]>
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 | ||
7859c45a KW |
45 | static void coroutine_fn GRAPH_RDLOCK |
46 | bdrv_parent_cb_resize(BlockDriverState *bs); | |
47 | ||
d05aa8bb | 48 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
5ae07b14 | 49 | int64_t offset, int64_t bytes, BdrvRequestFlags flags); |
61007b31 | 50 | |
d05ab380 EGE |
51 | static void GRAPH_RDLOCK |
52 | bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore) | |
61007b31 | 53 | { |
02d21300 | 54 | BdrvChild *c, *next; |
d05ab380 EGE |
55 | IO_OR_GS_CODE(); |
56 | assert_bdrv_graph_readable(); | |
27ccdd52 | 57 | |
02d21300 | 58 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { |
a82a3bd1 | 59 | if (c == ignore) { |
0152bf40 KW |
60 | continue; |
61 | } | |
606ed756 | 62 | bdrv_parent_drained_begin_single(c); |
ce0f1412 PB |
63 | } |
64 | } | |
61007b31 | 65 | |
2f65df6e | 66 | void bdrv_parent_drained_end_single(BdrvChild *c) |
804db8ea | 67 | { |
ab613350 | 68 | GLOBAL_STATE_CODE(); |
2f65df6e | 69 | |
57e05be3 KW |
70 | assert(c->quiesced_parent); |
71 | c->quiesced_parent = false; | |
72 | ||
bd86fb99 | 73 | if (c->klass->drained_end) { |
2f65df6e | 74 | c->klass->drained_end(c); |
804db8ea HR |
75 | } |
76 | } | |
77 | ||
d05ab380 EGE |
78 | static void GRAPH_RDLOCK |
79 | bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore) | |
ce0f1412 | 80 | { |
61ad631c | 81 | BdrvChild *c; |
d05ab380 EGE |
82 | IO_OR_GS_CODE(); |
83 | assert_bdrv_graph_readable(); | |
27ccdd52 | 84 | |
61ad631c | 85 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
a82a3bd1 | 86 | if (c == ignore) { |
0152bf40 KW |
87 | continue; |
88 | } | |
2f65df6e | 89 | bdrv_parent_drained_end_single(c); |
27ccdd52 | 90 | } |
61007b31 SH |
91 | } |
92 | ||
23987471 | 93 | bool bdrv_parent_drained_poll_single(BdrvChild *c) |
4be6a6d1 | 94 | { |
d05ab380 EGE |
95 | IO_OR_GS_CODE(); |
96 | ||
bd86fb99 HR |
97 | if (c->klass->drained_poll) { |
98 | return c->klass->drained_poll(c); | |
4be6a6d1 KW |
99 | } |
100 | return false; | |
101 | } | |
102 | ||
d05ab380 EGE |
103 | static bool GRAPH_RDLOCK |
104 | bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore, | |
105 | bool ignore_bds_parents) | |
89bd0305 KW |
106 | { |
107 | BdrvChild *c, *next; | |
108 | bool busy = false; | |
d05ab380 EGE |
109 | IO_OR_GS_CODE(); |
110 | assert_bdrv_graph_readable(); | |
89bd0305 KW |
111 | |
112 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { | |
bd86fb99 | 113 | if (c == ignore || (ignore_bds_parents && c->klass->parent_is_bds)) { |
89bd0305 KW |
114 | continue; |
115 | } | |
4be6a6d1 | 116 | busy |= bdrv_parent_drained_poll_single(c); |
89bd0305 KW |
117 | } |
118 | ||
119 | return busy; | |
120 | } | |
121 | ||
606ed756 | 122 | void bdrv_parent_drained_begin_single(BdrvChild *c) |
4be6a6d1 | 123 | { |
ab613350 | 124 | GLOBAL_STATE_CODE(); |
57e05be3 KW |
125 | |
126 | assert(!c->quiesced_parent); | |
127 | c->quiesced_parent = true; | |
128 | ||
bd86fb99 | 129 | if (c->klass->drained_begin) { |
d05ab380 | 130 | /* called with rdlock taken, but it doesn't really need it. */ |
bd86fb99 | 131 | c->klass->drained_begin(c); |
4be6a6d1 | 132 | } |
4be6a6d1 KW |
133 | } |
134 | ||
d9e0dfa2 EB |
135 | static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) |
136 | { | |
9f460c64 AO |
137 | dst->pdiscard_alignment = MAX(dst->pdiscard_alignment, |
138 | src->pdiscard_alignment); | |
d9e0dfa2 EB |
139 | dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); |
140 | dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer); | |
24b36e98 PB |
141 | dst->max_hw_transfer = MIN_NON_ZERO(dst->max_hw_transfer, |
142 | src->max_hw_transfer); | |
d9e0dfa2 EB |
143 | dst->opt_mem_alignment = MAX(dst->opt_mem_alignment, |
144 | src->opt_mem_alignment); | |
145 | dst->min_mem_alignment = MAX(dst->min_mem_alignment, | |
146 | src->min_mem_alignment); | |
147 | dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov); | |
cc071629 | 148 | dst->max_hw_iov = MIN_NON_ZERO(dst->max_hw_iov, src->max_hw_iov); |
d9e0dfa2 EB |
149 | } |
150 | ||
1e4c797c VSO |
151 | typedef struct BdrvRefreshLimitsState { |
152 | BlockDriverState *bs; | |
153 | BlockLimits old_bl; | |
154 | } BdrvRefreshLimitsState; | |
155 | ||
156 | static void bdrv_refresh_limits_abort(void *opaque) | |
157 | { | |
158 | BdrvRefreshLimitsState *s = opaque; | |
159 | ||
160 | s->bs->bl = s->old_bl; | |
161 | } | |
162 | ||
163 | static TransactionActionDrv bdrv_refresh_limits_drv = { | |
164 | .abort = bdrv_refresh_limits_abort, | |
165 | .clean = g_free, | |
166 | }; | |
167 | ||
168 | /* @tran is allowed to be NULL, in this case no rollback is possible. */ | |
169 | void bdrv_refresh_limits(BlockDriverState *bs, Transaction *tran, Error **errp) | |
61007b31 | 170 | { |
33985614 | 171 | ERRP_GUARD(); |
61007b31 | 172 | BlockDriver *drv = bs->drv; |
66b129ac HR |
173 | BdrvChild *c; |
174 | bool have_limits; | |
61007b31 | 175 | |
f791bf7f EGE |
176 | GLOBAL_STATE_CODE(); |
177 | ||
1e4c797c VSO |
178 | if (tran) { |
179 | BdrvRefreshLimitsState *s = g_new(BdrvRefreshLimitsState, 1); | |
180 | *s = (BdrvRefreshLimitsState) { | |
181 | .bs = bs, | |
182 | .old_bl = bs->bl, | |
183 | }; | |
184 | tran_add(tran, &bdrv_refresh_limits_drv, s); | |
185 | } | |
186 | ||
61007b31 SH |
187 | memset(&bs->bl, 0, sizeof(bs->bl)); |
188 | ||
189 | if (!drv) { | |
190 | return; | |
191 | } | |
192 | ||
79ba8c98 | 193 | /* Default alignment based on whether driver has byte interface */ |
e31f6864 | 194 | bs->bl.request_alignment = (drv->bdrv_co_preadv || |
ac850bf0 VSO |
195 | drv->bdrv_aio_preadv || |
196 | drv->bdrv_co_preadv_part) ? 1 : 512; | |
79ba8c98 | 197 | |
61007b31 | 198 | /* Take some limits from the children as a default */ |
66b129ac HR |
199 | have_limits = false; |
200 | QLIST_FOREACH(c, &bs->children, next) { | |
201 | if (c->role & (BDRV_CHILD_DATA | BDRV_CHILD_FILTERED | BDRV_CHILD_COW)) | |
202 | { | |
66b129ac HR |
203 | bdrv_merge_limits(&bs->bl, &c->bs->bl); |
204 | have_limits = true; | |
61007b31 | 205 | } |
160a29e2 PB |
206 | |
207 | if (c->role & BDRV_CHILD_FILTERED) { | |
208 | bs->bl.has_variable_length |= c->bs->bl.has_variable_length; | |
209 | } | |
66b129ac HR |
210 | } |
211 | ||
212 | if (!have_limits) { | |
4196d2f0 | 213 | bs->bl.min_mem_alignment = 512; |
8e3b0cbb | 214 | bs->bl.opt_mem_alignment = qemu_real_host_page_size(); |
bd44feb7 SH |
215 | |
216 | /* Safe default since most protocols use readv()/writev()/etc */ | |
217 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
218 | } |
219 | ||
61007b31 SH |
220 | /* Then let the driver override it */ |
221 | if (drv->bdrv_refresh_limits) { | |
222 | drv->bdrv_refresh_limits(bs, errp); | |
8b117001 VSO |
223 | if (*errp) { |
224 | return; | |
225 | } | |
226 | } | |
227 | ||
228 | if (bs->bl.request_alignment > BDRV_MAX_ALIGNMENT) { | |
229 | error_setg(errp, "Driver requires too large request alignment"); | |
61007b31 SH |
230 | } |
231 | } | |
232 | ||
233 | /** | |
234 | * The copy-on-read flag is actually a reference count so multiple users may | |
235 | * use the feature without worrying about clobbering its previous state. | |
236 | * Copy-on-read stays enabled until all users have called to disable it. | |
237 | */ | |
238 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
239 | { | |
384a48fb | 240 | IO_CODE(); |
d73415a3 | 241 | qatomic_inc(&bs->copy_on_read); |
61007b31 SH |
242 | } |
243 | ||
244 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
245 | { | |
d73415a3 | 246 | int old = qatomic_fetch_dec(&bs->copy_on_read); |
384a48fb | 247 | IO_CODE(); |
d3faa13e | 248 | assert(old >= 1); |
61007b31 SH |
249 | } |
250 | ||
61124f03 PB |
251 | typedef struct { |
252 | Coroutine *co; | |
253 | BlockDriverState *bs; | |
254 | bool done; | |
481cad48 | 255 | bool begin; |
fe4f0614 | 256 | bool poll; |
0152bf40 | 257 | BdrvChild *parent; |
61124f03 PB |
258 | } BdrvCoDrainData; |
259 | ||
1cc8e54a | 260 | /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ |
299403ae KW |
261 | bool bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent, |
262 | bool ignore_bds_parents) | |
89bd0305 | 263 | { |
ab613350 | 264 | GLOBAL_STATE_CODE(); |
fe4f0614 | 265 | |
6cd5c9d7 | 266 | if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) { |
89bd0305 KW |
267 | return true; |
268 | } | |
269 | ||
d73415a3 | 270 | if (qatomic_read(&bs->in_flight)) { |
fe4f0614 KW |
271 | return true; |
272 | } | |
273 | ||
fe4f0614 | 274 | return false; |
89bd0305 KW |
275 | } |
276 | ||
299403ae | 277 | static bool bdrv_drain_poll_top_level(BlockDriverState *bs, |
89bd0305 | 278 | BdrvChild *ignore_parent) |
1cc8e54a | 279 | { |
d05ab380 EGE |
280 | GLOBAL_STATE_CODE(); |
281 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
282 | ||
299403ae | 283 | return bdrv_drain_poll(bs, ignore_parent, false); |
1cc8e54a KW |
284 | } |
285 | ||
299403ae | 286 | static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent, |
a82a3bd1 KW |
287 | bool poll); |
288 | static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent); | |
0152bf40 | 289 | |
a77fd4bb FZ |
290 | static void bdrv_co_drain_bh_cb(void *opaque) |
291 | { | |
292 | BdrvCoDrainData *data = opaque; | |
293 | Coroutine *co = data->co; | |
99723548 | 294 | BlockDriverState *bs = data->bs; |
a77fd4bb | 295 | |
c8ca33d0 | 296 | if (bs) { |
aa1361d5 | 297 | AioContext *ctx = bdrv_get_aio_context(bs); |
960d5fb3 | 298 | aio_context_acquire(ctx); |
c8ca33d0 KW |
299 | bdrv_dec_in_flight(bs); |
300 | if (data->begin) { | |
a82a3bd1 | 301 | bdrv_do_drained_begin(bs, data->parent, data->poll); |
c8ca33d0 | 302 | } else { |
e037c09c | 303 | assert(!data->poll); |
a82a3bd1 | 304 | bdrv_do_drained_end(bs, data->parent); |
c8ca33d0 | 305 | } |
960d5fb3 | 306 | aio_context_release(ctx); |
481cad48 | 307 | } else { |
c8ca33d0 KW |
308 | assert(data->begin); |
309 | bdrv_drain_all_begin(); | |
481cad48 MP |
310 | } |
311 | ||
a77fd4bb | 312 | data->done = true; |
1919631e | 313 | aio_co_wake(co); |
a77fd4bb FZ |
314 | } |
315 | ||
481cad48 | 316 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, |
299403ae | 317 | bool begin, |
6cd5c9d7 | 318 | BdrvChild *parent, |
2f65df6e | 319 | bool poll) |
a77fd4bb FZ |
320 | { |
321 | BdrvCoDrainData data; | |
960d5fb3 KW |
322 | Coroutine *self = qemu_coroutine_self(); |
323 | AioContext *ctx = bdrv_get_aio_context(bs); | |
324 | AioContext *co_ctx = qemu_coroutine_get_aio_context(self); | |
a77fd4bb FZ |
325 | |
326 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
c40a2545 | 327 | * other coroutines run if they were queued by aio_co_enter(). */ |
a77fd4bb FZ |
328 | |
329 | assert(qemu_in_coroutine()); | |
330 | data = (BdrvCoDrainData) { | |
960d5fb3 | 331 | .co = self, |
a77fd4bb FZ |
332 | .bs = bs, |
333 | .done = false, | |
481cad48 | 334 | .begin = begin, |
0152bf40 | 335 | .parent = parent, |
fe4f0614 | 336 | .poll = poll, |
a77fd4bb | 337 | }; |
8e1da77e | 338 | |
c8ca33d0 KW |
339 | if (bs) { |
340 | bdrv_inc_in_flight(bs); | |
341 | } | |
960d5fb3 KW |
342 | |
343 | /* | |
344 | * Temporarily drop the lock across yield or we would get deadlocks. | |
345 | * bdrv_co_drain_bh_cb() reaquires the lock as needed. | |
346 | * | |
347 | * When we yield below, the lock for the current context will be | |
348 | * released, so if this is actually the lock that protects bs, don't drop | |
349 | * it a second time. | |
350 | */ | |
351 | if (ctx != co_ctx) { | |
352 | aio_context_release(ctx); | |
353 | } | |
ab613350 SH |
354 | replay_bh_schedule_oneshot_event(qemu_get_aio_context(), |
355 | bdrv_co_drain_bh_cb, &data); | |
a77fd4bb FZ |
356 | |
357 | qemu_coroutine_yield(); | |
358 | /* If we are resumed from some other event (such as an aio completion or a | |
359 | * timer callback), it is a bug in the caller that should be fixed. */ | |
360 | assert(data.done); | |
960d5fb3 | 361 | |
3202d8e4 | 362 | /* Reacquire the AioContext of bs if we dropped it */ |
960d5fb3 KW |
363 | if (ctx != co_ctx) { |
364 | aio_context_acquire(ctx); | |
365 | } | |
a77fd4bb FZ |
366 | } |
367 | ||
05c272ff KW |
368 | static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent, |
369 | bool poll) | |
6820643f | 370 | { |
384a48fb | 371 | IO_OR_GS_CODE(); |
05c272ff KW |
372 | |
373 | if (qemu_in_coroutine()) { | |
374 | bdrv_co_yield_to_drain(bs, true, parent, poll); | |
375 | return; | |
376 | } | |
d42cf288 | 377 | |
ab613350 SH |
378 | GLOBAL_STATE_CODE(); |
379 | ||
60369b86 | 380 | /* Stop things in parent-to-child order */ |
d73415a3 | 381 | if (qatomic_fetch_inc(&bs->quiesce_counter) == 0) { |
d05ab380 | 382 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
a82a3bd1 | 383 | bdrv_parent_drained_begin(bs, parent); |
57e05be3 KW |
384 | if (bs->drv && bs->drv->bdrv_drain_begin) { |
385 | bs->drv->bdrv_drain_begin(bs); | |
386 | } | |
c7bc05f7 | 387 | } |
d30b8e64 | 388 | |
fe4f0614 KW |
389 | /* |
390 | * Wait for drained requests to finish. | |
391 | * | |
392 | * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The | |
393 | * call is needed so things in this AioContext can make progress even | |
394 | * though we don't return to the main AioContext loop - this automatically | |
395 | * includes other nodes in the same AioContext and therefore all child | |
396 | * nodes. | |
397 | */ | |
398 | if (poll) { | |
299403ae | 399 | BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, parent)); |
fe4f0614 | 400 | } |
6820643f KW |
401 | } |
402 | ||
05c272ff KW |
403 | void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, BdrvChild *parent) |
404 | { | |
405 | bdrv_do_drained_begin(bs, parent, false); | |
406 | } | |
407 | ||
e2dbca03 PB |
408 | void coroutine_mixed_fn |
409 | bdrv_drained_begin(BlockDriverState *bs) | |
0152bf40 | 410 | { |
384a48fb | 411 | IO_OR_GS_CODE(); |
a82a3bd1 | 412 | bdrv_do_drained_begin(bs, NULL, true); |
0152bf40 KW |
413 | } |
414 | ||
e037c09c HR |
415 | /** |
416 | * This function does not poll, nor must any of its recursively called | |
2f65df6e | 417 | * functions. |
e037c09c | 418 | */ |
a82a3bd1 | 419 | static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent) |
6820643f | 420 | { |
0f115168 KW |
421 | int old_quiesce_counter; |
422 | ||
ab613350 SH |
423 | IO_OR_GS_CODE(); |
424 | ||
481cad48 | 425 | if (qemu_in_coroutine()) { |
a82a3bd1 | 426 | bdrv_co_yield_to_drain(bs, false, parent, false); |
481cad48 MP |
427 | return; |
428 | } | |
d05ab380 EGE |
429 | |
430 | /* At this point, we should be always running in the main loop. */ | |
431 | GLOBAL_STATE_CODE(); | |
6820643f | 432 | assert(bs->quiesce_counter > 0); |
ab613350 | 433 | GLOBAL_STATE_CODE(); |
6820643f | 434 | |
60369b86 | 435 | /* Re-enable things in child-to-parent order */ |
d73415a3 | 436 | old_quiesce_counter = qatomic_fetch_dec(&bs->quiesce_counter); |
0f115168 | 437 | if (old_quiesce_counter == 1) { |
d05ab380 | 438 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
57e05be3 KW |
439 | if (bs->drv && bs->drv->bdrv_drain_end) { |
440 | bs->drv->bdrv_drain_end(bs); | |
441 | } | |
a82a3bd1 | 442 | bdrv_parent_drained_end(bs, parent); |
0f115168 | 443 | } |
6820643f KW |
444 | } |
445 | ||
0152bf40 KW |
446 | void bdrv_drained_end(BlockDriverState *bs) |
447 | { | |
384a48fb | 448 | IO_OR_GS_CODE(); |
a82a3bd1 | 449 | bdrv_do_drained_end(bs, NULL); |
d736f119 KW |
450 | } |
451 | ||
b6e84c97 PB |
452 | void bdrv_drain(BlockDriverState *bs) |
453 | { | |
384a48fb | 454 | IO_OR_GS_CODE(); |
6820643f KW |
455 | bdrv_drained_begin(bs); |
456 | bdrv_drained_end(bs); | |
61007b31 SH |
457 | } |
458 | ||
c13ad59f KW |
459 | static void bdrv_drain_assert_idle(BlockDriverState *bs) |
460 | { | |
461 | BdrvChild *child, *next; | |
d05ab380 EGE |
462 | GLOBAL_STATE_CODE(); |
463 | GRAPH_RDLOCK_GUARD_MAINLOOP(); | |
c13ad59f | 464 | |
d73415a3 | 465 | assert(qatomic_read(&bs->in_flight) == 0); |
c13ad59f KW |
466 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
467 | bdrv_drain_assert_idle(child->bs); | |
468 | } | |
469 | } | |
470 | ||
0f12264e KW |
471 | unsigned int bdrv_drain_all_count = 0; |
472 | ||
473 | static bool bdrv_drain_all_poll(void) | |
474 | { | |
475 | BlockDriverState *bs = NULL; | |
476 | bool result = false; | |
d05ab380 | 477 | |
f791bf7f | 478 | GLOBAL_STATE_CODE(); |
d05ab380 | 479 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
0f12264e | 480 | |
0f12264e KW |
481 | /* bdrv_drain_poll() can't make changes to the graph and we are holding the |
482 | * main AioContext lock, so iterating bdrv_next_all_states() is safe. */ | |
483 | while ((bs = bdrv_next_all_states(bs))) { | |
484 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
485 | aio_context_acquire(aio_context); | |
299403ae | 486 | result |= bdrv_drain_poll(bs, NULL, true); |
0f12264e KW |
487 | aio_context_release(aio_context); |
488 | } | |
489 | ||
490 | return result; | |
491 | } | |
492 | ||
61007b31 SH |
493 | /* |
494 | * Wait for pending requests to complete across all BlockDriverStates | |
495 | * | |
496 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
497 | * after calling this function. | |
c0778f66 AG |
498 | * |
499 | * This pauses all block jobs and disables external clients. It must | |
500 | * be paired with bdrv_drain_all_end(). | |
501 | * | |
502 | * NOTE: no new block jobs or BlockDriverStates can be created between | |
503 | * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls. | |
61007b31 | 504 | */ |
da0bd744 | 505 | void bdrv_drain_all_begin_nopoll(void) |
61007b31 | 506 | { |
0f12264e | 507 | BlockDriverState *bs = NULL; |
f791bf7f | 508 | GLOBAL_STATE_CODE(); |
61007b31 | 509 | |
c8aa7895 PD |
510 | /* |
511 | * bdrv queue is managed by record/replay, | |
512 | * waiting for finishing the I/O requests may | |
513 | * be infinite | |
514 | */ | |
515 | if (replay_events_enabled()) { | |
516 | return; | |
517 | } | |
518 | ||
0f12264e KW |
519 | /* AIO_WAIT_WHILE() with a NULL context can only be called from the main |
520 | * loop AioContext, so make sure we're in the main context. */ | |
9a7e86c8 | 521 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
522 | assert(bdrv_drain_all_count < INT_MAX); |
523 | bdrv_drain_all_count++; | |
9a7e86c8 | 524 | |
0f12264e KW |
525 | /* Quiesce all nodes, without polling in-flight requests yet. The graph |
526 | * cannot change during this loop. */ | |
527 | while ((bs = bdrv_next_all_states(bs))) { | |
61007b31 SH |
528 | AioContext *aio_context = bdrv_get_aio_context(bs); |
529 | ||
530 | aio_context_acquire(aio_context); | |
a82a3bd1 | 531 | bdrv_do_drained_begin(bs, NULL, false); |
61007b31 SH |
532 | aio_context_release(aio_context); |
533 | } | |
da0bd744 KW |
534 | } |
535 | ||
e2dbca03 | 536 | void coroutine_mixed_fn bdrv_drain_all_begin(void) |
da0bd744 KW |
537 | { |
538 | BlockDriverState *bs = NULL; | |
539 | ||
540 | if (qemu_in_coroutine()) { | |
541 | bdrv_co_yield_to_drain(NULL, true, NULL, true); | |
542 | return; | |
543 | } | |
544 | ||
63945789 PM |
545 | /* |
546 | * bdrv queue is managed by record/replay, | |
547 | * waiting for finishing the I/O requests may | |
548 | * be infinite | |
549 | */ | |
550 | if (replay_events_enabled()) { | |
551 | return; | |
552 | } | |
553 | ||
da0bd744 | 554 | bdrv_drain_all_begin_nopoll(); |
61007b31 | 555 | |
0f12264e | 556 | /* Now poll the in-flight requests */ |
263d5e12 | 557 | AIO_WAIT_WHILE_UNLOCKED(NULL, bdrv_drain_all_poll()); |
0f12264e KW |
558 | |
559 | while ((bs = bdrv_next_all_states(bs))) { | |
c13ad59f | 560 | bdrv_drain_assert_idle(bs); |
61007b31 | 561 | } |
c0778f66 AG |
562 | } |
563 | ||
1a6d3bd2 GK |
564 | void bdrv_drain_all_end_quiesce(BlockDriverState *bs) |
565 | { | |
b4ad82aa | 566 | GLOBAL_STATE_CODE(); |
1a6d3bd2 GK |
567 | |
568 | g_assert(bs->quiesce_counter > 0); | |
569 | g_assert(!bs->refcnt); | |
570 | ||
571 | while (bs->quiesce_counter) { | |
a82a3bd1 | 572 | bdrv_do_drained_end(bs, NULL); |
1a6d3bd2 | 573 | } |
1a6d3bd2 GK |
574 | } |
575 | ||
c0778f66 AG |
576 | void bdrv_drain_all_end(void) |
577 | { | |
0f12264e | 578 | BlockDriverState *bs = NULL; |
f791bf7f | 579 | GLOBAL_STATE_CODE(); |
c0778f66 | 580 | |
c8aa7895 PD |
581 | /* |
582 | * bdrv queue is managed by record/replay, | |
583 | * waiting for finishing the I/O requests may | |
584 | * be endless | |
585 | */ | |
586 | if (replay_events_enabled()) { | |
587 | return; | |
588 | } | |
589 | ||
0f12264e | 590 | while ((bs = bdrv_next_all_states(bs))) { |
61007b31 SH |
591 | AioContext *aio_context = bdrv_get_aio_context(bs); |
592 | ||
593 | aio_context_acquire(aio_context); | |
a82a3bd1 | 594 | bdrv_do_drained_end(bs, NULL); |
61007b31 SH |
595 | aio_context_release(aio_context); |
596 | } | |
0f12264e | 597 | |
e037c09c | 598 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
599 | assert(bdrv_drain_all_count > 0); |
600 | bdrv_drain_all_count--; | |
61007b31 SH |
601 | } |
602 | ||
c0778f66 AG |
603 | void bdrv_drain_all(void) |
604 | { | |
f791bf7f | 605 | GLOBAL_STATE_CODE(); |
c0778f66 AG |
606 | bdrv_drain_all_begin(); |
607 | bdrv_drain_all_end(); | |
608 | } | |
609 | ||
61007b31 SH |
610 | /** |
611 | * Remove an active request from the tracked requests list | |
612 | * | |
613 | * This function should be called when a tracked request is completing. | |
614 | */ | |
f0d43b1e | 615 | static void coroutine_fn tracked_request_end(BdrvTrackedRequest *req) |
61007b31 SH |
616 | { |
617 | if (req->serialising) { | |
d73415a3 | 618 | qatomic_dec(&req->bs->serialising_in_flight); |
61007b31 SH |
619 | } |
620 | ||
fa9185fc | 621 | qemu_mutex_lock(&req->bs->reqs_lock); |
61007b31 | 622 | QLIST_REMOVE(req, list); |
fa9185fc | 623 | qemu_mutex_unlock(&req->bs->reqs_lock); |
3480ce69 SH |
624 | |
625 | /* | |
626 | * At this point qemu_co_queue_wait(&req->wait_queue, ...) won't be called | |
627 | * anymore because the request has been removed from the list, so it's safe | |
628 | * to restart the queue outside reqs_lock to minimize the critical section. | |
629 | */ | |
61007b31 SH |
630 | qemu_co_queue_restart_all(&req->wait_queue); |
631 | } | |
632 | ||
633 | /** | |
634 | * Add an active request to the tracked requests list | |
635 | */ | |
881a4c55 PB |
636 | static void coroutine_fn tracked_request_begin(BdrvTrackedRequest *req, |
637 | BlockDriverState *bs, | |
638 | int64_t offset, | |
639 | int64_t bytes, | |
640 | enum BdrvTrackedRequestType type) | |
61007b31 | 641 | { |
80247264 | 642 | bdrv_check_request(offset, bytes, &error_abort); |
22931a15 | 643 | |
61007b31 SH |
644 | *req = (BdrvTrackedRequest){ |
645 | .bs = bs, | |
646 | .offset = offset, | |
647 | .bytes = bytes, | |
ebde595c | 648 | .type = type, |
61007b31 SH |
649 | .co = qemu_coroutine_self(), |
650 | .serialising = false, | |
651 | .overlap_offset = offset, | |
652 | .overlap_bytes = bytes, | |
653 | }; | |
654 | ||
655 | qemu_co_queue_init(&req->wait_queue); | |
656 | ||
fa9185fc | 657 | qemu_mutex_lock(&bs->reqs_lock); |
61007b31 | 658 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
fa9185fc | 659 | qemu_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
660 | } |
661 | ||
3ba0e1a0 | 662 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, |
80247264 | 663 | int64_t offset, int64_t bytes) |
3ba0e1a0 | 664 | { |
80247264 EB |
665 | bdrv_check_request(offset, bytes, &error_abort); |
666 | ||
3ba0e1a0 PB |
667 | /* aaaa bbbb */ |
668 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
669 | return false; | |
670 | } | |
671 | /* bbbb aaaa */ | |
672 | if (req->overlap_offset >= offset + bytes) { | |
673 | return false; | |
674 | } | |
675 | return true; | |
676 | } | |
677 | ||
3183937f | 678 | /* Called with self->bs->reqs_lock held */ |
881a4c55 | 679 | static coroutine_fn BdrvTrackedRequest * |
3183937f VSO |
680 | bdrv_find_conflicting_request(BdrvTrackedRequest *self) |
681 | { | |
682 | BdrvTrackedRequest *req; | |
683 | ||
684 | QLIST_FOREACH(req, &self->bs->tracked_requests, list) { | |
685 | if (req == self || (!req->serialising && !self->serialising)) { | |
686 | continue; | |
687 | } | |
688 | if (tracked_request_overlaps(req, self->overlap_offset, | |
689 | self->overlap_bytes)) | |
690 | { | |
691 | /* | |
692 | * Hitting this means there was a reentrant request, for | |
693 | * example, a block driver issuing nested requests. This must | |
694 | * never happen since it means deadlock. | |
695 | */ | |
696 | assert(qemu_coroutine_self() != req->co); | |
697 | ||
698 | /* | |
699 | * If the request is already (indirectly) waiting for us, or | |
700 | * will wait for us as soon as it wakes up, then just go on | |
701 | * (instead of producing a deadlock in the former case). | |
702 | */ | |
703 | if (!req->waiting_for) { | |
704 | return req; | |
705 | } | |
706 | } | |
707 | } | |
708 | ||
709 | return NULL; | |
710 | } | |
711 | ||
ec1c8868 | 712 | /* Called with self->bs->reqs_lock held */ |
131498f7 | 713 | static void coroutine_fn |
ec1c8868 | 714 | bdrv_wait_serialising_requests_locked(BdrvTrackedRequest *self) |
3ba0e1a0 PB |
715 | { |
716 | BdrvTrackedRequest *req; | |
3ba0e1a0 | 717 | |
3183937f VSO |
718 | while ((req = bdrv_find_conflicting_request(self))) { |
719 | self->waiting_for = req; | |
ec1c8868 | 720 | qemu_co_queue_wait(&req->wait_queue, &self->bs->reqs_lock); |
3183937f | 721 | self->waiting_for = NULL; |
3183937f | 722 | } |
3ba0e1a0 PB |
723 | } |
724 | ||
8ac5aab2 VSO |
725 | /* Called with req->bs->reqs_lock held */ |
726 | static void tracked_request_set_serialising(BdrvTrackedRequest *req, | |
727 | uint64_t align) | |
61007b31 SH |
728 | { |
729 | int64_t overlap_offset = req->offset & ~(align - 1); | |
80247264 EB |
730 | int64_t overlap_bytes = |
731 | ROUND_UP(req->offset + req->bytes, align) - overlap_offset; | |
732 | ||
733 | bdrv_check_request(req->offset, req->bytes, &error_abort); | |
61007b31 SH |
734 | |
735 | if (!req->serialising) { | |
d73415a3 | 736 | qatomic_inc(&req->bs->serialising_in_flight); |
61007b31 SH |
737 | req->serialising = true; |
738 | } | |
739 | ||
740 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
741 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
09d2f948 VSO |
742 | } |
743 | ||
c28107e9 HR |
744 | /** |
745 | * Return the tracked request on @bs for the current coroutine, or | |
746 | * NULL if there is none. | |
747 | */ | |
748 | BdrvTrackedRequest *coroutine_fn bdrv_co_get_self_request(BlockDriverState *bs) | |
749 | { | |
750 | BdrvTrackedRequest *req; | |
751 | Coroutine *self = qemu_coroutine_self(); | |
967d7905 | 752 | IO_CODE(); |
c28107e9 HR |
753 | |
754 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
755 | if (req->co == self) { | |
756 | return req; | |
757 | } | |
758 | } | |
759 | ||
760 | return NULL; | |
761 | } | |
762 | ||
244483e6 | 763 | /** |
fc6b211f | 764 | * Round a region to subcluster (if supported) or cluster boundaries |
244483e6 | 765 | */ |
a00e70c0 | 766 | void coroutine_fn GRAPH_RDLOCK |
fc6b211f AD |
767 | bdrv_round_to_subclusters(BlockDriverState *bs, int64_t offset, int64_t bytes, |
768 | int64_t *align_offset, int64_t *align_bytes) | |
244483e6 KW |
769 | { |
770 | BlockDriverInfo bdi; | |
384a48fb | 771 | IO_CODE(); |
fc6b211f AD |
772 | if (bdrv_co_get_info(bs, &bdi) < 0 || bdi.subcluster_size == 0) { |
773 | *align_offset = offset; | |
774 | *align_bytes = bytes; | |
244483e6 | 775 | } else { |
fc6b211f AD |
776 | int64_t c = bdi.subcluster_size; |
777 | *align_offset = QEMU_ALIGN_DOWN(offset, c); | |
778 | *align_bytes = QEMU_ALIGN_UP(offset - *align_offset + bytes, c); | |
244483e6 KW |
779 | } |
780 | } | |
781 | ||
a00e70c0 | 782 | static int coroutine_fn GRAPH_RDLOCK bdrv_get_cluster_size(BlockDriverState *bs) |
61007b31 SH |
783 | { |
784 | BlockDriverInfo bdi; | |
785 | int ret; | |
786 | ||
3d47eb0a | 787 | ret = bdrv_co_get_info(bs, &bdi); |
61007b31 | 788 | if (ret < 0 || bdi.cluster_size == 0) { |
a5b8dd2c | 789 | return bs->bl.request_alignment; |
61007b31 SH |
790 | } else { |
791 | return bdi.cluster_size; | |
792 | } | |
793 | } | |
794 | ||
99723548 PB |
795 | void bdrv_inc_in_flight(BlockDriverState *bs) |
796 | { | |
967d7905 | 797 | IO_CODE(); |
d73415a3 | 798 | qatomic_inc(&bs->in_flight); |
99723548 PB |
799 | } |
800 | ||
c9d1a561 PB |
801 | void bdrv_wakeup(BlockDriverState *bs) |
802 | { | |
967d7905 | 803 | IO_CODE(); |
cfe29d82 | 804 | aio_wait_kick(); |
c9d1a561 PB |
805 | } |
806 | ||
99723548 PB |
807 | void bdrv_dec_in_flight(BlockDriverState *bs) |
808 | { | |
967d7905 | 809 | IO_CODE(); |
d73415a3 | 810 | qatomic_dec(&bs->in_flight); |
c9d1a561 | 811 | bdrv_wakeup(bs); |
99723548 PB |
812 | } |
813 | ||
131498f7 DL |
814 | static void coroutine_fn |
815 | bdrv_wait_serialising_requests(BdrvTrackedRequest *self) | |
61007b31 SH |
816 | { |
817 | BlockDriverState *bs = self->bs; | |
61007b31 | 818 | |
d73415a3 | 819 | if (!qatomic_read(&bs->serialising_in_flight)) { |
131498f7 | 820 | return; |
61007b31 SH |
821 | } |
822 | ||
fa9185fc | 823 | qemu_mutex_lock(&bs->reqs_lock); |
131498f7 | 824 | bdrv_wait_serialising_requests_locked(self); |
fa9185fc | 825 | qemu_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
826 | } |
827 | ||
131498f7 | 828 | void coroutine_fn bdrv_make_request_serialising(BdrvTrackedRequest *req, |
8ac5aab2 VSO |
829 | uint64_t align) |
830 | { | |
967d7905 | 831 | IO_CODE(); |
8ac5aab2 | 832 | |
fa9185fc | 833 | qemu_mutex_lock(&req->bs->reqs_lock); |
8ac5aab2 VSO |
834 | |
835 | tracked_request_set_serialising(req, align); | |
131498f7 | 836 | bdrv_wait_serialising_requests_locked(req); |
8ac5aab2 | 837 | |
fa9185fc | 838 | qemu_mutex_unlock(&req->bs->reqs_lock); |
8ac5aab2 VSO |
839 | } |
840 | ||
558902cc VSO |
841 | int bdrv_check_qiov_request(int64_t offset, int64_t bytes, |
842 | QEMUIOVector *qiov, size_t qiov_offset, | |
843 | Error **errp) | |
61007b31 | 844 | { |
63f4ad11 VSO |
845 | /* |
846 | * Check generic offset/bytes correctness | |
847 | */ | |
848 | ||
69b55e03 VSO |
849 | if (offset < 0) { |
850 | error_setg(errp, "offset is negative: %" PRIi64, offset); | |
851 | return -EIO; | |
852 | } | |
853 | ||
854 | if (bytes < 0) { | |
855 | error_setg(errp, "bytes is negative: %" PRIi64, bytes); | |
61007b31 SH |
856 | return -EIO; |
857 | } | |
858 | ||
8b117001 | 859 | if (bytes > BDRV_MAX_LENGTH) { |
69b55e03 VSO |
860 | error_setg(errp, "bytes(%" PRIi64 ") exceeds maximum(%" PRIi64 ")", |
861 | bytes, BDRV_MAX_LENGTH); | |
862 | return -EIO; | |
863 | } | |
864 | ||
865 | if (offset > BDRV_MAX_LENGTH) { | |
866 | error_setg(errp, "offset(%" PRIi64 ") exceeds maximum(%" PRIi64 ")", | |
867 | offset, BDRV_MAX_LENGTH); | |
8b117001 VSO |
868 | return -EIO; |
869 | } | |
870 | ||
871 | if (offset > BDRV_MAX_LENGTH - bytes) { | |
69b55e03 VSO |
872 | error_setg(errp, "sum of offset(%" PRIi64 ") and bytes(%" PRIi64 ") " |
873 | "exceeds maximum(%" PRIi64 ")", offset, bytes, | |
874 | BDRV_MAX_LENGTH); | |
8b117001 VSO |
875 | return -EIO; |
876 | } | |
877 | ||
63f4ad11 VSO |
878 | if (!qiov) { |
879 | return 0; | |
880 | } | |
881 | ||
882 | /* | |
883 | * Check qiov and qiov_offset | |
884 | */ | |
885 | ||
886 | if (qiov_offset > qiov->size) { | |
887 | error_setg(errp, "qiov_offset(%zu) overflow io vector size(%zu)", | |
888 | qiov_offset, qiov->size); | |
889 | return -EIO; | |
890 | } | |
891 | ||
892 | if (bytes > qiov->size - qiov_offset) { | |
893 | error_setg(errp, "bytes(%" PRIi64 ") + qiov_offset(%zu) overflow io " | |
894 | "vector size(%zu)", bytes, qiov_offset, qiov->size); | |
895 | return -EIO; | |
896 | } | |
897 | ||
8b117001 VSO |
898 | return 0; |
899 | } | |
900 | ||
63f4ad11 VSO |
901 | int bdrv_check_request(int64_t offset, int64_t bytes, Error **errp) |
902 | { | |
903 | return bdrv_check_qiov_request(offset, bytes, NULL, 0, errp); | |
904 | } | |
905 | ||
906 | static int bdrv_check_request32(int64_t offset, int64_t bytes, | |
907 | QEMUIOVector *qiov, size_t qiov_offset) | |
8b117001 | 908 | { |
63f4ad11 | 909 | int ret = bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, NULL); |
8b117001 VSO |
910 | if (ret < 0) { |
911 | return ret; | |
912 | } | |
913 | ||
914 | if (bytes > BDRV_REQUEST_MAX_BYTES) { | |
61007b31 SH |
915 | return -EIO; |
916 | } | |
917 | ||
918 | return 0; | |
919 | } | |
920 | ||
61007b31 | 921 | /* |
74021bc4 | 922 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
923 | * The operation is sped up by checking the block status and only writing |
924 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 925 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 926 | * BDRV_REQ_FUA). |
61007b31 | 927 | * |
f4649069 | 928 | * Returns < 0 on error, 0 on success. For error codes see bdrv_pwrite(). |
61007b31 | 929 | */ |
720ff280 | 930 | int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags) |
61007b31 | 931 | { |
237d78f8 EB |
932 | int ret; |
933 | int64_t target_size, bytes, offset = 0; | |
720ff280 | 934 | BlockDriverState *bs = child->bs; |
384a48fb | 935 | IO_CODE(); |
61007b31 | 936 | |
7286d610 EB |
937 | target_size = bdrv_getlength(bs); |
938 | if (target_size < 0) { | |
939 | return target_size; | |
61007b31 SH |
940 | } |
941 | ||
942 | for (;;) { | |
7286d610 EB |
943 | bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES); |
944 | if (bytes <= 0) { | |
61007b31 SH |
945 | return 0; |
946 | } | |
237d78f8 | 947 | ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL); |
61007b31 | 948 | if (ret < 0) { |
61007b31 SH |
949 | return ret; |
950 | } | |
951 | if (ret & BDRV_BLOCK_ZERO) { | |
237d78f8 | 952 | offset += bytes; |
61007b31 SH |
953 | continue; |
954 | } | |
237d78f8 | 955 | ret = bdrv_pwrite_zeroes(child, offset, bytes, flags); |
61007b31 | 956 | if (ret < 0) { |
61007b31 SH |
957 | return ret; |
958 | } | |
237d78f8 | 959 | offset += bytes; |
61007b31 SH |
960 | } |
961 | } | |
962 | ||
61007b31 SH |
963 | /* |
964 | * Writes to the file and ensures that no writes are reordered across this | |
965 | * request (acts as a barrier) | |
966 | * | |
967 | * Returns 0 on success, -errno in error cases. | |
968 | */ | |
e97190a4 AF |
969 | int coroutine_fn bdrv_co_pwrite_sync(BdrvChild *child, int64_t offset, |
970 | int64_t bytes, const void *buf, | |
971 | BdrvRequestFlags flags) | |
61007b31 SH |
972 | { |
973 | int ret; | |
384a48fb | 974 | IO_CODE(); |
b24a4c41 | 975 | assert_bdrv_graph_readable(); |
88095349 | 976 | |
e97190a4 | 977 | ret = bdrv_co_pwrite(child, offset, bytes, buf, flags); |
61007b31 SH |
978 | if (ret < 0) { |
979 | return ret; | |
980 | } | |
981 | ||
e97190a4 | 982 | ret = bdrv_co_flush(child->bs); |
855a6a93 KW |
983 | if (ret < 0) { |
984 | return ret; | |
61007b31 SH |
985 | } |
986 | ||
987 | return 0; | |
988 | } | |
989 | ||
08844473 KW |
990 | typedef struct CoroutineIOCompletion { |
991 | Coroutine *coroutine; | |
992 | int ret; | |
993 | } CoroutineIOCompletion; | |
994 | ||
995 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
996 | { | |
997 | CoroutineIOCompletion *co = opaque; | |
998 | ||
999 | co->ret = ret; | |
b9e413dd | 1000 | aio_co_wake(co->coroutine); |
08844473 KW |
1001 | } |
1002 | ||
7b1fb72e KW |
1003 | static int coroutine_fn GRAPH_RDLOCK |
1004 | bdrv_driver_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1005 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
166fe960 KW |
1006 | { |
1007 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
1008 | int64_t sector_num; |
1009 | unsigned int nb_sectors; | |
ac850bf0 VSO |
1010 | QEMUIOVector local_qiov; |
1011 | int ret; | |
b9b10c35 | 1012 | assert_bdrv_graph_readable(); |
3fb06697 | 1013 | |
17abcbee | 1014 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
e8b65355 | 1015 | assert(!(flags & ~bs->supported_read_flags)); |
fa166538 | 1016 | |
d470ad42 HR |
1017 | if (!drv) { |
1018 | return -ENOMEDIUM; | |
1019 | } | |
1020 | ||
ac850bf0 VSO |
1021 | if (drv->bdrv_co_preadv_part) { |
1022 | return drv->bdrv_co_preadv_part(bs, offset, bytes, qiov, qiov_offset, | |
1023 | flags); | |
1024 | } | |
1025 | ||
1026 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1027 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1028 | qiov = &local_qiov; | |
1029 | } | |
1030 | ||
3fb06697 | 1031 | if (drv->bdrv_co_preadv) { |
ac850bf0 VSO |
1032 | ret = drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); |
1033 | goto out; | |
3fb06697 KW |
1034 | } |
1035 | ||
edfab6a0 | 1036 | if (drv->bdrv_aio_preadv) { |
08844473 KW |
1037 | BlockAIOCB *acb; |
1038 | CoroutineIOCompletion co = { | |
1039 | .coroutine = qemu_coroutine_self(), | |
1040 | }; | |
1041 | ||
edfab6a0 EB |
1042 | acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags, |
1043 | bdrv_co_io_em_complete, &co); | |
08844473 | 1044 | if (acb == NULL) { |
ac850bf0 VSO |
1045 | ret = -EIO; |
1046 | goto out; | |
08844473 KW |
1047 | } else { |
1048 | qemu_coroutine_yield(); | |
ac850bf0 VSO |
1049 | ret = co.ret; |
1050 | goto out; | |
08844473 KW |
1051 | } |
1052 | } | |
edfab6a0 EB |
1053 | |
1054 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1055 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1056 | ||
1bbbf32d NS |
1057 | assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)); |
1058 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
41ae31e3 | 1059 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 EB |
1060 | assert(drv->bdrv_co_readv); |
1061 | ||
ac850bf0 VSO |
1062 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); |
1063 | ||
1064 | out: | |
1065 | if (qiov == &local_qiov) { | |
1066 | qemu_iovec_destroy(&local_qiov); | |
1067 | } | |
1068 | ||
1069 | return ret; | |
166fe960 KW |
1070 | } |
1071 | ||
7b1fb72e KW |
1072 | static int coroutine_fn GRAPH_RDLOCK |
1073 | bdrv_driver_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1074 | QEMUIOVector *qiov, size_t qiov_offset, | |
1075 | BdrvRequestFlags flags) | |
78a07294 KW |
1076 | { |
1077 | BlockDriver *drv = bs->drv; | |
e8b65355 | 1078 | bool emulate_fua = false; |
3fb06697 KW |
1079 | int64_t sector_num; |
1080 | unsigned int nb_sectors; | |
ac850bf0 | 1081 | QEMUIOVector local_qiov; |
78a07294 | 1082 | int ret; |
b9b10c35 | 1083 | assert_bdrv_graph_readable(); |
78a07294 | 1084 | |
17abcbee | 1085 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
fa166538 | 1086 | |
d470ad42 HR |
1087 | if (!drv) { |
1088 | return -ENOMEDIUM; | |
1089 | } | |
1090 | ||
e8b65355 SH |
1091 | if ((flags & BDRV_REQ_FUA) && |
1092 | (~bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1093 | flags &= ~BDRV_REQ_FUA; | |
1094 | emulate_fua = true; | |
1095 | } | |
1096 | ||
1097 | flags &= bs->supported_write_flags; | |
1098 | ||
ac850bf0 VSO |
1099 | if (drv->bdrv_co_pwritev_part) { |
1100 | ret = drv->bdrv_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, | |
e8b65355 | 1101 | flags); |
ac850bf0 VSO |
1102 | goto emulate_flags; |
1103 | } | |
1104 | ||
1105 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1106 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1107 | qiov = &local_qiov; | |
1108 | } | |
1109 | ||
3fb06697 | 1110 | if (drv->bdrv_co_pwritev) { |
e8b65355 | 1111 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, flags); |
3fb06697 KW |
1112 | goto emulate_flags; |
1113 | } | |
1114 | ||
edfab6a0 | 1115 | if (drv->bdrv_aio_pwritev) { |
08844473 KW |
1116 | BlockAIOCB *acb; |
1117 | CoroutineIOCompletion co = { | |
1118 | .coroutine = qemu_coroutine_self(), | |
1119 | }; | |
1120 | ||
e8b65355 | 1121 | acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov, flags, |
edfab6a0 | 1122 | bdrv_co_io_em_complete, &co); |
08844473 | 1123 | if (acb == NULL) { |
3fb06697 | 1124 | ret = -EIO; |
08844473 KW |
1125 | } else { |
1126 | qemu_coroutine_yield(); | |
3fb06697 | 1127 | ret = co.ret; |
08844473 | 1128 | } |
edfab6a0 EB |
1129 | goto emulate_flags; |
1130 | } | |
1131 | ||
1132 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1133 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1134 | ||
1bbbf32d NS |
1135 | assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)); |
1136 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
41ae31e3 | 1137 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 | 1138 | |
e18a58b4 | 1139 | assert(drv->bdrv_co_writev); |
e8b65355 | 1140 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov, flags); |
78a07294 | 1141 | |
3fb06697 | 1142 | emulate_flags: |
e8b65355 | 1143 | if (ret == 0 && emulate_fua) { |
78a07294 KW |
1144 | ret = bdrv_co_flush(bs); |
1145 | } | |
1146 | ||
ac850bf0 VSO |
1147 | if (qiov == &local_qiov) { |
1148 | qemu_iovec_destroy(&local_qiov); | |
1149 | } | |
1150 | ||
78a07294 KW |
1151 | return ret; |
1152 | } | |
1153 | ||
7b1fb72e | 1154 | static int coroutine_fn GRAPH_RDLOCK |
17abcbee VSO |
1155 | bdrv_driver_pwritev_compressed(BlockDriverState *bs, int64_t offset, |
1156 | int64_t bytes, QEMUIOVector *qiov, | |
ac850bf0 | 1157 | size_t qiov_offset) |
29a298af PB |
1158 | { |
1159 | BlockDriver *drv = bs->drv; | |
ac850bf0 VSO |
1160 | QEMUIOVector local_qiov; |
1161 | int ret; | |
b9b10c35 | 1162 | assert_bdrv_graph_readable(); |
29a298af | 1163 | |
17abcbee VSO |
1164 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1165 | ||
d470ad42 HR |
1166 | if (!drv) { |
1167 | return -ENOMEDIUM; | |
1168 | } | |
1169 | ||
ac850bf0 | 1170 | if (!block_driver_can_compress(drv)) { |
29a298af PB |
1171 | return -ENOTSUP; |
1172 | } | |
1173 | ||
ac850bf0 VSO |
1174 | if (drv->bdrv_co_pwritev_compressed_part) { |
1175 | return drv->bdrv_co_pwritev_compressed_part(bs, offset, bytes, | |
1176 | qiov, qiov_offset); | |
1177 | } | |
1178 | ||
1179 | if (qiov_offset == 0) { | |
1180 | return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov); | |
1181 | } | |
1182 | ||
1183 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1184 | ret = drv->bdrv_co_pwritev_compressed(bs, offset, bytes, &local_qiov); | |
1185 | qemu_iovec_destroy(&local_qiov); | |
1186 | ||
1187 | return ret; | |
29a298af PB |
1188 | } |
1189 | ||
7b1fb72e KW |
1190 | static int coroutine_fn GRAPH_RDLOCK |
1191 | bdrv_co_do_copy_on_readv(BdrvChild *child, int64_t offset, int64_t bytes, | |
1192 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
61007b31 | 1193 | { |
85c97ca7 KW |
1194 | BlockDriverState *bs = child->bs; |
1195 | ||
61007b31 SH |
1196 | /* Perform I/O through a temporary buffer so that users who scribble over |
1197 | * their read buffer while the operation is in progress do not end up | |
1198 | * modifying the image file. This is critical for zero-copy guest I/O | |
1199 | * where anything might happen inside guest memory. | |
1200 | */ | |
2275cc90 | 1201 | void *bounce_buffer = NULL; |
61007b31 SH |
1202 | |
1203 | BlockDriver *drv = bs->drv; | |
fc6b211f AD |
1204 | int64_t align_offset; |
1205 | int64_t align_bytes; | |
9df5afbd | 1206 | int64_t skip_bytes; |
61007b31 | 1207 | int ret; |
cb2e2878 EB |
1208 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
1209 | BDRV_REQUEST_MAX_BYTES); | |
9df5afbd | 1210 | int64_t progress = 0; |
8644476e | 1211 | bool skip_write; |
61007b31 | 1212 | |
9df5afbd VSO |
1213 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1214 | ||
d470ad42 HR |
1215 | if (!drv) { |
1216 | return -ENOMEDIUM; | |
1217 | } | |
1218 | ||
8644476e HR |
1219 | /* |
1220 | * Do not write anything when the BDS is inactive. That is not | |
1221 | * allowed, and it would not help. | |
1222 | */ | |
1223 | skip_write = (bs->open_flags & BDRV_O_INACTIVE); | |
1224 | ||
1bf03e66 KW |
1225 | /* FIXME We cannot require callers to have write permissions when all they |
1226 | * are doing is a read request. If we did things right, write permissions | |
1227 | * would be obtained anyway, but internally by the copy-on-read code. As | |
765d9df9 | 1228 | * long as it is implemented here rather than in a separate filter driver, |
1bf03e66 KW |
1229 | * the copy-on-read code doesn't have its own BdrvChild, however, for which |
1230 | * it could request permissions. Therefore we have to bypass the permission | |
1231 | * system for the moment. */ | |
1232 | // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
afa4b293 | 1233 | |
61007b31 | 1234 | /* Cover entire cluster so no additional backing file I/O is required when |
cb2e2878 EB |
1235 | * allocating cluster in the image file. Note that this value may exceed |
1236 | * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which | |
1237 | * is one reason we loop rather than doing it all at once. | |
61007b31 | 1238 | */ |
fc6b211f AD |
1239 | bdrv_round_to_subclusters(bs, offset, bytes, &align_offset, &align_bytes); |
1240 | skip_bytes = offset - align_offset; | |
61007b31 | 1241 | |
244483e6 | 1242 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
fc6b211f | 1243 | align_offset, align_bytes); |
61007b31 | 1244 | |
fc6b211f | 1245 | while (align_bytes) { |
cb2e2878 | 1246 | int64_t pnum; |
61007b31 | 1247 | |
8644476e HR |
1248 | if (skip_write) { |
1249 | ret = 1; /* "already allocated", so nothing will be copied */ | |
fc6b211f | 1250 | pnum = MIN(align_bytes, max_transfer); |
8644476e | 1251 | } else { |
cc323997 PB |
1252 | ret = bdrv_co_is_allocated(bs, align_offset, |
1253 | MIN(align_bytes, max_transfer), &pnum); | |
8644476e HR |
1254 | if (ret < 0) { |
1255 | /* | |
1256 | * Safe to treat errors in querying allocation as if | |
1257 | * unallocated; we'll probably fail again soon on the | |
1258 | * read, but at least that will set a decent errno. | |
1259 | */ | |
fc6b211f | 1260 | pnum = MIN(align_bytes, max_transfer); |
8644476e | 1261 | } |
61007b31 | 1262 | |
8644476e HR |
1263 | /* Stop at EOF if the image ends in the middle of the cluster */ |
1264 | if (ret == 0 && pnum == 0) { | |
1265 | assert(progress >= bytes); | |
1266 | break; | |
1267 | } | |
b0ddcbbb | 1268 | |
8644476e HR |
1269 | assert(skip_bytes < pnum); |
1270 | } | |
61007b31 | 1271 | |
cb2e2878 | 1272 | if (ret <= 0) { |
1143ec5e VSO |
1273 | QEMUIOVector local_qiov; |
1274 | ||
cb2e2878 | 1275 | /* Must copy-on-read; use the bounce buffer */ |
0d93ed08 | 1276 | pnum = MIN(pnum, MAX_BOUNCE_BUFFER); |
2275cc90 | 1277 | if (!bounce_buffer) { |
fc6b211f | 1278 | int64_t max_we_need = MAX(pnum, align_bytes - pnum); |
2275cc90 VSO |
1279 | int64_t max_allowed = MIN(max_transfer, MAX_BOUNCE_BUFFER); |
1280 | int64_t bounce_buffer_len = MIN(max_we_need, max_allowed); | |
1281 | ||
1282 | bounce_buffer = qemu_try_blockalign(bs, bounce_buffer_len); | |
1283 | if (!bounce_buffer) { | |
1284 | ret = -ENOMEM; | |
1285 | goto err; | |
1286 | } | |
1287 | } | |
0d93ed08 | 1288 | qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum); |
61007b31 | 1289 | |
fc6b211f | 1290 | ret = bdrv_driver_preadv(bs, align_offset, pnum, |
ac850bf0 | 1291 | &local_qiov, 0, 0); |
cb2e2878 EB |
1292 | if (ret < 0) { |
1293 | goto err; | |
1294 | } | |
1295 | ||
c834dc05 | 1296 | bdrv_co_debug_event(bs, BLKDBG_COR_WRITE); |
cb2e2878 EB |
1297 | if (drv->bdrv_co_pwrite_zeroes && |
1298 | buffer_is_zero(bounce_buffer, pnum)) { | |
1299 | /* FIXME: Should we (perhaps conditionally) be setting | |
1300 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
1301 | * that still correctly reads as zero? */ | |
fc6b211f | 1302 | ret = bdrv_co_do_pwrite_zeroes(bs, align_offset, pnum, |
7adcf59f | 1303 | BDRV_REQ_WRITE_UNCHANGED); |
cb2e2878 EB |
1304 | } else { |
1305 | /* This does not change the data on the disk, it is not | |
1306 | * necessary to flush even in cache=writethrough mode. | |
1307 | */ | |
fc6b211f | 1308 | ret = bdrv_driver_pwritev(bs, align_offset, pnum, |
ac850bf0 | 1309 | &local_qiov, 0, |
7adcf59f | 1310 | BDRV_REQ_WRITE_UNCHANGED); |
cb2e2878 EB |
1311 | } |
1312 | ||
1313 | if (ret < 0) { | |
1314 | /* It might be okay to ignore write errors for guest | |
1315 | * requests. If this is a deliberate copy-on-read | |
1316 | * then we don't want to ignore the error. Simply | |
1317 | * report it in all cases. | |
1318 | */ | |
1319 | goto err; | |
1320 | } | |
1321 | ||
3299e5ec | 1322 | if (!(flags & BDRV_REQ_PREFETCH)) { |
1143ec5e VSO |
1323 | qemu_iovec_from_buf(qiov, qiov_offset + progress, |
1324 | bounce_buffer + skip_bytes, | |
4ab78b19 | 1325 | MIN(pnum - skip_bytes, bytes - progress)); |
3299e5ec VSO |
1326 | } |
1327 | } else if (!(flags & BDRV_REQ_PREFETCH)) { | |
cb2e2878 | 1328 | /* Read directly into the destination */ |
1143ec5e VSO |
1329 | ret = bdrv_driver_preadv(bs, offset + progress, |
1330 | MIN(pnum - skip_bytes, bytes - progress), | |
1331 | qiov, qiov_offset + progress, 0); | |
cb2e2878 EB |
1332 | if (ret < 0) { |
1333 | goto err; | |
1334 | } | |
1335 | } | |
1336 | ||
fc6b211f AD |
1337 | align_offset += pnum; |
1338 | align_bytes -= pnum; | |
cb2e2878 EB |
1339 | progress += pnum - skip_bytes; |
1340 | skip_bytes = 0; | |
1341 | } | |
1342 | ret = 0; | |
61007b31 SH |
1343 | |
1344 | err: | |
1345 | qemu_vfree(bounce_buffer); | |
1346 | return ret; | |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * Forwards an already correctly aligned request to the BlockDriver. This | |
1a62d0ac EB |
1351 | * handles copy on read, zeroing after EOF, and fragmentation of large |
1352 | * reads; any other features must be implemented by the caller. | |
61007b31 | 1353 | */ |
7b1fb72e KW |
1354 | static int coroutine_fn GRAPH_RDLOCK |
1355 | bdrv_aligned_preadv(BdrvChild *child, BdrvTrackedRequest *req, | |
1356 | int64_t offset, int64_t bytes, int64_t align, | |
1357 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
61007b31 | 1358 | { |
85c97ca7 | 1359 | BlockDriverState *bs = child->bs; |
c9d20029 | 1360 | int64_t total_bytes, max_bytes; |
1a62d0ac | 1361 | int ret = 0; |
8b0c5d76 | 1362 | int64_t bytes_remaining = bytes; |
1a62d0ac | 1363 | int max_transfer; |
61007b31 | 1364 | |
8b0c5d76 | 1365 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
49c07526 KW |
1366 | assert(is_power_of_2(align)); |
1367 | assert((offset & (align - 1)) == 0); | |
1368 | assert((bytes & (align - 1)) == 0); | |
abb06c5a | 1369 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
1a62d0ac EB |
1370 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1371 | align); | |
a604fa2b | 1372 | |
e8b65355 SH |
1373 | /* |
1374 | * TODO: We would need a per-BDS .supported_read_flags and | |
a604fa2b EB |
1375 | * potential fallback support, if we ever implement any read flags |
1376 | * to pass through to drivers. For now, there aren't any | |
e8b65355 SH |
1377 | * passthrough flags except the BDRV_REQ_REGISTERED_BUF optimization hint. |
1378 | */ | |
1379 | assert(!(flags & ~(BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH | | |
1380 | BDRV_REQ_REGISTERED_BUF))); | |
61007b31 SH |
1381 | |
1382 | /* Handle Copy on Read and associated serialisation */ | |
1383 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
1384 | /* If we touch the same cluster it counts as an overlap. This | |
1385 | * guarantees that allocating writes will be serialized and not race | |
1386 | * with each other for the same cluster. For example, in copy-on-read | |
1387 | * it ensures that the CoR read and write operations are atomic and | |
1388 | * guest writes cannot interleave between them. */ | |
8ac5aab2 | 1389 | bdrv_make_request_serialising(req, bdrv_get_cluster_size(bs)); |
18fbd0de PB |
1390 | } else { |
1391 | bdrv_wait_serialising_requests(req); | |
61007b31 SH |
1392 | } |
1393 | ||
61007b31 | 1394 | if (flags & BDRV_REQ_COPY_ON_READ) { |
d6a644bb | 1395 | int64_t pnum; |
61007b31 | 1396 | |
897dd0ec AS |
1397 | /* The flag BDRV_REQ_COPY_ON_READ has reached its addressee */ |
1398 | flags &= ~BDRV_REQ_COPY_ON_READ; | |
1399 | ||
cc323997 | 1400 | ret = bdrv_co_is_allocated(bs, offset, bytes, &pnum); |
61007b31 SH |
1401 | if (ret < 0) { |
1402 | goto out; | |
1403 | } | |
1404 | ||
88e63df2 | 1405 | if (!ret || pnum != bytes) { |
65cd4424 VSO |
1406 | ret = bdrv_co_do_copy_on_readv(child, offset, bytes, |
1407 | qiov, qiov_offset, flags); | |
3299e5ec VSO |
1408 | goto out; |
1409 | } else if (flags & BDRV_REQ_PREFETCH) { | |
61007b31 SH |
1410 | goto out; |
1411 | } | |
1412 | } | |
1413 | ||
1a62d0ac | 1414 | /* Forward the request to the BlockDriver, possibly fragmenting it */ |
0af02bd1 | 1415 | total_bytes = bdrv_co_getlength(bs); |
c9d20029 KW |
1416 | if (total_bytes < 0) { |
1417 | ret = total_bytes; | |
1418 | goto out; | |
1419 | } | |
61007b31 | 1420 | |
e8b65355 | 1421 | assert(!(flags & ~(bs->supported_read_flags | BDRV_REQ_REGISTERED_BUF))); |
897dd0ec | 1422 | |
c9d20029 | 1423 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
1a62d0ac | 1424 | if (bytes <= max_bytes && bytes <= max_transfer) { |
897dd0ec | 1425 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, qiov_offset, flags); |
1a62d0ac EB |
1426 | goto out; |
1427 | } | |
61007b31 | 1428 | |
1a62d0ac | 1429 | while (bytes_remaining) { |
8b0c5d76 | 1430 | int64_t num; |
61007b31 | 1431 | |
1a62d0ac | 1432 | if (max_bytes) { |
1a62d0ac EB |
1433 | num = MIN(bytes_remaining, MIN(max_bytes, max_transfer)); |
1434 | assert(num); | |
61007b31 | 1435 | |
1a62d0ac | 1436 | ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining, |
134b7dec | 1437 | num, qiov, |
897dd0ec AS |
1438 | qiov_offset + bytes - bytes_remaining, |
1439 | flags); | |
1a62d0ac | 1440 | max_bytes -= num; |
1a62d0ac EB |
1441 | } else { |
1442 | num = bytes_remaining; | |
134b7dec HR |
1443 | ret = qemu_iovec_memset(qiov, qiov_offset + bytes - bytes_remaining, |
1444 | 0, bytes_remaining); | |
1a62d0ac EB |
1445 | } |
1446 | if (ret < 0) { | |
1447 | goto out; | |
1448 | } | |
1449 | bytes_remaining -= num; | |
61007b31 SH |
1450 | } |
1451 | ||
1452 | out: | |
1a62d0ac | 1453 | return ret < 0 ? ret : 0; |
61007b31 SH |
1454 | } |
1455 | ||
61007b31 | 1456 | /* |
7a3f542f VSO |
1457 | * Request padding |
1458 | * | |
1459 | * |<---- align ----->| |<----- align ---->| | |
1460 | * |<- head ->|<------------- bytes ------------->|<-- tail -->| | |
1461 | * | | | | | | | |
1462 | * -*----------$-------*-------- ... --------*-----$------------*--- | |
1463 | * | | | | | | | |
1464 | * | offset | | end | | |
1465 | * ALIGN_DOWN(offset) ALIGN_UP(offset) ALIGN_DOWN(end) ALIGN_UP(end) | |
1466 | * [buf ... ) [tail_buf ) | |
1467 | * | |
1468 | * @buf is an aligned allocation needed to store @head and @tail paddings. @head | |
1469 | * is placed at the beginning of @buf and @tail at the @end. | |
1470 | * | |
1471 | * @tail_buf is a pointer to sub-buffer, corresponding to align-sized chunk | |
1472 | * around tail, if tail exists. | |
1473 | * | |
1474 | * @merge_reads is true for small requests, | |
1475 | * if @buf_len == @head + bytes + @tail. In this case it is possible that both | |
1476 | * head and tail exist but @buf_len == align and @tail_buf == @buf. | |
18743311 HC |
1477 | * |
1478 | * @write is true for write requests, false for read requests. | |
1479 | * | |
1480 | * If padding makes the vector too long (exceeding IOV_MAX), then we need to | |
1481 | * merge existing vector elements into a single one. @collapse_bounce_buf acts | |
1482 | * as the bounce buffer in such cases. @pre_collapse_qiov has the pre-collapse | |
1483 | * I/O vector elements so for read requests, the data can be copied back after | |
1484 | * the read is done. | |
7a3f542f VSO |
1485 | */ |
1486 | typedef struct BdrvRequestPadding { | |
1487 | uint8_t *buf; | |
1488 | size_t buf_len; | |
1489 | uint8_t *tail_buf; | |
1490 | size_t head; | |
1491 | size_t tail; | |
1492 | bool merge_reads; | |
18743311 | 1493 | bool write; |
7a3f542f | 1494 | QEMUIOVector local_qiov; |
18743311 HC |
1495 | |
1496 | uint8_t *collapse_bounce_buf; | |
1497 | size_t collapse_len; | |
1498 | QEMUIOVector pre_collapse_qiov; | |
7a3f542f VSO |
1499 | } BdrvRequestPadding; |
1500 | ||
1501 | static bool bdrv_init_padding(BlockDriverState *bs, | |
1502 | int64_t offset, int64_t bytes, | |
18743311 | 1503 | bool write, |
7a3f542f VSO |
1504 | BdrvRequestPadding *pad) |
1505 | { | |
a56ed80c VSO |
1506 | int64_t align = bs->bl.request_alignment; |
1507 | int64_t sum; | |
1508 | ||
1509 | bdrv_check_request(offset, bytes, &error_abort); | |
1510 | assert(align <= INT_MAX); /* documented in block/block_int.h */ | |
1511 | assert(align <= SIZE_MAX / 2); /* so we can allocate the buffer */ | |
7a3f542f VSO |
1512 | |
1513 | memset(pad, 0, sizeof(*pad)); | |
1514 | ||
1515 | pad->head = offset & (align - 1); | |
1516 | pad->tail = ((offset + bytes) & (align - 1)); | |
1517 | if (pad->tail) { | |
1518 | pad->tail = align - pad->tail; | |
1519 | } | |
1520 | ||
ac9d00bf | 1521 | if (!pad->head && !pad->tail) { |
7a3f542f VSO |
1522 | return false; |
1523 | } | |
1524 | ||
ac9d00bf VSO |
1525 | assert(bytes); /* Nothing good in aligning zero-length requests */ |
1526 | ||
7a3f542f VSO |
1527 | sum = pad->head + bytes + pad->tail; |
1528 | pad->buf_len = (sum > align && pad->head && pad->tail) ? 2 * align : align; | |
1529 | pad->buf = qemu_blockalign(bs, pad->buf_len); | |
1530 | pad->merge_reads = sum == pad->buf_len; | |
1531 | if (pad->tail) { | |
1532 | pad->tail_buf = pad->buf + pad->buf_len - align; | |
1533 | } | |
1534 | ||
18743311 HC |
1535 | pad->write = write; |
1536 | ||
7a3f542f VSO |
1537 | return true; |
1538 | } | |
1539 | ||
7b1fb72e KW |
1540 | static int coroutine_fn GRAPH_RDLOCK |
1541 | bdrv_padding_rmw_read(BdrvChild *child, BdrvTrackedRequest *req, | |
1542 | BdrvRequestPadding *pad, bool zero_middle) | |
7a3f542f VSO |
1543 | { |
1544 | QEMUIOVector local_qiov; | |
1545 | BlockDriverState *bs = child->bs; | |
1546 | uint64_t align = bs->bl.request_alignment; | |
1547 | int ret; | |
1548 | ||
1549 | assert(req->serialising && pad->buf); | |
1550 | ||
1551 | if (pad->head || pad->merge_reads) { | |
8b0c5d76 | 1552 | int64_t bytes = pad->merge_reads ? pad->buf_len : align; |
7a3f542f VSO |
1553 | |
1554 | qemu_iovec_init_buf(&local_qiov, pad->buf, bytes); | |
1555 | ||
1556 | if (pad->head) { | |
c834dc05 | 1557 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
7a3f542f VSO |
1558 | } |
1559 | if (pad->merge_reads && pad->tail) { | |
c834dc05 | 1560 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
7a3f542f VSO |
1561 | } |
1562 | ret = bdrv_aligned_preadv(child, req, req->overlap_offset, bytes, | |
65cd4424 | 1563 | align, &local_qiov, 0, 0); |
7a3f542f VSO |
1564 | if (ret < 0) { |
1565 | return ret; | |
1566 | } | |
1567 | if (pad->head) { | |
c834dc05 | 1568 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
7a3f542f VSO |
1569 | } |
1570 | if (pad->merge_reads && pad->tail) { | |
c834dc05 | 1571 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
7a3f542f VSO |
1572 | } |
1573 | ||
1574 | if (pad->merge_reads) { | |
1575 | goto zero_mem; | |
1576 | } | |
1577 | } | |
1578 | ||
1579 | if (pad->tail) { | |
1580 | qemu_iovec_init_buf(&local_qiov, pad->tail_buf, align); | |
1581 | ||
c834dc05 | 1582 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
7a3f542f VSO |
1583 | ret = bdrv_aligned_preadv( |
1584 | child, req, | |
1585 | req->overlap_offset + req->overlap_bytes - align, | |
65cd4424 | 1586 | align, align, &local_qiov, 0, 0); |
7a3f542f VSO |
1587 | if (ret < 0) { |
1588 | return ret; | |
1589 | } | |
c834dc05 | 1590 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
7a3f542f VSO |
1591 | } |
1592 | ||
1593 | zero_mem: | |
1594 | if (zero_middle) { | |
1595 | memset(pad->buf + pad->head, 0, pad->buf_len - pad->head - pad->tail); | |
1596 | } | |
1597 | ||
1598 | return 0; | |
1599 | } | |
1600 | ||
18743311 HC |
1601 | /** |
1602 | * Free *pad's associated buffers, and perform any necessary finalization steps. | |
1603 | */ | |
1604 | static void bdrv_padding_finalize(BdrvRequestPadding *pad) | |
7a3f542f | 1605 | { |
18743311 HC |
1606 | if (pad->collapse_bounce_buf) { |
1607 | if (!pad->write) { | |
1608 | /* | |
1609 | * If padding required elements in the vector to be collapsed into a | |
1610 | * bounce buffer, copy the bounce buffer content back | |
1611 | */ | |
1612 | qemu_iovec_from_buf(&pad->pre_collapse_qiov, 0, | |
1613 | pad->collapse_bounce_buf, pad->collapse_len); | |
1614 | } | |
1615 | qemu_vfree(pad->collapse_bounce_buf); | |
1616 | qemu_iovec_destroy(&pad->pre_collapse_qiov); | |
1617 | } | |
7a3f542f VSO |
1618 | if (pad->buf) { |
1619 | qemu_vfree(pad->buf); | |
1620 | qemu_iovec_destroy(&pad->local_qiov); | |
1621 | } | |
98ca4549 | 1622 | memset(pad, 0, sizeof(*pad)); |
7a3f542f VSO |
1623 | } |
1624 | ||
18743311 HC |
1625 | /* |
1626 | * Create pad->local_qiov by wrapping @iov in the padding head and tail, while | |
1627 | * ensuring that the resulting vector will not exceed IOV_MAX elements. | |
1628 | * | |
1629 | * To ensure this, when necessary, the first two or three elements of @iov are | |
1630 | * merged into pad->collapse_bounce_buf and replaced by a reference to that | |
1631 | * bounce buffer in pad->local_qiov. | |
1632 | * | |
1633 | * After performing a read request, the data from the bounce buffer must be | |
1634 | * copied back into pad->pre_collapse_qiov (e.g. by bdrv_padding_finalize()). | |
1635 | */ | |
1636 | static int bdrv_create_padded_qiov(BlockDriverState *bs, | |
1637 | BdrvRequestPadding *pad, | |
1638 | struct iovec *iov, int niov, | |
1639 | size_t iov_offset, size_t bytes) | |
1640 | { | |
1641 | int padded_niov, surplus_count, collapse_count; | |
1642 | ||
1643 | /* Assert this invariant */ | |
1644 | assert(niov <= IOV_MAX); | |
1645 | ||
1646 | /* | |
1647 | * Cannot pad if resulting length would exceed SIZE_MAX. Returning an error | |
1648 | * to the guest is not ideal, but there is little else we can do. At least | |
1649 | * this will practically never happen on 64-bit systems. | |
1650 | */ | |
1651 | if (SIZE_MAX - pad->head < bytes || | |
1652 | SIZE_MAX - pad->head - bytes < pad->tail) | |
1653 | { | |
1654 | return -EINVAL; | |
1655 | } | |
1656 | ||
1657 | /* Length of the resulting IOV if we just concatenated everything */ | |
1658 | padded_niov = !!pad->head + niov + !!pad->tail; | |
1659 | ||
1660 | qemu_iovec_init(&pad->local_qiov, MIN(padded_niov, IOV_MAX)); | |
1661 | ||
1662 | if (pad->head) { | |
1663 | qemu_iovec_add(&pad->local_qiov, pad->buf, pad->head); | |
1664 | } | |
1665 | ||
1666 | /* | |
1667 | * If padded_niov > IOV_MAX, we cannot just concatenate everything. | |
1668 | * Instead, merge the first two or three elements of @iov to reduce the | |
1669 | * number of vector elements as necessary. | |
1670 | */ | |
1671 | if (padded_niov > IOV_MAX) { | |
1672 | /* | |
1673 | * Only head and tail can have lead to the number of entries exceeding | |
1674 | * IOV_MAX, so we can exceed it by the head and tail at most. We need | |
1675 | * to reduce the number of elements by `surplus_count`, so we merge that | |
1676 | * many elements plus one into one element. | |
1677 | */ | |
1678 | surplus_count = padded_niov - IOV_MAX; | |
1679 | assert(surplus_count <= !!pad->head + !!pad->tail); | |
1680 | collapse_count = surplus_count + 1; | |
1681 | ||
1682 | /* | |
1683 | * Move the elements to collapse into `pad->pre_collapse_qiov`, then | |
1684 | * advance `iov` (and associated variables) by those elements. | |
1685 | */ | |
1686 | qemu_iovec_init(&pad->pre_collapse_qiov, collapse_count); | |
1687 | qemu_iovec_concat_iov(&pad->pre_collapse_qiov, iov, | |
1688 | collapse_count, iov_offset, SIZE_MAX); | |
1689 | iov += collapse_count; | |
1690 | iov_offset = 0; | |
1691 | niov -= collapse_count; | |
1692 | bytes -= pad->pre_collapse_qiov.size; | |
1693 | ||
1694 | /* | |
1695 | * Construct the bounce buffer to match the length of the to-collapse | |
1696 | * vector elements, and for write requests, initialize it with the data | |
1697 | * from those elements. Then add it to `pad->local_qiov`. | |
1698 | */ | |
1699 | pad->collapse_len = pad->pre_collapse_qiov.size; | |
1700 | pad->collapse_bounce_buf = qemu_blockalign(bs, pad->collapse_len); | |
1701 | if (pad->write) { | |
1702 | qemu_iovec_to_buf(&pad->pre_collapse_qiov, 0, | |
1703 | pad->collapse_bounce_buf, pad->collapse_len); | |
1704 | } | |
1705 | qemu_iovec_add(&pad->local_qiov, | |
1706 | pad->collapse_bounce_buf, pad->collapse_len); | |
1707 | } | |
1708 | ||
1709 | qemu_iovec_concat_iov(&pad->local_qiov, iov, niov, iov_offset, bytes); | |
1710 | ||
1711 | if (pad->tail) { | |
1712 | qemu_iovec_add(&pad->local_qiov, | |
1713 | pad->buf + pad->buf_len - pad->tail, pad->tail); | |
1714 | } | |
1715 | ||
1716 | assert(pad->local_qiov.niov == MIN(padded_niov, IOV_MAX)); | |
1717 | return 0; | |
1718 | } | |
1719 | ||
7a3f542f VSO |
1720 | /* |
1721 | * bdrv_pad_request | |
1722 | * | |
1723 | * Exchange request parameters with padded request if needed. Don't include RMW | |
1724 | * read of padding, bdrv_padding_rmw_read() should be called separately if | |
1725 | * needed. | |
1726 | * | |
18743311 HC |
1727 | * @write is true for write requests, false for read requests. |
1728 | * | |
98ca4549 VSO |
1729 | * Request parameters (@qiov, &qiov_offset, &offset, &bytes) are in-out: |
1730 | * - on function start they represent original request | |
1731 | * - on failure or when padding is not needed they are unchanged | |
1732 | * - on success when padding is needed they represent padded request | |
61007b31 | 1733 | */ |
98ca4549 VSO |
1734 | static int bdrv_pad_request(BlockDriverState *bs, |
1735 | QEMUIOVector **qiov, size_t *qiov_offset, | |
37e9403e | 1736 | int64_t *offset, int64_t *bytes, |
18743311 | 1737 | bool write, |
e8b65355 SH |
1738 | BdrvRequestPadding *pad, bool *padded, |
1739 | BdrvRequestFlags *flags) | |
7a3f542f | 1740 | { |
4c002cef | 1741 | int ret; |
18743311 HC |
1742 | struct iovec *sliced_iov; |
1743 | int sliced_niov; | |
1744 | size_t sliced_head, sliced_tail; | |
4c002cef | 1745 | |
ef256751 HC |
1746 | /* Should have been checked by the caller already */ |
1747 | ret = bdrv_check_request32(*offset, *bytes, *qiov, *qiov_offset); | |
1748 | if (ret < 0) { | |
1749 | return ret; | |
1750 | } | |
37e9403e | 1751 | |
18743311 | 1752 | if (!bdrv_init_padding(bs, *offset, *bytes, write, pad)) { |
98ca4549 VSO |
1753 | if (padded) { |
1754 | *padded = false; | |
1755 | } | |
1756 | return 0; | |
7a3f542f VSO |
1757 | } |
1758 | ||
18743311 HC |
1759 | sliced_iov = qemu_iovec_slice(*qiov, *qiov_offset, *bytes, |
1760 | &sliced_head, &sliced_tail, | |
1761 | &sliced_niov); | |
1762 | ||
ef256751 | 1763 | /* Guaranteed by bdrv_check_request32() */ |
18743311 HC |
1764 | assert(*bytes <= SIZE_MAX); |
1765 | ret = bdrv_create_padded_qiov(bs, pad, sliced_iov, sliced_niov, | |
1766 | sliced_head, *bytes); | |
98ca4549 | 1767 | if (ret < 0) { |
18743311 | 1768 | bdrv_padding_finalize(pad); |
98ca4549 VSO |
1769 | return ret; |
1770 | } | |
7a3f542f VSO |
1771 | *bytes += pad->head + pad->tail; |
1772 | *offset -= pad->head; | |
1773 | *qiov = &pad->local_qiov; | |
1acc3466 | 1774 | *qiov_offset = 0; |
98ca4549 VSO |
1775 | if (padded) { |
1776 | *padded = true; | |
1777 | } | |
e8b65355 SH |
1778 | if (flags) { |
1779 | /* Can't use optimization hint with bounce buffer */ | |
1780 | *flags &= ~BDRV_REQ_REGISTERED_BUF; | |
1781 | } | |
7a3f542f | 1782 | |
98ca4549 | 1783 | return 0; |
7a3f542f VSO |
1784 | } |
1785 | ||
a03ef88f | 1786 | int coroutine_fn bdrv_co_preadv(BdrvChild *child, |
e9e52efd | 1787 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, |
61007b31 | 1788 | BdrvRequestFlags flags) |
1acc3466 | 1789 | { |
967d7905 | 1790 | IO_CODE(); |
1acc3466 VSO |
1791 | return bdrv_co_preadv_part(child, offset, bytes, qiov, 0, flags); |
1792 | } | |
1793 | ||
1794 | int coroutine_fn bdrv_co_preadv_part(BdrvChild *child, | |
37e9403e | 1795 | int64_t offset, int64_t bytes, |
1acc3466 VSO |
1796 | QEMUIOVector *qiov, size_t qiov_offset, |
1797 | BdrvRequestFlags flags) | |
61007b31 | 1798 | { |
a03ef88f | 1799 | BlockDriverState *bs = child->bs; |
61007b31 | 1800 | BdrvTrackedRequest req; |
7a3f542f | 1801 | BdrvRequestPadding pad; |
61007b31 | 1802 | int ret; |
967d7905 | 1803 | IO_CODE(); |
61007b31 | 1804 | |
37e9403e | 1805 | trace_bdrv_co_preadv_part(bs, offset, bytes, flags); |
61007b31 | 1806 | |
1e97be91 | 1807 | if (!bdrv_co_is_inserted(bs)) { |
f4dad307 VSO |
1808 | return -ENOMEDIUM; |
1809 | } | |
1810 | ||
63f4ad11 | 1811 | ret = bdrv_check_request32(offset, bytes, qiov, qiov_offset); |
61007b31 SH |
1812 | if (ret < 0) { |
1813 | return ret; | |
1814 | } | |
1815 | ||
ac9d00bf VSO |
1816 | if (bytes == 0 && !QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)) { |
1817 | /* | |
1818 | * Aligning zero request is nonsense. Even if driver has special meaning | |
1819 | * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass | |
1820 | * it to driver due to request_alignment. | |
1821 | * | |
1822 | * Still, no reason to return an error if someone do unaligned | |
1823 | * zero-length read occasionally. | |
1824 | */ | |
1825 | return 0; | |
1826 | } | |
1827 | ||
99723548 PB |
1828 | bdrv_inc_in_flight(bs); |
1829 | ||
9568b511 | 1830 | /* Don't do copy-on-read if we read data before write operation */ |
d73415a3 | 1831 | if (qatomic_read(&bs->copy_on_read)) { |
61007b31 SH |
1832 | flags |= BDRV_REQ_COPY_ON_READ; |
1833 | } | |
1834 | ||
18743311 HC |
1835 | ret = bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, false, |
1836 | &pad, NULL, &flags); | |
98ca4549 | 1837 | if (ret < 0) { |
87ab8802 | 1838 | goto fail; |
98ca4549 | 1839 | } |
61007b31 | 1840 | |
ebde595c | 1841 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
7a3f542f VSO |
1842 | ret = bdrv_aligned_preadv(child, &req, offset, bytes, |
1843 | bs->bl.request_alignment, | |
1acc3466 | 1844 | qiov, qiov_offset, flags); |
61007b31 | 1845 | tracked_request_end(&req); |
18743311 | 1846 | bdrv_padding_finalize(&pad); |
61007b31 | 1847 | |
87ab8802 KW |
1848 | fail: |
1849 | bdrv_dec_in_flight(bs); | |
1850 | ||
61007b31 SH |
1851 | return ret; |
1852 | } | |
1853 | ||
eeb47775 KW |
1854 | static int coroutine_fn GRAPH_RDLOCK |
1855 | bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1856 | BdrvRequestFlags flags) | |
61007b31 SH |
1857 | { |
1858 | BlockDriver *drv = bs->drv; | |
1859 | QEMUIOVector qiov; | |
0d93ed08 | 1860 | void *buf = NULL; |
61007b31 | 1861 | int ret = 0; |
465fe887 | 1862 | bool need_flush = false; |
443668ca DL |
1863 | int head = 0; |
1864 | int tail = 0; | |
61007b31 | 1865 | |
2aaa3f9b VSO |
1866 | int64_t max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, |
1867 | INT64_MAX); | |
a5b8dd2c EB |
1868 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1869 | bs->bl.request_alignment); | |
cb2e2878 | 1870 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER); |
d05aa8bb | 1871 | |
abaf8b75 | 1872 | assert_bdrv_graph_readable(); |
5ae07b14 VSO |
1873 | bdrv_check_request(offset, bytes, &error_abort); |
1874 | ||
d470ad42 HR |
1875 | if (!drv) { |
1876 | return -ENOMEDIUM; | |
1877 | } | |
1878 | ||
fe0480d6 KW |
1879 | if ((flags & ~bs->supported_zero_flags) & BDRV_REQ_NO_FALLBACK) { |
1880 | return -ENOTSUP; | |
1881 | } | |
1882 | ||
e8b65355 SH |
1883 | /* By definition there is no user buffer so this flag doesn't make sense */ |
1884 | if (flags & BDRV_REQ_REGISTERED_BUF) { | |
1885 | return -EINVAL; | |
1886 | } | |
1887 | ||
0bc329fb HR |
1888 | /* Invalidate the cached block-status data range if this write overlaps */ |
1889 | bdrv_bsc_invalidate_range(bs, offset, bytes); | |
1890 | ||
b8d0a980 EB |
1891 | assert(alignment % bs->bl.request_alignment == 0); |
1892 | head = offset % alignment; | |
f5a5ca79 | 1893 | tail = (offset + bytes) % alignment; |
b8d0a980 EB |
1894 | max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment); |
1895 | assert(max_write_zeroes >= bs->bl.request_alignment); | |
61007b31 | 1896 | |
f5a5ca79 | 1897 | while (bytes > 0 && !ret) { |
5ae07b14 | 1898 | int64_t num = bytes; |
61007b31 SH |
1899 | |
1900 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1901 | * to be aligned, and that unaligned requests do not cross cluster |
1902 | * boundaries. | |
61007b31 | 1903 | */ |
443668ca | 1904 | if (head) { |
b2f95fee EB |
1905 | /* Make a small request up to the first aligned sector. For |
1906 | * convenience, limit this request to max_transfer even if | |
1907 | * we don't need to fall back to writes. */ | |
f5a5ca79 | 1908 | num = MIN(MIN(bytes, max_transfer), alignment - head); |
b2f95fee EB |
1909 | head = (head + num) % alignment; |
1910 | assert(num < max_write_zeroes); | |
d05aa8bb | 1911 | } else if (tail && num > alignment) { |
443668ca DL |
1912 | /* Shorten the request to the last aligned sector. */ |
1913 | num -= tail; | |
61007b31 SH |
1914 | } |
1915 | ||
1916 | /* limit request size */ | |
1917 | if (num > max_write_zeroes) { | |
1918 | num = max_write_zeroes; | |
1919 | } | |
1920 | ||
1921 | ret = -ENOTSUP; | |
1922 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1923 | if (drv->bdrv_co_pwrite_zeroes) { |
1924 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1925 | flags & bs->supported_zero_flags); | |
1926 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1927 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1928 | need_flush = true; | |
1929 | } | |
465fe887 EB |
1930 | } else { |
1931 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1932 | } |
1933 | ||
294682cc | 1934 | if (ret == -ENOTSUP && !(flags & BDRV_REQ_NO_FALLBACK)) { |
61007b31 | 1935 | /* Fall back to bounce buffer if write zeroes is unsupported */ |
465fe887 EB |
1936 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1937 | ||
1938 | if ((flags & BDRV_REQ_FUA) && | |
1939 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1940 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1941 | * flush on each chunk; use just one at the end */ | |
1942 | write_flags &= ~BDRV_REQ_FUA; | |
1943 | need_flush = true; | |
1944 | } | |
5def6b80 | 1945 | num = MIN(num, max_transfer); |
0d93ed08 VSO |
1946 | if (buf == NULL) { |
1947 | buf = qemu_try_blockalign0(bs, num); | |
1948 | if (buf == NULL) { | |
61007b31 SH |
1949 | ret = -ENOMEM; |
1950 | goto fail; | |
1951 | } | |
61007b31 | 1952 | } |
0d93ed08 | 1953 | qemu_iovec_init_buf(&qiov, buf, num); |
61007b31 | 1954 | |
ac850bf0 | 1955 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, 0, write_flags); |
61007b31 SH |
1956 | |
1957 | /* Keep bounce buffer around if it is big enough for all | |
1958 | * all future requests. | |
1959 | */ | |
5def6b80 | 1960 | if (num < max_transfer) { |
0d93ed08 VSO |
1961 | qemu_vfree(buf); |
1962 | buf = NULL; | |
61007b31 SH |
1963 | } |
1964 | } | |
1965 | ||
d05aa8bb | 1966 | offset += num; |
f5a5ca79 | 1967 | bytes -= num; |
61007b31 SH |
1968 | } |
1969 | ||
1970 | fail: | |
465fe887 EB |
1971 | if (ret == 0 && need_flush) { |
1972 | ret = bdrv_co_flush(bs); | |
1973 | } | |
0d93ed08 | 1974 | qemu_vfree(buf); |
61007b31 SH |
1975 | return ret; |
1976 | } | |
1977 | ||
a00e70c0 | 1978 | static inline int coroutine_fn GRAPH_RDLOCK |
fcfd9ade | 1979 | bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, int64_t bytes, |
85fe2479 FZ |
1980 | BdrvTrackedRequest *req, int flags) |
1981 | { | |
1982 | BlockDriverState *bs = child->bs; | |
fcfd9ade VSO |
1983 | |
1984 | bdrv_check_request(offset, bytes, &error_abort); | |
85fe2479 | 1985 | |
307261b2 | 1986 | if (bdrv_is_read_only(bs)) { |
85fe2479 FZ |
1987 | return -EPERM; |
1988 | } | |
1989 | ||
85fe2479 FZ |
1990 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
1991 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); | |
1992 | assert(!(flags & ~BDRV_REQ_MASK)); | |
d1a764d1 | 1993 | assert(!((flags & BDRV_REQ_NO_WAIT) && !(flags & BDRV_REQ_SERIALISING))); |
85fe2479 FZ |
1994 | |
1995 | if (flags & BDRV_REQ_SERIALISING) { | |
d1a764d1 VSO |
1996 | QEMU_LOCK_GUARD(&bs->reqs_lock); |
1997 | ||
1998 | tracked_request_set_serialising(req, bdrv_get_cluster_size(bs)); | |
1999 | ||
2000 | if ((flags & BDRV_REQ_NO_WAIT) && bdrv_find_conflicting_request(req)) { | |
2001 | return -EBUSY; | |
2002 | } | |
2003 | ||
2004 | bdrv_wait_serialising_requests_locked(req); | |
18fbd0de PB |
2005 | } else { |
2006 | bdrv_wait_serialising_requests(req); | |
85fe2479 FZ |
2007 | } |
2008 | ||
85fe2479 FZ |
2009 | assert(req->overlap_offset <= offset); |
2010 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
fcfd9ade VSO |
2011 | assert(offset + bytes <= bs->total_sectors * BDRV_SECTOR_SIZE || |
2012 | child->perm & BLK_PERM_RESIZE); | |
85fe2479 | 2013 | |
cd47d792 FZ |
2014 | switch (req->type) { |
2015 | case BDRV_TRACKED_WRITE: | |
2016 | case BDRV_TRACKED_DISCARD: | |
2017 | if (flags & BDRV_REQ_WRITE_UNCHANGED) { | |
2018 | assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
2019 | } else { | |
2020 | assert(child->perm & BLK_PERM_WRITE); | |
2021 | } | |
94783301 VSO |
2022 | bdrv_write_threshold_check_write(bs, offset, bytes); |
2023 | return 0; | |
cd47d792 FZ |
2024 | case BDRV_TRACKED_TRUNCATE: |
2025 | assert(child->perm & BLK_PERM_RESIZE); | |
2026 | return 0; | |
2027 | default: | |
2028 | abort(); | |
85fe2479 | 2029 | } |
85fe2479 FZ |
2030 | } |
2031 | ||
7859c45a | 2032 | static inline void coroutine_fn GRAPH_RDLOCK |
fcfd9ade | 2033 | bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, int64_t bytes, |
85fe2479 FZ |
2034 | BdrvTrackedRequest *req, int ret) |
2035 | { | |
2036 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
2037 | BlockDriverState *bs = child->bs; | |
2038 | ||
fcfd9ade VSO |
2039 | bdrv_check_request(offset, bytes, &error_abort); |
2040 | ||
d73415a3 | 2041 | qatomic_inc(&bs->write_gen); |
85fe2479 | 2042 | |
00695c27 FZ |
2043 | /* |
2044 | * Discard cannot extend the image, but in error handling cases, such as | |
2045 | * when reverting a qcow2 cluster allocation, the discarded range can pass | |
2046 | * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD | |
2047 | * here. Instead, just skip it, since semantically a discard request | |
2048 | * beyond EOF cannot expand the image anyway. | |
2049 | */ | |
7f8f03ef | 2050 | if (ret == 0 && |
cd47d792 FZ |
2051 | (req->type == BDRV_TRACKED_TRUNCATE || |
2052 | end_sector > bs->total_sectors) && | |
2053 | req->type != BDRV_TRACKED_DISCARD) { | |
7f8f03ef FZ |
2054 | bs->total_sectors = end_sector; |
2055 | bdrv_parent_cb_resize(bs); | |
2056 | bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS); | |
85fe2479 | 2057 | } |
00695c27 FZ |
2058 | if (req->bytes) { |
2059 | switch (req->type) { | |
2060 | case BDRV_TRACKED_WRITE: | |
2061 | stat64_max(&bs->wr_highest_offset, offset + bytes); | |
2062 | /* fall through, to set dirty bits */ | |
2063 | case BDRV_TRACKED_DISCARD: | |
2064 | bdrv_set_dirty(bs, offset, bytes); | |
2065 | break; | |
2066 | default: | |
2067 | break; | |
2068 | } | |
2069 | } | |
85fe2479 FZ |
2070 | } |
2071 | ||
61007b31 | 2072 | /* |
04ed95f4 EB |
2073 | * Forwards an already correctly aligned write request to the BlockDriver, |
2074 | * after possibly fragmenting it. | |
61007b31 | 2075 | */ |
7b1fb72e KW |
2076 | static int coroutine_fn GRAPH_RDLOCK |
2077 | bdrv_aligned_pwritev(BdrvChild *child, BdrvTrackedRequest *req, | |
2078 | int64_t offset, int64_t bytes, int64_t align, | |
2079 | QEMUIOVector *qiov, size_t qiov_offset, | |
2080 | BdrvRequestFlags flags) | |
61007b31 | 2081 | { |
85c97ca7 | 2082 | BlockDriverState *bs = child->bs; |
61007b31 | 2083 | BlockDriver *drv = bs->drv; |
61007b31 SH |
2084 | int ret; |
2085 | ||
fcfd9ade | 2086 | int64_t bytes_remaining = bytes; |
04ed95f4 | 2087 | int max_transfer; |
61007b31 | 2088 | |
fcfd9ade VSO |
2089 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
2090 | ||
d470ad42 HR |
2091 | if (!drv) { |
2092 | return -ENOMEDIUM; | |
2093 | } | |
2094 | ||
d6883bc9 VSO |
2095 | if (bdrv_has_readonly_bitmaps(bs)) { |
2096 | return -EPERM; | |
2097 | } | |
2098 | ||
cff86b38 EB |
2099 | assert(is_power_of_2(align)); |
2100 | assert((offset & (align - 1)) == 0); | |
2101 | assert((bytes & (align - 1)) == 0); | |
04ed95f4 EB |
2102 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
2103 | align); | |
61007b31 | 2104 | |
85fe2479 | 2105 | ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags); |
61007b31 SH |
2106 | |
2107 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 2108 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
28c4da28 | 2109 | qemu_iovec_is_zero(qiov, qiov_offset, bytes)) { |
61007b31 SH |
2110 | flags |= BDRV_REQ_ZERO_WRITE; |
2111 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
2112 | flags |= BDRV_REQ_MAY_UNMAP; | |
2113 | } | |
3c586715 SH |
2114 | |
2115 | /* Can't use optimization hint with bufferless zero write */ | |
2116 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
61007b31 SH |
2117 | } |
2118 | ||
2119 | if (ret < 0) { | |
2120 | /* Do nothing, write notifier decided to fail this request */ | |
2121 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
c834dc05 | 2122 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 2123 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
3ea1a091 | 2124 | } else if (flags & BDRV_REQ_WRITE_COMPRESSED) { |
28c4da28 VSO |
2125 | ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, |
2126 | qiov, qiov_offset); | |
04ed95f4 | 2127 | } else if (bytes <= max_transfer) { |
c834dc05 | 2128 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV); |
28c4da28 | 2129 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, qiov_offset, flags); |
04ed95f4 | 2130 | } else { |
c834dc05 | 2131 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV); |
04ed95f4 EB |
2132 | while (bytes_remaining) { |
2133 | int num = MIN(bytes_remaining, max_transfer); | |
04ed95f4 EB |
2134 | int local_flags = flags; |
2135 | ||
2136 | assert(num); | |
2137 | if (num < bytes_remaining && (flags & BDRV_REQ_FUA) && | |
2138 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
2139 | /* If FUA is going to be emulated by flush, we only | |
2140 | * need to flush on the last iteration */ | |
2141 | local_flags &= ~BDRV_REQ_FUA; | |
2142 | } | |
04ed95f4 EB |
2143 | |
2144 | ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining, | |
134b7dec HR |
2145 | num, qiov, |
2146 | qiov_offset + bytes - bytes_remaining, | |
28c4da28 | 2147 | local_flags); |
04ed95f4 EB |
2148 | if (ret < 0) { |
2149 | break; | |
2150 | } | |
2151 | bytes_remaining -= num; | |
2152 | } | |
61007b31 | 2153 | } |
c834dc05 | 2154 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 2155 | |
61007b31 | 2156 | if (ret >= 0) { |
04ed95f4 | 2157 | ret = 0; |
61007b31 | 2158 | } |
85fe2479 | 2159 | bdrv_co_write_req_finish(child, offset, bytes, req, ret); |
61007b31 SH |
2160 | |
2161 | return ret; | |
2162 | } | |
2163 | ||
7b1fb72e KW |
2164 | static int coroutine_fn GRAPH_RDLOCK |
2165 | bdrv_co_do_zero_pwritev(BdrvChild *child, int64_t offset, int64_t bytes, | |
2166 | BdrvRequestFlags flags, BdrvTrackedRequest *req) | |
9eeb6dd1 | 2167 | { |
85c97ca7 | 2168 | BlockDriverState *bs = child->bs; |
9eeb6dd1 | 2169 | QEMUIOVector local_qiov; |
a5b8dd2c | 2170 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 | 2171 | int ret = 0; |
7a3f542f VSO |
2172 | bool padding; |
2173 | BdrvRequestPadding pad; | |
9eeb6dd1 | 2174 | |
e8b65355 SH |
2175 | /* This flag doesn't make sense for padding or zero writes */ |
2176 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
2177 | ||
18743311 | 2178 | padding = bdrv_init_padding(bs, offset, bytes, true, &pad); |
7a3f542f | 2179 | if (padding) { |
45e62b46 | 2180 | assert(!(flags & BDRV_REQ_NO_WAIT)); |
8ac5aab2 | 2181 | bdrv_make_request_serialising(req, align); |
9eeb6dd1 | 2182 | |
7a3f542f VSO |
2183 | bdrv_padding_rmw_read(child, req, &pad, true); |
2184 | ||
2185 | if (pad.head || pad.merge_reads) { | |
2186 | int64_t aligned_offset = offset & ~(align - 1); | |
2187 | int64_t write_bytes = pad.merge_reads ? pad.buf_len : align; | |
2188 | ||
2189 | qemu_iovec_init_buf(&local_qiov, pad.buf, write_bytes); | |
2190 | ret = bdrv_aligned_pwritev(child, req, aligned_offset, write_bytes, | |
28c4da28 | 2191 | align, &local_qiov, 0, |
7a3f542f VSO |
2192 | flags & ~BDRV_REQ_ZERO_WRITE); |
2193 | if (ret < 0 || pad.merge_reads) { | |
2194 | /* Error or all work is done */ | |
2195 | goto out; | |
2196 | } | |
2197 | offset += write_bytes - pad.head; | |
2198 | bytes -= write_bytes - pad.head; | |
9eeb6dd1 | 2199 | } |
9eeb6dd1 FZ |
2200 | } |
2201 | ||
2202 | assert(!bytes || (offset & (align - 1)) == 0); | |
2203 | if (bytes >= align) { | |
2204 | /* Write the aligned part in the middle. */ | |
fcfd9ade | 2205 | int64_t aligned_bytes = bytes & ~(align - 1); |
85c97ca7 | 2206 | ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align, |
28c4da28 | 2207 | NULL, 0, flags); |
9eeb6dd1 | 2208 | if (ret < 0) { |
7a3f542f | 2209 | goto out; |
9eeb6dd1 FZ |
2210 | } |
2211 | bytes -= aligned_bytes; | |
2212 | offset += aligned_bytes; | |
2213 | } | |
2214 | ||
2215 | assert(!bytes || (offset & (align - 1)) == 0); | |
2216 | if (bytes) { | |
7a3f542f | 2217 | assert(align == pad.tail + bytes); |
9eeb6dd1 | 2218 | |
7a3f542f | 2219 | qemu_iovec_init_buf(&local_qiov, pad.tail_buf, align); |
85c97ca7 | 2220 | ret = bdrv_aligned_pwritev(child, req, offset, align, align, |
28c4da28 VSO |
2221 | &local_qiov, 0, |
2222 | flags & ~BDRV_REQ_ZERO_WRITE); | |
9eeb6dd1 | 2223 | } |
9eeb6dd1 | 2224 | |
7a3f542f | 2225 | out: |
18743311 | 2226 | bdrv_padding_finalize(&pad); |
7a3f542f VSO |
2227 | |
2228 | return ret; | |
9eeb6dd1 FZ |
2229 | } |
2230 | ||
61007b31 SH |
2231 | /* |
2232 | * Handle a write request in coroutine context | |
2233 | */ | |
a03ef88f | 2234 | int coroutine_fn bdrv_co_pwritev(BdrvChild *child, |
e9e52efd | 2235 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, |
61007b31 | 2236 | BdrvRequestFlags flags) |
1acc3466 | 2237 | { |
967d7905 | 2238 | IO_CODE(); |
1acc3466 VSO |
2239 | return bdrv_co_pwritev_part(child, offset, bytes, qiov, 0, flags); |
2240 | } | |
2241 | ||
2242 | int coroutine_fn bdrv_co_pwritev_part(BdrvChild *child, | |
37e9403e | 2243 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, size_t qiov_offset, |
1acc3466 | 2244 | BdrvRequestFlags flags) |
61007b31 | 2245 | { |
a03ef88f | 2246 | BlockDriverState *bs = child->bs; |
61007b31 | 2247 | BdrvTrackedRequest req; |
a5b8dd2c | 2248 | uint64_t align = bs->bl.request_alignment; |
7a3f542f | 2249 | BdrvRequestPadding pad; |
61007b31 | 2250 | int ret; |
f0deecff | 2251 | bool padded = false; |
967d7905 | 2252 | IO_CODE(); |
61007b31 | 2253 | |
37e9403e | 2254 | trace_bdrv_co_pwritev_part(child->bs, offset, bytes, flags); |
f42cf447 | 2255 | |
1e97be91 | 2256 | if (!bdrv_co_is_inserted(bs)) { |
61007b31 SH |
2257 | return -ENOMEDIUM; |
2258 | } | |
61007b31 | 2259 | |
2aaa3f9b VSO |
2260 | if (flags & BDRV_REQ_ZERO_WRITE) { |
2261 | ret = bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, NULL); | |
2262 | } else { | |
2263 | ret = bdrv_check_request32(offset, bytes, qiov, qiov_offset); | |
2264 | } | |
61007b31 SH |
2265 | if (ret < 0) { |
2266 | return ret; | |
2267 | } | |
2268 | ||
f2208fdc AG |
2269 | /* If the request is misaligned then we can't make it efficient */ |
2270 | if ((flags & BDRV_REQ_NO_FALLBACK) && | |
2271 | !QEMU_IS_ALIGNED(offset | bytes, align)) | |
2272 | { | |
2273 | return -ENOTSUP; | |
2274 | } | |
2275 | ||
ac9d00bf VSO |
2276 | if (bytes == 0 && !QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)) { |
2277 | /* | |
2278 | * Aligning zero request is nonsense. Even if driver has special meaning | |
2279 | * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass | |
2280 | * it to driver due to request_alignment. | |
2281 | * | |
2282 | * Still, no reason to return an error if someone do unaligned | |
2283 | * zero-length write occasionally. | |
2284 | */ | |
2285 | return 0; | |
2286 | } | |
2287 | ||
f0deecff VSO |
2288 | if (!(flags & BDRV_REQ_ZERO_WRITE)) { |
2289 | /* | |
2290 | * Pad request for following read-modify-write cycle. | |
2291 | * bdrv_co_do_zero_pwritev() does aligning by itself, so, we do | |
2292 | * alignment only if there is no ZERO flag. | |
2293 | */ | |
18743311 HC |
2294 | ret = bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, true, |
2295 | &pad, &padded, &flags); | |
98ca4549 VSO |
2296 | if (ret < 0) { |
2297 | return ret; | |
2298 | } | |
f0deecff VSO |
2299 | } |
2300 | ||
99723548 | 2301 | bdrv_inc_in_flight(bs); |
ebde595c | 2302 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 2303 | |
18a59f03 | 2304 | if (flags & BDRV_REQ_ZERO_WRITE) { |
f0deecff | 2305 | assert(!padded); |
85c97ca7 | 2306 | ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req); |
9eeb6dd1 FZ |
2307 | goto out; |
2308 | } | |
2309 | ||
f0deecff VSO |
2310 | if (padded) { |
2311 | /* | |
2312 | * Request was unaligned to request_alignment and therefore | |
2313 | * padded. We are going to do read-modify-write, and must | |
2314 | * serialize the request to prevent interactions of the | |
2315 | * widened region with other transactions. | |
2316 | */ | |
45e62b46 | 2317 | assert(!(flags & BDRV_REQ_NO_WAIT)); |
8ac5aab2 | 2318 | bdrv_make_request_serialising(&req, align); |
7a3f542f | 2319 | bdrv_padding_rmw_read(child, &req, &pad, false); |
61007b31 SH |
2320 | } |
2321 | ||
85c97ca7 | 2322 | ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align, |
1acc3466 | 2323 | qiov, qiov_offset, flags); |
61007b31 | 2324 | |
18743311 | 2325 | bdrv_padding_finalize(&pad); |
61007b31 | 2326 | |
9eeb6dd1 FZ |
2327 | out: |
2328 | tracked_request_end(&req); | |
99723548 | 2329 | bdrv_dec_in_flight(bs); |
7a3f542f | 2330 | |
61007b31 SH |
2331 | return ret; |
2332 | } | |
2333 | ||
a03ef88f | 2334 | int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, |
e9e52efd | 2335 | int64_t bytes, BdrvRequestFlags flags) |
61007b31 | 2336 | { |
384a48fb | 2337 | IO_CODE(); |
f5a5ca79 | 2338 | trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags); |
abaf8b75 | 2339 | assert_bdrv_graph_readable(); |
61007b31 | 2340 | |
a03ef88f | 2341 | if (!(child->bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
2342 | flags &= ~BDRV_REQ_MAY_UNMAP; |
2343 | } | |
61007b31 | 2344 | |
f5a5ca79 | 2345 | return bdrv_co_pwritev(child, offset, bytes, NULL, |
74021bc4 | 2346 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
2347 | } |
2348 | ||
4085f5c7 JS |
2349 | /* |
2350 | * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not. | |
2351 | */ | |
2352 | int bdrv_flush_all(void) | |
2353 | { | |
2354 | BdrvNextIterator it; | |
2355 | BlockDriverState *bs = NULL; | |
2356 | int result = 0; | |
2357 | ||
f791bf7f | 2358 | GLOBAL_STATE_CODE(); |
2b3912f1 | 2359 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
f791bf7f | 2360 | |
c8aa7895 PD |
2361 | /* |
2362 | * bdrv queue is managed by record/replay, | |
2363 | * creating new flush request for stopping | |
2364 | * the VM may break the determinism | |
2365 | */ | |
2366 | if (replay_events_enabled()) { | |
2367 | return result; | |
2368 | } | |
2369 | ||
4085f5c7 JS |
2370 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
2371 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2372 | int ret; | |
2373 | ||
2374 | aio_context_acquire(aio_context); | |
2375 | ret = bdrv_flush(bs); | |
2376 | if (ret < 0 && !result) { | |
2377 | result = ret; | |
2378 | } | |
2379 | aio_context_release(aio_context); | |
2380 | } | |
2381 | ||
2382 | return result; | |
2383 | } | |
2384 | ||
61007b31 SH |
2385 | /* |
2386 | * Returns the allocation status of the specified sectors. | |
2387 | * Drivers not implementing the functionality are assumed to not support | |
2388 | * backing files, hence all their sectors are reported as allocated. | |
2389 | * | |
86a3d5c6 EB |
2390 | * If 'want_zero' is true, the caller is querying for mapping |
2391 | * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and | |
2392 | * _ZERO where possible; otherwise, the result favors larger 'pnum', | |
2393 | * with a focus on accurate BDRV_BLOCK_ALLOCATED. | |
c9ce8c4d | 2394 | * |
2e8bc787 | 2395 | * If 'offset' is beyond the end of the disk image the return value is |
fb0d8654 | 2396 | * BDRV_BLOCK_EOF and 'pnum' is set to 0. |
61007b31 | 2397 | * |
2e8bc787 | 2398 | * 'bytes' is the max value 'pnum' should be set to. If bytes goes |
fb0d8654 EB |
2399 | * beyond the end of the disk image it will be clamped; if 'pnum' is set to |
2400 | * the end of the image, then the returned value will include BDRV_BLOCK_EOF. | |
67a0fd2a | 2401 | * |
2e8bc787 EB |
2402 | * 'pnum' is set to the number of bytes (including and immediately |
2403 | * following the specified offset) that are easily known to be in the | |
2404 | * same allocated/unallocated state. Note that a second call starting | |
2405 | * at the original offset plus returned pnum may have the same status. | |
2406 | * The returned value is non-zero on success except at end-of-file. | |
2407 | * | |
2408 | * Returns negative errno on failure. Otherwise, if the | |
2409 | * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are | |
2410 | * set to the host mapping and BDS corresponding to the guest offset. | |
61007b31 | 2411 | */ |
7ff9579e | 2412 | static int coroutine_fn GRAPH_RDLOCK |
b170e929 PB |
2413 | bdrv_co_do_block_status(BlockDriverState *bs, bool want_zero, |
2414 | int64_t offset, int64_t bytes, | |
2415 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
2e8bc787 EB |
2416 | { |
2417 | int64_t total_size; | |
2418 | int64_t n; /* bytes */ | |
efa6e2ed | 2419 | int ret; |
2e8bc787 | 2420 | int64_t local_map = 0; |
298a1665 | 2421 | BlockDriverState *local_file = NULL; |
efa6e2ed EB |
2422 | int64_t aligned_offset, aligned_bytes; |
2423 | uint32_t align; | |
549ec0d9 | 2424 | bool has_filtered_child; |
61007b31 | 2425 | |
298a1665 | 2426 | assert(pnum); |
7ff9579e | 2427 | assert_bdrv_graph_readable(); |
298a1665 | 2428 | *pnum = 0; |
0af02bd1 | 2429 | total_size = bdrv_co_getlength(bs); |
2e8bc787 EB |
2430 | if (total_size < 0) { |
2431 | ret = total_size; | |
298a1665 | 2432 | goto early_out; |
61007b31 SH |
2433 | } |
2434 | ||
2e8bc787 | 2435 | if (offset >= total_size) { |
298a1665 EB |
2436 | ret = BDRV_BLOCK_EOF; |
2437 | goto early_out; | |
61007b31 | 2438 | } |
2e8bc787 | 2439 | if (!bytes) { |
298a1665 EB |
2440 | ret = 0; |
2441 | goto early_out; | |
9cdcfd9f | 2442 | } |
61007b31 | 2443 | |
2e8bc787 EB |
2444 | n = total_size - offset; |
2445 | if (n < bytes) { | |
2446 | bytes = n; | |
61007b31 SH |
2447 | } |
2448 | ||
0af02bd1 | 2449 | /* Must be non-NULL or bdrv_co_getlength() would have failed */ |
d470ad42 | 2450 | assert(bs->drv); |
549ec0d9 HR |
2451 | has_filtered_child = bdrv_filter_child(bs); |
2452 | if (!bs->drv->bdrv_co_block_status && !has_filtered_child) { | |
2e8bc787 | 2453 | *pnum = bytes; |
61007b31 | 2454 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; |
2e8bc787 | 2455 | if (offset + bytes == total_size) { |
fb0d8654 EB |
2456 | ret |= BDRV_BLOCK_EOF; |
2457 | } | |
61007b31 | 2458 | if (bs->drv->protocol_name) { |
2e8bc787 EB |
2459 | ret |= BDRV_BLOCK_OFFSET_VALID; |
2460 | local_map = offset; | |
298a1665 | 2461 | local_file = bs; |
61007b31 | 2462 | } |
298a1665 | 2463 | goto early_out; |
61007b31 SH |
2464 | } |
2465 | ||
99723548 | 2466 | bdrv_inc_in_flight(bs); |
efa6e2ed EB |
2467 | |
2468 | /* Round out to request_alignment boundaries */ | |
86a3d5c6 | 2469 | align = bs->bl.request_alignment; |
efa6e2ed EB |
2470 | aligned_offset = QEMU_ALIGN_DOWN(offset, align); |
2471 | aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset; | |
2472 | ||
549ec0d9 | 2473 | if (bs->drv->bdrv_co_block_status) { |
0bc329fb HR |
2474 | /* |
2475 | * Use the block-status cache only for protocol nodes: Format | |
2476 | * drivers are generally quick to inquire the status, but protocol | |
2477 | * drivers often need to get information from outside of qemu, so | |
2478 | * we do not have control over the actual implementation. There | |
2479 | * have been cases where inquiring the status took an unreasonably | |
2480 | * long time, and we can do nothing in qemu to fix it. | |
2481 | * This is especially problematic for images with large data areas, | |
2482 | * because finding the few holes in them and giving them special | |
2483 | * treatment does not gain much performance. Therefore, we try to | |
2484 | * cache the last-identified data region. | |
2485 | * | |
2486 | * Second, limiting ourselves to protocol nodes allows us to assume | |
2487 | * the block status for data regions to be DATA | OFFSET_VALID, and | |
2488 | * that the host offset is the same as the guest offset. | |
2489 | * | |
2490 | * Note that it is possible that external writers zero parts of | |
2491 | * the cached regions without the cache being invalidated, and so | |
2492 | * we may report zeroes as data. This is not catastrophic, | |
2493 | * however, because reporting zeroes as data is fine. | |
2494 | */ | |
2495 | if (QLIST_EMPTY(&bs->children) && | |
2496 | bdrv_bsc_is_data(bs, aligned_offset, pnum)) | |
2497 | { | |
2498 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
2499 | local_file = bs; | |
2500 | local_map = aligned_offset; | |
2501 | } else { | |
2502 | ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset, | |
2503 | aligned_bytes, pnum, &local_map, | |
2504 | &local_file); | |
2505 | ||
2506 | /* | |
2507 | * Note that checking QLIST_EMPTY(&bs->children) is also done when | |
2508 | * the cache is queried above. Technically, we do not need to check | |
2509 | * it here; the worst that can happen is that we fill the cache for | |
2510 | * non-protocol nodes, and then it is never used. However, filling | |
2511 | * the cache requires an RCU update, so double check here to avoid | |
2512 | * such an update if possible. | |
113b727c HR |
2513 | * |
2514 | * Check want_zero, because we only want to update the cache when we | |
2515 | * have accurate information about what is zero and what is data. | |
0bc329fb | 2516 | */ |
113b727c HR |
2517 | if (want_zero && |
2518 | ret == (BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID) && | |
0bc329fb HR |
2519 | QLIST_EMPTY(&bs->children)) |
2520 | { | |
2521 | /* | |
2522 | * When a protocol driver reports BLOCK_OFFSET_VALID, the | |
2523 | * returned local_map value must be the same as the offset we | |
2524 | * have passed (aligned_offset), and local_bs must be the node | |
2525 | * itself. | |
2526 | * Assert this, because we follow this rule when reading from | |
2527 | * the cache (see the `local_file = bs` and | |
2528 | * `local_map = aligned_offset` assignments above), and the | |
2529 | * result the cache delivers must be the same as the driver | |
2530 | * would deliver. | |
2531 | */ | |
2532 | assert(local_file == bs); | |
2533 | assert(local_map == aligned_offset); | |
2534 | bdrv_bsc_fill(bs, aligned_offset, *pnum); | |
2535 | } | |
2536 | } | |
549ec0d9 HR |
2537 | } else { |
2538 | /* Default code for filters */ | |
2539 | ||
2540 | local_file = bdrv_filter_bs(bs); | |
2541 | assert(local_file); | |
2542 | ||
2543 | *pnum = aligned_bytes; | |
2544 | local_map = aligned_offset; | |
2545 | ret = BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; | |
2546 | } | |
636cb512 EB |
2547 | if (ret < 0) { |
2548 | *pnum = 0; | |
2549 | goto out; | |
efa6e2ed EB |
2550 | } |
2551 | ||
2e8bc787 | 2552 | /* |
636cb512 | 2553 | * The driver's result must be a non-zero multiple of request_alignment. |
efa6e2ed | 2554 | * Clamp pnum and adjust map to original request. |
2e8bc787 | 2555 | */ |
636cb512 EB |
2556 | assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) && |
2557 | align > offset - aligned_offset); | |
69f47505 VSO |
2558 | if (ret & BDRV_BLOCK_RECURSE) { |
2559 | assert(ret & BDRV_BLOCK_DATA); | |
2560 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
2561 | assert(!(ret & BDRV_BLOCK_ZERO)); | |
2562 | } | |
2563 | ||
efa6e2ed EB |
2564 | *pnum -= offset - aligned_offset; |
2565 | if (*pnum > bytes) { | |
2566 | *pnum = bytes; | |
61007b31 | 2567 | } |
2e8bc787 | 2568 | if (ret & BDRV_BLOCK_OFFSET_VALID) { |
efa6e2ed | 2569 | local_map += offset - aligned_offset; |
2e8bc787 | 2570 | } |
61007b31 SH |
2571 | |
2572 | if (ret & BDRV_BLOCK_RAW) { | |
298a1665 | 2573 | assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file); |
b170e929 PB |
2574 | ret = bdrv_co_do_block_status(local_file, want_zero, local_map, |
2575 | *pnum, pnum, &local_map, &local_file); | |
99723548 | 2576 | goto out; |
61007b31 SH |
2577 | } |
2578 | ||
2579 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
2580 | ret |= BDRV_BLOCK_ALLOCATED; | |
d40f4a56 | 2581 | } else if (bs->drv->supports_backing) { |
cb850315 HR |
2582 | BlockDriverState *cow_bs = bdrv_cow_bs(bs); |
2583 | ||
d40f4a56 AG |
2584 | if (!cow_bs) { |
2585 | ret |= BDRV_BLOCK_ZERO; | |
2586 | } else if (want_zero) { | |
0af02bd1 | 2587 | int64_t size2 = bdrv_co_getlength(cow_bs); |
c9ce8c4d | 2588 | |
2e8bc787 | 2589 | if (size2 >= 0 && offset >= size2) { |
61007b31 SH |
2590 | ret |= BDRV_BLOCK_ZERO; |
2591 | } | |
2592 | } | |
2593 | } | |
2594 | ||
69f47505 VSO |
2595 | if (want_zero && ret & BDRV_BLOCK_RECURSE && |
2596 | local_file && local_file != bs && | |
61007b31 SH |
2597 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
2598 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
2e8bc787 EB |
2599 | int64_t file_pnum; |
2600 | int ret2; | |
61007b31 | 2601 | |
b170e929 PB |
2602 | ret2 = bdrv_co_do_block_status(local_file, want_zero, local_map, |
2603 | *pnum, &file_pnum, NULL, NULL); | |
61007b31 SH |
2604 | if (ret2 >= 0) { |
2605 | /* Ignore errors. This is just providing extra information, it | |
2606 | * is useful but not necessary. | |
2607 | */ | |
c61e684e EB |
2608 | if (ret2 & BDRV_BLOCK_EOF && |
2609 | (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) { | |
2610 | /* | |
2611 | * It is valid for the format block driver to read | |
2612 | * beyond the end of the underlying file's current | |
2613 | * size; such areas read as zero. | |
2614 | */ | |
61007b31 SH |
2615 | ret |= BDRV_BLOCK_ZERO; |
2616 | } else { | |
2617 | /* Limit request to the range reported by the protocol driver */ | |
2618 | *pnum = file_pnum; | |
2619 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
2620 | } | |
2621 | } | |
2622 | } | |
2623 | ||
99723548 PB |
2624 | out: |
2625 | bdrv_dec_in_flight(bs); | |
2e8bc787 | 2626 | if (ret >= 0 && offset + *pnum == total_size) { |
fb0d8654 EB |
2627 | ret |= BDRV_BLOCK_EOF; |
2628 | } | |
298a1665 EB |
2629 | early_out: |
2630 | if (file) { | |
2631 | *file = local_file; | |
2632 | } | |
2e8bc787 EB |
2633 | if (map) { |
2634 | *map = local_map; | |
2635 | } | |
61007b31 SH |
2636 | return ret; |
2637 | } | |
2638 | ||
21c2283e | 2639 | int coroutine_fn |
f9e694cb VSO |
2640 | bdrv_co_common_block_status_above(BlockDriverState *bs, |
2641 | BlockDriverState *base, | |
3555a432 | 2642 | bool include_base, |
f9e694cb VSO |
2643 | bool want_zero, |
2644 | int64_t offset, | |
2645 | int64_t bytes, | |
2646 | int64_t *pnum, | |
2647 | int64_t *map, | |
a92b1b06 EB |
2648 | BlockDriverState **file, |
2649 | int *depth) | |
ba3f0e25 | 2650 | { |
67c095c8 | 2651 | int ret; |
ba3f0e25 | 2652 | BlockDriverState *p; |
67c095c8 | 2653 | int64_t eof = 0; |
a92b1b06 | 2654 | int dummy; |
1581a70d | 2655 | IO_CODE(); |
ba3f0e25 | 2656 | |
3555a432 | 2657 | assert(!include_base || base); /* Can't include NULL base */ |
7ff9579e | 2658 | assert_bdrv_graph_readable(); |
67c095c8 | 2659 | |
a92b1b06 EB |
2660 | if (!depth) { |
2661 | depth = &dummy; | |
2662 | } | |
2663 | *depth = 0; | |
2664 | ||
624f27bb VSO |
2665 | if (!include_base && bs == base) { |
2666 | *pnum = bytes; | |
2667 | return 0; | |
2668 | } | |
2669 | ||
b170e929 PB |
2670 | ret = bdrv_co_do_block_status(bs, want_zero, offset, bytes, pnum, |
2671 | map, file); | |
a92b1b06 | 2672 | ++*depth; |
3555a432 | 2673 | if (ret < 0 || *pnum == 0 || ret & BDRV_BLOCK_ALLOCATED || bs == base) { |
67c095c8 VSO |
2674 | return ret; |
2675 | } | |
2676 | ||
2677 | if (ret & BDRV_BLOCK_EOF) { | |
2678 | eof = offset + *pnum; | |
2679 | } | |
2680 | ||
2681 | assert(*pnum <= bytes); | |
2682 | bytes = *pnum; | |
2683 | ||
3555a432 | 2684 | for (p = bdrv_filter_or_cow_bs(bs); include_base || p != base; |
67c095c8 VSO |
2685 | p = bdrv_filter_or_cow_bs(p)) |
2686 | { | |
b170e929 PB |
2687 | ret = bdrv_co_do_block_status(p, want_zero, offset, bytes, pnum, |
2688 | map, file); | |
a92b1b06 | 2689 | ++*depth; |
c61e684e | 2690 | if (ret < 0) { |
67c095c8 | 2691 | return ret; |
c61e684e | 2692 | } |
67c095c8 | 2693 | if (*pnum == 0) { |
c61e684e | 2694 | /* |
67c095c8 VSO |
2695 | * The top layer deferred to this layer, and because this layer is |
2696 | * short, any zeroes that we synthesize beyond EOF behave as if they | |
2697 | * were allocated at this layer. | |
2698 | * | |
2699 | * We don't include BDRV_BLOCK_EOF into ret, as upper layer may be | |
2700 | * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see | |
2701 | * below. | |
c61e684e | 2702 | */ |
67c095c8 | 2703 | assert(ret & BDRV_BLOCK_EOF); |
5b648c67 | 2704 | *pnum = bytes; |
67c095c8 VSO |
2705 | if (file) { |
2706 | *file = p; | |
2707 | } | |
2708 | ret = BDRV_BLOCK_ZERO | BDRV_BLOCK_ALLOCATED; | |
2709 | break; | |
c61e684e | 2710 | } |
67c095c8 VSO |
2711 | if (ret & BDRV_BLOCK_ALLOCATED) { |
2712 | /* | |
2713 | * We've found the node and the status, we must break. | |
2714 | * | |
2715 | * Drop BDRV_BLOCK_EOF, as it's not for upper layer, which may be | |
2716 | * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see | |
2717 | * below. | |
2718 | */ | |
2719 | ret &= ~BDRV_BLOCK_EOF; | |
ba3f0e25 FZ |
2720 | break; |
2721 | } | |
67c095c8 | 2722 | |
3555a432 VSO |
2723 | if (p == base) { |
2724 | assert(include_base); | |
2725 | break; | |
2726 | } | |
2727 | ||
67c095c8 VSO |
2728 | /* |
2729 | * OK, [offset, offset + *pnum) region is unallocated on this layer, | |
2730 | * let's continue the diving. | |
2731 | */ | |
2732 | assert(*pnum <= bytes); | |
2733 | bytes = *pnum; | |
ba3f0e25 | 2734 | } |
67c095c8 VSO |
2735 | |
2736 | if (offset + *pnum == eof) { | |
2737 | ret |= BDRV_BLOCK_EOF; | |
2738 | } | |
2739 | ||
ba3f0e25 FZ |
2740 | return ret; |
2741 | } | |
2742 | ||
7b52a921 EGE |
2743 | int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs, |
2744 | BlockDriverState *base, | |
2745 | int64_t offset, int64_t bytes, | |
2746 | int64_t *pnum, int64_t *map, | |
2747 | BlockDriverState **file) | |
2748 | { | |
2749 | IO_CODE(); | |
2750 | return bdrv_co_common_block_status_above(bs, base, false, true, offset, | |
2751 | bytes, pnum, map, file, NULL); | |
2752 | } | |
2753 | ||
1b88457e PB |
2754 | int coroutine_fn bdrv_co_block_status(BlockDriverState *bs, int64_t offset, |
2755 | int64_t bytes, int64_t *pnum, | |
2756 | int64_t *map, BlockDriverState **file) | |
c9ce8c4d | 2757 | { |
384a48fb | 2758 | IO_CODE(); |
1b88457e PB |
2759 | return bdrv_co_block_status_above(bs, bdrv_filter_or_cow_bs(bs), |
2760 | offset, bytes, pnum, map, file); | |
ba3f0e25 FZ |
2761 | } |
2762 | ||
46cd1e8a AG |
2763 | /* |
2764 | * Check @bs (and its backing chain) to see if the range defined | |
2765 | * by @offset and @bytes is known to read as zeroes. | |
2766 | * Return 1 if that is the case, 0 otherwise and -errno on error. | |
2767 | * This test is meant to be fast rather than accurate so returning 0 | |
2768 | * does not guarantee non-zero data. | |
2769 | */ | |
2770 | int coroutine_fn bdrv_co_is_zero_fast(BlockDriverState *bs, int64_t offset, | |
2771 | int64_t bytes) | |
2772 | { | |
2773 | int ret; | |
2774 | int64_t pnum = bytes; | |
384a48fb | 2775 | IO_CODE(); |
46cd1e8a AG |
2776 | |
2777 | if (!bytes) { | |
2778 | return 1; | |
2779 | } | |
2780 | ||
ce47ff20 AF |
2781 | ret = bdrv_co_common_block_status_above(bs, NULL, false, false, offset, |
2782 | bytes, &pnum, NULL, NULL, NULL); | |
46cd1e8a AG |
2783 | |
2784 | if (ret < 0) { | |
2785 | return ret; | |
2786 | } | |
2787 | ||
2788 | return (pnum == bytes) && (ret & BDRV_BLOCK_ZERO); | |
2789 | } | |
2790 | ||
7b52a921 EGE |
2791 | int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t offset, |
2792 | int64_t bytes, int64_t *pnum) | |
2793 | { | |
2794 | int ret; | |
2795 | int64_t dummy; | |
2796 | IO_CODE(); | |
2797 | ||
2798 | ret = bdrv_co_common_block_status_above(bs, bs, true, false, offset, | |
2799 | bytes, pnum ? pnum : &dummy, NULL, | |
2800 | NULL, NULL); | |
2801 | if (ret < 0) { | |
2802 | return ret; | |
2803 | } | |
2804 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2805 | } | |
2806 | ||
61007b31 SH |
2807 | /* |
2808 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
2809 | * | |
a92b1b06 EB |
2810 | * Return a positive depth if (a prefix of) the given range is allocated |
2811 | * in any image between BASE and TOP (BASE is only included if include_base | |
2812 | * is set). Depth 1 is TOP, 2 is the first backing layer, and so forth. | |
170d3bd3 AS |
2813 | * BASE can be NULL to check if the given offset is allocated in any |
2814 | * image of the chain. Return 0 otherwise, or negative errno on | |
2815 | * failure. | |
61007b31 | 2816 | * |
51b0a488 EB |
2817 | * 'pnum' is set to the number of bytes (including and immediately |
2818 | * following the specified offset) that are known to be in the same | |
2819 | * allocated/unallocated state. Note that a subsequent call starting | |
2820 | * at 'offset + *pnum' may return the same allocation status (in other | |
2821 | * words, the result is not necessarily the maximum possible range); | |
2822 | * but 'pnum' will only be 0 when end of file is reached. | |
61007b31 | 2823 | */ |
578ffa9f PB |
2824 | int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *bs, |
2825 | BlockDriverState *base, | |
2826 | bool include_base, int64_t offset, | |
2827 | int64_t bytes, int64_t *pnum) | |
61007b31 | 2828 | { |
a92b1b06 | 2829 | int depth; |
7b52a921 | 2830 | int ret; |
384a48fb | 2831 | IO_CODE(); |
7b52a921 | 2832 | |
578ffa9f PB |
2833 | ret = bdrv_co_common_block_status_above(bs, base, include_base, false, |
2834 | offset, bytes, pnum, NULL, NULL, | |
2835 | &depth); | |
7e7e5100 VSO |
2836 | if (ret < 0) { |
2837 | return ret; | |
61007b31 SH |
2838 | } |
2839 | ||
a92b1b06 EB |
2840 | if (ret & BDRV_BLOCK_ALLOCATED) { |
2841 | return depth; | |
2842 | } | |
2843 | return 0; | |
61007b31 SH |
2844 | } |
2845 | ||
21c2283e | 2846 | int coroutine_fn |
b33b354f | 2847 | bdrv_co_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) |
1a8ae822 KW |
2848 | { |
2849 | BlockDriver *drv = bs->drv; | |
c4db2e25 | 2850 | BlockDriverState *child_bs = bdrv_primary_bs(bs); |
b984b296 | 2851 | int ret; |
1581a70d | 2852 | IO_CODE(); |
1b3ff9fe | 2853 | assert_bdrv_graph_readable(); |
b984b296 VSO |
2854 | |
2855 | ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); | |
2856 | if (ret < 0) { | |
2857 | return ret; | |
2858 | } | |
dc88a467 | 2859 | |
b33b354f VSO |
2860 | if (!drv) { |
2861 | return -ENOMEDIUM; | |
2862 | } | |
2863 | ||
dc88a467 | 2864 | bdrv_inc_in_flight(bs); |
1a8ae822 | 2865 | |
ca5e2ad9 EGE |
2866 | if (drv->bdrv_co_load_vmstate) { |
2867 | ret = drv->bdrv_co_load_vmstate(bs, qiov, pos); | |
c4db2e25 | 2868 | } else if (child_bs) { |
b33b354f | 2869 | ret = bdrv_co_readv_vmstate(child_bs, qiov, pos); |
b984b296 VSO |
2870 | } else { |
2871 | ret = -ENOTSUP; | |
1a8ae822 KW |
2872 | } |
2873 | ||
dc88a467 | 2874 | bdrv_dec_in_flight(bs); |
b33b354f | 2875 | |
dc88a467 | 2876 | return ret; |
1a8ae822 KW |
2877 | } |
2878 | ||
b33b354f VSO |
2879 | int coroutine_fn |
2880 | bdrv_co_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 2881 | { |
b33b354f VSO |
2882 | BlockDriver *drv = bs->drv; |
2883 | BlockDriverState *child_bs = bdrv_primary_bs(bs); | |
b984b296 | 2884 | int ret; |
1581a70d | 2885 | IO_CODE(); |
1b3ff9fe | 2886 | assert_bdrv_graph_readable(); |
b984b296 VSO |
2887 | |
2888 | ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); | |
2889 | if (ret < 0) { | |
2890 | return ret; | |
2891 | } | |
61007b31 | 2892 | |
b33b354f VSO |
2893 | if (!drv) { |
2894 | return -ENOMEDIUM; | |
b433d942 KW |
2895 | } |
2896 | ||
b33b354f | 2897 | bdrv_inc_in_flight(bs); |
61007b31 | 2898 | |
ca5e2ad9 EGE |
2899 | if (drv->bdrv_co_save_vmstate) { |
2900 | ret = drv->bdrv_co_save_vmstate(bs, qiov, pos); | |
b33b354f VSO |
2901 | } else if (child_bs) { |
2902 | ret = bdrv_co_writev_vmstate(child_bs, qiov, pos); | |
b984b296 VSO |
2903 | } else { |
2904 | ret = -ENOTSUP; | |
b33b354f VSO |
2905 | } |
2906 | ||
2907 | bdrv_dec_in_flight(bs); | |
2908 | ||
2909 | return ret; | |
61007b31 SH |
2910 | } |
2911 | ||
b33b354f | 2912 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
61007b31 | 2913 | int64_t pos, int size) |
5ddda0b8 | 2914 | { |
0d93ed08 | 2915 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b33b354f | 2916 | int ret = bdrv_writev_vmstate(bs, &qiov, pos); |
384a48fb | 2917 | IO_CODE(); |
b433d942 | 2918 | |
b33b354f | 2919 | return ret < 0 ? ret : size; |
5ddda0b8 KW |
2920 | } |
2921 | ||
b33b354f VSO |
2922 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, |
2923 | int64_t pos, int size) | |
61007b31 | 2924 | { |
b33b354f VSO |
2925 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
2926 | int ret = bdrv_readv_vmstate(bs, &qiov, pos); | |
384a48fb | 2927 | IO_CODE(); |
b33b354f VSO |
2928 | |
2929 | return ret < 0 ? ret : size; | |
61007b31 SH |
2930 | } |
2931 | ||
2932 | /**************************************************************/ | |
2933 | /* async I/Os */ | |
2934 | ||
652b0dd8 SH |
2935 | /** |
2936 | * Synchronously cancels an acb. Must be called with the BQL held and the acb | |
2937 | * must be processed with the BQL held too (IOThreads are not allowed). | |
2938 | * | |
2939 | * Use bdrv_aio_cancel_async() instead when possible. | |
2940 | */ | |
61007b31 SH |
2941 | void bdrv_aio_cancel(BlockAIOCB *acb) |
2942 | { | |
652b0dd8 | 2943 | GLOBAL_STATE_CODE(); |
61007b31 SH |
2944 | qemu_aio_ref(acb); |
2945 | bdrv_aio_cancel_async(acb); | |
652b0dd8 | 2946 | AIO_WAIT_WHILE_UNLOCKED(NULL, acb->refcnt > 1); |
61007b31 SH |
2947 | qemu_aio_unref(acb); |
2948 | } | |
2949 | ||
2950 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2951 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2952 | * In either case the completion callback must be called. */ | |
2953 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2954 | { | |
384a48fb | 2955 | IO_CODE(); |
61007b31 SH |
2956 | if (acb->aiocb_info->cancel_async) { |
2957 | acb->aiocb_info->cancel_async(acb); | |
2958 | } | |
2959 | } | |
2960 | ||
61007b31 SH |
2961 | /**************************************************************/ |
2962 | /* Coroutine block device emulation */ | |
2963 | ||
61007b31 SH |
2964 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) |
2965 | { | |
883833e2 HR |
2966 | BdrvChild *primary_child = bdrv_primary_child(bs); |
2967 | BdrvChild *child; | |
49ca6259 FZ |
2968 | int current_gen; |
2969 | int ret = 0; | |
384a48fb | 2970 | IO_CODE(); |
49ca6259 | 2971 | |
88095349 | 2972 | assert_bdrv_graph_readable(); |
49ca6259 | 2973 | bdrv_inc_in_flight(bs); |
61007b31 | 2974 | |
1e97be91 | 2975 | if (!bdrv_co_is_inserted(bs) || bdrv_is_read_only(bs) || |
1b6bc94d | 2976 | bdrv_is_sg(bs)) { |
49ca6259 | 2977 | goto early_exit; |
61007b31 SH |
2978 | } |
2979 | ||
fa9185fc | 2980 | qemu_mutex_lock(&bs->reqs_lock); |
d73415a3 | 2981 | current_gen = qatomic_read(&bs->write_gen); |
3ff2f67a EY |
2982 | |
2983 | /* Wait until any previous flushes are completed */ | |
99723548 | 2984 | while (bs->active_flush_req) { |
3783fa3d | 2985 | qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock); |
3ff2f67a EY |
2986 | } |
2987 | ||
3783fa3d | 2988 | /* Flushes reach this point in nondecreasing current_gen order. */ |
99723548 | 2989 | bs->active_flush_req = true; |
fa9185fc | 2990 | qemu_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2991 | |
c32b82af PD |
2992 | /* Write back all layers by calling one driver function */ |
2993 | if (bs->drv->bdrv_co_flush) { | |
2994 | ret = bs->drv->bdrv_co_flush(bs); | |
2995 | goto out; | |
2996 | } | |
2997 | ||
61007b31 | 2998 | /* Write back cached data to the OS even with cache=unsafe */ |
17362398 | 2999 | BLKDBG_CO_EVENT(primary_child, BLKDBG_FLUSH_TO_OS); |
61007b31 SH |
3000 | if (bs->drv->bdrv_co_flush_to_os) { |
3001 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
3002 | if (ret < 0) { | |
cdb5e315 | 3003 | goto out; |
61007b31 SH |
3004 | } |
3005 | } | |
3006 | ||
3007 | /* But don't actually force it to the disk with cache=unsafe */ | |
3008 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
883833e2 | 3009 | goto flush_children; |
61007b31 SH |
3010 | } |
3011 | ||
3ff2f67a EY |
3012 | /* Check if we really need to flush anything */ |
3013 | if (bs->flushed_gen == current_gen) { | |
883833e2 | 3014 | goto flush_children; |
3ff2f67a EY |
3015 | } |
3016 | ||
17362398 | 3017 | BLKDBG_CO_EVENT(primary_child, BLKDBG_FLUSH_TO_DISK); |
d470ad42 HR |
3018 | if (!bs->drv) { |
3019 | /* bs->drv->bdrv_co_flush() might have ejected the BDS | |
3020 | * (even in case of apparent success) */ | |
3021 | ret = -ENOMEDIUM; | |
3022 | goto out; | |
3023 | } | |
61007b31 SH |
3024 | if (bs->drv->bdrv_co_flush_to_disk) { |
3025 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
3026 | } else if (bs->drv->bdrv_aio_flush) { | |
3027 | BlockAIOCB *acb; | |
3028 | CoroutineIOCompletion co = { | |
3029 | .coroutine = qemu_coroutine_self(), | |
3030 | }; | |
3031 | ||
3032 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
3033 | if (acb == NULL) { | |
3034 | ret = -EIO; | |
3035 | } else { | |
3036 | qemu_coroutine_yield(); | |
3037 | ret = co.ret; | |
3038 | } | |
3039 | } else { | |
3040 | /* | |
3041 | * Some block drivers always operate in either writethrough or unsafe | |
3042 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
3043 | * know how the server works (because the behaviour is hardcoded or | |
3044 | * depends on server-side configuration), so we can't ensure that | |
3045 | * everything is safe on disk. Returning an error doesn't work because | |
3046 | * that would break guests even if the server operates in writethrough | |
3047 | * mode. | |
3048 | * | |
3049 | * Let's hope the user knows what he's doing. | |
3050 | */ | |
3051 | ret = 0; | |
3052 | } | |
3ff2f67a | 3053 | |
61007b31 | 3054 | if (ret < 0) { |
cdb5e315 | 3055 | goto out; |
61007b31 SH |
3056 | } |
3057 | ||
3058 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
3059 | * in the case of cache=unsafe, so there are no useless flushes. | |
3060 | */ | |
883833e2 HR |
3061 | flush_children: |
3062 | ret = 0; | |
3063 | QLIST_FOREACH(child, &bs->children, next) { | |
3064 | if (child->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
3065 | int this_child_ret = bdrv_co_flush(child->bs); | |
3066 | if (!ret) { | |
3067 | ret = this_child_ret; | |
3068 | } | |
3069 | } | |
3070 | } | |
3071 | ||
cdb5e315 | 3072 | out: |
3ff2f67a | 3073 | /* Notify any pending flushes that we have completed */ |
e6af1e08 KW |
3074 | if (ret == 0) { |
3075 | bs->flushed_gen = current_gen; | |
3076 | } | |
3783fa3d | 3077 | |
fa9185fc | 3078 | qemu_mutex_lock(&bs->reqs_lock); |
99723548 | 3079 | bs->active_flush_req = false; |
156af3ac DL |
3080 | /* Return value is ignored - it's ok if wait queue is empty */ |
3081 | qemu_co_queue_next(&bs->flush_queue); | |
fa9185fc | 3082 | qemu_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 3083 | |
49ca6259 | 3084 | early_exit: |
99723548 | 3085 | bdrv_dec_in_flight(bs); |
cdb5e315 | 3086 | return ret; |
61007b31 SH |
3087 | } |
3088 | ||
d93e5726 VSO |
3089 | int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, |
3090 | int64_t bytes) | |
61007b31 | 3091 | { |
b1066c87 | 3092 | BdrvTrackedRequest req; |
39af49c0 VSO |
3093 | int ret; |
3094 | int64_t max_pdiscard; | |
3482b9bc | 3095 | int head, tail, align; |
0b9fd3f4 | 3096 | BlockDriverState *bs = child->bs; |
384a48fb | 3097 | IO_CODE(); |
9a5a1c62 | 3098 | assert_bdrv_graph_readable(); |
61007b31 | 3099 | |
1e97be91 | 3100 | if (!bs || !bs->drv || !bdrv_co_is_inserted(bs)) { |
61007b31 SH |
3101 | return -ENOMEDIUM; |
3102 | } | |
3103 | ||
d6883bc9 VSO |
3104 | if (bdrv_has_readonly_bitmaps(bs)) { |
3105 | return -EPERM; | |
3106 | } | |
3107 | ||
69b55e03 | 3108 | ret = bdrv_check_request(offset, bytes, NULL); |
8b117001 VSO |
3109 | if (ret < 0) { |
3110 | return ret; | |
61007b31 SH |
3111 | } |
3112 | ||
61007b31 SH |
3113 | /* Do nothing if disabled. */ |
3114 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
3115 | return 0; | |
3116 | } | |
3117 | ||
02aefe43 | 3118 | if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) { |
61007b31 SH |
3119 | return 0; |
3120 | } | |
3121 | ||
0bc329fb HR |
3122 | /* Invalidate the cached block-status data range if this discard overlaps */ |
3123 | bdrv_bsc_invalidate_range(bs, offset, bytes); | |
3124 | ||
3482b9bc EB |
3125 | /* Discard is advisory, but some devices track and coalesce |
3126 | * unaligned requests, so we must pass everything down rather than | |
3127 | * round here. Still, most devices will just silently ignore | |
3128 | * unaligned requests (by returning -ENOTSUP), so we must fragment | |
3129 | * the request accordingly. */ | |
02aefe43 | 3130 | align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment); |
b8d0a980 EB |
3131 | assert(align % bs->bl.request_alignment == 0); |
3132 | head = offset % align; | |
f5a5ca79 | 3133 | tail = (offset + bytes) % align; |
9f1963b3 | 3134 | |
99723548 | 3135 | bdrv_inc_in_flight(bs); |
f5a5ca79 | 3136 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD); |
50824995 | 3137 | |
00695c27 | 3138 | ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0); |
ec050f77 DL |
3139 | if (ret < 0) { |
3140 | goto out; | |
3141 | } | |
3142 | ||
6a8f3dbb | 3143 | max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT64_MAX), |
9f1963b3 | 3144 | align); |
3482b9bc | 3145 | assert(max_pdiscard >= bs->bl.request_alignment); |
61007b31 | 3146 | |
f5a5ca79 | 3147 | while (bytes > 0) { |
d93e5726 | 3148 | int64_t num = bytes; |
3482b9bc EB |
3149 | |
3150 | if (head) { | |
3151 | /* Make small requests to get to alignment boundaries. */ | |
f5a5ca79 | 3152 | num = MIN(bytes, align - head); |
3482b9bc EB |
3153 | if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) { |
3154 | num %= bs->bl.request_alignment; | |
3155 | } | |
3156 | head = (head + num) % align; | |
3157 | assert(num < max_pdiscard); | |
3158 | } else if (tail) { | |
3159 | if (num > align) { | |
3160 | /* Shorten the request to the last aligned cluster. */ | |
3161 | num -= tail; | |
3162 | } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) && | |
3163 | tail > bs->bl.request_alignment) { | |
3164 | tail %= bs->bl.request_alignment; | |
3165 | num -= tail; | |
3166 | } | |
3167 | } | |
3168 | /* limit request size */ | |
3169 | if (num > max_pdiscard) { | |
3170 | num = max_pdiscard; | |
3171 | } | |
61007b31 | 3172 | |
d470ad42 HR |
3173 | if (!bs->drv) { |
3174 | ret = -ENOMEDIUM; | |
3175 | goto out; | |
3176 | } | |
47a5486d EB |
3177 | if (bs->drv->bdrv_co_pdiscard) { |
3178 | ret = bs->drv->bdrv_co_pdiscard(bs, offset, num); | |
61007b31 SH |
3179 | } else { |
3180 | BlockAIOCB *acb; | |
3181 | CoroutineIOCompletion co = { | |
3182 | .coroutine = qemu_coroutine_self(), | |
3183 | }; | |
3184 | ||
4da444a0 EB |
3185 | acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num, |
3186 | bdrv_co_io_em_complete, &co); | |
61007b31 | 3187 | if (acb == NULL) { |
b1066c87 FZ |
3188 | ret = -EIO; |
3189 | goto out; | |
61007b31 SH |
3190 | } else { |
3191 | qemu_coroutine_yield(); | |
3192 | ret = co.ret; | |
3193 | } | |
3194 | } | |
3195 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 3196 | goto out; |
61007b31 SH |
3197 | } |
3198 | ||
9f1963b3 | 3199 | offset += num; |
f5a5ca79 | 3200 | bytes -= num; |
61007b31 | 3201 | } |
b1066c87 FZ |
3202 | ret = 0; |
3203 | out: | |
00695c27 | 3204 | bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret); |
b1066c87 | 3205 | tracked_request_end(&req); |
99723548 | 3206 | bdrv_dec_in_flight(bs); |
b1066c87 | 3207 | return ret; |
61007b31 SH |
3208 | } |
3209 | ||
881a4c55 | 3210 | int coroutine_fn bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
3211 | { |
3212 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
3213 | CoroutineIOCompletion co = { |
3214 | .coroutine = qemu_coroutine_self(), | |
3215 | }; | |
3216 | BlockAIOCB *acb; | |
384a48fb | 3217 | IO_CODE(); |
26c518ab | 3218 | assert_bdrv_graph_readable(); |
61007b31 | 3219 | |
99723548 | 3220 | bdrv_inc_in_flight(bs); |
16a389dc | 3221 | if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) { |
5c5ae76a FZ |
3222 | co.ret = -ENOTSUP; |
3223 | goto out; | |
3224 | } | |
3225 | ||
16a389dc KW |
3226 | if (drv->bdrv_co_ioctl) { |
3227 | co.ret = drv->bdrv_co_ioctl(bs, req, buf); | |
3228 | } else { | |
3229 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
3230 | if (!acb) { | |
3231 | co.ret = -ENOTSUP; | |
3232 | goto out; | |
3233 | } | |
3234 | qemu_coroutine_yield(); | |
5c5ae76a | 3235 | } |
5c5ae76a | 3236 | out: |
99723548 | 3237 | bdrv_dec_in_flight(bs); |
5c5ae76a FZ |
3238 | return co.ret; |
3239 | } | |
3240 | ||
6d43eaa3 SL |
3241 | int coroutine_fn bdrv_co_zone_report(BlockDriverState *bs, int64_t offset, |
3242 | unsigned int *nr_zones, | |
3243 | BlockZoneDescriptor *zones) | |
3244 | { | |
3245 | BlockDriver *drv = bs->drv; | |
3246 | CoroutineIOCompletion co = { | |
3247 | .coroutine = qemu_coroutine_self(), | |
3248 | }; | |
3249 | IO_CODE(); | |
3250 | ||
3251 | bdrv_inc_in_flight(bs); | |
3252 | if (!drv || !drv->bdrv_co_zone_report || bs->bl.zoned == BLK_Z_NONE) { | |
3253 | co.ret = -ENOTSUP; | |
3254 | goto out; | |
3255 | } | |
3256 | co.ret = drv->bdrv_co_zone_report(bs, offset, nr_zones, zones); | |
3257 | out: | |
3258 | bdrv_dec_in_flight(bs); | |
3259 | return co.ret; | |
3260 | } | |
3261 | ||
3262 | int coroutine_fn bdrv_co_zone_mgmt(BlockDriverState *bs, BlockZoneOp op, | |
3263 | int64_t offset, int64_t len) | |
3264 | { | |
3265 | BlockDriver *drv = bs->drv; | |
3266 | CoroutineIOCompletion co = { | |
3267 | .coroutine = qemu_coroutine_self(), | |
3268 | }; | |
3269 | IO_CODE(); | |
3270 | ||
3271 | bdrv_inc_in_flight(bs); | |
3272 | if (!drv || !drv->bdrv_co_zone_mgmt || bs->bl.zoned == BLK_Z_NONE) { | |
3273 | co.ret = -ENOTSUP; | |
3274 | goto out; | |
3275 | } | |
3276 | co.ret = drv->bdrv_co_zone_mgmt(bs, op, offset, len); | |
3277 | out: | |
3278 | bdrv_dec_in_flight(bs); | |
3279 | return co.ret; | |
3280 | } | |
3281 | ||
4751d09a SL |
3282 | int coroutine_fn bdrv_co_zone_append(BlockDriverState *bs, int64_t *offset, |
3283 | QEMUIOVector *qiov, | |
3284 | BdrvRequestFlags flags) | |
3285 | { | |
3286 | int ret; | |
3287 | BlockDriver *drv = bs->drv; | |
3288 | CoroutineIOCompletion co = { | |
3289 | .coroutine = qemu_coroutine_self(), | |
3290 | }; | |
3291 | IO_CODE(); | |
3292 | ||
3293 | ret = bdrv_check_qiov_request(*offset, qiov->size, qiov, 0, NULL); | |
3294 | if (ret < 0) { | |
3295 | return ret; | |
3296 | } | |
3297 | ||
3298 | bdrv_inc_in_flight(bs); | |
3299 | if (!drv || !drv->bdrv_co_zone_append || bs->bl.zoned == BLK_Z_NONE) { | |
3300 | co.ret = -ENOTSUP; | |
3301 | goto out; | |
3302 | } | |
3303 | co.ret = drv->bdrv_co_zone_append(bs, offset, qiov, flags); | |
3304 | out: | |
3305 | bdrv_dec_in_flight(bs); | |
3306 | return co.ret; | |
3307 | } | |
3308 | ||
61007b31 SH |
3309 | void *qemu_blockalign(BlockDriverState *bs, size_t size) |
3310 | { | |
384a48fb | 3311 | IO_CODE(); |
61007b31 SH |
3312 | return qemu_memalign(bdrv_opt_mem_align(bs), size); |
3313 | } | |
3314 | ||
3315 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
3316 | { | |
384a48fb | 3317 | IO_CODE(); |
61007b31 SH |
3318 | return memset(qemu_blockalign(bs, size), 0, size); |
3319 | } | |
3320 | ||
3321 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
3322 | { | |
3323 | size_t align = bdrv_opt_mem_align(bs); | |
384a48fb | 3324 | IO_CODE(); |
61007b31 SH |
3325 | |
3326 | /* Ensure that NULL is never returned on success */ | |
3327 | assert(align > 0); | |
3328 | if (size == 0) { | |
3329 | size = align; | |
3330 | } | |
3331 | ||
3332 | return qemu_try_memalign(align, size); | |
3333 | } | |
3334 | ||
3335 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
3336 | { | |
3337 | void *mem = qemu_try_blockalign(bs, size); | |
384a48fb | 3338 | IO_CODE(); |
61007b31 SH |
3339 | |
3340 | if (mem) { | |
3341 | memset(mem, 0, size); | |
3342 | } | |
3343 | ||
3344 | return mem; | |
3345 | } | |
3346 | ||
f4ec04ba | 3347 | /* Helper that undoes bdrv_register_buf() when it fails partway through */ |
d9249c25 KW |
3348 | static void GRAPH_RDLOCK |
3349 | bdrv_register_buf_rollback(BlockDriverState *bs, void *host, size_t size, | |
3350 | BdrvChild *final_child) | |
f4ec04ba SH |
3351 | { |
3352 | BdrvChild *child; | |
3353 | ||
d9249c25 KW |
3354 | GLOBAL_STATE_CODE(); |
3355 | assert_bdrv_graph_readable(); | |
3356 | ||
f4ec04ba SH |
3357 | QLIST_FOREACH(child, &bs->children, next) { |
3358 | if (child == final_child) { | |
3359 | break; | |
3360 | } | |
3361 | ||
3362 | bdrv_unregister_buf(child->bs, host, size); | |
3363 | } | |
3364 | ||
3365 | if (bs->drv && bs->drv->bdrv_unregister_buf) { | |
3366 | bs->drv->bdrv_unregister_buf(bs, host, size); | |
3367 | } | |
3368 | } | |
3369 | ||
3370 | bool bdrv_register_buf(BlockDriverState *bs, void *host, size_t size, | |
3371 | Error **errp) | |
23d0ba93 FZ |
3372 | { |
3373 | BdrvChild *child; | |
3374 | ||
f791bf7f | 3375 | GLOBAL_STATE_CODE(); |
d9249c25 KW |
3376 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
3377 | ||
23d0ba93 | 3378 | if (bs->drv && bs->drv->bdrv_register_buf) { |
f4ec04ba SH |
3379 | if (!bs->drv->bdrv_register_buf(bs, host, size, errp)) { |
3380 | return false; | |
3381 | } | |
23d0ba93 FZ |
3382 | } |
3383 | QLIST_FOREACH(child, &bs->children, next) { | |
f4ec04ba SH |
3384 | if (!bdrv_register_buf(child->bs, host, size, errp)) { |
3385 | bdrv_register_buf_rollback(bs, host, size, child); | |
3386 | return false; | |
3387 | } | |
23d0ba93 | 3388 | } |
f4ec04ba | 3389 | return true; |
23d0ba93 FZ |
3390 | } |
3391 | ||
4f384011 | 3392 | void bdrv_unregister_buf(BlockDriverState *bs, void *host, size_t size) |
23d0ba93 FZ |
3393 | { |
3394 | BdrvChild *child; | |
3395 | ||
f791bf7f | 3396 | GLOBAL_STATE_CODE(); |
d9249c25 KW |
3397 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
3398 | ||
23d0ba93 | 3399 | if (bs->drv && bs->drv->bdrv_unregister_buf) { |
4f384011 | 3400 | bs->drv->bdrv_unregister_buf(bs, host, size); |
23d0ba93 FZ |
3401 | } |
3402 | QLIST_FOREACH(child, &bs->children, next) { | |
4f384011 | 3403 | bdrv_unregister_buf(child->bs, host, size); |
23d0ba93 FZ |
3404 | } |
3405 | } | |
fcc67678 | 3406 | |
abaf8b75 | 3407 | static int coroutine_fn GRAPH_RDLOCK bdrv_co_copy_range_internal( |
a5215b8f VSO |
3408 | BdrvChild *src, int64_t src_offset, BdrvChild *dst, |
3409 | int64_t dst_offset, int64_t bytes, | |
67b51fb9 VSO |
3410 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags, |
3411 | bool recurse_src) | |
fcc67678 | 3412 | { |
999658a0 | 3413 | BdrvTrackedRequest req; |
fcc67678 | 3414 | int ret; |
742bf09b | 3415 | assert_bdrv_graph_readable(); |
fcc67678 | 3416 | |
fe0480d6 KW |
3417 | /* TODO We can support BDRV_REQ_NO_FALLBACK here */ |
3418 | assert(!(read_flags & BDRV_REQ_NO_FALLBACK)); | |
3419 | assert(!(write_flags & BDRV_REQ_NO_FALLBACK)); | |
45e62b46 VSO |
3420 | assert(!(read_flags & BDRV_REQ_NO_WAIT)); |
3421 | assert(!(write_flags & BDRV_REQ_NO_WAIT)); | |
fe0480d6 | 3422 | |
1e97be91 | 3423 | if (!dst || !dst->bs || !bdrv_co_is_inserted(dst->bs)) { |
fcc67678 FZ |
3424 | return -ENOMEDIUM; |
3425 | } | |
63f4ad11 | 3426 | ret = bdrv_check_request32(dst_offset, bytes, NULL, 0); |
fcc67678 FZ |
3427 | if (ret) { |
3428 | return ret; | |
3429 | } | |
67b51fb9 VSO |
3430 | if (write_flags & BDRV_REQ_ZERO_WRITE) { |
3431 | return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags); | |
fcc67678 FZ |
3432 | } |
3433 | ||
1e97be91 | 3434 | if (!src || !src->bs || !bdrv_co_is_inserted(src->bs)) { |
d4d3e5a0 FZ |
3435 | return -ENOMEDIUM; |
3436 | } | |
63f4ad11 | 3437 | ret = bdrv_check_request32(src_offset, bytes, NULL, 0); |
d4d3e5a0 FZ |
3438 | if (ret) { |
3439 | return ret; | |
3440 | } | |
3441 | ||
fcc67678 FZ |
3442 | if (!src->bs->drv->bdrv_co_copy_range_from |
3443 | || !dst->bs->drv->bdrv_co_copy_range_to | |
3444 | || src->bs->encrypted || dst->bs->encrypted) { | |
3445 | return -ENOTSUP; | |
3446 | } | |
37aec7d7 | 3447 | |
fcc67678 | 3448 | if (recurse_src) { |
999658a0 VSO |
3449 | bdrv_inc_in_flight(src->bs); |
3450 | tracked_request_begin(&req, src->bs, src_offset, bytes, | |
3451 | BDRV_TRACKED_READ); | |
3452 | ||
09d2f948 VSO |
3453 | /* BDRV_REQ_SERIALISING is only for write operation */ |
3454 | assert(!(read_flags & BDRV_REQ_SERIALISING)); | |
c53cb427 | 3455 | bdrv_wait_serialising_requests(&req); |
999658a0 | 3456 | |
37aec7d7 FZ |
3457 | ret = src->bs->drv->bdrv_co_copy_range_from(src->bs, |
3458 | src, src_offset, | |
3459 | dst, dst_offset, | |
67b51fb9 VSO |
3460 | bytes, |
3461 | read_flags, write_flags); | |
999658a0 VSO |
3462 | |
3463 | tracked_request_end(&req); | |
3464 | bdrv_dec_in_flight(src->bs); | |
fcc67678 | 3465 | } else { |
999658a0 VSO |
3466 | bdrv_inc_in_flight(dst->bs); |
3467 | tracked_request_begin(&req, dst->bs, dst_offset, bytes, | |
3468 | BDRV_TRACKED_WRITE); | |
0eb1e891 FZ |
3469 | ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req, |
3470 | write_flags); | |
3471 | if (!ret) { | |
3472 | ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs, | |
3473 | src, src_offset, | |
3474 | dst, dst_offset, | |
3475 | bytes, | |
3476 | read_flags, write_flags); | |
3477 | } | |
3478 | bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret); | |
999658a0 VSO |
3479 | tracked_request_end(&req); |
3480 | bdrv_dec_in_flight(dst->bs); | |
fcc67678 | 3481 | } |
999658a0 | 3482 | |
37aec7d7 | 3483 | return ret; |
fcc67678 FZ |
3484 | } |
3485 | ||
3486 | /* Copy range from @src to @dst. | |
3487 | * | |
3488 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3489 | * semantics. */ | |
a5215b8f VSO |
3490 | int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, int64_t src_offset, |
3491 | BdrvChild *dst, int64_t dst_offset, | |
3492 | int64_t bytes, | |
67b51fb9 VSO |
3493 | BdrvRequestFlags read_flags, |
3494 | BdrvRequestFlags write_flags) | |
fcc67678 | 3495 | { |
967d7905 | 3496 | IO_CODE(); |
742bf09b | 3497 | assert_bdrv_graph_readable(); |
ecc983a5 FZ |
3498 | trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes, |
3499 | read_flags, write_flags); | |
fcc67678 | 3500 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3501 | bytes, read_flags, write_flags, true); |
fcc67678 FZ |
3502 | } |
3503 | ||
3504 | /* Copy range from @src to @dst. | |
3505 | * | |
3506 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3507 | * semantics. */ | |
a5215b8f VSO |
3508 | int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, int64_t src_offset, |
3509 | BdrvChild *dst, int64_t dst_offset, | |
3510 | int64_t bytes, | |
67b51fb9 VSO |
3511 | BdrvRequestFlags read_flags, |
3512 | BdrvRequestFlags write_flags) | |
fcc67678 | 3513 | { |
967d7905 | 3514 | IO_CODE(); |
742bf09b | 3515 | assert_bdrv_graph_readable(); |
ecc983a5 FZ |
3516 | trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3517 | read_flags, write_flags); | |
fcc67678 | 3518 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3519 | bytes, read_flags, write_flags, false); |
fcc67678 FZ |
3520 | } |
3521 | ||
a5215b8f VSO |
3522 | int coroutine_fn bdrv_co_copy_range(BdrvChild *src, int64_t src_offset, |
3523 | BdrvChild *dst, int64_t dst_offset, | |
3524 | int64_t bytes, BdrvRequestFlags read_flags, | |
67b51fb9 | 3525 | BdrvRequestFlags write_flags) |
fcc67678 | 3526 | { |
384a48fb | 3527 | IO_CODE(); |
742bf09b EGE |
3528 | assert_bdrv_graph_readable(); |
3529 | ||
37aec7d7 FZ |
3530 | return bdrv_co_copy_range_from(src, src_offset, |
3531 | dst, dst_offset, | |
67b51fb9 | 3532 | bytes, read_flags, write_flags); |
fcc67678 | 3533 | } |
3d9f2d2a | 3534 | |
7859c45a KW |
3535 | static void coroutine_fn GRAPH_RDLOCK |
3536 | bdrv_parent_cb_resize(BlockDriverState *bs) | |
3d9f2d2a KW |
3537 | { |
3538 | BdrvChild *c; | |
7859c45a KW |
3539 | |
3540 | assert_bdrv_graph_readable(); | |
3541 | ||
3d9f2d2a | 3542 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
bd86fb99 HR |
3543 | if (c->klass->resize) { |
3544 | c->klass->resize(c); | |
3d9f2d2a KW |
3545 | } |
3546 | } | |
3547 | } | |
3548 | ||
3549 | /** | |
3550 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
c80d8b06 HR |
3551 | * |
3552 | * If 'exact' is true, the file must be resized to exactly the given | |
3553 | * 'offset'. Otherwise, it is sufficient for the node to be at least | |
3554 | * 'offset' bytes in length. | |
3d9f2d2a | 3555 | */ |
c80d8b06 | 3556 | int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, |
7b8e4857 KW |
3557 | PreallocMode prealloc, BdrvRequestFlags flags, |
3558 | Error **errp) | |
3d9f2d2a KW |
3559 | { |
3560 | BlockDriverState *bs = child->bs; | |
23b93525 | 3561 | BdrvChild *filtered, *backing; |
3d9f2d2a | 3562 | BlockDriver *drv = bs->drv; |
1bc5f09f KW |
3563 | BdrvTrackedRequest req; |
3564 | int64_t old_size, new_bytes; | |
3d9f2d2a | 3565 | int ret; |
384a48fb | 3566 | IO_CODE(); |
c2b8e315 | 3567 | assert_bdrv_graph_readable(); |
3d9f2d2a KW |
3568 | |
3569 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ | |
3570 | if (!drv) { | |
3571 | error_setg(errp, "No medium inserted"); | |
3572 | return -ENOMEDIUM; | |
3573 | } | |
3574 | if (offset < 0) { | |
3575 | error_setg(errp, "Image size cannot be negative"); | |
3576 | return -EINVAL; | |
3577 | } | |
3578 | ||
69b55e03 | 3579 | ret = bdrv_check_request(offset, 0, errp); |
8b117001 | 3580 | if (ret < 0) { |
8b117001 VSO |
3581 | return ret; |
3582 | } | |
3583 | ||
0af02bd1 | 3584 | old_size = bdrv_co_getlength(bs); |
1bc5f09f KW |
3585 | if (old_size < 0) { |
3586 | error_setg_errno(errp, -old_size, "Failed to get old image size"); | |
3587 | return old_size; | |
3588 | } | |
3589 | ||
97efa869 EB |
3590 | if (bdrv_is_read_only(bs)) { |
3591 | error_setg(errp, "Image is read-only"); | |
3592 | return -EACCES; | |
3593 | } | |
3594 | ||
1bc5f09f KW |
3595 | if (offset > old_size) { |
3596 | new_bytes = offset - old_size; | |
3597 | } else { | |
3598 | new_bytes = 0; | |
3599 | } | |
3600 | ||
3d9f2d2a | 3601 | bdrv_inc_in_flight(bs); |
5416a11e FZ |
3602 | tracked_request_begin(&req, bs, offset - new_bytes, new_bytes, |
3603 | BDRV_TRACKED_TRUNCATE); | |
1bc5f09f KW |
3604 | |
3605 | /* If we are growing the image and potentially using preallocation for the | |
3606 | * new area, we need to make sure that no write requests are made to it | |
3607 | * concurrently or they might be overwritten by preallocation. */ | |
3608 | if (new_bytes) { | |
8ac5aab2 | 3609 | bdrv_make_request_serialising(&req, 1); |
cd47d792 | 3610 | } |
cd47d792 FZ |
3611 | ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req, |
3612 | 0); | |
3613 | if (ret < 0) { | |
3614 | error_setg_errno(errp, -ret, | |
3615 | "Failed to prepare request for truncation"); | |
3616 | goto out; | |
1bc5f09f | 3617 | } |
3d9f2d2a | 3618 | |
93393e69 | 3619 | filtered = bdrv_filter_child(bs); |
23b93525 | 3620 | backing = bdrv_cow_child(bs); |
93393e69 | 3621 | |
955c7d66 KW |
3622 | /* |
3623 | * If the image has a backing file that is large enough that it would | |
3624 | * provide data for the new area, we cannot leave it unallocated because | |
3625 | * then the backing file content would become visible. Instead, zero-fill | |
3626 | * the new area. | |
3627 | * | |
3628 | * Note that if the image has a backing file, but was opened without the | |
3629 | * backing file, taking care of keeping things consistent with that backing | |
3630 | * file is the user's responsibility. | |
3631 | */ | |
23b93525 | 3632 | if (new_bytes && backing) { |
955c7d66 KW |
3633 | int64_t backing_len; |
3634 | ||
bd53086e | 3635 | backing_len = bdrv_co_getlength(backing->bs); |
955c7d66 KW |
3636 | if (backing_len < 0) { |
3637 | ret = backing_len; | |
3638 | error_setg_errno(errp, -ret, "Could not get backing file size"); | |
3639 | goto out; | |
3640 | } | |
3641 | ||
3642 | if (backing_len > old_size) { | |
3643 | flags |= BDRV_REQ_ZERO_WRITE; | |
3644 | } | |
3645 | } | |
3646 | ||
6b7e8f8b | 3647 | if (drv->bdrv_co_truncate) { |
92b92799 KW |
3648 | if (flags & ~bs->supported_truncate_flags) { |
3649 | error_setg(errp, "Block driver does not support requested flags"); | |
3650 | ret = -ENOTSUP; | |
3651 | goto out; | |
3652 | } | |
3653 | ret = drv->bdrv_co_truncate(bs, offset, exact, prealloc, flags, errp); | |
93393e69 HR |
3654 | } else if (filtered) { |
3655 | ret = bdrv_co_truncate(filtered, offset, exact, prealloc, flags, errp); | |
6b7e8f8b | 3656 | } else { |
3d9f2d2a KW |
3657 | error_setg(errp, "Image format driver does not support resize"); |
3658 | ret = -ENOTSUP; | |
3659 | goto out; | |
3660 | } | |
3d9f2d2a KW |
3661 | if (ret < 0) { |
3662 | goto out; | |
3663 | } | |
6b7e8f8b | 3664 | |
bd53086e | 3665 | ret = bdrv_co_refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); |
3d9f2d2a KW |
3666 | if (ret < 0) { |
3667 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
3668 | } else { | |
3669 | offset = bs->total_sectors * BDRV_SECTOR_SIZE; | |
3670 | } | |
c057960c EGE |
3671 | /* |
3672 | * It's possible that truncation succeeded but bdrv_refresh_total_sectors | |
cd47d792 | 3673 | * failed, but the latter doesn't affect how we should finish the request. |
c057960c EGE |
3674 | * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. |
3675 | */ | |
cd47d792 | 3676 | bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0); |
3d9f2d2a KW |
3677 | |
3678 | out: | |
1bc5f09f | 3679 | tracked_request_end(&req); |
3d9f2d2a | 3680 | bdrv_dec_in_flight(bs); |
1bc5f09f | 3681 | |
3d9f2d2a KW |
3682 | return ret; |
3683 | } | |
bd54669a VSO |
3684 | |
3685 | void bdrv_cancel_in_flight(BlockDriverState *bs) | |
3686 | { | |
f791bf7f | 3687 | GLOBAL_STATE_CODE(); |
bd54669a VSO |
3688 | if (!bs || !bs->drv) { |
3689 | return; | |
3690 | } | |
3691 | ||
3692 | if (bs->drv->bdrv_cancel_in_flight) { | |
3693 | bs->drv->bdrv_cancel_in_flight(bs); | |
3694 | } | |
3695 | } | |
ce14f3b4 VSO |
3696 | |
3697 | int coroutine_fn | |
3698 | bdrv_co_preadv_snapshot(BdrvChild *child, int64_t offset, int64_t bytes, | |
3699 | QEMUIOVector *qiov, size_t qiov_offset) | |
3700 | { | |
3701 | BlockDriverState *bs = child->bs; | |
3702 | BlockDriver *drv = bs->drv; | |
3703 | int ret; | |
3704 | IO_CODE(); | |
7b9e8b22 | 3705 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3706 | |
3707 | if (!drv) { | |
3708 | return -ENOMEDIUM; | |
3709 | } | |
3710 | ||
3711 | if (!drv->bdrv_co_preadv_snapshot) { | |
3712 | return -ENOTSUP; | |
3713 | } | |
3714 | ||
3715 | bdrv_inc_in_flight(bs); | |
3716 | ret = drv->bdrv_co_preadv_snapshot(bs, offset, bytes, qiov, qiov_offset); | |
3717 | bdrv_dec_in_flight(bs); | |
3718 | ||
3719 | return ret; | |
3720 | } | |
3721 | ||
3722 | int coroutine_fn | |
3723 | bdrv_co_snapshot_block_status(BlockDriverState *bs, | |
3724 | bool want_zero, int64_t offset, int64_t bytes, | |
3725 | int64_t *pnum, int64_t *map, | |
3726 | BlockDriverState **file) | |
3727 | { | |
3728 | BlockDriver *drv = bs->drv; | |
3729 | int ret; | |
3730 | IO_CODE(); | |
7b9e8b22 | 3731 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3732 | |
3733 | if (!drv) { | |
3734 | return -ENOMEDIUM; | |
3735 | } | |
3736 | ||
3737 | if (!drv->bdrv_co_snapshot_block_status) { | |
3738 | return -ENOTSUP; | |
3739 | } | |
3740 | ||
3741 | bdrv_inc_in_flight(bs); | |
3742 | ret = drv->bdrv_co_snapshot_block_status(bs, want_zero, offset, bytes, | |
3743 | pnum, map, file); | |
3744 | bdrv_dec_in_flight(bs); | |
3745 | ||
3746 | return ret; | |
3747 | } | |
3748 | ||
3749 | int coroutine_fn | |
3750 | bdrv_co_pdiscard_snapshot(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
3751 | { | |
3752 | BlockDriver *drv = bs->drv; | |
3753 | int ret; | |
3754 | IO_CODE(); | |
9a5a1c62 | 3755 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3756 | |
3757 | if (!drv) { | |
3758 | return -ENOMEDIUM; | |
3759 | } | |
3760 | ||
3761 | if (!drv->bdrv_co_pdiscard_snapshot) { | |
3762 | return -ENOTSUP; | |
3763 | } | |
3764 | ||
3765 | bdrv_inc_in_flight(bs); | |
3766 | ret = drv->bdrv_co_pdiscard_snapshot(bs, offset, bytes); | |
3767 | bdrv_dec_in_flight(bs); | |
3768 | ||
3769 | return ret; | |
3770 | } |