]>
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" |
f348b6d1 | 32 | #include "qemu/cutils.h" |
da34e65c | 33 | #include "qapi/error.h" |
d49b6836 | 34 | #include "qemu/error-report.h" |
db725815 | 35 | #include "qemu/main-loop.h" |
61007b31 SH |
36 | |
37 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
38 | ||
cb2e2878 EB |
39 | /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */ |
40 | #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS) | |
41 | ||
7f8f03ef | 42 | static void bdrv_parent_cb_resize(BlockDriverState *bs); |
d05aa8bb | 43 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 44 | int64_t offset, int bytes, BdrvRequestFlags flags); |
61007b31 | 45 | |
f4c8a43b HR |
46 | static void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore, |
47 | bool ignore_bds_parents) | |
61007b31 | 48 | { |
02d21300 | 49 | BdrvChild *c, *next; |
27ccdd52 | 50 | |
02d21300 | 51 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { |
6cd5c9d7 | 52 | if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { |
0152bf40 KW |
53 | continue; |
54 | } | |
4be6a6d1 | 55 | bdrv_parent_drained_begin_single(c, false); |
ce0f1412 PB |
56 | } |
57 | } | |
61007b31 | 58 | |
e037c09c HR |
59 | static void bdrv_parent_drained_end_single_no_poll(BdrvChild *c, |
60 | int *drained_end_counter) | |
804db8ea HR |
61 | { |
62 | assert(c->parent_quiesce_counter > 0); | |
63 | c->parent_quiesce_counter--; | |
64 | if (c->role->drained_end) { | |
e037c09c | 65 | c->role->drained_end(c, drained_end_counter); |
804db8ea HR |
66 | } |
67 | } | |
68 | ||
e037c09c HR |
69 | void bdrv_parent_drained_end_single(BdrvChild *c) |
70 | { | |
71 | int drained_end_counter = 0; | |
72 | bdrv_parent_drained_end_single_no_poll(c, &drained_end_counter); | |
73 | BDRV_POLL_WHILE(c->bs, atomic_read(&drained_end_counter) > 0); | |
74 | } | |
75 | ||
f4c8a43b | 76 | static void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore, |
e037c09c HR |
77 | bool ignore_bds_parents, |
78 | int *drained_end_counter) | |
ce0f1412 | 79 | { |
61ad631c | 80 | BdrvChild *c; |
27ccdd52 | 81 | |
61ad631c | 82 | QLIST_FOREACH(c, &bs->parents, next_parent) { |
6cd5c9d7 | 83 | if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { |
0152bf40 KW |
84 | continue; |
85 | } | |
e037c09c | 86 | bdrv_parent_drained_end_single_no_poll(c, drained_end_counter); |
27ccdd52 | 87 | } |
61007b31 SH |
88 | } |
89 | ||
4be6a6d1 KW |
90 | static bool bdrv_parent_drained_poll_single(BdrvChild *c) |
91 | { | |
92 | if (c->role->drained_poll) { | |
93 | return c->role->drained_poll(c); | |
94 | } | |
95 | return false; | |
96 | } | |
97 | ||
6cd5c9d7 KW |
98 | static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore, |
99 | bool ignore_bds_parents) | |
89bd0305 KW |
100 | { |
101 | BdrvChild *c, *next; | |
102 | bool busy = false; | |
103 | ||
104 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { | |
6cd5c9d7 | 105 | if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { |
89bd0305 KW |
106 | continue; |
107 | } | |
4be6a6d1 | 108 | busy |= bdrv_parent_drained_poll_single(c); |
89bd0305 KW |
109 | } |
110 | ||
111 | return busy; | |
112 | } | |
113 | ||
4be6a6d1 KW |
114 | void bdrv_parent_drained_begin_single(BdrvChild *c, bool poll) |
115 | { | |
804db8ea | 116 | c->parent_quiesce_counter++; |
4be6a6d1 KW |
117 | if (c->role->drained_begin) { |
118 | c->role->drained_begin(c); | |
119 | } | |
120 | if (poll) { | |
121 | BDRV_POLL_WHILE(c->bs, bdrv_parent_drained_poll_single(c)); | |
122 | } | |
123 | } | |
124 | ||
d9e0dfa2 EB |
125 | static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) |
126 | { | |
127 | dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); | |
128 | dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer); | |
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); | |
134 | } | |
135 | ||
61007b31 SH |
136 | void bdrv_refresh_limits(BlockDriverState *bs, Error **errp) |
137 | { | |
138 | BlockDriver *drv = bs->drv; | |
139 | Error *local_err = NULL; | |
140 | ||
141 | memset(&bs->bl, 0, sizeof(bs->bl)); | |
142 | ||
143 | if (!drv) { | |
144 | return; | |
145 | } | |
146 | ||
79ba8c98 | 147 | /* Default alignment based on whether driver has byte interface */ |
e31f6864 | 148 | bs->bl.request_alignment = (drv->bdrv_co_preadv || |
ac850bf0 VSO |
149 | drv->bdrv_aio_preadv || |
150 | drv->bdrv_co_preadv_part) ? 1 : 512; | |
79ba8c98 | 151 | |
61007b31 SH |
152 | /* Take some limits from the children as a default */ |
153 | if (bs->file) { | |
9a4f4c31 | 154 | bdrv_refresh_limits(bs->file->bs, &local_err); |
61007b31 SH |
155 | if (local_err) { |
156 | error_propagate(errp, local_err); | |
157 | return; | |
158 | } | |
d9e0dfa2 | 159 | bdrv_merge_limits(&bs->bl, &bs->file->bs->bl); |
61007b31 | 160 | } else { |
4196d2f0 | 161 | bs->bl.min_mem_alignment = 512; |
459b4e66 | 162 | bs->bl.opt_mem_alignment = getpagesize(); |
bd44feb7 SH |
163 | |
164 | /* Safe default since most protocols use readv()/writev()/etc */ | |
165 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
166 | } |
167 | ||
760e0063 KW |
168 | if (bs->backing) { |
169 | bdrv_refresh_limits(bs->backing->bs, &local_err); | |
61007b31 SH |
170 | if (local_err) { |
171 | error_propagate(errp, local_err); | |
172 | return; | |
173 | } | |
d9e0dfa2 | 174 | bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl); |
61007b31 SH |
175 | } |
176 | ||
177 | /* Then let the driver override it */ | |
178 | if (drv->bdrv_refresh_limits) { | |
179 | drv->bdrv_refresh_limits(bs, errp); | |
180 | } | |
181 | } | |
182 | ||
183 | /** | |
184 | * The copy-on-read flag is actually a reference count so multiple users may | |
185 | * use the feature without worrying about clobbering its previous state. | |
186 | * Copy-on-read stays enabled until all users have called to disable it. | |
187 | */ | |
188 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
189 | { | |
d3faa13e | 190 | atomic_inc(&bs->copy_on_read); |
61007b31 SH |
191 | } |
192 | ||
193 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
194 | { | |
d3faa13e PB |
195 | int old = atomic_fetch_dec(&bs->copy_on_read); |
196 | assert(old >= 1); | |
61007b31 SH |
197 | } |
198 | ||
61124f03 PB |
199 | typedef struct { |
200 | Coroutine *co; | |
201 | BlockDriverState *bs; | |
202 | bool done; | |
481cad48 | 203 | bool begin; |
b0165585 | 204 | bool recursive; |
fe4f0614 | 205 | bool poll; |
0152bf40 | 206 | BdrvChild *parent; |
6cd5c9d7 | 207 | bool ignore_bds_parents; |
8e1da77e | 208 | int *drained_end_counter; |
61124f03 PB |
209 | } BdrvCoDrainData; |
210 | ||
211 | static void coroutine_fn bdrv_drain_invoke_entry(void *opaque) | |
212 | { | |
213 | BdrvCoDrainData *data = opaque; | |
214 | BlockDriverState *bs = data->bs; | |
215 | ||
481cad48 | 216 | if (data->begin) { |
f8ea8dac | 217 | bs->drv->bdrv_co_drain_begin(bs); |
481cad48 MP |
218 | } else { |
219 | bs->drv->bdrv_co_drain_end(bs); | |
220 | } | |
61124f03 | 221 | |
65181d63 | 222 | /* Set data->done and decrement drained_end_counter before bdrv_wakeup() */ |
61124f03 | 223 | atomic_mb_set(&data->done, true); |
e037c09c | 224 | if (!data->begin) { |
8e1da77e HR |
225 | atomic_dec(data->drained_end_counter); |
226 | } | |
65181d63 | 227 | bdrv_dec_in_flight(bs); |
8e1da77e | 228 | |
e037c09c | 229 | g_free(data); |
61124f03 PB |
230 | } |
231 | ||
db0289b9 | 232 | /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */ |
8e1da77e HR |
233 | static void bdrv_drain_invoke(BlockDriverState *bs, bool begin, |
234 | int *drained_end_counter) | |
61124f03 | 235 | { |
0109e7e6 | 236 | BdrvCoDrainData *data; |
61124f03 | 237 | |
f8ea8dac | 238 | if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) || |
481cad48 | 239 | (!begin && !bs->drv->bdrv_co_drain_end)) { |
61124f03 PB |
240 | return; |
241 | } | |
242 | ||
0109e7e6 KW |
243 | data = g_new(BdrvCoDrainData, 1); |
244 | *data = (BdrvCoDrainData) { | |
245 | .bs = bs, | |
246 | .done = false, | |
8e1da77e HR |
247 | .begin = begin, |
248 | .drained_end_counter = drained_end_counter, | |
0109e7e6 KW |
249 | }; |
250 | ||
e037c09c | 251 | if (!begin) { |
8e1da77e HR |
252 | atomic_inc(drained_end_counter); |
253 | } | |
254 | ||
0109e7e6 KW |
255 | /* Make sure the driver callback completes during the polling phase for |
256 | * drain_begin. */ | |
257 | bdrv_inc_in_flight(bs); | |
258 | data->co = qemu_coroutine_create(bdrv_drain_invoke_entry, data); | |
259 | aio_co_schedule(bdrv_get_aio_context(bs), data->co); | |
61124f03 PB |
260 | } |
261 | ||
1cc8e54a | 262 | /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ |
fe4f0614 | 263 | bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, |
6cd5c9d7 | 264 | BdrvChild *ignore_parent, bool ignore_bds_parents) |
89bd0305 | 265 | { |
fe4f0614 KW |
266 | BdrvChild *child, *next; |
267 | ||
6cd5c9d7 | 268 | if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) { |
89bd0305 KW |
269 | return true; |
270 | } | |
271 | ||
fe4f0614 KW |
272 | if (atomic_read(&bs->in_flight)) { |
273 | return true; | |
274 | } | |
275 | ||
276 | if (recursive) { | |
6cd5c9d7 | 277 | assert(!ignore_bds_parents); |
fe4f0614 | 278 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
6cd5c9d7 | 279 | if (bdrv_drain_poll(child->bs, recursive, child, false)) { |
fe4f0614 KW |
280 | return true; |
281 | } | |
282 | } | |
283 | } | |
284 | ||
285 | return false; | |
89bd0305 KW |
286 | } |
287 | ||
fe4f0614 | 288 | static bool bdrv_drain_poll_top_level(BlockDriverState *bs, bool recursive, |
89bd0305 | 289 | BdrvChild *ignore_parent) |
1cc8e54a | 290 | { |
6cd5c9d7 | 291 | return bdrv_drain_poll(bs, recursive, ignore_parent, false); |
1cc8e54a KW |
292 | } |
293 | ||
b0165585 | 294 | static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, |
6cd5c9d7 KW |
295 | BdrvChild *parent, bool ignore_bds_parents, |
296 | bool poll); | |
b0165585 | 297 | static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, |
8e1da77e HR |
298 | BdrvChild *parent, bool ignore_bds_parents, |
299 | int *drained_end_counter); | |
0152bf40 | 300 | |
a77fd4bb FZ |
301 | static void bdrv_co_drain_bh_cb(void *opaque) |
302 | { | |
303 | BdrvCoDrainData *data = opaque; | |
304 | Coroutine *co = data->co; | |
99723548 | 305 | BlockDriverState *bs = data->bs; |
a77fd4bb | 306 | |
c8ca33d0 | 307 | if (bs) { |
aa1361d5 KW |
308 | AioContext *ctx = bdrv_get_aio_context(bs); |
309 | AioContext *co_ctx = qemu_coroutine_get_aio_context(co); | |
310 | ||
311 | /* | |
312 | * When the coroutine yielded, the lock for its home context was | |
313 | * released, so we need to re-acquire it here. If it explicitly | |
314 | * acquired a different context, the lock is still held and we don't | |
315 | * want to lock it a second time (or AIO_WAIT_WHILE() would hang). | |
316 | */ | |
317 | if (ctx == co_ctx) { | |
318 | aio_context_acquire(ctx); | |
319 | } | |
c8ca33d0 KW |
320 | bdrv_dec_in_flight(bs); |
321 | if (data->begin) { | |
e037c09c | 322 | assert(!data->drained_end_counter); |
6cd5c9d7 KW |
323 | bdrv_do_drained_begin(bs, data->recursive, data->parent, |
324 | data->ignore_bds_parents, data->poll); | |
c8ca33d0 | 325 | } else { |
e037c09c | 326 | assert(!data->poll); |
6cd5c9d7 | 327 | bdrv_do_drained_end(bs, data->recursive, data->parent, |
8e1da77e HR |
328 | data->ignore_bds_parents, |
329 | data->drained_end_counter); | |
c8ca33d0 | 330 | } |
aa1361d5 KW |
331 | if (ctx == co_ctx) { |
332 | aio_context_release(ctx); | |
333 | } | |
481cad48 | 334 | } else { |
c8ca33d0 KW |
335 | assert(data->begin); |
336 | bdrv_drain_all_begin(); | |
481cad48 MP |
337 | } |
338 | ||
a77fd4bb | 339 | data->done = true; |
1919631e | 340 | aio_co_wake(co); |
a77fd4bb FZ |
341 | } |
342 | ||
481cad48 | 343 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, |
b0165585 | 344 | bool begin, bool recursive, |
6cd5c9d7 KW |
345 | BdrvChild *parent, |
346 | bool ignore_bds_parents, | |
8e1da77e HR |
347 | bool poll, |
348 | int *drained_end_counter) | |
a77fd4bb FZ |
349 | { |
350 | BdrvCoDrainData data; | |
351 | ||
352 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
c40a2545 | 353 | * other coroutines run if they were queued by aio_co_enter(). */ |
a77fd4bb FZ |
354 | |
355 | assert(qemu_in_coroutine()); | |
356 | data = (BdrvCoDrainData) { | |
357 | .co = qemu_coroutine_self(), | |
358 | .bs = bs, | |
359 | .done = false, | |
481cad48 | 360 | .begin = begin, |
b0165585 | 361 | .recursive = recursive, |
0152bf40 | 362 | .parent = parent, |
6cd5c9d7 | 363 | .ignore_bds_parents = ignore_bds_parents, |
fe4f0614 | 364 | .poll = poll, |
8e1da77e | 365 | .drained_end_counter = drained_end_counter, |
a77fd4bb | 366 | }; |
8e1da77e | 367 | |
c8ca33d0 KW |
368 | if (bs) { |
369 | bdrv_inc_in_flight(bs); | |
370 | } | |
fffb6e12 PB |
371 | aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), |
372 | bdrv_co_drain_bh_cb, &data); | |
a77fd4bb FZ |
373 | |
374 | qemu_coroutine_yield(); | |
375 | /* If we are resumed from some other event (such as an aio completion or a | |
376 | * timer callback), it is a bug in the caller that should be fixed. */ | |
377 | assert(data.done); | |
378 | } | |
379 | ||
dcf94a23 | 380 | void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, |
6cd5c9d7 | 381 | BdrvChild *parent, bool ignore_bds_parents) |
6820643f | 382 | { |
dcf94a23 | 383 | assert(!qemu_in_coroutine()); |
d42cf288 | 384 | |
60369b86 | 385 | /* Stop things in parent-to-child order */ |
414c2ec3 | 386 | if (atomic_fetch_inc(&bs->quiesce_counter) == 0) { |
6820643f | 387 | aio_disable_external(bdrv_get_aio_context(bs)); |
6820643f KW |
388 | } |
389 | ||
6cd5c9d7 | 390 | bdrv_parent_drained_begin(bs, parent, ignore_bds_parents); |
8e1da77e | 391 | bdrv_drain_invoke(bs, true, NULL); |
dcf94a23 KW |
392 | } |
393 | ||
394 | static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, | |
6cd5c9d7 KW |
395 | BdrvChild *parent, bool ignore_bds_parents, |
396 | bool poll) | |
dcf94a23 KW |
397 | { |
398 | BdrvChild *child, *next; | |
399 | ||
400 | if (qemu_in_coroutine()) { | |
6cd5c9d7 | 401 | bdrv_co_yield_to_drain(bs, true, recursive, parent, ignore_bds_parents, |
8e1da77e | 402 | poll, NULL); |
dcf94a23 KW |
403 | return; |
404 | } | |
405 | ||
6cd5c9d7 | 406 | bdrv_do_drained_begin_quiesce(bs, parent, ignore_bds_parents); |
d30b8e64 | 407 | |
b0165585 | 408 | if (recursive) { |
6cd5c9d7 | 409 | assert(!ignore_bds_parents); |
d736f119 | 410 | bs->recursive_quiesce_counter++; |
b0165585 | 411 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
6cd5c9d7 KW |
412 | bdrv_do_drained_begin(child->bs, true, child, ignore_bds_parents, |
413 | false); | |
b0165585 KW |
414 | } |
415 | } | |
fe4f0614 KW |
416 | |
417 | /* | |
418 | * Wait for drained requests to finish. | |
419 | * | |
420 | * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The | |
421 | * call is needed so things in this AioContext can make progress even | |
422 | * though we don't return to the main AioContext loop - this automatically | |
423 | * includes other nodes in the same AioContext and therefore all child | |
424 | * nodes. | |
425 | */ | |
426 | if (poll) { | |
6cd5c9d7 | 427 | assert(!ignore_bds_parents); |
fe4f0614 KW |
428 | BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, recursive, parent)); |
429 | } | |
6820643f KW |
430 | } |
431 | ||
0152bf40 KW |
432 | void bdrv_drained_begin(BlockDriverState *bs) |
433 | { | |
6cd5c9d7 | 434 | bdrv_do_drained_begin(bs, false, NULL, false, true); |
b0165585 KW |
435 | } |
436 | ||
437 | void bdrv_subtree_drained_begin(BlockDriverState *bs) | |
438 | { | |
6cd5c9d7 | 439 | bdrv_do_drained_begin(bs, true, NULL, false, true); |
0152bf40 KW |
440 | } |
441 | ||
e037c09c HR |
442 | /** |
443 | * This function does not poll, nor must any of its recursively called | |
444 | * functions. The *drained_end_counter pointee will be incremented | |
445 | * once for every background operation scheduled, and decremented once | |
446 | * the operation settles. Therefore, the pointer must remain valid | |
447 | * until the pointee reaches 0. That implies that whoever sets up the | |
448 | * pointee has to poll until it is 0. | |
449 | * | |
450 | * We use atomic operations to access *drained_end_counter, because | |
451 | * (1) when called from bdrv_set_aio_context_ignore(), the subgraph of | |
452 | * @bs may contain nodes in different AioContexts, | |
453 | * (2) bdrv_drain_all_end() uses the same counter for all nodes, | |
454 | * regardless of which AioContext they are in. | |
455 | */ | |
6cd5c9d7 | 456 | static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, |
8e1da77e HR |
457 | BdrvChild *parent, bool ignore_bds_parents, |
458 | int *drained_end_counter) | |
6820643f | 459 | { |
61ad631c | 460 | BdrvChild *child; |
0f115168 KW |
461 | int old_quiesce_counter; |
462 | ||
e037c09c HR |
463 | assert(drained_end_counter != NULL); |
464 | ||
481cad48 | 465 | if (qemu_in_coroutine()) { |
6cd5c9d7 | 466 | bdrv_co_yield_to_drain(bs, false, recursive, parent, ignore_bds_parents, |
8e1da77e | 467 | false, drained_end_counter); |
481cad48 MP |
468 | return; |
469 | } | |
6820643f | 470 | assert(bs->quiesce_counter > 0); |
6820643f | 471 | |
60369b86 | 472 | /* Re-enable things in child-to-parent order */ |
8e1da77e | 473 | bdrv_drain_invoke(bs, false, drained_end_counter); |
e037c09c HR |
474 | bdrv_parent_drained_end(bs, parent, ignore_bds_parents, |
475 | drained_end_counter); | |
5cb2737e HR |
476 | |
477 | old_quiesce_counter = atomic_fetch_dec(&bs->quiesce_counter); | |
0f115168 | 478 | if (old_quiesce_counter == 1) { |
0f115168 KW |
479 | aio_enable_external(bdrv_get_aio_context(bs)); |
480 | } | |
b0165585 KW |
481 | |
482 | if (recursive) { | |
6cd5c9d7 | 483 | assert(!ignore_bds_parents); |
d736f119 | 484 | bs->recursive_quiesce_counter--; |
61ad631c | 485 | QLIST_FOREACH(child, &bs->children, next) { |
8e1da77e HR |
486 | bdrv_do_drained_end(child->bs, true, child, ignore_bds_parents, |
487 | drained_end_counter); | |
b0165585 KW |
488 | } |
489 | } | |
6820643f KW |
490 | } |
491 | ||
0152bf40 KW |
492 | void bdrv_drained_end(BlockDriverState *bs) |
493 | { | |
e037c09c HR |
494 | int drained_end_counter = 0; |
495 | bdrv_do_drained_end(bs, false, NULL, false, &drained_end_counter); | |
496 | BDRV_POLL_WHILE(bs, atomic_read(&drained_end_counter) > 0); | |
497 | } | |
498 | ||
499 | void bdrv_drained_end_no_poll(BlockDriverState *bs, int *drained_end_counter) | |
500 | { | |
501 | bdrv_do_drained_end(bs, false, NULL, false, drained_end_counter); | |
b0165585 KW |
502 | } |
503 | ||
504 | void bdrv_subtree_drained_end(BlockDriverState *bs) | |
505 | { | |
e037c09c HR |
506 | int drained_end_counter = 0; |
507 | bdrv_do_drained_end(bs, true, NULL, false, &drained_end_counter); | |
508 | BDRV_POLL_WHILE(bs, atomic_read(&drained_end_counter) > 0); | |
0152bf40 KW |
509 | } |
510 | ||
d736f119 KW |
511 | void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent) |
512 | { | |
513 | int i; | |
514 | ||
515 | for (i = 0; i < new_parent->recursive_quiesce_counter; i++) { | |
6cd5c9d7 | 516 | bdrv_do_drained_begin(child->bs, true, child, false, true); |
d736f119 KW |
517 | } |
518 | } | |
519 | ||
520 | void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent) | |
521 | { | |
e037c09c | 522 | int drained_end_counter = 0; |
d736f119 KW |
523 | int i; |
524 | ||
525 | for (i = 0; i < old_parent->recursive_quiesce_counter; i++) { | |
e037c09c HR |
526 | bdrv_do_drained_end(child->bs, true, child, false, |
527 | &drained_end_counter); | |
d736f119 | 528 | } |
e037c09c HR |
529 | |
530 | BDRV_POLL_WHILE(child->bs, atomic_read(&drained_end_counter) > 0); | |
d736f119 KW |
531 | } |
532 | ||
61007b31 | 533 | /* |
67da1dc5 FZ |
534 | * Wait for pending requests to complete on a single BlockDriverState subtree, |
535 | * and suspend block driver's internal I/O until next request arrives. | |
61007b31 | 536 | * |
61007b31 SH |
537 | * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState |
538 | * AioContext. | |
539 | */ | |
b6e84c97 | 540 | void coroutine_fn bdrv_co_drain(BlockDriverState *bs) |
61007b31 | 541 | { |
6820643f KW |
542 | assert(qemu_in_coroutine()); |
543 | bdrv_drained_begin(bs); | |
544 | bdrv_drained_end(bs); | |
b6e84c97 | 545 | } |
f406c03c | 546 | |
b6e84c97 PB |
547 | void bdrv_drain(BlockDriverState *bs) |
548 | { | |
6820643f KW |
549 | bdrv_drained_begin(bs); |
550 | bdrv_drained_end(bs); | |
61007b31 SH |
551 | } |
552 | ||
c13ad59f KW |
553 | static void bdrv_drain_assert_idle(BlockDriverState *bs) |
554 | { | |
555 | BdrvChild *child, *next; | |
556 | ||
557 | assert(atomic_read(&bs->in_flight) == 0); | |
558 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { | |
559 | bdrv_drain_assert_idle(child->bs); | |
560 | } | |
561 | } | |
562 | ||
0f12264e KW |
563 | unsigned int bdrv_drain_all_count = 0; |
564 | ||
565 | static bool bdrv_drain_all_poll(void) | |
566 | { | |
567 | BlockDriverState *bs = NULL; | |
568 | bool result = false; | |
569 | ||
0f12264e KW |
570 | /* bdrv_drain_poll() can't make changes to the graph and we are holding the |
571 | * main AioContext lock, so iterating bdrv_next_all_states() is safe. */ | |
572 | while ((bs = bdrv_next_all_states(bs))) { | |
573 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
574 | aio_context_acquire(aio_context); | |
575 | result |= bdrv_drain_poll(bs, false, NULL, true); | |
576 | aio_context_release(aio_context); | |
577 | } | |
578 | ||
579 | return result; | |
580 | } | |
581 | ||
61007b31 SH |
582 | /* |
583 | * Wait for pending requests to complete across all BlockDriverStates | |
584 | * | |
585 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
586 | * after calling this function. | |
c0778f66 AG |
587 | * |
588 | * This pauses all block jobs and disables external clients. It must | |
589 | * be paired with bdrv_drain_all_end(). | |
590 | * | |
591 | * NOTE: no new block jobs or BlockDriverStates can be created between | |
592 | * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls. | |
61007b31 | 593 | */ |
c0778f66 | 594 | void bdrv_drain_all_begin(void) |
61007b31 | 595 | { |
0f12264e | 596 | BlockDriverState *bs = NULL; |
61007b31 | 597 | |
c8ca33d0 | 598 | if (qemu_in_coroutine()) { |
8e1da77e | 599 | bdrv_co_yield_to_drain(NULL, true, false, NULL, true, true, NULL); |
c8ca33d0 KW |
600 | return; |
601 | } | |
602 | ||
0f12264e KW |
603 | /* AIO_WAIT_WHILE() with a NULL context can only be called from the main |
604 | * loop AioContext, so make sure we're in the main context. */ | |
9a7e86c8 | 605 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
606 | assert(bdrv_drain_all_count < INT_MAX); |
607 | bdrv_drain_all_count++; | |
9a7e86c8 | 608 | |
0f12264e KW |
609 | /* Quiesce all nodes, without polling in-flight requests yet. The graph |
610 | * cannot change during this loop. */ | |
611 | while ((bs = bdrv_next_all_states(bs))) { | |
61007b31 SH |
612 | AioContext *aio_context = bdrv_get_aio_context(bs); |
613 | ||
614 | aio_context_acquire(aio_context); | |
0f12264e | 615 | bdrv_do_drained_begin(bs, false, NULL, true, false); |
61007b31 SH |
616 | aio_context_release(aio_context); |
617 | } | |
618 | ||
0f12264e | 619 | /* Now poll the in-flight requests */ |
cfe29d82 | 620 | AIO_WAIT_WHILE(NULL, bdrv_drain_all_poll()); |
0f12264e KW |
621 | |
622 | while ((bs = bdrv_next_all_states(bs))) { | |
c13ad59f | 623 | bdrv_drain_assert_idle(bs); |
61007b31 | 624 | } |
c0778f66 AG |
625 | } |
626 | ||
627 | void bdrv_drain_all_end(void) | |
628 | { | |
0f12264e | 629 | BlockDriverState *bs = NULL; |
e037c09c | 630 | int drained_end_counter = 0; |
c0778f66 | 631 | |
0f12264e | 632 | while ((bs = bdrv_next_all_states(bs))) { |
61007b31 SH |
633 | AioContext *aio_context = bdrv_get_aio_context(bs); |
634 | ||
635 | aio_context_acquire(aio_context); | |
e037c09c | 636 | bdrv_do_drained_end(bs, false, NULL, true, &drained_end_counter); |
61007b31 SH |
637 | aio_context_release(aio_context); |
638 | } | |
0f12264e | 639 | |
e037c09c HR |
640 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
641 | AIO_WAIT_WHILE(NULL, atomic_read(&drained_end_counter) > 0); | |
642 | ||
0f12264e KW |
643 | assert(bdrv_drain_all_count > 0); |
644 | bdrv_drain_all_count--; | |
61007b31 SH |
645 | } |
646 | ||
c0778f66 AG |
647 | void bdrv_drain_all(void) |
648 | { | |
649 | bdrv_drain_all_begin(); | |
650 | bdrv_drain_all_end(); | |
651 | } | |
652 | ||
61007b31 SH |
653 | /** |
654 | * Remove an active request from the tracked requests list | |
655 | * | |
656 | * This function should be called when a tracked request is completing. | |
657 | */ | |
658 | static void tracked_request_end(BdrvTrackedRequest *req) | |
659 | { | |
660 | if (req->serialising) { | |
20fc71b2 | 661 | atomic_dec(&req->bs->serialising_in_flight); |
61007b31 SH |
662 | } |
663 | ||
3783fa3d | 664 | qemu_co_mutex_lock(&req->bs->reqs_lock); |
61007b31 SH |
665 | QLIST_REMOVE(req, list); |
666 | qemu_co_queue_restart_all(&req->wait_queue); | |
3783fa3d | 667 | qemu_co_mutex_unlock(&req->bs->reqs_lock); |
61007b31 SH |
668 | } |
669 | ||
670 | /** | |
671 | * Add an active request to the tracked requests list | |
672 | */ | |
673 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
674 | BlockDriverState *bs, | |
675 | int64_t offset, | |
22931a15 | 676 | uint64_t bytes, |
ebde595c | 677 | enum BdrvTrackedRequestType type) |
61007b31 | 678 | { |
22931a15 FZ |
679 | assert(bytes <= INT64_MAX && offset <= INT64_MAX - bytes); |
680 | ||
61007b31 SH |
681 | *req = (BdrvTrackedRequest){ |
682 | .bs = bs, | |
683 | .offset = offset, | |
684 | .bytes = bytes, | |
ebde595c | 685 | .type = type, |
61007b31 SH |
686 | .co = qemu_coroutine_self(), |
687 | .serialising = false, | |
688 | .overlap_offset = offset, | |
689 | .overlap_bytes = bytes, | |
690 | }; | |
691 | ||
692 | qemu_co_queue_init(&req->wait_queue); | |
693 | ||
3783fa3d | 694 | qemu_co_mutex_lock(&bs->reqs_lock); |
61007b31 | 695 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
3783fa3d | 696 | qemu_co_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
697 | } |
698 | ||
699 | static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align) | |
700 | { | |
701 | int64_t overlap_offset = req->offset & ~(align - 1); | |
22931a15 | 702 | uint64_t overlap_bytes = ROUND_UP(req->offset + req->bytes, align) |
61007b31 SH |
703 | - overlap_offset; |
704 | ||
705 | if (!req->serialising) { | |
20fc71b2 | 706 | atomic_inc(&req->bs->serialising_in_flight); |
61007b31 SH |
707 | req->serialising = true; |
708 | } | |
709 | ||
710 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
711 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
712 | } | |
713 | ||
09d2f948 VSO |
714 | static bool is_request_serialising_and_aligned(BdrvTrackedRequest *req) |
715 | { | |
716 | /* | |
717 | * If the request is serialising, overlap_offset and overlap_bytes are set, | |
718 | * so we can check if the request is aligned. Otherwise, don't care and | |
719 | * return false. | |
720 | */ | |
721 | ||
722 | return req->serialising && (req->offset == req->overlap_offset) && | |
723 | (req->bytes == req->overlap_bytes); | |
724 | } | |
725 | ||
244483e6 KW |
726 | /** |
727 | * Round a region to cluster boundaries | |
728 | */ | |
729 | void bdrv_round_to_clusters(BlockDriverState *bs, | |
7cfd5275 | 730 | int64_t offset, int64_t bytes, |
244483e6 | 731 | int64_t *cluster_offset, |
7cfd5275 | 732 | int64_t *cluster_bytes) |
244483e6 KW |
733 | { |
734 | BlockDriverInfo bdi; | |
735 | ||
736 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
737 | *cluster_offset = offset; | |
738 | *cluster_bytes = bytes; | |
739 | } else { | |
740 | int64_t c = bdi.cluster_size; | |
741 | *cluster_offset = QEMU_ALIGN_DOWN(offset, c); | |
742 | *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c); | |
743 | } | |
744 | } | |
745 | ||
61007b31 SH |
746 | static int bdrv_get_cluster_size(BlockDriverState *bs) |
747 | { | |
748 | BlockDriverInfo bdi; | |
749 | int ret; | |
750 | ||
751 | ret = bdrv_get_info(bs, &bdi); | |
752 | if (ret < 0 || bdi.cluster_size == 0) { | |
a5b8dd2c | 753 | return bs->bl.request_alignment; |
61007b31 SH |
754 | } else { |
755 | return bdi.cluster_size; | |
756 | } | |
757 | } | |
758 | ||
759 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
22931a15 | 760 | int64_t offset, uint64_t bytes) |
61007b31 SH |
761 | { |
762 | /* aaaa bbbb */ | |
763 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
764 | return false; | |
765 | } | |
766 | /* bbbb aaaa */ | |
767 | if (req->overlap_offset >= offset + bytes) { | |
768 | return false; | |
769 | } | |
770 | return true; | |
771 | } | |
772 | ||
99723548 PB |
773 | void bdrv_inc_in_flight(BlockDriverState *bs) |
774 | { | |
775 | atomic_inc(&bs->in_flight); | |
776 | } | |
777 | ||
c9d1a561 PB |
778 | void bdrv_wakeup(BlockDriverState *bs) |
779 | { | |
cfe29d82 | 780 | aio_wait_kick(); |
c9d1a561 PB |
781 | } |
782 | ||
99723548 PB |
783 | void bdrv_dec_in_flight(BlockDriverState *bs) |
784 | { | |
785 | atomic_dec(&bs->in_flight); | |
c9d1a561 | 786 | bdrv_wakeup(bs); |
99723548 PB |
787 | } |
788 | ||
61007b31 SH |
789 | static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self) |
790 | { | |
791 | BlockDriverState *bs = self->bs; | |
792 | BdrvTrackedRequest *req; | |
793 | bool retry; | |
794 | bool waited = false; | |
795 | ||
20fc71b2 | 796 | if (!atomic_read(&bs->serialising_in_flight)) { |
61007b31 SH |
797 | return false; |
798 | } | |
799 | ||
800 | do { | |
801 | retry = false; | |
3783fa3d | 802 | qemu_co_mutex_lock(&bs->reqs_lock); |
61007b31 SH |
803 | QLIST_FOREACH(req, &bs->tracked_requests, list) { |
804 | if (req == self || (!req->serialising && !self->serialising)) { | |
805 | continue; | |
806 | } | |
807 | if (tracked_request_overlaps(req, self->overlap_offset, | |
808 | self->overlap_bytes)) | |
809 | { | |
810 | /* Hitting this means there was a reentrant request, for | |
811 | * example, a block driver issuing nested requests. This must | |
812 | * never happen since it means deadlock. | |
813 | */ | |
814 | assert(qemu_coroutine_self() != req->co); | |
815 | ||
816 | /* If the request is already (indirectly) waiting for us, or | |
817 | * will wait for us as soon as it wakes up, then just go on | |
818 | * (instead of producing a deadlock in the former case). */ | |
819 | if (!req->waiting_for) { | |
820 | self->waiting_for = req; | |
3783fa3d | 821 | qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock); |
61007b31 SH |
822 | self->waiting_for = NULL; |
823 | retry = true; | |
824 | waited = true; | |
825 | break; | |
826 | } | |
827 | } | |
828 | } | |
3783fa3d | 829 | qemu_co_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
830 | } while (retry); |
831 | ||
832 | return waited; | |
833 | } | |
834 | ||
835 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
836 | size_t size) | |
837 | { | |
41ae31e3 | 838 | if (size > BDRV_REQUEST_MAX_BYTES) { |
61007b31 SH |
839 | return -EIO; |
840 | } | |
841 | ||
842 | if (!bdrv_is_inserted(bs)) { | |
843 | return -ENOMEDIUM; | |
844 | } | |
845 | ||
846 | if (offset < 0) { | |
847 | return -EIO; | |
848 | } | |
849 | ||
850 | return 0; | |
851 | } | |
852 | ||
61007b31 | 853 | typedef struct RwCo { |
e293b7a3 | 854 | BdrvChild *child; |
61007b31 SH |
855 | int64_t offset; |
856 | QEMUIOVector *qiov; | |
857 | bool is_write; | |
858 | int ret; | |
859 | BdrvRequestFlags flags; | |
860 | } RwCo; | |
861 | ||
862 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
863 | { | |
864 | RwCo *rwco = opaque; | |
865 | ||
866 | if (!rwco->is_write) { | |
a03ef88f | 867 | rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset, |
cab3a356 KW |
868 | rwco->qiov->size, rwco->qiov, |
869 | rwco->flags); | |
61007b31 | 870 | } else { |
a03ef88f | 871 | rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset, |
cab3a356 KW |
872 | rwco->qiov->size, rwco->qiov, |
873 | rwco->flags); | |
61007b31 | 874 | } |
4720cbee | 875 | aio_wait_kick(); |
61007b31 SH |
876 | } |
877 | ||
878 | /* | |
879 | * Process a vectored synchronous request using coroutines | |
880 | */ | |
e293b7a3 | 881 | static int bdrv_prwv_co(BdrvChild *child, int64_t offset, |
61007b31 SH |
882 | QEMUIOVector *qiov, bool is_write, |
883 | BdrvRequestFlags flags) | |
884 | { | |
885 | Coroutine *co; | |
886 | RwCo rwco = { | |
e293b7a3 | 887 | .child = child, |
61007b31 SH |
888 | .offset = offset, |
889 | .qiov = qiov, | |
890 | .is_write = is_write, | |
891 | .ret = NOT_DONE, | |
892 | .flags = flags, | |
893 | }; | |
894 | ||
61007b31 SH |
895 | if (qemu_in_coroutine()) { |
896 | /* Fast-path if already in coroutine context */ | |
897 | bdrv_rw_co_entry(&rwco); | |
898 | } else { | |
0b8b8753 | 899 | co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco); |
e92f0e19 | 900 | bdrv_coroutine_enter(child->bs, co); |
88b062c2 | 901 | BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE); |
61007b31 SH |
902 | } |
903 | return rwco.ret; | |
904 | } | |
905 | ||
720ff280 | 906 | int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset, |
f5a5ca79 | 907 | int bytes, BdrvRequestFlags flags) |
61007b31 | 908 | { |
0d93ed08 | 909 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, NULL, bytes); |
74021bc4 | 910 | |
e293b7a3 | 911 | return bdrv_prwv_co(child, offset, &qiov, true, |
74021bc4 | 912 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
913 | } |
914 | ||
915 | /* | |
74021bc4 | 916 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
917 | * The operation is sped up by checking the block status and only writing |
918 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 919 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 920 | * BDRV_REQ_FUA). |
61007b31 SH |
921 | * |
922 | * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). | |
923 | */ | |
720ff280 | 924 | int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags) |
61007b31 | 925 | { |
237d78f8 EB |
926 | int ret; |
927 | int64_t target_size, bytes, offset = 0; | |
720ff280 | 928 | BlockDriverState *bs = child->bs; |
61007b31 | 929 | |
7286d610 EB |
930 | target_size = bdrv_getlength(bs); |
931 | if (target_size < 0) { | |
932 | return target_size; | |
61007b31 SH |
933 | } |
934 | ||
935 | for (;;) { | |
7286d610 EB |
936 | bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES); |
937 | if (bytes <= 0) { | |
61007b31 SH |
938 | return 0; |
939 | } | |
237d78f8 | 940 | ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL); |
61007b31 | 941 | if (ret < 0) { |
61007b31 SH |
942 | return ret; |
943 | } | |
944 | if (ret & BDRV_BLOCK_ZERO) { | |
237d78f8 | 945 | offset += bytes; |
61007b31 SH |
946 | continue; |
947 | } | |
237d78f8 | 948 | ret = bdrv_pwrite_zeroes(child, offset, bytes, flags); |
61007b31 | 949 | if (ret < 0) { |
61007b31 SH |
950 | return ret; |
951 | } | |
237d78f8 | 952 | offset += bytes; |
61007b31 SH |
953 | } |
954 | } | |
955 | ||
cf2ab8fc | 956 | int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
f1e84741 KW |
957 | { |
958 | int ret; | |
959 | ||
e293b7a3 | 960 | ret = bdrv_prwv_co(child, offset, qiov, false, 0); |
f1e84741 KW |
961 | if (ret < 0) { |
962 | return ret; | |
963 | } | |
964 | ||
965 | return qiov->size; | |
966 | } | |
967 | ||
2e11d756 | 968 | /* See bdrv_pwrite() for the return codes */ |
cf2ab8fc | 969 | int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes) |
61007b31 | 970 | { |
0d93ed08 | 971 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes); |
61007b31 SH |
972 | |
973 | if (bytes < 0) { | |
974 | return -EINVAL; | |
975 | } | |
976 | ||
cf2ab8fc | 977 | return bdrv_preadv(child, offset, &qiov); |
61007b31 SH |
978 | } |
979 | ||
d9ca2ea2 | 980 | int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
61007b31 SH |
981 | { |
982 | int ret; | |
983 | ||
e293b7a3 | 984 | ret = bdrv_prwv_co(child, offset, qiov, true, 0); |
61007b31 SH |
985 | if (ret < 0) { |
986 | return ret; | |
987 | } | |
988 | ||
989 | return qiov->size; | |
990 | } | |
991 | ||
2e11d756 AG |
992 | /* Return no. of bytes on success or < 0 on error. Important errors are: |
993 | -EIO generic I/O error (may happen for all errors) | |
994 | -ENOMEDIUM No media inserted. | |
995 | -EINVAL Invalid offset or number of bytes | |
996 | -EACCES Trying to write a read-only device | |
997 | */ | |
d9ca2ea2 | 998 | int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes) |
61007b31 | 999 | { |
0d93ed08 | 1000 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes); |
61007b31 SH |
1001 | |
1002 | if (bytes < 0) { | |
1003 | return -EINVAL; | |
1004 | } | |
1005 | ||
d9ca2ea2 | 1006 | return bdrv_pwritev(child, offset, &qiov); |
61007b31 SH |
1007 | } |
1008 | ||
1009 | /* | |
1010 | * Writes to the file and ensures that no writes are reordered across this | |
1011 | * request (acts as a barrier) | |
1012 | * | |
1013 | * Returns 0 on success, -errno in error cases. | |
1014 | */ | |
d9ca2ea2 KW |
1015 | int bdrv_pwrite_sync(BdrvChild *child, int64_t offset, |
1016 | const void *buf, int count) | |
61007b31 SH |
1017 | { |
1018 | int ret; | |
1019 | ||
d9ca2ea2 | 1020 | ret = bdrv_pwrite(child, offset, buf, count); |
61007b31 SH |
1021 | if (ret < 0) { |
1022 | return ret; | |
1023 | } | |
1024 | ||
d9ca2ea2 | 1025 | ret = bdrv_flush(child->bs); |
855a6a93 KW |
1026 | if (ret < 0) { |
1027 | return ret; | |
61007b31 SH |
1028 | } |
1029 | ||
1030 | return 0; | |
1031 | } | |
1032 | ||
08844473 KW |
1033 | typedef struct CoroutineIOCompletion { |
1034 | Coroutine *coroutine; | |
1035 | int ret; | |
1036 | } CoroutineIOCompletion; | |
1037 | ||
1038 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
1039 | { | |
1040 | CoroutineIOCompletion *co = opaque; | |
1041 | ||
1042 | co->ret = ret; | |
b9e413dd | 1043 | aio_co_wake(co->coroutine); |
08844473 KW |
1044 | } |
1045 | ||
166fe960 KW |
1046 | static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs, |
1047 | uint64_t offset, uint64_t bytes, | |
ac850bf0 VSO |
1048 | QEMUIOVector *qiov, |
1049 | size_t qiov_offset, int flags) | |
166fe960 KW |
1050 | { |
1051 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
1052 | int64_t sector_num; |
1053 | unsigned int nb_sectors; | |
ac850bf0 VSO |
1054 | QEMUIOVector local_qiov; |
1055 | int ret; | |
3fb06697 | 1056 | |
fa166538 | 1057 | assert(!(flags & ~BDRV_REQ_MASK)); |
fe0480d6 | 1058 | assert(!(flags & BDRV_REQ_NO_FALLBACK)); |
fa166538 | 1059 | |
d470ad42 HR |
1060 | if (!drv) { |
1061 | return -ENOMEDIUM; | |
1062 | } | |
1063 | ||
ac850bf0 VSO |
1064 | if (drv->bdrv_co_preadv_part) { |
1065 | return drv->bdrv_co_preadv_part(bs, offset, bytes, qiov, qiov_offset, | |
1066 | flags); | |
1067 | } | |
1068 | ||
1069 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1070 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1071 | qiov = &local_qiov; | |
1072 | } | |
1073 | ||
3fb06697 | 1074 | if (drv->bdrv_co_preadv) { |
ac850bf0 VSO |
1075 | ret = drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); |
1076 | goto out; | |
3fb06697 KW |
1077 | } |
1078 | ||
edfab6a0 | 1079 | if (drv->bdrv_aio_preadv) { |
08844473 KW |
1080 | BlockAIOCB *acb; |
1081 | CoroutineIOCompletion co = { | |
1082 | .coroutine = qemu_coroutine_self(), | |
1083 | }; | |
1084 | ||
edfab6a0 EB |
1085 | acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags, |
1086 | bdrv_co_io_em_complete, &co); | |
08844473 | 1087 | if (acb == NULL) { |
ac850bf0 VSO |
1088 | ret = -EIO; |
1089 | goto out; | |
08844473 KW |
1090 | } else { |
1091 | qemu_coroutine_yield(); | |
ac850bf0 VSO |
1092 | ret = co.ret; |
1093 | goto out; | |
08844473 KW |
1094 | } |
1095 | } | |
edfab6a0 EB |
1096 | |
1097 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1098 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1099 | ||
1100 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1101 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
41ae31e3 | 1102 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 EB |
1103 | assert(drv->bdrv_co_readv); |
1104 | ||
ac850bf0 VSO |
1105 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); |
1106 | ||
1107 | out: | |
1108 | if (qiov == &local_qiov) { | |
1109 | qemu_iovec_destroy(&local_qiov); | |
1110 | } | |
1111 | ||
1112 | return ret; | |
166fe960 KW |
1113 | } |
1114 | ||
78a07294 KW |
1115 | static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs, |
1116 | uint64_t offset, uint64_t bytes, | |
ac850bf0 VSO |
1117 | QEMUIOVector *qiov, |
1118 | size_t qiov_offset, int flags) | |
78a07294 KW |
1119 | { |
1120 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
1121 | int64_t sector_num; |
1122 | unsigned int nb_sectors; | |
ac850bf0 | 1123 | QEMUIOVector local_qiov; |
78a07294 KW |
1124 | int ret; |
1125 | ||
fa166538 | 1126 | assert(!(flags & ~BDRV_REQ_MASK)); |
fe0480d6 | 1127 | assert(!(flags & BDRV_REQ_NO_FALLBACK)); |
fa166538 | 1128 | |
d470ad42 HR |
1129 | if (!drv) { |
1130 | return -ENOMEDIUM; | |
1131 | } | |
1132 | ||
ac850bf0 VSO |
1133 | if (drv->bdrv_co_pwritev_part) { |
1134 | ret = drv->bdrv_co_pwritev_part(bs, offset, bytes, qiov, qiov_offset, | |
1135 | flags & bs->supported_write_flags); | |
1136 | flags &= ~bs->supported_write_flags; | |
1137 | goto emulate_flags; | |
1138 | } | |
1139 | ||
1140 | if (qiov_offset > 0 || bytes != qiov->size) { | |
1141 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1142 | qiov = &local_qiov; | |
1143 | } | |
1144 | ||
3fb06697 | 1145 | if (drv->bdrv_co_pwritev) { |
515c2f43 KW |
1146 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, |
1147 | flags & bs->supported_write_flags); | |
1148 | flags &= ~bs->supported_write_flags; | |
3fb06697 KW |
1149 | goto emulate_flags; |
1150 | } | |
1151 | ||
edfab6a0 | 1152 | if (drv->bdrv_aio_pwritev) { |
08844473 KW |
1153 | BlockAIOCB *acb; |
1154 | CoroutineIOCompletion co = { | |
1155 | .coroutine = qemu_coroutine_self(), | |
1156 | }; | |
1157 | ||
edfab6a0 EB |
1158 | acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov, |
1159 | flags & bs->supported_write_flags, | |
1160 | bdrv_co_io_em_complete, &co); | |
1161 | flags &= ~bs->supported_write_flags; | |
08844473 | 1162 | if (acb == NULL) { |
3fb06697 | 1163 | ret = -EIO; |
08844473 KW |
1164 | } else { |
1165 | qemu_coroutine_yield(); | |
3fb06697 | 1166 | ret = co.ret; |
08844473 | 1167 | } |
edfab6a0 EB |
1168 | goto emulate_flags; |
1169 | } | |
1170 | ||
1171 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1172 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1173 | ||
1174 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1175 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
41ae31e3 | 1176 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 | 1177 | |
e18a58b4 EB |
1178 | assert(drv->bdrv_co_writev); |
1179 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov, | |
1180 | flags & bs->supported_write_flags); | |
1181 | flags &= ~bs->supported_write_flags; | |
78a07294 | 1182 | |
3fb06697 | 1183 | emulate_flags: |
4df863f3 | 1184 | if (ret == 0 && (flags & BDRV_REQ_FUA)) { |
78a07294 KW |
1185 | ret = bdrv_co_flush(bs); |
1186 | } | |
1187 | ||
ac850bf0 VSO |
1188 | if (qiov == &local_qiov) { |
1189 | qemu_iovec_destroy(&local_qiov); | |
1190 | } | |
1191 | ||
78a07294 KW |
1192 | return ret; |
1193 | } | |
1194 | ||
29a298af PB |
1195 | static int coroutine_fn |
1196 | bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset, | |
ac850bf0 VSO |
1197 | uint64_t bytes, QEMUIOVector *qiov, |
1198 | size_t qiov_offset) | |
29a298af PB |
1199 | { |
1200 | BlockDriver *drv = bs->drv; | |
ac850bf0 VSO |
1201 | QEMUIOVector local_qiov; |
1202 | int ret; | |
29a298af | 1203 | |
d470ad42 HR |
1204 | if (!drv) { |
1205 | return -ENOMEDIUM; | |
1206 | } | |
1207 | ||
ac850bf0 | 1208 | if (!block_driver_can_compress(drv)) { |
29a298af PB |
1209 | return -ENOTSUP; |
1210 | } | |
1211 | ||
ac850bf0 VSO |
1212 | if (drv->bdrv_co_pwritev_compressed_part) { |
1213 | return drv->bdrv_co_pwritev_compressed_part(bs, offset, bytes, | |
1214 | qiov, qiov_offset); | |
1215 | } | |
1216 | ||
1217 | if (qiov_offset == 0) { | |
1218 | return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov); | |
1219 | } | |
1220 | ||
1221 | qemu_iovec_init_slice(&local_qiov, qiov, qiov_offset, bytes); | |
1222 | ret = drv->bdrv_co_pwritev_compressed(bs, offset, bytes, &local_qiov); | |
1223 | qemu_iovec_destroy(&local_qiov); | |
1224 | ||
1225 | return ret; | |
29a298af PB |
1226 | } |
1227 | ||
85c97ca7 | 1228 | static int coroutine_fn bdrv_co_do_copy_on_readv(BdrvChild *child, |
3299e5ec | 1229 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1143ec5e | 1230 | size_t qiov_offset, int flags) |
61007b31 | 1231 | { |
85c97ca7 KW |
1232 | BlockDriverState *bs = child->bs; |
1233 | ||
61007b31 SH |
1234 | /* Perform I/O through a temporary buffer so that users who scribble over |
1235 | * their read buffer while the operation is in progress do not end up | |
1236 | * modifying the image file. This is critical for zero-copy guest I/O | |
1237 | * where anything might happen inside guest memory. | |
1238 | */ | |
2275cc90 | 1239 | void *bounce_buffer = NULL; |
61007b31 SH |
1240 | |
1241 | BlockDriver *drv = bs->drv; | |
244483e6 | 1242 | int64_t cluster_offset; |
7cfd5275 | 1243 | int64_t cluster_bytes; |
61007b31 SH |
1244 | size_t skip_bytes; |
1245 | int ret; | |
cb2e2878 EB |
1246 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
1247 | BDRV_REQUEST_MAX_BYTES); | |
1248 | unsigned int progress = 0; | |
61007b31 | 1249 | |
d470ad42 HR |
1250 | if (!drv) { |
1251 | return -ENOMEDIUM; | |
1252 | } | |
1253 | ||
1bf03e66 KW |
1254 | /* FIXME We cannot require callers to have write permissions when all they |
1255 | * are doing is a read request. If we did things right, write permissions | |
1256 | * would be obtained anyway, but internally by the copy-on-read code. As | |
765d9df9 | 1257 | * long as it is implemented here rather than in a separate filter driver, |
1bf03e66 KW |
1258 | * the copy-on-read code doesn't have its own BdrvChild, however, for which |
1259 | * it could request permissions. Therefore we have to bypass the permission | |
1260 | * system for the moment. */ | |
1261 | // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
afa4b293 | 1262 | |
61007b31 | 1263 | /* Cover entire cluster so no additional backing file I/O is required when |
cb2e2878 EB |
1264 | * allocating cluster in the image file. Note that this value may exceed |
1265 | * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which | |
1266 | * is one reason we loop rather than doing it all at once. | |
61007b31 | 1267 | */ |
244483e6 | 1268 | bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes); |
cb2e2878 | 1269 | skip_bytes = offset - cluster_offset; |
61007b31 | 1270 | |
244483e6 KW |
1271 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
1272 | cluster_offset, cluster_bytes); | |
61007b31 | 1273 | |
cb2e2878 EB |
1274 | while (cluster_bytes) { |
1275 | int64_t pnum; | |
61007b31 | 1276 | |
cb2e2878 EB |
1277 | ret = bdrv_is_allocated(bs, cluster_offset, |
1278 | MIN(cluster_bytes, max_transfer), &pnum); | |
1279 | if (ret < 0) { | |
1280 | /* Safe to treat errors in querying allocation as if | |
1281 | * unallocated; we'll probably fail again soon on the | |
1282 | * read, but at least that will set a decent errno. | |
1283 | */ | |
1284 | pnum = MIN(cluster_bytes, max_transfer); | |
1285 | } | |
61007b31 | 1286 | |
b0ddcbbb KW |
1287 | /* Stop at EOF if the image ends in the middle of the cluster */ |
1288 | if (ret == 0 && pnum == 0) { | |
1289 | assert(progress >= bytes); | |
1290 | break; | |
1291 | } | |
1292 | ||
cb2e2878 | 1293 | assert(skip_bytes < pnum); |
61007b31 | 1294 | |
cb2e2878 | 1295 | if (ret <= 0) { |
1143ec5e VSO |
1296 | QEMUIOVector local_qiov; |
1297 | ||
cb2e2878 | 1298 | /* Must copy-on-read; use the bounce buffer */ |
0d93ed08 | 1299 | pnum = MIN(pnum, MAX_BOUNCE_BUFFER); |
2275cc90 VSO |
1300 | if (!bounce_buffer) { |
1301 | int64_t max_we_need = MAX(pnum, cluster_bytes - pnum); | |
1302 | int64_t max_allowed = MIN(max_transfer, MAX_BOUNCE_BUFFER); | |
1303 | int64_t bounce_buffer_len = MIN(max_we_need, max_allowed); | |
1304 | ||
1305 | bounce_buffer = qemu_try_blockalign(bs, bounce_buffer_len); | |
1306 | if (!bounce_buffer) { | |
1307 | ret = -ENOMEM; | |
1308 | goto err; | |
1309 | } | |
1310 | } | |
0d93ed08 | 1311 | qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum); |
61007b31 | 1312 | |
cb2e2878 | 1313 | ret = bdrv_driver_preadv(bs, cluster_offset, pnum, |
ac850bf0 | 1314 | &local_qiov, 0, 0); |
cb2e2878 EB |
1315 | if (ret < 0) { |
1316 | goto err; | |
1317 | } | |
1318 | ||
1319 | bdrv_debug_event(bs, BLKDBG_COR_WRITE); | |
1320 | if (drv->bdrv_co_pwrite_zeroes && | |
1321 | buffer_is_zero(bounce_buffer, pnum)) { | |
1322 | /* FIXME: Should we (perhaps conditionally) be setting | |
1323 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
1324 | * that still correctly reads as zero? */ | |
7adcf59f HR |
1325 | ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, pnum, |
1326 | BDRV_REQ_WRITE_UNCHANGED); | |
cb2e2878 EB |
1327 | } else { |
1328 | /* This does not change the data on the disk, it is not | |
1329 | * necessary to flush even in cache=writethrough mode. | |
1330 | */ | |
1331 | ret = bdrv_driver_pwritev(bs, cluster_offset, pnum, | |
ac850bf0 | 1332 | &local_qiov, 0, |
7adcf59f | 1333 | BDRV_REQ_WRITE_UNCHANGED); |
cb2e2878 EB |
1334 | } |
1335 | ||
1336 | if (ret < 0) { | |
1337 | /* It might be okay to ignore write errors for guest | |
1338 | * requests. If this is a deliberate copy-on-read | |
1339 | * then we don't want to ignore the error. Simply | |
1340 | * report it in all cases. | |
1341 | */ | |
1342 | goto err; | |
1343 | } | |
1344 | ||
3299e5ec | 1345 | if (!(flags & BDRV_REQ_PREFETCH)) { |
1143ec5e VSO |
1346 | qemu_iovec_from_buf(qiov, qiov_offset + progress, |
1347 | bounce_buffer + skip_bytes, | |
3299e5ec VSO |
1348 | pnum - skip_bytes); |
1349 | } | |
1350 | } else if (!(flags & BDRV_REQ_PREFETCH)) { | |
cb2e2878 | 1351 | /* Read directly into the destination */ |
1143ec5e VSO |
1352 | ret = bdrv_driver_preadv(bs, offset + progress, |
1353 | MIN(pnum - skip_bytes, bytes - progress), | |
1354 | qiov, qiov_offset + progress, 0); | |
cb2e2878 EB |
1355 | if (ret < 0) { |
1356 | goto err; | |
1357 | } | |
1358 | } | |
1359 | ||
1360 | cluster_offset += pnum; | |
1361 | cluster_bytes -= pnum; | |
1362 | progress += pnum - skip_bytes; | |
1363 | skip_bytes = 0; | |
1364 | } | |
1365 | ret = 0; | |
61007b31 SH |
1366 | |
1367 | err: | |
1368 | qemu_vfree(bounce_buffer); | |
1369 | return ret; | |
1370 | } | |
1371 | ||
1372 | /* | |
1373 | * Forwards an already correctly aligned request to the BlockDriver. This | |
1a62d0ac EB |
1374 | * handles copy on read, zeroing after EOF, and fragmentation of large |
1375 | * reads; any other features must be implemented by the caller. | |
61007b31 | 1376 | */ |
85c97ca7 | 1377 | static int coroutine_fn bdrv_aligned_preadv(BdrvChild *child, |
61007b31 | 1378 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, |
65cd4424 | 1379 | int64_t align, QEMUIOVector *qiov, size_t qiov_offset, int flags) |
61007b31 | 1380 | { |
85c97ca7 | 1381 | BlockDriverState *bs = child->bs; |
c9d20029 | 1382 | int64_t total_bytes, max_bytes; |
1a62d0ac EB |
1383 | int ret = 0; |
1384 | uint64_t bytes_remaining = bytes; | |
1385 | int max_transfer; | |
61007b31 | 1386 | |
49c07526 KW |
1387 | assert(is_power_of_2(align)); |
1388 | assert((offset & (align - 1)) == 0); | |
1389 | assert((bytes & (align - 1)) == 0); | |
abb06c5a | 1390 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
1a62d0ac EB |
1391 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1392 | align); | |
a604fa2b EB |
1393 | |
1394 | /* TODO: We would need a per-BDS .supported_read_flags and | |
1395 | * potential fallback support, if we ever implement any read flags | |
1396 | * to pass through to drivers. For now, there aren't any | |
1397 | * passthrough flags. */ | |
3299e5ec VSO |
1398 | assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ | |
1399 | BDRV_REQ_PREFETCH))); | |
61007b31 SH |
1400 | |
1401 | /* Handle Copy on Read and associated serialisation */ | |
1402 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
1403 | /* If we touch the same cluster it counts as an overlap. This | |
1404 | * guarantees that allocating writes will be serialized and not race | |
1405 | * with each other for the same cluster. For example, in copy-on-read | |
1406 | * it ensures that the CoR read and write operations are atomic and | |
1407 | * guest writes cannot interleave between them. */ | |
1408 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
1409 | } | |
1410 | ||
09d2f948 VSO |
1411 | /* BDRV_REQ_SERIALISING is only for write operation */ |
1412 | assert(!(flags & BDRV_REQ_SERIALISING)); | |
1413 | ||
61408b25 FZ |
1414 | if (!(flags & BDRV_REQ_NO_SERIALISING)) { |
1415 | wait_serialising_requests(req); | |
1416 | } | |
61007b31 SH |
1417 | |
1418 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
d6a644bb | 1419 | int64_t pnum; |
61007b31 | 1420 | |
88e63df2 | 1421 | ret = bdrv_is_allocated(bs, offset, bytes, &pnum); |
61007b31 SH |
1422 | if (ret < 0) { |
1423 | goto out; | |
1424 | } | |
1425 | ||
88e63df2 | 1426 | if (!ret || pnum != bytes) { |
65cd4424 VSO |
1427 | ret = bdrv_co_do_copy_on_readv(child, offset, bytes, |
1428 | qiov, qiov_offset, flags); | |
3299e5ec VSO |
1429 | goto out; |
1430 | } else if (flags & BDRV_REQ_PREFETCH) { | |
61007b31 SH |
1431 | goto out; |
1432 | } | |
1433 | } | |
1434 | ||
1a62d0ac | 1435 | /* Forward the request to the BlockDriver, possibly fragmenting it */ |
c9d20029 KW |
1436 | total_bytes = bdrv_getlength(bs); |
1437 | if (total_bytes < 0) { | |
1438 | ret = total_bytes; | |
1439 | goto out; | |
1440 | } | |
61007b31 | 1441 | |
c9d20029 | 1442 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
1a62d0ac | 1443 | if (bytes <= max_bytes && bytes <= max_transfer) { |
65cd4424 | 1444 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, qiov_offset, 0); |
1a62d0ac EB |
1445 | goto out; |
1446 | } | |
61007b31 | 1447 | |
1a62d0ac EB |
1448 | while (bytes_remaining) { |
1449 | int num; | |
61007b31 | 1450 | |
1a62d0ac | 1451 | if (max_bytes) { |
1a62d0ac EB |
1452 | num = MIN(bytes_remaining, MIN(max_bytes, max_transfer)); |
1453 | assert(num); | |
61007b31 | 1454 | |
1a62d0ac | 1455 | ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining, |
65cd4424 | 1456 | num, qiov, bytes - bytes_remaining, 0); |
1a62d0ac | 1457 | max_bytes -= num; |
1a62d0ac EB |
1458 | } else { |
1459 | num = bytes_remaining; | |
1460 | ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0, | |
1461 | bytes_remaining); | |
1462 | } | |
1463 | if (ret < 0) { | |
1464 | goto out; | |
1465 | } | |
1466 | bytes_remaining -= num; | |
61007b31 SH |
1467 | } |
1468 | ||
1469 | out: | |
1a62d0ac | 1470 | return ret < 0 ? ret : 0; |
61007b31 SH |
1471 | } |
1472 | ||
61007b31 | 1473 | /* |
7a3f542f VSO |
1474 | * Request padding |
1475 | * | |
1476 | * |<---- align ----->| |<----- align ---->| | |
1477 | * |<- head ->|<------------- bytes ------------->|<-- tail -->| | |
1478 | * | | | | | | | |
1479 | * -*----------$-------*-------- ... --------*-----$------------*--- | |
1480 | * | | | | | | | |
1481 | * | offset | | end | | |
1482 | * ALIGN_DOWN(offset) ALIGN_UP(offset) ALIGN_DOWN(end) ALIGN_UP(end) | |
1483 | * [buf ... ) [tail_buf ) | |
1484 | * | |
1485 | * @buf is an aligned allocation needed to store @head and @tail paddings. @head | |
1486 | * is placed at the beginning of @buf and @tail at the @end. | |
1487 | * | |
1488 | * @tail_buf is a pointer to sub-buffer, corresponding to align-sized chunk | |
1489 | * around tail, if tail exists. | |
1490 | * | |
1491 | * @merge_reads is true for small requests, | |
1492 | * if @buf_len == @head + bytes + @tail. In this case it is possible that both | |
1493 | * head and tail exist but @buf_len == align and @tail_buf == @buf. | |
1494 | */ | |
1495 | typedef struct BdrvRequestPadding { | |
1496 | uint8_t *buf; | |
1497 | size_t buf_len; | |
1498 | uint8_t *tail_buf; | |
1499 | size_t head; | |
1500 | size_t tail; | |
1501 | bool merge_reads; | |
1502 | QEMUIOVector local_qiov; | |
1503 | } BdrvRequestPadding; | |
1504 | ||
1505 | static bool bdrv_init_padding(BlockDriverState *bs, | |
1506 | int64_t offset, int64_t bytes, | |
1507 | BdrvRequestPadding *pad) | |
1508 | { | |
1509 | uint64_t align = bs->bl.request_alignment; | |
1510 | size_t sum; | |
1511 | ||
1512 | memset(pad, 0, sizeof(*pad)); | |
1513 | ||
1514 | pad->head = offset & (align - 1); | |
1515 | pad->tail = ((offset + bytes) & (align - 1)); | |
1516 | if (pad->tail) { | |
1517 | pad->tail = align - pad->tail; | |
1518 | } | |
1519 | ||
1520 | if ((!pad->head && !pad->tail) || !bytes) { | |
1521 | return false; | |
1522 | } | |
1523 | ||
1524 | sum = pad->head + bytes + pad->tail; | |
1525 | pad->buf_len = (sum > align && pad->head && pad->tail) ? 2 * align : align; | |
1526 | pad->buf = qemu_blockalign(bs, pad->buf_len); | |
1527 | pad->merge_reads = sum == pad->buf_len; | |
1528 | if (pad->tail) { | |
1529 | pad->tail_buf = pad->buf + pad->buf_len - align; | |
1530 | } | |
1531 | ||
1532 | return true; | |
1533 | } | |
1534 | ||
1535 | static int bdrv_padding_rmw_read(BdrvChild *child, | |
1536 | BdrvTrackedRequest *req, | |
1537 | BdrvRequestPadding *pad, | |
1538 | bool zero_middle) | |
1539 | { | |
1540 | QEMUIOVector local_qiov; | |
1541 | BlockDriverState *bs = child->bs; | |
1542 | uint64_t align = bs->bl.request_alignment; | |
1543 | int ret; | |
1544 | ||
1545 | assert(req->serialising && pad->buf); | |
1546 | ||
1547 | if (pad->head || pad->merge_reads) { | |
1548 | uint64_t bytes = pad->merge_reads ? pad->buf_len : align; | |
1549 | ||
1550 | qemu_iovec_init_buf(&local_qiov, pad->buf, bytes); | |
1551 | ||
1552 | if (pad->head) { | |
1553 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); | |
1554 | } | |
1555 | if (pad->merge_reads && pad->tail) { | |
1556 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); | |
1557 | } | |
1558 | ret = bdrv_aligned_preadv(child, req, req->overlap_offset, bytes, | |
65cd4424 | 1559 | align, &local_qiov, 0, 0); |
7a3f542f VSO |
1560 | if (ret < 0) { |
1561 | return ret; | |
1562 | } | |
1563 | if (pad->head) { | |
1564 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); | |
1565 | } | |
1566 | if (pad->merge_reads && pad->tail) { | |
1567 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); | |
1568 | } | |
1569 | ||
1570 | if (pad->merge_reads) { | |
1571 | goto zero_mem; | |
1572 | } | |
1573 | } | |
1574 | ||
1575 | if (pad->tail) { | |
1576 | qemu_iovec_init_buf(&local_qiov, pad->tail_buf, align); | |
1577 | ||
1578 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); | |
1579 | ret = bdrv_aligned_preadv( | |
1580 | child, req, | |
1581 | req->overlap_offset + req->overlap_bytes - align, | |
65cd4424 | 1582 | align, align, &local_qiov, 0, 0); |
7a3f542f VSO |
1583 | if (ret < 0) { |
1584 | return ret; | |
1585 | } | |
1586 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); | |
1587 | } | |
1588 | ||
1589 | zero_mem: | |
1590 | if (zero_middle) { | |
1591 | memset(pad->buf + pad->head, 0, pad->buf_len - pad->head - pad->tail); | |
1592 | } | |
1593 | ||
1594 | return 0; | |
1595 | } | |
1596 | ||
1597 | static void bdrv_padding_destroy(BdrvRequestPadding *pad) | |
1598 | { | |
1599 | if (pad->buf) { | |
1600 | qemu_vfree(pad->buf); | |
1601 | qemu_iovec_destroy(&pad->local_qiov); | |
1602 | } | |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * bdrv_pad_request | |
1607 | * | |
1608 | * Exchange request parameters with padded request if needed. Don't include RMW | |
1609 | * read of padding, bdrv_padding_rmw_read() should be called separately if | |
1610 | * needed. | |
1611 | * | |
1612 | * All parameters except @bs are in-out: they represent original request at | |
1613 | * function call and padded (if padding needed) at function finish. | |
1614 | * | |
1615 | * Function always succeeds. | |
61007b31 | 1616 | */ |
7a3f542f VSO |
1617 | static bool bdrv_pad_request(BlockDriverState *bs, QEMUIOVector **qiov, |
1618 | int64_t *offset, unsigned int *bytes, | |
1619 | BdrvRequestPadding *pad) | |
1620 | { | |
1621 | if (!bdrv_init_padding(bs, *offset, *bytes, pad)) { | |
1622 | return false; | |
1623 | } | |
1624 | ||
1625 | qemu_iovec_init_extended(&pad->local_qiov, pad->buf, pad->head, | |
1626 | *qiov, 0, *bytes, | |
1627 | pad->buf + pad->buf_len - pad->tail, pad->tail); | |
1628 | *bytes += pad->head + pad->tail; | |
1629 | *offset -= pad->head; | |
1630 | *qiov = &pad->local_qiov; | |
1631 | ||
1632 | return true; | |
1633 | } | |
1634 | ||
a03ef88f | 1635 | int coroutine_fn bdrv_co_preadv(BdrvChild *child, |
61007b31 SH |
1636 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1637 | BdrvRequestFlags flags) | |
1638 | { | |
a03ef88f | 1639 | BlockDriverState *bs = child->bs; |
61007b31 | 1640 | BdrvTrackedRequest req; |
7a3f542f | 1641 | BdrvRequestPadding pad; |
61007b31 SH |
1642 | int ret; |
1643 | ||
7a3f542f | 1644 | trace_bdrv_co_preadv(bs, offset, bytes, flags); |
61007b31 SH |
1645 | |
1646 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1647 | if (ret < 0) { | |
1648 | return ret; | |
1649 | } | |
1650 | ||
99723548 PB |
1651 | bdrv_inc_in_flight(bs); |
1652 | ||
9568b511 | 1653 | /* Don't do copy-on-read if we read data before write operation */ |
d3faa13e | 1654 | if (atomic_read(&bs->copy_on_read) && !(flags & BDRV_REQ_NO_SERIALISING)) { |
61007b31 SH |
1655 | flags |= BDRV_REQ_COPY_ON_READ; |
1656 | } | |
1657 | ||
7a3f542f | 1658 | bdrv_pad_request(bs, &qiov, &offset, &bytes, &pad); |
61007b31 | 1659 | |
ebde595c | 1660 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
7a3f542f VSO |
1661 | ret = bdrv_aligned_preadv(child, &req, offset, bytes, |
1662 | bs->bl.request_alignment, | |
65cd4424 | 1663 | qiov, 0, flags); |
61007b31 | 1664 | tracked_request_end(&req); |
99723548 | 1665 | bdrv_dec_in_flight(bs); |
61007b31 | 1666 | |
7a3f542f | 1667 | bdrv_padding_destroy(&pad); |
61007b31 SH |
1668 | |
1669 | return ret; | |
1670 | } | |
1671 | ||
d05aa8bb | 1672 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 1673 | int64_t offset, int bytes, BdrvRequestFlags flags) |
61007b31 SH |
1674 | { |
1675 | BlockDriver *drv = bs->drv; | |
1676 | QEMUIOVector qiov; | |
0d93ed08 | 1677 | void *buf = NULL; |
61007b31 | 1678 | int ret = 0; |
465fe887 | 1679 | bool need_flush = false; |
443668ca DL |
1680 | int head = 0; |
1681 | int tail = 0; | |
61007b31 | 1682 | |
cf081fca | 1683 | int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX); |
a5b8dd2c EB |
1684 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1685 | bs->bl.request_alignment); | |
cb2e2878 | 1686 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER); |
d05aa8bb | 1687 | |
d470ad42 HR |
1688 | if (!drv) { |
1689 | return -ENOMEDIUM; | |
1690 | } | |
1691 | ||
fe0480d6 KW |
1692 | if ((flags & ~bs->supported_zero_flags) & BDRV_REQ_NO_FALLBACK) { |
1693 | return -ENOTSUP; | |
1694 | } | |
1695 | ||
b8d0a980 EB |
1696 | assert(alignment % bs->bl.request_alignment == 0); |
1697 | head = offset % alignment; | |
f5a5ca79 | 1698 | tail = (offset + bytes) % alignment; |
b8d0a980 EB |
1699 | max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment); |
1700 | assert(max_write_zeroes >= bs->bl.request_alignment); | |
61007b31 | 1701 | |
f5a5ca79 MP |
1702 | while (bytes > 0 && !ret) { |
1703 | int num = bytes; | |
61007b31 SH |
1704 | |
1705 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1706 | * to be aligned, and that unaligned requests do not cross cluster |
1707 | * boundaries. | |
61007b31 | 1708 | */ |
443668ca | 1709 | if (head) { |
b2f95fee EB |
1710 | /* Make a small request up to the first aligned sector. For |
1711 | * convenience, limit this request to max_transfer even if | |
1712 | * we don't need to fall back to writes. */ | |
f5a5ca79 | 1713 | num = MIN(MIN(bytes, max_transfer), alignment - head); |
b2f95fee EB |
1714 | head = (head + num) % alignment; |
1715 | assert(num < max_write_zeroes); | |
d05aa8bb | 1716 | } else if (tail && num > alignment) { |
443668ca DL |
1717 | /* Shorten the request to the last aligned sector. */ |
1718 | num -= tail; | |
61007b31 SH |
1719 | } |
1720 | ||
1721 | /* limit request size */ | |
1722 | if (num > max_write_zeroes) { | |
1723 | num = max_write_zeroes; | |
1724 | } | |
1725 | ||
1726 | ret = -ENOTSUP; | |
1727 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1728 | if (drv->bdrv_co_pwrite_zeroes) { |
1729 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1730 | flags & bs->supported_zero_flags); | |
1731 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1732 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1733 | need_flush = true; | |
1734 | } | |
465fe887 EB |
1735 | } else { |
1736 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1737 | } |
1738 | ||
118f9944 | 1739 | if (ret < 0 && !(flags & BDRV_REQ_NO_FALLBACK)) { |
61007b31 | 1740 | /* Fall back to bounce buffer if write zeroes is unsupported */ |
465fe887 EB |
1741 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1742 | ||
1743 | if ((flags & BDRV_REQ_FUA) && | |
1744 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1745 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1746 | * flush on each chunk; use just one at the end */ | |
1747 | write_flags &= ~BDRV_REQ_FUA; | |
1748 | need_flush = true; | |
1749 | } | |
5def6b80 | 1750 | num = MIN(num, max_transfer); |
0d93ed08 VSO |
1751 | if (buf == NULL) { |
1752 | buf = qemu_try_blockalign0(bs, num); | |
1753 | if (buf == NULL) { | |
61007b31 SH |
1754 | ret = -ENOMEM; |
1755 | goto fail; | |
1756 | } | |
61007b31 | 1757 | } |
0d93ed08 | 1758 | qemu_iovec_init_buf(&qiov, buf, num); |
61007b31 | 1759 | |
ac850bf0 | 1760 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, 0, write_flags); |
61007b31 SH |
1761 | |
1762 | /* Keep bounce buffer around if it is big enough for all | |
1763 | * all future requests. | |
1764 | */ | |
5def6b80 | 1765 | if (num < max_transfer) { |
0d93ed08 VSO |
1766 | qemu_vfree(buf); |
1767 | buf = NULL; | |
61007b31 SH |
1768 | } |
1769 | } | |
1770 | ||
d05aa8bb | 1771 | offset += num; |
f5a5ca79 | 1772 | bytes -= num; |
61007b31 SH |
1773 | } |
1774 | ||
1775 | fail: | |
465fe887 EB |
1776 | if (ret == 0 && need_flush) { |
1777 | ret = bdrv_co_flush(bs); | |
1778 | } | |
0d93ed08 | 1779 | qemu_vfree(buf); |
61007b31 SH |
1780 | return ret; |
1781 | } | |
1782 | ||
85fe2479 FZ |
1783 | static inline int coroutine_fn |
1784 | bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes, | |
1785 | BdrvTrackedRequest *req, int flags) | |
1786 | { | |
1787 | BlockDriverState *bs = child->bs; | |
1788 | bool waited; | |
1789 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1790 | ||
1791 | if (bs->read_only) { | |
1792 | return -EPERM; | |
1793 | } | |
1794 | ||
1795 | /* BDRV_REQ_NO_SERIALISING is only for read operation */ | |
1796 | assert(!(flags & BDRV_REQ_NO_SERIALISING)); | |
1797 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
1798 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); | |
1799 | assert(!(flags & ~BDRV_REQ_MASK)); | |
1800 | ||
1801 | if (flags & BDRV_REQ_SERIALISING) { | |
1802 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
1803 | } | |
1804 | ||
1805 | waited = wait_serialising_requests(req); | |
1806 | ||
1807 | assert(!waited || !req->serialising || | |
1808 | is_request_serialising_and_aligned(req)); | |
1809 | assert(req->overlap_offset <= offset); | |
1810 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
cd47d792 | 1811 | assert(end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE); |
85fe2479 | 1812 | |
cd47d792 FZ |
1813 | switch (req->type) { |
1814 | case BDRV_TRACKED_WRITE: | |
1815 | case BDRV_TRACKED_DISCARD: | |
1816 | if (flags & BDRV_REQ_WRITE_UNCHANGED) { | |
1817 | assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
1818 | } else { | |
1819 | assert(child->perm & BLK_PERM_WRITE); | |
1820 | } | |
1821 | return notifier_with_return_list_notify(&bs->before_write_notifiers, | |
1822 | req); | |
1823 | case BDRV_TRACKED_TRUNCATE: | |
1824 | assert(child->perm & BLK_PERM_RESIZE); | |
1825 | return 0; | |
1826 | default: | |
1827 | abort(); | |
85fe2479 | 1828 | } |
85fe2479 FZ |
1829 | } |
1830 | ||
1831 | static inline void coroutine_fn | |
1832 | bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, uint64_t bytes, | |
1833 | BdrvTrackedRequest *req, int ret) | |
1834 | { | |
1835 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1836 | BlockDriverState *bs = child->bs; | |
1837 | ||
1838 | atomic_inc(&bs->write_gen); | |
85fe2479 | 1839 | |
00695c27 FZ |
1840 | /* |
1841 | * Discard cannot extend the image, but in error handling cases, such as | |
1842 | * when reverting a qcow2 cluster allocation, the discarded range can pass | |
1843 | * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD | |
1844 | * here. Instead, just skip it, since semantically a discard request | |
1845 | * beyond EOF cannot expand the image anyway. | |
1846 | */ | |
7f8f03ef | 1847 | if (ret == 0 && |
cd47d792 FZ |
1848 | (req->type == BDRV_TRACKED_TRUNCATE || |
1849 | end_sector > bs->total_sectors) && | |
1850 | req->type != BDRV_TRACKED_DISCARD) { | |
7f8f03ef FZ |
1851 | bs->total_sectors = end_sector; |
1852 | bdrv_parent_cb_resize(bs); | |
1853 | bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS); | |
85fe2479 | 1854 | } |
00695c27 FZ |
1855 | if (req->bytes) { |
1856 | switch (req->type) { | |
1857 | case BDRV_TRACKED_WRITE: | |
1858 | stat64_max(&bs->wr_highest_offset, offset + bytes); | |
1859 | /* fall through, to set dirty bits */ | |
1860 | case BDRV_TRACKED_DISCARD: | |
1861 | bdrv_set_dirty(bs, offset, bytes); | |
1862 | break; | |
1863 | default: | |
1864 | break; | |
1865 | } | |
1866 | } | |
85fe2479 FZ |
1867 | } |
1868 | ||
61007b31 | 1869 | /* |
04ed95f4 EB |
1870 | * Forwards an already correctly aligned write request to the BlockDriver, |
1871 | * after possibly fragmenting it. | |
61007b31 | 1872 | */ |
85c97ca7 | 1873 | static int coroutine_fn bdrv_aligned_pwritev(BdrvChild *child, |
61007b31 | 1874 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, |
28c4da28 | 1875 | int64_t align, QEMUIOVector *qiov, size_t qiov_offset, int flags) |
61007b31 | 1876 | { |
85c97ca7 | 1877 | BlockDriverState *bs = child->bs; |
61007b31 | 1878 | BlockDriver *drv = bs->drv; |
61007b31 SH |
1879 | int ret; |
1880 | ||
04ed95f4 EB |
1881 | uint64_t bytes_remaining = bytes; |
1882 | int max_transfer; | |
61007b31 | 1883 | |
d470ad42 HR |
1884 | if (!drv) { |
1885 | return -ENOMEDIUM; | |
1886 | } | |
1887 | ||
d6883bc9 VSO |
1888 | if (bdrv_has_readonly_bitmaps(bs)) { |
1889 | return -EPERM; | |
1890 | } | |
1891 | ||
cff86b38 EB |
1892 | assert(is_power_of_2(align)); |
1893 | assert((offset & (align - 1)) == 0); | |
1894 | assert((bytes & (align - 1)) == 0); | |
28c4da28 | 1895 | assert(!qiov || qiov_offset + bytes <= qiov->size); |
04ed95f4 EB |
1896 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1897 | align); | |
61007b31 | 1898 | |
85fe2479 | 1899 | ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags); |
61007b31 SH |
1900 | |
1901 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 1902 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
28c4da28 | 1903 | qemu_iovec_is_zero(qiov, qiov_offset, bytes)) { |
61007b31 SH |
1904 | flags |= BDRV_REQ_ZERO_WRITE; |
1905 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
1906 | flags |= BDRV_REQ_MAY_UNMAP; | |
1907 | } | |
1908 | } | |
1909 | ||
1910 | if (ret < 0) { | |
1911 | /* Do nothing, write notifier decided to fail this request */ | |
1912 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
9a4f4c31 | 1913 | bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 1914 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
3ea1a091 | 1915 | } else if (flags & BDRV_REQ_WRITE_COMPRESSED) { |
28c4da28 VSO |
1916 | ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, |
1917 | qiov, qiov_offset); | |
04ed95f4 | 1918 | } else if (bytes <= max_transfer) { |
9a4f4c31 | 1919 | bdrv_debug_event(bs, BLKDBG_PWRITEV); |
28c4da28 | 1920 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, qiov_offset, flags); |
04ed95f4 EB |
1921 | } else { |
1922 | bdrv_debug_event(bs, BLKDBG_PWRITEV); | |
1923 | while (bytes_remaining) { | |
1924 | int num = MIN(bytes_remaining, max_transfer); | |
04ed95f4 EB |
1925 | int local_flags = flags; |
1926 | ||
1927 | assert(num); | |
1928 | if (num < bytes_remaining && (flags & BDRV_REQ_FUA) && | |
1929 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1930 | /* If FUA is going to be emulated by flush, we only | |
1931 | * need to flush on the last iteration */ | |
1932 | local_flags &= ~BDRV_REQ_FUA; | |
1933 | } | |
04ed95f4 EB |
1934 | |
1935 | ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining, | |
28c4da28 VSO |
1936 | num, qiov, bytes - bytes_remaining, |
1937 | local_flags); | |
04ed95f4 EB |
1938 | if (ret < 0) { |
1939 | break; | |
1940 | } | |
1941 | bytes_remaining -= num; | |
1942 | } | |
61007b31 | 1943 | } |
9a4f4c31 | 1944 | bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 1945 | |
61007b31 | 1946 | if (ret >= 0) { |
04ed95f4 | 1947 | ret = 0; |
61007b31 | 1948 | } |
85fe2479 | 1949 | bdrv_co_write_req_finish(child, offset, bytes, req, ret); |
61007b31 SH |
1950 | |
1951 | return ret; | |
1952 | } | |
1953 | ||
85c97ca7 | 1954 | static int coroutine_fn bdrv_co_do_zero_pwritev(BdrvChild *child, |
9eeb6dd1 FZ |
1955 | int64_t offset, |
1956 | unsigned int bytes, | |
1957 | BdrvRequestFlags flags, | |
1958 | BdrvTrackedRequest *req) | |
1959 | { | |
85c97ca7 | 1960 | BlockDriverState *bs = child->bs; |
9eeb6dd1 | 1961 | QEMUIOVector local_qiov; |
a5b8dd2c | 1962 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 | 1963 | int ret = 0; |
7a3f542f VSO |
1964 | bool padding; |
1965 | BdrvRequestPadding pad; | |
9eeb6dd1 | 1966 | |
7a3f542f VSO |
1967 | padding = bdrv_init_padding(bs, offset, bytes, &pad); |
1968 | if (padding) { | |
9eeb6dd1 FZ |
1969 | mark_request_serialising(req, align); |
1970 | wait_serialising_requests(req); | |
9eeb6dd1 | 1971 | |
7a3f542f VSO |
1972 | bdrv_padding_rmw_read(child, req, &pad, true); |
1973 | ||
1974 | if (pad.head || pad.merge_reads) { | |
1975 | int64_t aligned_offset = offset & ~(align - 1); | |
1976 | int64_t write_bytes = pad.merge_reads ? pad.buf_len : align; | |
1977 | ||
1978 | qemu_iovec_init_buf(&local_qiov, pad.buf, write_bytes); | |
1979 | ret = bdrv_aligned_pwritev(child, req, aligned_offset, write_bytes, | |
28c4da28 | 1980 | align, &local_qiov, 0, |
7a3f542f VSO |
1981 | flags & ~BDRV_REQ_ZERO_WRITE); |
1982 | if (ret < 0 || pad.merge_reads) { | |
1983 | /* Error or all work is done */ | |
1984 | goto out; | |
1985 | } | |
1986 | offset += write_bytes - pad.head; | |
1987 | bytes -= write_bytes - pad.head; | |
9eeb6dd1 | 1988 | } |
9eeb6dd1 FZ |
1989 | } |
1990 | ||
1991 | assert(!bytes || (offset & (align - 1)) == 0); | |
1992 | if (bytes >= align) { | |
1993 | /* Write the aligned part in the middle. */ | |
1994 | uint64_t aligned_bytes = bytes & ~(align - 1); | |
85c97ca7 | 1995 | ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align, |
28c4da28 | 1996 | NULL, 0, flags); |
9eeb6dd1 | 1997 | if (ret < 0) { |
7a3f542f | 1998 | goto out; |
9eeb6dd1 FZ |
1999 | } |
2000 | bytes -= aligned_bytes; | |
2001 | offset += aligned_bytes; | |
2002 | } | |
2003 | ||
2004 | assert(!bytes || (offset & (align - 1)) == 0); | |
2005 | if (bytes) { | |
7a3f542f | 2006 | assert(align == pad.tail + bytes); |
9eeb6dd1 | 2007 | |
7a3f542f | 2008 | qemu_iovec_init_buf(&local_qiov, pad.tail_buf, align); |
85c97ca7 | 2009 | ret = bdrv_aligned_pwritev(child, req, offset, align, align, |
28c4da28 VSO |
2010 | &local_qiov, 0, |
2011 | flags & ~BDRV_REQ_ZERO_WRITE); | |
9eeb6dd1 | 2012 | } |
9eeb6dd1 | 2013 | |
7a3f542f VSO |
2014 | out: |
2015 | bdrv_padding_destroy(&pad); | |
2016 | ||
2017 | return ret; | |
9eeb6dd1 FZ |
2018 | } |
2019 | ||
61007b31 SH |
2020 | /* |
2021 | * Handle a write request in coroutine context | |
2022 | */ | |
a03ef88f | 2023 | int coroutine_fn bdrv_co_pwritev(BdrvChild *child, |
61007b31 SH |
2024 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
2025 | BdrvRequestFlags flags) | |
2026 | { | |
a03ef88f | 2027 | BlockDriverState *bs = child->bs; |
61007b31 | 2028 | BdrvTrackedRequest req; |
a5b8dd2c | 2029 | uint64_t align = bs->bl.request_alignment; |
7a3f542f | 2030 | BdrvRequestPadding pad; |
61007b31 SH |
2031 | int ret; |
2032 | ||
f42cf447 DB |
2033 | trace_bdrv_co_pwritev(child->bs, offset, bytes, flags); |
2034 | ||
61007b31 SH |
2035 | if (!bs->drv) { |
2036 | return -ENOMEDIUM; | |
2037 | } | |
61007b31 SH |
2038 | |
2039 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
2040 | if (ret < 0) { | |
2041 | return ret; | |
2042 | } | |
2043 | ||
99723548 | 2044 | bdrv_inc_in_flight(bs); |
61007b31 SH |
2045 | /* |
2046 | * Align write if necessary by performing a read-modify-write cycle. | |
2047 | * Pad qiov with the read parts and be sure to have a tracked request not | |
2048 | * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle. | |
2049 | */ | |
ebde595c | 2050 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 2051 | |
18a59f03 | 2052 | if (flags & BDRV_REQ_ZERO_WRITE) { |
85c97ca7 | 2053 | ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req); |
9eeb6dd1 FZ |
2054 | goto out; |
2055 | } | |
2056 | ||
7a3f542f | 2057 | if (bdrv_pad_request(bs, &qiov, &offset, &bytes, &pad)) { |
61007b31 SH |
2058 | mark_request_serialising(&req, align); |
2059 | wait_serialising_requests(&req); | |
7a3f542f | 2060 | bdrv_padding_rmw_read(child, &req, &pad, false); |
61007b31 SH |
2061 | } |
2062 | ||
85c97ca7 | 2063 | ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align, |
28c4da28 | 2064 | qiov, 0, flags); |
61007b31 | 2065 | |
7a3f542f | 2066 | bdrv_padding_destroy(&pad); |
61007b31 | 2067 | |
9eeb6dd1 FZ |
2068 | out: |
2069 | tracked_request_end(&req); | |
99723548 | 2070 | bdrv_dec_in_flight(bs); |
7a3f542f | 2071 | |
61007b31 SH |
2072 | return ret; |
2073 | } | |
2074 | ||
a03ef88f | 2075 | int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, |
f5a5ca79 | 2076 | int bytes, BdrvRequestFlags flags) |
61007b31 | 2077 | { |
f5a5ca79 | 2078 | trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags); |
61007b31 | 2079 | |
a03ef88f | 2080 | if (!(child->bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
2081 | flags &= ~BDRV_REQ_MAY_UNMAP; |
2082 | } | |
61007b31 | 2083 | |
f5a5ca79 | 2084 | return bdrv_co_pwritev(child, offset, bytes, NULL, |
74021bc4 | 2085 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
2086 | } |
2087 | ||
4085f5c7 JS |
2088 | /* |
2089 | * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not. | |
2090 | */ | |
2091 | int bdrv_flush_all(void) | |
2092 | { | |
2093 | BdrvNextIterator it; | |
2094 | BlockDriverState *bs = NULL; | |
2095 | int result = 0; | |
2096 | ||
2097 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
2098 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2099 | int ret; | |
2100 | ||
2101 | aio_context_acquire(aio_context); | |
2102 | ret = bdrv_flush(bs); | |
2103 | if (ret < 0 && !result) { | |
2104 | result = ret; | |
2105 | } | |
2106 | aio_context_release(aio_context); | |
2107 | } | |
2108 | ||
2109 | return result; | |
2110 | } | |
2111 | ||
2112 | ||
4bcd936e | 2113 | typedef struct BdrvCoBlockStatusData { |
61007b31 SH |
2114 | BlockDriverState *bs; |
2115 | BlockDriverState *base; | |
c9ce8c4d | 2116 | bool want_zero; |
4bcd936e EB |
2117 | int64_t offset; |
2118 | int64_t bytes; | |
2119 | int64_t *pnum; | |
2120 | int64_t *map; | |
c9ce8c4d | 2121 | BlockDriverState **file; |
4bcd936e | 2122 | int ret; |
61007b31 | 2123 | bool done; |
4bcd936e | 2124 | } BdrvCoBlockStatusData; |
61007b31 | 2125 | |
3e4d0e72 EB |
2126 | int coroutine_fn bdrv_co_block_status_from_file(BlockDriverState *bs, |
2127 | bool want_zero, | |
2128 | int64_t offset, | |
2129 | int64_t bytes, | |
2130 | int64_t *pnum, | |
2131 | int64_t *map, | |
2132 | BlockDriverState **file) | |
f7cc69b3 MP |
2133 | { |
2134 | assert(bs->file && bs->file->bs); | |
3e4d0e72 EB |
2135 | *pnum = bytes; |
2136 | *map = offset; | |
f7cc69b3 | 2137 | *file = bs->file->bs; |
3e4d0e72 | 2138 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; |
f7cc69b3 MP |
2139 | } |
2140 | ||
3e4d0e72 EB |
2141 | int coroutine_fn bdrv_co_block_status_from_backing(BlockDriverState *bs, |
2142 | bool want_zero, | |
2143 | int64_t offset, | |
2144 | int64_t bytes, | |
2145 | int64_t *pnum, | |
2146 | int64_t *map, | |
2147 | BlockDriverState **file) | |
f7cc69b3 MP |
2148 | { |
2149 | assert(bs->backing && bs->backing->bs); | |
3e4d0e72 EB |
2150 | *pnum = bytes; |
2151 | *map = offset; | |
f7cc69b3 | 2152 | *file = bs->backing->bs; |
3e4d0e72 | 2153 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; |
f7cc69b3 MP |
2154 | } |
2155 | ||
61007b31 SH |
2156 | /* |
2157 | * Returns the allocation status of the specified sectors. | |
2158 | * Drivers not implementing the functionality are assumed to not support | |
2159 | * backing files, hence all their sectors are reported as allocated. | |
2160 | * | |
86a3d5c6 EB |
2161 | * If 'want_zero' is true, the caller is querying for mapping |
2162 | * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and | |
2163 | * _ZERO where possible; otherwise, the result favors larger 'pnum', | |
2164 | * with a focus on accurate BDRV_BLOCK_ALLOCATED. | |
c9ce8c4d | 2165 | * |
2e8bc787 | 2166 | * If 'offset' is beyond the end of the disk image the return value is |
fb0d8654 | 2167 | * BDRV_BLOCK_EOF and 'pnum' is set to 0. |
61007b31 | 2168 | * |
2e8bc787 | 2169 | * 'bytes' is the max value 'pnum' should be set to. If bytes goes |
fb0d8654 EB |
2170 | * beyond the end of the disk image it will be clamped; if 'pnum' is set to |
2171 | * the end of the image, then the returned value will include BDRV_BLOCK_EOF. | |
67a0fd2a | 2172 | * |
2e8bc787 EB |
2173 | * 'pnum' is set to the number of bytes (including and immediately |
2174 | * following the specified offset) that are easily known to be in the | |
2175 | * same allocated/unallocated state. Note that a second call starting | |
2176 | * at the original offset plus returned pnum may have the same status. | |
2177 | * The returned value is non-zero on success except at end-of-file. | |
2178 | * | |
2179 | * Returns negative errno on failure. Otherwise, if the | |
2180 | * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are | |
2181 | * set to the host mapping and BDS corresponding to the guest offset. | |
61007b31 | 2182 | */ |
2e8bc787 EB |
2183 | static int coroutine_fn bdrv_co_block_status(BlockDriverState *bs, |
2184 | bool want_zero, | |
2185 | int64_t offset, int64_t bytes, | |
2186 | int64_t *pnum, int64_t *map, | |
2187 | BlockDriverState **file) | |
2188 | { | |
2189 | int64_t total_size; | |
2190 | int64_t n; /* bytes */ | |
efa6e2ed | 2191 | int ret; |
2e8bc787 | 2192 | int64_t local_map = 0; |
298a1665 | 2193 | BlockDriverState *local_file = NULL; |
efa6e2ed EB |
2194 | int64_t aligned_offset, aligned_bytes; |
2195 | uint32_t align; | |
61007b31 | 2196 | |
298a1665 EB |
2197 | assert(pnum); |
2198 | *pnum = 0; | |
2e8bc787 EB |
2199 | total_size = bdrv_getlength(bs); |
2200 | if (total_size < 0) { | |
2201 | ret = total_size; | |
298a1665 | 2202 | goto early_out; |
61007b31 SH |
2203 | } |
2204 | ||
2e8bc787 | 2205 | if (offset >= total_size) { |
298a1665 EB |
2206 | ret = BDRV_BLOCK_EOF; |
2207 | goto early_out; | |
61007b31 | 2208 | } |
2e8bc787 | 2209 | if (!bytes) { |
298a1665 EB |
2210 | ret = 0; |
2211 | goto early_out; | |
9cdcfd9f | 2212 | } |
61007b31 | 2213 | |
2e8bc787 EB |
2214 | n = total_size - offset; |
2215 | if (n < bytes) { | |
2216 | bytes = n; | |
61007b31 SH |
2217 | } |
2218 | ||
d470ad42 HR |
2219 | /* Must be non-NULL or bdrv_getlength() would have failed */ |
2220 | assert(bs->drv); | |
636cb512 | 2221 | if (!bs->drv->bdrv_co_block_status) { |
2e8bc787 | 2222 | *pnum = bytes; |
61007b31 | 2223 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; |
2e8bc787 | 2224 | if (offset + bytes == total_size) { |
fb0d8654 EB |
2225 | ret |= BDRV_BLOCK_EOF; |
2226 | } | |
61007b31 | 2227 | if (bs->drv->protocol_name) { |
2e8bc787 EB |
2228 | ret |= BDRV_BLOCK_OFFSET_VALID; |
2229 | local_map = offset; | |
298a1665 | 2230 | local_file = bs; |
61007b31 | 2231 | } |
298a1665 | 2232 | goto early_out; |
61007b31 SH |
2233 | } |
2234 | ||
99723548 | 2235 | bdrv_inc_in_flight(bs); |
efa6e2ed EB |
2236 | |
2237 | /* Round out to request_alignment boundaries */ | |
86a3d5c6 | 2238 | align = bs->bl.request_alignment; |
efa6e2ed EB |
2239 | aligned_offset = QEMU_ALIGN_DOWN(offset, align); |
2240 | aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset; | |
2241 | ||
636cb512 EB |
2242 | ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset, |
2243 | aligned_bytes, pnum, &local_map, | |
2244 | &local_file); | |
2245 | if (ret < 0) { | |
2246 | *pnum = 0; | |
2247 | goto out; | |
efa6e2ed EB |
2248 | } |
2249 | ||
2e8bc787 | 2250 | /* |
636cb512 | 2251 | * The driver's result must be a non-zero multiple of request_alignment. |
efa6e2ed | 2252 | * Clamp pnum and adjust map to original request. |
2e8bc787 | 2253 | */ |
636cb512 EB |
2254 | assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) && |
2255 | align > offset - aligned_offset); | |
69f47505 VSO |
2256 | if (ret & BDRV_BLOCK_RECURSE) { |
2257 | assert(ret & BDRV_BLOCK_DATA); | |
2258 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
2259 | assert(!(ret & BDRV_BLOCK_ZERO)); | |
2260 | } | |
2261 | ||
efa6e2ed EB |
2262 | *pnum -= offset - aligned_offset; |
2263 | if (*pnum > bytes) { | |
2264 | *pnum = bytes; | |
61007b31 | 2265 | } |
2e8bc787 | 2266 | if (ret & BDRV_BLOCK_OFFSET_VALID) { |
efa6e2ed | 2267 | local_map += offset - aligned_offset; |
2e8bc787 | 2268 | } |
61007b31 SH |
2269 | |
2270 | if (ret & BDRV_BLOCK_RAW) { | |
298a1665 | 2271 | assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file); |
2e8bc787 EB |
2272 | ret = bdrv_co_block_status(local_file, want_zero, local_map, |
2273 | *pnum, pnum, &local_map, &local_file); | |
99723548 | 2274 | goto out; |
61007b31 SH |
2275 | } |
2276 | ||
2277 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
2278 | ret |= BDRV_BLOCK_ALLOCATED; | |
c9ce8c4d | 2279 | } else if (want_zero) { |
61007b31 SH |
2280 | if (bdrv_unallocated_blocks_are_zero(bs)) { |
2281 | ret |= BDRV_BLOCK_ZERO; | |
760e0063 KW |
2282 | } else if (bs->backing) { |
2283 | BlockDriverState *bs2 = bs->backing->bs; | |
2e8bc787 | 2284 | int64_t size2 = bdrv_getlength(bs2); |
c9ce8c4d | 2285 | |
2e8bc787 | 2286 | if (size2 >= 0 && offset >= size2) { |
61007b31 SH |
2287 | ret |= BDRV_BLOCK_ZERO; |
2288 | } | |
2289 | } | |
2290 | } | |
2291 | ||
69f47505 VSO |
2292 | if (want_zero && ret & BDRV_BLOCK_RECURSE && |
2293 | local_file && local_file != bs && | |
61007b31 SH |
2294 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
2295 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
2e8bc787 EB |
2296 | int64_t file_pnum; |
2297 | int ret2; | |
61007b31 | 2298 | |
2e8bc787 EB |
2299 | ret2 = bdrv_co_block_status(local_file, want_zero, local_map, |
2300 | *pnum, &file_pnum, NULL, NULL); | |
61007b31 SH |
2301 | if (ret2 >= 0) { |
2302 | /* Ignore errors. This is just providing extra information, it | |
2303 | * is useful but not necessary. | |
2304 | */ | |
c61e684e EB |
2305 | if (ret2 & BDRV_BLOCK_EOF && |
2306 | (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) { | |
2307 | /* | |
2308 | * It is valid for the format block driver to read | |
2309 | * beyond the end of the underlying file's current | |
2310 | * size; such areas read as zero. | |
2311 | */ | |
61007b31 SH |
2312 | ret |= BDRV_BLOCK_ZERO; |
2313 | } else { | |
2314 | /* Limit request to the range reported by the protocol driver */ | |
2315 | *pnum = file_pnum; | |
2316 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
2317 | } | |
2318 | } | |
2319 | } | |
2320 | ||
99723548 PB |
2321 | out: |
2322 | bdrv_dec_in_flight(bs); | |
2e8bc787 | 2323 | if (ret >= 0 && offset + *pnum == total_size) { |
fb0d8654 EB |
2324 | ret |= BDRV_BLOCK_EOF; |
2325 | } | |
298a1665 EB |
2326 | early_out: |
2327 | if (file) { | |
2328 | *file = local_file; | |
2329 | } | |
2e8bc787 EB |
2330 | if (map) { |
2331 | *map = local_map; | |
2332 | } | |
61007b31 SH |
2333 | return ret; |
2334 | } | |
2335 | ||
5b648c67 EB |
2336 | static int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs, |
2337 | BlockDriverState *base, | |
2338 | bool want_zero, | |
2339 | int64_t offset, | |
2340 | int64_t bytes, | |
2341 | int64_t *pnum, | |
2342 | int64_t *map, | |
2343 | BlockDriverState **file) | |
ba3f0e25 FZ |
2344 | { |
2345 | BlockDriverState *p; | |
5b648c67 | 2346 | int ret = 0; |
c61e684e | 2347 | bool first = true; |
ba3f0e25 FZ |
2348 | |
2349 | assert(bs != base); | |
760e0063 | 2350 | for (p = bs; p != base; p = backing_bs(p)) { |
5b648c67 EB |
2351 | ret = bdrv_co_block_status(p, want_zero, offset, bytes, pnum, map, |
2352 | file); | |
c61e684e EB |
2353 | if (ret < 0) { |
2354 | break; | |
2355 | } | |
2356 | if (ret & BDRV_BLOCK_ZERO && ret & BDRV_BLOCK_EOF && !first) { | |
2357 | /* | |
2358 | * Reading beyond the end of the file continues to read | |
2359 | * zeroes, but we can only widen the result to the | |
2360 | * unallocated length we learned from an earlier | |
2361 | * iteration. | |
2362 | */ | |
5b648c67 | 2363 | *pnum = bytes; |
c61e684e EB |
2364 | } |
2365 | if (ret & (BDRV_BLOCK_ZERO | BDRV_BLOCK_DATA)) { | |
ba3f0e25 FZ |
2366 | break; |
2367 | } | |
5b648c67 EB |
2368 | /* [offset, pnum] unallocated on this layer, which could be only |
2369 | * the first part of [offset, bytes]. */ | |
2370 | bytes = MIN(bytes, *pnum); | |
c61e684e | 2371 | first = false; |
ba3f0e25 FZ |
2372 | } |
2373 | return ret; | |
2374 | } | |
2375 | ||
31826642 | 2376 | /* Coroutine wrapper for bdrv_block_status_above() */ |
5b648c67 | 2377 | static void coroutine_fn bdrv_block_status_above_co_entry(void *opaque) |
61007b31 | 2378 | { |
4bcd936e | 2379 | BdrvCoBlockStatusData *data = opaque; |
61007b31 | 2380 | |
5b648c67 EB |
2381 | data->ret = bdrv_co_block_status_above(data->bs, data->base, |
2382 | data->want_zero, | |
2383 | data->offset, data->bytes, | |
2384 | data->pnum, data->map, data->file); | |
61007b31 | 2385 | data->done = true; |
4720cbee | 2386 | aio_wait_kick(); |
61007b31 SH |
2387 | } |
2388 | ||
2389 | /* | |
5b648c67 | 2390 | * Synchronous wrapper around bdrv_co_block_status_above(). |
61007b31 | 2391 | * |
5b648c67 | 2392 | * See bdrv_co_block_status_above() for details. |
61007b31 | 2393 | */ |
7ddb99b9 EB |
2394 | static int bdrv_common_block_status_above(BlockDriverState *bs, |
2395 | BlockDriverState *base, | |
2396 | bool want_zero, int64_t offset, | |
2397 | int64_t bytes, int64_t *pnum, | |
2398 | int64_t *map, | |
2399 | BlockDriverState **file) | |
61007b31 SH |
2400 | { |
2401 | Coroutine *co; | |
4bcd936e | 2402 | BdrvCoBlockStatusData data = { |
61007b31 | 2403 | .bs = bs, |
ba3f0e25 | 2404 | .base = base, |
c9ce8c4d | 2405 | .want_zero = want_zero, |
7ddb99b9 EB |
2406 | .offset = offset, |
2407 | .bytes = bytes, | |
2408 | .pnum = pnum, | |
2409 | .map = map, | |
c9ce8c4d | 2410 | .file = file, |
61007b31 SH |
2411 | .done = false, |
2412 | }; | |
2413 | ||
2414 | if (qemu_in_coroutine()) { | |
2415 | /* Fast-path if already in coroutine context */ | |
5b648c67 | 2416 | bdrv_block_status_above_co_entry(&data); |
61007b31 | 2417 | } else { |
5b648c67 | 2418 | co = qemu_coroutine_create(bdrv_block_status_above_co_entry, &data); |
e92f0e19 | 2419 | bdrv_coroutine_enter(bs, co); |
88b062c2 | 2420 | BDRV_POLL_WHILE(bs, !data.done); |
61007b31 | 2421 | } |
7ddb99b9 | 2422 | return data.ret; |
61007b31 SH |
2423 | } |
2424 | ||
31826642 EB |
2425 | int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base, |
2426 | int64_t offset, int64_t bytes, int64_t *pnum, | |
2427 | int64_t *map, BlockDriverState **file) | |
c9ce8c4d | 2428 | { |
31826642 EB |
2429 | return bdrv_common_block_status_above(bs, base, true, offset, bytes, |
2430 | pnum, map, file); | |
c9ce8c4d EB |
2431 | } |
2432 | ||
237d78f8 EB |
2433 | int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2434 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
ba3f0e25 | 2435 | { |
31826642 EB |
2436 | return bdrv_block_status_above(bs, backing_bs(bs), |
2437 | offset, bytes, pnum, map, file); | |
ba3f0e25 FZ |
2438 | } |
2439 | ||
d6a644bb EB |
2440 | int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t offset, |
2441 | int64_t bytes, int64_t *pnum) | |
61007b31 | 2442 | { |
7ddb99b9 EB |
2443 | int ret; |
2444 | int64_t dummy; | |
d6a644bb | 2445 | |
7ddb99b9 EB |
2446 | ret = bdrv_common_block_status_above(bs, backing_bs(bs), false, offset, |
2447 | bytes, pnum ? pnum : &dummy, NULL, | |
c9ce8c4d | 2448 | NULL); |
61007b31 SH |
2449 | if (ret < 0) { |
2450 | return ret; | |
2451 | } | |
2452 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2453 | } | |
2454 | ||
2455 | /* | |
2456 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
2457 | * | |
170d3bd3 AS |
2458 | * Return 1 if (a prefix of) the given range is allocated in any image |
2459 | * between BASE and TOP (BASE is only included if include_base is set). | |
2460 | * BASE can be NULL to check if the given offset is allocated in any | |
2461 | * image of the chain. Return 0 otherwise, or negative errno on | |
2462 | * failure. | |
61007b31 | 2463 | * |
51b0a488 EB |
2464 | * 'pnum' is set to the number of bytes (including and immediately |
2465 | * following the specified offset) that are known to be in the same | |
2466 | * allocated/unallocated state. Note that a subsequent call starting | |
2467 | * at 'offset + *pnum' may return the same allocation status (in other | |
2468 | * words, the result is not necessarily the maximum possible range); | |
2469 | * but 'pnum' will only be 0 when end of file is reached. | |
61007b31 SH |
2470 | * |
2471 | */ | |
2472 | int bdrv_is_allocated_above(BlockDriverState *top, | |
2473 | BlockDriverState *base, | |
170d3bd3 AS |
2474 | bool include_base, int64_t offset, |
2475 | int64_t bytes, int64_t *pnum) | |
61007b31 SH |
2476 | { |
2477 | BlockDriverState *intermediate; | |
51b0a488 EB |
2478 | int ret; |
2479 | int64_t n = bytes; | |
61007b31 | 2480 | |
170d3bd3 AS |
2481 | assert(base || !include_base); |
2482 | ||
61007b31 | 2483 | intermediate = top; |
170d3bd3 | 2484 | while (include_base || intermediate != base) { |
d6a644bb | 2485 | int64_t pnum_inter; |
c00716be | 2486 | int64_t size_inter; |
d6a644bb | 2487 | |
170d3bd3 | 2488 | assert(intermediate); |
51b0a488 | 2489 | ret = bdrv_is_allocated(intermediate, offset, bytes, &pnum_inter); |
61007b31 SH |
2490 | if (ret < 0) { |
2491 | return ret; | |
d6a644bb | 2492 | } |
d6a644bb | 2493 | if (ret) { |
51b0a488 | 2494 | *pnum = pnum_inter; |
61007b31 SH |
2495 | return 1; |
2496 | } | |
2497 | ||
51b0a488 | 2498 | size_inter = bdrv_getlength(intermediate); |
c00716be EB |
2499 | if (size_inter < 0) { |
2500 | return size_inter; | |
2501 | } | |
51b0a488 EB |
2502 | if (n > pnum_inter && |
2503 | (intermediate == top || offset + pnum_inter < size_inter)) { | |
2504 | n = pnum_inter; | |
61007b31 SH |
2505 | } |
2506 | ||
170d3bd3 AS |
2507 | if (intermediate == base) { |
2508 | break; | |
2509 | } | |
2510 | ||
760e0063 | 2511 | intermediate = backing_bs(intermediate); |
61007b31 SH |
2512 | } |
2513 | ||
2514 | *pnum = n; | |
2515 | return 0; | |
2516 | } | |
2517 | ||
1a8ae822 KW |
2518 | typedef struct BdrvVmstateCo { |
2519 | BlockDriverState *bs; | |
2520 | QEMUIOVector *qiov; | |
2521 | int64_t pos; | |
2522 | bool is_read; | |
2523 | int ret; | |
2524 | } BdrvVmstateCo; | |
2525 | ||
2526 | static int coroutine_fn | |
2527 | bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
2528 | bool is_read) | |
2529 | { | |
2530 | BlockDriver *drv = bs->drv; | |
dc88a467 SH |
2531 | int ret = -ENOTSUP; |
2532 | ||
2533 | bdrv_inc_in_flight(bs); | |
1a8ae822 KW |
2534 | |
2535 | if (!drv) { | |
dc88a467 | 2536 | ret = -ENOMEDIUM; |
1a8ae822 | 2537 | } else if (drv->bdrv_load_vmstate) { |
dc88a467 SH |
2538 | if (is_read) { |
2539 | ret = drv->bdrv_load_vmstate(bs, qiov, pos); | |
2540 | } else { | |
2541 | ret = drv->bdrv_save_vmstate(bs, qiov, pos); | |
2542 | } | |
1a8ae822 | 2543 | } else if (bs->file) { |
dc88a467 | 2544 | ret = bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read); |
1a8ae822 KW |
2545 | } |
2546 | ||
dc88a467 SH |
2547 | bdrv_dec_in_flight(bs); |
2548 | return ret; | |
1a8ae822 KW |
2549 | } |
2550 | ||
2551 | static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque) | |
2552 | { | |
2553 | BdrvVmstateCo *co = opaque; | |
2554 | co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read); | |
4720cbee | 2555 | aio_wait_kick(); |
1a8ae822 KW |
2556 | } |
2557 | ||
2558 | static inline int | |
2559 | bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
2560 | bool is_read) | |
2561 | { | |
2562 | if (qemu_in_coroutine()) { | |
2563 | return bdrv_co_rw_vmstate(bs, qiov, pos, is_read); | |
2564 | } else { | |
2565 | BdrvVmstateCo data = { | |
2566 | .bs = bs, | |
2567 | .qiov = qiov, | |
2568 | .pos = pos, | |
2569 | .is_read = is_read, | |
2570 | .ret = -EINPROGRESS, | |
2571 | }; | |
0b8b8753 | 2572 | Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data); |
1a8ae822 | 2573 | |
e92f0e19 | 2574 | bdrv_coroutine_enter(bs, co); |
ea17c9d2 | 2575 | BDRV_POLL_WHILE(bs, data.ret == -EINPROGRESS); |
1a8ae822 KW |
2576 | return data.ret; |
2577 | } | |
2578 | } | |
2579 | ||
61007b31 SH |
2580 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
2581 | int64_t pos, int size) | |
2582 | { | |
0d93ed08 | 2583 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b433d942 | 2584 | int ret; |
61007b31 | 2585 | |
b433d942 KW |
2586 | ret = bdrv_writev_vmstate(bs, &qiov, pos); |
2587 | if (ret < 0) { | |
2588 | return ret; | |
2589 | } | |
2590 | ||
2591 | return size; | |
61007b31 SH |
2592 | } |
2593 | ||
2594 | int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
2595 | { | |
1a8ae822 | 2596 | return bdrv_rw_vmstate(bs, qiov, pos, false); |
61007b31 SH |
2597 | } |
2598 | ||
2599 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
2600 | int64_t pos, int size) | |
5ddda0b8 | 2601 | { |
0d93ed08 | 2602 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b433d942 | 2603 | int ret; |
5ddda0b8 | 2604 | |
b433d942 KW |
2605 | ret = bdrv_readv_vmstate(bs, &qiov, pos); |
2606 | if (ret < 0) { | |
2607 | return ret; | |
2608 | } | |
2609 | ||
2610 | return size; | |
5ddda0b8 KW |
2611 | } |
2612 | ||
2613 | int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 2614 | { |
1a8ae822 | 2615 | return bdrv_rw_vmstate(bs, qiov, pos, true); |
61007b31 SH |
2616 | } |
2617 | ||
2618 | /**************************************************************/ | |
2619 | /* async I/Os */ | |
2620 | ||
61007b31 SH |
2621 | void bdrv_aio_cancel(BlockAIOCB *acb) |
2622 | { | |
2623 | qemu_aio_ref(acb); | |
2624 | bdrv_aio_cancel_async(acb); | |
2625 | while (acb->refcnt > 1) { | |
2626 | if (acb->aiocb_info->get_aio_context) { | |
2627 | aio_poll(acb->aiocb_info->get_aio_context(acb), true); | |
2628 | } else if (acb->bs) { | |
2f47da5f PB |
2629 | /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so |
2630 | * assert that we're not using an I/O thread. Thread-safe | |
2631 | * code should use bdrv_aio_cancel_async exclusively. | |
2632 | */ | |
2633 | assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context()); | |
61007b31 SH |
2634 | aio_poll(bdrv_get_aio_context(acb->bs), true); |
2635 | } else { | |
2636 | abort(); | |
2637 | } | |
2638 | } | |
2639 | qemu_aio_unref(acb); | |
2640 | } | |
2641 | ||
2642 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2643 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2644 | * In either case the completion callback must be called. */ | |
2645 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2646 | { | |
2647 | if (acb->aiocb_info->cancel_async) { | |
2648 | acb->aiocb_info->cancel_async(acb); | |
2649 | } | |
2650 | } | |
2651 | ||
61007b31 SH |
2652 | /**************************************************************/ |
2653 | /* Coroutine block device emulation */ | |
2654 | ||
e293b7a3 KW |
2655 | typedef struct FlushCo { |
2656 | BlockDriverState *bs; | |
2657 | int ret; | |
2658 | } FlushCo; | |
2659 | ||
2660 | ||
61007b31 SH |
2661 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) |
2662 | { | |
e293b7a3 | 2663 | FlushCo *rwco = opaque; |
61007b31 SH |
2664 | |
2665 | rwco->ret = bdrv_co_flush(rwco->bs); | |
4720cbee | 2666 | aio_wait_kick(); |
61007b31 SH |
2667 | } |
2668 | ||
2669 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
2670 | { | |
49ca6259 FZ |
2671 | int current_gen; |
2672 | int ret = 0; | |
2673 | ||
2674 | bdrv_inc_in_flight(bs); | |
61007b31 | 2675 | |
e914404e | 2676 | if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs) || |
1b6bc94d | 2677 | bdrv_is_sg(bs)) { |
49ca6259 | 2678 | goto early_exit; |
61007b31 SH |
2679 | } |
2680 | ||
3783fa3d | 2681 | qemu_co_mutex_lock(&bs->reqs_lock); |
47fec599 | 2682 | current_gen = atomic_read(&bs->write_gen); |
3ff2f67a EY |
2683 | |
2684 | /* Wait until any previous flushes are completed */ | |
99723548 | 2685 | while (bs->active_flush_req) { |
3783fa3d | 2686 | qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock); |
3ff2f67a EY |
2687 | } |
2688 | ||
3783fa3d | 2689 | /* Flushes reach this point in nondecreasing current_gen order. */ |
99723548 | 2690 | bs->active_flush_req = true; |
3783fa3d | 2691 | qemu_co_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2692 | |
c32b82af PD |
2693 | /* Write back all layers by calling one driver function */ |
2694 | if (bs->drv->bdrv_co_flush) { | |
2695 | ret = bs->drv->bdrv_co_flush(bs); | |
2696 | goto out; | |
2697 | } | |
2698 | ||
61007b31 SH |
2699 | /* Write back cached data to the OS even with cache=unsafe */ |
2700 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); | |
2701 | if (bs->drv->bdrv_co_flush_to_os) { | |
2702 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
2703 | if (ret < 0) { | |
cdb5e315 | 2704 | goto out; |
61007b31 SH |
2705 | } |
2706 | } | |
2707 | ||
2708 | /* But don't actually force it to the disk with cache=unsafe */ | |
2709 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
2710 | goto flush_parent; | |
2711 | } | |
2712 | ||
3ff2f67a EY |
2713 | /* Check if we really need to flush anything */ |
2714 | if (bs->flushed_gen == current_gen) { | |
2715 | goto flush_parent; | |
2716 | } | |
2717 | ||
61007b31 | 2718 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); |
d470ad42 HR |
2719 | if (!bs->drv) { |
2720 | /* bs->drv->bdrv_co_flush() might have ejected the BDS | |
2721 | * (even in case of apparent success) */ | |
2722 | ret = -ENOMEDIUM; | |
2723 | goto out; | |
2724 | } | |
61007b31 SH |
2725 | if (bs->drv->bdrv_co_flush_to_disk) { |
2726 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
2727 | } else if (bs->drv->bdrv_aio_flush) { | |
2728 | BlockAIOCB *acb; | |
2729 | CoroutineIOCompletion co = { | |
2730 | .coroutine = qemu_coroutine_self(), | |
2731 | }; | |
2732 | ||
2733 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
2734 | if (acb == NULL) { | |
2735 | ret = -EIO; | |
2736 | } else { | |
2737 | qemu_coroutine_yield(); | |
2738 | ret = co.ret; | |
2739 | } | |
2740 | } else { | |
2741 | /* | |
2742 | * Some block drivers always operate in either writethrough or unsafe | |
2743 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
2744 | * know how the server works (because the behaviour is hardcoded or | |
2745 | * depends on server-side configuration), so we can't ensure that | |
2746 | * everything is safe on disk. Returning an error doesn't work because | |
2747 | * that would break guests even if the server operates in writethrough | |
2748 | * mode. | |
2749 | * | |
2750 | * Let's hope the user knows what he's doing. | |
2751 | */ | |
2752 | ret = 0; | |
2753 | } | |
3ff2f67a | 2754 | |
61007b31 | 2755 | if (ret < 0) { |
cdb5e315 | 2756 | goto out; |
61007b31 SH |
2757 | } |
2758 | ||
2759 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
2760 | * in the case of cache=unsafe, so there are no useless flushes. | |
2761 | */ | |
2762 | flush_parent: | |
cdb5e315 FZ |
2763 | ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0; |
2764 | out: | |
3ff2f67a | 2765 | /* Notify any pending flushes that we have completed */ |
e6af1e08 KW |
2766 | if (ret == 0) { |
2767 | bs->flushed_gen = current_gen; | |
2768 | } | |
3783fa3d PB |
2769 | |
2770 | qemu_co_mutex_lock(&bs->reqs_lock); | |
99723548 | 2771 | bs->active_flush_req = false; |
156af3ac DL |
2772 | /* Return value is ignored - it's ok if wait queue is empty */ |
2773 | qemu_co_queue_next(&bs->flush_queue); | |
3783fa3d | 2774 | qemu_co_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2775 | |
49ca6259 | 2776 | early_exit: |
99723548 | 2777 | bdrv_dec_in_flight(bs); |
cdb5e315 | 2778 | return ret; |
61007b31 SH |
2779 | } |
2780 | ||
2781 | int bdrv_flush(BlockDriverState *bs) | |
2782 | { | |
2783 | Coroutine *co; | |
e293b7a3 | 2784 | FlushCo flush_co = { |
61007b31 SH |
2785 | .bs = bs, |
2786 | .ret = NOT_DONE, | |
2787 | }; | |
2788 | ||
2789 | if (qemu_in_coroutine()) { | |
2790 | /* Fast-path if already in coroutine context */ | |
e293b7a3 | 2791 | bdrv_flush_co_entry(&flush_co); |
61007b31 | 2792 | } else { |
0b8b8753 | 2793 | co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co); |
e92f0e19 | 2794 | bdrv_coroutine_enter(bs, co); |
88b062c2 | 2795 | BDRV_POLL_WHILE(bs, flush_co.ret == NOT_DONE); |
61007b31 SH |
2796 | } |
2797 | ||
e293b7a3 | 2798 | return flush_co.ret; |
61007b31 SH |
2799 | } |
2800 | ||
2801 | typedef struct DiscardCo { | |
0b9fd3f4 | 2802 | BdrvChild *child; |
0c51a893 | 2803 | int64_t offset; |
d93e5726 | 2804 | int64_t bytes; |
61007b31 SH |
2805 | int ret; |
2806 | } DiscardCo; | |
0c51a893 | 2807 | static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque) |
61007b31 SH |
2808 | { |
2809 | DiscardCo *rwco = opaque; | |
2810 | ||
0b9fd3f4 | 2811 | rwco->ret = bdrv_co_pdiscard(rwco->child, rwco->offset, rwco->bytes); |
4720cbee | 2812 | aio_wait_kick(); |
61007b31 SH |
2813 | } |
2814 | ||
d93e5726 VSO |
2815 | int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, |
2816 | int64_t bytes) | |
61007b31 | 2817 | { |
b1066c87 | 2818 | BdrvTrackedRequest req; |
9f1963b3 | 2819 | int max_pdiscard, ret; |
3482b9bc | 2820 | int head, tail, align; |
0b9fd3f4 | 2821 | BlockDriverState *bs = child->bs; |
61007b31 | 2822 | |
d93e5726 | 2823 | if (!bs || !bs->drv || !bdrv_is_inserted(bs)) { |
61007b31 SH |
2824 | return -ENOMEDIUM; |
2825 | } | |
2826 | ||
d6883bc9 VSO |
2827 | if (bdrv_has_readonly_bitmaps(bs)) { |
2828 | return -EPERM; | |
2829 | } | |
2830 | ||
d93e5726 VSO |
2831 | if (offset < 0 || bytes < 0 || bytes > INT64_MAX - offset) { |
2832 | return -EIO; | |
61007b31 SH |
2833 | } |
2834 | ||
61007b31 SH |
2835 | /* Do nothing if disabled. */ |
2836 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
2837 | return 0; | |
2838 | } | |
2839 | ||
02aefe43 | 2840 | if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) { |
61007b31 SH |
2841 | return 0; |
2842 | } | |
2843 | ||
3482b9bc EB |
2844 | /* Discard is advisory, but some devices track and coalesce |
2845 | * unaligned requests, so we must pass everything down rather than | |
2846 | * round here. Still, most devices will just silently ignore | |
2847 | * unaligned requests (by returning -ENOTSUP), so we must fragment | |
2848 | * the request accordingly. */ | |
02aefe43 | 2849 | align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment); |
b8d0a980 EB |
2850 | assert(align % bs->bl.request_alignment == 0); |
2851 | head = offset % align; | |
f5a5ca79 | 2852 | tail = (offset + bytes) % align; |
9f1963b3 | 2853 | |
99723548 | 2854 | bdrv_inc_in_flight(bs); |
f5a5ca79 | 2855 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD); |
50824995 | 2856 | |
00695c27 | 2857 | ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0); |
ec050f77 DL |
2858 | if (ret < 0) { |
2859 | goto out; | |
2860 | } | |
2861 | ||
9f1963b3 EB |
2862 | max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX), |
2863 | align); | |
3482b9bc | 2864 | assert(max_pdiscard >= bs->bl.request_alignment); |
61007b31 | 2865 | |
f5a5ca79 | 2866 | while (bytes > 0) { |
d93e5726 | 2867 | int64_t num = bytes; |
3482b9bc EB |
2868 | |
2869 | if (head) { | |
2870 | /* Make small requests to get to alignment boundaries. */ | |
f5a5ca79 | 2871 | num = MIN(bytes, align - head); |
3482b9bc EB |
2872 | if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) { |
2873 | num %= bs->bl.request_alignment; | |
2874 | } | |
2875 | head = (head + num) % align; | |
2876 | assert(num < max_pdiscard); | |
2877 | } else if (tail) { | |
2878 | if (num > align) { | |
2879 | /* Shorten the request to the last aligned cluster. */ | |
2880 | num -= tail; | |
2881 | } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) && | |
2882 | tail > bs->bl.request_alignment) { | |
2883 | tail %= bs->bl.request_alignment; | |
2884 | num -= tail; | |
2885 | } | |
2886 | } | |
2887 | /* limit request size */ | |
2888 | if (num > max_pdiscard) { | |
2889 | num = max_pdiscard; | |
2890 | } | |
61007b31 | 2891 | |
d470ad42 HR |
2892 | if (!bs->drv) { |
2893 | ret = -ENOMEDIUM; | |
2894 | goto out; | |
2895 | } | |
47a5486d EB |
2896 | if (bs->drv->bdrv_co_pdiscard) { |
2897 | ret = bs->drv->bdrv_co_pdiscard(bs, offset, num); | |
61007b31 SH |
2898 | } else { |
2899 | BlockAIOCB *acb; | |
2900 | CoroutineIOCompletion co = { | |
2901 | .coroutine = qemu_coroutine_self(), | |
2902 | }; | |
2903 | ||
4da444a0 EB |
2904 | acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num, |
2905 | bdrv_co_io_em_complete, &co); | |
61007b31 | 2906 | if (acb == NULL) { |
b1066c87 FZ |
2907 | ret = -EIO; |
2908 | goto out; | |
61007b31 SH |
2909 | } else { |
2910 | qemu_coroutine_yield(); | |
2911 | ret = co.ret; | |
2912 | } | |
2913 | } | |
2914 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 2915 | goto out; |
61007b31 SH |
2916 | } |
2917 | ||
9f1963b3 | 2918 | offset += num; |
f5a5ca79 | 2919 | bytes -= num; |
61007b31 | 2920 | } |
b1066c87 FZ |
2921 | ret = 0; |
2922 | out: | |
00695c27 | 2923 | bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret); |
b1066c87 | 2924 | tracked_request_end(&req); |
99723548 | 2925 | bdrv_dec_in_flight(bs); |
b1066c87 | 2926 | return ret; |
61007b31 SH |
2927 | } |
2928 | ||
d93e5726 | 2929 | int bdrv_pdiscard(BdrvChild *child, int64_t offset, int64_t bytes) |
61007b31 SH |
2930 | { |
2931 | Coroutine *co; | |
2932 | DiscardCo rwco = { | |
0b9fd3f4 | 2933 | .child = child, |
0c51a893 | 2934 | .offset = offset, |
f5a5ca79 | 2935 | .bytes = bytes, |
61007b31 SH |
2936 | .ret = NOT_DONE, |
2937 | }; | |
2938 | ||
2939 | if (qemu_in_coroutine()) { | |
2940 | /* Fast-path if already in coroutine context */ | |
0c51a893 | 2941 | bdrv_pdiscard_co_entry(&rwco); |
61007b31 | 2942 | } else { |
0c51a893 | 2943 | co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco); |
0b9fd3f4 FZ |
2944 | bdrv_coroutine_enter(child->bs, co); |
2945 | BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE); | |
61007b31 SH |
2946 | } |
2947 | ||
2948 | return rwco.ret; | |
2949 | } | |
2950 | ||
48af776a | 2951 | int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
2952 | { |
2953 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
2954 | CoroutineIOCompletion co = { |
2955 | .coroutine = qemu_coroutine_self(), | |
2956 | }; | |
2957 | BlockAIOCB *acb; | |
61007b31 | 2958 | |
99723548 | 2959 | bdrv_inc_in_flight(bs); |
16a389dc | 2960 | if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) { |
5c5ae76a FZ |
2961 | co.ret = -ENOTSUP; |
2962 | goto out; | |
2963 | } | |
2964 | ||
16a389dc KW |
2965 | if (drv->bdrv_co_ioctl) { |
2966 | co.ret = drv->bdrv_co_ioctl(bs, req, buf); | |
2967 | } else { | |
2968 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
2969 | if (!acb) { | |
2970 | co.ret = -ENOTSUP; | |
2971 | goto out; | |
2972 | } | |
2973 | qemu_coroutine_yield(); | |
5c5ae76a | 2974 | } |
5c5ae76a | 2975 | out: |
99723548 | 2976 | bdrv_dec_in_flight(bs); |
5c5ae76a FZ |
2977 | return co.ret; |
2978 | } | |
2979 | ||
61007b31 SH |
2980 | void *qemu_blockalign(BlockDriverState *bs, size_t size) |
2981 | { | |
2982 | return qemu_memalign(bdrv_opt_mem_align(bs), size); | |
2983 | } | |
2984 | ||
2985 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
2986 | { | |
2987 | return memset(qemu_blockalign(bs, size), 0, size); | |
2988 | } | |
2989 | ||
2990 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
2991 | { | |
2992 | size_t align = bdrv_opt_mem_align(bs); | |
2993 | ||
2994 | /* Ensure that NULL is never returned on success */ | |
2995 | assert(align > 0); | |
2996 | if (size == 0) { | |
2997 | size = align; | |
2998 | } | |
2999 | ||
3000 | return qemu_try_memalign(align, size); | |
3001 | } | |
3002 | ||
3003 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
3004 | { | |
3005 | void *mem = qemu_try_blockalign(bs, size); | |
3006 | ||
3007 | if (mem) { | |
3008 | memset(mem, 0, size); | |
3009 | } | |
3010 | ||
3011 | return mem; | |
3012 | } | |
3013 | ||
3014 | /* | |
3015 | * Check if all memory in this vector is sector aligned. | |
3016 | */ | |
3017 | bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
3018 | { | |
3019 | int i; | |
4196d2f0 | 3020 | size_t alignment = bdrv_min_mem_align(bs); |
61007b31 SH |
3021 | |
3022 | for (i = 0; i < qiov->niov; i++) { | |
3023 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
3024 | return false; | |
3025 | } | |
3026 | if (qiov->iov[i].iov_len % alignment) { | |
3027 | return false; | |
3028 | } | |
3029 | } | |
3030 | ||
3031 | return true; | |
3032 | } | |
3033 | ||
3034 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
3035 | NotifierWithReturn *notifier) | |
3036 | { | |
3037 | notifier_with_return_list_add(&bs->before_write_notifiers, notifier); | |
3038 | } | |
3039 | ||
3040 | void bdrv_io_plug(BlockDriverState *bs) | |
3041 | { | |
6b98bd64 PB |
3042 | BdrvChild *child; |
3043 | ||
3044 | QLIST_FOREACH(child, &bs->children, next) { | |
3045 | bdrv_io_plug(child->bs); | |
3046 | } | |
3047 | ||
850d54a2 | 3048 | if (atomic_fetch_inc(&bs->io_plugged) == 0) { |
6b98bd64 PB |
3049 | BlockDriver *drv = bs->drv; |
3050 | if (drv && drv->bdrv_io_plug) { | |
3051 | drv->bdrv_io_plug(bs); | |
3052 | } | |
61007b31 SH |
3053 | } |
3054 | } | |
3055 | ||
3056 | void bdrv_io_unplug(BlockDriverState *bs) | |
3057 | { | |
6b98bd64 PB |
3058 | BdrvChild *child; |
3059 | ||
3060 | assert(bs->io_plugged); | |
850d54a2 | 3061 | if (atomic_fetch_dec(&bs->io_plugged) == 1) { |
6b98bd64 PB |
3062 | BlockDriver *drv = bs->drv; |
3063 | if (drv && drv->bdrv_io_unplug) { | |
3064 | drv->bdrv_io_unplug(bs); | |
3065 | } | |
3066 | } | |
3067 | ||
3068 | QLIST_FOREACH(child, &bs->children, next) { | |
3069 | bdrv_io_unplug(child->bs); | |
61007b31 SH |
3070 | } |
3071 | } | |
23d0ba93 FZ |
3072 | |
3073 | void bdrv_register_buf(BlockDriverState *bs, void *host, size_t size) | |
3074 | { | |
3075 | BdrvChild *child; | |
3076 | ||
3077 | if (bs->drv && bs->drv->bdrv_register_buf) { | |
3078 | bs->drv->bdrv_register_buf(bs, host, size); | |
3079 | } | |
3080 | QLIST_FOREACH(child, &bs->children, next) { | |
3081 | bdrv_register_buf(child->bs, host, size); | |
3082 | } | |
3083 | } | |
3084 | ||
3085 | void bdrv_unregister_buf(BlockDriverState *bs, void *host) | |
3086 | { | |
3087 | BdrvChild *child; | |
3088 | ||
3089 | if (bs->drv && bs->drv->bdrv_unregister_buf) { | |
3090 | bs->drv->bdrv_unregister_buf(bs, host); | |
3091 | } | |
3092 | QLIST_FOREACH(child, &bs->children, next) { | |
3093 | bdrv_unregister_buf(child->bs, host); | |
3094 | } | |
3095 | } | |
fcc67678 | 3096 | |
67b51fb9 VSO |
3097 | static int coroutine_fn bdrv_co_copy_range_internal( |
3098 | BdrvChild *src, uint64_t src_offset, BdrvChild *dst, | |
3099 | uint64_t dst_offset, uint64_t bytes, | |
3100 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags, | |
3101 | bool recurse_src) | |
fcc67678 | 3102 | { |
999658a0 | 3103 | BdrvTrackedRequest req; |
fcc67678 FZ |
3104 | int ret; |
3105 | ||
fe0480d6 KW |
3106 | /* TODO We can support BDRV_REQ_NO_FALLBACK here */ |
3107 | assert(!(read_flags & BDRV_REQ_NO_FALLBACK)); | |
3108 | assert(!(write_flags & BDRV_REQ_NO_FALLBACK)); | |
3109 | ||
d4d3e5a0 | 3110 | if (!dst || !dst->bs) { |
fcc67678 FZ |
3111 | return -ENOMEDIUM; |
3112 | } | |
fcc67678 FZ |
3113 | ret = bdrv_check_byte_request(dst->bs, dst_offset, bytes); |
3114 | if (ret) { | |
3115 | return ret; | |
3116 | } | |
67b51fb9 VSO |
3117 | if (write_flags & BDRV_REQ_ZERO_WRITE) { |
3118 | return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags); | |
fcc67678 FZ |
3119 | } |
3120 | ||
d4d3e5a0 FZ |
3121 | if (!src || !src->bs) { |
3122 | return -ENOMEDIUM; | |
3123 | } | |
3124 | ret = bdrv_check_byte_request(src->bs, src_offset, bytes); | |
3125 | if (ret) { | |
3126 | return ret; | |
3127 | } | |
3128 | ||
fcc67678 FZ |
3129 | if (!src->bs->drv->bdrv_co_copy_range_from |
3130 | || !dst->bs->drv->bdrv_co_copy_range_to | |
3131 | || src->bs->encrypted || dst->bs->encrypted) { | |
3132 | return -ENOTSUP; | |
3133 | } | |
37aec7d7 | 3134 | |
fcc67678 | 3135 | if (recurse_src) { |
999658a0 VSO |
3136 | bdrv_inc_in_flight(src->bs); |
3137 | tracked_request_begin(&req, src->bs, src_offset, bytes, | |
3138 | BDRV_TRACKED_READ); | |
3139 | ||
09d2f948 VSO |
3140 | /* BDRV_REQ_SERIALISING is only for write operation */ |
3141 | assert(!(read_flags & BDRV_REQ_SERIALISING)); | |
67b51fb9 | 3142 | if (!(read_flags & BDRV_REQ_NO_SERIALISING)) { |
999658a0 VSO |
3143 | wait_serialising_requests(&req); |
3144 | } | |
3145 | ||
37aec7d7 FZ |
3146 | ret = src->bs->drv->bdrv_co_copy_range_from(src->bs, |
3147 | src, src_offset, | |
3148 | dst, dst_offset, | |
67b51fb9 VSO |
3149 | bytes, |
3150 | read_flags, write_flags); | |
999658a0 VSO |
3151 | |
3152 | tracked_request_end(&req); | |
3153 | bdrv_dec_in_flight(src->bs); | |
fcc67678 | 3154 | } else { |
999658a0 VSO |
3155 | bdrv_inc_in_flight(dst->bs); |
3156 | tracked_request_begin(&req, dst->bs, dst_offset, bytes, | |
3157 | BDRV_TRACKED_WRITE); | |
0eb1e891 FZ |
3158 | ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req, |
3159 | write_flags); | |
3160 | if (!ret) { | |
3161 | ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs, | |
3162 | src, src_offset, | |
3163 | dst, dst_offset, | |
3164 | bytes, | |
3165 | read_flags, write_flags); | |
3166 | } | |
3167 | bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret); | |
999658a0 VSO |
3168 | tracked_request_end(&req); |
3169 | bdrv_dec_in_flight(dst->bs); | |
fcc67678 | 3170 | } |
999658a0 | 3171 | |
37aec7d7 | 3172 | return ret; |
fcc67678 FZ |
3173 | } |
3174 | ||
3175 | /* Copy range from @src to @dst. | |
3176 | * | |
3177 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3178 | * semantics. */ | |
3179 | int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, uint64_t src_offset, | |
3180 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3181 | uint64_t bytes, |
3182 | BdrvRequestFlags read_flags, | |
3183 | BdrvRequestFlags write_flags) | |
fcc67678 | 3184 | { |
ecc983a5 FZ |
3185 | trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes, |
3186 | read_flags, write_flags); | |
fcc67678 | 3187 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3188 | bytes, read_flags, write_flags, true); |
fcc67678 FZ |
3189 | } |
3190 | ||
3191 | /* Copy range from @src to @dst. | |
3192 | * | |
3193 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3194 | * semantics. */ | |
3195 | int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, uint64_t src_offset, | |
3196 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3197 | uint64_t bytes, |
3198 | BdrvRequestFlags read_flags, | |
3199 | BdrvRequestFlags write_flags) | |
fcc67678 | 3200 | { |
ecc983a5 FZ |
3201 | trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3202 | read_flags, write_flags); | |
fcc67678 | 3203 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3204 | bytes, read_flags, write_flags, false); |
fcc67678 FZ |
3205 | } |
3206 | ||
3207 | int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset, | |
3208 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3209 | uint64_t bytes, BdrvRequestFlags read_flags, |
3210 | BdrvRequestFlags write_flags) | |
fcc67678 | 3211 | { |
37aec7d7 FZ |
3212 | return bdrv_co_copy_range_from(src, src_offset, |
3213 | dst, dst_offset, | |
67b51fb9 | 3214 | bytes, read_flags, write_flags); |
fcc67678 | 3215 | } |
3d9f2d2a KW |
3216 | |
3217 | static void bdrv_parent_cb_resize(BlockDriverState *bs) | |
3218 | { | |
3219 | BdrvChild *c; | |
3220 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
3221 | if (c->role->resize) { | |
3222 | c->role->resize(c); | |
3223 | } | |
3224 | } | |
3225 | } | |
3226 | ||
3227 | /** | |
3228 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
3229 | */ | |
3230 | int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, | |
3231 | PreallocMode prealloc, Error **errp) | |
3232 | { | |
3233 | BlockDriverState *bs = child->bs; | |
3234 | BlockDriver *drv = bs->drv; | |
1bc5f09f KW |
3235 | BdrvTrackedRequest req; |
3236 | int64_t old_size, new_bytes; | |
3d9f2d2a KW |
3237 | int ret; |
3238 | ||
3d9f2d2a KW |
3239 | |
3240 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ | |
3241 | if (!drv) { | |
3242 | error_setg(errp, "No medium inserted"); | |
3243 | return -ENOMEDIUM; | |
3244 | } | |
3245 | if (offset < 0) { | |
3246 | error_setg(errp, "Image size cannot be negative"); | |
3247 | return -EINVAL; | |
3248 | } | |
3249 | ||
1bc5f09f KW |
3250 | old_size = bdrv_getlength(bs); |
3251 | if (old_size < 0) { | |
3252 | error_setg_errno(errp, -old_size, "Failed to get old image size"); | |
3253 | return old_size; | |
3254 | } | |
3255 | ||
3256 | if (offset > old_size) { | |
3257 | new_bytes = offset - old_size; | |
3258 | } else { | |
3259 | new_bytes = 0; | |
3260 | } | |
3261 | ||
3d9f2d2a | 3262 | bdrv_inc_in_flight(bs); |
5416a11e FZ |
3263 | tracked_request_begin(&req, bs, offset - new_bytes, new_bytes, |
3264 | BDRV_TRACKED_TRUNCATE); | |
1bc5f09f KW |
3265 | |
3266 | /* If we are growing the image and potentially using preallocation for the | |
3267 | * new area, we need to make sure that no write requests are made to it | |
3268 | * concurrently or they might be overwritten by preallocation. */ | |
3269 | if (new_bytes) { | |
3270 | mark_request_serialising(&req, 1); | |
cd47d792 FZ |
3271 | } |
3272 | if (bs->read_only) { | |
3273 | error_setg(errp, "Image is read-only"); | |
3274 | ret = -EACCES; | |
3275 | goto out; | |
3276 | } | |
3277 | ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req, | |
3278 | 0); | |
3279 | if (ret < 0) { | |
3280 | error_setg_errno(errp, -ret, | |
3281 | "Failed to prepare request for truncation"); | |
3282 | goto out; | |
1bc5f09f | 3283 | } |
3d9f2d2a KW |
3284 | |
3285 | if (!drv->bdrv_co_truncate) { | |
3286 | if (bs->file && drv->is_filter) { | |
3287 | ret = bdrv_co_truncate(bs->file, offset, prealloc, errp); | |
3288 | goto out; | |
3289 | } | |
3290 | error_setg(errp, "Image format driver does not support resize"); | |
3291 | ret = -ENOTSUP; | |
3292 | goto out; | |
3293 | } | |
3d9f2d2a KW |
3294 | |
3295 | ret = drv->bdrv_co_truncate(bs, offset, prealloc, errp); | |
3296 | if (ret < 0) { | |
3297 | goto out; | |
3298 | } | |
3299 | ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); | |
3300 | if (ret < 0) { | |
3301 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
3302 | } else { | |
3303 | offset = bs->total_sectors * BDRV_SECTOR_SIZE; | |
3304 | } | |
cd47d792 FZ |
3305 | /* It's possible that truncation succeeded but refresh_total_sectors |
3306 | * failed, but the latter doesn't affect how we should finish the request. | |
3307 | * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. */ | |
3308 | bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0); | |
3d9f2d2a KW |
3309 | |
3310 | out: | |
1bc5f09f | 3311 | tracked_request_end(&req); |
3d9f2d2a | 3312 | bdrv_dec_in_flight(bs); |
1bc5f09f | 3313 | |
3d9f2d2a KW |
3314 | return ret; |
3315 | } | |
3316 | ||
3317 | typedef struct TruncateCo { | |
3318 | BdrvChild *child; | |
3319 | int64_t offset; | |
3320 | PreallocMode prealloc; | |
3321 | Error **errp; | |
3322 | int ret; | |
3323 | } TruncateCo; | |
3324 | ||
3325 | static void coroutine_fn bdrv_truncate_co_entry(void *opaque) | |
3326 | { | |
3327 | TruncateCo *tco = opaque; | |
3328 | tco->ret = bdrv_co_truncate(tco->child, tco->offset, tco->prealloc, | |
3329 | tco->errp); | |
4720cbee | 3330 | aio_wait_kick(); |
3d9f2d2a KW |
3331 | } |
3332 | ||
3333 | int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc, | |
3334 | Error **errp) | |
3335 | { | |
3336 | Coroutine *co; | |
3337 | TruncateCo tco = { | |
3338 | .child = child, | |
3339 | .offset = offset, | |
3340 | .prealloc = prealloc, | |
3341 | .errp = errp, | |
3342 | .ret = NOT_DONE, | |
3343 | }; | |
3344 | ||
3345 | if (qemu_in_coroutine()) { | |
3346 | /* Fast-path if already in coroutine context */ | |
3347 | bdrv_truncate_co_entry(&tco); | |
3348 | } else { | |
3349 | co = qemu_coroutine_create(bdrv_truncate_co_entry, &tco); | |
4720cbee | 3350 | bdrv_coroutine_enter(child->bs, co); |
3d9f2d2a KW |
3351 | BDRV_POLL_WHILE(child->bs, tco.ret == NOT_DONE); |
3352 | } | |
3353 | ||
3354 | return tco.ret; | |
3355 | } |