]>
Commit | Line | Data |
---|---|---|
61007b31 SH |
1 | /* |
2 | * Block layer I/O functions | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * | |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
24 | ||
80c71a24 | 25 | #include "qemu/osdep.h" |
61007b31 | 26 | #include "trace.h" |
7f0e9da6 | 27 | #include "sysemu/block-backend.h" |
7719f3c9 | 28 | #include "block/aio-wait.h" |
61007b31 | 29 | #include "block/blockjob.h" |
f321dcb5 | 30 | #include "block/blockjob_int.h" |
61007b31 | 31 | #include "block/block_int.h" |
21c2283e | 32 | #include "block/coroutines.h" |
e2c1c34f | 33 | #include "block/dirty-bitmap.h" |
94783301 | 34 | #include "block/write-threshold.h" |
f348b6d1 | 35 | #include "qemu/cutils.h" |
5df022cf | 36 | #include "qemu/memalign.h" |
da34e65c | 37 | #include "qapi/error.h" |
d49b6836 | 38 | #include "qemu/error-report.h" |
db725815 | 39 | #include "qemu/main-loop.h" |
c8aa7895 | 40 | #include "sysemu/replay.h" |
61007b31 | 41 | |
cb2e2878 EB |
42 | /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */ |
43 | #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS) | |
44 | ||
7f8f03ef | 45 | static void bdrv_parent_cb_resize(BlockDriverState *bs); |
d05aa8bb | 46 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
5ae07b14 | 47 | int64_t offset, int64_t bytes, BdrvRequestFlags flags); |
61007b31 | 48 | |
a82a3bd1 | 49 | static void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore) |
61007b31 | 50 | { |
02d21300 | 51 | BdrvChild *c, *next; |
27ccdd52 | 52 | |
02d21300 | 53 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { |
a82a3bd1 | 54 | if (c == ignore) { |
0152bf40 KW |
55 | continue; |
56 | } | |
606ed756 | 57 | bdrv_parent_drained_begin_single(c); |
ce0f1412 PB |
58 | } |
59 | } | |
61007b31 | 60 | |
2f65df6e | 61 | void bdrv_parent_drained_end_single(BdrvChild *c) |
804db8ea | 62 | { |
ab613350 | 63 | GLOBAL_STATE_CODE(); |
2f65df6e | 64 | |
57e05be3 KW |
65 | assert(c->quiesced_parent); |
66 | c->quiesced_parent = false; | |
67 | ||
bd86fb99 | 68 | if (c->klass->drained_end) { |
2f65df6e | 69 | c->klass->drained_end(c); |
804db8ea HR |
70 | } |
71 | } | |
72 | ||
a82a3bd1 | 73 | static void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore) |
ce0f1412 | 74 | { |
61ad631c | 75 | BdrvChild *c; |
27ccdd52 | 76 | |
61ad631c | 77 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
a82a3bd1 | 78 | if (c == ignore) { |
0152bf40 KW |
79 | continue; |
80 | } | |
2f65df6e | 81 | bdrv_parent_drained_end_single(c); |
27ccdd52 | 82 | } |
61007b31 SH |
83 | } |
84 | ||
23987471 | 85 | bool bdrv_parent_drained_poll_single(BdrvChild *c) |
4be6a6d1 | 86 | { |
bd86fb99 HR |
87 | if (c->klass->drained_poll) { |
88 | return c->klass->drained_poll(c); | |
4be6a6d1 KW |
89 | } |
90 | return false; | |
91 | } | |
92 | ||
6cd5c9d7 KW |
93 | static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore, |
94 | bool ignore_bds_parents) | |
89bd0305 KW |
95 | { |
96 | BdrvChild *c, *next; | |
97 | bool busy = false; | |
98 | ||
99 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { | |
bd86fb99 | 100 | if (c == ignore || (ignore_bds_parents && c->klass->parent_is_bds)) { |
89bd0305 KW |
101 | continue; |
102 | } | |
4be6a6d1 | 103 | busy |= bdrv_parent_drained_poll_single(c); |
89bd0305 KW |
104 | } |
105 | ||
106 | return busy; | |
107 | } | |
108 | ||
606ed756 | 109 | void bdrv_parent_drained_begin_single(BdrvChild *c) |
4be6a6d1 | 110 | { |
ab613350 | 111 | GLOBAL_STATE_CODE(); |
57e05be3 KW |
112 | |
113 | assert(!c->quiesced_parent); | |
114 | c->quiesced_parent = true; | |
115 | ||
bd86fb99 HR |
116 | if (c->klass->drained_begin) { |
117 | c->klass->drained_begin(c); | |
4be6a6d1 | 118 | } |
4be6a6d1 KW |
119 | } |
120 | ||
d9e0dfa2 EB |
121 | static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) |
122 | { | |
9f460c64 AO |
123 | dst->pdiscard_alignment = MAX(dst->pdiscard_alignment, |
124 | src->pdiscard_alignment); | |
d9e0dfa2 EB |
125 | dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); |
126 | dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer); | |
24b36e98 PB |
127 | dst->max_hw_transfer = MIN_NON_ZERO(dst->max_hw_transfer, |
128 | src->max_hw_transfer); | |
d9e0dfa2 EB |
129 | dst->opt_mem_alignment = MAX(dst->opt_mem_alignment, |
130 | src->opt_mem_alignment); | |
131 | dst->min_mem_alignment = MAX(dst->min_mem_alignment, | |
132 | src->min_mem_alignment); | |
133 | dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov); | |
cc071629 | 134 | dst->max_hw_iov = MIN_NON_ZERO(dst->max_hw_iov, src->max_hw_iov); |
d9e0dfa2 EB |
135 | } |
136 | ||
1e4c797c VSO |
137 | typedef struct BdrvRefreshLimitsState { |
138 | BlockDriverState *bs; | |
139 | BlockLimits old_bl; | |
140 | } BdrvRefreshLimitsState; | |
141 | ||
142 | static void bdrv_refresh_limits_abort(void *opaque) | |
143 | { | |
144 | BdrvRefreshLimitsState *s = opaque; | |
145 | ||
146 | s->bs->bl = s->old_bl; | |
147 | } | |
148 | ||
149 | static TransactionActionDrv bdrv_refresh_limits_drv = { | |
150 | .abort = bdrv_refresh_limits_abort, | |
151 | .clean = g_free, | |
152 | }; | |
153 | ||
154 | /* @tran is allowed to be NULL, in this case no rollback is possible. */ | |
155 | void bdrv_refresh_limits(BlockDriverState *bs, Transaction *tran, Error **errp) | |
61007b31 | 156 | { |
33985614 | 157 | ERRP_GUARD(); |
61007b31 | 158 | BlockDriver *drv = bs->drv; |
66b129ac HR |
159 | BdrvChild *c; |
160 | bool have_limits; | |
61007b31 | 161 | |
f791bf7f EGE |
162 | GLOBAL_STATE_CODE(); |
163 | ||
1e4c797c VSO |
164 | if (tran) { |
165 | BdrvRefreshLimitsState *s = g_new(BdrvRefreshLimitsState, 1); | |
166 | *s = (BdrvRefreshLimitsState) { | |
167 | .bs = bs, | |
168 | .old_bl = bs->bl, | |
169 | }; | |
170 | tran_add(tran, &bdrv_refresh_limits_drv, s); | |
171 | } | |
172 | ||
61007b31 SH |
173 | memset(&bs->bl, 0, sizeof(bs->bl)); |
174 | ||
175 | if (!drv) { | |
176 | return; | |
177 | } | |
178 | ||
79ba8c98 | 179 | /* Default alignment based on whether driver has byte interface */ |
e31f6864 | 180 | bs->bl.request_alignment = (drv->bdrv_co_preadv || |
ac850bf0 VSO |
181 | drv->bdrv_aio_preadv || |
182 | drv->bdrv_co_preadv_part) ? 1 : 512; | |
79ba8c98 | 183 | |
61007b31 | 184 | /* Take some limits from the children as a default */ |
66b129ac HR |
185 | have_limits = false; |
186 | QLIST_FOREACH(c, &bs->children, next) { | |
187 | if (c->role & (BDRV_CHILD_DATA | BDRV_CHILD_FILTERED | BDRV_CHILD_COW)) | |
188 | { | |
66b129ac HR |
189 | bdrv_merge_limits(&bs->bl, &c->bs->bl); |
190 | have_limits = true; | |
61007b31 | 191 | } |
160a29e2 PB |
192 | |
193 | if (c->role & BDRV_CHILD_FILTERED) { | |
194 | bs->bl.has_variable_length |= c->bs->bl.has_variable_length; | |
195 | } | |
66b129ac HR |
196 | } |
197 | ||
198 | if (!have_limits) { | |
4196d2f0 | 199 | bs->bl.min_mem_alignment = 512; |
8e3b0cbb | 200 | bs->bl.opt_mem_alignment = qemu_real_host_page_size(); |
bd44feb7 SH |
201 | |
202 | /* Safe default since most protocols use readv()/writev()/etc */ | |
203 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
204 | } |
205 | ||
61007b31 SH |
206 | /* Then let the driver override it */ |
207 | if (drv->bdrv_refresh_limits) { | |
208 | drv->bdrv_refresh_limits(bs, errp); | |
8b117001 VSO |
209 | if (*errp) { |
210 | return; | |
211 | } | |
212 | } | |
213 | ||
214 | if (bs->bl.request_alignment > BDRV_MAX_ALIGNMENT) { | |
215 | error_setg(errp, "Driver requires too large request alignment"); | |
61007b31 SH |
216 | } |
217 | } | |
218 | ||
219 | /** | |
220 | * The copy-on-read flag is actually a reference count so multiple users may | |
221 | * use the feature without worrying about clobbering its previous state. | |
222 | * Copy-on-read stays enabled until all users have called to disable it. | |
223 | */ | |
224 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
225 | { | |
384a48fb | 226 | IO_CODE(); |
d73415a3 | 227 | qatomic_inc(&bs->copy_on_read); |
61007b31 SH |
228 | } |
229 | ||
230 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
231 | { | |
d73415a3 | 232 | int old = qatomic_fetch_dec(&bs->copy_on_read); |
384a48fb | 233 | IO_CODE(); |
d3faa13e | 234 | assert(old >= 1); |
61007b31 SH |
235 | } |
236 | ||
61124f03 PB |
237 | typedef struct { |
238 | Coroutine *co; | |
239 | BlockDriverState *bs; | |
240 | bool done; | |
481cad48 | 241 | bool begin; |
fe4f0614 | 242 | bool poll; |
0152bf40 | 243 | BdrvChild *parent; |
61124f03 PB |
244 | } BdrvCoDrainData; |
245 | ||
1cc8e54a | 246 | /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ |
299403ae KW |
247 | bool bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent, |
248 | bool ignore_bds_parents) | |
89bd0305 | 249 | { |
ab613350 | 250 | GLOBAL_STATE_CODE(); |
fe4f0614 | 251 | |
6cd5c9d7 | 252 | if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) { |
89bd0305 KW |
253 | return true; |
254 | } | |
255 | ||
d73415a3 | 256 | if (qatomic_read(&bs->in_flight)) { |
fe4f0614 KW |
257 | return true; |
258 | } | |
259 | ||
fe4f0614 | 260 | return false; |
89bd0305 KW |
261 | } |
262 | ||
299403ae | 263 | static bool bdrv_drain_poll_top_level(BlockDriverState *bs, |
89bd0305 | 264 | BdrvChild *ignore_parent) |
1cc8e54a | 265 | { |
299403ae | 266 | return bdrv_drain_poll(bs, ignore_parent, false); |
1cc8e54a KW |
267 | } |
268 | ||
299403ae | 269 | static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent, |
a82a3bd1 KW |
270 | bool poll); |
271 | static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent); | |
0152bf40 | 272 | |
a77fd4bb FZ |
273 | static void bdrv_co_drain_bh_cb(void *opaque) |
274 | { | |
275 | BdrvCoDrainData *data = opaque; | |
276 | Coroutine *co = data->co; | |
99723548 | 277 | BlockDriverState *bs = data->bs; |
a77fd4bb | 278 | |
c8ca33d0 | 279 | if (bs) { |
aa1361d5 | 280 | AioContext *ctx = bdrv_get_aio_context(bs); |
960d5fb3 | 281 | aio_context_acquire(ctx); |
c8ca33d0 KW |
282 | bdrv_dec_in_flight(bs); |
283 | if (data->begin) { | |
a82a3bd1 | 284 | bdrv_do_drained_begin(bs, data->parent, data->poll); |
c8ca33d0 | 285 | } else { |
e037c09c | 286 | assert(!data->poll); |
a82a3bd1 | 287 | bdrv_do_drained_end(bs, data->parent); |
c8ca33d0 | 288 | } |
960d5fb3 | 289 | aio_context_release(ctx); |
481cad48 | 290 | } else { |
c8ca33d0 KW |
291 | assert(data->begin); |
292 | bdrv_drain_all_begin(); | |
481cad48 MP |
293 | } |
294 | ||
a77fd4bb | 295 | data->done = true; |
1919631e | 296 | aio_co_wake(co); |
a77fd4bb FZ |
297 | } |
298 | ||
481cad48 | 299 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, |
299403ae | 300 | bool begin, |
6cd5c9d7 | 301 | BdrvChild *parent, |
2f65df6e | 302 | bool poll) |
a77fd4bb FZ |
303 | { |
304 | BdrvCoDrainData data; | |
960d5fb3 KW |
305 | Coroutine *self = qemu_coroutine_self(); |
306 | AioContext *ctx = bdrv_get_aio_context(bs); | |
307 | AioContext *co_ctx = qemu_coroutine_get_aio_context(self); | |
a77fd4bb FZ |
308 | |
309 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
c40a2545 | 310 | * other coroutines run if they were queued by aio_co_enter(). */ |
a77fd4bb FZ |
311 | |
312 | assert(qemu_in_coroutine()); | |
313 | data = (BdrvCoDrainData) { | |
960d5fb3 | 314 | .co = self, |
a77fd4bb FZ |
315 | .bs = bs, |
316 | .done = false, | |
481cad48 | 317 | .begin = begin, |
0152bf40 | 318 | .parent = parent, |
fe4f0614 | 319 | .poll = poll, |
a77fd4bb | 320 | }; |
8e1da77e | 321 | |
c8ca33d0 KW |
322 | if (bs) { |
323 | bdrv_inc_in_flight(bs); | |
324 | } | |
960d5fb3 KW |
325 | |
326 | /* | |
327 | * Temporarily drop the lock across yield or we would get deadlocks. | |
328 | * bdrv_co_drain_bh_cb() reaquires the lock as needed. | |
329 | * | |
330 | * When we yield below, the lock for the current context will be | |
331 | * released, so if this is actually the lock that protects bs, don't drop | |
332 | * it a second time. | |
333 | */ | |
334 | if (ctx != co_ctx) { | |
335 | aio_context_release(ctx); | |
336 | } | |
ab613350 SH |
337 | replay_bh_schedule_oneshot_event(qemu_get_aio_context(), |
338 | bdrv_co_drain_bh_cb, &data); | |
a77fd4bb FZ |
339 | |
340 | qemu_coroutine_yield(); | |
341 | /* If we are resumed from some other event (such as an aio completion or a | |
342 | * timer callback), it is a bug in the caller that should be fixed. */ | |
343 | assert(data.done); | |
960d5fb3 KW |
344 | |
345 | /* Reaquire the AioContext of bs if we dropped it */ | |
346 | if (ctx != co_ctx) { | |
347 | aio_context_acquire(ctx); | |
348 | } | |
a77fd4bb FZ |
349 | } |
350 | ||
05c272ff KW |
351 | static void bdrv_do_drained_begin(BlockDriverState *bs, BdrvChild *parent, |
352 | bool poll) | |
6820643f | 353 | { |
384a48fb | 354 | IO_OR_GS_CODE(); |
05c272ff KW |
355 | |
356 | if (qemu_in_coroutine()) { | |
357 | bdrv_co_yield_to_drain(bs, true, parent, poll); | |
358 | return; | |
359 | } | |
d42cf288 | 360 | |
ab613350 SH |
361 | GLOBAL_STATE_CODE(); |
362 | ||
60369b86 | 363 | /* Stop things in parent-to-child order */ |
d73415a3 | 364 | if (qatomic_fetch_inc(&bs->quiesce_counter) == 0) { |
a82a3bd1 | 365 | bdrv_parent_drained_begin(bs, parent); |
57e05be3 KW |
366 | if (bs->drv && bs->drv->bdrv_drain_begin) { |
367 | bs->drv->bdrv_drain_begin(bs); | |
368 | } | |
c7bc05f7 | 369 | } |
d30b8e64 | 370 | |
fe4f0614 KW |
371 | /* |
372 | * Wait for drained requests to finish. | |
373 | * | |
374 | * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The | |
375 | * call is needed so things in this AioContext can make progress even | |
376 | * though we don't return to the main AioContext loop - this automatically | |
377 | * includes other nodes in the same AioContext and therefore all child | |
378 | * nodes. | |
379 | */ | |
380 | if (poll) { | |
299403ae | 381 | BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, parent)); |
fe4f0614 | 382 | } |
6820643f KW |
383 | } |
384 | ||
05c272ff KW |
385 | void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, BdrvChild *parent) |
386 | { | |
387 | bdrv_do_drained_begin(bs, parent, false); | |
388 | } | |
389 | ||
0152bf40 KW |
390 | void bdrv_drained_begin(BlockDriverState *bs) |
391 | { | |
384a48fb | 392 | IO_OR_GS_CODE(); |
a82a3bd1 | 393 | bdrv_do_drained_begin(bs, NULL, true); |
0152bf40 KW |
394 | } |
395 | ||
e037c09c HR |
396 | /** |
397 | * This function does not poll, nor must any of its recursively called | |
2f65df6e | 398 | * functions. |
e037c09c | 399 | */ |
a82a3bd1 | 400 | static void bdrv_do_drained_end(BlockDriverState *bs, BdrvChild *parent) |
6820643f | 401 | { |
0f115168 KW |
402 | int old_quiesce_counter; |
403 | ||
ab613350 SH |
404 | IO_OR_GS_CODE(); |
405 | ||
481cad48 | 406 | if (qemu_in_coroutine()) { |
a82a3bd1 | 407 | bdrv_co_yield_to_drain(bs, false, parent, false); |
481cad48 MP |
408 | return; |
409 | } | |
6820643f | 410 | assert(bs->quiesce_counter > 0); |
ab613350 | 411 | GLOBAL_STATE_CODE(); |
6820643f | 412 | |
60369b86 | 413 | /* Re-enable things in child-to-parent order */ |
d73415a3 | 414 | old_quiesce_counter = qatomic_fetch_dec(&bs->quiesce_counter); |
0f115168 | 415 | if (old_quiesce_counter == 1) { |
57e05be3 KW |
416 | if (bs->drv && bs->drv->bdrv_drain_end) { |
417 | bs->drv->bdrv_drain_end(bs); | |
418 | } | |
a82a3bd1 | 419 | bdrv_parent_drained_end(bs, parent); |
0f115168 | 420 | } |
6820643f KW |
421 | } |
422 | ||
0152bf40 KW |
423 | void bdrv_drained_end(BlockDriverState *bs) |
424 | { | |
384a48fb | 425 | IO_OR_GS_CODE(); |
a82a3bd1 | 426 | bdrv_do_drained_end(bs, NULL); |
d736f119 KW |
427 | } |
428 | ||
b6e84c97 PB |
429 | void bdrv_drain(BlockDriverState *bs) |
430 | { | |
384a48fb | 431 | IO_OR_GS_CODE(); |
6820643f KW |
432 | bdrv_drained_begin(bs); |
433 | bdrv_drained_end(bs); | |
61007b31 SH |
434 | } |
435 | ||
c13ad59f KW |
436 | static void bdrv_drain_assert_idle(BlockDriverState *bs) |
437 | { | |
438 | BdrvChild *child, *next; | |
439 | ||
d73415a3 | 440 | assert(qatomic_read(&bs->in_flight) == 0); |
c13ad59f KW |
441 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
442 | bdrv_drain_assert_idle(child->bs); | |
443 | } | |
444 | } | |
445 | ||
0f12264e KW |
446 | unsigned int bdrv_drain_all_count = 0; |
447 | ||
448 | static bool bdrv_drain_all_poll(void) | |
449 | { | |
450 | BlockDriverState *bs = NULL; | |
451 | bool result = false; | |
f791bf7f | 452 | GLOBAL_STATE_CODE(); |
0f12264e | 453 | |
0f12264e KW |
454 | /* bdrv_drain_poll() can't make changes to the graph and we are holding the |
455 | * main AioContext lock, so iterating bdrv_next_all_states() is safe. */ | |
456 | while ((bs = bdrv_next_all_states(bs))) { | |
457 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
458 | aio_context_acquire(aio_context); | |
299403ae | 459 | result |= bdrv_drain_poll(bs, NULL, true); |
0f12264e KW |
460 | aio_context_release(aio_context); |
461 | } | |
462 | ||
463 | return result; | |
464 | } | |
465 | ||
61007b31 SH |
466 | /* |
467 | * Wait for pending requests to complete across all BlockDriverStates | |
468 | * | |
469 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
470 | * after calling this function. | |
c0778f66 AG |
471 | * |
472 | * This pauses all block jobs and disables external clients. It must | |
473 | * be paired with bdrv_drain_all_end(). | |
474 | * | |
475 | * NOTE: no new block jobs or BlockDriverStates can be created between | |
476 | * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls. | |
61007b31 | 477 | */ |
da0bd744 | 478 | void bdrv_drain_all_begin_nopoll(void) |
61007b31 | 479 | { |
0f12264e | 480 | BlockDriverState *bs = NULL; |
f791bf7f | 481 | GLOBAL_STATE_CODE(); |
61007b31 | 482 | |
c8aa7895 PD |
483 | /* |
484 | * bdrv queue is managed by record/replay, | |
485 | * waiting for finishing the I/O requests may | |
486 | * be infinite | |
487 | */ | |
488 | if (replay_events_enabled()) { | |
489 | return; | |
490 | } | |
491 | ||
0f12264e KW |
492 | /* AIO_WAIT_WHILE() with a NULL context can only be called from the main |
493 | * loop AioContext, so make sure we're in the main context. */ | |
9a7e86c8 | 494 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
495 | assert(bdrv_drain_all_count < INT_MAX); |
496 | bdrv_drain_all_count++; | |
9a7e86c8 | 497 | |
0f12264e KW |
498 | /* Quiesce all nodes, without polling in-flight requests yet. The graph |
499 | * cannot change during this loop. */ | |
500 | while ((bs = bdrv_next_all_states(bs))) { | |
61007b31 SH |
501 | AioContext *aio_context = bdrv_get_aio_context(bs); |
502 | ||
503 | aio_context_acquire(aio_context); | |
a82a3bd1 | 504 | bdrv_do_drained_begin(bs, NULL, false); |
61007b31 SH |
505 | aio_context_release(aio_context); |
506 | } | |
da0bd744 KW |
507 | } |
508 | ||
509 | void bdrv_drain_all_begin(void) | |
510 | { | |
511 | BlockDriverState *bs = NULL; | |
512 | ||
513 | if (qemu_in_coroutine()) { | |
514 | bdrv_co_yield_to_drain(NULL, true, NULL, true); | |
515 | return; | |
516 | } | |
517 | ||
63945789 PM |
518 | /* |
519 | * bdrv queue is managed by record/replay, | |
520 | * waiting for finishing the I/O requests may | |
521 | * be infinite | |
522 | */ | |
523 | if (replay_events_enabled()) { | |
524 | return; | |
525 | } | |
526 | ||
da0bd744 | 527 | bdrv_drain_all_begin_nopoll(); |
61007b31 | 528 | |
0f12264e | 529 | /* Now poll the in-flight requests */ |
263d5e12 | 530 | AIO_WAIT_WHILE_UNLOCKED(NULL, bdrv_drain_all_poll()); |
0f12264e KW |
531 | |
532 | while ((bs = bdrv_next_all_states(bs))) { | |
c13ad59f | 533 | bdrv_drain_assert_idle(bs); |
61007b31 | 534 | } |
c0778f66 AG |
535 | } |
536 | ||
1a6d3bd2 GK |
537 | void bdrv_drain_all_end_quiesce(BlockDriverState *bs) |
538 | { | |
b4ad82aa | 539 | GLOBAL_STATE_CODE(); |
1a6d3bd2 GK |
540 | |
541 | g_assert(bs->quiesce_counter > 0); | |
542 | g_assert(!bs->refcnt); | |
543 | ||
544 | while (bs->quiesce_counter) { | |
a82a3bd1 | 545 | bdrv_do_drained_end(bs, NULL); |
1a6d3bd2 | 546 | } |
1a6d3bd2 GK |
547 | } |
548 | ||
c0778f66 AG |
549 | void bdrv_drain_all_end(void) |
550 | { | |
0f12264e | 551 | BlockDriverState *bs = NULL; |
f791bf7f | 552 | GLOBAL_STATE_CODE(); |
c0778f66 | 553 | |
c8aa7895 PD |
554 | /* |
555 | * bdrv queue is managed by record/replay, | |
556 | * waiting for finishing the I/O requests may | |
557 | * be endless | |
558 | */ | |
559 | if (replay_events_enabled()) { | |
560 | return; | |
561 | } | |
562 | ||
0f12264e | 563 | while ((bs = bdrv_next_all_states(bs))) { |
61007b31 SH |
564 | AioContext *aio_context = bdrv_get_aio_context(bs); |
565 | ||
566 | aio_context_acquire(aio_context); | |
a82a3bd1 | 567 | bdrv_do_drained_end(bs, NULL); |
61007b31 SH |
568 | aio_context_release(aio_context); |
569 | } | |
0f12264e | 570 | |
e037c09c | 571 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
572 | assert(bdrv_drain_all_count > 0); |
573 | bdrv_drain_all_count--; | |
61007b31 SH |
574 | } |
575 | ||
c0778f66 AG |
576 | void bdrv_drain_all(void) |
577 | { | |
f791bf7f | 578 | GLOBAL_STATE_CODE(); |
c0778f66 AG |
579 | bdrv_drain_all_begin(); |
580 | bdrv_drain_all_end(); | |
581 | } | |
582 | ||
61007b31 SH |
583 | /** |
584 | * Remove an active request from the tracked requests list | |
585 | * | |
586 | * This function should be called when a tracked request is completing. | |
587 | */ | |
f0d43b1e | 588 | static void coroutine_fn tracked_request_end(BdrvTrackedRequest *req) |
61007b31 SH |
589 | { |
590 | if (req->serialising) { | |
d73415a3 | 591 | qatomic_dec(&req->bs->serialising_in_flight); |
61007b31 SH |
592 | } |
593 | ||
3783fa3d | 594 | qemu_co_mutex_lock(&req->bs->reqs_lock); |
61007b31 SH |
595 | QLIST_REMOVE(req, list); |
596 | qemu_co_queue_restart_all(&req->wait_queue); | |
3783fa3d | 597 | qemu_co_mutex_unlock(&req->bs->reqs_lock); |
61007b31 SH |
598 | } |
599 | ||
600 | /** | |
601 | * Add an active request to the tracked requests list | |
602 | */ | |
881a4c55 PB |
603 | static void coroutine_fn tracked_request_begin(BdrvTrackedRequest *req, |
604 | BlockDriverState *bs, | |
605 | int64_t offset, | |
606 | int64_t bytes, | |
607 | enum BdrvTrackedRequestType type) | |
61007b31 | 608 | { |
80247264 | 609 | bdrv_check_request(offset, bytes, &error_abort); |
22931a15 | 610 | |
61007b31 SH |
611 | *req = (BdrvTrackedRequest){ |
612 | .bs = bs, | |
613 | .offset = offset, | |
614 | .bytes = bytes, | |
ebde595c | 615 | .type = type, |
61007b31 SH |
616 | .co = qemu_coroutine_self(), |
617 | .serialising = false, | |
618 | .overlap_offset = offset, | |
619 | .overlap_bytes = bytes, | |
620 | }; | |
621 | ||
622 | qemu_co_queue_init(&req->wait_queue); | |
623 | ||
3783fa3d | 624 | qemu_co_mutex_lock(&bs->reqs_lock); |
61007b31 | 625 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
3783fa3d | 626 | qemu_co_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
627 | } |
628 | ||
3ba0e1a0 | 629 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, |
80247264 | 630 | int64_t offset, int64_t bytes) |
3ba0e1a0 | 631 | { |
80247264 EB |
632 | bdrv_check_request(offset, bytes, &error_abort); |
633 | ||
3ba0e1a0 PB |
634 | /* aaaa bbbb */ |
635 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
636 | return false; | |
637 | } | |
638 | /* bbbb aaaa */ | |
639 | if (req->overlap_offset >= offset + bytes) { | |
640 | return false; | |
641 | } | |
642 | return true; | |
643 | } | |
644 | ||
3183937f | 645 | /* Called with self->bs->reqs_lock held */ |
881a4c55 | 646 | static coroutine_fn BdrvTrackedRequest * |
3183937f VSO |
647 | bdrv_find_conflicting_request(BdrvTrackedRequest *self) |
648 | { | |
649 | BdrvTrackedRequest *req; | |
650 | ||
651 | QLIST_FOREACH(req, &self->bs->tracked_requests, list) { | |
652 | if (req == self || (!req->serialising && !self->serialising)) { | |
653 | continue; | |
654 | } | |
655 | if (tracked_request_overlaps(req, self->overlap_offset, | |
656 | self->overlap_bytes)) | |
657 | { | |
658 | /* | |
659 | * Hitting this means there was a reentrant request, for | |
660 | * example, a block driver issuing nested requests. This must | |
661 | * never happen since it means deadlock. | |
662 | */ | |
663 | assert(qemu_coroutine_self() != req->co); | |
664 | ||
665 | /* | |
666 | * If the request is already (indirectly) waiting for us, or | |
667 | * will wait for us as soon as it wakes up, then just go on | |
668 | * (instead of producing a deadlock in the former case). | |
669 | */ | |
670 | if (!req->waiting_for) { | |
671 | return req; | |
672 | } | |
673 | } | |
674 | } | |
675 | ||
676 | return NULL; | |
677 | } | |
678 | ||
ec1c8868 | 679 | /* Called with self->bs->reqs_lock held */ |
131498f7 | 680 | static void coroutine_fn |
ec1c8868 | 681 | bdrv_wait_serialising_requests_locked(BdrvTrackedRequest *self) |
3ba0e1a0 PB |
682 | { |
683 | BdrvTrackedRequest *req; | |
3ba0e1a0 | 684 | |
3183937f VSO |
685 | while ((req = bdrv_find_conflicting_request(self))) { |
686 | self->waiting_for = req; | |
ec1c8868 | 687 | qemu_co_queue_wait(&req->wait_queue, &self->bs->reqs_lock); |
3183937f | 688 | self->waiting_for = NULL; |
3183937f | 689 | } |
3ba0e1a0 PB |
690 | } |
691 | ||
8ac5aab2 VSO |
692 | /* Called with req->bs->reqs_lock held */ |
693 | static void tracked_request_set_serialising(BdrvTrackedRequest *req, | |
694 | uint64_t align) | |
61007b31 SH |
695 | { |
696 | int64_t overlap_offset = req->offset & ~(align - 1); | |
80247264 EB |
697 | int64_t overlap_bytes = |
698 | ROUND_UP(req->offset + req->bytes, align) - overlap_offset; | |
699 | ||
700 | bdrv_check_request(req->offset, req->bytes, &error_abort); | |
61007b31 SH |
701 | |
702 | if (!req->serialising) { | |
d73415a3 | 703 | qatomic_inc(&req->bs->serialising_in_flight); |
61007b31 SH |
704 | req->serialising = true; |
705 | } | |
706 | ||
707 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
708 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
09d2f948 VSO |
709 | } |
710 | ||
c28107e9 HR |
711 | /** |
712 | * Return the tracked request on @bs for the current coroutine, or | |
713 | * NULL if there is none. | |
714 | */ | |
715 | BdrvTrackedRequest *coroutine_fn bdrv_co_get_self_request(BlockDriverState *bs) | |
716 | { | |
717 | BdrvTrackedRequest *req; | |
718 | Coroutine *self = qemu_coroutine_self(); | |
967d7905 | 719 | IO_CODE(); |
c28107e9 HR |
720 | |
721 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
722 | if (req->co == self) { | |
723 | return req; | |
724 | } | |
725 | } | |
726 | ||
727 | return NULL; | |
728 | } | |
729 | ||
244483e6 KW |
730 | /** |
731 | * Round a region to cluster boundaries | |
732 | */ | |
a00e70c0 EGE |
733 | void coroutine_fn GRAPH_RDLOCK |
734 | bdrv_round_to_clusters(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
735 | int64_t *cluster_offset, int64_t *cluster_bytes) | |
244483e6 KW |
736 | { |
737 | BlockDriverInfo bdi; | |
384a48fb | 738 | IO_CODE(); |
3d47eb0a | 739 | if (bdrv_co_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { |
244483e6 KW |
740 | *cluster_offset = offset; |
741 | *cluster_bytes = bytes; | |
742 | } else { | |
743 | int64_t c = bdi.cluster_size; | |
744 | *cluster_offset = QEMU_ALIGN_DOWN(offset, c); | |
745 | *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c); | |
746 | } | |
747 | } | |
748 | ||
a00e70c0 | 749 | static int coroutine_fn GRAPH_RDLOCK bdrv_get_cluster_size(BlockDriverState *bs) |
61007b31 SH |
750 | { |
751 | BlockDriverInfo bdi; | |
752 | int ret; | |
753 | ||
3d47eb0a | 754 | ret = bdrv_co_get_info(bs, &bdi); |
61007b31 | 755 | if (ret < 0 || bdi.cluster_size == 0) { |
a5b8dd2c | 756 | return bs->bl.request_alignment; |
61007b31 SH |
757 | } else { |
758 | return bdi.cluster_size; | |
759 | } | |
760 | } | |
761 | ||
99723548 PB |
762 | void bdrv_inc_in_flight(BlockDriverState *bs) |
763 | { | |
967d7905 | 764 | IO_CODE(); |
d73415a3 | 765 | qatomic_inc(&bs->in_flight); |
99723548 PB |
766 | } |
767 | ||
c9d1a561 PB |
768 | void bdrv_wakeup(BlockDriverState *bs) |
769 | { | |
967d7905 | 770 | IO_CODE(); |
cfe29d82 | 771 | aio_wait_kick(); |
c9d1a561 PB |
772 | } |
773 | ||
99723548 PB |
774 | void bdrv_dec_in_flight(BlockDriverState *bs) |
775 | { | |
967d7905 | 776 | IO_CODE(); |
d73415a3 | 777 | qatomic_dec(&bs->in_flight); |
c9d1a561 | 778 | bdrv_wakeup(bs); |
99723548 PB |
779 | } |
780 | ||
131498f7 DL |
781 | static void coroutine_fn |
782 | bdrv_wait_serialising_requests(BdrvTrackedRequest *self) | |
61007b31 SH |
783 | { |
784 | BlockDriverState *bs = self->bs; | |
61007b31 | 785 | |
d73415a3 | 786 | if (!qatomic_read(&bs->serialising_in_flight)) { |
131498f7 | 787 | return; |
61007b31 SH |
788 | } |
789 | ||
3ba0e1a0 | 790 | qemu_co_mutex_lock(&bs->reqs_lock); |
131498f7 | 791 | bdrv_wait_serialising_requests_locked(self); |
3ba0e1a0 | 792 | qemu_co_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
793 | } |
794 | ||
131498f7 | 795 | void coroutine_fn bdrv_make_request_serialising(BdrvTrackedRequest *req, |
8ac5aab2 VSO |
796 | uint64_t align) |
797 | { | |
967d7905 | 798 | IO_CODE(); |
8ac5aab2 VSO |
799 | |
800 | qemu_co_mutex_lock(&req->bs->reqs_lock); | |
801 | ||
802 | tracked_request_set_serialising(req, align); | |
131498f7 | 803 | bdrv_wait_serialising_requests_locked(req); |
8ac5aab2 VSO |
804 | |
805 | qemu_co_mutex_unlock(&req->bs->reqs_lock); | |
8ac5aab2 VSO |
806 | } |
807 | ||
558902cc VSO |
808 | int bdrv_check_qiov_request(int64_t offset, int64_t bytes, |
809 | QEMUIOVector *qiov, size_t qiov_offset, | |
810 | Error **errp) | |
61007b31 | 811 | { |
63f4ad11 VSO |
812 | /* |
813 | * Check generic offset/bytes correctness | |
814 | */ | |
815 | ||
69b55e03 VSO |
816 | if (offset < 0) { |
817 | error_setg(errp, "offset is negative: %" PRIi64, offset); | |
818 | return -EIO; | |
819 | } | |
820 | ||
821 | if (bytes < 0) { | |
822 | error_setg(errp, "bytes is negative: %" PRIi64, bytes); | |
61007b31 SH |
823 | return -EIO; |
824 | } | |
825 | ||
8b117001 | 826 | if (bytes > BDRV_MAX_LENGTH) { |
69b55e03 VSO |
827 | error_setg(errp, "bytes(%" PRIi64 ") exceeds maximum(%" PRIi64 ")", |
828 | bytes, BDRV_MAX_LENGTH); | |
829 | return -EIO; | |
830 | } | |
831 | ||
832 | if (offset > BDRV_MAX_LENGTH) { | |
833 | error_setg(errp, "offset(%" PRIi64 ") exceeds maximum(%" PRIi64 ")", | |
834 | offset, BDRV_MAX_LENGTH); | |
8b117001 VSO |
835 | return -EIO; |
836 | } | |
837 | ||
838 | if (offset > BDRV_MAX_LENGTH - bytes) { | |
69b55e03 VSO |
839 | error_setg(errp, "sum of offset(%" PRIi64 ") and bytes(%" PRIi64 ") " |
840 | "exceeds maximum(%" PRIi64 ")", offset, bytes, | |
841 | BDRV_MAX_LENGTH); | |
8b117001 VSO |
842 | return -EIO; |
843 | } | |
844 | ||
63f4ad11 VSO |
845 | if (!qiov) { |
846 | return 0; | |
847 | } | |
848 | ||
849 | /* | |
850 | * Check qiov and qiov_offset | |
851 | */ | |
852 | ||
853 | if (qiov_offset > qiov->size) { | |
854 | error_setg(errp, "qiov_offset(%zu) overflow io vector size(%zu)", | |
855 | qiov_offset, qiov->size); | |
856 | return -EIO; | |
857 | } | |
858 | ||
859 | if (bytes > qiov->size - qiov_offset) { | |
860 | error_setg(errp, "bytes(%" PRIi64 ") + qiov_offset(%zu) overflow io " | |
861 | "vector size(%zu)", bytes, qiov_offset, qiov->size); | |
862 | return -EIO; | |
863 | } | |
864 | ||
8b117001 VSO |
865 | return 0; |
866 | } | |
867 | ||
63f4ad11 VSO |
868 | int bdrv_check_request(int64_t offset, int64_t bytes, Error **errp) |
869 | { | |
870 | return bdrv_check_qiov_request(offset, bytes, NULL, 0, errp); | |
871 | } | |
872 | ||
873 | static int bdrv_check_request32(int64_t offset, int64_t bytes, | |
874 | QEMUIOVector *qiov, size_t qiov_offset) | |
8b117001 | 875 | { |
63f4ad11 | 876 | int ret = bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, NULL); |
8b117001 VSO |
877 | if (ret < 0) { |
878 | return ret; | |
879 | } | |
880 | ||
881 | if (bytes > BDRV_REQUEST_MAX_BYTES) { | |
61007b31 SH |
882 | return -EIO; |
883 | } | |
884 | ||
885 | return 0; | |
886 | } | |
887 | ||
61007b31 | 888 | /* |
74021bc4 | 889 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
890 | * The operation is sped up by checking the block status and only writing |
891 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 892 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 893 | * BDRV_REQ_FUA). |
61007b31 | 894 | * |
f4649069 | 895 | * Returns < 0 on error, 0 on success. For error codes see bdrv_pwrite(). |
61007b31 | 896 | */ |
720ff280 | 897 | int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags) |
61007b31 | 898 | { |
237d78f8 EB |
899 | int ret; |
900 | int64_t target_size, bytes, offset = 0; | |
720ff280 | 901 | BlockDriverState *bs = child->bs; |
384a48fb | 902 | IO_CODE(); |
61007b31 | 903 | |
7286d610 EB |
904 | target_size = bdrv_getlength(bs); |
905 | if (target_size < 0) { | |
906 | return target_size; | |
61007b31 SH |
907 | } |
908 | ||
909 | for (;;) { | |
7286d610 EB |
910 | bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES); |
911 | if (bytes <= 0) { | |
61007b31 SH |
912 | return 0; |
913 | } | |
237d78f8 | 914 | ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL); |
61007b31 | 915 | if (ret < 0) { |
61007b31 SH |
916 | return ret; |
917 | } | |
918 | if (ret & BDRV_BLOCK_ZERO) { | |
237d78f8 | 919 | offset += bytes; |
61007b31 SH |
920 | continue; |
921 | } | |
237d78f8 | 922 | ret = bdrv_pwrite_zeroes(child, offset, bytes, flags); |
61007b31 | 923 | if (ret < 0) { |
61007b31 SH |
924 | return ret; |
925 | } | |
237d78f8 | 926 | offset += bytes; |
61007b31 SH |
927 | } |
928 | } | |
929 | ||
61007b31 SH |
930 | /* |
931 | * Writes to the file and ensures that no writes are reordered across this | |
932 | * request (acts as a barrier) | |
933 | * | |
934 | * Returns 0 on success, -errno in error cases. | |
935 | */ | |
e97190a4 AF |
936 | int coroutine_fn bdrv_co_pwrite_sync(BdrvChild *child, int64_t offset, |
937 | int64_t bytes, const void *buf, | |
938 | BdrvRequestFlags flags) | |
61007b31 SH |
939 | { |
940 | int ret; | |
384a48fb | 941 | IO_CODE(); |
b24a4c41 | 942 | assert_bdrv_graph_readable(); |
88095349 | 943 | |
e97190a4 | 944 | ret = bdrv_co_pwrite(child, offset, bytes, buf, flags); |
61007b31 SH |
945 | if (ret < 0) { |
946 | return ret; | |
947 | } | |
948 | ||
e97190a4 | 949 | ret = bdrv_co_flush(child->bs); |
855a6a93 KW |
950 | if (ret < 0) { |
951 | return ret; | |
61007b31 SH |
952 | } |
953 | ||
954 | return 0; | |
955 | } | |
956 | ||
08844473 KW |
957 | typedef struct CoroutineIOCompletion { |
958 | Coroutine *coroutine; | |
959 | int ret; | |
960 | } CoroutineIOCompletion; | |
961 | ||
962 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
963 | { | |
964 | CoroutineIOCompletion *co = opaque; | |
965 | ||
966 | co->ret = ret; | |
b9e413dd | 967 | aio_co_wake(co->coroutine); |
08844473 KW |
968 | } |
969 | ||
7b1fb72e KW |
970 | static int coroutine_fn GRAPH_RDLOCK |
971 | bdrv_driver_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
972 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
166fe960 KW |
973 | { |
974 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
975 | int64_t sector_num; |
976 | unsigned int nb_sectors; | |
ac850bf0 VSO |
977 | QEMUIOVector local_qiov; |
978 | int ret; | |
b9b10c35 | 979 | assert_bdrv_graph_readable(); |
3fb06697 | 980 | |
17abcbee | 981 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
e8b65355 | 982 | assert(!(flags & ~bs->supported_read_flags)); |
fa166538 | 983 | |
d470ad42 HR |
984 | if (!drv) { |
985 | return -ENOMEDIUM; | |
986 | } | |
987 | ||
ac850bf0 VSO |
988 | if (drv->bdrv_co_preadv_part) { |
989 | return drv->bdrv_co_preadv_part(bs, offset, bytes, qiov, qiov_offset, | |
990 | flags); | |
991 | } | |
992 | ||
993 | if (qiov_offset > 0 || bytes != qiov->size) { | |
994 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
995 | qiov = &local_qiov; | |
996 | } | |
997 | ||
3fb06697 | 998 | if (drv->bdrv_co_preadv) { |
ac850bf0 VSO |
999 | ret = drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); |
1000 | goto out; | |
3fb06697 KW |
1001 | } |
1002 | ||
edfab6a0 | 1003 | if (drv->bdrv_aio_preadv) { |
08844473 KW |
1004 | BlockAIOCB *acb; |
1005 | CoroutineIOCompletion co = { | |
1006 | .coroutine = qemu_coroutine_self(), | |
1007 | }; | |
1008 | ||
edfab6a0 EB |
1009 | acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags, |
1010 | bdrv_co_io_em_complete, &co); | |
08844473 | 1011 | if (acb == NULL) { |
ac850bf0 VSO |
1012 | ret = -EIO; |
1013 | goto out; | |
08844473 KW |
1014 | } else { |
1015 | qemu_coroutine_yield(); | |
ac850bf0 VSO |
1016 | ret = co.ret; |
1017 | goto out; | |
08844473 KW |
1018 | } |
1019 | } | |
edfab6a0 EB |
1020 | |
1021 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1022 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1023 | ||
1bbbf32d NS |
1024 | assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)); |
1025 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
41ae31e3 | 1026 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 EB |
1027 | assert(drv->bdrv_co_readv); |
1028 | ||
ac850bf0 VSO |
1029 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); |
1030 | ||
1031 | out: | |
1032 | if (qiov == &local_qiov) { | |
1033 | qemu_iovec_destroy(&local_qiov); | |
1034 | } | |
1035 | ||
1036 | return ret; | |
166fe960 KW |
1037 | } |
1038 | ||
7b1fb72e KW |
1039 | static int coroutine_fn GRAPH_RDLOCK |
1040 | bdrv_driver_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1041 | QEMUIOVector *qiov, size_t qiov_offset, | |
1042 | BdrvRequestFlags flags) | |
78a07294 KW |
1043 | { |
1044 | BlockDriver *drv = bs->drv; | |
e8b65355 | 1045 | bool emulate_fua = false; |
3fb06697 KW |
1046 | int64_t sector_num; |
1047 | unsigned int nb_sectors; | |
ac850bf0 | 1048 | QEMUIOVector local_qiov; |
78a07294 | 1049 | int ret; |
b9b10c35 | 1050 | assert_bdrv_graph_readable(); |
78a07294 | 1051 | |
17abcbee | 1052 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
fa166538 | 1053 | |
d470ad42 HR |
1054 | if (!drv) { |
1055 | return -ENOMEDIUM; | |
1056 | } | |
1057 | ||
e8b65355 SH |
1058 | if ((flags & BDRV_REQ_FUA) && |
1059 | (~bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1060 | flags &= ~BDRV_REQ_FUA; | |
1061 | emulate_fua = true; | |
1062 | } | |
1063 | ||
1064 | flags &= bs->supported_write_flags; | |
1065 | ||
ac850bf0 VSO |
1066 | if (drv->bdrv_co_pwritev_part) { |
1067 | ret = drv->bdrv_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, | |
e8b65355 | 1068 | flags); |
ac850bf0 VSO |
1069 | goto emulate_flags; |
1070 | } | |
1071 | ||
1072 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1073 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1074 | qiov = &local_qiov; | |
1075 | } | |
1076 | ||
3fb06697 | 1077 | if (drv->bdrv_co_pwritev) { |
e8b65355 | 1078 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, flags); |
3fb06697 KW |
1079 | goto emulate_flags; |
1080 | } | |
1081 | ||
edfab6a0 | 1082 | if (drv->bdrv_aio_pwritev) { |
08844473 KW |
1083 | BlockAIOCB *acb; |
1084 | CoroutineIOCompletion co = { | |
1085 | .coroutine = qemu_coroutine_self(), | |
1086 | }; | |
1087 | ||
e8b65355 | 1088 | acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov, flags, |
edfab6a0 | 1089 | bdrv_co_io_em_complete, &co); |
08844473 | 1090 | if (acb == NULL) { |
3fb06697 | 1091 | ret = -EIO; |
08844473 KW |
1092 | } else { |
1093 | qemu_coroutine_yield(); | |
3fb06697 | 1094 | ret = co.ret; |
08844473 | 1095 | } |
edfab6a0 EB |
1096 | goto emulate_flags; |
1097 | } | |
1098 | ||
1099 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1100 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1101 | ||
1bbbf32d NS |
1102 | assert(QEMU_IS_ALIGNED(offset, BDRV_SECTOR_SIZE)); |
1103 | assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); | |
41ae31e3 | 1104 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 | 1105 | |
e18a58b4 | 1106 | assert(drv->bdrv_co_writev); |
e8b65355 | 1107 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov, flags); |
78a07294 | 1108 | |
3fb06697 | 1109 | emulate_flags: |
e8b65355 | 1110 | if (ret == 0 && emulate_fua) { |
78a07294 KW |
1111 | ret = bdrv_co_flush(bs); |
1112 | } | |
1113 | ||
ac850bf0 VSO |
1114 | if (qiov == &local_qiov) { |
1115 | qemu_iovec_destroy(&local_qiov); | |
1116 | } | |
1117 | ||
78a07294 KW |
1118 | return ret; |
1119 | } | |
1120 | ||
7b1fb72e | 1121 | static int coroutine_fn GRAPH_RDLOCK |
17abcbee VSO |
1122 | bdrv_driver_pwritev_compressed(BlockDriverState *bs, int64_t offset, |
1123 | int64_t bytes, QEMUIOVector *qiov, | |
ac850bf0 | 1124 | size_t qiov_offset) |
29a298af PB |
1125 | { |
1126 | BlockDriver *drv = bs->drv; | |
ac850bf0 VSO |
1127 | QEMUIOVector local_qiov; |
1128 | int ret; | |
b9b10c35 | 1129 | assert_bdrv_graph_readable(); |
29a298af | 1130 | |
17abcbee VSO |
1131 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1132 | ||
d470ad42 HR |
1133 | if (!drv) { |
1134 | return -ENOMEDIUM; | |
1135 | } | |
1136 | ||
ac850bf0 | 1137 | if (!block_driver_can_compress(drv)) { |
29a298af PB |
1138 | return -ENOTSUP; |
1139 | } | |
1140 | ||
ac850bf0 VSO |
1141 | if (drv->bdrv_co_pwritev_compressed_part) { |
1142 | return drv->bdrv_co_pwritev_compressed_part(bs, offset, bytes, | |
1143 | qiov, qiov_offset); | |
1144 | } | |
1145 | ||
1146 | if (qiov_offset == 0) { | |
1147 | return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov); | |
1148 | } | |
1149 | ||
1150 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1151 | ret = drv->bdrv_co_pwritev_compressed(bs, offset, bytes, &local_qiov); | |
1152 | qemu_iovec_destroy(&local_qiov); | |
1153 | ||
1154 | return ret; | |
29a298af PB |
1155 | } |
1156 | ||
7b1fb72e KW |
1157 | static int coroutine_fn GRAPH_RDLOCK |
1158 | bdrv_co_do_copy_on_readv(BdrvChild *child, int64_t offset, int64_t bytes, | |
1159 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
61007b31 | 1160 | { |
85c97ca7 KW |
1161 | BlockDriverState *bs = child->bs; |
1162 | ||
61007b31 SH |
1163 | /* Perform I/O through a temporary buffer so that users who scribble over |
1164 | * their read buffer while the operation is in progress do not end up | |
1165 | * modifying the image file. This is critical for zero-copy guest I/O | |
1166 | * where anything might happen inside guest memory. | |
1167 | */ | |
2275cc90 | 1168 | void *bounce_buffer = NULL; |
61007b31 SH |
1169 | |
1170 | BlockDriver *drv = bs->drv; | |
244483e6 | 1171 | int64_t cluster_offset; |
7cfd5275 | 1172 | int64_t cluster_bytes; |
9df5afbd | 1173 | int64_t skip_bytes; |
61007b31 | 1174 | int ret; |
cb2e2878 EB |
1175 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
1176 | BDRV_REQUEST_MAX_BYTES); | |
9df5afbd | 1177 | int64_t progress = 0; |
8644476e | 1178 | bool skip_write; |
61007b31 | 1179 | |
9df5afbd VSO |
1180 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1181 | ||
d470ad42 HR |
1182 | if (!drv) { |
1183 | return -ENOMEDIUM; | |
1184 | } | |
1185 | ||
8644476e HR |
1186 | /* |
1187 | * Do not write anything when the BDS is inactive. That is not | |
1188 | * allowed, and it would not help. | |
1189 | */ | |
1190 | skip_write = (bs->open_flags & BDRV_O_INACTIVE); | |
1191 | ||
1bf03e66 KW |
1192 | /* FIXME We cannot require callers to have write permissions when all they |
1193 | * are doing is a read request. If we did things right, write permissions | |
1194 | * would be obtained anyway, but internally by the copy-on-read code. As | |
765d9df9 | 1195 | * long as it is implemented here rather than in a separate filter driver, |
1bf03e66 KW |
1196 | * the copy-on-read code doesn't have its own BdrvChild, however, for which |
1197 | * it could request permissions. Therefore we have to bypass the permission | |
1198 | * system for the moment. */ | |
1199 | // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
afa4b293 | 1200 | |
61007b31 | 1201 | /* Cover entire cluster so no additional backing file I/O is required when |
cb2e2878 EB |
1202 | * allocating cluster in the image file. Note that this value may exceed |
1203 | * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which | |
1204 | * is one reason we loop rather than doing it all at once. | |
61007b31 | 1205 | */ |
244483e6 | 1206 | bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes); |
cb2e2878 | 1207 | skip_bytes = offset - cluster_offset; |
61007b31 | 1208 | |
244483e6 KW |
1209 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
1210 | cluster_offset, cluster_bytes); | |
61007b31 | 1211 | |
cb2e2878 EB |
1212 | while (cluster_bytes) { |
1213 | int64_t pnum; | |
61007b31 | 1214 | |
8644476e HR |
1215 | if (skip_write) { |
1216 | ret = 1; /* "already allocated", so nothing will be copied */ | |
cb2e2878 | 1217 | pnum = MIN(cluster_bytes, max_transfer); |
8644476e HR |
1218 | } else { |
1219 | ret = bdrv_is_allocated(bs, cluster_offset, | |
1220 | MIN(cluster_bytes, max_transfer), &pnum); | |
1221 | if (ret < 0) { | |
1222 | /* | |
1223 | * Safe to treat errors in querying allocation as if | |
1224 | * unallocated; we'll probably fail again soon on the | |
1225 | * read, but at least that will set a decent errno. | |
1226 | */ | |
1227 | pnum = MIN(cluster_bytes, max_transfer); | |
1228 | } | |
61007b31 | 1229 | |
8644476e HR |
1230 | /* Stop at EOF if the image ends in the middle of the cluster */ |
1231 | if (ret == 0 && pnum == 0) { | |
1232 | assert(progress >= bytes); | |
1233 | break; | |
1234 | } | |
b0ddcbbb | 1235 | |
8644476e HR |
1236 | assert(skip_bytes < pnum); |
1237 | } | |
61007b31 | 1238 | |
cb2e2878 | 1239 | if (ret <= 0) { |
1143ec5e VSO |
1240 | QEMUIOVector local_qiov; |
1241 | ||
cb2e2878 | 1242 | /* Must copy-on-read; use the bounce buffer */ |
0d93ed08 | 1243 | pnum = MIN(pnum, MAX_BOUNCE_BUFFER); |
2275cc90 VSO |
1244 | if (!bounce_buffer) { |
1245 | int64_t max_we_need = MAX(pnum, cluster_bytes - pnum); | |
1246 | int64_t max_allowed = MIN(max_transfer, MAX_BOUNCE_BUFFER); | |
1247 | int64_t bounce_buffer_len = MIN(max_we_need, max_allowed); | |
1248 | ||
1249 | bounce_buffer = qemu_try_blockalign(bs, bounce_buffer_len); | |
1250 | if (!bounce_buffer) { | |
1251 | ret = -ENOMEM; | |
1252 | goto err; | |
1253 | } | |
1254 | } | |
0d93ed08 | 1255 | qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum); |
61007b31 | 1256 | |
cb2e2878 | 1257 | ret = bdrv_driver_preadv(bs, cluster_offset, pnum, |
ac850bf0 | 1258 | &local_qiov, 0, 0); |
cb2e2878 EB |
1259 | if (ret < 0) { |
1260 | goto err; | |
1261 | } | |
1262 | ||
c834dc05 | 1263 | bdrv_co_debug_event(bs, BLKDBG_COR_WRITE); |
cb2e2878 EB |
1264 | if (drv->bdrv_co_pwrite_zeroes && |
1265 | buffer_is_zero(bounce_buffer, pnum)) { | |
1266 | /* FIXME: Should we (perhaps conditionally) be setting | |
1267 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
1268 | * that still correctly reads as zero? */ | |
7adcf59f HR |
1269 | ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, pnum, |
1270 | BDRV_REQ_WRITE_UNCHANGED); | |
cb2e2878 EB |
1271 | } else { |
1272 | /* This does not change the data on the disk, it is not | |
1273 | * necessary to flush even in cache=writethrough mode. | |
1274 | */ | |
1275 | ret = bdrv_driver_pwritev(bs, cluster_offset, pnum, | |
ac850bf0 | 1276 | &local_qiov, 0, |
7adcf59f | 1277 | BDRV_REQ_WRITE_UNCHANGED); |
cb2e2878 EB |
1278 | } |
1279 | ||
1280 | if (ret < 0) { | |
1281 | /* It might be okay to ignore write errors for guest | |
1282 | * requests. If this is a deliberate copy-on-read | |
1283 | * then we don't want to ignore the error. Simply | |
1284 | * report it in all cases. | |
1285 | */ | |
1286 | goto err; | |
1287 | } | |
1288 | ||
3299e5ec | 1289 | if (!(flags & BDRV_REQ_PREFETCH)) { |
1143ec5e VSO |
1290 | qemu_iovec_from_buf(qiov, qiov_offset + progress, |
1291 | bounce_buffer + skip_bytes, | |
4ab78b19 | 1292 | MIN(pnum - skip_bytes, bytes - progress)); |
3299e5ec VSO |
1293 | } |
1294 | } else if (!(flags & BDRV_REQ_PREFETCH)) { | |
cb2e2878 | 1295 | /* Read directly into the destination */ |
1143ec5e VSO |
1296 | ret = bdrv_driver_preadv(bs, offset + progress, |
1297 | MIN(pnum - skip_bytes, bytes - progress), | |
1298 | qiov, qiov_offset + progress, 0); | |
cb2e2878 EB |
1299 | if (ret < 0) { |
1300 | goto err; | |
1301 | } | |
1302 | } | |
1303 | ||
1304 | cluster_offset += pnum; | |
1305 | cluster_bytes -= pnum; | |
1306 | progress += pnum - skip_bytes; | |
1307 | skip_bytes = 0; | |
1308 | } | |
1309 | ret = 0; | |
61007b31 SH |
1310 | |
1311 | err: | |
1312 | qemu_vfree(bounce_buffer); | |
1313 | return ret; | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * Forwards an already correctly aligned request to the BlockDriver. This | |
1a62d0ac EB |
1318 | * handles copy on read, zeroing after EOF, and fragmentation of large |
1319 | * reads; any other features must be implemented by the caller. | |
61007b31 | 1320 | */ |
7b1fb72e KW |
1321 | static int coroutine_fn GRAPH_RDLOCK |
1322 | bdrv_aligned_preadv(BdrvChild *child, BdrvTrackedRequest *req, | |
1323 | int64_t offset, int64_t bytes, int64_t align, | |
1324 | QEMUIOVector *qiov, size_t qiov_offset, int flags) | |
61007b31 | 1325 | { |
85c97ca7 | 1326 | BlockDriverState *bs = child->bs; |
c9d20029 | 1327 | int64_t total_bytes, max_bytes; |
1a62d0ac | 1328 | int ret = 0; |
8b0c5d76 | 1329 | int64_t bytes_remaining = bytes; |
1a62d0ac | 1330 | int max_transfer; |
61007b31 | 1331 | |
8b0c5d76 | 1332 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
49c07526 KW |
1333 | assert(is_power_of_2(align)); |
1334 | assert((offset & (align - 1)) == 0); | |
1335 | assert((bytes & (align - 1)) == 0); | |
abb06c5a | 1336 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
1a62d0ac EB |
1337 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1338 | align); | |
a604fa2b | 1339 | |
e8b65355 SH |
1340 | /* |
1341 | * TODO: We would need a per-BDS .supported_read_flags and | |
a604fa2b EB |
1342 | * potential fallback support, if we ever implement any read flags |
1343 | * to pass through to drivers. For now, there aren't any | |
e8b65355 SH |
1344 | * passthrough flags except the BDRV_REQ_REGISTERED_BUF optimization hint. |
1345 | */ | |
1346 | assert(!(flags & ~(BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH | | |
1347 | BDRV_REQ_REGISTERED_BUF))); | |
61007b31 SH |
1348 | |
1349 | /* Handle Copy on Read and associated serialisation */ | |
1350 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
1351 | /* If we touch the same cluster it counts as an overlap. This | |
1352 | * guarantees that allocating writes will be serialized and not race | |
1353 | * with each other for the same cluster. For example, in copy-on-read | |
1354 | * it ensures that the CoR read and write operations are atomic and | |
1355 | * guest writes cannot interleave between them. */ | |
8ac5aab2 | 1356 | bdrv_make_request_serialising(req, bdrv_get_cluster_size(bs)); |
18fbd0de PB |
1357 | } else { |
1358 | bdrv_wait_serialising_requests(req); | |
61007b31 SH |
1359 | } |
1360 | ||
61007b31 | 1361 | if (flags & BDRV_REQ_COPY_ON_READ) { |
d6a644bb | 1362 | int64_t pnum; |
61007b31 | 1363 | |
897dd0ec AS |
1364 | /* The flag BDRV_REQ_COPY_ON_READ has reached its addressee */ |
1365 | flags &= ~BDRV_REQ_COPY_ON_READ; | |
1366 | ||
88e63df2 | 1367 | ret = bdrv_is_allocated(bs, offset, bytes, &pnum); |
61007b31 SH |
1368 | if (ret < 0) { |
1369 | goto out; | |
1370 | } | |
1371 | ||
88e63df2 | 1372 | if (!ret || pnum != bytes) { |
65cd4424 VSO |
1373 | ret = bdrv_co_do_copy_on_readv(child, offset, bytes, |
1374 | qiov, qiov_offset, flags); | |
3299e5ec VSO |
1375 | goto out; |
1376 | } else if (flags & BDRV_REQ_PREFETCH) { | |
61007b31 SH |
1377 | goto out; |
1378 | } | |
1379 | } | |
1380 | ||
1a62d0ac | 1381 | /* Forward the request to the BlockDriver, possibly fragmenting it */ |
c9d20029 KW |
1382 | total_bytes = bdrv_getlength(bs); |
1383 | if (total_bytes < 0) { | |
1384 | ret = total_bytes; | |
1385 | goto out; | |
1386 | } | |
61007b31 | 1387 | |
e8b65355 | 1388 | assert(!(flags & ~(bs->supported_read_flags | BDRV_REQ_REGISTERED_BUF))); |
897dd0ec | 1389 | |
c9d20029 | 1390 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
1a62d0ac | 1391 | if (bytes <= max_bytes && bytes <= max_transfer) { |
897dd0ec | 1392 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, qiov_offset, flags); |
1a62d0ac EB |
1393 | goto out; |
1394 | } | |
61007b31 | 1395 | |
1a62d0ac | 1396 | while (bytes_remaining) { |
8b0c5d76 | 1397 | int64_t num; |
61007b31 | 1398 | |
1a62d0ac | 1399 | if (max_bytes) { |
1a62d0ac EB |
1400 | num = MIN(bytes_remaining, MIN(max_bytes, max_transfer)); |
1401 | assert(num); | |
61007b31 | 1402 | |
1a62d0ac | 1403 | ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining, |
134b7dec | 1404 | num, qiov, |
897dd0ec AS |
1405 | qiov_offset + bytes - bytes_remaining, |
1406 | flags); | |
1a62d0ac | 1407 | max_bytes -= num; |
1a62d0ac EB |
1408 | } else { |
1409 | num = bytes_remaining; | |
134b7dec HR |
1410 | ret = qemu_iovec_memset(qiov, qiov_offset + bytes - bytes_remaining, |
1411 | 0, bytes_remaining); | |
1a62d0ac EB |
1412 | } |
1413 | if (ret < 0) { | |
1414 | goto out; | |
1415 | } | |
1416 | bytes_remaining -= num; | |
61007b31 SH |
1417 | } |
1418 | ||
1419 | out: | |
1a62d0ac | 1420 | return ret < 0 ? ret : 0; |
61007b31 SH |
1421 | } |
1422 | ||
61007b31 | 1423 | /* |
7a3f542f VSO |
1424 | * Request padding |
1425 | * | |
1426 | * |<---- align ----->| |<----- align ---->| | |
1427 | * |<- head ->|<------------- bytes ------------->|<-- tail -->| | |
1428 | * | | | | | | | |
1429 | * -*----------$-------*-------- ... --------*-----$------------*--- | |
1430 | * | | | | | | | |
1431 | * | offset | | end | | |
1432 | * ALIGN_DOWN(offset) ALIGN_UP(offset) ALIGN_DOWN(end) ALIGN_UP(end) | |
1433 | * [buf ... ) [tail_buf ) | |
1434 | * | |
1435 | * @buf is an aligned allocation needed to store @head and @tail paddings. @head | |
1436 | * is placed at the beginning of @buf and @tail at the @end. | |
1437 | * | |
1438 | * @tail_buf is a pointer to sub-buffer, corresponding to align-sized chunk | |
1439 | * around tail, if tail exists. | |
1440 | * | |
1441 | * @merge_reads is true for small requests, | |
1442 | * if @buf_len == @head + bytes + @tail. In this case it is possible that both | |
1443 | * head and tail exist but @buf_len == align and @tail_buf == @buf. | |
1444 | */ | |
1445 | typedef struct BdrvRequestPadding { | |
1446 | uint8_t *buf; | |
1447 | size_t buf_len; | |
1448 | uint8_t *tail_buf; | |
1449 | size_t head; | |
1450 | size_t tail; | |
1451 | bool merge_reads; | |
1452 | QEMUIOVector local_qiov; | |
1453 | } BdrvRequestPadding; | |
1454 | ||
1455 | static bool bdrv_init_padding(BlockDriverState *bs, | |
1456 | int64_t offset, int64_t bytes, | |
1457 | BdrvRequestPadding *pad) | |
1458 | { | |
a56ed80c VSO |
1459 | int64_t align = bs->bl.request_alignment; |
1460 | int64_t sum; | |
1461 | ||
1462 | bdrv_check_request(offset, bytes, &error_abort); | |
1463 | assert(align <= INT_MAX); /* documented in block/block_int.h */ | |
1464 | assert(align <= SIZE_MAX / 2); /* so we can allocate the buffer */ | |
7a3f542f VSO |
1465 | |
1466 | memset(pad, 0, sizeof(*pad)); | |
1467 | ||
1468 | pad->head = offset & (align - 1); | |
1469 | pad->tail = ((offset + bytes) & (align - 1)); | |
1470 | if (pad->tail) { | |
1471 | pad->tail = align - pad->tail; | |
1472 | } | |
1473 | ||
ac9d00bf | 1474 | if (!pad->head && !pad->tail) { |
7a3f542f VSO |
1475 | return false; |
1476 | } | |
1477 | ||
ac9d00bf VSO |
1478 | assert(bytes); /* Nothing good in aligning zero-length requests */ |
1479 | ||
7a3f542f VSO |
1480 | sum = pad->head + bytes + pad->tail; |
1481 | pad->buf_len = (sum > align && pad->head && pad->tail) ? 2 * align : align; | |
1482 | pad->buf = qemu_blockalign(bs, pad->buf_len); | |
1483 | pad->merge_reads = sum == pad->buf_len; | |
1484 | if (pad->tail) { | |
1485 | pad->tail_buf = pad->buf + pad->buf_len - align; | |
1486 | } | |
1487 | ||
1488 | return true; | |
1489 | } | |
1490 | ||
7b1fb72e KW |
1491 | static int coroutine_fn GRAPH_RDLOCK |
1492 | bdrv_padding_rmw_read(BdrvChild *child, BdrvTrackedRequest *req, | |
1493 | BdrvRequestPadding *pad, bool zero_middle) | |
7a3f542f VSO |
1494 | { |
1495 | QEMUIOVector local_qiov; | |
1496 | BlockDriverState *bs = child->bs; | |
1497 | uint64_t align = bs->bl.request_alignment; | |
1498 | int ret; | |
1499 | ||
1500 | assert(req->serialising && pad->buf); | |
1501 | ||
1502 | if (pad->head || pad->merge_reads) { | |
8b0c5d76 | 1503 | int64_t bytes = pad->merge_reads ? pad->buf_len : align; |
7a3f542f VSO |
1504 | |
1505 | qemu_iovec_init_buf(&local_qiov, pad->buf, bytes); | |
1506 | ||
1507 | if (pad->head) { | |
c834dc05 | 1508 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
7a3f542f VSO |
1509 | } |
1510 | if (pad->merge_reads && pad->tail) { | |
c834dc05 | 1511 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
7a3f542f VSO |
1512 | } |
1513 | ret = bdrv_aligned_preadv(child, req, req->overlap_offset, bytes, | |
65cd4424 | 1514 | align, &local_qiov, 0, 0); |
7a3f542f VSO |
1515 | if (ret < 0) { |
1516 | return ret; | |
1517 | } | |
1518 | if (pad->head) { | |
c834dc05 | 1519 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
7a3f542f VSO |
1520 | } |
1521 | if (pad->merge_reads && pad->tail) { | |
c834dc05 | 1522 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
7a3f542f VSO |
1523 | } |
1524 | ||
1525 | if (pad->merge_reads) { | |
1526 | goto zero_mem; | |
1527 | } | |
1528 | } | |
1529 | ||
1530 | if (pad->tail) { | |
1531 | qemu_iovec_init_buf(&local_qiov, pad->tail_buf, align); | |
1532 | ||
c834dc05 | 1533 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
7a3f542f VSO |
1534 | ret = bdrv_aligned_preadv( |
1535 | child, req, | |
1536 | req->overlap_offset + req->overlap_bytes - align, | |
65cd4424 | 1537 | align, align, &local_qiov, 0, 0); |
7a3f542f VSO |
1538 | if (ret < 0) { |
1539 | return ret; | |
1540 | } | |
c834dc05 | 1541 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
7a3f542f VSO |
1542 | } |
1543 | ||
1544 | zero_mem: | |
1545 | if (zero_middle) { | |
1546 | memset(pad->buf + pad->head, 0, pad->buf_len - pad->head - pad->tail); | |
1547 | } | |
1548 | ||
1549 | return 0; | |
1550 | } | |
1551 | ||
1552 | static void bdrv_padding_destroy(BdrvRequestPadding *pad) | |
1553 | { | |
1554 | if (pad->buf) { | |
1555 | qemu_vfree(pad->buf); | |
1556 | qemu_iovec_destroy(&pad->local_qiov); | |
1557 | } | |
98ca4549 | 1558 | memset(pad, 0, sizeof(*pad)); |
7a3f542f VSO |
1559 | } |
1560 | ||
1561 | /* | |
1562 | * bdrv_pad_request | |
1563 | * | |
1564 | * Exchange request parameters with padded request if needed. Don't include RMW | |
1565 | * read of padding, bdrv_padding_rmw_read() should be called separately if | |
1566 | * needed. | |
1567 | * | |
98ca4549 VSO |
1568 | * Request parameters (@qiov, &qiov_offset, &offset, &bytes) are in-out: |
1569 | * - on function start they represent original request | |
1570 | * - on failure or when padding is not needed they are unchanged | |
1571 | * - on success when padding is needed they represent padded request | |
61007b31 | 1572 | */ |
98ca4549 VSO |
1573 | static int bdrv_pad_request(BlockDriverState *bs, |
1574 | QEMUIOVector **qiov, size_t *qiov_offset, | |
37e9403e | 1575 | int64_t *offset, int64_t *bytes, |
e8b65355 SH |
1576 | BdrvRequestPadding *pad, bool *padded, |
1577 | BdrvRequestFlags *flags) | |
7a3f542f | 1578 | { |
4c002cef VSO |
1579 | int ret; |
1580 | ||
37e9403e VSO |
1581 | bdrv_check_qiov_request(*offset, *bytes, *qiov, *qiov_offset, &error_abort); |
1582 | ||
7a3f542f | 1583 | if (!bdrv_init_padding(bs, *offset, *bytes, pad)) { |
98ca4549 VSO |
1584 | if (padded) { |
1585 | *padded = false; | |
1586 | } | |
1587 | return 0; | |
7a3f542f VSO |
1588 | } |
1589 | ||
4c002cef VSO |
1590 | ret = qemu_iovec_init_extended(&pad->local_qiov, pad->buf, pad->head, |
1591 | *qiov, *qiov_offset, *bytes, | |
1592 | pad->buf + pad->buf_len - pad->tail, | |
1593 | pad->tail); | |
98ca4549 VSO |
1594 | if (ret < 0) { |
1595 | bdrv_padding_destroy(pad); | |
1596 | return ret; | |
1597 | } | |
7a3f542f VSO |
1598 | *bytes += pad->head + pad->tail; |
1599 | *offset -= pad->head; | |
1600 | *qiov = &pad->local_qiov; | |
1acc3466 | 1601 | *qiov_offset = 0; |
98ca4549 VSO |
1602 | if (padded) { |
1603 | *padded = true; | |
1604 | } | |
e8b65355 SH |
1605 | if (flags) { |
1606 | /* Can't use optimization hint with bounce buffer */ | |
1607 | *flags &= ~BDRV_REQ_REGISTERED_BUF; | |
1608 | } | |
7a3f542f | 1609 | |
98ca4549 | 1610 | return 0; |
7a3f542f VSO |
1611 | } |
1612 | ||
a03ef88f | 1613 | int coroutine_fn bdrv_co_preadv(BdrvChild *child, |
e9e52efd | 1614 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, |
61007b31 | 1615 | BdrvRequestFlags flags) |
1acc3466 | 1616 | { |
967d7905 | 1617 | IO_CODE(); |
1acc3466 VSO |
1618 | return bdrv_co_preadv_part(child, offset, bytes, qiov, 0, flags); |
1619 | } | |
1620 | ||
1621 | int coroutine_fn bdrv_co_preadv_part(BdrvChild *child, | |
37e9403e | 1622 | int64_t offset, int64_t bytes, |
1acc3466 VSO |
1623 | QEMUIOVector *qiov, size_t qiov_offset, |
1624 | BdrvRequestFlags flags) | |
61007b31 | 1625 | { |
a03ef88f | 1626 | BlockDriverState *bs = child->bs; |
61007b31 | 1627 | BdrvTrackedRequest req; |
7a3f542f | 1628 | BdrvRequestPadding pad; |
61007b31 | 1629 | int ret; |
967d7905 | 1630 | IO_CODE(); |
61007b31 | 1631 | |
37e9403e | 1632 | trace_bdrv_co_preadv_part(bs, offset, bytes, flags); |
61007b31 | 1633 | |
1e97be91 | 1634 | if (!bdrv_co_is_inserted(bs)) { |
f4dad307 VSO |
1635 | return -ENOMEDIUM; |
1636 | } | |
1637 | ||
63f4ad11 | 1638 | ret = bdrv_check_request32(offset, bytes, qiov, qiov_offset); |
61007b31 SH |
1639 | if (ret < 0) { |
1640 | return ret; | |
1641 | } | |
1642 | ||
ac9d00bf VSO |
1643 | if (bytes == 0 && !QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)) { |
1644 | /* | |
1645 | * Aligning zero request is nonsense. Even if driver has special meaning | |
1646 | * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass | |
1647 | * it to driver due to request_alignment. | |
1648 | * | |
1649 | * Still, no reason to return an error if someone do unaligned | |
1650 | * zero-length read occasionally. | |
1651 | */ | |
1652 | return 0; | |
1653 | } | |
1654 | ||
99723548 PB |
1655 | bdrv_inc_in_flight(bs); |
1656 | ||
9568b511 | 1657 | /* Don't do copy-on-read if we read data before write operation */ |
d73415a3 | 1658 | if (qatomic_read(&bs->copy_on_read)) { |
61007b31 SH |
1659 | flags |= BDRV_REQ_COPY_ON_READ; |
1660 | } | |
1661 | ||
98ca4549 | 1662 | ret = bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, &pad, |
e8b65355 | 1663 | NULL, &flags); |
98ca4549 | 1664 | if (ret < 0) { |
87ab8802 | 1665 | goto fail; |
98ca4549 | 1666 | } |
61007b31 | 1667 | |
ebde595c | 1668 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
7a3f542f VSO |
1669 | ret = bdrv_aligned_preadv(child, &req, offset, bytes, |
1670 | bs->bl.request_alignment, | |
1acc3466 | 1671 | qiov, qiov_offset, flags); |
61007b31 | 1672 | tracked_request_end(&req); |
7a3f542f | 1673 | bdrv_padding_destroy(&pad); |
61007b31 | 1674 | |
87ab8802 KW |
1675 | fail: |
1676 | bdrv_dec_in_flight(bs); | |
1677 | ||
61007b31 SH |
1678 | return ret; |
1679 | } | |
1680 | ||
eeb47775 KW |
1681 | static int coroutine_fn GRAPH_RDLOCK |
1682 | bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
1683 | BdrvRequestFlags flags) | |
61007b31 SH |
1684 | { |
1685 | BlockDriver *drv = bs->drv; | |
1686 | QEMUIOVector qiov; | |
0d93ed08 | 1687 | void *buf = NULL; |
61007b31 | 1688 | int ret = 0; |
465fe887 | 1689 | bool need_flush = false; |
443668ca DL |
1690 | int head = 0; |
1691 | int tail = 0; | |
61007b31 | 1692 | |
2aaa3f9b VSO |
1693 | int64_t max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, |
1694 | INT64_MAX); | |
a5b8dd2c EB |
1695 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1696 | bs->bl.request_alignment); | |
cb2e2878 | 1697 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER); |
d05aa8bb | 1698 | |
abaf8b75 | 1699 | assert_bdrv_graph_readable(); |
5ae07b14 VSO |
1700 | bdrv_check_request(offset, bytes, &error_abort); |
1701 | ||
d470ad42 HR |
1702 | if (!drv) { |
1703 | return -ENOMEDIUM; | |
1704 | } | |
1705 | ||
fe0480d6 KW |
1706 | if ((flags & ~bs->supported_zero_flags) & BDRV_REQ_NO_FALLBACK) { |
1707 | return -ENOTSUP; | |
1708 | } | |
1709 | ||
e8b65355 SH |
1710 | /* By definition there is no user buffer so this flag doesn't make sense */ |
1711 | if (flags & BDRV_REQ_REGISTERED_BUF) { | |
1712 | return -EINVAL; | |
1713 | } | |
1714 | ||
0bc329fb HR |
1715 | /* Invalidate the cached block-status data range if this write overlaps */ |
1716 | bdrv_bsc_invalidate_range(bs, offset, bytes); | |
1717 | ||
b8d0a980 EB |
1718 | assert(alignment % bs->bl.request_alignment == 0); |
1719 | head = offset % alignment; | |
f5a5ca79 | 1720 | tail = (offset + bytes) % alignment; |
b8d0a980 EB |
1721 | max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment); |
1722 | assert(max_write_zeroes >= bs->bl.request_alignment); | |
61007b31 | 1723 | |
f5a5ca79 | 1724 | while (bytes > 0 && !ret) { |
5ae07b14 | 1725 | int64_t num = bytes; |
61007b31 SH |
1726 | |
1727 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1728 | * to be aligned, and that unaligned requests do not cross cluster |
1729 | * boundaries. | |
61007b31 | 1730 | */ |
443668ca | 1731 | if (head) { |
b2f95fee EB |
1732 | /* Make a small request up to the first aligned sector. For |
1733 | * convenience, limit this request to max_transfer even if | |
1734 | * we don't need to fall back to writes. */ | |
f5a5ca79 | 1735 | num = MIN(MIN(bytes, max_transfer), alignment - head); |
b2f95fee EB |
1736 | head = (head + num) % alignment; |
1737 | assert(num < max_write_zeroes); | |
d05aa8bb | 1738 | } else if (tail && num > alignment) { |
443668ca DL |
1739 | /* Shorten the request to the last aligned sector. */ |
1740 | num -= tail; | |
61007b31 SH |
1741 | } |
1742 | ||
1743 | /* limit request size */ | |
1744 | if (num > max_write_zeroes) { | |
1745 | num = max_write_zeroes; | |
1746 | } | |
1747 | ||
1748 | ret = -ENOTSUP; | |
1749 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1750 | if (drv->bdrv_co_pwrite_zeroes) { |
1751 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1752 | flags & bs->supported_zero_flags); | |
1753 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1754 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1755 | need_flush = true; | |
1756 | } | |
465fe887 EB |
1757 | } else { |
1758 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1759 | } |
1760 | ||
294682cc | 1761 | if (ret == -ENOTSUP && !(flags & BDRV_REQ_NO_FALLBACK)) { |
61007b31 | 1762 | /* Fall back to bounce buffer if write zeroes is unsupported */ |
465fe887 EB |
1763 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1764 | ||
1765 | if ((flags & BDRV_REQ_FUA) && | |
1766 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1767 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1768 | * flush on each chunk; use just one at the end */ | |
1769 | write_flags &= ~BDRV_REQ_FUA; | |
1770 | need_flush = true; | |
1771 | } | |
5def6b80 | 1772 | num = MIN(num, max_transfer); |
0d93ed08 VSO |
1773 | if (buf == NULL) { |
1774 | buf = qemu_try_blockalign0(bs, num); | |
1775 | if (buf == NULL) { | |
61007b31 SH |
1776 | ret = -ENOMEM; |
1777 | goto fail; | |
1778 | } | |
61007b31 | 1779 | } |
0d93ed08 | 1780 | qemu_iovec_init_buf(&qiov, buf, num); |
61007b31 | 1781 | |
ac850bf0 | 1782 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, 0, write_flags); |
61007b31 SH |
1783 | |
1784 | /* Keep bounce buffer around if it is big enough for all | |
1785 | * all future requests. | |
1786 | */ | |
5def6b80 | 1787 | if (num < max_transfer) { |
0d93ed08 VSO |
1788 | qemu_vfree(buf); |
1789 | buf = NULL; | |
61007b31 SH |
1790 | } |
1791 | } | |
1792 | ||
d05aa8bb | 1793 | offset += num; |
f5a5ca79 | 1794 | bytes -= num; |
61007b31 SH |
1795 | } |
1796 | ||
1797 | fail: | |
465fe887 EB |
1798 | if (ret == 0 && need_flush) { |
1799 | ret = bdrv_co_flush(bs); | |
1800 | } | |
0d93ed08 | 1801 | qemu_vfree(buf); |
61007b31 SH |
1802 | return ret; |
1803 | } | |
1804 | ||
a00e70c0 | 1805 | static inline int coroutine_fn GRAPH_RDLOCK |
fcfd9ade | 1806 | bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, int64_t bytes, |
85fe2479 FZ |
1807 | BdrvTrackedRequest *req, int flags) |
1808 | { | |
1809 | BlockDriverState *bs = child->bs; | |
fcfd9ade VSO |
1810 | |
1811 | bdrv_check_request(offset, bytes, &error_abort); | |
85fe2479 | 1812 | |
307261b2 | 1813 | if (bdrv_is_read_only(bs)) { |
85fe2479 FZ |
1814 | return -EPERM; |
1815 | } | |
1816 | ||
85fe2479 FZ |
1817 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); |
1818 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); | |
1819 | assert(!(flags & ~BDRV_REQ_MASK)); | |
d1a764d1 | 1820 | assert(!((flags & BDRV_REQ_NO_WAIT) && !(flags & BDRV_REQ_SERIALISING))); |
85fe2479 FZ |
1821 | |
1822 | if (flags & BDRV_REQ_SERIALISING) { | |
d1a764d1 VSO |
1823 | QEMU_LOCK_GUARD(&bs->reqs_lock); |
1824 | ||
1825 | tracked_request_set_serialising(req, bdrv_get_cluster_size(bs)); | |
1826 | ||
1827 | if ((flags & BDRV_REQ_NO_WAIT) && bdrv_find_conflicting_request(req)) { | |
1828 | return -EBUSY; | |
1829 | } | |
1830 | ||
1831 | bdrv_wait_serialising_requests_locked(req); | |
18fbd0de PB |
1832 | } else { |
1833 | bdrv_wait_serialising_requests(req); | |
85fe2479 FZ |
1834 | } |
1835 | ||
85fe2479 FZ |
1836 | assert(req->overlap_offset <= offset); |
1837 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
fcfd9ade VSO |
1838 | assert(offset + bytes <= bs->total_sectors * BDRV_SECTOR_SIZE || |
1839 | child->perm & BLK_PERM_RESIZE); | |
85fe2479 | 1840 | |
cd47d792 FZ |
1841 | switch (req->type) { |
1842 | case BDRV_TRACKED_WRITE: | |
1843 | case BDRV_TRACKED_DISCARD: | |
1844 | if (flags & BDRV_REQ_WRITE_UNCHANGED) { | |
1845 | assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
1846 | } else { | |
1847 | assert(child->perm & BLK_PERM_WRITE); | |
1848 | } | |
94783301 VSO |
1849 | bdrv_write_threshold_check_write(bs, offset, bytes); |
1850 | return 0; | |
cd47d792 FZ |
1851 | case BDRV_TRACKED_TRUNCATE: |
1852 | assert(child->perm & BLK_PERM_RESIZE); | |
1853 | return 0; | |
1854 | default: | |
1855 | abort(); | |
85fe2479 | 1856 | } |
85fe2479 FZ |
1857 | } |
1858 | ||
1859 | static inline void coroutine_fn | |
fcfd9ade | 1860 | bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, int64_t bytes, |
85fe2479 FZ |
1861 | BdrvTrackedRequest *req, int ret) |
1862 | { | |
1863 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1864 | BlockDriverState *bs = child->bs; | |
1865 | ||
fcfd9ade VSO |
1866 | bdrv_check_request(offset, bytes, &error_abort); |
1867 | ||
d73415a3 | 1868 | qatomic_inc(&bs->write_gen); |
85fe2479 | 1869 | |
00695c27 FZ |
1870 | /* |
1871 | * Discard cannot extend the image, but in error handling cases, such as | |
1872 | * when reverting a qcow2 cluster allocation, the discarded range can pass | |
1873 | * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD | |
1874 | * here. Instead, just skip it, since semantically a discard request | |
1875 | * beyond EOF cannot expand the image anyway. | |
1876 | */ | |
7f8f03ef | 1877 | if (ret == 0 && |
cd47d792 FZ |
1878 | (req->type == BDRV_TRACKED_TRUNCATE || |
1879 | end_sector > bs->total_sectors) && | |
1880 | req->type != BDRV_TRACKED_DISCARD) { | |
7f8f03ef FZ |
1881 | bs->total_sectors = end_sector; |
1882 | bdrv_parent_cb_resize(bs); | |
1883 | bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS); | |
85fe2479 | 1884 | } |
00695c27 FZ |
1885 | if (req->bytes) { |
1886 | switch (req->type) { | |
1887 | case BDRV_TRACKED_WRITE: | |
1888 | stat64_max(&bs->wr_highest_offset, offset + bytes); | |
1889 | /* fall through, to set dirty bits */ | |
1890 | case BDRV_TRACKED_DISCARD: | |
1891 | bdrv_set_dirty(bs, offset, bytes); | |
1892 | break; | |
1893 | default: | |
1894 | break; | |
1895 | } | |
1896 | } | |
85fe2479 FZ |
1897 | } |
1898 | ||
61007b31 | 1899 | /* |
04ed95f4 EB |
1900 | * Forwards an already correctly aligned write request to the BlockDriver, |
1901 | * after possibly fragmenting it. | |
61007b31 | 1902 | */ |
7b1fb72e KW |
1903 | static int coroutine_fn GRAPH_RDLOCK |
1904 | bdrv_aligned_pwritev(BdrvChild *child, BdrvTrackedRequest *req, | |
1905 | int64_t offset, int64_t bytes, int64_t align, | |
1906 | QEMUIOVector *qiov, size_t qiov_offset, | |
1907 | BdrvRequestFlags flags) | |
61007b31 | 1908 | { |
85c97ca7 | 1909 | BlockDriverState *bs = child->bs; |
61007b31 | 1910 | BlockDriver *drv = bs->drv; |
61007b31 SH |
1911 | int ret; |
1912 | ||
fcfd9ade | 1913 | int64_t bytes_remaining = bytes; |
04ed95f4 | 1914 | int max_transfer; |
61007b31 | 1915 | |
fcfd9ade VSO |
1916 | bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, &error_abort); |
1917 | ||
d470ad42 HR |
1918 | if (!drv) { |
1919 | return -ENOMEDIUM; | |
1920 | } | |
1921 | ||
d6883bc9 VSO |
1922 | if (bdrv_has_readonly_bitmaps(bs)) { |
1923 | return -EPERM; | |
1924 | } | |
1925 | ||
cff86b38 EB |
1926 | assert(is_power_of_2(align)); |
1927 | assert((offset & (align - 1)) == 0); | |
1928 | assert((bytes & (align - 1)) == 0); | |
04ed95f4 EB |
1929 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1930 | align); | |
61007b31 | 1931 | |
85fe2479 | 1932 | ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags); |
61007b31 SH |
1933 | |
1934 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 1935 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
28c4da28 | 1936 | qemu_iovec_is_zero(qiov, qiov_offset, bytes)) { |
61007b31 SH |
1937 | flags |= BDRV_REQ_ZERO_WRITE; |
1938 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
1939 | flags |= BDRV_REQ_MAY_UNMAP; | |
1940 | } | |
3c586715 SH |
1941 | |
1942 | /* Can't use optimization hint with bufferless zero write */ | |
1943 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
61007b31 SH |
1944 | } |
1945 | ||
1946 | if (ret < 0) { | |
1947 | /* Do nothing, write notifier decided to fail this request */ | |
1948 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
c834dc05 | 1949 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 1950 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
3ea1a091 | 1951 | } else if (flags & BDRV_REQ_WRITE_COMPRESSED) { |
28c4da28 VSO |
1952 | ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, |
1953 | qiov, qiov_offset); | |
04ed95f4 | 1954 | } else if (bytes <= max_transfer) { |
c834dc05 | 1955 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV); |
28c4da28 | 1956 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, qiov_offset, flags); |
04ed95f4 | 1957 | } else { |
c834dc05 | 1958 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV); |
04ed95f4 EB |
1959 | while (bytes_remaining) { |
1960 | int num = MIN(bytes_remaining, max_transfer); | |
04ed95f4 EB |
1961 | int local_flags = flags; |
1962 | ||
1963 | assert(num); | |
1964 | if (num < bytes_remaining && (flags & BDRV_REQ_FUA) && | |
1965 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1966 | /* If FUA is going to be emulated by flush, we only | |
1967 | * need to flush on the last iteration */ | |
1968 | local_flags &= ~BDRV_REQ_FUA; | |
1969 | } | |
04ed95f4 EB |
1970 | |
1971 | ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining, | |
134b7dec HR |
1972 | num, qiov, |
1973 | qiov_offset + bytes - bytes_remaining, | |
28c4da28 | 1974 | local_flags); |
04ed95f4 EB |
1975 | if (ret < 0) { |
1976 | break; | |
1977 | } | |
1978 | bytes_remaining -= num; | |
1979 | } | |
61007b31 | 1980 | } |
c834dc05 | 1981 | bdrv_co_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 1982 | |
61007b31 | 1983 | if (ret >= 0) { |
04ed95f4 | 1984 | ret = 0; |
61007b31 | 1985 | } |
85fe2479 | 1986 | bdrv_co_write_req_finish(child, offset, bytes, req, ret); |
61007b31 SH |
1987 | |
1988 | return ret; | |
1989 | } | |
1990 | ||
7b1fb72e KW |
1991 | static int coroutine_fn GRAPH_RDLOCK |
1992 | bdrv_co_do_zero_pwritev(BdrvChild *child, int64_t offset, int64_t bytes, | |
1993 | BdrvRequestFlags flags, BdrvTrackedRequest *req) | |
9eeb6dd1 | 1994 | { |
85c97ca7 | 1995 | BlockDriverState *bs = child->bs; |
9eeb6dd1 | 1996 | QEMUIOVector local_qiov; |
a5b8dd2c | 1997 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 | 1998 | int ret = 0; |
7a3f542f VSO |
1999 | bool padding; |
2000 | BdrvRequestPadding pad; | |
9eeb6dd1 | 2001 | |
e8b65355 SH |
2002 | /* This flag doesn't make sense for padding or zero writes */ |
2003 | flags &= ~BDRV_REQ_REGISTERED_BUF; | |
2004 | ||
7a3f542f VSO |
2005 | padding = bdrv_init_padding(bs, offset, bytes, &pad); |
2006 | if (padding) { | |
45e62b46 | 2007 | assert(!(flags & BDRV_REQ_NO_WAIT)); |
8ac5aab2 | 2008 | bdrv_make_request_serialising(req, align); |
9eeb6dd1 | 2009 | |
7a3f542f VSO |
2010 | bdrv_padding_rmw_read(child, req, &pad, true); |
2011 | ||
2012 | if (pad.head || pad.merge_reads) { | |
2013 | int64_t aligned_offset = offset & ~(align - 1); | |
2014 | int64_t write_bytes = pad.merge_reads ? pad.buf_len : align; | |
2015 | ||
2016 | qemu_iovec_init_buf(&local_qiov, pad.buf, write_bytes); | |
2017 | ret = bdrv_aligned_pwritev(child, req, aligned_offset, write_bytes, | |
28c4da28 | 2018 | align, &local_qiov, 0, |
7a3f542f VSO |
2019 | flags & ~BDRV_REQ_ZERO_WRITE); |
2020 | if (ret < 0 || pad.merge_reads) { | |
2021 | /* Error or all work is done */ | |
2022 | goto out; | |
2023 | } | |
2024 | offset += write_bytes - pad.head; | |
2025 | bytes -= write_bytes - pad.head; | |
9eeb6dd1 | 2026 | } |
9eeb6dd1 FZ |
2027 | } |
2028 | ||
2029 | assert(!bytes || (offset & (align - 1)) == 0); | |
2030 | if (bytes >= align) { | |
2031 | /* Write the aligned part in the middle. */ | |
fcfd9ade | 2032 | int64_t aligned_bytes = bytes & ~(align - 1); |
85c97ca7 | 2033 | ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align, |
28c4da28 | 2034 | NULL, 0, flags); |
9eeb6dd1 | 2035 | if (ret < 0) { |
7a3f542f | 2036 | goto out; |
9eeb6dd1 FZ |
2037 | } |
2038 | bytes -= aligned_bytes; | |
2039 | offset += aligned_bytes; | |
2040 | } | |
2041 | ||
2042 | assert(!bytes || (offset & (align - 1)) == 0); | |
2043 | if (bytes) { | |
7a3f542f | 2044 | assert(align == pad.tail + bytes); |
9eeb6dd1 | 2045 | |
7a3f542f | 2046 | qemu_iovec_init_buf(&local_qiov, pad.tail_buf, align); |
85c97ca7 | 2047 | ret = bdrv_aligned_pwritev(child, req, offset, align, align, |
28c4da28 VSO |
2048 | &local_qiov, 0, |
2049 | flags & ~BDRV_REQ_ZERO_WRITE); | |
9eeb6dd1 | 2050 | } |
9eeb6dd1 | 2051 | |
7a3f542f VSO |
2052 | out: |
2053 | bdrv_padding_destroy(&pad); | |
2054 | ||
2055 | return ret; | |
9eeb6dd1 FZ |
2056 | } |
2057 | ||
61007b31 SH |
2058 | /* |
2059 | * Handle a write request in coroutine context | |
2060 | */ | |
a03ef88f | 2061 | int coroutine_fn bdrv_co_pwritev(BdrvChild *child, |
e9e52efd | 2062 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, |
61007b31 | 2063 | BdrvRequestFlags flags) |
1acc3466 | 2064 | { |
967d7905 | 2065 | IO_CODE(); |
1acc3466 VSO |
2066 | return bdrv_co_pwritev_part(child, offset, bytes, qiov, 0, flags); |
2067 | } | |
2068 | ||
2069 | int coroutine_fn bdrv_co_pwritev_part(BdrvChild *child, | |
37e9403e | 2070 | int64_t offset, int64_t bytes, QEMUIOVector *qiov, size_t qiov_offset, |
1acc3466 | 2071 | BdrvRequestFlags flags) |
61007b31 | 2072 | { |
a03ef88f | 2073 | BlockDriverState *bs = child->bs; |
61007b31 | 2074 | BdrvTrackedRequest req; |
a5b8dd2c | 2075 | uint64_t align = bs->bl.request_alignment; |
7a3f542f | 2076 | BdrvRequestPadding pad; |
61007b31 | 2077 | int ret; |
f0deecff | 2078 | bool padded = false; |
967d7905 | 2079 | IO_CODE(); |
61007b31 | 2080 | |
37e9403e | 2081 | trace_bdrv_co_pwritev_part(child->bs, offset, bytes, flags); |
f42cf447 | 2082 | |
1e97be91 | 2083 | if (!bdrv_co_is_inserted(bs)) { |
61007b31 SH |
2084 | return -ENOMEDIUM; |
2085 | } | |
61007b31 | 2086 | |
2aaa3f9b VSO |
2087 | if (flags & BDRV_REQ_ZERO_WRITE) { |
2088 | ret = bdrv_check_qiov_request(offset, bytes, qiov, qiov_offset, NULL); | |
2089 | } else { | |
2090 | ret = bdrv_check_request32(offset, bytes, qiov, qiov_offset); | |
2091 | } | |
61007b31 SH |
2092 | if (ret < 0) { |
2093 | return ret; | |
2094 | } | |
2095 | ||
f2208fdc AG |
2096 | /* If the request is misaligned then we can't make it efficient */ |
2097 | if ((flags & BDRV_REQ_NO_FALLBACK) && | |
2098 | !QEMU_IS_ALIGNED(offset | bytes, align)) | |
2099 | { | |
2100 | return -ENOTSUP; | |
2101 | } | |
2102 | ||
ac9d00bf VSO |
2103 | if (bytes == 0 && !QEMU_IS_ALIGNED(offset, bs->bl.request_alignment)) { |
2104 | /* | |
2105 | * Aligning zero request is nonsense. Even if driver has special meaning | |
2106 | * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass | |
2107 | * it to driver due to request_alignment. | |
2108 | * | |
2109 | * Still, no reason to return an error if someone do unaligned | |
2110 | * zero-length write occasionally. | |
2111 | */ | |
2112 | return 0; | |
2113 | } | |
2114 | ||
f0deecff VSO |
2115 | if (!(flags & BDRV_REQ_ZERO_WRITE)) { |
2116 | /* | |
2117 | * Pad request for following read-modify-write cycle. | |
2118 | * bdrv_co_do_zero_pwritev() does aligning by itself, so, we do | |
2119 | * alignment only if there is no ZERO flag. | |
2120 | */ | |
98ca4549 | 2121 | ret = bdrv_pad_request(bs, &qiov, &qiov_offset, &offset, &bytes, &pad, |
e8b65355 | 2122 | &padded, &flags); |
98ca4549 VSO |
2123 | if (ret < 0) { |
2124 | return ret; | |
2125 | } | |
f0deecff VSO |
2126 | } |
2127 | ||
99723548 | 2128 | bdrv_inc_in_flight(bs); |
ebde595c | 2129 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 2130 | |
18a59f03 | 2131 | if (flags & BDRV_REQ_ZERO_WRITE) { |
f0deecff | 2132 | assert(!padded); |
85c97ca7 | 2133 | ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req); |
9eeb6dd1 FZ |
2134 | goto out; |
2135 | } | |
2136 | ||
f0deecff VSO |
2137 | if (padded) { |
2138 | /* | |
2139 | * Request was unaligned to request_alignment and therefore | |
2140 | * padded. We are going to do read-modify-write, and must | |
2141 | * serialize the request to prevent interactions of the | |
2142 | * widened region with other transactions. | |
2143 | */ | |
45e62b46 | 2144 | assert(!(flags & BDRV_REQ_NO_WAIT)); |
8ac5aab2 | 2145 | bdrv_make_request_serialising(&req, align); |
7a3f542f | 2146 | bdrv_padding_rmw_read(child, &req, &pad, false); |
61007b31 SH |
2147 | } |
2148 | ||
85c97ca7 | 2149 | ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align, |
1acc3466 | 2150 | qiov, qiov_offset, flags); |
61007b31 | 2151 | |
7a3f542f | 2152 | bdrv_padding_destroy(&pad); |
61007b31 | 2153 | |
9eeb6dd1 FZ |
2154 | out: |
2155 | tracked_request_end(&req); | |
99723548 | 2156 | bdrv_dec_in_flight(bs); |
7a3f542f | 2157 | |
61007b31 SH |
2158 | return ret; |
2159 | } | |
2160 | ||
a03ef88f | 2161 | int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, |
e9e52efd | 2162 | int64_t bytes, BdrvRequestFlags flags) |
61007b31 | 2163 | { |
384a48fb | 2164 | IO_CODE(); |
f5a5ca79 | 2165 | trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags); |
abaf8b75 | 2166 | assert_bdrv_graph_readable(); |
61007b31 | 2167 | |
a03ef88f | 2168 | if (!(child->bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
2169 | flags &= ~BDRV_REQ_MAY_UNMAP; |
2170 | } | |
61007b31 | 2171 | |
f5a5ca79 | 2172 | return bdrv_co_pwritev(child, offset, bytes, NULL, |
74021bc4 | 2173 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
2174 | } |
2175 | ||
4085f5c7 JS |
2176 | /* |
2177 | * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not. | |
2178 | */ | |
2179 | int bdrv_flush_all(void) | |
2180 | { | |
2181 | BdrvNextIterator it; | |
2182 | BlockDriverState *bs = NULL; | |
2183 | int result = 0; | |
2184 | ||
f791bf7f EGE |
2185 | GLOBAL_STATE_CODE(); |
2186 | ||
c8aa7895 PD |
2187 | /* |
2188 | * bdrv queue is managed by record/replay, | |
2189 | * creating new flush request for stopping | |
2190 | * the VM may break the determinism | |
2191 | */ | |
2192 | if (replay_events_enabled()) { | |
2193 | return result; | |
2194 | } | |
2195 | ||
4085f5c7 JS |
2196 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
2197 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2198 | int ret; | |
2199 | ||
2200 | aio_context_acquire(aio_context); | |
2201 | ret = bdrv_flush(bs); | |
2202 | if (ret < 0 && !result) { | |
2203 | result = ret; | |
2204 | } | |
2205 | aio_context_release(aio_context); | |
2206 | } | |
2207 | ||
2208 | return result; | |
2209 | } | |
2210 | ||
61007b31 SH |
2211 | /* |
2212 | * Returns the allocation status of the specified sectors. | |
2213 | * Drivers not implementing the functionality are assumed to not support | |
2214 | * backing files, hence all their sectors are reported as allocated. | |
2215 | * | |
86a3d5c6 EB |
2216 | * If 'want_zero' is true, the caller is querying for mapping |
2217 | * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and | |
2218 | * _ZERO where possible; otherwise, the result favors larger 'pnum', | |
2219 | * with a focus on accurate BDRV_BLOCK_ALLOCATED. | |
c9ce8c4d | 2220 | * |
2e8bc787 | 2221 | * If 'offset' is beyond the end of the disk image the return value is |
fb0d8654 | 2222 | * BDRV_BLOCK_EOF and 'pnum' is set to 0. |
61007b31 | 2223 | * |
2e8bc787 | 2224 | * 'bytes' is the max value 'pnum' should be set to. If bytes goes |
fb0d8654 EB |
2225 | * beyond the end of the disk image it will be clamped; if 'pnum' is set to |
2226 | * the end of the image, then the returned value will include BDRV_BLOCK_EOF. | |
67a0fd2a | 2227 | * |
2e8bc787 EB |
2228 | * 'pnum' is set to the number of bytes (including and immediately |
2229 | * following the specified offset) that are easily known to be in the | |
2230 | * same allocated/unallocated state. Note that a second call starting | |
2231 | * at the original offset plus returned pnum may have the same status. | |
2232 | * The returned value is non-zero on success except at end-of-file. | |
2233 | * | |
2234 | * Returns negative errno on failure. Otherwise, if the | |
2235 | * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are | |
2236 | * set to the host mapping and BDS corresponding to the guest offset. | |
61007b31 | 2237 | */ |
7ff9579e KW |
2238 | static int coroutine_fn GRAPH_RDLOCK |
2239 | bdrv_co_block_status(BlockDriverState *bs, bool want_zero, | |
2240 | int64_t offset, int64_t bytes, | |
2241 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
2e8bc787 EB |
2242 | { |
2243 | int64_t total_size; | |
2244 | int64_t n; /* bytes */ | |
efa6e2ed | 2245 | int ret; |
2e8bc787 | 2246 | int64_t local_map = 0; |
298a1665 | 2247 | BlockDriverState *local_file = NULL; |
efa6e2ed EB |
2248 | int64_t aligned_offset, aligned_bytes; |
2249 | uint32_t align; | |
549ec0d9 | 2250 | bool has_filtered_child; |
61007b31 | 2251 | |
298a1665 | 2252 | assert(pnum); |
7ff9579e | 2253 | assert_bdrv_graph_readable(); |
298a1665 | 2254 | *pnum = 0; |
2e8bc787 EB |
2255 | total_size = bdrv_getlength(bs); |
2256 | if (total_size < 0) { | |
2257 | ret = total_size; | |
298a1665 | 2258 | goto early_out; |
61007b31 SH |
2259 | } |
2260 | ||
2e8bc787 | 2261 | if (offset >= total_size) { |
298a1665 EB |
2262 | ret = BDRV_BLOCK_EOF; |
2263 | goto early_out; | |
61007b31 | 2264 | } |
2e8bc787 | 2265 | if (!bytes) { |
298a1665 EB |
2266 | ret = 0; |
2267 | goto early_out; | |
9cdcfd9f | 2268 | } |
61007b31 | 2269 | |
2e8bc787 EB |
2270 | n = total_size - offset; |
2271 | if (n < bytes) { | |
2272 | bytes = n; | |
61007b31 SH |
2273 | } |
2274 | ||
d470ad42 HR |
2275 | /* Must be non-NULL or bdrv_getlength() would have failed */ |
2276 | assert(bs->drv); | |
549ec0d9 HR |
2277 | has_filtered_child = bdrv_filter_child(bs); |
2278 | if (!bs->drv->bdrv_co_block_status && !has_filtered_child) { | |
2e8bc787 | 2279 | *pnum = bytes; |
61007b31 | 2280 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; |
2e8bc787 | 2281 | if (offset + bytes == total_size) { |
fb0d8654 EB |
2282 | ret |= BDRV_BLOCK_EOF; |
2283 | } | |
61007b31 | 2284 | if (bs->drv->protocol_name) { |
2e8bc787 EB |
2285 | ret |= BDRV_BLOCK_OFFSET_VALID; |
2286 | local_map = offset; | |
298a1665 | 2287 | local_file = bs; |
61007b31 | 2288 | } |
298a1665 | 2289 | goto early_out; |
61007b31 SH |
2290 | } |
2291 | ||
99723548 | 2292 | bdrv_inc_in_flight(bs); |
efa6e2ed EB |
2293 | |
2294 | /* Round out to request_alignment boundaries */ | |
86a3d5c6 | 2295 | align = bs->bl.request_alignment; |
efa6e2ed EB |
2296 | aligned_offset = QEMU_ALIGN_DOWN(offset, align); |
2297 | aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset; | |
2298 | ||
549ec0d9 | 2299 | if (bs->drv->bdrv_co_block_status) { |
0bc329fb HR |
2300 | /* |
2301 | * Use the block-status cache only for protocol nodes: Format | |
2302 | * drivers are generally quick to inquire the status, but protocol | |
2303 | * drivers often need to get information from outside of qemu, so | |
2304 | * we do not have control over the actual implementation. There | |
2305 | * have been cases where inquiring the status took an unreasonably | |
2306 | * long time, and we can do nothing in qemu to fix it. | |
2307 | * This is especially problematic for images with large data areas, | |
2308 | * because finding the few holes in them and giving them special | |
2309 | * treatment does not gain much performance. Therefore, we try to | |
2310 | * cache the last-identified data region. | |
2311 | * | |
2312 | * Second, limiting ourselves to protocol nodes allows us to assume | |
2313 | * the block status for data regions to be DATA | OFFSET_VALID, and | |
2314 | * that the host offset is the same as the guest offset. | |
2315 | * | |
2316 | * Note that it is possible that external writers zero parts of | |
2317 | * the cached regions without the cache being invalidated, and so | |
2318 | * we may report zeroes as data. This is not catastrophic, | |
2319 | * however, because reporting zeroes as data is fine. | |
2320 | */ | |
2321 | if (QLIST_EMPTY(&bs->children) && | |
2322 | bdrv_bsc_is_data(bs, aligned_offset, pnum)) | |
2323 | { | |
2324 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
2325 | local_file = bs; | |
2326 | local_map = aligned_offset; | |
2327 | } else { | |
2328 | ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset, | |
2329 | aligned_bytes, pnum, &local_map, | |
2330 | &local_file); | |
2331 | ||
2332 | /* | |
2333 | * Note that checking QLIST_EMPTY(&bs->children) is also done when | |
2334 | * the cache is queried above. Technically, we do not need to check | |
2335 | * it here; the worst that can happen is that we fill the cache for | |
2336 | * non-protocol nodes, and then it is never used. However, filling | |
2337 | * the cache requires an RCU update, so double check here to avoid | |
2338 | * such an update if possible. | |
113b727c HR |
2339 | * |
2340 | * Check want_zero, because we only want to update the cache when we | |
2341 | * have accurate information about what is zero and what is data. | |
0bc329fb | 2342 | */ |
113b727c HR |
2343 | if (want_zero && |
2344 | ret == (BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID) && | |
0bc329fb HR |
2345 | QLIST_EMPTY(&bs->children)) |
2346 | { | |
2347 | /* | |
2348 | * When a protocol driver reports BLOCK_OFFSET_VALID, the | |
2349 | * returned local_map value must be the same as the offset we | |
2350 | * have passed (aligned_offset), and local_bs must be the node | |
2351 | * itself. | |
2352 | * Assert this, because we follow this rule when reading from | |
2353 | * the cache (see the `local_file = bs` and | |
2354 | * `local_map = aligned_offset` assignments above), and the | |
2355 | * result the cache delivers must be the same as the driver | |
2356 | * would deliver. | |
2357 | */ | |
2358 | assert(local_file == bs); | |
2359 | assert(local_map == aligned_offset); | |
2360 | bdrv_bsc_fill(bs, aligned_offset, *pnum); | |
2361 | } | |
2362 | } | |
549ec0d9 HR |
2363 | } else { |
2364 | /* Default code for filters */ | |
2365 | ||
2366 | local_file = bdrv_filter_bs(bs); | |
2367 | assert(local_file); | |
2368 | ||
2369 | *pnum = aligned_bytes; | |
2370 | local_map = aligned_offset; | |
2371 | ret = BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; | |
2372 | } | |
636cb512 EB |
2373 | if (ret < 0) { |
2374 | *pnum = 0; | |
2375 | goto out; | |
efa6e2ed EB |
2376 | } |
2377 | ||
2e8bc787 | 2378 | /* |
636cb512 | 2379 | * The driver's result must be a non-zero multiple of request_alignment. |
efa6e2ed | 2380 | * Clamp pnum and adjust map to original request. |
2e8bc787 | 2381 | */ |
636cb512 EB |
2382 | assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) && |
2383 | align > offset - aligned_offset); | |
69f47505 VSO |
2384 | if (ret & BDRV_BLOCK_RECURSE) { |
2385 | assert(ret & BDRV_BLOCK_DATA); | |
2386 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
2387 | assert(!(ret & BDRV_BLOCK_ZERO)); | |
2388 | } | |
2389 | ||
efa6e2ed EB |
2390 | *pnum -= offset - aligned_offset; |
2391 | if (*pnum > bytes) { | |
2392 | *pnum = bytes; | |
61007b31 | 2393 | } |
2e8bc787 | 2394 | if (ret & BDRV_BLOCK_OFFSET_VALID) { |
efa6e2ed | 2395 | local_map += offset - aligned_offset; |
2e8bc787 | 2396 | } |
61007b31 SH |
2397 | |
2398 | if (ret & BDRV_BLOCK_RAW) { | |
298a1665 | 2399 | assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file); |
2e8bc787 EB |
2400 | ret = bdrv_co_block_status(local_file, want_zero, local_map, |
2401 | *pnum, pnum, &local_map, &local_file); | |
99723548 | 2402 | goto out; |
61007b31 SH |
2403 | } |
2404 | ||
2405 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
2406 | ret |= BDRV_BLOCK_ALLOCATED; | |
d40f4a56 | 2407 | } else if (bs->drv->supports_backing) { |
cb850315 HR |
2408 | BlockDriverState *cow_bs = bdrv_cow_bs(bs); |
2409 | ||
d40f4a56 AG |
2410 | if (!cow_bs) { |
2411 | ret |= BDRV_BLOCK_ZERO; | |
2412 | } else if (want_zero) { | |
cb850315 | 2413 | int64_t size2 = bdrv_getlength(cow_bs); |
c9ce8c4d | 2414 | |
2e8bc787 | 2415 | if (size2 >= 0 && offset >= size2) { |
61007b31 SH |
2416 | ret |= BDRV_BLOCK_ZERO; |
2417 | } | |
2418 | } | |
2419 | } | |
2420 | ||
69f47505 VSO |
2421 | if (want_zero && ret & BDRV_BLOCK_RECURSE && |
2422 | local_file && local_file != bs && | |
61007b31 SH |
2423 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
2424 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
2e8bc787 EB |
2425 | int64_t file_pnum; |
2426 | int ret2; | |
61007b31 | 2427 | |
2e8bc787 EB |
2428 | ret2 = bdrv_co_block_status(local_file, want_zero, local_map, |
2429 | *pnum, &file_pnum, NULL, NULL); | |
61007b31 SH |
2430 | if (ret2 >= 0) { |
2431 | /* Ignore errors. This is just providing extra information, it | |
2432 | * is useful but not necessary. | |
2433 | */ | |
c61e684e EB |
2434 | if (ret2 & BDRV_BLOCK_EOF && |
2435 | (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) { | |
2436 | /* | |
2437 | * It is valid for the format block driver to read | |
2438 | * beyond the end of the underlying file's current | |
2439 | * size; such areas read as zero. | |
2440 | */ | |
61007b31 SH |
2441 | ret |= BDRV_BLOCK_ZERO; |
2442 | } else { | |
2443 | /* Limit request to the range reported by the protocol driver */ | |
2444 | *pnum = file_pnum; | |
2445 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
2446 | } | |
2447 | } | |
2448 | } | |
2449 | ||
99723548 PB |
2450 | out: |
2451 | bdrv_dec_in_flight(bs); | |
2e8bc787 | 2452 | if (ret >= 0 && offset + *pnum == total_size) { |
fb0d8654 EB |
2453 | ret |= BDRV_BLOCK_EOF; |
2454 | } | |
298a1665 EB |
2455 | early_out: |
2456 | if (file) { | |
2457 | *file = local_file; | |
2458 | } | |
2e8bc787 EB |
2459 | if (map) { |
2460 | *map = local_map; | |
2461 | } | |
61007b31 SH |
2462 | return ret; |
2463 | } | |
2464 | ||
21c2283e | 2465 | int coroutine_fn |
f9e694cb VSO |
2466 | bdrv_co_common_block_status_above(BlockDriverState *bs, |
2467 | BlockDriverState *base, | |
3555a432 | 2468 | bool include_base, |
f9e694cb VSO |
2469 | bool want_zero, |
2470 | int64_t offset, | |
2471 | int64_t bytes, | |
2472 | int64_t *pnum, | |
2473 | int64_t *map, | |
a92b1b06 EB |
2474 | BlockDriverState **file, |
2475 | int *depth) | |
ba3f0e25 | 2476 | { |
67c095c8 | 2477 | int ret; |
ba3f0e25 | 2478 | BlockDriverState *p; |
67c095c8 | 2479 | int64_t eof = 0; |
a92b1b06 | 2480 | int dummy; |
1581a70d | 2481 | IO_CODE(); |
ba3f0e25 | 2482 | |
3555a432 | 2483 | assert(!include_base || base); /* Can't include NULL base */ |
7ff9579e | 2484 | assert_bdrv_graph_readable(); |
67c095c8 | 2485 | |
a92b1b06 EB |
2486 | if (!depth) { |
2487 | depth = &dummy; | |
2488 | } | |
2489 | *depth = 0; | |
2490 | ||
624f27bb VSO |
2491 | if (!include_base && bs == base) { |
2492 | *pnum = bytes; | |
2493 | return 0; | |
2494 | } | |
2495 | ||
67c095c8 | 2496 | ret = bdrv_co_block_status(bs, want_zero, offset, bytes, pnum, map, file); |
a92b1b06 | 2497 | ++*depth; |
3555a432 | 2498 | if (ret < 0 || *pnum == 0 || ret & BDRV_BLOCK_ALLOCATED || bs == base) { |
67c095c8 VSO |
2499 | return ret; |
2500 | } | |
2501 | ||
2502 | if (ret & BDRV_BLOCK_EOF) { | |
2503 | eof = offset + *pnum; | |
2504 | } | |
2505 | ||
2506 | assert(*pnum <= bytes); | |
2507 | bytes = *pnum; | |
2508 | ||
3555a432 | 2509 | for (p = bdrv_filter_or_cow_bs(bs); include_base || p != base; |
67c095c8 VSO |
2510 | p = bdrv_filter_or_cow_bs(p)) |
2511 | { | |
5b648c67 EB |
2512 | ret = bdrv_co_block_status(p, want_zero, offset, bytes, pnum, map, |
2513 | file); | |
a92b1b06 | 2514 | ++*depth; |
c61e684e | 2515 | if (ret < 0) { |
67c095c8 | 2516 | return ret; |
c61e684e | 2517 | } |
67c095c8 | 2518 | if (*pnum == 0) { |
c61e684e | 2519 | /* |
67c095c8 VSO |
2520 | * The top layer deferred to this layer, and because this layer is |
2521 | * short, any zeroes that we synthesize beyond EOF behave as if they | |
2522 | * were allocated at this layer. | |
2523 | * | |
2524 | * We don't include BDRV_BLOCK_EOF into ret, as upper layer may be | |
2525 | * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see | |
2526 | * below. | |
c61e684e | 2527 | */ |
67c095c8 | 2528 | assert(ret & BDRV_BLOCK_EOF); |
5b648c67 | 2529 | *pnum = bytes; |
67c095c8 VSO |
2530 | if (file) { |
2531 | *file = p; | |
2532 | } | |
2533 | ret = BDRV_BLOCK_ZERO | BDRV_BLOCK_ALLOCATED; | |
2534 | break; | |
c61e684e | 2535 | } |
67c095c8 VSO |
2536 | if (ret & BDRV_BLOCK_ALLOCATED) { |
2537 | /* | |
2538 | * We've found the node and the status, we must break. | |
2539 | * | |
2540 | * Drop BDRV_BLOCK_EOF, as it's not for upper layer, which may be | |
2541 | * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see | |
2542 | * below. | |
2543 | */ | |
2544 | ret &= ~BDRV_BLOCK_EOF; | |
ba3f0e25 FZ |
2545 | break; |
2546 | } | |
67c095c8 | 2547 | |
3555a432 VSO |
2548 | if (p == base) { |
2549 | assert(include_base); | |
2550 | break; | |
2551 | } | |
2552 | ||
67c095c8 VSO |
2553 | /* |
2554 | * OK, [offset, offset + *pnum) region is unallocated on this layer, | |
2555 | * let's continue the diving. | |
2556 | */ | |
2557 | assert(*pnum <= bytes); | |
2558 | bytes = *pnum; | |
ba3f0e25 | 2559 | } |
67c095c8 VSO |
2560 | |
2561 | if (offset + *pnum == eof) { | |
2562 | ret |= BDRV_BLOCK_EOF; | |
2563 | } | |
2564 | ||
ba3f0e25 FZ |
2565 | return ret; |
2566 | } | |
2567 | ||
7b52a921 EGE |
2568 | int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs, |
2569 | BlockDriverState *base, | |
2570 | int64_t offset, int64_t bytes, | |
2571 | int64_t *pnum, int64_t *map, | |
2572 | BlockDriverState **file) | |
2573 | { | |
2574 | IO_CODE(); | |
2575 | return bdrv_co_common_block_status_above(bs, base, false, true, offset, | |
2576 | bytes, pnum, map, file, NULL); | |
2577 | } | |
2578 | ||
31826642 EB |
2579 | int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base, |
2580 | int64_t offset, int64_t bytes, int64_t *pnum, | |
2581 | int64_t *map, BlockDriverState **file) | |
c9ce8c4d | 2582 | { |
384a48fb | 2583 | IO_CODE(); |
3555a432 | 2584 | return bdrv_common_block_status_above(bs, base, false, true, offset, bytes, |
a92b1b06 | 2585 | pnum, map, file, NULL); |
c9ce8c4d EB |
2586 | } |
2587 | ||
237d78f8 EB |
2588 | int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2589 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
ba3f0e25 | 2590 | { |
384a48fb | 2591 | IO_CODE(); |
cb850315 | 2592 | return bdrv_block_status_above(bs, bdrv_filter_or_cow_bs(bs), |
31826642 | 2593 | offset, bytes, pnum, map, file); |
ba3f0e25 FZ |
2594 | } |
2595 | ||
46cd1e8a AG |
2596 | /* |
2597 | * Check @bs (and its backing chain) to see if the range defined | |
2598 | * by @offset and @bytes is known to read as zeroes. | |
2599 | * Return 1 if that is the case, 0 otherwise and -errno on error. | |
2600 | * This test is meant to be fast rather than accurate so returning 0 | |
2601 | * does not guarantee non-zero data. | |
2602 | */ | |
2603 | int coroutine_fn bdrv_co_is_zero_fast(BlockDriverState *bs, int64_t offset, | |
2604 | int64_t bytes) | |
2605 | { | |
2606 | int ret; | |
2607 | int64_t pnum = bytes; | |
384a48fb | 2608 | IO_CODE(); |
46cd1e8a AG |
2609 | |
2610 | if (!bytes) { | |
2611 | return 1; | |
2612 | } | |
2613 | ||
ce47ff20 AF |
2614 | ret = bdrv_co_common_block_status_above(bs, NULL, false, false, offset, |
2615 | bytes, &pnum, NULL, NULL, NULL); | |
46cd1e8a AG |
2616 | |
2617 | if (ret < 0) { | |
2618 | return ret; | |
2619 | } | |
2620 | ||
2621 | return (pnum == bytes) && (ret & BDRV_BLOCK_ZERO); | |
2622 | } | |
2623 | ||
7b52a921 EGE |
2624 | int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t offset, |
2625 | int64_t bytes, int64_t *pnum) | |
2626 | { | |
2627 | int ret; | |
2628 | int64_t dummy; | |
2629 | IO_CODE(); | |
2630 | ||
2631 | ret = bdrv_co_common_block_status_above(bs, bs, true, false, offset, | |
2632 | bytes, pnum ? pnum : &dummy, NULL, | |
2633 | NULL, NULL); | |
2634 | if (ret < 0) { | |
2635 | return ret; | |
2636 | } | |
2637 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2638 | } | |
2639 | ||
7c85803c AF |
2640 | int bdrv_is_allocated(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2641 | int64_t *pnum) | |
61007b31 | 2642 | { |
7ddb99b9 EB |
2643 | int ret; |
2644 | int64_t dummy; | |
384a48fb | 2645 | IO_CODE(); |
d6a644bb | 2646 | |
3555a432 VSO |
2647 | ret = bdrv_common_block_status_above(bs, bs, true, false, offset, |
2648 | bytes, pnum ? pnum : &dummy, NULL, | |
a92b1b06 | 2649 | NULL, NULL); |
61007b31 SH |
2650 | if (ret < 0) { |
2651 | return ret; | |
2652 | } | |
2653 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2654 | } | |
2655 | ||
7b52a921 EGE |
2656 | /* See bdrv_is_allocated_above for documentation */ |
2657 | int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *top, | |
2658 | BlockDriverState *base, | |
2659 | bool include_base, int64_t offset, | |
2660 | int64_t bytes, int64_t *pnum) | |
2661 | { | |
2662 | int depth; | |
2663 | int ret; | |
2664 | IO_CODE(); | |
2665 | ||
2666 | ret = bdrv_co_common_block_status_above(top, base, include_base, false, | |
2667 | offset, bytes, pnum, NULL, NULL, | |
2668 | &depth); | |
2669 | if (ret < 0) { | |
2670 | return ret; | |
2671 | } | |
2672 | ||
2673 | if (ret & BDRV_BLOCK_ALLOCATED) { | |
2674 | return depth; | |
2675 | } | |
2676 | return 0; | |
2677 | } | |
2678 | ||
61007b31 SH |
2679 | /* |
2680 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
2681 | * | |
a92b1b06 EB |
2682 | * Return a positive depth if (a prefix of) the given range is allocated |
2683 | * in any image between BASE and TOP (BASE is only included if include_base | |
2684 | * is set). Depth 1 is TOP, 2 is the first backing layer, and so forth. | |
170d3bd3 AS |
2685 | * BASE can be NULL to check if the given offset is allocated in any |
2686 | * image of the chain. Return 0 otherwise, or negative errno on | |
2687 | * failure. | |
61007b31 | 2688 | * |
51b0a488 EB |
2689 | * 'pnum' is set to the number of bytes (including and immediately |
2690 | * following the specified offset) that are known to be in the same | |
2691 | * allocated/unallocated state. Note that a subsequent call starting | |
2692 | * at 'offset + *pnum' may return the same allocation status (in other | |
2693 | * words, the result is not necessarily the maximum possible range); | |
2694 | * but 'pnum' will only be 0 when end of file is reached. | |
61007b31 SH |
2695 | */ |
2696 | int bdrv_is_allocated_above(BlockDriverState *top, | |
2697 | BlockDriverState *base, | |
170d3bd3 AS |
2698 | bool include_base, int64_t offset, |
2699 | int64_t bytes, int64_t *pnum) | |
61007b31 | 2700 | { |
a92b1b06 | 2701 | int depth; |
7b52a921 | 2702 | int ret; |
384a48fb | 2703 | IO_CODE(); |
7b52a921 EGE |
2704 | |
2705 | ret = bdrv_common_block_status_above(top, base, include_base, false, | |
2706 | offset, bytes, pnum, NULL, NULL, | |
2707 | &depth); | |
7e7e5100 VSO |
2708 | if (ret < 0) { |
2709 | return ret; | |
61007b31 SH |
2710 | } |
2711 | ||
a92b1b06 EB |
2712 | if (ret & BDRV_BLOCK_ALLOCATED) { |
2713 | return depth; | |
2714 | } | |
2715 | return 0; | |
61007b31 SH |
2716 | } |
2717 | ||
21c2283e | 2718 | int coroutine_fn |
b33b354f | 2719 | bdrv_co_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) |
1a8ae822 KW |
2720 | { |
2721 | BlockDriver *drv = bs->drv; | |
c4db2e25 | 2722 | BlockDriverState *child_bs = bdrv_primary_bs(bs); |
b984b296 | 2723 | int ret; |
1581a70d | 2724 | IO_CODE(); |
1b3ff9fe | 2725 | assert_bdrv_graph_readable(); |
b984b296 VSO |
2726 | |
2727 | ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); | |
2728 | if (ret < 0) { | |
2729 | return ret; | |
2730 | } | |
dc88a467 | 2731 | |
b33b354f VSO |
2732 | if (!drv) { |
2733 | return -ENOMEDIUM; | |
2734 | } | |
2735 | ||
dc88a467 | 2736 | bdrv_inc_in_flight(bs); |
1a8ae822 | 2737 | |
ca5e2ad9 EGE |
2738 | if (drv->bdrv_co_load_vmstate) { |
2739 | ret = drv->bdrv_co_load_vmstate(bs, qiov, pos); | |
c4db2e25 | 2740 | } else if (child_bs) { |
b33b354f | 2741 | ret = bdrv_co_readv_vmstate(child_bs, qiov, pos); |
b984b296 VSO |
2742 | } else { |
2743 | ret = -ENOTSUP; | |
1a8ae822 KW |
2744 | } |
2745 | ||
dc88a467 | 2746 | bdrv_dec_in_flight(bs); |
b33b354f | 2747 | |
dc88a467 | 2748 | return ret; |
1a8ae822 KW |
2749 | } |
2750 | ||
b33b354f VSO |
2751 | int coroutine_fn |
2752 | bdrv_co_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 2753 | { |
b33b354f VSO |
2754 | BlockDriver *drv = bs->drv; |
2755 | BlockDriverState *child_bs = bdrv_primary_bs(bs); | |
b984b296 | 2756 | int ret; |
1581a70d | 2757 | IO_CODE(); |
1b3ff9fe | 2758 | assert_bdrv_graph_readable(); |
b984b296 VSO |
2759 | |
2760 | ret = bdrv_check_qiov_request(pos, qiov->size, qiov, 0, NULL); | |
2761 | if (ret < 0) { | |
2762 | return ret; | |
2763 | } | |
61007b31 | 2764 | |
b33b354f VSO |
2765 | if (!drv) { |
2766 | return -ENOMEDIUM; | |
b433d942 KW |
2767 | } |
2768 | ||
b33b354f | 2769 | bdrv_inc_in_flight(bs); |
61007b31 | 2770 | |
ca5e2ad9 EGE |
2771 | if (drv->bdrv_co_save_vmstate) { |
2772 | ret = drv->bdrv_co_save_vmstate(bs, qiov, pos); | |
b33b354f VSO |
2773 | } else if (child_bs) { |
2774 | ret = bdrv_co_writev_vmstate(child_bs, qiov, pos); | |
b984b296 VSO |
2775 | } else { |
2776 | ret = -ENOTSUP; | |
b33b354f VSO |
2777 | } |
2778 | ||
2779 | bdrv_dec_in_flight(bs); | |
2780 | ||
2781 | return ret; | |
61007b31 SH |
2782 | } |
2783 | ||
b33b354f | 2784 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
61007b31 | 2785 | int64_t pos, int size) |
5ddda0b8 | 2786 | { |
0d93ed08 | 2787 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b33b354f | 2788 | int ret = bdrv_writev_vmstate(bs, &qiov, pos); |
384a48fb | 2789 | IO_CODE(); |
b433d942 | 2790 | |
b33b354f | 2791 | return ret < 0 ? ret : size; |
5ddda0b8 KW |
2792 | } |
2793 | ||
b33b354f VSO |
2794 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, |
2795 | int64_t pos, int size) | |
61007b31 | 2796 | { |
b33b354f VSO |
2797 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
2798 | int ret = bdrv_readv_vmstate(bs, &qiov, pos); | |
384a48fb | 2799 | IO_CODE(); |
b33b354f VSO |
2800 | |
2801 | return ret < 0 ? ret : size; | |
61007b31 SH |
2802 | } |
2803 | ||
2804 | /**************************************************************/ | |
2805 | /* async I/Os */ | |
2806 | ||
61007b31 SH |
2807 | void bdrv_aio_cancel(BlockAIOCB *acb) |
2808 | { | |
384a48fb | 2809 | IO_CODE(); |
61007b31 SH |
2810 | qemu_aio_ref(acb); |
2811 | bdrv_aio_cancel_async(acb); | |
2812 | while (acb->refcnt > 1) { | |
2813 | if (acb->aiocb_info->get_aio_context) { | |
2814 | aio_poll(acb->aiocb_info->get_aio_context(acb), true); | |
2815 | } else if (acb->bs) { | |
2f47da5f PB |
2816 | /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so |
2817 | * assert that we're not using an I/O thread. Thread-safe | |
2818 | * code should use bdrv_aio_cancel_async exclusively. | |
2819 | */ | |
2820 | assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context()); | |
61007b31 SH |
2821 | aio_poll(bdrv_get_aio_context(acb->bs), true); |
2822 | } else { | |
2823 | abort(); | |
2824 | } | |
2825 | } | |
2826 | qemu_aio_unref(acb); | |
2827 | } | |
2828 | ||
2829 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2830 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2831 | * In either case the completion callback must be called. */ | |
2832 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2833 | { | |
384a48fb | 2834 | IO_CODE(); |
61007b31 SH |
2835 | if (acb->aiocb_info->cancel_async) { |
2836 | acb->aiocb_info->cancel_async(acb); | |
2837 | } | |
2838 | } | |
2839 | ||
61007b31 SH |
2840 | /**************************************************************/ |
2841 | /* Coroutine block device emulation */ | |
2842 | ||
61007b31 SH |
2843 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) |
2844 | { | |
883833e2 HR |
2845 | BdrvChild *primary_child = bdrv_primary_child(bs); |
2846 | BdrvChild *child; | |
49ca6259 FZ |
2847 | int current_gen; |
2848 | int ret = 0; | |
384a48fb | 2849 | IO_CODE(); |
49ca6259 | 2850 | |
88095349 | 2851 | assert_bdrv_graph_readable(); |
49ca6259 | 2852 | bdrv_inc_in_flight(bs); |
61007b31 | 2853 | |
1e97be91 | 2854 | if (!bdrv_co_is_inserted(bs) || bdrv_is_read_only(bs) || |
1b6bc94d | 2855 | bdrv_is_sg(bs)) { |
49ca6259 | 2856 | goto early_exit; |
61007b31 SH |
2857 | } |
2858 | ||
3783fa3d | 2859 | qemu_co_mutex_lock(&bs->reqs_lock); |
d73415a3 | 2860 | current_gen = qatomic_read(&bs->write_gen); |
3ff2f67a EY |
2861 | |
2862 | /* Wait until any previous flushes are completed */ | |
99723548 | 2863 | while (bs->active_flush_req) { |
3783fa3d | 2864 | qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock); |
3ff2f67a EY |
2865 | } |
2866 | ||
3783fa3d | 2867 | /* Flushes reach this point in nondecreasing current_gen order. */ |
99723548 | 2868 | bs->active_flush_req = true; |
3783fa3d | 2869 | qemu_co_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2870 | |
c32b82af PD |
2871 | /* Write back all layers by calling one driver function */ |
2872 | if (bs->drv->bdrv_co_flush) { | |
2873 | ret = bs->drv->bdrv_co_flush(bs); | |
2874 | goto out; | |
2875 | } | |
2876 | ||
61007b31 | 2877 | /* Write back cached data to the OS even with cache=unsafe */ |
883833e2 | 2878 | BLKDBG_EVENT(primary_child, BLKDBG_FLUSH_TO_OS); |
61007b31 SH |
2879 | if (bs->drv->bdrv_co_flush_to_os) { |
2880 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
2881 | if (ret < 0) { | |
cdb5e315 | 2882 | goto out; |
61007b31 SH |
2883 | } |
2884 | } | |
2885 | ||
2886 | /* But don't actually force it to the disk with cache=unsafe */ | |
2887 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
883833e2 | 2888 | goto flush_children; |
61007b31 SH |
2889 | } |
2890 | ||
3ff2f67a EY |
2891 | /* Check if we really need to flush anything */ |
2892 | if (bs->flushed_gen == current_gen) { | |
883833e2 | 2893 | goto flush_children; |
3ff2f67a EY |
2894 | } |
2895 | ||
883833e2 | 2896 | BLKDBG_EVENT(primary_child, BLKDBG_FLUSH_TO_DISK); |
d470ad42 HR |
2897 | if (!bs->drv) { |
2898 | /* bs->drv->bdrv_co_flush() might have ejected the BDS | |
2899 | * (even in case of apparent success) */ | |
2900 | ret = -ENOMEDIUM; | |
2901 | goto out; | |
2902 | } | |
61007b31 SH |
2903 | if (bs->drv->bdrv_co_flush_to_disk) { |
2904 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
2905 | } else if (bs->drv->bdrv_aio_flush) { | |
2906 | BlockAIOCB *acb; | |
2907 | CoroutineIOCompletion co = { | |
2908 | .coroutine = qemu_coroutine_self(), | |
2909 | }; | |
2910 | ||
2911 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
2912 | if (acb == NULL) { | |
2913 | ret = -EIO; | |
2914 | } else { | |
2915 | qemu_coroutine_yield(); | |
2916 | ret = co.ret; | |
2917 | } | |
2918 | } else { | |
2919 | /* | |
2920 | * Some block drivers always operate in either writethrough or unsafe | |
2921 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
2922 | * know how the server works (because the behaviour is hardcoded or | |
2923 | * depends on server-side configuration), so we can't ensure that | |
2924 | * everything is safe on disk. Returning an error doesn't work because | |
2925 | * that would break guests even if the server operates in writethrough | |
2926 | * mode. | |
2927 | * | |
2928 | * Let's hope the user knows what he's doing. | |
2929 | */ | |
2930 | ret = 0; | |
2931 | } | |
3ff2f67a | 2932 | |
61007b31 | 2933 | if (ret < 0) { |
cdb5e315 | 2934 | goto out; |
61007b31 SH |
2935 | } |
2936 | ||
2937 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
2938 | * in the case of cache=unsafe, so there are no useless flushes. | |
2939 | */ | |
883833e2 HR |
2940 | flush_children: |
2941 | ret = 0; | |
2942 | QLIST_FOREACH(child, &bs->children, next) { | |
2943 | if (child->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED)) { | |
2944 | int this_child_ret = bdrv_co_flush(child->bs); | |
2945 | if (!ret) { | |
2946 | ret = this_child_ret; | |
2947 | } | |
2948 | } | |
2949 | } | |
2950 | ||
cdb5e315 | 2951 | out: |
3ff2f67a | 2952 | /* Notify any pending flushes that we have completed */ |
e6af1e08 KW |
2953 | if (ret == 0) { |
2954 | bs->flushed_gen = current_gen; | |
2955 | } | |
3783fa3d PB |
2956 | |
2957 | qemu_co_mutex_lock(&bs->reqs_lock); | |
99723548 | 2958 | bs->active_flush_req = false; |
156af3ac DL |
2959 | /* Return value is ignored - it's ok if wait queue is empty */ |
2960 | qemu_co_queue_next(&bs->flush_queue); | |
3783fa3d | 2961 | qemu_co_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2962 | |
49ca6259 | 2963 | early_exit: |
99723548 | 2964 | bdrv_dec_in_flight(bs); |
cdb5e315 | 2965 | return ret; |
61007b31 SH |
2966 | } |
2967 | ||
d93e5726 VSO |
2968 | int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, |
2969 | int64_t bytes) | |
61007b31 | 2970 | { |
b1066c87 | 2971 | BdrvTrackedRequest req; |
39af49c0 VSO |
2972 | int ret; |
2973 | int64_t max_pdiscard; | |
3482b9bc | 2974 | int head, tail, align; |
0b9fd3f4 | 2975 | BlockDriverState *bs = child->bs; |
384a48fb | 2976 | IO_CODE(); |
9a5a1c62 | 2977 | assert_bdrv_graph_readable(); |
61007b31 | 2978 | |
1e97be91 | 2979 | if (!bs || !bs->drv || !bdrv_co_is_inserted(bs)) { |
61007b31 SH |
2980 | return -ENOMEDIUM; |
2981 | } | |
2982 | ||
d6883bc9 VSO |
2983 | if (bdrv_has_readonly_bitmaps(bs)) { |
2984 | return -EPERM; | |
2985 | } | |
2986 | ||
69b55e03 | 2987 | ret = bdrv_check_request(offset, bytes, NULL); |
8b117001 VSO |
2988 | if (ret < 0) { |
2989 | return ret; | |
61007b31 SH |
2990 | } |
2991 | ||
61007b31 SH |
2992 | /* Do nothing if disabled. */ |
2993 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
2994 | return 0; | |
2995 | } | |
2996 | ||
02aefe43 | 2997 | if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) { |
61007b31 SH |
2998 | return 0; |
2999 | } | |
3000 | ||
0bc329fb HR |
3001 | /* Invalidate the cached block-status data range if this discard overlaps */ |
3002 | bdrv_bsc_invalidate_range(bs, offset, bytes); | |
3003 | ||
3482b9bc EB |
3004 | /* Discard is advisory, but some devices track and coalesce |
3005 | * unaligned requests, so we must pass everything down rather than | |
3006 | * round here. Still, most devices will just silently ignore | |
3007 | * unaligned requests (by returning -ENOTSUP), so we must fragment | |
3008 | * the request accordingly. */ | |
02aefe43 | 3009 | align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment); |
b8d0a980 EB |
3010 | assert(align % bs->bl.request_alignment == 0); |
3011 | head = offset % align; | |
f5a5ca79 | 3012 | tail = (offset + bytes) % align; |
9f1963b3 | 3013 | |
99723548 | 3014 | bdrv_inc_in_flight(bs); |
f5a5ca79 | 3015 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD); |
50824995 | 3016 | |
00695c27 | 3017 | ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0); |
ec050f77 DL |
3018 | if (ret < 0) { |
3019 | goto out; | |
3020 | } | |
3021 | ||
6a8f3dbb | 3022 | max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT64_MAX), |
9f1963b3 | 3023 | align); |
3482b9bc | 3024 | assert(max_pdiscard >= bs->bl.request_alignment); |
61007b31 | 3025 | |
f5a5ca79 | 3026 | while (bytes > 0) { |
d93e5726 | 3027 | int64_t num = bytes; |
3482b9bc EB |
3028 | |
3029 | if (head) { | |
3030 | /* Make small requests to get to alignment boundaries. */ | |
f5a5ca79 | 3031 | num = MIN(bytes, align - head); |
3482b9bc EB |
3032 | if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) { |
3033 | num %= bs->bl.request_alignment; | |
3034 | } | |
3035 | head = (head + num) % align; | |
3036 | assert(num < max_pdiscard); | |
3037 | } else if (tail) { | |
3038 | if (num > align) { | |
3039 | /* Shorten the request to the last aligned cluster. */ | |
3040 | num -= tail; | |
3041 | } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) && | |
3042 | tail > bs->bl.request_alignment) { | |
3043 | tail %= bs->bl.request_alignment; | |
3044 | num -= tail; | |
3045 | } | |
3046 | } | |
3047 | /* limit request size */ | |
3048 | if (num > max_pdiscard) { | |
3049 | num = max_pdiscard; | |
3050 | } | |
61007b31 | 3051 | |
d470ad42 HR |
3052 | if (!bs->drv) { |
3053 | ret = -ENOMEDIUM; | |
3054 | goto out; | |
3055 | } | |
47a5486d EB |
3056 | if (bs->drv->bdrv_co_pdiscard) { |
3057 | ret = bs->drv->bdrv_co_pdiscard(bs, offset, num); | |
61007b31 SH |
3058 | } else { |
3059 | BlockAIOCB *acb; | |
3060 | CoroutineIOCompletion co = { | |
3061 | .coroutine = qemu_coroutine_self(), | |
3062 | }; | |
3063 | ||
4da444a0 EB |
3064 | acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num, |
3065 | bdrv_co_io_em_complete, &co); | |
61007b31 | 3066 | if (acb == NULL) { |
b1066c87 FZ |
3067 | ret = -EIO; |
3068 | goto out; | |
61007b31 SH |
3069 | } else { |
3070 | qemu_coroutine_yield(); | |
3071 | ret = co.ret; | |
3072 | } | |
3073 | } | |
3074 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 3075 | goto out; |
61007b31 SH |
3076 | } |
3077 | ||
9f1963b3 | 3078 | offset += num; |
f5a5ca79 | 3079 | bytes -= num; |
61007b31 | 3080 | } |
b1066c87 FZ |
3081 | ret = 0; |
3082 | out: | |
00695c27 | 3083 | bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret); |
b1066c87 | 3084 | tracked_request_end(&req); |
99723548 | 3085 | bdrv_dec_in_flight(bs); |
b1066c87 | 3086 | return ret; |
61007b31 SH |
3087 | } |
3088 | ||
881a4c55 | 3089 | int coroutine_fn bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
3090 | { |
3091 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
3092 | CoroutineIOCompletion co = { |
3093 | .coroutine = qemu_coroutine_self(), | |
3094 | }; | |
3095 | BlockAIOCB *acb; | |
384a48fb | 3096 | IO_CODE(); |
26c518ab | 3097 | assert_bdrv_graph_readable(); |
61007b31 | 3098 | |
99723548 | 3099 | bdrv_inc_in_flight(bs); |
16a389dc | 3100 | if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) { |
5c5ae76a FZ |
3101 | co.ret = -ENOTSUP; |
3102 | goto out; | |
3103 | } | |
3104 | ||
16a389dc KW |
3105 | if (drv->bdrv_co_ioctl) { |
3106 | co.ret = drv->bdrv_co_ioctl(bs, req, buf); | |
3107 | } else { | |
3108 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
3109 | if (!acb) { | |
3110 | co.ret = -ENOTSUP; | |
3111 | goto out; | |
3112 | } | |
3113 | qemu_coroutine_yield(); | |
5c5ae76a | 3114 | } |
5c5ae76a | 3115 | out: |
99723548 | 3116 | bdrv_dec_in_flight(bs); |
5c5ae76a FZ |
3117 | return co.ret; |
3118 | } | |
3119 | ||
6d43eaa3 SL |
3120 | int coroutine_fn bdrv_co_zone_report(BlockDriverState *bs, int64_t offset, |
3121 | unsigned int *nr_zones, | |
3122 | BlockZoneDescriptor *zones) | |
3123 | { | |
3124 | BlockDriver *drv = bs->drv; | |
3125 | CoroutineIOCompletion co = { | |
3126 | .coroutine = qemu_coroutine_self(), | |
3127 | }; | |
3128 | IO_CODE(); | |
3129 | ||
3130 | bdrv_inc_in_flight(bs); | |
3131 | if (!drv || !drv->bdrv_co_zone_report || bs->bl.zoned == BLK_Z_NONE) { | |
3132 | co.ret = -ENOTSUP; | |
3133 | goto out; | |
3134 | } | |
3135 | co.ret = drv->bdrv_co_zone_report(bs, offset, nr_zones, zones); | |
3136 | out: | |
3137 | bdrv_dec_in_flight(bs); | |
3138 | return co.ret; | |
3139 | } | |
3140 | ||
3141 | int coroutine_fn bdrv_co_zone_mgmt(BlockDriverState *bs, BlockZoneOp op, | |
3142 | int64_t offset, int64_t len) | |
3143 | { | |
3144 | BlockDriver *drv = bs->drv; | |
3145 | CoroutineIOCompletion co = { | |
3146 | .coroutine = qemu_coroutine_self(), | |
3147 | }; | |
3148 | IO_CODE(); | |
3149 | ||
3150 | bdrv_inc_in_flight(bs); | |
3151 | if (!drv || !drv->bdrv_co_zone_mgmt || bs->bl.zoned == BLK_Z_NONE) { | |
3152 | co.ret = -ENOTSUP; | |
3153 | goto out; | |
3154 | } | |
3155 | co.ret = drv->bdrv_co_zone_mgmt(bs, op, offset, len); | |
3156 | out: | |
3157 | bdrv_dec_in_flight(bs); | |
3158 | return co.ret; | |
3159 | } | |
3160 | ||
4751d09a SL |
3161 | int coroutine_fn bdrv_co_zone_append(BlockDriverState *bs, int64_t *offset, |
3162 | QEMUIOVector *qiov, | |
3163 | BdrvRequestFlags flags) | |
3164 | { | |
3165 | int ret; | |
3166 | BlockDriver *drv = bs->drv; | |
3167 | CoroutineIOCompletion co = { | |
3168 | .coroutine = qemu_coroutine_self(), | |
3169 | }; | |
3170 | IO_CODE(); | |
3171 | ||
3172 | ret = bdrv_check_qiov_request(*offset, qiov->size, qiov, 0, NULL); | |
3173 | if (ret < 0) { | |
3174 | return ret; | |
3175 | } | |
3176 | ||
3177 | bdrv_inc_in_flight(bs); | |
3178 | if (!drv || !drv->bdrv_co_zone_append || bs->bl.zoned == BLK_Z_NONE) { | |
3179 | co.ret = -ENOTSUP; | |
3180 | goto out; | |
3181 | } | |
3182 | co.ret = drv->bdrv_co_zone_append(bs, offset, qiov, flags); | |
3183 | out: | |
3184 | bdrv_dec_in_flight(bs); | |
3185 | return co.ret; | |
3186 | } | |
3187 | ||
61007b31 SH |
3188 | void *qemu_blockalign(BlockDriverState *bs, size_t size) |
3189 | { | |
384a48fb | 3190 | IO_CODE(); |
61007b31 SH |
3191 | return qemu_memalign(bdrv_opt_mem_align(bs), size); |
3192 | } | |
3193 | ||
3194 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
3195 | { | |
384a48fb | 3196 | IO_CODE(); |
61007b31 SH |
3197 | return memset(qemu_blockalign(bs, size), 0, size); |
3198 | } | |
3199 | ||
3200 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
3201 | { | |
3202 | size_t align = bdrv_opt_mem_align(bs); | |
384a48fb | 3203 | IO_CODE(); |
61007b31 SH |
3204 | |
3205 | /* Ensure that NULL is never returned on success */ | |
3206 | assert(align > 0); | |
3207 | if (size == 0) { | |
3208 | size = align; | |
3209 | } | |
3210 | ||
3211 | return qemu_try_memalign(align, size); | |
3212 | } | |
3213 | ||
3214 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
3215 | { | |
3216 | void *mem = qemu_try_blockalign(bs, size); | |
384a48fb | 3217 | IO_CODE(); |
61007b31 SH |
3218 | |
3219 | if (mem) { | |
3220 | memset(mem, 0, size); | |
3221 | } | |
3222 | ||
3223 | return mem; | |
3224 | } | |
3225 | ||
f4ec04ba | 3226 | /* Helper that undoes bdrv_register_buf() when it fails partway through */ |
d9249c25 KW |
3227 | static void GRAPH_RDLOCK |
3228 | bdrv_register_buf_rollback(BlockDriverState *bs, void *host, size_t size, | |
3229 | BdrvChild *final_child) | |
f4ec04ba SH |
3230 | { |
3231 | BdrvChild *child; | |
3232 | ||
d9249c25 KW |
3233 | GLOBAL_STATE_CODE(); |
3234 | assert_bdrv_graph_readable(); | |
3235 | ||
f4ec04ba SH |
3236 | QLIST_FOREACH(child, &bs->children, next) { |
3237 | if (child == final_child) { | |
3238 | break; | |
3239 | } | |
3240 | ||
3241 | bdrv_unregister_buf(child->bs, host, size); | |
3242 | } | |
3243 | ||
3244 | if (bs->drv && bs->drv->bdrv_unregister_buf) { | |
3245 | bs->drv->bdrv_unregister_buf(bs, host, size); | |
3246 | } | |
3247 | } | |
3248 | ||
3249 | bool bdrv_register_buf(BlockDriverState *bs, void *host, size_t size, | |
3250 | Error **errp) | |
23d0ba93 FZ |
3251 | { |
3252 | BdrvChild *child; | |
3253 | ||
f791bf7f | 3254 | GLOBAL_STATE_CODE(); |
d9249c25 KW |
3255 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
3256 | ||
23d0ba93 | 3257 | if (bs->drv && bs->drv->bdrv_register_buf) { |
f4ec04ba SH |
3258 | if (!bs->drv->bdrv_register_buf(bs, host, size, errp)) { |
3259 | return false; | |
3260 | } | |
23d0ba93 FZ |
3261 | } |
3262 | QLIST_FOREACH(child, &bs->children, next) { | |
f4ec04ba SH |
3263 | if (!bdrv_register_buf(child->bs, host, size, errp)) { |
3264 | bdrv_register_buf_rollback(bs, host, size, child); | |
3265 | return false; | |
3266 | } | |
23d0ba93 | 3267 | } |
f4ec04ba | 3268 | return true; |
23d0ba93 FZ |
3269 | } |
3270 | ||
4f384011 | 3271 | void bdrv_unregister_buf(BlockDriverState *bs, void *host, size_t size) |
23d0ba93 FZ |
3272 | { |
3273 | BdrvChild *child; | |
3274 | ||
f791bf7f | 3275 | GLOBAL_STATE_CODE(); |
d9249c25 KW |
3276 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
3277 | ||
23d0ba93 | 3278 | if (bs->drv && bs->drv->bdrv_unregister_buf) { |
4f384011 | 3279 | bs->drv->bdrv_unregister_buf(bs, host, size); |
23d0ba93 FZ |
3280 | } |
3281 | QLIST_FOREACH(child, &bs->children, next) { | |
4f384011 | 3282 | bdrv_unregister_buf(child->bs, host, size); |
23d0ba93 FZ |
3283 | } |
3284 | } | |
fcc67678 | 3285 | |
abaf8b75 | 3286 | static int coroutine_fn GRAPH_RDLOCK bdrv_co_copy_range_internal( |
a5215b8f VSO |
3287 | BdrvChild *src, int64_t src_offset, BdrvChild *dst, |
3288 | int64_t dst_offset, int64_t bytes, | |
67b51fb9 VSO |
3289 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags, |
3290 | bool recurse_src) | |
fcc67678 | 3291 | { |
999658a0 | 3292 | BdrvTrackedRequest req; |
fcc67678 | 3293 | int ret; |
742bf09b | 3294 | assert_bdrv_graph_readable(); |
fcc67678 | 3295 | |
fe0480d6 KW |
3296 | /* TODO We can support BDRV_REQ_NO_FALLBACK here */ |
3297 | assert(!(read_flags & BDRV_REQ_NO_FALLBACK)); | |
3298 | assert(!(write_flags & BDRV_REQ_NO_FALLBACK)); | |
45e62b46 VSO |
3299 | assert(!(read_flags & BDRV_REQ_NO_WAIT)); |
3300 | assert(!(write_flags & BDRV_REQ_NO_WAIT)); | |
fe0480d6 | 3301 | |
1e97be91 | 3302 | if (!dst || !dst->bs || !bdrv_co_is_inserted(dst->bs)) { |
fcc67678 FZ |
3303 | return -ENOMEDIUM; |
3304 | } | |
63f4ad11 | 3305 | ret = bdrv_check_request32(dst_offset, bytes, NULL, 0); |
fcc67678 FZ |
3306 | if (ret) { |
3307 | return ret; | |
3308 | } | |
67b51fb9 VSO |
3309 | if (write_flags & BDRV_REQ_ZERO_WRITE) { |
3310 | return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags); | |
fcc67678 FZ |
3311 | } |
3312 | ||
1e97be91 | 3313 | if (!src || !src->bs || !bdrv_co_is_inserted(src->bs)) { |
d4d3e5a0 FZ |
3314 | return -ENOMEDIUM; |
3315 | } | |
63f4ad11 | 3316 | ret = bdrv_check_request32(src_offset, bytes, NULL, 0); |
d4d3e5a0 FZ |
3317 | if (ret) { |
3318 | return ret; | |
3319 | } | |
3320 | ||
fcc67678 FZ |
3321 | if (!src->bs->drv->bdrv_co_copy_range_from |
3322 | || !dst->bs->drv->bdrv_co_copy_range_to | |
3323 | || src->bs->encrypted || dst->bs->encrypted) { | |
3324 | return -ENOTSUP; | |
3325 | } | |
37aec7d7 | 3326 | |
fcc67678 | 3327 | if (recurse_src) { |
999658a0 VSO |
3328 | bdrv_inc_in_flight(src->bs); |
3329 | tracked_request_begin(&req, src->bs, src_offset, bytes, | |
3330 | BDRV_TRACKED_READ); | |
3331 | ||
09d2f948 VSO |
3332 | /* BDRV_REQ_SERIALISING is only for write operation */ |
3333 | assert(!(read_flags & BDRV_REQ_SERIALISING)); | |
c53cb427 | 3334 | bdrv_wait_serialising_requests(&req); |
999658a0 | 3335 | |
37aec7d7 FZ |
3336 | ret = src->bs->drv->bdrv_co_copy_range_from(src->bs, |
3337 | src, src_offset, | |
3338 | dst, dst_offset, | |
67b51fb9 VSO |
3339 | bytes, |
3340 | read_flags, write_flags); | |
999658a0 VSO |
3341 | |
3342 | tracked_request_end(&req); | |
3343 | bdrv_dec_in_flight(src->bs); | |
fcc67678 | 3344 | } else { |
999658a0 VSO |
3345 | bdrv_inc_in_flight(dst->bs); |
3346 | tracked_request_begin(&req, dst->bs, dst_offset, bytes, | |
3347 | BDRV_TRACKED_WRITE); | |
0eb1e891 FZ |
3348 | ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req, |
3349 | write_flags); | |
3350 | if (!ret) { | |
3351 | ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs, | |
3352 | src, src_offset, | |
3353 | dst, dst_offset, | |
3354 | bytes, | |
3355 | read_flags, write_flags); | |
3356 | } | |
3357 | bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret); | |
999658a0 VSO |
3358 | tracked_request_end(&req); |
3359 | bdrv_dec_in_flight(dst->bs); | |
fcc67678 | 3360 | } |
999658a0 | 3361 | |
37aec7d7 | 3362 | return ret; |
fcc67678 FZ |
3363 | } |
3364 | ||
3365 | /* Copy range from @src to @dst. | |
3366 | * | |
3367 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3368 | * semantics. */ | |
a5215b8f VSO |
3369 | int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, int64_t src_offset, |
3370 | BdrvChild *dst, int64_t dst_offset, | |
3371 | int64_t bytes, | |
67b51fb9 VSO |
3372 | BdrvRequestFlags read_flags, |
3373 | BdrvRequestFlags write_flags) | |
fcc67678 | 3374 | { |
967d7905 | 3375 | IO_CODE(); |
742bf09b | 3376 | assert_bdrv_graph_readable(); |
ecc983a5 FZ |
3377 | trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes, |
3378 | read_flags, write_flags); | |
fcc67678 | 3379 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3380 | bytes, read_flags, write_flags, true); |
fcc67678 FZ |
3381 | } |
3382 | ||
3383 | /* Copy range from @src to @dst. | |
3384 | * | |
3385 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3386 | * semantics. */ | |
a5215b8f VSO |
3387 | int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, int64_t src_offset, |
3388 | BdrvChild *dst, int64_t dst_offset, | |
3389 | int64_t bytes, | |
67b51fb9 VSO |
3390 | BdrvRequestFlags read_flags, |
3391 | BdrvRequestFlags write_flags) | |
fcc67678 | 3392 | { |
967d7905 | 3393 | IO_CODE(); |
742bf09b | 3394 | assert_bdrv_graph_readable(); |
ecc983a5 FZ |
3395 | trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3396 | read_flags, write_flags); | |
fcc67678 | 3397 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3398 | bytes, read_flags, write_flags, false); |
fcc67678 FZ |
3399 | } |
3400 | ||
a5215b8f VSO |
3401 | int coroutine_fn bdrv_co_copy_range(BdrvChild *src, int64_t src_offset, |
3402 | BdrvChild *dst, int64_t dst_offset, | |
3403 | int64_t bytes, BdrvRequestFlags read_flags, | |
67b51fb9 | 3404 | BdrvRequestFlags write_flags) |
fcc67678 | 3405 | { |
384a48fb | 3406 | IO_CODE(); |
742bf09b EGE |
3407 | assert_bdrv_graph_readable(); |
3408 | ||
37aec7d7 FZ |
3409 | return bdrv_co_copy_range_from(src, src_offset, |
3410 | dst, dst_offset, | |
67b51fb9 | 3411 | bytes, read_flags, write_flags); |
fcc67678 | 3412 | } |
3d9f2d2a KW |
3413 | |
3414 | static void bdrv_parent_cb_resize(BlockDriverState *bs) | |
3415 | { | |
3416 | BdrvChild *c; | |
3417 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
bd86fb99 HR |
3418 | if (c->klass->resize) { |
3419 | c->klass->resize(c); | |
3d9f2d2a KW |
3420 | } |
3421 | } | |
3422 | } | |
3423 | ||
3424 | /** | |
3425 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
c80d8b06 HR |
3426 | * |
3427 | * If 'exact' is true, the file must be resized to exactly the given | |
3428 | * 'offset'. Otherwise, it is sufficient for the node to be at least | |
3429 | * 'offset' bytes in length. | |
3d9f2d2a | 3430 | */ |
c80d8b06 | 3431 | int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact, |
7b8e4857 KW |
3432 | PreallocMode prealloc, BdrvRequestFlags flags, |
3433 | Error **errp) | |
3d9f2d2a KW |
3434 | { |
3435 | BlockDriverState *bs = child->bs; | |
23b93525 | 3436 | BdrvChild *filtered, *backing; |
3d9f2d2a | 3437 | BlockDriver *drv = bs->drv; |
1bc5f09f KW |
3438 | BdrvTrackedRequest req; |
3439 | int64_t old_size, new_bytes; | |
3d9f2d2a | 3440 | int ret; |
384a48fb | 3441 | IO_CODE(); |
c2b8e315 | 3442 | assert_bdrv_graph_readable(); |
3d9f2d2a KW |
3443 | |
3444 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ | |
3445 | if (!drv) { | |
3446 | error_setg(errp, "No medium inserted"); | |
3447 | return -ENOMEDIUM; | |
3448 | } | |
3449 | if (offset < 0) { | |
3450 | error_setg(errp, "Image size cannot be negative"); | |
3451 | return -EINVAL; | |
3452 | } | |
3453 | ||
69b55e03 | 3454 | ret = bdrv_check_request(offset, 0, errp); |
8b117001 | 3455 | if (ret < 0) { |
8b117001 VSO |
3456 | return ret; |
3457 | } | |
3458 | ||
1bc5f09f KW |
3459 | old_size = bdrv_getlength(bs); |
3460 | if (old_size < 0) { | |
3461 | error_setg_errno(errp, -old_size, "Failed to get old image size"); | |
3462 | return old_size; | |
3463 | } | |
3464 | ||
97efa869 EB |
3465 | if (bdrv_is_read_only(bs)) { |
3466 | error_setg(errp, "Image is read-only"); | |
3467 | return -EACCES; | |
3468 | } | |
3469 | ||
1bc5f09f KW |
3470 | if (offset > old_size) { |
3471 | new_bytes = offset - old_size; | |
3472 | } else { | |
3473 | new_bytes = 0; | |
3474 | } | |
3475 | ||
3d9f2d2a | 3476 | bdrv_inc_in_flight(bs); |
5416a11e FZ |
3477 | tracked_request_begin(&req, bs, offset - new_bytes, new_bytes, |
3478 | BDRV_TRACKED_TRUNCATE); | |
1bc5f09f KW |
3479 | |
3480 | /* If we are growing the image and potentially using preallocation for the | |
3481 | * new area, we need to make sure that no write requests are made to it | |
3482 | * concurrently or they might be overwritten by preallocation. */ | |
3483 | if (new_bytes) { | |
8ac5aab2 | 3484 | bdrv_make_request_serialising(&req, 1); |
cd47d792 | 3485 | } |
cd47d792 FZ |
3486 | ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req, |
3487 | 0); | |
3488 | if (ret < 0) { | |
3489 | error_setg_errno(errp, -ret, | |
3490 | "Failed to prepare request for truncation"); | |
3491 | goto out; | |
1bc5f09f | 3492 | } |
3d9f2d2a | 3493 | |
93393e69 | 3494 | filtered = bdrv_filter_child(bs); |
23b93525 | 3495 | backing = bdrv_cow_child(bs); |
93393e69 | 3496 | |
955c7d66 KW |
3497 | /* |
3498 | * If the image has a backing file that is large enough that it would | |
3499 | * provide data for the new area, we cannot leave it unallocated because | |
3500 | * then the backing file content would become visible. Instead, zero-fill | |
3501 | * the new area. | |
3502 | * | |
3503 | * Note that if the image has a backing file, but was opened without the | |
3504 | * backing file, taking care of keeping things consistent with that backing | |
3505 | * file is the user's responsibility. | |
3506 | */ | |
23b93525 | 3507 | if (new_bytes && backing) { |
955c7d66 KW |
3508 | int64_t backing_len; |
3509 | ||
bd53086e | 3510 | backing_len = bdrv_co_getlength(backing->bs); |
955c7d66 KW |
3511 | if (backing_len < 0) { |
3512 | ret = backing_len; | |
3513 | error_setg_errno(errp, -ret, "Could not get backing file size"); | |
3514 | goto out; | |
3515 | } | |
3516 | ||
3517 | if (backing_len > old_size) { | |
3518 | flags |= BDRV_REQ_ZERO_WRITE; | |
3519 | } | |
3520 | } | |
3521 | ||
6b7e8f8b | 3522 | if (drv->bdrv_co_truncate) { |
92b92799 KW |
3523 | if (flags & ~bs->supported_truncate_flags) { |
3524 | error_setg(errp, "Block driver does not support requested flags"); | |
3525 | ret = -ENOTSUP; | |
3526 | goto out; | |
3527 | } | |
3528 | ret = drv->bdrv_co_truncate(bs, offset, exact, prealloc, flags, errp); | |
93393e69 HR |
3529 | } else if (filtered) { |
3530 | ret = bdrv_co_truncate(filtered, offset, exact, prealloc, flags, errp); | |
6b7e8f8b | 3531 | } else { |
3d9f2d2a KW |
3532 | error_setg(errp, "Image format driver does not support resize"); |
3533 | ret = -ENOTSUP; | |
3534 | goto out; | |
3535 | } | |
3d9f2d2a KW |
3536 | if (ret < 0) { |
3537 | goto out; | |
3538 | } | |
6b7e8f8b | 3539 | |
bd53086e | 3540 | ret = bdrv_co_refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); |
3d9f2d2a KW |
3541 | if (ret < 0) { |
3542 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
3543 | } else { | |
3544 | offset = bs->total_sectors * BDRV_SECTOR_SIZE; | |
3545 | } | |
c057960c EGE |
3546 | /* |
3547 | * It's possible that truncation succeeded but bdrv_refresh_total_sectors | |
cd47d792 | 3548 | * failed, but the latter doesn't affect how we should finish the request. |
c057960c EGE |
3549 | * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. |
3550 | */ | |
cd47d792 | 3551 | bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0); |
3d9f2d2a KW |
3552 | |
3553 | out: | |
1bc5f09f | 3554 | tracked_request_end(&req); |
3d9f2d2a | 3555 | bdrv_dec_in_flight(bs); |
1bc5f09f | 3556 | |
3d9f2d2a KW |
3557 | return ret; |
3558 | } | |
bd54669a VSO |
3559 | |
3560 | void bdrv_cancel_in_flight(BlockDriverState *bs) | |
3561 | { | |
f791bf7f | 3562 | GLOBAL_STATE_CODE(); |
bd54669a VSO |
3563 | if (!bs || !bs->drv) { |
3564 | return; | |
3565 | } | |
3566 | ||
3567 | if (bs->drv->bdrv_cancel_in_flight) { | |
3568 | bs->drv->bdrv_cancel_in_flight(bs); | |
3569 | } | |
3570 | } | |
ce14f3b4 VSO |
3571 | |
3572 | int coroutine_fn | |
3573 | bdrv_co_preadv_snapshot(BdrvChild *child, int64_t offset, int64_t bytes, | |
3574 | QEMUIOVector *qiov, size_t qiov_offset) | |
3575 | { | |
3576 | BlockDriverState *bs = child->bs; | |
3577 | BlockDriver *drv = bs->drv; | |
3578 | int ret; | |
3579 | IO_CODE(); | |
7b9e8b22 | 3580 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3581 | |
3582 | if (!drv) { | |
3583 | return -ENOMEDIUM; | |
3584 | } | |
3585 | ||
3586 | if (!drv->bdrv_co_preadv_snapshot) { | |
3587 | return -ENOTSUP; | |
3588 | } | |
3589 | ||
3590 | bdrv_inc_in_flight(bs); | |
3591 | ret = drv->bdrv_co_preadv_snapshot(bs, offset, bytes, qiov, qiov_offset); | |
3592 | bdrv_dec_in_flight(bs); | |
3593 | ||
3594 | return ret; | |
3595 | } | |
3596 | ||
3597 | int coroutine_fn | |
3598 | bdrv_co_snapshot_block_status(BlockDriverState *bs, | |
3599 | bool want_zero, int64_t offset, int64_t bytes, | |
3600 | int64_t *pnum, int64_t *map, | |
3601 | BlockDriverState **file) | |
3602 | { | |
3603 | BlockDriver *drv = bs->drv; | |
3604 | int ret; | |
3605 | IO_CODE(); | |
7b9e8b22 | 3606 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3607 | |
3608 | if (!drv) { | |
3609 | return -ENOMEDIUM; | |
3610 | } | |
3611 | ||
3612 | if (!drv->bdrv_co_snapshot_block_status) { | |
3613 | return -ENOTSUP; | |
3614 | } | |
3615 | ||
3616 | bdrv_inc_in_flight(bs); | |
3617 | ret = drv->bdrv_co_snapshot_block_status(bs, want_zero, offset, bytes, | |
3618 | pnum, map, file); | |
3619 | bdrv_dec_in_flight(bs); | |
3620 | ||
3621 | return ret; | |
3622 | } | |
3623 | ||
3624 | int coroutine_fn | |
3625 | bdrv_co_pdiscard_snapshot(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
3626 | { | |
3627 | BlockDriver *drv = bs->drv; | |
3628 | int ret; | |
3629 | IO_CODE(); | |
9a5a1c62 | 3630 | assert_bdrv_graph_readable(); |
ce14f3b4 VSO |
3631 | |
3632 | if (!drv) { | |
3633 | return -ENOMEDIUM; | |
3634 | } | |
3635 | ||
3636 | if (!drv->bdrv_co_pdiscard_snapshot) { | |
3637 | return -ENOTSUP; | |
3638 | } | |
3639 | ||
3640 | bdrv_inc_in_flight(bs); | |
3641 | ret = drv->bdrv_co_pdiscard_snapshot(bs, offset, bytes); | |
3642 | bdrv_dec_in_flight(bs); | |
3643 | ||
3644 | return ret; | |
3645 | } |