254 | bool poll; |
0152bf40 | 255 | BdrvChild *parent; |
61124f03 PB |
256 | } BdrvCoDrainData; |
257 | ||
1cc8e54a | 258 | /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ |
299403ae KW |
259 | bool bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent, |
260 | bool ignore_bds_parents) | |
89bd0305 | 261 | { |
ab613350 | 262 | GLOBAL_STATE_CODE(); |
fe4f0614 | 263 | |
6cd5c9d7 | 264 | if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) { |
89bd0305 KW |
265 | return true; |
266 | } | |
267 | ||
d73415a3 | 268 | if (qatomic_read(&bs->in_flight)) { |
fe4f0614 KW |
269 | return true; |
270 | } | |
271 | ||
fe4f0614 | 272 | return false; |
89bd0305 KW |
273 | } |
274 | ||
299403ae | 275 | static bool bdrv_drain_poll_top_level(BlockDriverState *bs, |
89bd0305 | 276 | BdrvChild *ignore_parent) |
1cc8e54a | 277 | { |
d05ab380 EGE |
278 | GLOBAL_STATE_CODE(); |
279 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
280 | ||
299403ae | 281 | return bdrv_drain_poll(bs, ignore_parent, false); |
1cc8e54a KW |
282 | } |
283 | ||
299403ae | 284 | static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent, |
a82a3bd1 KW |
285 | bool poll); |
286 | static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent); | |
0152bf40 | 287 | |
a77fd4bb FZ |
288 | static void bdrv_co_drain_bh_cb(void *opaque) |
289 | { | |
290 | BdrvCoDrainData *data = opaque; | |
291 | Coroutine *co = data->co; | |
99723548 | 292 | BlockDriverState *bs = data->bs; |
a77fd4bb | 293 | |
c8ca33d0 | 294 | if (bs) { |
aa1361d5 | 295 | AioContext *ctx = bdrv_get_aio_context(bs); |
960d5fb3 | 296 | aio_context_acquire(ctx); |
c8ca33d0 KW |
297 | bdrv_dec_in_flight(bs); |
298 | if (data->begin) { | |
a82a3bd1 | 299 | bdrv_do_drained_begin(bs, data->parent, data->poll); |
c8ca33d0 | 300 | } else { |
e037c09c | 301 | assert(!data->poll); |
a82a3bd1 | 302 | bdrv_do_drained_end(bs, data->parent); |
c8ca33d0 | 303 | } |
960d5fb3 | 304 | aio_context_release(ctx); |
481cad48 | 305 | } else { |
c8ca33d0 KW |
306 | assert(data->begin); |
307 | bdrv_drain_all_begin(); | |
481cad48 MP |
308 | } |
309 | ||
a77fd4bb | 310 | data->done = true; |
1919631e | 311 | aio_co_wake(co); |
a77fd4bb FZ |
312 | } |
313 | ||
481cad48 | 314 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, |
299403ae | 315 | bool begin, |
6cd5c9d7 | 316 | BdrvChild *parent, |
2f65df6e | 317 | bool poll) |
a77fd4bb FZ |
318 | { |
319 | BdrvCoDrainData data; | |
960d5fb3 KW |
320 | Coroutine *self = qemu_coroutine_self(); |
321 | AioContext *ctx = bdrv_get_aio_context(bs); | |
322 | AioContext *co_ctx = qemu_coroutine_get_aio_context(self); | |
a77fd4bb FZ |
323 | |
324 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
c40a2545 | 325 | * other coroutines run if they were queued by aio_co_enter(). */ |
a77fd4bb FZ |
326 | |
327 | assert(qemu_in_coroutine()); | |
328 | data = (BdrvCoDrainData) { | |
960d5fb3 | 329 | .co = self, |
a77fd4bb FZ |
330 | .bs = bs, |
331 | .done = false, | |
481cad48 | 332 | .begin = begin, |
0152bf40 | 333 | .parent = parent, |
fe4f0614 | 334 | .poll = poll, |
a77fd4bb | 335 | }; |
8e1da77e | 336 | |
c8ca33d0 KW |
337 | if (bs) { |
338 | bdrv_inc_in_flight(bs); | |
339 | } | |
960d5fb3 KW |
340 | |
341 | /* | |
342 | * Temporarily drop the lock across yield or we would get deadlocks. | |
343 | * bdrv_co_drain_bh_cb() reaquires the lock as needed. | |
344 | * | |
345 | * When we yield below, the lock for the current context will be | |
346 | * released, so if this is actually the lock that protects bs, don't drop | |
347 | * it a second time. | |
348 | */ | |
349 | if (ctx != co_ctx) { | |
350 | aio_context_release(ctx); | |
351 | } | |
ab613350 SH |
352 | replay_bh_schedule_oneshot_event(qemu_get_aio_context(), |
353 | bdrv_co_drain_bh_cb, &data); | |
a77fd4bb FZ |
354 | |
355 | qemu_coroutine_yield(); | |
356 | /* If we are resumed from some other event (such as an aio completion or a | |
357 | * timer callback), it is a bug in the caller that should be fixed. */ | |
358 | assert(data.done); | |
960d5fb3 | 359 | |
3202d8e4 | 360 | /* Reacquire the AioContext of bs if we dropped it */ |
960d5fb3 KW |
361 | if (ctx != co_ctx) { |
362 | aio_context_acquire(ctx); | |
363 | } | |
a77fd4bb FZ |
364 | } |
365 | ||
05c272ff KW |
366 | static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent, |
367 | bool poll) | |
6820643f | 368 | { |
384a48fb | 369 | IO_OR_GS_CODE(); |
05c272ff KW |
370 | |
371 | if (qemu_in_coroutine()) { | |
372 | bdrv_co_yield_to_drain(bs, true, parent, poll); | |
373 | return; | |
374 | } | |
d42cf288 | 375 | |
ab613350 SH |
376 | GLOBAL_STATE_CODE(); |
377 | ||
60369b86 | 378 | /* Stop things in parent-to-child order */ |
d73415a3 | 379 | if (qatomic_fetch_inc(&bs->quiesce_counter) == 0) { |
d05ab380 | 380 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
a82a3bd1 | 381 | bdrv_parent_drained_begin(bs, parent); |
57e05be3 KW |
382 | if (bs->drv && bs->drv->bdrv_drain_begin) { |
383 | bs->drv->bdrv_drain_begin(bs); | |
384 | } | |
c7bc05f7 | 385 | } |
d30b8e64 | 386 | |
fe4f0614 KW |
387 | /* |
388 | * Wait for drained requests to finish. | |
389 | * | |
390 | * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The | |
391 | * call is needed so things in this AioContext can make progress even | |
392 | * though we don't return to the main AioContext loop - this automatically | |
393 | * includes other nodes in the same AioContext and therefore all child | |
394 | * nodes. | |
395 | */ | |
396 | if (poll) { | |
299403ae | 397 | BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, parent)); |
fe4f0614 | 398 | } |
6820643f KW |
399 | } |
400 | ||
05c272ff KW |
401 | void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, BdrvChild *parent) |
402 | { | |
403 | bdrv_do_drained_begin(bs, parent, false); | |
404 | } | |
405 | ||
e2dbca03 PB |
406 | void coroutine_mixed_fn |
407 | bdrv_drained_begin(BlockDriverState *bs) | |
0152bf40 | 408 | { |
384a48fb | 409 | IO_OR_GS_CODE(); |
a82a3bd1 | 410 | bdrv_do_drained_begin(bs, NULL, true); |
0152bf40 KW |
411 | } |
412 | ||
e037c09c HR |
413 | /** |
414 | * This function does not poll, nor must any of its recursively called | |
2f65df6e | 415 | * functions. |
e037c09c | 416 | */ |
a82a3bd1 | 417 | static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent) |
6820643f | 418 | { |
0f115168 KW |
419 | int old_quiesce_counter; |
420 | ||
ab613350 SH |
421 | IO_OR_GS_CODE(); |
422 | ||
481cad48 | 423 | if (qemu_in_coroutine()) { |
a82a3bd1 | 424 | bdrv_co_yield_to_drain(bs, false, parent, false); |
481cad48 MP |
425 | return; |
426 | } | |
d05ab380 EGE |
427 | |
428 | /* At this point, we should be always running in the main loop. */ | |
429 | GLOBAL_STATE_CODE(); | |
6820643f | 430 | assert(bs->quiesce_counter > 0); |
ab613350 | 431 | GLOBAL_STATE_CODE(); |
6820643f | 432 | |
60369b86 | 433 | /* Re-enable things in child-to-parent order */ |
d73415a3 | 434 | old_quiesce_counter = qatomic_fetch_dec(&bs->quiesce_counter); |
0f115168 | 435 | if (old_quiesce_counter == 1) { |
d05ab380 | 436 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
57e05be3 KW |
437 | if (bs->drv && bs->drv->bdrv_drain_end) { |
438 | bs->drv->bdrv_drain_end(bs); | |
439 | } | |
a82a3bd1 | 440 | bdrv_parent_drained_end(bs, parent); |
0f115168 | 441 | } |
6820643f KW |
442 | } |
443 | ||
0152bf40 KW |
444 | void bdrv_drained_end(BlockDriverState *bs) |
445 | { | |
384a48fb | 446 | IO_OR_GS_CODE(); |
a82a3bd1 | 447 | bdrv_do_drained_end(bs, NULL); |
d736f119 KW |
448 | } |
449 | ||
b6e84c97 PB |
450 | void bdrv_drain(BlockDriverState *bs) |
451 | { | |
384a48fb | 452 | IO_OR_GS_CODE(); |
6820643f KW |
453 | bdrv_drained_begin(bs); |
454 | bdrv_drained_end(bs); | |
61007b31 SH |
455 | } |
456 | ||
c13ad59f KW |
457 | static void bdrv_drain_assert_idle(BlockDriverState *bs) |
458 | { | |
459 | BdrvChild *child, *next; | |
d05ab380 EGE |
460 | GLOBAL_STATE_CODE(); |
461 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
c13ad59f | 462 | |
d73415a3 | 463 | assert(qatomic_read(&bs->in_flight) == 0); |
c13ad59f KW |
464 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
465 | bdrv_drain_assert_idle(child->bs); | |
466 | } | |
467 | } | |
468 | ||
0f12264e KW |
469 | unsigned int bdrv_drain_all_count = 0; |
470 | ||
471 | static bool bdrv_drain_all_poll(void) | |
472 | { | |
473 | BlockDriverState *bs = NULL; | |
474 | bool result = false; | |
d05ab380 | 475 | |
f791bf7f | 476 | GLOBAL_STATE_CODE(); |
d05ab380 | 477 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
0f12264e | 478 | |
0f12264e KW |
479 | /* bdrv_drain_poll() can't make changes to the graph and we are holding the |
480 | * main AioContext lock, so iterating bdrv_next_all_states() is safe. */ | |
481 | while ((bs = bdrv_next_all_states(bs))) { | |
482 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
483 | aio_context_acquire(aio_context); | |
299403ae | 484 | result |= bdrv_drain_poll(bs, NULL, true); |
0f12264e KW |
485 | aio_context_release(aio_context); |
486 | } | |
487 | ||
488 | return result; | |
489 | } | |
490 | ||
61007b31 SH |
491 | /* |
492 | * Wait for pending requests to complete across all BlockDriverStates | |
493 | * | |
494 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
495 | * after calling this function. | |
c0778f66 AG |
496 | * |
497 | * This pauses all block jobs and disables external clients. It must | |
498 | * be paired with bdrv_drain_all_end(). | |
499 | * | |
500 | * NOTE: no new block jobs or BlockDriverStates can be created between | |
501 | * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls. | |
61007b31 | 502 | */ |
da0bd744 | 503 | void bdrv_drain_all_begin_nopoll(void) |
61007b31 | 504 | { |
0f12264e | 505 | BlockDriverState *bs = NULL; |
f791bf7f | 506 | GLOBAL_STATE_CODE(); |
61007b31 | 507 | |
c8aa7895 PD |
508 | /* |
509 | * bdrv queue is managed by record/replay, | |
510 | * waiting for finishing the I/O requests may | |
511 | * be infinite | |
512 | */ | |
513 | if (replay_events_enabled()) { | |
514 | return; | |
515 | } | |
516 | ||
0f12264e KW |
517 | /* AIO_WAIT_WHILE() with a NULL context can only be called from the main |
518 | * loop AioContext, so make sure we're in the main context. */ | |
9a7e86c8 | 519 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
520 | assert(bdrv_drain_all_count < INT_MAX); |
521 | bdrv_drain_all_count++; | |
9a7e86c8 | 522 | |
0f12264e KW |
523 | /* Quiesce all nodes, without polling in-flight requests yet. The graph |
524 | * cannot change during this loop. */ | |
525 | while ((bs = bdrv_next_all_states(bs))) { | |
61007b31 SH |
526 | AioContext *aio_context = bdrv_get_aio_context(bs); |
527 | ||
528 | aio_context_acquire(aio_context); | |
a82a3bd1 | 529 | bdrv_do_drained_begin(bs, NULL, false); |
61007b31 SH |
530 | aio_context_release(aio_context); |
531 | } | |
da0bd744 KW |
532 | } |
533 | ||
e2dbca03 | 534 | void coroutine_mixed_fn bdrv_drain_all_begin(void) |
da0bd744 KW |
535 | { |
536 | BlockDriverState *bs = NULL; | |
537 | ||
538 | if (qemu_in_coroutine()) { | |
539 | bdrv_co_yield_to_drain(NULL, true, NULL, true); | |
540 | return; | |
541 | } | |
542 | ||
63945789 PM |
543 | /* |
544 | * bdrv queue is managed by record/replay, | |
545 | * waiting for finishing the I/O requests may | |
546 | * be infinite | |
547 | */ | |
548 | if (replay_events_enabled()) { | |
549 | return; | |
550 | } | |
551 | ||
da0bd744 | 552 | bdrv_drain_all_begin_nopoll(); |
61007b31 | 553 | |
0f12264e | 554 | /* Now poll the in-flight requests */ |
263d5e12 | 555 | AIO_WAIT_WHILE_UNLOCKED(NULL, bdrv_drain_all_poll()); |
0f12264e KW |
556 | |
557 | while ((bs = bdrv_next_all_states(bs))) { | |
c13ad59f | 558 | bdrv_drain_assert_idle(bs); |
61007b31 | 559 | } |
c0778f66 AG |
560 | } |
561 | ||
1a6d3bd2 GK |
562 | void bdrv_drain_all_end_quiesce(BlockDriverState *bs) |
563 | { | |
b4ad82aa | 564 | GLOBAL_STATE_CODE(); |
1a6d3bd2 GK |
565 | |
566 | g_assert(bs->quiesce_counter > 0); | |
567 | g_assert(!bs->refcnt); | |
568 | ||
569 | while (bs->quiesce_counter) { | |
a82a3bd1 | 570 | bdrv_do_drained_end(bs, NULL); |
1a6d3bd2 | 571 | } |
1a6d3bd2 GK |
572 | } |
573 | ||
c0778f66 AG |
574 | void bdrv_drain_all_end(void) |
575 | { | |
0f12264e | 576 | BlockDriverState *bs = NULL; |
f791bf7f | 577 | GLOBAL_STATE_CODE(); |
c0778f66 | 578 | |
c8aa7895 PD |
579 | /* |
580 | * bdrv queue is managed by record/replay, | |
581 | * waiting for finishing the I/O requests may | |
582 | * be endless | |
583 | */ | |
584 | if (replay_events_enabled()) { | |
585 | return; | |
586 | } | |
587 | ||
0f12264e | 588 | while ((bs = bdrv_next_all_states(bs))) { |
61007b31 SH |
589 | AioContext *aio_context = bdrv_get_aio_context(bs); |
590 | ||
591 | aio_context_acquire(aio_context); | |
a82a3bd1 | 592 | bdrv_do_drained_end(bs, NULL); |
61007b31 SH |
593 | aio_context_release(aio_context); |
594 | } | |
0f12264e | 595 | |
e037c09c | 596 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
597 | assert(bdrv_drain_all_count > 0); |
598 | bdrv_drain_all_count--; | |
61007b31 SH |
599 | } |
600 | ||
c0778f66 AG |
601 | void bdrv_drain_all(void) |
602 | { | |
f791bf7f | 603 | GLOBAL_STATE_CODE(); |
c0778f66 AG |
604 | bdrv_drain_all_begin(); |
605 | bdrv_drain_all_end(); | |
606 | } | |
607 | ||
61007b31 SH |
608 | /** |
609 | * Remove an active request from the tracked requests list | |
610 | * | |
611 | * This function should be called when a tracked request is completing. | |
612 | */ | |
f0d43b1e | 613 | static void coroutine_fn tracked_request_end(BdrvTrackedRequest *req) |
61007b31 SH |
614 | { |
615 | if (req->serialising) { | |
d73415a3 | 616 | qatomic_dec(&req->bs->serialising_in_flight); |
61007b31 SH |
617 | } |
618 | ||
fa9185fc | 619 | qemu_mutex_lock(&req->bs->reqs_lock); |
61007b31 | 620 | QLIST_REMOVE(req, list); |
fa9185fc | 621 | qemu_mutex_unlock(&req->bs->reqs_lock); |
3480ce69 SH |
622 | |
623 | /* | |
624 | * At this point qemu_co_queue_wait(&req->wait_queue, ...) won't be called | |
625 | * anymore because the request has been removed from the list, so it's safe | |
626 | * to restart the queue outside reqs_lock to minimize the critical section. | |
627 | */ | |
61007b31 SH |
628 | qemu_co_queue_restart_all(&req->wait_queue); |
629 | } | |
630 | ||
631 | /** | |
632 | * Add an active request to the tracked requests list | |
633 | */ | |
881a4c55 PB |
634 | static void coroutine_fn tracked_request_begin(BdrvTrackedRequest *req, |
635 | BlockDriverState *bs, | |
636 | int64_t offset, | |
637 | int64_t bytes, | |
638 | enum BdrvTrackedRequestType type) | |
61007b31 | 639 | { |
80247264 | 640 | bdrv_check_request(offset, bytes, &error_abort); |
22931a15 | 641 | |
61007b31 SH |
642 | *req = (BdrvTrackedRequest){ |
643 | .bs = bs, | |
644 | .offset = offset, | |
645 | .bytes = bytes, | |
ebde595c | 646 | .type = type, |
61007b31 SH |
647 | .co = qemu_coroutine_self(), |
648 | .serialising = false, | |
649 | .overlap_offset = offset, | |
650 | .overlap_bytes = bytes, | |
651 | }; | |
652 | ||
653 | qemu_co_queue_init(&req->wait_queue); | |
654 | ||
fa9185fc | 655 | qemu_mutex_lock(&bs->reqs_lock); |
61007b31 | 656 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
fa9185fc | 657 | qemu_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
658 | } |
659 | ||
3ba0e1a0 | 660 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, |
80247264 | 661 | int64_t offset, int64_t bytes) |
3ba0e1a0 | 662 | { |
80247264 EB |
663 | bdrv_check_request(offset, bytes, &error_abort); |
664 | ||
3ba0e1a0 PB |
665 | /* aaaa bbbb */ |
666 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
667 | return false; | |
668 | } | |
669 | /* bbbb aaaa */ | |
670 | if (req->overlap_offset >= offset + bytes) { | |
671 | return false; | |
672 | } | |
673 | return true; | |
674 | } | |
675 | ||
3183937f | 676 | /* Called with self->bs->reqs_lock held */ |
881a4c55 | 677 | static coroutine_fn BdrvTrackedRequest * |
3183937f VSO |
678 | bdrv_find_conflicting_request(BdrvTrackedRequest *self) |
679 | { | |
680 | BdrvTrackedRequest *req; | |
681 | ||
682 | QLIST_FOREACH(req, &self->bs->tracked_requests, list) { | |
683 | if (req == self || (!req->serialising && !self->serialising)) { | |
684 | continue; | |
685 | } | |
686 | if (tracked_request_overlaps(req, self->overlap_offset, | |
687 | self->overlap_bytes)) | |
688 | { | |
689 | /* | |
690 | * Hitting this means there was a reentrant request, for | |
691 | * example, a block driver issuing nested requests. This must | |
692 | * never happen since it means deadlock. | |
693 | */ | |
694 | assert(qemu_coroutine_self() != req->co); | |
695 | ||
696 | /* | |
697 | * If the request is already (indirectly) waiting for us, or | |
698 | * will wait for us as soon as it wakes up, then just go on | |
699 | * (instead of producing a deadlock in the former case). | |
700 | */ | |
701 | if (!req->waiting_for) { | |
702 | return req; | |
703 | } | |
704 | } | |
705 | } | |
706 | ||
707 | return NULL; | |
708 | } | |
709 | ||
ec1c8868 | 710 | /* Called with self->bs->reqs_lock held */ |
131498f7 | 711 | static void coroutine_fn |
ec1c8868 | 712 | bdrv_wait_serialising_requests_locked(BdrvTrackedRequest *self) |
3ba0e1a0 PB |
713 | { |
714 | BdrvTrackedRequest *req; | |
3ba0e1a0 | 715 | |
3183937f VSO |
716 | while ((req = bdrv_find_conflicting_request(self))) { |
717 | self->waiting_for = req; | |
ec1c8868 | 718 | qemu_co_queue_wait(&req->wait_queue, &self->bs->reqs_lock); |
3183937f | 719 | self->waiting_for = NULL; |
3183937f | 720 | } |
3ba0e1a0 PB |
721 | } |
722 | ||
8ac5aab2 VSO |
723 | /* Called with req->bs->reqs_lock held */ |
724 | static void tracked_request_set_serialising(BdrvTrackedRequest *req, | |
725 | uint64_t align) | |
61007b31 SH |
726 | { |
727 | int64_t overlap_offset = req->offset & ~(align - 1); | |
80247264 EB |
728 | int64_t overlap_bytes = |
729 | ROUND_UP(req->offset + req->bytes, align) - overlap_offset; | |
730 | ||
731 | bdrv_check_request(req->offset, req->bytes, &error_abort); | |
61007b31 SH |
732 | |
733 | if (!req->serialising) { | |
d73415a3 | 734 | qatomic_inc(&req->bs->serialising_in_flight); |
61007b31 SH |
735 | req->serialising = true; |
736 | } | |
737 | ||
738 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
739 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
09d2f948 VSO |
740 | } |
741 | ||
c28107e9 HR |
742 | /** |
743 | * Return the tracked request on @bs for the current coroutine, or | |
744 | * NULL if there is none. | |
745 | */ | |
746 | BdrvTrackedRequest *coroutine_fn bdrv_co_get_self_request(BlockDriverState *bs) | |
747 | { | |
748 | BdrvTrackedRequest *req; | |
749 | Coroutine *self = qemu_coroutine_self(); | |
967d7905 | 750 | IO_CODE(); |
c28107e9 HR |
751 | |
752 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
753 | if (req->co == self) { | |
754 | return req; | |
755 | } | |
756 | } | |
757 | ||
758 | return NULL; | |
759 | } | |
760 | ||
244483e6 | 761 | /** |
fc6b211f | 762 | * Round a region to subcluster (if supported) or cluster boundaries |
244483e6 | 763 | */ |
a00e70c0 | 764 | void coroutine_fn GRAPH_RDLOCK |
fc6b211f AD |
765 | bdrv_round_to_subclusters(BlockDriverState *bs, int64_t offset, int64_t bytes, |
766 | int64_t *align_offset, int64_t *align_bytes) | |
244483e6 KW |
767 | { |
768 | BlockDriverInfo bdi; | |
384a48fb | 769 | IO_CODE(); |
fc6b211f AD |
770 | if (bdrv_co_get_info(bs, &bdi) < 0 || bdi.subcluster_size == 0) { |
771 | *align_offset = offset; | |
772 | *align_bytes = bytes; | |
244483e6 | 773 | } else { |
fc6b211f AD |
774 | int64_t c = bdi.subcluster_size; |
775 | *align_offset = QEMU_ALIGN_DOWN(offset, c); | |
776 | *align_bytes = QEMU_ALIGN_UP(offset - *align_offset + bytes, c); | |
244483e6 KW |
777 | } |
778 | } | |
779 | ||
a00e70c0 | 780 | static int coroutine_fn GRAPH_RDLOCK bdrv_get_cluster_size(BlockDriverState *bs) |
61007b31 SH |
781 | { |
782 | BlockDriverInfo bdi; | |
783 | int ret; | |
784 | ||
3d47eb0a | 785 | ret = bdrv_co_get_info(bs, &bdi); |
61007b31 | 786 | if (ret < 0 || bdi.cluster_size == 0) { |
a5b8dd2c | 787 | return bs->bl.request_alignment; |
61007b31 SH |
788 | } else { |
789 | return bdi.cluster_size; | |
790 | } | |
791 | } | |
792 | ||
99723548 PB |
793 | void bdrv_inc_in_flight(BlockDriverState *bs) |
794 | { | |
967d7905 | 795 | IO_CODE(); |
d73415a3 | 796 | qatomic_inc(&bs->in_flight); |
99723548 PB |
797 | } |
798 | ||
c9d1a561 PB |
799 | void bdrv_wakeup(BlockDriverState *bs) |
800 | { | |
967d7905 | 801 | IO_CODE(); |
cfe29d82 | 802 | aio_wait_kick(); |
c9d1a561 PB |
803 | } |
804 | ||
99723548 PB |
805 | void bdrv_dec_in_flight(BlockDriverState *bs) |
806 | { | |
967d7905 | 807 | IO_CODE(); |
d73415a3 | 808 | qatomic_dec(&bs->in_flight); |
c9d1a561 | 809 | bdrv_wakeup(bs); |
99723548 PB |
810 | } |
811 | ||
131498f7 DL |
812 | static void coroutine_fn |
813 | bdrv_wait_serialising_requests(BdrvTrackedRequest *self) | |
61007b31 SH |
814 | { |
815 | BlockDriverState *bs = self->bs; | |
61007b31 | 816 | |
d73415a3 | 817 | if (!qatomic_read(&bs->serialising_in_flight)) { |
131498f7 | 818 | return; |
61007b31 SH |
819 | } |
820 | ||
fa9185fc | 821 | qemu_mutex_lock(&bs->reqs_lock); |
131498f7 | 822 | bdrv_wait_serialising_requests_locked(self); |
fa9185fc | 823 | qemu_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
824 | } |
825 | ||
131498f7 | 826 | void coroutine_fn bdrv_make_request_serialising(BdrvTrackedRequest *req, |
8ac5aab2 VSO |
827 | uint64_t align) |
828 | { | |
967d7905 | 829 | IO_CODE(); |
8ac5aab2 | 830 | |
fa9185fc | 831 | qemu_mutex_lock(&req->bs->reqs_lock); |
8ac5aab2 VSO |
832 | |
833 | tracked_request_set_serialising(req, align); | |
131498f7 | 834 | bdrv_wait_serialising_requests_locked(req); |
8ac5aab2 | 835 | |
fa9185fc | 836 | qemu_mutex_unlock(&req->bs->reqs_lock); |
8ac5aab2 VSO |
837 | } |
838 | ||
558902cc VSO |
839 | int bdrv_check_qiov_request(int64_t offset, int64_t bytes, |
840 | QEMUIOVector *qiov, size_t qiov_offset, | |
841 | Error **errp) | |
61007b31 | 842 | { |
63f4ad11 VSO |
843 | /* |
844 | * Check generic offset/bytes correctness | |
845 | */ | |
846 | ||
69b55e03 VSO |
847 | if (offset < 0) { |
848 | error_setg(errp, "offset is negative: %" PRIi64, offset); | |
849 | return -EIO; | |
850 | } | |
851 | ||
852 | if (bytes < 0) { | |
853 | error_setg(errp, "bytes is negative: %" PRIi64, bytes); | |
61007b31 SH |
854 | return -EIO; |
855 | } | |
856 | ||
8b117001 | 857 | if (bytes > BDRV_MAX_LENGTH) { |
69b55e03 VSO |
858 | error_setg(errp, "bytes(%" PRIi64 ") exceeds maximum(%" PRIi64 ")", |
859 | bytes, BDRV_MAX_LENGTH); | |
860 | return -EIO; | |
861 | } | |
862 | ||
863 | if (offset > BDRV_MAX_LENGTH) { | |
864 | error_setg(errp, "offset(%" PRIi64 ") exceeds maximum(%" PRIi64 ")", | |
865 | offset, BDRV_MAX_LENGTH); | |
8b117001 VSO |
866 | return -EIO; |
867 | } | |
868 | ||
869 | if (offset > BDRV_MAX_LENGTH - bytes) { | |
69b55e03 VSO |
870 | error_setg(errp, "sum of offset(%" PRIi64 ") and bytes(%" PRIi64 ") " |
871 | "exceeds maximum(%" PRIi64 ")", offset, bytes, | |
872 | BDRV_MAX_LENGTH); | |
8b117001 VSO |
873 | return -EIO; |
874 | } | |
875 | ||
63f4ad11 VSO |
876 | if (!qiov) { |
877 | return 0; | |
878 | } | |
879 | ||
880 | /* | |
881 | * Check qiov and qiov_offset | |
882 | */ | |
883 | ||
884 | if (qiov_offset > qiov->size) { | |
885 | error_setg(errp, "qiov_offset(%zu) overflow io vector size(%zu)", | |
886 | qiov_offset, qiov->size); | |
887 | return -EIO; | |
888 | } | |
889 | ||
890 | if (bytes > qiov->size - qiov_offset) { | |
891 | error_setg(errp, "bytes(%" PRIi64 ") + qiov_offset(%zu) overflow io " | |
892 | "vector size(%zu)", bytes, qiov_offset, qiov->size); | |
893 | return -EIO; | |
894 | } | |
895 | ||
8b117001 VSO |
896 | return 0; |
897 | } | |
898 | ||
63f4ad11 VSO |
899 | int bdrv_check_request(int64_t offset, int64_t bytes, Error **errp) |
900 | { | |
901 | return bdrv_check_qiov_request(offset, bytes, NULL, 0, errp); | |
902 | } | |
903 | ||
904 | static int bdrv_check_request32(int64_t offset, int64_t bytes, | |
905 | QEMUIOVector *qiov, size_t qiov_offset) | |
8b117001 | 906 | { |
63f4ad11 | 907 | int ret = bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, NULL); |
8b117001 VSO |
908 | if (ret < 0) { |
909 | return ret; | |
910 | } | |
911 | ||
912 | if (bytes > BDRV_REQUEST_MAX_BYTES) { | |
61007b31 SH |
913 | return -EIO; |
914 | } | |
915 | ||
916 | return 0; | |
917 | } | |
918 | ||
61007b31 | 919 | /* |
74021bc4 | 920 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
921 | * The operation is sped up by checking the block status and only writing |
922 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 923 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 924 | * BDRV_REQ_FUA). |
61007b31 | 925 | * |
f4649069 | 926 | * Returns < 0 on error, 0 on success. For error codes see bdrv_pwrite(). |
61007b31 | 927 | */ |
720ff280 | 928 | int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags) |
61007b31 | 929 | { |
237d78f8 EB |
930 | int ret; |
931 | int64_t target_size, bytes, offset = 0; | |
720ff280 | 932 | BlockDriverState *bs = child->bs; |
384a48fb | 933 | IO_CODE(); |
61007b31 | 934 | |
7286d610 EB |
935 | target_size = bdrv_getlength(bs); |
936 | if (target_size < 0) { | |
937 | return target_size; | |
61007b31 SH |
938 | } |
939 | ||
940 | for (;;) { | |
7286d610 EB |
941 | bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES); |
942 | if (bytes <= 0) { | |
61007b31 SH |
943 | return 0; |
944 | } | |
237d78f8 | 945 | ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL); |
61007b31 | 946 | if (ret < 0) { |
61007b31 SH |
947 | return ret; |
948 | } | |
949 | if (ret & BDRV_BLOCK_ZERO) { | |
237d78f8 | 950 | offset += bytes; |
61007b31 SH |
951 | continue; |
952 | } | |
237d78f8 | 953 | ret = bdrv_pwrite_zeroes(child, offset, bytes, flags); |
61007b31 | 954 | if (ret < 0) { |
61007b31 SH |
955 | return ret; |
956 | } | |
237d78f8 | 957 | offset += bytes; |
61007b31 SH |
958 | } |
959 | } | |
960 | ||
61007b31 SH |
961 | /* |
962 | * Writes to the file and ensures that no writes are reordered across this | |
963 | * request (acts as a barrier) | |
964 | * | |
965 | * Returns 0 on success, -errno in error cases. | |
966 | */ | |
e97190a4 AF |
967 | int coroutine_fn bdrv_co_pwrite_sync(BdrvChild *child, int64_t offset, |
968 | int64_t bytes, const void *buf, | |
969 | BdrvRequestFlags flags) | |
61007b31 SH |
970 | { |
971 | int ret; | |
384a48fb | 972 | IO_CODE(); |
b24a4c41 | 973 | assert_bdrv_graph_readable(); |
88095349 | 974 | |
e97190a4 | 975 | ret = bdrv_co_pwrite(child, offset, bytes, buf, flags); |
61007b31 SH |
976 | if (ret < 0) { |
977 | return ret; | |
978 | } | |
979 | ||
e97190a4 | 980 | ret = bdrv_co_flush(child->bs); |
855a6a93 KW |
981 | if (ret < 0) { |
982 | return ret; | |
61007b31 SH |
983 | } |
984 | ||
985 | return 0; | |
986 | } | |
987 | ||
08844473 KW |
988 | typedef struct CoroutineIOCompletion { |
989 | Coroutine *coroutine; | |
990 | int ret; | |
991 | } CoroutineIOCompletion; | |
992 | ||
993 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
994 | { | |
995 | CoroutineIOCompletion *co = opaque; | |
996 | ||
997 | co->ret = ret; | |
b9e413dd | 998 | aio_co_wake(co->coroutine); |
08844473 KW |
999 | } |
1000 | ||
7b1fb72e KW |
1001 | static int coroutine_fn GRAPH_RDLOCK |
1002 | bdrv_driver_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1003 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
166fe960 KW |
1004 | { |
1005 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
1006 | int64_t sector_num; |
1007 | unsigned int nb_sectors; | |
ac850bf0 VSO |
1008 | QEMUIOVector local_qiov; |
1009 | int ret; | |
b9b10c35 | 1010 | assert_bdrv_graph_readable(); |
3fb06697 | 1011 | |
17abcbee | 1012 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
e8b65355 | 1013 | assert(!(flags & ~bs->supported_read_flags)); |
fa166538 | 1014 | |
d470ad42 HR |
1015 | if (!drv) { |
1016 | return -ENOMEDIUM; | |
1017 | } | |
1018 | ||
ac850bf0 VSO |
1019 | if (drv->bdrv_co_preadv_part) { |
1020 | return drv->bdrv_co_preadv_part(bs, offset, bytes, qiov, qiov_offset, | |
1021 | flags); | |
1022 | } | |
1023 | ||
1024 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1025 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1026 | qiov = &local_qiov; | |
1027 | } | |
1028 | ||
3fb06697 | 1029 | if (drv->bdrv_co_preadv) { |
ac850bf0 VSO |
1030 | ret = drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); |
1031 | goto out; | |
3fb06697 KW |
1032 | } |
1033 | ||
edfab6a0 | 1034 | if (drv->bdrv_aio_preadv) { |
08844473 KW |
1035 | BlockAIOCB *acb; |
1036 | CoroutineIOCompletion co = { | |
1037 | .coroutine = qemu_coroutine_self(), | |
1038 | }; | |
1039 | ||
edfab6a0 EB |
1040 | acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags, |
1041 | bdrv_co_io_em_complete, &co); | |
08844473 | 1042 | if (acb == NULL) { |
ac850bf0 VSO |
1043 | ret = -EIO; |
1044 | goto out; | |
08844473 KW |
1045 | } else { |
1046 | qemu_coroutine_yield(); | |
ac850bf0 VSO |
1047 | ret = co.ret; |
1048 | goto out; | |
08844473 KW |
1049 | } |
1050 | } | |
edfab6a0 EB |
1051 | |
1052 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1053 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1054 | ||
1bbbf32d NS |
1055 | assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)); |
1056 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
41ae31e3 | 1057 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 EB |
1058 | assert(drv->bdrv_co_readv); |
1059 | ||
ac850bf0 VSO |
1060 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); |
1061 | ||
1062 | out: | |
1063 | if (qiov == &local_qiov) { | |
1064 | qemu_iovec_destroy(&local_qiov); | |
1065 | } | |
1066 | ||
1067 | return ret; | |
166fe960 KW |
1068 | } |
1069 | ||
7b1fb72e KW |
1070 | static int coroutine_fn GRAPH_RDLOCK |
1071 | bdrv_driver_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1072 | QEMUIOVector *qiov, size_t qiov_offset, | |
1073 | BdrvRequestFlags flags) | |
78a07294 KW |
1074 | { |
1075 | BlockDriver *drv = bs->drv; | |
e8b65355 | 1076 | bool emulate_fua = false; |
3fb06697 KW |
1077 | int64_t sector_num; |
1078 | unsigned int nb_sectors; | |
ac850bf0 | 1079 | QEMUIOVector local_qiov; |
78a07294 | 1080 | int ret; |
b9b10c35 | 1081 | assert_bdrv_graph_readable(); |
78a07294 | 1082 | |
17abcbee | 1083 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
fa166538 | 1084 | |
d470ad42 HR |
1085 | if (!drv) { |
1086 | return -ENOMEDIUM; | |
1087 | } | |
1088 | ||
e8b65355 SH |
1089 | if ((flags & BDRV_REQ_FUA) && |
1090 | (~bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1091 | flags &= ~BDRV_REQ_FUA; | |
1092 | emulate_fua = true; | |
1093 | } | |
1094 | ||
1095 | flags &= bs->supported_write_flags; | |
1096 | ||
ac850bf0 VSO |
1097 | if (drv->bdrv_co_pwritev_part) { |
1098 | ret = drv->bdrv_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, | |
e8b65355 | 1099 | flags); |
ac850bf0 VSO |
1100 | goto emulate_flags; |
1101 | } | |
1102 | ||
1103 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1104 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1105 | qiov = &local_qiov; | |
1106 | } | |
1107 | ||
3fb06697 | 1108 | if (drv->bdrv_co_pwritev) { |
e8b65355 | 1109 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, flags); |
3fb06697 KW |
1110 | goto emulate_flags; |
1111 | } | |
1112 | ||
edfab6a0 | 1113 | if (drv->bdrv_aio_pwritev) { |
08844473 KW |
1114 | BlockAIOCB *acb; |
1115 | CoroutineIOCompletion co = { | |
1116 | .coroutine = qemu_coroutine_self(), | |
1117 | }; | |
1118 | ||
e8b65355 | 1119 | acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov, flags, |
edfab6a0 | 1120 | bdrv_co_io_em_complete, &co); |
08844473 | 1121 | if (acb == NULL) { |
3fb06697 | 1122 | ret = -EIO; |
08844473 KW |
1123 | } else { |
1124 | qemu_coroutine_yield(); | |
3fb06697 | 1125 | ret = co.ret; |
08844473 | 1126 | } |
edfab6a0 EB |
1127 | goto emulate_flags; |
1128 | } | |
1129 | ||
1130 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1131 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1132 | ||
1bbbf32d NS |
1133 | assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)); |
1134 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
41ae31e3 | 1135 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 | 1136 | |
e18a58b4 | 1137 | assert(drv->bdrv_co_writev); |
e8b65355 | 1138 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov, flags); |
78a07294 | 1139 | |
3fb06697 | 1140 | emulate_flags: |
e8b65355 | 1141 | if (ret == 0 && emulate_fua) { |
78a07294 KW |
1142 | ret = bdrv_co_flush(bs); |
1143 | } | |
1144 | ||
ac850bf0 VSO |
1145 | if (qiov == &local_qiov) { |
1146 | qemu_iovec_destroy(&local_qiov); | |
1147 | } | |
1148 | ||
78a07294 KW |
1149 | return ret; |
1150 | } | |
1151 | ||
7b1fb72e | 1152 | static int coroutine_fn GRAPH_RDLOCK |
17abcbee VSO |
1153 | bdrv_driver_pwritev_compressed(BlockDriverState *bs, int64_t offset, |
1154 | int64_t bytes, QEMUIOVector *qiov, | |
ac850bf0 | 1155 | size_t qiov_offset) |
29a298af PB |
1156 | { |
1157 | BlockDriver *drv = bs->drv; | |
ac850bf0 VSO |
1158 | QEMUIOVector local_qiov; |
1159 | int ret; | |
b9b10c35 | 1160 | assert_bdrv_graph_readable(); |
29a298af | 1161 | |
17abcbee VSO |
1162 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1163 | ||
d470ad42 HR |
1164 | if (!drv) { |
1165 | return -ENOMEDIUM; | |
1166 | } | |
1167 | ||
ac850bf0 | 1168 | if (!block_driver_can_compress(drv)) { |
29a298af PB |
1169 | return -ENOTSUP; |
1170 | } | |
1171 | ||
ac850bf0 VSO |
1172 | if (drv->bdrv_co_pwritev_compressed_part) { |
1173 | return drv->bdrv_co_pwritev_compressed_part(bs, offset, bytes, | |
1174 | qiov, qiov_offset); | |
1175 | } | |
1176 | ||
1177 | if (qiov_offset == 0) { | |
1178 | return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov); | |
1179 | } | |
1180 | ||
1181 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1182 | ret = drv->bdrv_co_pwritev_compressed(bs, offset, bytes, &local_qiov); | |
1183 | qemu_iovec_destroy(&local_qiov); | |
1184 | ||
1185 | return ret; | |
29a298af PB |
1186 | } |
1187 | ||
7b1fb72e KW |
1188 | static int coroutine_fn GRAPH_RDLOCK |
1189 | bdrv_co_do_copy_on_readv(BdrvChild *child, int64_t offset, int64_t bytes, | |
1190 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
61007b31 | 1191 | { |
85c97ca7 KW |
1192 | BlockDriverState *bs = child->bs; |
1193 | ||
61007b31 SH |
1194 | /* Perform I/O through a temporary buffer so that users who scribble over |
1195 | * their read buffer while the operation is in progress do not end up | |
1196 | * modifying the image file. This is critical for zero-copy guest I/O | |
1197 | * where anything might happen inside guest memory. | |
1198 | */ | |
2275cc90 | 1199 | void *bounce_buffer = NULL; |
61007b31 SH |
1200 | |
1201 | BlockDriver *drv = bs->drv; | |
fc6b211f AD |
1202 | int64_t align_offset; |
1203 | int64_t align_bytes; | |
9df5afbd | 1204 | int64_t skip_bytes; |
61007b31 | 1205 | int ret; |
cb2e2878 EB |
1206 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
1207 | BDRV_REQUEST_MAX_BYTES); | |
9df5afbd | 1208 | int64_t progress = 0; |
8644476e | 1209 | bool skip_write; |
61007b31 | 1210 | |
9df5afbd VSO |
1211 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1212 | ||
d470ad42 HR |
1213 | if (!drv) { |
1214 | return -ENOMEDIUM; | |
1215 | } | |
1216 | ||
8644476e HR |
1217 | /* |
1218 | * Do not write anything when the BDS is inactive. That is not | |
1219 | * allowed, and it would not help. | |
1220 | */ | |
1221 | skip_write = (bs->open_flags & BDRV_O_INACTIVE); | |
1222 | ||
1bf03e66 KW |
1223 | /* FIXME We cannot require callers to have write permissions when all they |
1224 | * are doing is a read request. If we did things right, write permissions | |
1225 | * would be obtained anyway, but internally by the copy-on-read code. As | |
765d9df9 | 1226 | * long as it is implemented here rather than in a separate filter driver, |
1bf03e66 KW |
1227 | * the copy-on-read code doesn't have its own BdrvChild, however, for which |
1228 | * it could request permissions. Therefore we have to bypass the permission | |
1229 | * system for the moment. */ | |
1230 | // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
afa4b293 | 1231 | |
61007b31 | 1232 | /* Cover entire cluster so no additional backing file I/O is required when |
cb2e2878 EB |
1233 | * allocating cluster in the image file. Note that this value may exceed |
1234 | * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which | |
1235 | * is one reason we loop rather than doing it all at once. | |
61007b31 | 1236 | */ |
fc6b211f AD |
1237 | bdrv_round_to_subclusters(bs, offset, bytes, &align_offset, &align_bytes); |
1238 | skip_bytes = offset - align_offset; | |
61007b31 | 1239 | |
244483e6 | 1240 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
fc6b211f | 1241 | align_offset, align_bytes); |
61007b31 | 1242 | |
fc6b211f | 1243 | while (align_bytes) { |
cb2e2878 | 1244 | int64_t pnum; |
61007b31 | 1245 | |
8644476e HR |
1246 | if (skip_write) { |
1247 | ret = 1; /* "already allocated", so nothing will be copied */ | |
fc6b211f | 1248 | pnum = MIN(align_bytes, max_transfer); |
8644476e | 1249 | } else { |
cc323997 PB |
1250 | ret = bdrv_co_is_allocated(bs, align_offset, |
1251 | MIN(align_bytes, max_transfer), &pnum); | |
8644476e HR |
1252 | if (ret < 0) { |
1253 | /* | |
1254 | * Safe to treat errors in querying allocation as if | |
1255 | * unallocated; we'll probably fail again soon on the | |
1256 | * read, but at least that will set a decent errno. | |
1257 | */ | |
fc6b211f | 1258 | pnum = MIN(align_bytes, max_transfer); |
8644476e | 1259 | } |
61007b31 | 1260 | |
8644476e HR |
1261 | /* Stop at EOF if the image ends in the middle of the cluster */ |
1262 | if (ret == 0 && pnum == 0) { | |
1263 | assert(progress >= bytes); | |
1264 | break; | |
1265 | } | |
b0ddcbbb | 1266 | |
8644476e HR |
1267 | assert(skip_bytes < pnum); |
1268 | } | |
61007b31 | 1269 | |
cb2e2878 | 1270 | if (ret <= 0) { |
1143ec5e VSO |
1271 | QEMUIOVector local_qiov; |
1272 | ||
cb2e2878 | 1273 | /* Must copy-on-read; use the bounce buffer */ |
0d93ed08 | 1274 | pnum = MIN(pnum, MAX_BOUNCE_BUFFER); |
2275cc90 | 1275 | if (!bounce_buffer) { |
fc6b211f | 1276 | int64_t max_we_need = MAX(pnum, align_bytes - pnum); |
2275cc90 VSO |
1277 | int64_t max_allowed = MIN(max_transfer, MAX_BOUNCE_BUFFER); |
1278 | int64_t bounce_buffer_len = MIN(max_we_need, max_allowed); | |
1279 | ||
1280 | bounce_buffer = qemu_try_blockalign(bs, bounce_buffer_len); | |
1281 | if (!bounce_buffer) { | |
1282 | ret = -ENOMEM; | |
1283 | goto err; | |
1284 | } | |
1285 | } | |
0d93ed08 | 1286 | qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum); |
61007b31 | 1287 | |
fc6b211f | 1288 | ret = bdrv_driver_preadv(bs, align_offset, pnum, |
ac850bf0 | 1289 | &local_qiov, 0, 0); |
cb2e2878 EB |
1290 | if (ret < 0) { |
1291 | goto err; | |
1292 | } | |
1293 | ||
c834dc05 | 1294 | bdrv_co_debug_event(bs, BLKDBG_COR_WRITE); |
cb2e2878 EB |
1295 | if (drv->bdrv_co_pwrite_zeroes && |
1296 | buffer_is_zero(bounce_buffer, pnum)) { | |
1297 | /* FIXME: Should we (perhaps conditionally) be setting | |
1298 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
1299 | * that still correctly reads as zero? */ | |
fc6b211f | 1300 | ret = bdrv_co_do_pwrite_zeroes(bs, align_offset, pnum, |
7adcf59f | 1301 | BDRV_REQ_WRITE_UNCHANGED); |
cb2e2878 EB |
1302 | } else { |
1303 | /* This does not change the data on the disk, it is not | |
1304 | * necessary to flush even in cache=writethrough mode. | |
1305 | */ | |
fc6b211f | 1306 | ret = bdrv_driver_pwritev(bs, align_offset, pnum, |
ac850bf0 | 1307 | &local_qiov, 0, |
7adcf59f | 1308 | BDRV_REQ_WRITE_UNCHANGED); |
cb2e2878 EB |
1309 | } |
1310 | ||
1311 | if (ret < 0) { | |
1312 | /* It might be okay to ignore write errors for guest | |
1313 | * requests. If this is a deliberate copy-on-read | |
1314 | * then we don't want to ignore the error. Simply | |
1315 | * report it in all cases. | |
1316 | */ | |
1317 | goto err; | |
1318 | } | |
1319 | ||
3299e5ec | 1320 | if (!(flags & BDRV_REQ_PREFETCH)) { |
1143ec5e VSO |
1321 | qemu_iovec_from_buf(qiov, qiov_offset + progress, |
1322 | bounce_buffer + skip_bytes, | |
4ab78b19 | 1323 | MIN(pnum - skip_bytes, bytes - progress)); |
3299e5ec VSO |
1324 | } |
1325 | } else if (!(flags & BDRV_REQ_PREFETCH)) { | |
cb2e2878 | 1326 | /* Read directly into the destination */ |
1143ec5e VSO |
1327 | ret = bdrv_driver_preadv(bs, offset + progress, |
1328 | MIN(pnum - skip_bytes, bytes - progress), | |
1329 | qiov, qiov_offset + progress, 0); | |
cb2e2878 EB |
1330 | if (ret < 0) { |
1331 | goto err; | |
1332 | } | |
1333 | } | |
1334 | ||
fc6b211f AD |
1335 | align_offset += pnum; |
1336 | align_bytes -= pnum; | |
cb2e2878 EB |
1337 | progress += pnum - skip_bytes; |
1338 | skip_bytes = 0; | |
1339 | } | |
1340 | ret = 0; | |
61007b31 SH |
1341 | |
1342 | err: | |
1343 | qemu_vfree(bounce_buffer); | |
1344 | return ret; | |
1345 | } | |
1346 | ||
1347 | /* | |
1348 | * Forwards an already correctly aligned request to the BlockDriver. This | |
1a62d0ac EB |
1349 | * handles copy on read, zeroing after EOF, and fragmentation of large |
1350 | * reads; any other features must be implemented by the caller. | |
61007b31 | 1351 | */ |
7b1fb72e KW |
1352 | static int coroutine_fn GRAPH_RDLOCK |
1353 | bdrv_aligned_preadv(BdrvChild *child, BdrvTrackedRequest *req, | |
1354 | int64_t offset, int64_t bytes, int64_t align, | |
1355 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
61007b31 | 1356 | { |
85c97ca7 | 1357 | BlockDriverState *bs = child->bs; |
c9d20029 | 1358 | int64_t total_bytes, max_bytes; |
1a62d0ac | 1359 | int ret = 0; |
8b0c5d76 | 1360 | int64_t bytes_remaining = bytes; |
1a62d0ac | 1361 | int max_transfer; |
61007b31 | 1362 | |
8b0c5d76 | 1363 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
49c07526 KW |
1364 | assert(is_power_of_2(align)); |
1365 | assert((offset & (align - 1)) == 0); | |
1366 | assert((bytes & (align - 1)) == 0); | |
abb06c5a | 1367 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
1a62d0ac EB |
1368 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1369 | align); | |
a604fa2b | 1370 | |
e8b65355 SH |
1371 | /* |
1372 | * TODO: We would need a per-BDS .supported_read_flags and | |
a604fa2b EB |
1373 | * potential fallback support, if we ever implement any read flags |
1374 | * to pass through to drivers. For now, there aren't any | |
e8b65355 SH |
1375 | * passthrough flags except the BDRV_REQ_REGISTERED_BUF optimization hint. |
1376 | */ | |
1377 | assert(!(flags & ~(BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH | | |
1378 | BDRV_REQ_REGISTERED_BUF))); | |
61007b31 SH |
1379 | |
1380 | /* Handle Copy on Read and associated serialisation */ | |
1381 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
1382 | /* If we touch the same cluster it counts as an overlap. This | |
1383 | * guarantees that allocating writes will be serialized and not race | |
1384 | * with each other for the same cluster. For example, in copy-on-read | |
1385 | * it ensures that the CoR read and write operations are atomic and | |
1386 | * guest writes cannot interleave between them. */ | |
8ac5aab2 | 1387 | bdrv_make_request_serialising(req, bdrv_get_cluster_size(bs)); |
18fbd0de PB |
1388 | } else { |
1389 | bdrv_wait_serialising_requests(req); | |
61007b31 SH |
1390 | } |
1391 | ||
61007b31 | 1392 | if (flags & BDRV_REQ_COPY_ON_READ) { |
d6a644bb | 1393 | int64_t pnum; |
61007b31 | 1394 | |
897dd0ec AS |
1395 | /* The flag BDRV_REQ_COPY_ON_READ has reached its addressee */ |
1396 | flags &= ~BDRV_REQ_COPY_ON_READ; | |
1397 | ||
cc323997 | 1398 | ret = bdrv_co_is_allocated(bs, offset, bytes, &pnum); |
61007b31 SH |
1399 | if (ret < 0) { |
1400 | goto out; | |
1401 | } | |
1402 | ||
88e63df2 | 1403 | if (!ret || pnum != bytes) { |
65cd4424 VSO |
1404 | ret = bdrv_co_do_copy_on_readv(child, offset, bytes, |
1405 | qiov, qiov_offset, flags); | |
3299e5ec VSO |
1406 | goto out; |
1407 | } else if (flags & BDRV_REQ_PREFETCH) { | |
61007b31 SH |
1408 | goto out; |
1409 | } | |
1410 | } | |
1411 | ||
1a62d0ac | 1412 | /* Forward the request to the BlockDriver, possibly fragmenting it */ |
0af02bd1 | 1413 | total_bytes = bdrv_co_getlength(bs); |
c9d20029 KW |
1414 | if (total_bytes < 0) { |
1415 | ret = total_bytes; | |
1416 | goto out; | |
1417 | } | |
61007b31 | 1418 | |
e8b65355 | 1419 | assert(!(flags & ~(bs->supported_read_flags | BDRV_REQ_REGISTERED_BUF))); |
897dd0ec | 1420 | |
c9d20029 | 1421 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
1a62d0ac | 1422 | if (bytes <= max_bytes && bytes <= max_transfer) { |
897dd0ec | 1423 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, qiov_offset, flags); |
1a62d0ac EB |
1424 | goto out; |
1425 | } | |
61007b31 | 1426 | |
1a62d0ac | 1427 | while (bytes_remaining) { |
8b0c5d76 | 1428 | int64_t num; |
61007b31 | 1429 | |
1a62d0ac | 1430 | if (max_bytes) { |
1a62d0ac EB |
1431 | num = MIN(bytes_remaining, MIN(max_bytes, max_transfer)); |
1432 | assert(num); | |
61007b31 | 1433 | |
1a62d0ac | 1434 | ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining, |
134b7dec | 1435 | num, qiov, |
897dd0ec AS |
1436 | qiov_offset + bytes - bytes_remaining, |
1437 | flags); | |
1a62d0ac | 1438 | max_bytes -= num; |
1a62d0ac EB |
1439 | } else { |
1440 | num = bytes_remaining; | |
134b7dec HR |
1441 | ret = qemu_iovec_memset(qiov, qiov_offset + bytes - bytes_remaining, |
1442 | 0, bytes_remaining); | |
1a62d0ac EB |
1443 | } |
1444 | if (ret < 0) { | |
1445 | goto out; | |
1446 | } | |
1447 | bytes_remaining -= num; | |
61007b31 SH |
1448 | } |
1449 | ||
1450 | out: | |
1a62d0ac | 1451 | return ret < 0 ? ret : 0; |
61007b31 SH |
1452 | } |
1453 | ||
61007b31 | 1454 | /* |
7a3f542f VSO |
1455 | * Request padding |
1456 | * | |
1457 | * |<---- align ----->| |<----- align ---->| | |
1458 | * |<- head ->|<------------- bytes ------------->|<-- tail -->| | |
1459 | * | | | | | | | |
1460 | * -*----------$-------*-------- ... --------*-----$------------*--- | |
1461 | * | | | | | | | |
1462 | * | offset | | end | | |
1463 | * ALIGN_DOWN(offset) ALIGN_UP(offset) ALIGN_DOWN(end) ALIGN_UP(end) | |
1464 | * [buf ... ) [tail_buf ) | |
1465 | * | |
1466 | * @buf is an aligned allocation needed to store @head and @tail paddings. @head | |
1467 | * is placed at the beginning of @buf and @tail at the @end. | |
1468 | * | |
1469 | * @tail_buf is a pointer to sub-buffer, corresponding to align-sized chunk | |
1470 | * around tail, if tail exists. | |
1471 | * | |
1472 | * @merge_reads is true for small requests, | |
1473 | * if @buf_len == @head + bytes + @tail. In this case it is possible that both | |
1474 | * head and tail exist but @buf_len == align and @tail_buf == @buf. | |
18743311 HC |
1475 | * |
1476 | * @write is true for write requests, false for read requests. | |
1477 | * | |
1478 | * If padding makes the vector too long (exceeding IOV_MAX), then we need to | |
1479 | * merge existing vector elements into a single one. @collapse_bounce_buf acts | |
1480 | * as the bounce buffer in such cases. @pre_collapse_qiov has the pre-collapse | |
1481 | * I/O vector elements so for read requests, the data can be copied back after | |
1482 | * the read is done. | |
7a3f542f VSO |
1483 | */ |
1484 | typedef struct BdrvRequestPadding { | |
1485 | uint8_t *buf; | |
1486 | size_t buf_len; | |
1487 | uint8_t *tail_buf; | |
1488 | size_t head; | |
1489 | size_t tail; | |
1490 | bool merge_reads; | |
18743311 | 1491 | bool write; |
7a3f542f | 1492 | QEMUIOVector local_qiov; |
18743311 HC |
1493 | |
1494 | uint8_t *collapse_bounce_buf; | |
1495 | size_t collapse_len; | |
1496 | QEMUIOVector pre_collapse_qiov; | |
7a3f542f VSO |
1497 | } BdrvRequestPadding; |
1498 | ||
1499 | static bool bdrv_init_padding(BlockDriverState *bs, | |
1500 | int64_t offset, int64_t bytes, | |
18743311 | 1501 | bool write, |
7a3f542f VSO |
1502 | BdrvRequestPadding *pad) |
1503 | { | |
a56ed80c VSO |
1504 | int64_t align = bs->bl.request_alignment; |
1505 | int64_t sum; | |
1506 | ||
1507 | bdrv_check_request(offset, bytes, &error_abort); | |
1508 | assert(align <= INT_MAX); /* documented in block/block_int.h */ | |
1509 | assert(align <= SIZE_MAX / 2); /* so we can allocate the buffer */ | |
7a3f542f VSO |
1510 | |
1511 | memset(pad, 0, sizeof(*pad)); | |
1512 | ||
1513 | pad->head = offset & (align - 1); | |
1514 | pad->tail = ((offset + bytes) & (align - 1)); | |
1515 | if (pad->tail) { | |
1516 | pad->tail = align - pad->tail; | |
1517 | } | |
1518 | ||
ac9d00bf | 1519 | if (!pad->head && !pad->tail) { |
7a3f542f VSO |
1520 | return false; |
1521 | } | |
1522 | ||
ac9d00bf VSO |
1523 | assert(bytes); /* Nothing good in aligning zero-length requests */ |
1524 | ||
7a3f542f VSO |
1525 | sum = pad->head + bytes + pad->tail; |
1526 | pad->buf_len = (sum > align && pad->head && pad->tail) ? 2 * align : align; | |
1527 | pad->buf = qemu_blockalign(bs, pad->buf_len); | |
1528 | pad->merge_reads = sum == pad->buf_len; | |
1529 | if (pad->tail) { | |
1530 | pad->tail_buf = pad->buf + pad->buf_len - align; | |
1531 | } | |
1532 | ||
18743311 HC |
1533 | pad->write = write; |
1534 | ||
7a3f542f VSO |
1535 | return true; |
1536 | } | |
1537 | ||
7b1fb72e KW |
1538 | static int coroutine_fn GRAPH_RDLOCK |
1539 | bdrv_padding_rmw_read(BdrvChild *child, BdrvTrackedRequest *req, | |
1540 | BdrvRequestPadding *pad, bool zero_middle) | |
7a3f542f VSO |
1541 | { |
1542 | QEMUIOVector local_qiov; | |
1543 | BlockDriverState *bs = child->bs; | |
1544 | uint64_t align = bs->bl.request_alignment; | |
1545 | int ret; | |
1546 | ||
1547 | assert(req->serialising && pad->buf); | |
1548 | ||
1549 | if (pad->head || pad->merge_reads) { | |
8b0c5d76 | 1550 | int64_t bytes = pad->merge_reads ? pad->buf_len : align; |
7a3f542f VSO |
1551 | |
1552 | qemu_iovec_init_buf(&local_qiov, pad->buf, bytes); | |
1553 | ||
1554 | if (pad->head) { | |
c834dc05 | 1555 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
7a3f542f VSO |
1556 | } |
1557 | if (pad->merge_reads && pad->tail) { | |
c834dc05 | 1558 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
7a3f542f VSO |
1559 | } |
1560 | ret = bdrv_aligned_preadv(child, req, req->overlap_offset, bytes, | |
65cd4424 | 1561 | align, &local_qiov, 0, 0); |
7a3f542f VSO |
1562 | if (ret < 0) { |
1563 | return ret; | |
1564 | } | |
1565 | if (pad->head) { | |
c834dc05 | 1566 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
7a3f542f VSO |
1567 | } |
1568 | if (pad->merge_reads && pad->tail) { | |
c834dc05 | 1569 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
7a3f542f VSO |
1570 | } |
1571 | ||
1572 | if (pad->merge_reads) { | |
1573 | goto zero_mem; | |
1574 | } | |
1575 | } | |
1576 | ||
1577 | if (pad->tail) { | |
1578 | qemu_iovec_init_buf(&local_qiov, pad->tail_buf, align); | |
1579 | ||
c834dc05 | 1580 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
7a3f542f VSO |
1581 | ret = bdrv_aligned_preadv( |
1582 | child, req, | |
1583 | req->overlap_offset + req->overlap_bytes - align, | |
65cd4424 | 1584 | align, align, &local_qiov, 0, 0); |
7a3f542f VSO |
1585 | if (ret < 0) { |
1586 | return ret; | |
1587 | } | |
c834dc05 | 1588 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
7a3f542f VSO |
1589 | } |
1590 | ||
1591 | zero_mem: | |
1592 | if (zero_middle) { | |
1593 | memset(pad->buf + pad->head, 0, pad->buf_len - pad->head - pad->tail); | |
1594 | } | |
1595 | ||
1596 | return 0; | |
1597 | } | |
1598 | ||
18743311 HC |
1599 | /** |
1600 | * Free *pad's associated buffers, and perform any necessary finalization steps. | |
1601 | */ | |
1602 | static void bdrv_padding_finalize(BdrvRequestPadding *pad) | |
7a3f542f | 1603 | { |
18743311 HC |
1604 | if (pad->collapse_bounce_buf) { |
1605 | if (!pad->write) { | |
1606 | /* | |
1607 | * If padding required elements in the vector to be collapsed into a | |
1608 | * bounce buffer, copy the bounce buffer content back | |
1609 | */ | |
1610 | qemu_iovec_from_buf(&pad->pre_collapse_qiov, 0, | |
1611 | pad->collapse_bounce_buf, pad->collapse_len); | |
1612 | } | |
1613 | qemu_vfree(pad->collapse_bounce_buf); | |
1614 | qemu_iovec_destroy(&pad->pre_collapse_qiov); | |
1615 | } | |
7a3f542f VSO |
1616 | if (pad->buf) { |
1617 | qemu_vfree(pad->buf); | |
1618 | qemu_iovec_destroy(&pad->local_qiov); | |
1619 | } | |
98ca4549 | 1620 | memset(pad, 0, sizeof(*pad)); |
7a3f542f VSO |
1621 | } |
1622 | ||
18743311 HC |
1623 | /* |
1624 | * Create pad->local_qiov by wrapping @iov in the padding head and tail, while | |
1625 | * ensuring that the resulting vector will not exceed IOV_MAX elements. | |
1626 | * | |
1627 | * To ensure this, when necessary, the first two or three elements of @iov are | |
1628 | * merged into pad->collapse_bounce_buf and replaced by a reference to that | |
1629 | * bounce buffer in pad->local_qiov. | |
1630 | * | |
1631 | * After performing a read request, the data from the bounce buffer must be | |
1632 | * copied back into pad->pre_collapse_qiov (e.g. by bdrv_padding_finalize()). | |
1633 | */ | |
1634 | static int bdrv_create_padded_qiov(BlockDriverState *bs, | |
1635 | BdrvRequestPadding *pad, | |
1636 | struct iovec *iov, int niov, | |
1637 | size_t iov_offset, size_t bytes) | |
1638 | { | |
1639 | int padded_niov, surplus_count, collapse_count; | |
1640 | ||
1641 | /* Assert this invariant */ | |
1642 | assert(niov <= IOV_MAX); | |
1643 | ||
1644 | /* | |
1645 | * Cannot pad if resulting length would exceed SIZE_MAX. Returning an error | |
1646 | * to the guest is not ideal, but there is little else we can do. At least | |
1647 | * this will practically never happen on 64-bit systems. | |
1648 | */ | |
1649 | if (SIZE_MAX - pad->head < bytes || | |
1650 | SIZE_MAX - pad->head - bytes < pad->tail) | |
1651 | { | |
1652 | return -EINVAL; | |
1653 | } | |
1654 | ||
1655 | /* Length of the resulting IOV if we just concatenated everything */ | |
1656 | padded_niov = !!pad->head + niov + !!pad->tail; | |
1657 | ||
1658 | qemu_iovec_init(&pad->local_qiov, MIN(padded_niov, IOV_MAX)); | |
1659 | ||
1660 | if (pad->head) { | |
1661 | qemu_iovec_add(&pad->local_qiov, pad->buf, pad->head); | |
1662 | } | |
1663 | ||
1664 | /* | |
1665 | * If padded_niov > IOV_MAX, we cannot just concatenate everything. | |
1666 | * Instead, merge the first two or three elements of @iov to reduce the | |
1667 | * number of vector elements as necessary. | |
1668 | */ | |
1669 | if (padded_niov > IOV_MAX) { | |
1670 | /* | |
1671 | * Only head and tail can have lead to the number of entries exceeding | |
1672 | * IOV_MAX, so we can exceed it by the head and tail at most. We need | |
1673 | * to reduce the number of elements by `surplus_count`, so we merge that | |
1674 | * many elements plus one into one element. | |
1675 | */ | |
1676 | surplus_count = padded_niov - IOV_MAX; | |
1677 | assert(surplus_count <= !!pad->head + !!pad->tail); | |
1678 | collapse_count = surplus_count + 1; | |
1679 | ||
1680 | /* | |
1681 | * Move the elements to collapse into `pad->pre_collapse_qiov`, then | |
1682 | * advance `iov` (and associated variables) by those elements. | |
1683 | */ | |
1684 | qemu_iovec_init(&pad->pre_collapse_qiov, collapse_count); | |
1685 | qemu_iovec_concat_iov(&pad->pre_collapse_qiov, iov, | |
1686 | collapse_count, iov_offset, SIZE_MAX); | |
1687 | iov += collapse_count; | |
1688 | iov_offset = 0; | |
1689 | niov -= collapse_count; | |
1690 | bytes -= pad->pre_collapse_qiov.size; | |
1691 | ||
1692 | /* | |
1693 | * Construct the bounce buffer to match the length of the to-collapse | |
1694 | * vector elements, and for write requests, initialize it with the data | |
1695 | * from those elements. Then add it to `pad->local_qiov`. | |
1696 | */ | |
1697 | pad->collapse_len = pad->pre_collapse_qiov.size; | |
1698 | pad->collapse_bounce_buf = qemu_blockalign(bs, pad->collapse_len); | |
1699 | if (pad->write) { | |
1700 | qemu_iovec_to_buf(&pad->pre_collapse_qiov, 0, | |
1701 | pad->collapse_bounce_buf, pad->collapse_len); | |
1702 | } | |
1703 | qemu_iovec_add(&pad->local_qiov, | |
1704 | pad->collapse_bounce_buf, pad->collapse_len); | |
1705 | } | |
1706 | ||
1707 | qemu_iovec_concat_iov(&pad->local_qiov, iov, niov, iov_offset, bytes); | |
1708 | ||
1709 | if (pad->tail) { | |
1710 | qemu_iovec_add(&pad->local_qiov, | |
1711 | pad->buf + pad->buf_len - pad->tail, pad->tail); | |
1712 | } | |
1713 | ||
1714 | assert(pad->local_qiov.niov == MIN(padded_niov, IOV_MAX)); | |
1715 | return 0; | |
1716 | } | |
1717 | ||
7a3f542f VSO |
1718 | /* |
1719 | * bdrv_pad_request | |
1720 | * | |
1721 | * Exchange request parameters with padded request if needed. Don't include RMW | |
1722 | * read of padding, bdrv_padding_rmw_read() should be called separately if | |
1723 | * needed. | |
1724 | * | |
18743311 HC |
1725 | * @write is true for write requests, false for read requests. |
1726 | * | |
98ca4549 VSO |
1727 | * Request parameters (@qiov, &qiov_offset, &offset, &bytes) are in-out: |
1728 | * - on function start they represent original request | |
1729 | * - on failure or when padding is not needed they are unchanged | |
1730 | * - on success when padding is needed they represent padded request | |
61007b31 | 1731 | */ |
98ca4549 VSO |
1732 | static int bdrv_pad_request(BlockDriverState *bs, |
1733 | QEMUIOVector **qiov, size_t *qiov_offset, | |
37e9403e | 1734 | int64_t *offset, int64_t *bytes, |
18743311 | 1735 | bool write, |
e8b65355 SH |
1736 | BdrvRequestPadding *pad, bool *padded, |
1737 | BdrvRequestFlags *flags) | |
7a3f542f | 1738 | { |
4c002cef | 1739 | int ret; |
18743311 HC |
1740 | struct iovec *sliced_iov; |
1741 | int sliced_niov; | |
1742 | size_t sliced_head, sliced_tail; | |
4c002cef | 1743 | |
ef256751 HC |
1744 | /* Should have been checked by the caller already */ |
1745 | ret = bdrv_check_request32(*offset, *bytes, *qiov, *qiov_offset); | |
1746 | if (ret < 0) { | |
1747 | return ret; | |
1748 | } | |
37e9403e | 1749 | |
18743311 | 1750 | if (!bdrv_init_padding(bs, *offset, *bytes, write, pad)) { |
98ca4549 VSO |
1751 | if (padded) { |
1752 | *padded = false; | |
1753 | } | |
1754 | return 0; | |
7a3f542f VSO |
1755 | } |
1756 | ||
18743311 HC |
1757 | sliced_iov = qemu_iovec_slice(*qiov, *qiov_offset, *bytes, |
1758 | &sliced_head, &sliced_tail, | |
1759 | &sliced_niov); | |
1760 | ||
ef256751 | 1761 | /* Guaranteed by bdrv_check_request32() */ |
18743311 HC |
1762 | assert(*bytes <= SIZE_MAX); |
1763 | ret = bdrv_create_padded_qiov(bs, pad, sliced_iov, sliced_niov, | |
1764 | sliced_head, *bytes); | |
98ca4549 | 1765 | if (ret < 0) { |
18743311 | 1766 | bdrv_padding_finalize(pad); |
98ca4549 VSO |
1767 | return ret; |
1768 | } | |
7a3f542f VSO |
1769 | *bytes += pad->head + pad->tail; |
1770 | *offset -= pad->head; | |
1771 | *qiov = &pad->local_qiov; | |
1acc3466 | 1772 | *qiov_offset = 0; |
98ca4549 VSO |
1773 | if (padded) { |
1774 | *padded = true; | |
1775 | } | |
e8b65355 SH |
1776 | if (flags) { |
1777 | /* Can't use optimization hint with bounce buffer */ | |
1778 | *flags &= ~BDRV_REQ_REGISTERED_BUF; | |
1779 | } | |
7a3f542f | 1780 | |
98ca4549 | 1781 | return 0; |
7a3f542f VSO |
1782 | } |
1783 | ||
a03ef88f | 1784 | int coroutine_fn bdrv_co_preadv(BdrvChild *child, |
e9e52efd | 1785 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, |
61007b31 | 1786 | BdrvRequestFlags flags) |
1acc3466 | 1787 | { |
967d7905 | 1788 | IO_CODE(); |
1acc3466 VSO |
1789 | return bdrv_co_preadv_part(child, offset, bytes, qiov, 0, flags); |
1790 | } | |
1791 | ||
1792 | int coroutine_fn bdrv_co_preadv_part(BdrvChild *child, | |
37e9403e | 1793 | int64_t offset, int64_t bytes, |
1acc3466 VSO |
1794 | QEMUIOVector *qiov, size_t qiov_offset, |
1795 | BdrvRequestFlags flags) | |
61007b31 | 1796 | { |
a03ef88f | 1797 | BlockDriverState *bs = child->bs; |
61007b31 | 1798 | BdrvTrackedRequest req; |
7a3f542f | 1799 | BdrvRequestPadding pad; |
61007b31 | 1800 | int ret; |
967d7905 | 1801 | IO_CODE(); |
61007b31 | 1802 | |
37e9403e | 1803 | trace_bdrv_co_preadv_part(bs, offset, bytes, flags); |
61007b31 | 1804 | |
1e97be91 | 1805 | if (!bdrv_co_is_inserted(bs)) { |
f4dad307 VSO |
1806 | return -ENOMEDIUM; |
1807 | } | |
1808 | ||
63f4ad11 | 1809 | ret = bdrv_check_request32(offset, bytes, qiov, qiov_offset); |
61007b31 SH |
1810 | if (ret < 0) { |
1811 | return ret; | |
1812 | } | |
1813 | ||
ac9d00bf VSO |
1814 | if (bytes == 0 && !QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)) { |
1815 | /* | |
1816 | * Aligning zero request is nonsense. Even if driver has special meaning | |
1817 | * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass | |
1818 | * it to driver due to request_alignment. | |
1819 | * | |
1820 | * Still, no reason to return an error if someone do unaligned | |
1821 | * zero-length read occasionally. | |
1822 | */ | |
1823 | return 0; | |
1824 | } | |
1825 | ||
99723548 PB |
1826 | bdrv_inc_in_flight(bs); |
1827 | ||
9568b511 | 1828 | /* Don't do copy-on-read if we read data before write operation */ |
d73415a3 | 1829 | if (qatomic_read(&bs->copy_on_read)) { |
61007b31 SH |
1830 | flags |= BDRV_REQ_COPY_ON_READ; |
1831 | } | |
1832 | ||
18743311 HC |
1833 | ret = bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, false, |
1834 | &pad, NULL, &flags); | |
98ca4549 | 1835 | if (ret < 0) { |
87ab8802 | 1836 | goto fail; |
98ca4549 | 1837 | } |
61007b31 | 1838 | |
ebde595c | 1839 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
7a3f542f VSO |
1840 | ret = bdrv_aligned_preadv(child, &req, offset, bytes, |
1841 | bs->bl.request_alignment, | |
1acc3466 | 1842 | qiov, qiov_offset, flags); |
61007b31 | 1843 | tracked_request_end(&req); |
18743311 | 1844 | bdrv_padding_finalize(&pad); |
61007b31 | 1845 | |
87ab8802 KW |
1846 | fail: |
1847 | bdrv_dec_in_flight(bs); | |
1848 | ||
61007b31 SH |
1849 | return ret; |
1850 | } | |
1851 | ||
eeb47775 KW |
1852 | static int coroutine_fn GRAPH_RDLOCK |
1853 | bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1854 | BdrvRequestFlags flags) | |
61007b31 SH |
1855 | { |
1856 | BlockDriver *drv = bs->drv; | |
1857 | QEMUIOVector qiov; | |
0d93ed08 | 1858 | void *buf = NULL; |
61007b31 | 1859 | int ret = 0; |
465fe887 | 1860 | bool need_flush = false; |
443668ca DL |
1861 | int head = 0; |
1862 | int tail = 0; | |
61007b31 | 1863 | |
2aaa3f9b VSO |
1864 | int64_t max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, |
1865 | INT64_MAX); | |
a5b8dd2c EB |
1866 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1867 | bs->bl.request_alignment); | |
cb2e2878 | 1868 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER); |
d05aa8bb | 1869 | |
abaf8b75 | 1870 | assert_bdrv_graph_readable(); |
5ae07b14 VSO |
1871 | bdrv_check_request(offset, bytes, &error_abort); |
1872 | ||
d470ad42 HR |
1873 | if (!drv) { |
1874 | return -ENOMEDIUM; | |
1875 | } | |
1876 | ||
fe0480d6 KW |
1877 | if ((flags & ~bs->supported_zero_flags) & BDRV_REQ_NO_FALLBACK) { |
1878 | return -ENOTSUP; | |
1879 | } | |
1880 | ||
e8b65355 SH |
1881 | /* By definition there is no user buffer so this flag doesn't make sense */ |
1882 | if (flags & BDRV_REQ_REGISTERED_BUF) { | |
1883 | return -EINVAL; | |
1884 | } | |
1885 | ||
0bc329fb HR |
1886 | /* Invalidate the cached block-status data range if this write overlaps */ |
1887 | bdrv_bsc_invalidate_range(bs, offset, bytes); | |
1888 | ||
b8d0a980 EB |
1889 | assert(alignment % bs->bl.request_alignment == 0); |
1890 | head = offset % alignment; | |
f5a5ca79 | 1891 | tail = (offset + bytes) % alignment; |
b8d0a980 EB |
1892 | max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment); |
1893 | assert(max_write_zeroes >= bs->bl.request_alignment); | |
61007b31 | 1894 | |
f5a5ca79 | 1895 | while (bytes > 0 && !ret) { |
5ae07b14 | 1896 | int64_t num = bytes; |
61007b31 SH |
1897 | |
1898 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1899 | * to be aligned, and that unaligned requests do not cross cluster |
1900 | * boundaries. | |
61007b31 | 1901 | */ |
443668ca | 1902 | if (head) { |
b2f95fee EB |
1903 | /* Make a small request up to the first aligned sector. For |
1904 | * convenience, limit this request to max_transfer even if | |
1905 | * we don't need to fall back to writes. */ | |
f5a5ca79 | 1906 | num = MIN(MIN(bytes, max_transfer), alignment - head); |
b2f95fee EB |
1907 | head = (head + num) % alignment; |
1908 | assert(num < max_write_zeroes); | |
d05aa8bb | 1909 | } else if (tail && num > alignment) { |
443668ca DL |
1910 | /* Shorten the request to the last aligned sector. */ |
1911 | num -= tail; | |
61007b31 SH |
1912 | } |
1913 | ||
1914 | /* limit request size */ | |
1915 | if (num > max_write_zeroes) { | |
1916 | num = max_write_zeroes; | |
1917 | } | |
1918 | ||
1919 | ret = -ENOTSUP; | |
1920 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1921 | if (drv->bdrv_co_pwrite_zeroes) { |
1922 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1923 | flags & bs->supported_zero_flags); | |
1924 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1925 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1926 | need_flush = true; | |
1927 | } | |
465fe887 EB |
1928 | } else { |
1929 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1930 | } |
1931 | ||
294682cc | 1932 | if (ret == -ENOTSUP && !(flags & BDRV_REQ_NO_FALLBACK)) { |
61007b31 | 1933 | /* Fall back to bounce buffer if write zeroes is unsupported */ |
465fe887 EB |
1934 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1935 | ||
1936 | if ((flags & BDRV_REQ_FUA) && | |
1937 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1938 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1939 | * flush on each chunk; use just one at the end */ | |
1940 | write_flags &= ~BDRV_REQ_FUA; | |
1941 | need_flush = true; | |
1942 | } | |
5def6b80 | 1943 | num = MIN(num, max_transfer); |
0d93ed08 VSO |
1944 | if (buf == NULL) { |
1945 | buf = qemu_try_blockalign0(bs, num); | |
1946 | if (buf == NULL) { | |
61007b31 SH |
1947 | ret = -ENOMEM; |
1948 | goto fail; | |
1949 | } | |
61007b31 | 1950 | } |
0d93ed08 | 1951 | qemu_iovec_init_buf(&qiov, buf, num); |
61007b31 | 1952 | |
ac850bf0 | 1953 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, 0, write_flags); |
61007b31 SH |
1954 | |
1955 | /* Keep bounce buffer around if it is big enough for all | |
1956 | * all future requests. | |
1957 | */ | |
5def6b80 | 1958 | if (num < max_transfer) { |
0d93ed08 VSO |
1959 | qemu_vfree(buf); |
1960 | buf = NULL; | |
61007b31 SH |
1961 | } |
1962 | } | |
1963 | ||
d05aa8bb | 1964 | offset += num; |
f5a5ca79 | 1965 | bytes -= num; |
61007b31 SH |
1966 | } |
1967 | ||
1968 | fail: | |
465fe887 EB |
1969 | if (ret == 0 && need_flush) { |
1970 | ret = bdrv_co_flush(bs); | |
1971 | } | |
0d93ed08 | 1972 | qemu_vfree(buf); |
61007b31 SH |
1973 | return ret; |
1974 | } | |
1975 | ||
a00e70c0 | 1976 | static inline int coroutine_fn GRAPH_RDLOCK |
fcfd9ade | 1977 | bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, int64_t bytes, |
85fe2479 FZ |
1978 | BdrvTrackedRequest *req, int flags) |
1979 | { | |
1980 | BlockDriverState *bs = child->bs; | |
fcfd9ade VSO |
1981 | |
1982 | bdrv_check_request(offset, bytes, &error_abort); | |
85fe2479 | 1983 | |
307261b2 | 1984 | if (bdrv_is_read_only(bs)) { |
85fe2479 FZ |
1985 | return -EPERM; |
1986 | } | |
1987 | ||
85fe2479 FZ |
1988 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
1989 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); | |
1990 | assert(!(flags & ~BDRV_REQ_MASK)); | |
d1a764d1 | 1991 | assert(!((flags & BDRV_REQ_NO_WAIT) && !(flags & BDRV_REQ_SERIALISING))); |
85fe2479 FZ |
1992 | |
1993 | if (flags & BDRV_REQ_SERIALISING) { | |
d1a764d1 VSO |
1994 | QEMU_LOCK_GUARD(&bs->reqs_lock); |
1995 | ||
1996 | tracked_request_set_serialising(req, bdrv_get_cluster_size(bs)); | |
1997 | ||
1998 | if ((flags & BDRV_REQ_NO_WAIT) && bdrv_find_conflicting_request(req)) { | |
1999 | return -EBUSY; | |
2000 | } | |
2001 | ||
2002 | bdrv_wait_serialising_requests_locked(req); | |
18fbd0de PB |
2003 | } else { |
2004 | bdrv_wait_serialising_requests(req); | |
85fe2479 FZ |
2005 | } |
2006 | ||
85fe2479 FZ |
2007 | assert(req->overlap_offset <= offset); |
2008 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
fcfd9ade VSO |
2009 | assert(offset + bytes <= bs->total_sectors * BDRV_SECTOR_SIZE || |
2010 | child->perm & BLK_PERM_RESIZE); | |
85fe2479 | 2011 | |
cd47d792 FZ |
2012 | switch (req->type) { |
2013 | case BDRV_TRACKED_WRITE: | |
2014 | case BDRV_TRACKED_DISCARD: | |
2015 | if (flags & BDRV_REQ_WRITE_UNCHANGED) { | |
2016 | assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
2017 | } else { | |
2018 | assert(child->perm & BLK_PERM_WRITE); | |
2019 | } | |
94783301 VSO |
2020 | bdrv_write_threshold_check_write(bs, offset, bytes); |
2021 | return 0; | |
cd47d792 FZ |
2022 | case BDRV_TRACKED_TRUNCATE: |
2023 | assert(child->perm & BLK_PERM_RESIZE); | |
2024 | return 0; | |
2025 | default: | |
2026 | abort(); | |
85fe2479 | 2027 | } |
85fe2479 FZ |
2028 | } |
2029 | ||
2030 | static inline void coroutine_fn | |
fcfd9ade | 2031 | bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, int64_t bytes, |
85fe2479 FZ |
2032 | BdrvTrackedRequest *req, int ret) |
2033 | { | |
2034 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
2035 | BlockDriverState *bs = child->bs; | |
2036 | ||
fcfd9ade VSO |
2037 | bdrv_check_request(offset, bytes, &error_abort); |
2038 | ||
d73415a3 | 2039 | qatomic_inc(&bs->write_gen); |
85fe2479 | 2040 | |
00695c27 FZ |
2041 | /* |
2042 | * Discard cannot extend the image, but in error handling cases, such as | |
2043 | * when reverting a qcow2 cluster allocation, the discarded range can pass | |
2044 | * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD | |
2045 | * here. Instead, just skip it, since semantically a discard request | |
2046 | * beyond EOF cannot expand the image anyway. | |
2047 | */ | |
7f8f03ef | 2048 | if (ret == 0 && |
cd47d792 FZ |
2049 | (req->type == BDRV_TRACKED_TRUNCATE || |
2050 | end_sector > bs->total_sectors) && | |
2051 | req->type != BDRV_TRACKED_DISCARD) { | |
7f8f03ef FZ |
2052 | bs->total_sectors = end_sector; |
2053 | bdrv_parent_cb_resize(bs); | |
2054 | bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS); | |
85fe2479 | 2055 | } |
00695c27 FZ |
2056 | if (req->bytes) { |
2057 | switch (req->type) { | |
2058 | case BDRV_TRACKED_WRITE: | |
2059 | stat64_max(&bs->wr_highest_offset, offset + bytes); | |
2060 | /* fall through, to set dirty bits */ | |
2061 | case BDRV_TRACKED_DISCARD: | |
2062 | bdrv_set_dirty(bs, offset, bytes); | |
2063 | break; | |
2064 | default: | |
2065 | break; | |
2066 | } | |
2067 | } | |
85fe2479 FZ |
2068 | } |
2069 | ||
61007b31 | 2070 | /* |
04ed95f4 EB |
2071 | * Forwards an already correctly aligned write request to the BlockDriver, |
2072 | * after possibly fragmenting it. | |
61007b31 | 2073 | */ |
7b1fb72e KW |
2074 | static int coroutine_fn GRAPH_RDLOCK |
2075 | bdrv_aligned_pwritev(BdrvChild *child, BdrvTrackedRequest *req, | |
2076 | int64_t offset, int64_t bytes, int64_t align, | |
2077 | QEMUIOVector *qiov, size_t qiov_offset, | |
2078 | BdrvRequestFlags flags) | |
61007b31 | 2079 | { |
85c97ca7 | 2080 | BlockDriverState *bs = child->bs; |
61007b31 | 2081 | BlockDriver *drv = bs->drv; |
61007b31 SH |
2082 | int ret; |
2083 | ||
fcfd9ade | 2084 | int64_t bytes_remaining = bytes; |
04ed95f4 | 2085 | int max_transfer; |
61007b31 | 2086 | |
fcfd9ade VSO |
2087 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
2088 | ||
d470ad42 HR |
2089 | if (!drv) { |
2090 | return -ENOMEDIUM; | |
2091 | } | |
2092 | ||
d6883bc9 VSO |
2093 | if (bdrv_has_readonly_bitmaps(bs)) { |
2094 | return -EPERM; | |
2095 | } | |
2096 | ||
cff86b38 EB |
2097 | assert(is_power_of_2(align)); |
2098 | assert((offset & (align - 1)) == 0); | |
2099 | assert((bytes & (align - 1)) == 0); | |
04ed95f4 EB |
2100 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
2101 | align); | |
61007b31 | 2102 | |
85fe2479 | 2103 | ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags); |
61007b31 SH |
2104 | |
2105 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 2106 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
28c4da28 | 2107 | qemu_iovec_is_zero(qiov, qiov_offset, bytes)) { |
61007b31 SH |
2108 | flags |= BDRV_REQ_ZERO_WRITE; |
2109 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
2110 | flags |= BDRV_REQ_MAY_UNMAP; | |
2111 | } | |
3c586715 SH |
2112 | |
2113 | /* Can't use optimization hint with bufferless zero write */ | |
2114 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
61007b31 SH |
2115 | } |
2116 | ||
2117 | if (ret < 0) { | |
2118 | /* Do nothing, write notifier decided to fail this request */ | |
2119 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
c834dc05 | 2120 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 2121 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
3ea1a091 | 2122 | } else if (flags & BDRV_REQ_WRITE_COMPRESSED) { |
28c4da28 VSO |
2123 | ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, |
2124 | qiov, qiov_offset); | |
04ed95f4 | 2125 | } else if (bytes <= max_transfer) { |
c834dc05 | 2126 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV); |
28c4da28 | 2127 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, qiov_offset, flags); |
04ed95f4 | 2128 | } else { |
c834dc05 | 2129 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV); |
04ed95f4 EB |
2130 | while (bytes_remaining) { |
2131 | int num = MIN(bytes_remaining, max_transfer); | |
04ed95f4 EB |
2132 | int local_flags = flags; |
2133 | ||
2134 | assert(num); | |
2135 | if (num < bytes_remaining && (flags & BDRV_REQ_FUA) && | |
2136 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
2137 | /* If FUA is going to be emulated by flush, we only | |
2138 | * need to flush on the last iteration */ | |
2139 | local_flags &= ~BDRV_REQ_FUA; | |
2140 | } | |
04ed95f4 EB |
2141 | |
2142 | ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining, | |
134b7dec HR |
2143 | num, qiov, |
2144 | qiov_offset + bytes - bytes_remaining, | |
28c4da28 | 2145 | local_flags); |
04ed95f4 EB |
2146 | if (ret < 0) { |
2147 | break; | |
2148 | } | |
2149 | bytes_remaining -= num; | |
2150 | } | |
61007b31 | 2151 | } |
c834dc05 | 2152 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 2153 | |
61007b31 | 2154 | if (ret >= 0) { |
04ed95f4 | 2155 | ret = 0; |
61007b31 | 2156 | } |
85fe2479 | 2157 | bdrv_co_write_req_finish(child, offset, bytes, req, ret); |
61007b31 SH |
2158 | |
2159 | return ret; | |
2160 | } | |
2161 | ||
7b1fb72e KW |
2162 | static int coroutine_fn GRAPH_RDLOCK |
2163 | bdrv_co_do_zero_pwritev(BdrvChild *child, int64_t offset, int64_t bytes, | |
2164 | BdrvRequestFlags flags, BdrvTrackedRequest *req) | |
9eeb6dd1 | 2165 | { |
85c97ca7 | 2166 | BlockDriverState *bs = child->bs; |
9eeb6dd1 | 2167 | QEMUIOVector local_qiov; |
a5b8dd2c | 2168 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 | 2169 | int ret = 0; |
7a3f542f VSO |
2170 | bool padding; |
2171 | BdrvRequestPadding pad; | |
9eeb6dd1 | 2172 | |
e8b65355 SH |
2173 | /* This flag doesn't make sense for padding or zero writes */ |
2174 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
2175 | ||
18743311 | 2176 | padding = bdrv_init_padding(bs, offset, bytes, true, &pad); |
7a3f542f | 2177 | if (padding) { |
45e62b46 | 2178 | assert(!(flags & BDRV_REQ_NO_WAIT)); |
8ac5aab2 | 2179 | bdrv_make_request_serialising(req, align); |
9eeb6dd1 | 2180 | |
7a3f542f VSO |
2181 | bdrv_padding_rmw_read(child, req, &pad, true); |
2182 | ||
2183 | if (pad.head || pad.merge_reads) { | |
2184 | int64_t aligned_offset = offset & ~(align - 1); | |
2185 | int64_t write_bytes = pad.merge_reads ? pad.buf_len : align; | |
2186 | ||
2187 | qemu_iovec_init_buf(&local_qiov, pad.buf, write_bytes); | |
2188 | ret = bdrv_aligned_pwritev(child, req, aligned_offset, write_bytes, | |
28c4da28 | 2189 | align, &local_qiov, 0, |
7a3f542f VSO |
2190 | flags & ~BDRV_REQ_ZERO_WRITE); |
2191 | if (ret < 0 || pad.merge_reads) { | |
2192 | /* Error or all work is done */ | |
2193 | goto out; | |
2194 | } | |
2195 | offset += write_bytes - pad.head; | |
2196 | bytes -= write_bytes - pad.head; | |
9eeb6dd1 | 2197 | } |
9eeb6dd1 FZ |
2198 | } |
2199 | ||
2200 | assert(!bytes || (offset & (align - 1)) == 0); | |
2201 | if (bytes >= align) { | |
2202 | /* Write the aligned part in the middle. */ | |
fcfd9ade | 2203 | int64_t aligned_bytes = bytes & ~(align - 1); |
85c97ca7 | 2204 | ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align, |
28c4da28 | 2205 | NULL, 0, flags); |
9eeb6dd1 | 2206 | if (ret < 0) { |
7a3f542f | 2207 | goto out; |
9eeb6dd1 FZ |
2208 | } |
2209 | bytes -= aligned_bytes; | |
2210 | offset += aligned_bytes; | |
2211 | } | |
2212 | ||
2213 | assert(!bytes || (offset & (align - 1)) == 0); | |
2214 | if (bytes) { | |
7a3f542f | 2215 | assert(align == pad.tail + bytes); |
9eeb6dd1 | 2216 | |
7a3f542f | 2217 | qemu_iovec_init_buf(&local_qiov, pad.tail_buf, align); |
85c97ca7 | 2218 | ret = bdrv_aligned_pwritev(child, req, offset, align, align, |
28c4da28 VSO |
2219 | &local_qiov, 0, |
2220 | flags & ~BDRV_REQ_ZERO_WRITE); | |
9eeb6dd1 | 2221 | } |
9eeb6dd1 | 2222 | |
7a3f542f | 2223 | out: |
18743311 | 2224 | bdrv_padding_finalize(&pad); |
7a3f542f VSO |
2225 | |
2226 | return ret; | |
9eeb6dd1 FZ |
2227 | } |
2228 | ||
61007b31 SH |
2229 | /* |
2230 | * Handle a write request in coroutine context | |
2231 | */ | |
a03ef88f | 2232 | int coroutine_fn bdrv_co_pwritev(BdrvChild *child, |
e9e52efd | 2233 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, |
61007b31 | 2234 | BdrvRequestFlags flags) |
1acc3466 | 2235 | { |
967d7905 | 2236 | IO_CODE(); |
1acc3466 VSO |
2237 | return bdrv_co_pwritev_part(child, offset, bytes, qiov, 0, flags); |
2238 | } | |
2239 | ||
2240 | int coroutine_fn bdrv_co_pwritev_part(BdrvChild *child, | |
37e9403e | 2241 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, size_t qiov_offset, |
1acc3466 | 2242 | BdrvRequestFlags flags) |
61007b31 | 2243 | { |
a03ef88f | 2244 | BlockDriverState *bs = child->bs; |
61007b31 | 2245 | BdrvTrackedRequest req; |
a5b8dd2c | 2246 | uint64_t align = bs->bl.request_alignment; |
7a3f542f | 2247 | BdrvRequestPadding pad; |
61007b31 | 2248 | int ret; |
f0deecff | 2249 | bool padded = false; |
967d7905 | 2250 | IO_CODE(); |
61007b31 | 2251 | |
37e9403e | 2252 | trace_bdrv_co_pwritev_part(child->bs, offset, bytes, flags); |
f42cf447 | 2253 | |
1e97be91 | 2254 | if (!bdrv_co_is_inserted(bs)) { |
61007b31 SH |
2255 | return -ENOMEDIUM; |
2256 | } | |
61007b31 | 2257 | |
2aaa3f9b VSO |
2258 | if (flags & BDRV_REQ_ZERO_WRITE) { |
2259 | ret = bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, NULL); | |
2260 | } else { | |
2261 | ret = bdrv_check_request32(offset, bytes, qiov, qiov_offset); | |
2262 | } | |
61007b31 SH |
2263 | if (ret < 0) { |
2264 | return ret; | |
2265 | } | |
2266 | ||
f2208fdc AG |
2267 | /* If the request is misaligned then we can't make it efficient */ |
2268 | if ((flags & BDRV_REQ_NO_FALLBACK) && | |
2269 | !QEMU_IS_ALIGNED(offset | bytes, align)) | |
2270 | { | |
2271 | return -ENOTSUP; | |
2272 | } | |
2273 | ||
ac9d00bf VSO |
2274 | if (bytes == 0 && !QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)) { |
2275 | /* | |
2276 | * Aligning zero request is nonsense. Even if driver has special meaning | |
2277 | * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass | |
2278 | * it to driver due to request_alignment. | |
2279 | * | |
2280 | * Still, no reason to return an error if someone do unaligned | |
2281 | * zero-length write occasionally. | |
2282 | */ | |
2283 | return 0; | |
2284 | } | |
2285 | ||
f0deecff VSO |
2286 | if (!(flags & BDRV_REQ_ZERO_WRITE)) { |
2287 | /* | |
2288 | * Pad request for following read-modify-write cycle. | |
2289 | * bdrv_co_do_zero_pwritev() does aligning by itself, so, we do | |
2290 | * alignment only if there is no ZERO flag. | |
2291 | */ | |
18743311 HC |
2292 | ret = bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, true, |
2293 | &pad, &padded, &flags); | |
98ca4549 VSO |
2294 | if (ret < 0) { |
2295 | return ret; | |
2296 | } | |
f0deecff VSO |
2297 | } |
2298 | ||
99723548 | 2299 | bdrv_inc_in_flight(bs); |
ebde595c | 2300 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 2301 | |
18a59f03 | 2302 | if (flags & BDRV_REQ_ZERO_WRITE) { |
f0deecff | 2303 | assert(!padded); |
85c97ca7 | 2304 | ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req); |
9eeb6dd1 FZ |
2305 | goto out; |
2306 | } | |
2307 | ||
f0deecff VSO |
2308 | if (padded) { |
2309 | /* | |
2310 | * Request was unaligned to request_alignment and therefore | |
2311 | * padded. We are going to do read-modify-write, and must | |
2312 | * serialize the request to prevent interactions of the | |
2313 | * widened region with other transactions. | |
2314 | */ | |
45e62b46 | 2315 | assert(!(flags & BDRV_REQ_NO_WAIT)); |
8ac5aab2 | 2316 | bdrv_make_request_serialising(&req, align); |
7a3f542f | 2317 | bdrv_padding_rmw_read(child, &req, &pad, false); |
61007b31 SH |
2318 | } |
2319 | ||
85c97ca7 | 2320 | ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align, |
1acc3466 | 2321 | qiov, qiov_offset, flags); |
61007b31 | 2322 | |
18743311 | 2323 | bdrv_padding_finalize(&pad); |
61007b31 | 2324 | |
9eeb6dd1 FZ |
2325 | out: |
2326 | tracked_request_end(&req); | |
99723548 | 2327 | bdrv_dec_in_flight(bs); |
7a3f542f | 2328 | |
61007b31 SH |
2329 | return ret; |
2330 | } | |
2331 | ||
a03ef88f | 2332 | int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, |
e9e52efd | 2333 | int64_t bytes, BdrvRequestFlags flags) |
61007b31 | 2334 | { |
384a48fb | 2335 | IO_CODE(); |
f5a5ca79 | 2336 | trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags); |
abaf8b75 | 2337 | assert_bdrv_graph_readable(); |
61007b31 | 2338 | |
a03ef88f | 2339 | if (!(child->bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
2340 | flags &= ~BDRV_REQ_MAY_UNMAP; |
2341 | } | |
61007b31 | 2342 | |
f5a5ca79 | 2343 | return bdrv_co_pwritev(child, offset, bytes, NULL, |
74021bc4 | 2344 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
2345 | } |
2346 | ||
4085f5c7 JS |
2347 | /* |
2348 | * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not. | |
2349 | */ | |
2350 | int bdrv_flush_all(void) | |
2351 | { | |
2352 | BdrvNextIterator it; | |
2353 | BlockDriverState *bs = NULL; | |
2354 | int result = 0; | |
2355 | ||
f791bf7f | 2356 | GLOBAL_STATE_CODE(); |
2b3912f1 | 2357 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
f791bf7f | 2358 | |
c8aa7895 PD |
2359 | /* |
2360 | * bdrv queue is managed by record/replay, | |
2361 | * creating new flush request for stopping | |
2362 | * the VM may break the determinism | |
2363 | */ | |
2364 | if (replay_events_enabled()) { | |
2365 | return result; | |
2366 | } | |
2367 | ||
4085f5c7 JS |
2368 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
2369 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2370 | int ret; | |
2371 | ||
2372 | aio_context_acquire(aio_context); | |
2373 | ret = bdrv_flush(bs); | |
2374 | if (ret < 0 && !result) { | |
2375 | result = ret; | |
2376 | } | |
2377 | aio_context_release(aio_context); | |
2378 | } | |
2379 | ||
2380 | return result; | |
2381 | } | |
2382 | ||
61007b31 SH |
2383 | /* |
2384 | * Returns the allocation status of the specified sectors. | |
2385 | * Drivers not implementing the functionality are assumed to not support | |
2386 | * backing files, hence all their sectors are reported as allocated. | |
2387 | * | |
86a3d5c6 EB |
2388 | * If 'want_zero' is true, the caller is querying for mapping |
2389 | * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and | |
2390 | * _ZERO where possible; otherwise, the result favors larger 'pnum', | |
2391 | * with a focus on accurate BDRV_BLOCK_ALLOCATED. | |
c9ce8c4d | 2392 | * |
2e8bc787 | 2393 | * If 'offset' is beyond the end of the disk image the return value is |
fb0d8654 | 2394 | * BDRV_BLOCK_EOF and 'pnum' is set to 0. |
61007b31 | 2395 | * |
2e8bc787 | 2396 | * 'bytes' is the max value 'pnum' should be set to. If bytes goes |
fb0d8654 EB |
2397 | * beyond the end of the disk image it will be clamped; if 'pnum' is set to |
2398 | * the end of the image, then the returned value will include BDRV_BLOCK_EOF. | |
67a0fd2a | 2399 | * |
2e8bc787 EB |
2400 | * 'pnum' is set to the number of bytes (including and immediately |
2401 | * following the specified offset) that are easily known to be in the | |
2402 | * same allocated/unallocated state. Note that a second call starting | |
2403 | * at the original offset plus returned pnum may have the same status. | |
2404 | * The returned value is non-zero on success except at end-of-file. | |
2405 | * | |
2406 | * Returns negative errno on failure. Otherwise, if the | |
2407 | * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are | |
2408 | * set to the host mapping and BDS corresponding to the guest offset. | |
61007b31 | 2409 | */ |
7ff9579e | 2410 | static int coroutine_fn GRAPH_RDLOCK |
b170e929 PB |
2411 | bdrv_co_do_block_status(BlockDriverState *bs, bool want_zero, |
2412 | int64_t offset, int64_t bytes, | |
2413 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
2e8bc787 EB |
2414 | { |
2415 | int64_t total_size; | |
2416 | int64_t n; /* bytes */ | |
efa6e2ed | 2417 | int ret; |
2e8bc787 | 2418 | int64_t local_map = 0; |
298a1665 | 2419 | BlockDriverState *local_file = NULL; |
efa6e2ed EB |
2420 | int64_t aligned_offset, aligned_bytes; |
2421 | uint32_t align; | |
549ec0d9 | 2422 | bool has_filtered_child; |
61007b31 | 2423 | |
298a1665 | 2424 | assert(pnum); |
7ff9579e | 2425 | assert_bdrv_graph_readable(); |
298a1665 | 2426 | *pnum = 0; |
0af02bd1 | 2427 | total_size = bdrv_co_getlength(bs); |
2e8bc787 EB |
2428 | if (total_size < 0) { |
2429 | ret = total_size; | |
298a1665 | 2430 | goto early_out; |
61007b31 SH |
2431 | } |
2432 | ||
2e8bc787 | 2433 | if (offset >= total_size) { |
298a1665 EB |
2434 | ret = BDRV_BLOCK_EOF; |
2435 | goto early_out; | |
61007b31 | 2436 | } |
2e8bc787 | 2437 | if (!bytes) { |
298a1665 EB |
2438 | ret = 0; |
2439 | goto early_out; | |
9cdcfd9f | 2440 | } |
61007b31 | 2441 | |
2e8bc787 EB |
2442 | n = total_size - offset; |
2443 | if (n < bytes) { | |
2444 | bytes = n; | |
61007b31 SH |
2445 | } |
2446 | ||
0af02bd1 | 2447 | /* Must be non-NULL or bdrv_co_getlength() would have failed */ |
d470ad42 | 2448 | assert(bs->drv); |
549ec0d9 HR |
2449 | has_filtered_child = bdrv_filter_child(bs); |
2450 | if (!bs->drv->bdrv_co_block_status && !has_filtered_child) { | |
2e8bc787 | 2451 | *pnum = bytes; |
61007b31 | 2452 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; |
2e8bc787 | 2453 | if (offset + bytes == total_size) { |
fb0d8654 EB |
2454 | ret |= BDRV_BLOCK_EOF; |
2455 | } | |
61007b31 | 2456 | if (bs->drv->protocol_name) { |
2e8bc787 EB |
2457 | ret |= BDRV_BLOCK_OFFSET_VALID; |
2458 | local_map = offset; | |
298a1665 | 2459 | local_file = bs; |
61007b31 | 2460 | } |
298a1665 | 2461 | goto early_out; |
61007b31 SH |
2462 | } |
2463 | ||
99723548 | 2464 | bdrv_inc_in_flight(bs); |
efa6e2ed EB |
2465 | |
2466 | /* Round out to request_alignment boundaries */ | |
86a3d5c6 | 2467 | align = bs->bl.request_alignment; |
efa6e2ed EB |
2468 | aligned_offset = QEMU_ALIGN_DOWN(offset, align); |
2469 | aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset; | |
2470 | ||
549ec0d9 | 2471 | if (bs->drv->bdrv_co_block_status) { |
0bc329fb HR |
2472 | /* |
2473 | * Use the block-status cache only for protocol nodes: Format | |
2474 | * drivers are generally quick to inquire the status, but protocol | |
2475 | * drivers often need to get information from outside of qemu, so | |
2476 | * we do not have control over the actual implementation. There | |
2477 | * have been cases where inquiring the status took an unreasonably | |
2478 | * long time, and we can do nothing in qemu to fix it. | |
2479 | * This is especially problematic for images with large data areas, | |
2480 | * because finding the few holes in them and giving them special | |
2481 | * treatment does not gain much performance. Therefore, we try to | |
2482 | * cache the last-identified data region. | |
2483 | * | |
2484 | * Second, limiting ourselves to protocol nodes allows us to assume | |
2485 | * the block status for data regions to be DATA | OFFSET_VALID, and | |
2486 | * that the host offset is the same as the guest offset. | |
2487 | * | |
2488 | * Note that it is possible that external writers zero parts of | |
2489 | * the cached regions without the cache being invalidated, and so | |
2490 | * we may report zeroes as data. This is not catastrophic, | |
2491 | * however, because reporting zeroes as data is fine. | |
2492 | */ | |
2493 | if (QLIST_EMPTY(&bs->children) && | |
2494 | bdrv_bsc_is_data(bs, aligned_offset, pnum)) | |
2495 | { | |
2496 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
2497 | local_file = bs; | |
2498 | local_map = aligned_offset; | |
2499 | } else { | |
2500 | ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset, | |
2501 | aligned_bytes, pnum, &local_map, | |
2502 | &local_file); | |
2503 | ||
2504 | /* | |
2505 | * Note that checking QLIST_EMPTY(&bs->children) is also done when | |
2506 | * the cache is queried above. Technically, we do not need to check | |
2507 | * it here; the worst that can happen is that we fill the cache for | |
2508 | * non-protocol nodes, and then it is never used. However, filling | |
2509 | * the cache requires an RCU update, so double check here to avoid | |
2510 | * such an update if possible. | |
113b727c HR |
2511 | * |
2512 | * Check want_zero, because we only want to update the cache when we | |
2513 | * have accurate information about what is zero and what is data. | |
0bc329fb | 2514 | */ |
113b727c HR |
2515 | if (want_zero && |
2516 | ret == (BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID) && | |
0bc329fb HR |
2517 | QLIST_EMPTY(&bs->children)) |
2518 | { | |
2519 | /* | |
2520 | * When a protocol driver reports BLOCK_OFFSET_VALID, the | |
2521 | * returned local_map value must be the same as the offset we | |
2522 | * have passed (aligned_offset), and local_bs must be the node | |
2523 | * itself. | |
2524 | * Assert this, because we follow this rule when reading from | |
2525 | * the cache (see the `local_file = bs` and | |
2526 | * `local_map = aligned_offset` assignments above), and the | |
2527 | * result the cache delivers must be the same as the driver | |
2528 | * would deliver. | |
2529 | */ | |
2530 | assert(local_file == bs); | |
2531 | assert(local_map == aligned_offset); | |
2532 | bdrv_bsc_fill(bs, aligned_offset, *pnum); | |
2533 | } | |
2534 | } | |
549ec0d9 HR |
2535 | } else { |
2536 | /* Default code for filters */ | |
2537 | ||
2538 | local_file = bdrv_filter_bs(bs); | |
2539 | assert(local_file); | |
2540 | ||
2541 | *pnum = aligned_bytes; | |
2542 | local_map = aligned_offset; | |
2543 | ret = BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; | |
2544 | } | |
636cb512 EB |
2545 | if (ret < 0) { |
2546 | *pnum = 0; | |
2547 | goto out; | |
efa6e2ed EB |
2548 | } |
2549 | ||
2e8bc787 | 2550 | /* |
636cb512 | 2551 | * The driver's result must be a non-zero multiple of request_alignment. |
efa6e2ed | 2552 | * Clamp pnum and adjust map to original request. |
2e8bc787 | 2553 | */ |
636cb512 EB |
2554 | assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) && |
2555 | align > offset - aligned_offset); | |
69f47505 VSO |
2556 | if (ret & BDRV_BLOCK_RECURSE) { |
2557 | assert(ret & BDRV_BLOCK_DATA); | |
2558 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
2559 | assert(!(ret & BDRV_BLOCK_ZERO)); | |
2560 | } | |
2561 | ||
efa6e2ed EB |
2562 | *pnum -= offset - aligned_offset; |
2563 | if (*pnum > bytes) { | |
2564 | *pnum = bytes; | |
61007b31 | 2565 | } |
2e8bc787 | 2566 | if (ret & BDRV_BLOCK_OFFSET_VALID) { |
efa6e2ed | 2567 | local_map += offset - aligned_offset; |
2e8bc787 | 2568 | } |
61007b31 SH |
2569 | |
2570 | if (ret & BDRV_BLOCK_RAW) { | |
298a1665 | 2571 | assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file); |
b170e929 PB |
2572 | ret = bdrv_co_do_block_status(local_file, want_zero, local_map, |
2573 | *pnum, pnum, &local_map, &local_file); | |
99723548 | 2574 | goto out; |
61007b31 SH |
2575 | } |
2576 | ||
2577 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
2578 | ret |= BDRV_BLOCK_ALLOCATED; | |
d40f4a56 | 2579 | } else if (bs->drv->supports_backing) { |
cb850315 HR |
2580 | BlockDriverState *cow_bs = bdrv_cow_bs(bs); |
2581 | ||
d40f4a56 AG |
2582 | if (!cow_bs) { |
2583 | ret |= BDRV_BLOCK_ZERO; | |
2584 | } else if (want_zero) { | |
0af02bd1 | 2585 | int64_t size2 = bdrv_co_getlength(cow_bs); |
c9ce8c4d | 2586 | |
2e8bc787 | 2587 | if (size2 >= 0 && offset >= size2) { |
61007b31 SH |
2588 | ret |= BDRV_BLOCK_ZERO; |
2589 | } | |
2590 | } | |
2591 | } | |
2592 | ||
69f47505 VSO |
2593 | if (want_zero && ret & BDRV_BLOCK_RECURSE && |
2594 | local_file && local_file != bs && | |
61007b31 SH |
2595 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
2596 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
2e8bc787 EB |
2597 | int64_t file_pnum; |
2598 | int ret2; | |
61007b31 | 2599 | |
b170e929 PB |
2600 | ret2 = bdrv_co_do_block_status(local_file, want_zero, local_map, |
2601 | *pnum, &file_pnum, NULL, NULL); | |
61007b31 SH |
2602 | if (ret2 >= 0) { |
2603 | /* Ignore errors. This is just providing extra information, it | |
2604 | * is useful but not necessary. | |
2605 | */ | |
c61e684e EB |
2606 | if (ret2 & BDRV_BLOCK_EOF && |
2607 | (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) { | |
2608 | /* | |
2609 | * It is valid for the format block driver to read | |
2610 | * beyond the end of the underlying file's current | |
2611 | * size; such areas read as zero. | |
2612 | */ | |
61007b31 SH |
2613 | ret |= BDRV_BLOCK_ZERO; |
2614 | } else { | |
2615 | /* Limit request to the range reported by the protocol driver */ | |
2616 | *pnum = file_pnum; | |
2617 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
2618 | } | |
2619 | } | |
2620 | } | |
2621 | ||
99723548 PB |
2622 | out: |
2623 | bdrv_dec_in_flight(bs); | |
2e8bc787 | 2624 | if (ret >= 0 && offset + *pnum == total_size) { |
fb0d8654 EB |
2625 | ret |= BDRV_BLOCK_EOF; |
2626 | } | |
298a1665 EB |
2627 | early_out: |
2628 | if (file) { | |
2629 | *file = local_file; | |
2630 | } | |
2e8bc787 EB |
2631 | if (map) { |
2632 | *map = local_map; | |
2633 | } | |
61007b31 SH |
2634 | return ret; |
2635 | } | |
2636 | ||
21c2283e | 2637 | int coroutine_fn |
f9e694cb VSO |
2638 | bdrv_co_common_block_status_above(BlockDriverState *bs, |
2639 | BlockDriverState *base, | |
3555a432 | 2640 | bool include_base, |
f9e694cb VSO |
2641 | bool want_zero, |
2642 | int64_t offset, | |
2643 | int64_t bytes, | |
2644 | int64_t *pnum, | |
2645 | int64_t *map, | |
a92b1b06 EB |
2646 | BlockDriverState **file, |
2647 | int *depth) | |
ba3f0e25 | 2648 | { |
67c095c8 | 2649 | int ret; |
ba3f0e25 | 2650 | BlockDriverState *p; |
67c095c8 | 2651 | int64_t eof = 0; |
a92b1b06 | 2652 | int dummy; |
1581a70d | 2653 | IO_CODE(); |
ba3f0e25 | 2654 | |
3555a432 | 2655 | assert(!include_base || base); /* Can't include NULL base */ |
7ff9579e | 2656 | assert_bdrv_graph_readable(); |
67c095c8 | 2657 | |
a92b1b06 EB |
2658 | if (!depth) { |
2659 | depth = &dummy; | |
2660 | } | |
2661 | *depth = 0; | |
2662 | ||
624f27bb VSO |
2663 | if (!include_base && bs == base) { |
2664 | *pnum = bytes; | |
2665 | return 0; | |
2666 | } | |
2667 | ||
b170e929 PB |
2668 | ret = bdrv_co_do_block_status(bs, want_zero, offset, bytes, pnum, |
2669 | map, file); | |
a92b1b06 | 2670 | ++*depth; |
3555a432 | 2671 | if (ret < 0 || *pnum == 0 || ret & BDRV_BLOCK_ALLOCATED || bs == base) { |
67c095c8 VSO |
2672 | return ret; |
2673 | } | |
2674 | ||
2675 | if (ret & BDRV_BLOCK_EOF) { | |
2676 | eof = offset + *pnum; | |
2677 | } | |
2678 | ||
2679 | assert(*pnum <= bytes); | |
2680 | bytes = *pnum; | |
2681 | ||
3555a432 | 2682 | for (p = bdrv_filter_or_cow_bs(bs); include_base || p != base; |
67c095c8 VSO |
2683 | p = bdrv_filter_or_cow_bs(p)) |
2684 | { | |
b170e929 PB |
2685 | ret = bdrv_co_do_block_status(p, want_zero, offset, bytes, pnum, |
2686 | map, file); | |
a92b1b06 | 2687 | ++*depth; |
c61e684e | 2688 | if (ret < 0) { |
67c095c8 | 2689 | return ret; |
c61e684e | 2690 | } |
67c095c8 | 2691 | if (*pnum == 0) { |
c61e684e | 2692 | /* |
67c095c8 VSO |
2693 | * The top layer deferred to this layer, and because this layer is |
2694 | * short, any zeroes that we synthesize beyond EOF behave as if they | |
2695 | * were allocated at this layer. | |
2696 | * | |
2697 | * We don't include BDRV_BLOCK_EOF into ret, as upper layer may be | |
2698 | * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see | |
2699 | * below. | |
c61e684e | 2700 | */ |
67c095c8 | 2701 | assert(ret & BDRV_BLOCK_EOF); |
5b648c67 | 2702 | *pnum = bytes; |
67c095c8 VSO |
2703 | if (file) { |
2704 | *file = p; | |
2705 | } | |
2706 | ret = BDRV_BLOCK_ZERO | BDRV_BLOCK_ALLOCATED; | |
2707 | break; | |
c61e684e | 2708 | } |
67c095c8 VSO |
2709 | if (ret & BDRV_BLOCK_ALLOCATED) { |
2710 | /* | |
2711 | * We've found the node and the status, we must break. | |
2712 | * | |
2713 | * Drop BDRV_BLOCK_EOF, as it's not for upper layer, which may be | |
2714 | * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see | |
2715 | * below. | |
2716 | */ | |
2717 | ret &= ~BDRV_BLOCK_EOF; | |
ba3f0e25 FZ |
2718 | break; |
2719 | } | |
67c095c8 | 2720 | |
3555a432 VSO |
2721 | if (p == base) { |
2722 | assert(include_base); | |
2723 | break; | |
2724 | } | |
2725 | ||
67c095c8 VSO |
2726 | /* |
2727 | * OK, [offset, offset + *pnum) region is unallocated on this layer, | |
2728 | * let's continue the diving. | |
2729 | */ | |
2730 | assert(*pnum <= bytes); | |
2731 | bytes = *pnum; | |
ba3f0e25 | 2732 | } |
67c095c8 VSO |
2733 | |
2734 | if (offset + *pnum == eof) { | |
2735 | ret |= BDRV_BLOCK_EOF; | |
2736 | } | |
2737 | ||
ba3f0e25 FZ |
2738 | return ret; |
2739 | } | |
2740 | ||
7b52a921 EGE |
2741 | int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs, |
2742 | BlockDriverState *base, | |
2743 | int64_t offset, int64_t bytes, | |
2744 | int64_t *pnum, int64_t *map, | |
2745 | BlockDriverState **file) | |
2746 | { | |
2747 | IO_CODE(); | |
2748 | return bdrv_co_common_block_status_above(bs, base, false, true, offset, | |
2749 | bytes, pnum, map, file, NULL); | |
2750 | } | |
2751 | ||
1b88457e PB |
2752 | int coroutine_fn bdrv_co_block_status(BlockDriverState *bs, int64_t offset, |
2753 | int64_t bytes, int64_t *pnum, | |
2754 | int64_t *map, BlockDriverState **file) | |
c9ce8c4d | 2755 | { |
384a48fb | 2756 | IO_CODE(); |
1b88457e PB |
2757 | return bdrv_co_block_status_above(bs, bdrv_filter_or_cow_bs(bs), |
2758 | offset, bytes, pnum, map, file); | |
ba3f0e25 FZ |
2759 | } |
2760 | ||
46cd1e8a AG |
2761 | /* |
2762 | * Check @bs (and its backing chain) to see if the range defined | |
2763 | * by @offset and @bytes is known to read as zeroes. | |
2764 | * Return 1 if that is the case, 0 otherwise and -errno on error. | |
2765 | * This test is meant to be fast rather than accurate so returning 0 | |
2766 | * does not guarantee non-zero data. | |
2767 | */ | |
2768 | int coroutine_fn bdrv_co_is_zero_fast(BlockDriverState *bs, int64_t offset, | |
2769 | int64_t bytes) | |
2770 | { | |
2771 | int ret; | |
2772 | int64_t pnum = bytes; | |
384a48fb | 2773 | IO_CODE(); |
46cd1e8a AG |
2774 | |
2775 | if (!bytes) { | |
2776 | return 1; | |
2777 | } | |
2778 | ||
ce47ff20 AF |
2779 | ret = bdrv_co_common_block_status_above(bs, NULL, false, false, offset, |
2780 | bytes, &pnum, NULL, NULL, NULL); | |
46cd1e8a AG |
2781 | |
2782 | if (ret < 0) { | |
2783 | return ret; | |
2784 | } | |
2785 | ||
2786 | return (pnum == bytes) && (ret & BDRV_BLOCK_ZERO); | |
2787 | } | |
2788 | ||
7b52a921 EGE |
2789 | int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t offset, |
2790 | int64_t bytes, int64_t *pnum) | |
2791 | { | |
2792 | int ret; | |
2793 | int64_t dummy; | |
2794 | IO_CODE(); | |
2795 | ||
2796 | ret = bdrv_co_common_block_status_above(bs, bs, true, false, offset, | |
2797 | bytes, pnum ? pnum : &dummy, NULL, | |
2798 | NULL, NULL); | |
2799 | if (ret < 0) { | |
2800 | return ret; | |
2801 | } | |
2802 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2803 | } | |
2804 | ||
61007b31 SH |
2805 | /* |
2806 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
2807 | * | |
a92b1b06 EB |
2808 | * Return a positive depth if (a prefix of) the given range is allocated |
2809 | * in any image between BASE and TOP (BASE is only included if include_base | |
2810 | * is set). Depth 1 is TOP, 2 is the first backing layer, and so forth. | |
170d3bd3 AS |
2811 | * BASE can be NULL to check if the given offset is allocated in any |
2812 | * image of the chain. Return 0 otherwise, or negative errno on | |
2813 | * failure. | |
61007b31 | 2814 | * |
51b0a488 EB |
2815 | * 'pnum' is set to the number of bytes (including and immediately |
2816 | * following the specified offset) that are known to be in the same | |
2817 | * allocated/unallocated state. Note that a subsequent call starting | |
2818 | * at 'offset + *pnum' may return the same allocation status (in other | |
2819 | * words, the result is not necessarily the maximum possible range); | |
2820 | * but 'pnum' will only be 0 when end of file is reached. | |
61007b31 | 2821 | */ |
578ffa9f PB |
2822 | int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *bs, |
2823 | BlockDriverState *base, | |
2824 | bool include_base, int64_t offset, | |
2825 | int64_t bytes, int64_t *pnum) | |
61007b31 | 2826 | { |
a92b1b06 | 2827 | int depth; |
7b52a921 | 2828 | int ret; |
384a48fb | 2829 | IO_CODE(); |
7b52a921 | 2830 | |
578ffa9f PB |
2831 | ret = bdrv_co_common_block_status_above(bs, base, include_base, false, |
2832 | offset, bytes, pnum, NULL, NULL, | |
2833 | &depth); | |
7e7e5100 VSO |
2834 | if (ret < 0) { |
2835 | return ret; | |
61007b31 SH |
2836 | } |
2837 | ||
a92b1b06 EB |
2838 | if (ret & BDRV_BLOCK_ALLOCATED) { |
2839 | return depth; | |
2840 | } | |
2841 | return 0; | |
61007b31 SH |
2842 | } |
2843 | ||
21c2283e | 2844 | int coroutine_fn |
b33b354f | 2845 | bdrv_co_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) |
1a8ae822 KW |
2846 | { |
2847 | BlockDriver *drv = bs->drv; | |
c4db2e25 | 2848 | BlockDriverState *child_bs = bdrv_primary_bs(bs); |
b984b296 | 2849 | int ret; |
1581a70d | 2850 | IO_CODE(); |
1b3ff9fe | 2851 | assert_bdrv_graph_readable(); |
b984b296 VSO |
2852 | |
2853 | ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); | |
2854 | if (ret < 0) { | |
2855 | return ret; | |
2856 | } | |
dc88a467 | 2857 | |
b33b354f VSO |
2858 | if (!drv) { |
2859 | return -ENOMEDIUM; | |
2860 | } | |
2861 | ||
dc88a467 | 2862 | bdrv_inc_in_flight(bs); |
1a8ae822 | 2863 | |
ca5e2ad9 EGE |
2864 | if (drv->bdrv_co_load_vmstate) { |
2865 | ret = drv->bdrv_co_load_vmstate(bs, qiov, pos); | |
c4db2e25 | 2866 | } else if (child_bs) { |
b33b354f | 2867 | ret = bdrv_co_readv_vmstate(child_bs, qiov, pos); |
b984b296 VSO |
2868 | } else { |
2869 | ret = -ENOTSUP; | |
1a8ae822 KW |
2870 | } |
2871 | ||
dc88a467 | 2872 | bdrv_dec_in_flight(bs); |
b33b354f | 2873 | |
dc88a467 | 2874 | return ret; |
1a8ae822 KW |
2875 | } |
2876 | ||
b33b354f VSO |
2877 | int coroutine_fn |
2878 | bdrv_co_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 2879 | { |
b33b354f VSO |
2880 | BlockDriver *drv = bs->drv; |
2881 | BlockDriverState *child_bs = bdrv_primary_bs(bs); | |
b984b296 | 2882 | int ret; |
1581a70d | 2883 | IO_CODE(); |
1b3ff9fe | 2884 | assert_bdrv_graph_readable(); |
b984b296 VSO |
2885 | |
2886 | ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); | |
2887 | if (ret < 0) { | |
2888 | return ret; | |
2889 | } | |
61007b31 | 2890 | |
b33b354f VSO |
2891 | if (!drv) { |
2892 | return -ENOMEDIUM; | |
b433d942 KW |
2893 | } |
2894 | ||
b33b354f | 2895 | bdrv_inc_in_flight(bs); |
61007b31 | 2896 | |
ca5e2ad9 EGE |
2897 | if (drv->bdrv_co_save_vmstate) { |
2898 | ret = drv->bdrv_co_save_vmstate(bs, qiov, pos); | |
b33b354f VSO |
2899 | } else if (child_bs) { |
2900 | ret = bdrv_co_writev_vmstate(child_bs, qiov, pos); | |
b984b296 VSO |
2901 | } else { |
2902 | ret = -ENOTSUP; | |
b33b354f VSO |
2903 | } |
2904 | ||
2905 | bdrv_dec_in_flight(bs); | |
2906 | ||
2907 | return ret; | |
61007b31 SH |
2908 | } |
2909 | ||
b33b354f | 2910 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
61007b31 | 2911 | int64_t pos, int size) |
5ddda0b8 | 2912 | { |
0d93ed08 | 2913 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b33b354f | 2914 | int ret = bdrv_writev_vmstate(bs, &qiov, pos); |
384a48fb | 2915 | IO_CODE(); |
b433d942 | 2916 | |
b33b354f | 2917 | return ret < 0 ? ret : size; |
5ddda0b8 KW |
2918 | } |
2919 | ||
b33b354f VSO |
2920 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, |
2921 | int64_t pos, int size) | |
61007b31 | 2922 | { |
b33b354f VSO |
2923 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
2924 | int ret = bdrv_readv_vmstate(bs, &qiov, pos); | |
384a48fb | 2925 | IO_CODE(); |
b33b354f VSO |
2926 | |
2927 | return ret < 0 ? ret : size; | |
61007b31 SH |
2928 | } |
2929 | ||
2930 | /**************************************************************/ | |
2931 | /* async I/Os */ | |
2932 | ||
652b0dd8 SH |
2933 | /** |
2934 | * Synchronously cancels an acb. Must be called with the BQL held and the acb | |
2935 | * must be processed with the BQL held too (IOThreads are not allowed). | |
2936 | * | |
2937 | * Use bdrv_aio_cancel_async() instead when possible. | |
2938 | */ | |
61007b31 SH |
2939 | void bdrv_aio_cancel(BlockAIOCB *acb) |
2940 | { | |
652b0dd8 | 2941 | GLOBAL_STATE_CODE(); |
61007b31 SH |
2942 | qemu_aio_ref(acb); |
2943 | bdrv_aio_cancel_async(acb); | |
652b0dd8 | 2944 | AIO_WAIT_WHILE_UNLOCKED(NULL, acb->refcnt > 1); |
61007b31 SH |
2945 | qemu_aio_unref(acb); |
2946 | } | |
2947 | ||
2948 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2949 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2950 | * In either case the completion callback must be called. */ | |
2951 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2952 | { | |
384a48fb | 2953 | IO_CODE(); |
61007b31 SH |
2954 | if (acb->aiocb_info->cancel_async) { |
2955 | acb->aiocb_info->cancel_async(acb); | |
2956 | } | |
2957 | } | |
2958 | ||
61007b31 SH |
2959 | /**************************************************************/ |
2960 | /* Coroutine block device emulation */ | |
2961 | ||
61007b31 SH |
2962 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) |
2963 | { | |
883833e2 HR |
2964 | BdrvChild *primary_child = bdrv_primary_child(bs); |
2965 | BdrvChild *child; | |
49ca6259 FZ |
2966 | int current_gen; |
2967 | int ret = 0; | |
384a48fb | 2968 | IO_CODE(); |
49ca6259 | 2969 | |
88095349 | 2970 | assert_bdrv_graph_readable(); |
49ca6259 | 2971 | bdrv_inc_in_flight(bs); |
61007b31 | 2972 | |
1e97be91 | 2973 | if (!bdrv_co_is_inserted(bs) || bdrv_is_read_only(bs) || |
1b6bc94d | 2974 | bdrv_is_sg(bs)) { |
49ca6259 | 2975 | goto early_exit; |
61007b31 SH |
2976 | } |
2977 | ||
fa9185fc | 2978 | qemu_mutex_lock(&bs->reqs_lock); |
d73415a3 | 2979 | current_gen = qatomic_read(&bs->write_gen); |
3ff2f67a EY |
2980 | |
2981 | /* Wait until any previous flushes are completed */ | |
99723548 | 2982 | while (bs->active_flush_req) { |
3783fa3d | 2983 | qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock); |
3ff2f67a EY |
2984 | } |
2985 | ||
3783fa3d | 2986 | /* Flushes reach this point in nondecreasing current_gen order. */ |
99723548 | 2987 | bs->active_flush_req = true; |
fa9185fc | 2988 | qemu_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2989 | |
c32b82af PD |
2990 | /* Write back all layers by calling one driver function */ |
2991 | if (bs->drv->bdrv_co_flush) { | |
2992 | ret = bs->drv->bdrv_co_flush(bs); | |
2993 | goto out; | |
2994 | } | |
2995 | ||
61007b31 | 2996 | /* Write back cached data to the OS even with cache=unsafe */ |
17362398 | 2997 | BLKDBG_CO_EVENT(primary_child, BLKDBG_FLUSH_TO_OS); |
61007b31 SH |
2998 | if (bs->drv->bdrv_co_flush_to_os) { |
2999 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
3000 | if (ret < 0) { | |
cdb5e315 | 3001 | goto out; |
61007b31 SH |
3002 | } |
3003 | } | |
3004 | ||
3005 | /* But don't actually force it to the disk with cache=unsafe */ | |
3006 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
883833e2 | 3007 | goto flush_children; |
61007b31 SH |
3008 | } |
3009 | ||
3ff2f67a EY |
3010 | /* Check if we really need to flush anything */ |
3011 | if (bs->flushed_gen == current_gen) { | |
883833e2 | 3012 | goto flush_children; |
3ff2f67a EY |
3013 | } |
3014 | ||
17362398 | 3015 | BLKDBG_CO_EVENT(primary_child, BLKDBG_FLUSH_TO_DISK); |
d470ad42 HR |
3016 | if (!bs->drv) { |
3017 | /* bs->drv->bdrv_co_flush() might have ejected the BDS | |
3018 | * (even in case of apparent success) */ | |
3019 | ret = -ENOMEDIUM; | |
3020 | goto out; | |
3021 | } | |
61007b31 SH |
3022 | if (bs->drv->bdrv_co_flush_to_disk) { |
3023 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
3024 | } else if (bs->drv->bdrv_aio_flush) { | |
3025 | BlockAIOCB *acb; | |
3026 | CoroutineIOCompletion co = { | |
3027 | .coroutine = qemu_coroutine_self(), | |
3028 | }; | |
3029 | ||
3030 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
3031 | if (acb == NULL) { | |
3032 | ret = -EIO; | |
3033 | } else { | |
3034 | qemu_coroutine_yield(); | |
3035 | ret = co.ret; | |
3036 | } | |
3037 | } else { | |
3038 | /* | |
3039 | * Some block drivers always operate in either writethrough or unsafe | |
3040 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
3041 | * know how the server works (because the behaviour is hardcoded or | |
3042 | * depends on server-side configuration), so we can't ensure that | |
3043 | * everything is safe on disk. Returning an error doesn't work because | |
3044 | * that would break guests even if the server operates in writethrough | |
3045 | * mode. | |
3046 | * | |
3047 | * Let's hope the user knows what he's doing. | |
3048 | */ | |
3049 | ret = 0; | |
3050 | } | |
3ff2f67a | 3051 | |
61007b31 | 3052 | if (ret < 0) { |
cdb5e315 | 3053 | goto out; |
61007b31 SH |
3054 | } |
3055 | ||
3056 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
3057 | * in the case of cache=unsafe, so there are no useless flushes. | |
3058 | */ | |
883833e2 HR |
3059 | flush_children: |
3060 | ret = 0; | |
3061 | QLIST_FOREACH(child, &bs->children, next) { | |
3062 | if (child->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
3063 | int this_child_ret = bdrv_co_flush(child->bs); | |
3064 | if (!ret) { | |
3065 | ret = this_child_ret; | |
3066 | } | |
3067 | } | |
3068 | } | |
3069 | ||
cdb5e315 | 3070 | out: |
3ff2f67a | 3071 | /* Notify any pending flushes that we have completed */ |
e6af1e08 KW |
3072 | if (ret == 0) { |
3073 | bs->flushed_gen = current_gen; | |
3074 | } | |
3783fa3d | 3075 | |
fa9185fc | 3076 | qemu_mutex_lock(&bs->reqs_lock); |
99723548 | 3077 | bs->active_flush_req = false; |
156af3ac DL |
3078 | /* Return value is ignored - it's ok if wait queue is empty */ |
3079 | qemu_co_queue_next(&bs->flush_queue); | |
fa9185fc | 3080 | qemu_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 3081 | |
49ca6259 | 3082 | early_exit: |
99723548 | 3083 | bdrv_dec_in_flight(bs); |
cdb5e315 | 3084 | return ret; |
61007b31 SH |
3085 | } |
3086 | ||
d93e5726 VSO |
3087 | int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, |
3088 | int64_t bytes) | |
61007b31 | 3089 | { |
b1066c87 | 3090 | BdrvTrackedRequest req; |
39af49c0 VSO |
3091 | int ret; |
3092 | int64_t max_pdiscard; | |
3482b9bc | 3093 | int head, tail, align; |
0b9fd3f4 | 3094 | BlockDriverState *bs = child->bs; |
384a48fb | 3095 | IO_CODE(); |
9a5a1c62 | 3096 | assert_bdrv_graph_readable(); |
61007b31 | 3097 | |
1e97be91 | 3098 | if (!bs || !bs->drv || !bdrv_co_is_inserted(bs)) { |
61007b31 SH |
3099 | return -ENOMEDIUM; |
3100 | } | |
3101 | ||
d6883bc9 VSO |
3102 | if (bdrv_has_readonly_bitmaps(bs)) { |
3103 | return -EPERM; | |
3104 | } | |
3105 | ||
69b55e03 | 3106 | ret = bdrv_check_request(offset, bytes, NULL); |
8b117001 VSO |
3107 | if (ret < 0) { |
3108 | return ret; | |
61007b31 SH |
3109 | } |
3110 | ||
61007b31 SH |
3111 | /* Do nothing if disabled. */ |
3112 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
3113 | return 0; | |
3114 | } | |
3115 | ||
02aefe43 | 3116 | if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) { |
61007b31 SH |
3117 | return 0; |
3118 | } | |
3119 | ||
0bc329fb HR |
3120 | /* Invalidate the cached block-status data range if this discard overlaps */ |
3121 | bdrv_bsc_invalidate_range(bs, offset, bytes); | |
3122 | ||
3482b9bc EB |
3123 | /* Discard is advisory, but some devices track and coalesce |
3124 | * unaligned requests, so we must pass everything down rather than | |
3125 | * round here. Still, most devices will just silently ignore | |
3126 | * unaligned requests (by returning -ENOTSUP), so we must fragment | |
3127 | * the request accordingly. */ | |
02aefe43 | 3128 | align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment); |
b8d0a980 EB |
3129 | assert(align % bs->bl.request_alignment == 0); |
3130 | head = offset % align; | |
f5a5ca79 | 3131 | tail = (offset + bytes) % align; |
9f1963b3 | 3132 | |
99723548 | 3133 | bdrv_inc_in_flight(bs); |
f5a5ca79 | 3134 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD); |
50824995 | 3135 | |
00695c27 | 3136 | ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0); |
ec050f77 DL |
3137 | if (ret < 0) { |
3138 | goto out; | |
3139 | } | |
3140 | ||
6a8f3dbb | 3141 | max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT64_MAX), |
9f1963b3 | 3142 | align); |
3482b9bc | 3143 | assert(max_pdiscard >= bs->bl.request_alignment); |
61007b31 | 3144 | |
f5a5ca79 | 3145 | while (bytes > 0) { |
d93e5726 | 3146 | int64_t num = bytes; |
3482b9bc EB |
3147 | |
3148 | if (head) { | |
3149 | /* Make small requests to get to alignment boundaries. */ | |
f5a5ca79 | 3150 | num = MIN(bytes, align - head); |
3482b9bc EB |
3151 | if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) { |
3152 | num %= bs->bl.request_alignment; | |
3153 | } | |
3154 | head = (head + num) % align; | |
3155 | assert(num < max_pdiscard); | |
3156 | } else if (tail) { | |
3157 | if (num > align) { | |
3158 | /* Shorten the request to the last aligned cluster. */ | |
3159 | num -= tail; | |
3160 | } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) && | |
3161 | tail > bs->bl.request_alignment) { | |
3162 | tail %= bs->bl.request_alignment; | |
3163 | num -= tail; | |
3164 | } | |
3165 | } | |
3166 | /* limit request size */ | |
3167 | if (num > max_pdiscard) { | |
3168 | num = max_pdiscard; | |
3169 | } | |
61007b31 | 3170 | |
d470ad42 HR |
3171 | if (!bs->drv) { |
3172 | ret = -ENOMEDIUM; | |
3173 | goto out; | |
3174 | } | |
47a5486d EB |
3175 | if (bs->drv->bdrv_co_pdiscard) { |
3176 | ret = bs->drv->bdrv_co_pdiscard(bs, offset, num); | |
61007b31 SH |
3177 | } else { |
3178 | BlockAIOCB *acb; | |
3179 | CoroutineIOCompletion co = { | |
3180 | .coroutine = qemu_coroutine_self(), | |
3181 | }; | |
3182 | ||
4da444a0 EB |
3183 | acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num, |
3184 | bdrv_co_io_em_complete, &co); | |
61007b31 | 3185 | if (acb == NULL) { |
b1066c87 FZ |
3186 | ret = -EIO; |
3187 | goto out; | |
61007b31 SH |
3188 | } else { |
3189 | qemu_coroutine_yield(); | |
3190 | ret = co.ret; | |
3191 | } | |
3192 | } | |
3193 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 3194 | goto out; |
61007b31 SH |
3195 | } |
3196 | ||
9f1963b3 | 3197 | offset += num; |
f5a5ca79 | 3198 | bytes -= num; |
61007b31 | 3199 | } |
b1066c87 FZ |
3200 | ret = 0; |
3201 | out: | |
00695c27 | 3202 | bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret); |
b1066c87 | 3203 | tracked_request_end(&req); |
99723548 | 3204 | bdrv_dec_in_flight(bs); |
b1066c87 | 3205 | return ret; |
61007b31 SH |
3206 | } |
3207 | ||
881a4c55 | 3208 | int coroutine_fn bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
3209 | { |
3210 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
3211 | CoroutineIOCompletion co = { |
3212 | .coroutine = qemu_coroutine_self(), | |
3213 | }; | |
3214 | BlockAIOCB *acb; | |
384a48fb | 3215 | IO_CODE(); |
26c518ab | 3216 | assert_bdrv_graph_readable(); |
61007b31 | 3217 | |
99723548 | 3218 | bdrv_inc_in_flight(bs); |
16a389dc | 3219 | if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) { |
5c5ae76a FZ |
3220 | co.ret = -ENOTSUP; |
3221 | goto out; | |
3222 | } | |
3223 | ||
16a389dc KW |
3224 | if (drv->bdrv_co_ioctl) { |
3225 | co.ret = drv->bdrv_co_ioctl(bs, req, buf); | |
3226 | } else { | |
3227 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
3228 | if (!acb) { | |
3229 | co.ret = -ENOTSUP; | |
3230 | goto out; | |
3231 | } | |
3232 | qemu_coroutine_yield(); | |
5c5ae76a | 3233 | } |
5c5ae76a | 3234 | out: |
99723548 | 3235 | bdrv_dec_in_flight(bs); |
5c5ae76a FZ |
3236 | return co.ret; |
3237 | } | |
3238 | ||
6d43eaa3 SL |
3239 | int coroutine_fn bdrv_co_zone_report(BlockDriverState *bs, int64_t offset, |
3240 | unsigned int *nr_zones, | |
3241 | BlockZoneDescriptor *zones) | |
3242 | { | |
3243 | BlockDriver *drv = bs->drv; | |
3244 | CoroutineIOCompletion co = { | |
3245 | .coroutine = qemu_coroutine_self(), | |
3246 | }; | |
3247 | IO_CODE(); | |
3248 | ||
3249 | bdrv_inc_in_flight(bs); | |
3250 | if (!drv || !drv->bdrv_co_zone_report || bs->bl.zoned == BLK_Z_NONE) { | |
3251 | co.ret = -ENOTSUP; | |
3252 | goto out; | |
3253 | } | |
3254 | co.ret = drv->bdrv_co_zone_report(bs, offset, nr_zones, zones); | |
3255 | out: | |
3256 | bdrv_dec_in_flight(bs); | |
3257 | return co.ret; | |
3258 | } | |
3259 | ||
3260 | int coroutine_fn bdrv_co_zone_mgmt(BlockDriverState *bs, BlockZoneOp op, | |
3261 | int64_t offset, int64_t len) | |
3262 | { | |
3263 | BlockDriver *drv = bs->drv; | |
3264 | CoroutineIOCompletion co = { | |
3265 | .coroutine = qemu_coroutine_self(), | |
3266 | }; | |
3267 | IO_CODE(); | |
3268 | ||
3269 | bdrv_inc_in_flight(bs); | |
3270 | if (!drv || !drv->bdrv_co_zone_mgmt || bs->bl.zoned == BLK_Z_NONE) { | |
3271 | co.ret = -ENOTSUP; | |
3272 | goto out; | |
3273 | } | |
3274 | co.ret = drv->bdrv_co_zone_mgmt(bs, op, offset, len); | |
3275 | out: | |
3276 | bdrv_dec_in_flight(bs); | |
3277 | return co.ret; | |
3278 | } | |
3279 | ||
4751d09a SL |
3280 | int coroutine_fn bdrv_co_zone_append(BlockDriverState *bs, int64_t *offset, |
3281 | QEMUIOVector *qiov, | |
3282 | BdrvRequestFlags flags) | |
3283 | { | |
3284 | int ret; | |
3285 | BlockDriver *drv = bs->drv; | |
3286 | CoroutineIOCompletion co = { | |
3287 | .coroutine = qemu_coroutine_self(), | |
3288 | }; | |
3289 | IO_CODE(); | |
3290 | ||
3291 | ret = bdrv_check_qiov_request(*offset, qiov->size, qiov, 0, NULL); | |
3292 | if (ret < 0) { | |
3293 | return ret; | |
3294 | } | |
3295 | ||
3296 | bdrv_inc_in_flight(bs); | |
3297 | if (!drv || !drv->bdrv_co_zone_append || bs->bl.zoned == BLK_Z_NONE) { | |
3298 | co.ret = -ENOTSUP; | |
3299 | goto out; | |
3300 | } | |
3301 | co.ret = drv->bdrv_co_zone_append(bs, offset, qiov, flags); | |
3302 | out: | |
3303 | bdrv_dec_in_flight(bs); | |
3304 | return co.ret; | |
3305 | } | |
3306 | ||
61007b31 SH |
3307 | void *qemu_blockalign(BlockDriverState *bs, size_t size) |
3308 | { | |
384a48fb | 3309 | IO_CODE(); |
61007b31 SH |
3310 | return qemu_memalign(bdrv_opt_mem_align(bs), size); |
3311 | } | |
3312 | ||
3313 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
3314 | { | |
384a48fb | 3315 | IO_CODE(); |
61007b31 SH |
3316 | return memset(qemu_blockalign(bs, size), 0, size); |
3317 | } | |
3318 | ||
3319 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
3320 | { | |
3321 | size_t align = bdrv_opt_mem_align(bs); | |
384a48fb | 3322 | IO_CODE(); |
61007b31 SH |
3323 | |
3324 | /* Ensure that NULL is never returned on success */ | |
3325 | assert(align > 0); | |
3326 | if (size == 0) { | |
3327 | size = align; | |
3328 | } | |
3329 | ||
3330 | return qemu_try_memalign(align, size); | |
3331 | } | |
3332 | ||
3333 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
3334 | { | |
3335 | void *mem = qemu_try_blockalign(bs, size); | |
384a48fb | 3336 | IO_CODE(); |
61007b31 SH |
3337 | |
3338 | if (mem) { | |
3339 | memset(mem, 0, size); | |
3340 | } | |
3341 | ||
3342 | return mem; | |
3343 | } | |
3344 | ||
f4ec04ba | 3345 | /* Helper that undoes bdrv_register_buf() when it fails partway through */ |
d9249c25 KW |
3346 | static void GRAPH_RDLOCK |
3347 | bdrv_register_buf_rollback(BlockDriverState *bs, void *host, size_t size, | |
3348 | BdrvChild *final_child) | |
f4ec04ba SH |
3349 | { |
3350 | BdrvChild *child; | |
3351 | ||
d9249c25 KW |
3352 | GLOBAL_STATE_CODE(); |
3353 | assert_bdrv_graph_readable(); | |
3354 | ||
f4ec04ba SH |
3355 | QLIST_FOREACH(child, &bs->children, next) { |
3356 | if (child == final_child) { | |
3357 | break; | |
3358 | } | |
3359 | ||
3360 | bdrv_unregister_buf(child->bs, host, size); | |
3361 | } | |
3362 | ||
3363 | if (bs->drv && bs->drv->bdrv_unregister_buf) { | |
3364 | bs->drv->bdrv_unregister_buf(bs, host, size); | |
3365 | } | |
3366 | } | |
3367 | ||
3368 | bool bdrv_register_buf(BlockDriverState *bs, void *host, size_t size, | |
3369 | Error **errp) | |
23d0ba93 FZ |
3370 | { |
3371 | BdrvChild *child; | |
3372 | ||
f791bf7f | 3373 | GLOBAL_STATE_CODE(); |
d9249c25 KW |
3374 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
3375 | ||
23d0ba93 | 3376 | if (bs->drv && bs->drv->bdrv_register_buf) { |
f4ec04ba SH |
3377 | if (!bs->drv->bdrv_register_buf(bs, host, size, errp)) { |
3378 | return false; | |
3379 | } | |
23d0ba93 FZ |
3380 | } |
3381 | QLIST_FOREACH(child, &bs->children, next) { | |
f4ec04ba SH |
3382 | if (!bdrv_register_buf(child->bs, host, size, errp)) { |
3383 | bdrv_register_buf_rollback(bs, host, size, child); | |
3384 | return false; | |
3385 | } | |
23d0ba93 | 3386 | } |
f4ec04ba | 3387 | return true; |
23d0ba93 FZ |
3388 | } |
3389 | ||
4f384011 | 3390 | void bdrv_unregister_buf(BlockDriverState *bs, void *host, size_t size) |
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_unregister_buf) { |
4f384011 | 3398 | bs->drv->bdrv_unregister_buf(bs, host, size); |
23d0ba93 FZ |
3399 | } |
3400 | QLIST_FOREACH(child, &bs->children, next) { | |
4f384011 | 3401 | bdrv_unregister_buf(child->bs, host, size); |
23d0ba93 FZ |
3402 | } |
3403 | } | |
fcc67678 | 3404 | |
abaf8b75 | 3405 | static int coroutine_fn GRAPH_RDLOCK bdrv_co_copy_range_internal( |
a5215b8f VSO |
3406 | BdrvChild *src, int64_t src_offset, BdrvChild *dst, |
3407 | int64_t dst_offset, int64_t bytes, | |
67b51fb9 VSO |
3408 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags, |
3409 | bool recurse_src) | |
fcc67678 | 3410 | { |
999658a0 | 3411 | BdrvTrackedRequest req; |
fcc67678 | 3412 | int ret; |
742bf09b | 3413 | assert_bdrv_graph_readable(); |
fcc67678 | 3414 | |
fe0480d6 KW |
3415 | /* TODO We can support BDRV_REQ_NO_FALLBACK here */ |
3416 | assert(!(read_flags & BDRV_REQ_NO_FALLBACK)); | |
3417 | assert(!(write_flags & BDRV_REQ_NO_FALLBACK)); | |
45e62b46 VSO |
3418 | assert(!(read_flags & BDRV_REQ_NO_WAIT)); |
3419 | assert(!(write_flags & BDRV_REQ_NO_WAIT)); | |
fe0480d6 | 3420 | |
1e97be91 | 3421 | if (!dst || !dst->bs || !bdrv_co_is_inserted(dst->bs)) { |
fcc67678 FZ |
3422 | return -ENOMEDIUM; |
3423 | } | |
63f4ad11 | 3424 | ret = bdrv_check_request32(dst_offset, bytes, NULL, 0); |
fcc67678 FZ |
3425 | if (ret) { |
3426 | return ret; | |
3427 | } | |
67b51fb9 VSO |
3428 | if (write_flags & BDRV_REQ_ZERO_WRITE) { |
3429 | return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags); | |
fcc67678 FZ |
3430 | } |
3431 | ||
1e97be91 | 3432 | if (!src || !src->bs || !bdrv_co_is_inserted(src->bs)) { |
d4d3e5a0 FZ |
3433 | return -ENOMEDIUM; |
3434 | } | |
63f4ad11 | 3435 | ret = bdrv_check_request32(src_offset, bytes, NULL, 0); |
d4d3e5a0 FZ |
3436 | if (ret) { |
3437 | return ret; | |
3438 | } | |
3439 | ||
fcc67678 FZ |
3440 | if (!src->bs->drv->bdrv_co_copy_range_from |
3441 | || !dst->bs->drv->bdrv_co_copy_range_to | |
3442 | || src->bs->encrypted || dst->bs->encrypted) { | |
3443 | return -ENOTSUP; | |
3444 | } | |
37aec7d7 | 3445 | |
fcc67678 | 3446 | if (recurse_src) { |
999658a0 VSO |
3447 | bdrv_inc_in_flight(src->bs); |
3448 | tracked_request_begin(&req, src->bs, src_offset, bytes, | |
3449 | BDRV_TRACKED_READ); | |
3450 | ||
09d2f948 VSO |
3451 | /* BDRV_REQ_SERIALISING is only for write operation */ |
3452 | assert(!(read_flags & BDRV_REQ_SERIALISING)); | |
c53cb427 | 3453 | bdrv_wait_serialising_requests(&req); |
999658a0 | 3454 | |
37aec7d7 FZ |
3455 | ret = src->bs->drv->bdrv_co_copy_range_from(src->bs, |
3456 | src, src_offset, | |
3457 | dst, dst_offset, | |
67b51fb9 VSO |
3458 | bytes, |
3459 | read_flags, write_flags); | |
999658a0 VSO |
3460 | |
3461 | tracked_request_end(&req); | |
3462 | bdrv_dec_in_flight(src->bs); | |
fcc67678 | 3463 | } else { |
999658a0 VSO |
3464 | bdrv_inc_in_flight(dst->bs); |
3465 | tracked_request_begin(&req, dst->bs, dst_offset, bytes, | |
3466 | BDRV_TRACKED_WRITE); | |
0eb1e891 FZ |
3467 | ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req, |
3468 | write_flags); | |
3469 | if (!ret) { | |
3470 | ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs, | |
3471 | src, src_offset, | |
3472 | dst, dst_offset, | |
3473 | bytes, | |
3474 | read_flags, write_flags); | |
3475 | } | |
3476 | bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret); | |
999658a0 VSO |
3477 | tracked_request_end(&req); |
3478 | bdrv_dec_in_flight(dst->bs); | |
fcc67678 | 3479 | } |
999658a0 | 3480 | |
37aec7d7 | 3481 | return ret; |
fcc67678 FZ |
3482 | } |
3483 | ||
3484 | /* Copy range from @src to @dst. | |
3485 | * | |
3486 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3487 | * semantics. */ | |
a5215b8f VSO |
3488 | int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, int64_t src_offset, |
3489 | BdrvChild *dst, int64_t dst_offset, | |
3490 | int64_t bytes, | |
67b51fb9 VSO |
3491 | BdrvRequestFlags read_flags, |
3492 | BdrvRequestFlags write_flags) | |
fcc67678 | 3493 | { |
967d7905 | 3494 | IO_CODE(); |
742bf09b | 3495 | assert_bdrv_graph_readable(); |
ecc983a5 FZ |
3496 | trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes, |
3497 | read_flags, write_flags); | |
fcc67678 | 3498 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3499 | bytes, read_flags, write_flags, true); |
fcc67678 FZ |
3500 | } |
3501 | ||
3502 | /* Copy range from @src to @dst. | |
3503 | * | |
3504 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3505 | * semantics. */ | |
a5215b8f VSO |
3506 | int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, int64_t src_offset, |
3507 | BdrvChild *dst, int64_t dst_offset, | |
3508 | int64_t bytes, | |
67b51fb9 VSO |
3509 | BdrvRequestFlags read_flags, |
3510 | BdrvRequestFlags write_flags) | |
fcc67678 | 3511 | { |
967d7905 | 3512 | IO_CODE(); |
742bf09b | 3513 | assert_bdrv_graph_readable(); |
ecc983a5 FZ |
3514 | trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3515 | read_flags, write_flags); | |
fcc67678 | 3516 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3517 | bytes, read_flags, write_flags, false); |
fcc67678 FZ |
3518 | } |
3519 | ||
a5215b8f VSO |
3520 | int coroutine_fn bdrv_co_copy_range(BdrvChild *src, int64_t src_offset, |
3521 | BdrvChild *dst, int64_t dst_offset, | |
3522 | int64_t bytes, BdrvRequestFlags read_flags, | |
67b51fb9 | 3523 | BdrvRequestFlags write_flags) |
fcc67678 | 3524 | { |
384a48fb | 3525 | IO_CODE(); |
742bf09b EGE |
3526 | assert_bdrv_graph_readable(); |
3527 | ||
37aec7d7 FZ |
3528 | return bdrv_co_copy_range_from(src, src_offset, |
3529 | dst, dst_offset, | |
67b51fb9 | 3530 | bytes, read_flags, write_flags); |
fcc67678 | 3531 | } |
3d9f2d2a KW |
3532 | |
3533 | static void bdrv_parent_cb_resize(BlockDriverState *bs) | |
3534 | { | |
3535 | BdrvChild *c; | |
3536 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
3537 | if (c->klass->resize) { |
3538 | c->klass->resize(c); | |
3d9f2d2a KW |
3539 | } |
3540 | } | |
3541 | } | |
3542 | ||
3543 | /** | |
3544 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
c80d8b06 HR |
3545 | * |
3546 | * If 'exact' is true, the file must be resized to exactly the given | |
3547 | * 'offset'. Otherwise, it is sufficient for the node to be at least | |
3548 | * 'offset' bytes in length. | |
3d9f2d2a | 3549 | */ |
c80d8b06 | 3550 | int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, |
7b8e4857 KW |
3551 | PreallocMode prealloc, BdrvRequestFlags flags, |
3552 | Error **errp) | |
3d9f2d2a KW |
3553 | { |
3554 | BlockDriverState *bs = child->bs; | |
23b93525 | 3555 | BdrvChild *filtered, *backing; |
3d9f2d2a | 3556 | BlockDriver *drv = bs->drv; |
1bc5f09f KW |
3557 | BdrvTrackedRequest req; |
3558 | int64_t old_size, new_bytes; | |
3d9f2d2a | 3559 | int ret; |
384a48fb | 3560 | IO_CODE(); |
c2b8e315 | 3561 | assert_bdrv_graph_readable(); |
3d9f2d2a KW |
3562 | |
3563 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ | |
3564 | if (!drv) { | |
3565 | error_setg(errp, "No medium inserted"); | |
3566 | return -ENOMEDIUM; | |
3567 | } | |
3568 | if (offset < 0) { | |
3569 | error_setg(errp, "Image size cannot be negative"); | |
3570 | return -EINVAL; | |
3571 | } | |
3572 | ||
69b55e03 | 3573 | ret = bdrv_check_request(offset, 0, errp); |
8b117001 | 3574 | if (ret < 0) { |
8b117001 VSO |
3575 | return ret; |
3576 | } | |
3577 | ||
0af02bd1 | 3578 | old_size = bdrv_co_getlength(bs); |
1bc5f09f KW |
3579 | if (old_size < 0) { |
3580 | error_setg_errno(errp, -old_size, "Failed to get old image size"); | |
3581 | return old_size; | |
3582 | } | |
3583 | ||
97efa869 EB |
3584 | if (bdrv_is_read_only(bs)) { |
3585 | error_setg(errp, "Image is read-only"); | |
3586 | return -EACCES; | |
3587 | } | |
3588 | ||
1bc5f09f KW |
3589 | if (offset > old_size) { |
3590 | new_bytes = offset - old_size; | |
3591 | } else { | |
3592 | new_bytes = 0; | |
3593 | } | |
3594 | ||
3d9f2d2a | 3595 | bdrv_inc_in_flight(bs); |
5416a11e FZ |
3596 | tracked_request_begin(&req, bs, offset - new_bytes, new_bytes, |
3597 | BDRV_TRACKED_TRUNCATE); | |
1bc5f09f KW |
3598 | |
3599 | /* If we are growing the image and potentially using preallocation for the | |
3600 | * new area, we need to make sure that no write requests are made to it | |
3601 | * concurrently or they might be overwritten by preallocation. */ | |
3602 | if (new_bytes) { | |
8ac5aab2 | 3603 | bdrv_make_request_serialising(&req, 1); |
cd47d792 | 3604 | } |
cd47d792 FZ |
3605 | ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req, |
3606 | 0); | |
3607 | if (ret < 0) { | |
3608 | error_setg_errno(errp, -ret, | |
3609 | "Failed to prepare request for truncation"); | |
3610 | goto out; | |
1bc5f09f | 3611 | } |
3d9f2d2a | 3612 | |
93393e69 | 3613 | filtered = bdrv_filter_child(bs); |
23b93525 | 3614 | backing = bdrv_cow_child(bs); |
93393e69 | 3615 | |
955c7d66 KW |
3616 | /* |
3617 | * If the image has a backing file that is large enough that it would | |
3618 | * provide data for the new area, we cannot leave it unallocated because | |
3619 | * then the backing file content would become visible. Instead, zero-fill | |
3620 | * the new area. | |
3621 | * | |
3622 | * Note that if the image has a backing file, but was opened without the | |
3623 | * backing file, taking care of keeping things consistent with that backing | |
3624 | * file is the user's responsibility. | |
3625 | */ | |
23b93525 | 3626 | if (new_bytes && backing) { |
955c7d66 KW |
3627 | int64_t backing_len; |
3628 | ||
bd53086e | 3629 | backing_len = bdrv_co_getlength(backing->bs); |
955c7d66 KW |
3630 | if (backing_len < 0) { |
3631 | ret = backing_len; | |
3632 | error_setg_errno(errp, -ret, "Could not get backing file size"); | |
3633 | goto out; | |
3634 | } | |
3635 | ||
3636 | if (backing_len > old_size) { | |
3637 | flags |= BDRV_REQ_ZERO_WRITE; | |
3638 | } | |
3639 | } | |
3640 | ||
6b7e8f8b | 3641 | if (drv->bdrv_co_truncate) { |
92b92799 KW |
3642 | if (flags & ~bs->supported_truncate_flags) { |
3643 | error_setg(errp, "Block driver does not support requested flags"); | |
3644 | ret = -ENOTSUP; | |
3645 | goto out; | |
3646 | } | |
3647 | ret = drv->bdrv_co_truncate(bs, offset, exact, prealloc, flags, errp); | |
93393e69 HR |
3648 | } else if (filtered) { |
3649 | ret = bdrv_co_truncate(filtered, offset, exact, prealloc, flags, errp); | |
6b7e8f8b | 3650 | } else { |
3d9f2d2a KW |
3651 | error_setg(errp, "Image format driver does not support resize"); |
3652 | ret = -ENOTSUP; | |
3653 | goto out; | |
3654 | } | |
3d9f2d2a KW |
3655 | if (ret < 0) { |
3656 | goto out; | |
3657 | } | |
6b7e8f8b | 3658 | |
bd53086e | 3659 | ret = bdrv_co_refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); |
3d9f2d2a KW |
3660 | if (ret < 0) { |
3661 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
3662 | } else { | |
3663 | offset = bs->total_sectors * BDRV_SECTOR_SIZE; | |
3664 | } | |
c057960c EGE |
3665 | /* |
3666 | * It's possible that truncation succeeded but bdrv_refresh_total_sectors | |
cd47d792 | 3667 | * failed, but the latter doesn't affect how we should finish the request. |
c057960c EGE |
3668 | * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. |
3669 | */ | |
cd47d792 | 3670 | bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0); |
3d9f2d2a KW |
3671 | |
3672 | out: | |
1bc5f09f | 3673 | tracked_request_end(&req); |
3d9f2d2a | 3674 | bdrv_dec_in_flight(bs); |
1bc5f09f | 3675 | |
3d9f2d2a KW |
3676 | return ret; |
3677 | } | |
bd54669a VSO |
3678 | |
3679 | void bdrv_cancel_in_flight(BlockDriverState *bs) | |
3680 | { | |
f791bf7f | 3681 | GLOBAL_STATE_CODE(); |
bd54669a VSO |
3682 | if (!bs || !bs->drv) { |
3683 | return; | |
3684 | } | |
3685 | ||
3686 | if (bs->drv->bdrv_cancel_in_flight) { | |
3687 | bs->drv->bdrv_cancel_in_flight(bs); | |
3688 | } | |
3689 | } | |
ce14f3b4 VSO |
3690 | |
3691 | int coroutine_fn | |
3692 | bdrv_co_preadv_snapshot(BdrvChild *child, int64_t offset, int64_t bytes, | |
3693 | QEMUIOVector *qiov, size_t qiov_offset) | |
3694 | { | |
3695 | BlockDriverState *bs = child->bs; | |
3696 | BlockDriver *drv = bs->drv; | |
3697 | int ret; | |
3698 | IO_CODE(); | |
7b9e8b22 | 3699 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3700 | |
3701 | if (!drv) { | |
3702 | return -ENOMEDIUM; | |
3703 | } | |
3704 | ||
3705 | if (!drv->bdrv_co_preadv_snapshot) { | |
3706 | return -ENOTSUP; | |
3707 | } | |
3708 | ||
3709 | bdrv_inc_in_flight(bs); | |
3710 | ret = drv->bdrv_co_preadv_snapshot(bs, offset, bytes, qiov, qiov_offset); | |
3711 | bdrv_dec_in_flight(bs); | |
3712 | ||
3713 | return ret; | |
3714 | } | |
3715 | ||
3716 | int coroutine_fn | |
3717 | bdrv_co_snapshot_block_status(BlockDriverState *bs, | |
3718 | bool want_zero, int64_t offset, int64_t bytes, | |
3719 | int64_t *pnum, int64_t *map, | |
3720 | BlockDriverState **file) | |
3721 | { | |
3722 | BlockDriver *drv = bs->drv; | |
3723 | int ret; | |
3724 | IO_CODE(); | |
7b9e8b22 | 3725 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3726 | |
3727 | if (!drv) { | |
3728 | return -ENOMEDIUM; | |
3729 | } | |
3730 | ||
3731 | if (!drv->bdrv_co_snapshot_block_status) { | |
3732 | return -ENOTSUP; | |
3733 | } | |
3734 | ||
3735 | bdrv_inc_in_flight(bs); | |
3736 | ret = drv->bdrv_co_snapshot_block_status(bs, want_zero, offset, bytes, | |
3737 | pnum, map, file); | |
3738 | bdrv_dec_in_flight(bs); | |
3739 | ||
3740 | return ret; | |
3741 | } | |
3742 | ||
3743 | int coroutine_fn | |
3744 | bdrv_co_pdiscard_snapshot(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
3745 | { | |
3746 | BlockDriver *drv = bs->drv; | |
3747 | int ret; | |
3748 | IO_CODE(); | |
9a5a1c62 | 3749 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3750 | |
3751 | if (!drv) { | |
3752 | return -ENOMEDIUM; | |
3753 | } | |
3754 | ||
3755 | if (!drv->bdrv_co_pdiscard_snapshot) { | |
3756 | return -ENOTSUP; | |
3757 | } | |
3758 | ||
3759 | bdrv_inc_in_flight(bs); | |
3760 | ret = drv->bdrv_co_pdiscard_snapshot(bs, offset, bytes); | |
3761 | bdrv_dec_in_flight(bs); | |
3762 | ||
3763 | return ret; | |
3764 | } |