]>
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 EB |
148 | bs->bl.request_alignment = (drv->bdrv_co_preadv || |
149 | drv->bdrv_aio_preadv) ? 1 : 512; | |
79ba8c98 | 150 | |
61007b31 SH |
151 | /* Take some limits from the children as a default */ |
152 | if (bs->file) { | |
9a4f4c31 | 153 | bdrv_refresh_limits(bs->file->bs, &local_err); |
61007b31 SH |
154 | if (local_err) { |
155 | error_propagate(errp, local_err); | |
156 | return; | |
157 | } | |
d9e0dfa2 | 158 | bdrv_merge_limits(&bs->bl, &bs->file->bs->bl); |
61007b31 | 159 | } else { |
4196d2f0 | 160 | bs->bl.min_mem_alignment = 512; |
459b4e66 | 161 | bs->bl.opt_mem_alignment = getpagesize(); |
bd44feb7 SH |
162 | |
163 | /* Safe default since most protocols use readv()/writev()/etc */ | |
164 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
165 | } |
166 | ||
760e0063 KW |
167 | if (bs->backing) { |
168 | bdrv_refresh_limits(bs->backing->bs, &local_err); | |
61007b31 SH |
169 | if (local_err) { |
170 | error_propagate(errp, local_err); | |
171 | return; | |
172 | } | |
d9e0dfa2 | 173 | bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl); |
61007b31 SH |
174 | } |
175 | ||
176 | /* Then let the driver override it */ | |
177 | if (drv->bdrv_refresh_limits) { | |
178 | drv->bdrv_refresh_limits(bs, errp); | |
179 | } | |
180 | } | |
181 | ||
182 | /** | |
183 | * The copy-on-read flag is actually a reference count so multiple users may | |
184 | * use the feature without worrying about clobbering its previous state. | |
185 | * Copy-on-read stays enabled until all users have called to disable it. | |
186 | */ | |
187 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
188 | { | |
d3faa13e | 189 | atomic_inc(&bs->copy_on_read); |
61007b31 SH |
190 | } |
191 | ||
192 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
193 | { | |
d3faa13e PB |
194 | int old = atomic_fetch_dec(&bs->copy_on_read); |
195 | assert(old >= 1); | |
61007b31 SH |
196 | } |
197 | ||
61124f03 PB |
198 | typedef struct { |
199 | Coroutine *co; | |
200 | BlockDriverState *bs; | |
201 | bool done; | |
481cad48 | 202 | bool begin; |
b0165585 | 203 | bool recursive; |
fe4f0614 | 204 | bool poll; |
0152bf40 | 205 | BdrvChild *parent; |
6cd5c9d7 | 206 | bool ignore_bds_parents; |
8e1da77e | 207 | int *drained_end_counter; |
61124f03 PB |
208 | } BdrvCoDrainData; |
209 | ||
210 | static void coroutine_fn bdrv_drain_invoke_entry(void *opaque) | |
211 | { | |
212 | BdrvCoDrainData *data = opaque; | |
213 | BlockDriverState *bs = data->bs; | |
214 | ||
481cad48 | 215 | if (data->begin) { |
f8ea8dac | 216 | bs->drv->bdrv_co_drain_begin(bs); |
481cad48 MP |
217 | } else { |
218 | bs->drv->bdrv_co_drain_end(bs); | |
219 | } | |
61124f03 | 220 | |
65181d63 | 221 | /* Set data->done and decrement drained_end_counter before bdrv_wakeup() */ |
61124f03 | 222 | atomic_mb_set(&data->done, true); |
e037c09c | 223 | if (!data->begin) { |
8e1da77e HR |
224 | atomic_dec(data->drained_end_counter); |
225 | } | |
65181d63 | 226 | bdrv_dec_in_flight(bs); |
8e1da77e | 227 | |
e037c09c | 228 | g_free(data); |
61124f03 PB |
229 | } |
230 | ||
db0289b9 | 231 | /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */ |
8e1da77e HR |
232 | static void bdrv_drain_invoke(BlockDriverState *bs, bool begin, |
233 | int *drained_end_counter) | |
61124f03 | 234 | { |
0109e7e6 | 235 | BdrvCoDrainData *data; |
61124f03 | 236 | |
f8ea8dac | 237 | if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) || |
481cad48 | 238 | (!begin && !bs->drv->bdrv_co_drain_end)) { |
61124f03 PB |
239 | return; |
240 | } | |
241 | ||
0109e7e6 KW |
242 | data = g_new(BdrvCoDrainData, 1); |
243 | *data = (BdrvCoDrainData) { | |
244 | .bs = bs, | |
245 | .done = false, | |
8e1da77e HR |
246 | .begin = begin, |
247 | .drained_end_counter = drained_end_counter, | |
0109e7e6 KW |
248 | }; |
249 | ||
e037c09c | 250 | if (!begin) { |
8e1da77e HR |
251 | atomic_inc(drained_end_counter); |
252 | } | |
253 | ||
0109e7e6 KW |
254 | /* Make sure the driver callback completes during the polling phase for |
255 | * drain_begin. */ | |
256 | bdrv_inc_in_flight(bs); | |
257 | data->co = qemu_coroutine_create(bdrv_drain_invoke_entry, data); | |
258 | aio_co_schedule(bdrv_get_aio_context(bs), data->co); | |
61124f03 PB |
259 | } |
260 | ||
1cc8e54a | 261 | /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ |
fe4f0614 | 262 | bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, |
6cd5c9d7 | 263 | BdrvChild *ignore_parent, bool ignore_bds_parents) |
89bd0305 | 264 | { |
fe4f0614 KW |
265 | BdrvChild *child, *next; |
266 | ||
6cd5c9d7 | 267 | if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) { |
89bd0305 KW |
268 | return true; |
269 | } | |
270 | ||
fe4f0614 KW |
271 | if (atomic_read(&bs->in_flight)) { |
272 | return true; | |
273 | } | |
274 | ||
275 | if (recursive) { | |
6cd5c9d7 | 276 | assert(!ignore_bds_parents); |
fe4f0614 | 277 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
6cd5c9d7 | 278 | if (bdrv_drain_poll(child->bs, recursive, child, false)) { |
fe4f0614 KW |
279 | return true; |
280 | } | |
281 | } | |
282 | } | |
283 | ||
284 | return false; | |
89bd0305 KW |
285 | } |
286 | ||
fe4f0614 | 287 | static bool bdrv_drain_poll_top_level(BlockDriverState *bs, bool recursive, |
89bd0305 | 288 | BdrvChild *ignore_parent) |
1cc8e54a | 289 | { |
6cd5c9d7 | 290 | return bdrv_drain_poll(bs, recursive, ignore_parent, false); |
1cc8e54a KW |
291 | } |
292 | ||
b0165585 | 293 | static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, |
6cd5c9d7 KW |
294 | BdrvChild *parent, bool ignore_bds_parents, |
295 | bool poll); | |
b0165585 | 296 | static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, |
8e1da77e HR |
297 | BdrvChild *parent, bool ignore_bds_parents, |
298 | int *drained_end_counter); | |
0152bf40 | 299 | |
a77fd4bb FZ |
300 | static void bdrv_co_drain_bh_cb(void *opaque) |
301 | { | |
302 | BdrvCoDrainData *data = opaque; | |
303 | Coroutine *co = data->co; | |
99723548 | 304 | BlockDriverState *bs = data->bs; |
a77fd4bb | 305 | |
c8ca33d0 | 306 | if (bs) { |
aa1361d5 KW |
307 | AioContext *ctx = bdrv_get_aio_context(bs); |
308 | AioContext *co_ctx = qemu_coroutine_get_aio_context(co); | |
309 | ||
310 | /* | |
311 | * When the coroutine yielded, the lock for its home context was | |
312 | * released, so we need to re-acquire it here. If it explicitly | |
313 | * acquired a different context, the lock is still held and we don't | |
314 | * want to lock it a second time (or AIO_WAIT_WHILE() would hang). | |
315 | */ | |
316 | if (ctx == co_ctx) { | |
317 | aio_context_acquire(ctx); | |
318 | } | |
c8ca33d0 KW |
319 | bdrv_dec_in_flight(bs); |
320 | if (data->begin) { | |
e037c09c | 321 | assert(!data->drained_end_counter); |
6cd5c9d7 KW |
322 | bdrv_do_drained_begin(bs, data->recursive, data->parent, |
323 | data->ignore_bds_parents, data->poll); | |
c8ca33d0 | 324 | } else { |
e037c09c | 325 | assert(!data->poll); |
6cd5c9d7 | 326 | bdrv_do_drained_end(bs, data->recursive, data->parent, |
8e1da77e HR |
327 | data->ignore_bds_parents, |
328 | data->drained_end_counter); | |
c8ca33d0 | 329 | } |
aa1361d5 KW |
330 | if (ctx == co_ctx) { |
331 | aio_context_release(ctx); | |
332 | } | |
481cad48 | 333 | } else { |
c8ca33d0 KW |
334 | assert(data->begin); |
335 | bdrv_drain_all_begin(); | |
481cad48 MP |
336 | } |
337 | ||
a77fd4bb | 338 | data->done = true; |
1919631e | 339 | aio_co_wake(co); |
a77fd4bb FZ |
340 | } |
341 | ||
481cad48 | 342 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, |
b0165585 | 343 | bool begin, bool recursive, |
6cd5c9d7 KW |
344 | BdrvChild *parent, |
345 | bool ignore_bds_parents, | |
8e1da77e HR |
346 | bool poll, |
347 | int *drained_end_counter) | |
a77fd4bb FZ |
348 | { |
349 | BdrvCoDrainData data; | |
350 | ||
351 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
c40a2545 | 352 | * other coroutines run if they were queued by aio_co_enter(). */ |
a77fd4bb FZ |
353 | |
354 | assert(qemu_in_coroutine()); | |
355 | data = (BdrvCoDrainData) { | |
356 | .co = qemu_coroutine_self(), | |
357 | .bs = bs, | |
358 | .done = false, | |
481cad48 | 359 | .begin = begin, |
b0165585 | 360 | .recursive = recursive, |
0152bf40 | 361 | .parent = parent, |
6cd5c9d7 | 362 | .ignore_bds_parents = ignore_bds_parents, |
fe4f0614 | 363 | .poll = poll, |
8e1da77e | 364 | .drained_end_counter = drained_end_counter, |
a77fd4bb | 365 | }; |
8e1da77e | 366 | |
c8ca33d0 KW |
367 | if (bs) { |
368 | bdrv_inc_in_flight(bs); | |
369 | } | |
fffb6e12 PB |
370 | aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), |
371 | bdrv_co_drain_bh_cb, &data); | |
a77fd4bb FZ |
372 | |
373 | qemu_coroutine_yield(); | |
374 | /* If we are resumed from some other event (such as an aio completion or a | |
375 | * timer callback), it is a bug in the caller that should be fixed. */ | |
376 | assert(data.done); | |
377 | } | |
378 | ||
dcf94a23 | 379 | void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, |
6cd5c9d7 | 380 | BdrvChild *parent, bool ignore_bds_parents) |
6820643f | 381 | { |
dcf94a23 | 382 | assert(!qemu_in_coroutine()); |
d42cf288 | 383 | |
60369b86 | 384 | /* Stop things in parent-to-child order */ |
414c2ec3 | 385 | if (atomic_fetch_inc(&bs->quiesce_counter) == 0) { |
6820643f | 386 | aio_disable_external(bdrv_get_aio_context(bs)); |
6820643f KW |
387 | } |
388 | ||
6cd5c9d7 | 389 | bdrv_parent_drained_begin(bs, parent, ignore_bds_parents); |
8e1da77e | 390 | bdrv_drain_invoke(bs, true, NULL); |
dcf94a23 KW |
391 | } |
392 | ||
393 | static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, | |
6cd5c9d7 KW |
394 | BdrvChild *parent, bool ignore_bds_parents, |
395 | bool poll) | |
dcf94a23 KW |
396 | { |
397 | BdrvChild *child, *next; | |
398 | ||
399 | if (qemu_in_coroutine()) { | |
6cd5c9d7 | 400 | bdrv_co_yield_to_drain(bs, true, recursive, parent, ignore_bds_parents, |
8e1da77e | 401 | poll, NULL); |
dcf94a23 KW |
402 | return; |
403 | } | |
404 | ||
6cd5c9d7 | 405 | bdrv_do_drained_begin_quiesce(bs, parent, ignore_bds_parents); |
d30b8e64 | 406 | |
b0165585 | 407 | if (recursive) { |
6cd5c9d7 | 408 | assert(!ignore_bds_parents); |
d736f119 | 409 | bs->recursive_quiesce_counter++; |
b0165585 | 410 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
6cd5c9d7 KW |
411 | bdrv_do_drained_begin(child->bs, true, child, ignore_bds_parents, |
412 | false); | |
b0165585 KW |
413 | } |
414 | } | |
fe4f0614 KW |
415 | |
416 | /* | |
417 | * Wait for drained requests to finish. | |
418 | * | |
419 | * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The | |
420 | * call is needed so things in this AioContext can make progress even | |
421 | * though we don't return to the main AioContext loop - this automatically | |
422 | * includes other nodes in the same AioContext and therefore all child | |
423 | * nodes. | |
424 | */ | |
425 | if (poll) { | |
6cd5c9d7 | 426 | assert(!ignore_bds_parents); |
fe4f0614 KW |
427 | BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, recursive, parent)); |
428 | } | |
6820643f KW |
429 | } |
430 | ||
0152bf40 KW |
431 | void bdrv_drained_begin(BlockDriverState *bs) |
432 | { | |
6cd5c9d7 | 433 | bdrv_do_drained_begin(bs, false, NULL, false, true); |
b0165585 KW |
434 | } |
435 | ||
436 | void bdrv_subtree_drained_begin(BlockDriverState *bs) | |
437 | { | |
6cd5c9d7 | 438 | bdrv_do_drained_begin(bs, true, NULL, false, true); |
0152bf40 KW |
439 | } |
440 | ||
e037c09c HR |
441 | /** |
442 | * This function does not poll, nor must any of its recursively called | |
443 | * functions. The *drained_end_counter pointee will be incremented | |
444 | * once for every background operation scheduled, and decremented once | |
445 | * the operation settles. Therefore, the pointer must remain valid | |
446 | * until the pointee reaches 0. That implies that whoever sets up the | |
447 | * pointee has to poll until it is 0. | |
448 | * | |
449 | * We use atomic operations to access *drained_end_counter, because | |
450 | * (1) when called from bdrv_set_aio_context_ignore(), the subgraph of | |
451 | * @bs may contain nodes in different AioContexts, | |
452 | * (2) bdrv_drain_all_end() uses the same counter for all nodes, | |
453 | * regardless of which AioContext they are in. | |
454 | */ | |
6cd5c9d7 | 455 | static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, |
8e1da77e HR |
456 | BdrvChild *parent, bool ignore_bds_parents, |
457 | int *drained_end_counter) | |
6820643f | 458 | { |
61ad631c | 459 | BdrvChild *child; |
0f115168 KW |
460 | int old_quiesce_counter; |
461 | ||
e037c09c HR |
462 | assert(drained_end_counter != NULL); |
463 | ||
481cad48 | 464 | if (qemu_in_coroutine()) { |
6cd5c9d7 | 465 | bdrv_co_yield_to_drain(bs, false, recursive, parent, ignore_bds_parents, |
8e1da77e | 466 | false, drained_end_counter); |
481cad48 MP |
467 | return; |
468 | } | |
6820643f | 469 | assert(bs->quiesce_counter > 0); |
6820643f | 470 | |
60369b86 | 471 | /* Re-enable things in child-to-parent order */ |
8e1da77e | 472 | bdrv_drain_invoke(bs, false, drained_end_counter); |
e037c09c HR |
473 | bdrv_parent_drained_end(bs, parent, ignore_bds_parents, |
474 | drained_end_counter); | |
5cb2737e HR |
475 | |
476 | old_quiesce_counter = atomic_fetch_dec(&bs->quiesce_counter); | |
0f115168 | 477 | if (old_quiesce_counter == 1) { |
0f115168 KW |
478 | aio_enable_external(bdrv_get_aio_context(bs)); |
479 | } | |
b0165585 KW |
480 | |
481 | if (recursive) { | |
6cd5c9d7 | 482 | assert(!ignore_bds_parents); |
d736f119 | 483 | bs->recursive_quiesce_counter--; |
61ad631c | 484 | QLIST_FOREACH(child, &bs->children, next) { |
8e1da77e HR |
485 | bdrv_do_drained_end(child->bs, true, child, ignore_bds_parents, |
486 | drained_end_counter); | |
b0165585 KW |
487 | } |
488 | } | |
6820643f KW |
489 | } |
490 | ||
0152bf40 KW |
491 | void bdrv_drained_end(BlockDriverState *bs) |
492 | { | |
e037c09c HR |
493 | int drained_end_counter = 0; |
494 | bdrv_do_drained_end(bs, false, NULL, false, &drained_end_counter); | |
495 | BDRV_POLL_WHILE(bs, atomic_read(&drained_end_counter) > 0); | |
496 | } | |
497 | ||
498 | void bdrv_drained_end_no_poll(BlockDriverState *bs, int *drained_end_counter) | |
499 | { | |
500 | bdrv_do_drained_end(bs, false, NULL, false, drained_end_counter); | |
b0165585 KW |
501 | } |
502 | ||
503 | void bdrv_subtree_drained_end(BlockDriverState *bs) | |
504 | { | |
e037c09c HR |
505 | int drained_end_counter = 0; |
506 | bdrv_do_drained_end(bs, true, NULL, false, &drained_end_counter); | |
507 | BDRV_POLL_WHILE(bs, atomic_read(&drained_end_counter) > 0); | |
0152bf40 KW |
508 | } |
509 | ||
d736f119 KW |
510 | void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent) |
511 | { | |
512 | int i; | |
513 | ||
514 | for (i = 0; i < new_parent->recursive_quiesce_counter; i++) { | |
6cd5c9d7 | 515 | bdrv_do_drained_begin(child->bs, true, child, false, true); |
d736f119 KW |
516 | } |
517 | } | |
518 | ||
519 | void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent) | |
520 | { | |
e037c09c | 521 | int drained_end_counter = 0; |
d736f119 KW |
522 | int i; |
523 | ||
524 | for (i = 0; i < old_parent->recursive_quiesce_counter; i++) { | |
e037c09c HR |
525 | bdrv_do_drained_end(child->bs, true, child, false, |
526 | &drained_end_counter); | |
d736f119 | 527 | } |
e037c09c HR |
528 | |
529 | BDRV_POLL_WHILE(child->bs, atomic_read(&drained_end_counter) > 0); | |
d736f119 KW |
530 | } |
531 | ||
61007b31 | 532 | /* |
67da1dc5 FZ |
533 | * Wait for pending requests to complete on a single BlockDriverState subtree, |
534 | * and suspend block driver's internal I/O until next request arrives. | |
61007b31 | 535 | * |
61007b31 SH |
536 | * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState |
537 | * AioContext. | |
538 | */ | |
b6e84c97 | 539 | void coroutine_fn bdrv_co_drain(BlockDriverState *bs) |
61007b31 | 540 | { |
6820643f KW |
541 | assert(qemu_in_coroutine()); |
542 | bdrv_drained_begin(bs); | |
543 | bdrv_drained_end(bs); | |
b6e84c97 | 544 | } |
f406c03c | 545 | |
b6e84c97 PB |
546 | void bdrv_drain(BlockDriverState *bs) |
547 | { | |
6820643f KW |
548 | bdrv_drained_begin(bs); |
549 | bdrv_drained_end(bs); | |
61007b31 SH |
550 | } |
551 | ||
c13ad59f KW |
552 | static void bdrv_drain_assert_idle(BlockDriverState *bs) |
553 | { | |
554 | BdrvChild *child, *next; | |
555 | ||
556 | assert(atomic_read(&bs->in_flight) == 0); | |
557 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { | |
558 | bdrv_drain_assert_idle(child->bs); | |
559 | } | |
560 | } | |
561 | ||
0f12264e KW |
562 | unsigned int bdrv_drain_all_count = 0; |
563 | ||
564 | static bool bdrv_drain_all_poll(void) | |
565 | { | |
566 | BlockDriverState *bs = NULL; | |
567 | bool result = false; | |
568 | ||
0f12264e KW |
569 | /* bdrv_drain_poll() can't make changes to the graph and we are holding the |
570 | * main AioContext lock, so iterating bdrv_next_all_states() is safe. */ | |
571 | while ((bs = bdrv_next_all_states(bs))) { | |
572 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
573 | aio_context_acquire(aio_context); | |
574 | result |= bdrv_drain_poll(bs, false, NULL, true); | |
575 | aio_context_release(aio_context); | |
576 | } | |
577 | ||
578 | return result; | |
579 | } | |
580 | ||
61007b31 SH |
581 | /* |
582 | * Wait for pending requests to complete across all BlockDriverStates | |
583 | * | |
584 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
585 | * after calling this function. | |
c0778f66 AG |
586 | * |
587 | * This pauses all block jobs and disables external clients. It must | |
588 | * be paired with bdrv_drain_all_end(). | |
589 | * | |
590 | * NOTE: no new block jobs or BlockDriverStates can be created between | |
591 | * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls. | |
61007b31 | 592 | */ |
c0778f66 | 593 | void bdrv_drain_all_begin(void) |
61007b31 | 594 | { |
0f12264e | 595 | BlockDriverState *bs = NULL; |
61007b31 | 596 | |
c8ca33d0 | 597 | if (qemu_in_coroutine()) { |
8e1da77e | 598 | bdrv_co_yield_to_drain(NULL, true, false, NULL, true, true, NULL); |
c8ca33d0 KW |
599 | return; |
600 | } | |
601 | ||
0f12264e KW |
602 | /* AIO_WAIT_WHILE() with a NULL context can only be called from the main |
603 | * loop AioContext, so make sure we're in the main context. */ | |
9a7e86c8 | 604 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
605 | assert(bdrv_drain_all_count < INT_MAX); |
606 | bdrv_drain_all_count++; | |
9a7e86c8 | 607 | |
0f12264e KW |
608 | /* Quiesce all nodes, without polling in-flight requests yet. The graph |
609 | * cannot change during this loop. */ | |
610 | while ((bs = bdrv_next_all_states(bs))) { | |
61007b31 SH |
611 | AioContext *aio_context = bdrv_get_aio_context(bs); |
612 | ||
613 | aio_context_acquire(aio_context); | |
0f12264e | 614 | bdrv_do_drained_begin(bs, false, NULL, true, false); |
61007b31 SH |
615 | aio_context_release(aio_context); |
616 | } | |
617 | ||
0f12264e | 618 | /* Now poll the in-flight requests */ |
cfe29d82 | 619 | AIO_WAIT_WHILE(NULL, bdrv_drain_all_poll()); |
0f12264e KW |
620 | |
621 | while ((bs = bdrv_next_all_states(bs))) { | |
c13ad59f | 622 | bdrv_drain_assert_idle(bs); |
61007b31 | 623 | } |
c0778f66 AG |
624 | } |
625 | ||
626 | void bdrv_drain_all_end(void) | |
627 | { | |
0f12264e | 628 | BlockDriverState *bs = NULL; |
e037c09c | 629 | int drained_end_counter = 0; |
c0778f66 | 630 | |
0f12264e | 631 | while ((bs = bdrv_next_all_states(bs))) { |
61007b31 SH |
632 | AioContext *aio_context = bdrv_get_aio_context(bs); |
633 | ||
634 | aio_context_acquire(aio_context); | |
e037c09c | 635 | bdrv_do_drained_end(bs, false, NULL, true, &drained_end_counter); |
61007b31 SH |
636 | aio_context_release(aio_context); |
637 | } | |
0f12264e | 638 | |
e037c09c HR |
639 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
640 | AIO_WAIT_WHILE(NULL, atomic_read(&drained_end_counter) > 0); | |
641 | ||
0f12264e KW |
642 | assert(bdrv_drain_all_count > 0); |
643 | bdrv_drain_all_count--; | |
61007b31 SH |
644 | } |
645 | ||
c0778f66 AG |
646 | void bdrv_drain_all(void) |
647 | { | |
648 | bdrv_drain_all_begin(); | |
649 | bdrv_drain_all_end(); | |
650 | } | |
651 | ||
61007b31 SH |
652 | /** |
653 | * Remove an active request from the tracked requests list | |
654 | * | |
655 | * This function should be called when a tracked request is completing. | |
656 | */ | |
657 | static void tracked_request_end(BdrvTrackedRequest *req) | |
658 | { | |
659 | if (req->serialising) { | |
20fc71b2 | 660 | atomic_dec(&req->bs->serialising_in_flight); |
61007b31 SH |
661 | } |
662 | ||
3783fa3d | 663 | qemu_co_mutex_lock(&req->bs->reqs_lock); |
61007b31 SH |
664 | QLIST_REMOVE(req, list); |
665 | qemu_co_queue_restart_all(&req->wait_queue); | |
3783fa3d | 666 | qemu_co_mutex_unlock(&req->bs->reqs_lock); |
61007b31 SH |
667 | } |
668 | ||
669 | /** | |
670 | * Add an active request to the tracked requests list | |
671 | */ | |
672 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
673 | BlockDriverState *bs, | |
674 | int64_t offset, | |
22931a15 | 675 | uint64_t bytes, |
ebde595c | 676 | enum BdrvTrackedRequestType type) |
61007b31 | 677 | { |
22931a15 FZ |
678 | assert(bytes <= INT64_MAX && offset <= INT64_MAX - bytes); |
679 | ||
61007b31 SH |
680 | *req = (BdrvTrackedRequest){ |
681 | .bs = bs, | |
682 | .offset = offset, | |
683 | .bytes = bytes, | |
ebde595c | 684 | .type = type, |
61007b31 SH |
685 | .co = qemu_coroutine_self(), |
686 | .serialising = false, | |
687 | .overlap_offset = offset, | |
688 | .overlap_bytes = bytes, | |
689 | }; | |
690 | ||
691 | qemu_co_queue_init(&req->wait_queue); | |
692 | ||
3783fa3d | 693 | qemu_co_mutex_lock(&bs->reqs_lock); |
61007b31 | 694 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
3783fa3d | 695 | qemu_co_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
696 | } |
697 | ||
698 | static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align) | |
699 | { | |
700 | int64_t overlap_offset = req->offset & ~(align - 1); | |
22931a15 | 701 | uint64_t overlap_bytes = ROUND_UP(req->offset + req->bytes, align) |
61007b31 SH |
702 | - overlap_offset; |
703 | ||
704 | if (!req->serialising) { | |
20fc71b2 | 705 | atomic_inc(&req->bs->serialising_in_flight); |
61007b31 SH |
706 | req->serialising = true; |
707 | } | |
708 | ||
709 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
710 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
711 | } | |
712 | ||
09d2f948 VSO |
713 | static bool is_request_serialising_and_aligned(BdrvTrackedRequest *req) |
714 | { | |
715 | /* | |
716 | * If the request is serialising, overlap_offset and overlap_bytes are set, | |
717 | * so we can check if the request is aligned. Otherwise, don't care and | |
718 | * return false. | |
719 | */ | |
720 | ||
721 | return req->serialising && (req->offset == req->overlap_offset) && | |
722 | (req->bytes == req->overlap_bytes); | |
723 | } | |
724 | ||
244483e6 KW |
725 | /** |
726 | * Round a region to cluster boundaries | |
727 | */ | |
728 | void bdrv_round_to_clusters(BlockDriverState *bs, | |
7cfd5275 | 729 | int64_t offset, int64_t bytes, |
244483e6 | 730 | int64_t *cluster_offset, |
7cfd5275 | 731 | int64_t *cluster_bytes) |
244483e6 KW |
732 | { |
733 | BlockDriverInfo bdi; | |
734 | ||
735 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
736 | *cluster_offset = offset; | |
737 | *cluster_bytes = bytes; | |
738 | } else { | |
739 | int64_t c = bdi.cluster_size; | |
740 | *cluster_offset = QEMU_ALIGN_DOWN(offset, c); | |
741 | *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c); | |
742 | } | |
743 | } | |
744 | ||
61007b31 SH |
745 | static int bdrv_get_cluster_size(BlockDriverState *bs) |
746 | { | |
747 | BlockDriverInfo bdi; | |
748 | int ret; | |
749 | ||
750 | ret = bdrv_get_info(bs, &bdi); | |
751 | if (ret < 0 || bdi.cluster_size == 0) { | |
a5b8dd2c | 752 | return bs->bl.request_alignment; |
61007b31 SH |
753 | } else { |
754 | return bdi.cluster_size; | |
755 | } | |
756 | } | |
757 | ||
758 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
22931a15 | 759 | int64_t offset, uint64_t bytes) |
61007b31 SH |
760 | { |
761 | /* aaaa bbbb */ | |
762 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
763 | return false; | |
764 | } | |
765 | /* bbbb aaaa */ | |
766 | if (req->overlap_offset >= offset + bytes) { | |
767 | return false; | |
768 | } | |
769 | return true; | |
770 | } | |
771 | ||
99723548 PB |
772 | void bdrv_inc_in_flight(BlockDriverState *bs) |
773 | { | |
774 | atomic_inc(&bs->in_flight); | |
775 | } | |
776 | ||
c9d1a561 PB |
777 | void bdrv_wakeup(BlockDriverState *bs) |
778 | { | |
cfe29d82 | 779 | aio_wait_kick(); |
c9d1a561 PB |
780 | } |
781 | ||
99723548 PB |
782 | void bdrv_dec_in_flight(BlockDriverState *bs) |
783 | { | |
784 | atomic_dec(&bs->in_flight); | |
c9d1a561 | 785 | bdrv_wakeup(bs); |
99723548 PB |
786 | } |
787 | ||
61007b31 SH |
788 | static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self) |
789 | { | |
790 | BlockDriverState *bs = self->bs; | |
791 | BdrvTrackedRequest *req; | |
792 | bool retry; | |
793 | bool waited = false; | |
794 | ||
20fc71b2 | 795 | if (!atomic_read(&bs->serialising_in_flight)) { |
61007b31 SH |
796 | return false; |
797 | } | |
798 | ||
799 | do { | |
800 | retry = false; | |
3783fa3d | 801 | qemu_co_mutex_lock(&bs->reqs_lock); |
61007b31 SH |
802 | QLIST_FOREACH(req, &bs->tracked_requests, list) { |
803 | if (req == self || (!req->serialising && !self->serialising)) { | |
804 | continue; | |
805 | } | |
806 | if (tracked_request_overlaps(req, self->overlap_offset, | |
807 | self->overlap_bytes)) | |
808 | { | |
809 | /* Hitting this means there was a reentrant request, for | |
810 | * example, a block driver issuing nested requests. This must | |
811 | * never happen since it means deadlock. | |
812 | */ | |
813 | assert(qemu_coroutine_self() != req->co); | |
814 | ||
815 | /* If the request is already (indirectly) waiting for us, or | |
816 | * will wait for us as soon as it wakes up, then just go on | |
817 | * (instead of producing a deadlock in the former case). */ | |
818 | if (!req->waiting_for) { | |
819 | self->waiting_for = req; | |
3783fa3d | 820 | qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock); |
61007b31 SH |
821 | self->waiting_for = NULL; |
822 | retry = true; | |
823 | waited = true; | |
824 | break; | |
825 | } | |
826 | } | |
827 | } | |
3783fa3d | 828 | qemu_co_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
829 | } while (retry); |
830 | ||
831 | return waited; | |
832 | } | |
833 | ||
834 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
835 | size_t size) | |
836 | { | |
41ae31e3 | 837 | if (size > BDRV_REQUEST_MAX_BYTES) { |
61007b31 SH |
838 | return -EIO; |
839 | } | |
840 | ||
841 | if (!bdrv_is_inserted(bs)) { | |
842 | return -ENOMEDIUM; | |
843 | } | |
844 | ||
845 | if (offset < 0) { | |
846 | return -EIO; | |
847 | } | |
848 | ||
849 | return 0; | |
850 | } | |
851 | ||
61007b31 | 852 | typedef struct RwCo { |
e293b7a3 | 853 | BdrvChild *child; |
61007b31 SH |
854 | int64_t offset; |
855 | QEMUIOVector *qiov; | |
856 | bool is_write; | |
857 | int ret; | |
858 | BdrvRequestFlags flags; | |
859 | } RwCo; | |
860 | ||
861 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
862 | { | |
863 | RwCo *rwco = opaque; | |
864 | ||
865 | if (!rwco->is_write) { | |
a03ef88f | 866 | rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset, |
cab3a356 KW |
867 | rwco->qiov->size, rwco->qiov, |
868 | rwco->flags); | |
61007b31 | 869 | } else { |
a03ef88f | 870 | rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset, |
cab3a356 KW |
871 | rwco->qiov->size, rwco->qiov, |
872 | rwco->flags); | |
61007b31 | 873 | } |
4720cbee | 874 | aio_wait_kick(); |
61007b31 SH |
875 | } |
876 | ||
877 | /* | |
878 | * Process a vectored synchronous request using coroutines | |
879 | */ | |
e293b7a3 | 880 | static int bdrv_prwv_co(BdrvChild *child, int64_t offset, |
61007b31 SH |
881 | QEMUIOVector *qiov, bool is_write, |
882 | BdrvRequestFlags flags) | |
883 | { | |
884 | Coroutine *co; | |
885 | RwCo rwco = { | |
e293b7a3 | 886 | .child = child, |
61007b31 SH |
887 | .offset = offset, |
888 | .qiov = qiov, | |
889 | .is_write = is_write, | |
890 | .ret = NOT_DONE, | |
891 | .flags = flags, | |
892 | }; | |
893 | ||
61007b31 SH |
894 | if (qemu_in_coroutine()) { |
895 | /* Fast-path if already in coroutine context */ | |
896 | bdrv_rw_co_entry(&rwco); | |
897 | } else { | |
0b8b8753 | 898 | co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco); |
e92f0e19 | 899 | bdrv_coroutine_enter(child->bs, co); |
88b062c2 | 900 | BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE); |
61007b31 SH |
901 | } |
902 | return rwco.ret; | |
903 | } | |
904 | ||
720ff280 | 905 | int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset, |
f5a5ca79 | 906 | int bytes, BdrvRequestFlags flags) |
61007b31 | 907 | { |
0d93ed08 | 908 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, NULL, bytes); |
74021bc4 | 909 | |
e293b7a3 | 910 | return bdrv_prwv_co(child, offset, &qiov, true, |
74021bc4 | 911 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
912 | } |
913 | ||
914 | /* | |
74021bc4 | 915 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
916 | * The operation is sped up by checking the block status and only writing |
917 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 918 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 919 | * BDRV_REQ_FUA). |
61007b31 SH |
920 | * |
921 | * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). | |
922 | */ | |
720ff280 | 923 | int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags) |
61007b31 | 924 | { |
237d78f8 EB |
925 | int ret; |
926 | int64_t target_size, bytes, offset = 0; | |
720ff280 | 927 | BlockDriverState *bs = child->bs; |
61007b31 | 928 | |
7286d610 EB |
929 | target_size = bdrv_getlength(bs); |
930 | if (target_size < 0) { | |
931 | return target_size; | |
61007b31 SH |
932 | } |
933 | ||
934 | for (;;) { | |
7286d610 EB |
935 | bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES); |
936 | if (bytes <= 0) { | |
61007b31 SH |
937 | return 0; |
938 | } | |
237d78f8 | 939 | ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL); |
61007b31 | 940 | if (ret < 0) { |
61007b31 SH |
941 | return ret; |
942 | } | |
943 | if (ret & BDRV_BLOCK_ZERO) { | |
237d78f8 | 944 | offset += bytes; |
61007b31 SH |
945 | continue; |
946 | } | |
237d78f8 | 947 | ret = bdrv_pwrite_zeroes(child, offset, bytes, flags); |
61007b31 | 948 | if (ret < 0) { |
61007b31 SH |
949 | return ret; |
950 | } | |
237d78f8 | 951 | offset += bytes; |
61007b31 SH |
952 | } |
953 | } | |
954 | ||
cf2ab8fc | 955 | int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
f1e84741 KW |
956 | { |
957 | int ret; | |
958 | ||
e293b7a3 | 959 | ret = bdrv_prwv_co(child, offset, qiov, false, 0); |
f1e84741 KW |
960 | if (ret < 0) { |
961 | return ret; | |
962 | } | |
963 | ||
964 | return qiov->size; | |
965 | } | |
966 | ||
2e11d756 | 967 | /* See bdrv_pwrite() for the return codes */ |
cf2ab8fc | 968 | int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes) |
61007b31 | 969 | { |
0d93ed08 | 970 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes); |
61007b31 SH |
971 | |
972 | if (bytes < 0) { | |
973 | return -EINVAL; | |
974 | } | |
975 | ||
cf2ab8fc | 976 | return bdrv_preadv(child, offset, &qiov); |
61007b31 SH |
977 | } |
978 | ||
d9ca2ea2 | 979 | int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
61007b31 SH |
980 | { |
981 | int ret; | |
982 | ||
e293b7a3 | 983 | ret = bdrv_prwv_co(child, offset, qiov, true, 0); |
61007b31 SH |
984 | if (ret < 0) { |
985 | return ret; | |
986 | } | |
987 | ||
988 | return qiov->size; | |
989 | } | |
990 | ||
2e11d756 AG |
991 | /* Return no. of bytes on success or < 0 on error. Important errors are: |
992 | -EIO generic I/O error (may happen for all errors) | |
993 | -ENOMEDIUM No media inserted. | |
994 | -EINVAL Invalid offset or number of bytes | |
995 | -EACCES Trying to write a read-only device | |
996 | */ | |
d9ca2ea2 | 997 | int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes) |
61007b31 | 998 | { |
0d93ed08 | 999 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes); |
61007b31 SH |
1000 | |
1001 | if (bytes < 0) { | |
1002 | return -EINVAL; | |
1003 | } | |
1004 | ||
d9ca2ea2 | 1005 | return bdrv_pwritev(child, offset, &qiov); |
61007b31 SH |
1006 | } |
1007 | ||
1008 | /* | |
1009 | * Writes to the file and ensures that no writes are reordered across this | |
1010 | * request (acts as a barrier) | |
1011 | * | |
1012 | * Returns 0 on success, -errno in error cases. | |
1013 | */ | |
d9ca2ea2 KW |
1014 | int bdrv_pwrite_sync(BdrvChild *child, int64_t offset, |
1015 | const void *buf, int count) | |
61007b31 SH |
1016 | { |
1017 | int ret; | |
1018 | ||
d9ca2ea2 | 1019 | ret = bdrv_pwrite(child, offset, buf, count); |
61007b31 SH |
1020 | if (ret < 0) { |
1021 | return ret; | |
1022 | } | |
1023 | ||
d9ca2ea2 | 1024 | ret = bdrv_flush(child->bs); |
855a6a93 KW |
1025 | if (ret < 0) { |
1026 | return ret; | |
61007b31 SH |
1027 | } |
1028 | ||
1029 | return 0; | |
1030 | } | |
1031 | ||
08844473 KW |
1032 | typedef struct CoroutineIOCompletion { |
1033 | Coroutine *coroutine; | |
1034 | int ret; | |
1035 | } CoroutineIOCompletion; | |
1036 | ||
1037 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
1038 | { | |
1039 | CoroutineIOCompletion *co = opaque; | |
1040 | ||
1041 | co->ret = ret; | |
b9e413dd | 1042 | aio_co_wake(co->coroutine); |
08844473 KW |
1043 | } |
1044 | ||
166fe960 KW |
1045 | static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs, |
1046 | uint64_t offset, uint64_t bytes, | |
1047 | QEMUIOVector *qiov, int flags) | |
1048 | { | |
1049 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
1050 | int64_t sector_num; |
1051 | unsigned int nb_sectors; | |
1052 | ||
fa166538 | 1053 | assert(!(flags & ~BDRV_REQ_MASK)); |
fe0480d6 | 1054 | assert(!(flags & BDRV_REQ_NO_FALLBACK)); |
fa166538 | 1055 | |
d470ad42 HR |
1056 | if (!drv) { |
1057 | return -ENOMEDIUM; | |
1058 | } | |
1059 | ||
3fb06697 KW |
1060 | if (drv->bdrv_co_preadv) { |
1061 | return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); | |
1062 | } | |
1063 | ||
edfab6a0 | 1064 | if (drv->bdrv_aio_preadv) { |
08844473 KW |
1065 | BlockAIOCB *acb; |
1066 | CoroutineIOCompletion co = { | |
1067 | .coroutine = qemu_coroutine_self(), | |
1068 | }; | |
1069 | ||
edfab6a0 EB |
1070 | acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags, |
1071 | bdrv_co_io_em_complete, &co); | |
08844473 KW |
1072 | if (acb == NULL) { |
1073 | return -EIO; | |
1074 | } else { | |
1075 | qemu_coroutine_yield(); | |
1076 | return co.ret; | |
1077 | } | |
1078 | } | |
edfab6a0 EB |
1079 | |
1080 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1081 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1082 | ||
1083 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1084 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
41ae31e3 | 1085 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 EB |
1086 | assert(drv->bdrv_co_readv); |
1087 | ||
1088 | return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
166fe960 KW |
1089 | } |
1090 | ||
78a07294 KW |
1091 | static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs, |
1092 | uint64_t offset, uint64_t bytes, | |
1093 | QEMUIOVector *qiov, int flags) | |
1094 | { | |
1095 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
1096 | int64_t sector_num; |
1097 | unsigned int nb_sectors; | |
78a07294 KW |
1098 | int ret; |
1099 | ||
fa166538 | 1100 | assert(!(flags & ~BDRV_REQ_MASK)); |
fe0480d6 | 1101 | assert(!(flags & BDRV_REQ_NO_FALLBACK)); |
fa166538 | 1102 | |
d470ad42 HR |
1103 | if (!drv) { |
1104 | return -ENOMEDIUM; | |
1105 | } | |
1106 | ||
3fb06697 | 1107 | if (drv->bdrv_co_pwritev) { |
515c2f43 KW |
1108 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, |
1109 | flags & bs->supported_write_flags); | |
1110 | flags &= ~bs->supported_write_flags; | |
3fb06697 KW |
1111 | goto emulate_flags; |
1112 | } | |
1113 | ||
edfab6a0 | 1114 | if (drv->bdrv_aio_pwritev) { |
08844473 KW |
1115 | BlockAIOCB *acb; |
1116 | CoroutineIOCompletion co = { | |
1117 | .coroutine = qemu_coroutine_self(), | |
1118 | }; | |
1119 | ||
edfab6a0 EB |
1120 | acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov, |
1121 | flags & bs->supported_write_flags, | |
1122 | bdrv_co_io_em_complete, &co); | |
1123 | flags &= ~bs->supported_write_flags; | |
08844473 | 1124 | if (acb == NULL) { |
3fb06697 | 1125 | ret = -EIO; |
08844473 KW |
1126 | } else { |
1127 | qemu_coroutine_yield(); | |
3fb06697 | 1128 | ret = co.ret; |
08844473 | 1129 | } |
edfab6a0 EB |
1130 | goto emulate_flags; |
1131 | } | |
1132 | ||
1133 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1134 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1135 | ||
1136 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1137 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
41ae31e3 | 1138 | assert(bytes <= BDRV_REQUEST_MAX_BYTES); |
edfab6a0 | 1139 | |
e18a58b4 EB |
1140 | assert(drv->bdrv_co_writev); |
1141 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov, | |
1142 | flags & bs->supported_write_flags); | |
1143 | flags &= ~bs->supported_write_flags; | |
78a07294 | 1144 | |
3fb06697 | 1145 | emulate_flags: |
4df863f3 | 1146 | if (ret == 0 && (flags & BDRV_REQ_FUA)) { |
78a07294 KW |
1147 | ret = bdrv_co_flush(bs); |
1148 | } | |
1149 | ||
1150 | return ret; | |
1151 | } | |
1152 | ||
29a298af PB |
1153 | static int coroutine_fn |
1154 | bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset, | |
1155 | uint64_t bytes, QEMUIOVector *qiov) | |
1156 | { | |
1157 | BlockDriver *drv = bs->drv; | |
1158 | ||
d470ad42 HR |
1159 | if (!drv) { |
1160 | return -ENOMEDIUM; | |
1161 | } | |
1162 | ||
29a298af PB |
1163 | if (!drv->bdrv_co_pwritev_compressed) { |
1164 | return -ENOTSUP; | |
1165 | } | |
1166 | ||
29a298af PB |
1167 | return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov); |
1168 | } | |
1169 | ||
85c97ca7 | 1170 | static int coroutine_fn bdrv_co_do_copy_on_readv(BdrvChild *child, |
3299e5ec VSO |
1171 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1172 | int flags) | |
61007b31 | 1173 | { |
85c97ca7 KW |
1174 | BlockDriverState *bs = child->bs; |
1175 | ||
61007b31 SH |
1176 | /* Perform I/O through a temporary buffer so that users who scribble over |
1177 | * their read buffer while the operation is in progress do not end up | |
1178 | * modifying the image file. This is critical for zero-copy guest I/O | |
1179 | * where anything might happen inside guest memory. | |
1180 | */ | |
1181 | void *bounce_buffer; | |
1182 | ||
1183 | BlockDriver *drv = bs->drv; | |
cb2e2878 | 1184 | QEMUIOVector local_qiov; |
244483e6 | 1185 | int64_t cluster_offset; |
7cfd5275 | 1186 | int64_t cluster_bytes; |
61007b31 SH |
1187 | size_t skip_bytes; |
1188 | int ret; | |
cb2e2878 EB |
1189 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
1190 | BDRV_REQUEST_MAX_BYTES); | |
1191 | unsigned int progress = 0; | |
61007b31 | 1192 | |
d470ad42 HR |
1193 | if (!drv) { |
1194 | return -ENOMEDIUM; | |
1195 | } | |
1196 | ||
1bf03e66 KW |
1197 | /* FIXME We cannot require callers to have write permissions when all they |
1198 | * are doing is a read request. If we did things right, write permissions | |
1199 | * would be obtained anyway, but internally by the copy-on-read code. As | |
765d9df9 | 1200 | * long as it is implemented here rather than in a separate filter driver, |
1bf03e66 KW |
1201 | * the copy-on-read code doesn't have its own BdrvChild, however, for which |
1202 | * it could request permissions. Therefore we have to bypass the permission | |
1203 | * system for the moment. */ | |
1204 | // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
afa4b293 | 1205 | |
61007b31 | 1206 | /* Cover entire cluster so no additional backing file I/O is required when |
cb2e2878 EB |
1207 | * allocating cluster in the image file. Note that this value may exceed |
1208 | * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which | |
1209 | * is one reason we loop rather than doing it all at once. | |
61007b31 | 1210 | */ |
244483e6 | 1211 | bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes); |
cb2e2878 | 1212 | skip_bytes = offset - cluster_offset; |
61007b31 | 1213 | |
244483e6 KW |
1214 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
1215 | cluster_offset, cluster_bytes); | |
61007b31 | 1216 | |
cb2e2878 EB |
1217 | bounce_buffer = qemu_try_blockalign(bs, |
1218 | MIN(MIN(max_transfer, cluster_bytes), | |
1219 | MAX_BOUNCE_BUFFER)); | |
61007b31 SH |
1220 | if (bounce_buffer == NULL) { |
1221 | ret = -ENOMEM; | |
1222 | goto err; | |
1223 | } | |
1224 | ||
cb2e2878 EB |
1225 | while (cluster_bytes) { |
1226 | int64_t pnum; | |
61007b31 | 1227 | |
cb2e2878 EB |
1228 | ret = bdrv_is_allocated(bs, cluster_offset, |
1229 | MIN(cluster_bytes, max_transfer), &pnum); | |
1230 | if (ret < 0) { | |
1231 | /* Safe to treat errors in querying allocation as if | |
1232 | * unallocated; we'll probably fail again soon on the | |
1233 | * read, but at least that will set a decent errno. | |
1234 | */ | |
1235 | pnum = MIN(cluster_bytes, max_transfer); | |
1236 | } | |
61007b31 | 1237 | |
b0ddcbbb KW |
1238 | /* Stop at EOF if the image ends in the middle of the cluster */ |
1239 | if (ret == 0 && pnum == 0) { | |
1240 | assert(progress >= bytes); | |
1241 | break; | |
1242 | } | |
1243 | ||
cb2e2878 | 1244 | assert(skip_bytes < pnum); |
61007b31 | 1245 | |
cb2e2878 EB |
1246 | if (ret <= 0) { |
1247 | /* Must copy-on-read; use the bounce buffer */ | |
0d93ed08 VSO |
1248 | pnum = MIN(pnum, MAX_BOUNCE_BUFFER); |
1249 | qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum); | |
61007b31 | 1250 | |
cb2e2878 EB |
1251 | ret = bdrv_driver_preadv(bs, cluster_offset, pnum, |
1252 | &local_qiov, 0); | |
1253 | if (ret < 0) { | |
1254 | goto err; | |
1255 | } | |
1256 | ||
1257 | bdrv_debug_event(bs, BLKDBG_COR_WRITE); | |
1258 | if (drv->bdrv_co_pwrite_zeroes && | |
1259 | buffer_is_zero(bounce_buffer, pnum)) { | |
1260 | /* FIXME: Should we (perhaps conditionally) be setting | |
1261 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
1262 | * that still correctly reads as zero? */ | |
7adcf59f HR |
1263 | ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, pnum, |
1264 | BDRV_REQ_WRITE_UNCHANGED); | |
cb2e2878 EB |
1265 | } else { |
1266 | /* This does not change the data on the disk, it is not | |
1267 | * necessary to flush even in cache=writethrough mode. | |
1268 | */ | |
1269 | ret = bdrv_driver_pwritev(bs, cluster_offset, pnum, | |
7adcf59f HR |
1270 | &local_qiov, |
1271 | BDRV_REQ_WRITE_UNCHANGED); | |
cb2e2878 EB |
1272 | } |
1273 | ||
1274 | if (ret < 0) { | |
1275 | /* It might be okay to ignore write errors for guest | |
1276 | * requests. If this is a deliberate copy-on-read | |
1277 | * then we don't want to ignore the error. Simply | |
1278 | * report it in all cases. | |
1279 | */ | |
1280 | goto err; | |
1281 | } | |
1282 | ||
3299e5ec VSO |
1283 | if (!(flags & BDRV_REQ_PREFETCH)) { |
1284 | qemu_iovec_from_buf(qiov, progress, bounce_buffer + skip_bytes, | |
1285 | pnum - skip_bytes); | |
1286 | } | |
1287 | } else if (!(flags & BDRV_REQ_PREFETCH)) { | |
cb2e2878 EB |
1288 | /* Read directly into the destination */ |
1289 | qemu_iovec_init(&local_qiov, qiov->niov); | |
1290 | qemu_iovec_concat(&local_qiov, qiov, progress, pnum - skip_bytes); | |
1291 | ret = bdrv_driver_preadv(bs, offset + progress, local_qiov.size, | |
1292 | &local_qiov, 0); | |
1293 | qemu_iovec_destroy(&local_qiov); | |
1294 | if (ret < 0) { | |
1295 | goto err; | |
1296 | } | |
1297 | } | |
1298 | ||
1299 | cluster_offset += pnum; | |
1300 | cluster_bytes -= pnum; | |
1301 | progress += pnum - skip_bytes; | |
1302 | skip_bytes = 0; | |
1303 | } | |
1304 | ret = 0; | |
61007b31 SH |
1305 | |
1306 | err: | |
1307 | qemu_vfree(bounce_buffer); | |
1308 | return ret; | |
1309 | } | |
1310 | ||
1311 | /* | |
1312 | * Forwards an already correctly aligned request to the BlockDriver. This | |
1a62d0ac EB |
1313 | * handles copy on read, zeroing after EOF, and fragmentation of large |
1314 | * reads; any other features must be implemented by the caller. | |
61007b31 | 1315 | */ |
85c97ca7 | 1316 | static int coroutine_fn bdrv_aligned_preadv(BdrvChild *child, |
61007b31 SH |
1317 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, |
1318 | int64_t align, QEMUIOVector *qiov, int flags) | |
1319 | { | |
85c97ca7 | 1320 | BlockDriverState *bs = child->bs; |
c9d20029 | 1321 | int64_t total_bytes, max_bytes; |
1a62d0ac EB |
1322 | int ret = 0; |
1323 | uint64_t bytes_remaining = bytes; | |
1324 | int max_transfer; | |
61007b31 | 1325 | |
49c07526 KW |
1326 | assert(is_power_of_2(align)); |
1327 | assert((offset & (align - 1)) == 0); | |
1328 | assert((bytes & (align - 1)) == 0); | |
61007b31 | 1329 | assert(!qiov || bytes == qiov->size); |
abb06c5a | 1330 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
1a62d0ac EB |
1331 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1332 | align); | |
a604fa2b EB |
1333 | |
1334 | /* TODO: We would need a per-BDS .supported_read_flags and | |
1335 | * potential fallback support, if we ever implement any read flags | |
1336 | * to pass through to drivers. For now, there aren't any | |
1337 | * passthrough flags. */ | |
3299e5ec VSO |
1338 | assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ | |
1339 | BDRV_REQ_PREFETCH))); | |
61007b31 SH |
1340 | |
1341 | /* Handle Copy on Read and associated serialisation */ | |
1342 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
1343 | /* If we touch the same cluster it counts as an overlap. This | |
1344 | * guarantees that allocating writes will be serialized and not race | |
1345 | * with each other for the same cluster. For example, in copy-on-read | |
1346 | * it ensures that the CoR read and write operations are atomic and | |
1347 | * guest writes cannot interleave between them. */ | |
1348 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
1349 | } | |
1350 | ||
09d2f948 VSO |
1351 | /* BDRV_REQ_SERIALISING is only for write operation */ |
1352 | assert(!(flags & BDRV_REQ_SERIALISING)); | |
1353 | ||
61408b25 FZ |
1354 | if (!(flags & BDRV_REQ_NO_SERIALISING)) { |
1355 | wait_serialising_requests(req); | |
1356 | } | |
61007b31 SH |
1357 | |
1358 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
d6a644bb | 1359 | int64_t pnum; |
61007b31 | 1360 | |
88e63df2 | 1361 | ret = bdrv_is_allocated(bs, offset, bytes, &pnum); |
61007b31 SH |
1362 | if (ret < 0) { |
1363 | goto out; | |
1364 | } | |
1365 | ||
88e63df2 | 1366 | if (!ret || pnum != bytes) { |
3299e5ec VSO |
1367 | ret = bdrv_co_do_copy_on_readv(child, offset, bytes, qiov, flags); |
1368 | goto out; | |
1369 | } else if (flags & BDRV_REQ_PREFETCH) { | |
61007b31 SH |
1370 | goto out; |
1371 | } | |
1372 | } | |
1373 | ||
1a62d0ac | 1374 | /* Forward the request to the BlockDriver, possibly fragmenting it */ |
c9d20029 KW |
1375 | total_bytes = bdrv_getlength(bs); |
1376 | if (total_bytes < 0) { | |
1377 | ret = total_bytes; | |
1378 | goto out; | |
1379 | } | |
61007b31 | 1380 | |
c9d20029 | 1381 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
1a62d0ac | 1382 | if (bytes <= max_bytes && bytes <= max_transfer) { |
c9d20029 | 1383 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0); |
1a62d0ac EB |
1384 | goto out; |
1385 | } | |
61007b31 | 1386 | |
1a62d0ac EB |
1387 | while (bytes_remaining) { |
1388 | int num; | |
61007b31 | 1389 | |
1a62d0ac EB |
1390 | if (max_bytes) { |
1391 | QEMUIOVector local_qiov; | |
61007b31 | 1392 | |
1a62d0ac EB |
1393 | num = MIN(bytes_remaining, MIN(max_bytes, max_transfer)); |
1394 | assert(num); | |
1395 | qemu_iovec_init(&local_qiov, qiov->niov); | |
1396 | qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num); | |
61007b31 | 1397 | |
1a62d0ac EB |
1398 | ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining, |
1399 | num, &local_qiov, 0); | |
1400 | max_bytes -= num; | |
1401 | qemu_iovec_destroy(&local_qiov); | |
1402 | } else { | |
1403 | num = bytes_remaining; | |
1404 | ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0, | |
1405 | bytes_remaining); | |
1406 | } | |
1407 | if (ret < 0) { | |
1408 | goto out; | |
1409 | } | |
1410 | bytes_remaining -= num; | |
61007b31 SH |
1411 | } |
1412 | ||
1413 | out: | |
1a62d0ac | 1414 | return ret < 0 ? ret : 0; |
61007b31 SH |
1415 | } |
1416 | ||
61007b31 | 1417 | /* |
7a3f542f VSO |
1418 | * Request padding |
1419 | * | |
1420 | * |<---- align ----->| |<----- align ---->| | |
1421 | * |<- head ->|<------------- bytes ------------->|<-- tail -->| | |
1422 | * | | | | | | | |
1423 | * -*----------$-------*-------- ... --------*-----$------------*--- | |
1424 | * | | | | | | | |
1425 | * | offset | | end | | |
1426 | * ALIGN_DOWN(offset) ALIGN_UP(offset) ALIGN_DOWN(end) ALIGN_UP(end) | |
1427 | * [buf ... ) [tail_buf ) | |
1428 | * | |
1429 | * @buf is an aligned allocation needed to store @head and @tail paddings. @head | |
1430 | * is placed at the beginning of @buf and @tail at the @end. | |
1431 | * | |
1432 | * @tail_buf is a pointer to sub-buffer, corresponding to align-sized chunk | |
1433 | * around tail, if tail exists. | |
1434 | * | |
1435 | * @merge_reads is true for small requests, | |
1436 | * if @buf_len == @head + bytes + @tail. In this case it is possible that both | |
1437 | * head and tail exist but @buf_len == align and @tail_buf == @buf. | |
1438 | */ | |
1439 | typedef struct BdrvRequestPadding { | |
1440 | uint8_t *buf; | |
1441 | size_t buf_len; | |
1442 | uint8_t *tail_buf; | |
1443 | size_t head; | |
1444 | size_t tail; | |
1445 | bool merge_reads; | |
1446 | QEMUIOVector local_qiov; | |
1447 | } BdrvRequestPadding; | |
1448 | ||
1449 | static bool bdrv_init_padding(BlockDriverState *bs, | |
1450 | int64_t offset, int64_t bytes, | |
1451 | BdrvRequestPadding *pad) | |
1452 | { | |
1453 | uint64_t align = bs->bl.request_alignment; | |
1454 | size_t sum; | |
1455 | ||
1456 | memset(pad, 0, sizeof(*pad)); | |
1457 | ||
1458 | pad->head = offset & (align - 1); | |
1459 | pad->tail = ((offset + bytes) & (align - 1)); | |
1460 | if (pad->tail) { | |
1461 | pad->tail = align - pad->tail; | |
1462 | } | |
1463 | ||
1464 | if ((!pad->head && !pad->tail) || !bytes) { | |
1465 | return false; | |
1466 | } | |
1467 | ||
1468 | sum = pad->head + bytes + pad->tail; | |
1469 | pad->buf_len = (sum > align && pad->head && pad->tail) ? 2 * align : align; | |
1470 | pad->buf = qemu_blockalign(bs, pad->buf_len); | |
1471 | pad->merge_reads = sum == pad->buf_len; | |
1472 | if (pad->tail) { | |
1473 | pad->tail_buf = pad->buf + pad->buf_len - align; | |
1474 | } | |
1475 | ||
1476 | return true; | |
1477 | } | |
1478 | ||
1479 | static int bdrv_padding_rmw_read(BdrvChild *child, | |
1480 | BdrvTrackedRequest *req, | |
1481 | BdrvRequestPadding *pad, | |
1482 | bool zero_middle) | |
1483 | { | |
1484 | QEMUIOVector local_qiov; | |
1485 | BlockDriverState *bs = child->bs; | |
1486 | uint64_t align = bs->bl.request_alignment; | |
1487 | int ret; | |
1488 | ||
1489 | assert(req->serialising && pad->buf); | |
1490 | ||
1491 | if (pad->head || pad->merge_reads) { | |
1492 | uint64_t bytes = pad->merge_reads ? pad->buf_len : align; | |
1493 | ||
1494 | qemu_iovec_init_buf(&local_qiov, pad->buf, bytes); | |
1495 | ||
1496 | if (pad->head) { | |
1497 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); | |
1498 | } | |
1499 | if (pad->merge_reads && pad->tail) { | |
1500 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); | |
1501 | } | |
1502 | ret = bdrv_aligned_preadv(child, req, req->overlap_offset, bytes, | |
1503 | align, &local_qiov, 0); | |
1504 | if (ret < 0) { | |
1505 | return ret; | |
1506 | } | |
1507 | if (pad->head) { | |
1508 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); | |
1509 | } | |
1510 | if (pad->merge_reads && pad->tail) { | |
1511 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); | |
1512 | } | |
1513 | ||
1514 | if (pad->merge_reads) { | |
1515 | goto zero_mem; | |
1516 | } | |
1517 | } | |
1518 | ||
1519 | if (pad->tail) { | |
1520 | qemu_iovec_init_buf(&local_qiov, pad->tail_buf, align); | |
1521 | ||
1522 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); | |
1523 | ret = bdrv_aligned_preadv( | |
1524 | child, req, | |
1525 | req->overlap_offset + req->overlap_bytes - align, | |
1526 | align, align, &local_qiov, 0); | |
1527 | if (ret < 0) { | |
1528 | return ret; | |
1529 | } | |
1530 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); | |
1531 | } | |
1532 | ||
1533 | zero_mem: | |
1534 | if (zero_middle) { | |
1535 | memset(pad->buf + pad->head, 0, pad->buf_len - pad->head - pad->tail); | |
1536 | } | |
1537 | ||
1538 | return 0; | |
1539 | } | |
1540 | ||
1541 | static void bdrv_padding_destroy(BdrvRequestPadding *pad) | |
1542 | { | |
1543 | if (pad->buf) { | |
1544 | qemu_vfree(pad->buf); | |
1545 | qemu_iovec_destroy(&pad->local_qiov); | |
1546 | } | |
1547 | } | |
1548 | ||
1549 | /* | |
1550 | * bdrv_pad_request | |
1551 | * | |
1552 | * Exchange request parameters with padded request if needed. Don't include RMW | |
1553 | * read of padding, bdrv_padding_rmw_read() should be called separately if | |
1554 | * needed. | |
1555 | * | |
1556 | * All parameters except @bs are in-out: they represent original request at | |
1557 | * function call and padded (if padding needed) at function finish. | |
1558 | * | |
1559 | * Function always succeeds. | |
61007b31 | 1560 | */ |
7a3f542f VSO |
1561 | static bool bdrv_pad_request(BlockDriverState *bs, QEMUIOVector **qiov, |
1562 | int64_t *offset, unsigned int *bytes, | |
1563 | BdrvRequestPadding *pad) | |
1564 | { | |
1565 | if (!bdrv_init_padding(bs, *offset, *bytes, pad)) { | |
1566 | return false; | |
1567 | } | |
1568 | ||
1569 | qemu_iovec_init_extended(&pad->local_qiov, pad->buf, pad->head, | |
1570 | *qiov, 0, *bytes, | |
1571 | pad->buf + pad->buf_len - pad->tail, pad->tail); | |
1572 | *bytes += pad->head + pad->tail; | |
1573 | *offset -= pad->head; | |
1574 | *qiov = &pad->local_qiov; | |
1575 | ||
1576 | return true; | |
1577 | } | |
1578 | ||
a03ef88f | 1579 | int coroutine_fn bdrv_co_preadv(BdrvChild *child, |
61007b31 SH |
1580 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1581 | BdrvRequestFlags flags) | |
1582 | { | |
a03ef88f | 1583 | BlockDriverState *bs = child->bs; |
61007b31 | 1584 | BdrvTrackedRequest req; |
7a3f542f | 1585 | BdrvRequestPadding pad; |
61007b31 SH |
1586 | int ret; |
1587 | ||
7a3f542f | 1588 | trace_bdrv_co_preadv(bs, offset, bytes, flags); |
61007b31 SH |
1589 | |
1590 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1591 | if (ret < 0) { | |
1592 | return ret; | |
1593 | } | |
1594 | ||
99723548 PB |
1595 | bdrv_inc_in_flight(bs); |
1596 | ||
9568b511 | 1597 | /* Don't do copy-on-read if we read data before write operation */ |
d3faa13e | 1598 | if (atomic_read(&bs->copy_on_read) && !(flags & BDRV_REQ_NO_SERIALISING)) { |
61007b31 SH |
1599 | flags |= BDRV_REQ_COPY_ON_READ; |
1600 | } | |
1601 | ||
7a3f542f | 1602 | bdrv_pad_request(bs, &qiov, &offset, &bytes, &pad); |
61007b31 | 1603 | |
ebde595c | 1604 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
7a3f542f VSO |
1605 | ret = bdrv_aligned_preadv(child, &req, offset, bytes, |
1606 | bs->bl.request_alignment, | |
1607 | qiov, flags); | |
61007b31 | 1608 | tracked_request_end(&req); |
99723548 | 1609 | bdrv_dec_in_flight(bs); |
61007b31 | 1610 | |
7a3f542f | 1611 | bdrv_padding_destroy(&pad); |
61007b31 SH |
1612 | |
1613 | return ret; | |
1614 | } | |
1615 | ||
d05aa8bb | 1616 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 1617 | int64_t offset, int bytes, BdrvRequestFlags flags) |
61007b31 SH |
1618 | { |
1619 | BlockDriver *drv = bs->drv; | |
1620 | QEMUIOVector qiov; | |
0d93ed08 | 1621 | void *buf = NULL; |
61007b31 | 1622 | int ret = 0; |
465fe887 | 1623 | bool need_flush = false; |
443668ca DL |
1624 | int head = 0; |
1625 | int tail = 0; | |
61007b31 | 1626 | |
cf081fca | 1627 | int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX); |
a5b8dd2c EB |
1628 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1629 | bs->bl.request_alignment); | |
cb2e2878 | 1630 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER); |
d05aa8bb | 1631 | |
d470ad42 HR |
1632 | if (!drv) { |
1633 | return -ENOMEDIUM; | |
1634 | } | |
1635 | ||
fe0480d6 KW |
1636 | if ((flags & ~bs->supported_zero_flags) & BDRV_REQ_NO_FALLBACK) { |
1637 | return -ENOTSUP; | |
1638 | } | |
1639 | ||
b8d0a980 EB |
1640 | assert(alignment % bs->bl.request_alignment == 0); |
1641 | head = offset % alignment; | |
f5a5ca79 | 1642 | tail = (offset + bytes) % alignment; |
b8d0a980 EB |
1643 | max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment); |
1644 | assert(max_write_zeroes >= bs->bl.request_alignment); | |
61007b31 | 1645 | |
f5a5ca79 MP |
1646 | while (bytes > 0 && !ret) { |
1647 | int num = bytes; | |
61007b31 SH |
1648 | |
1649 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1650 | * to be aligned, and that unaligned requests do not cross cluster |
1651 | * boundaries. | |
61007b31 | 1652 | */ |
443668ca | 1653 | if (head) { |
b2f95fee EB |
1654 | /* Make a small request up to the first aligned sector. For |
1655 | * convenience, limit this request to max_transfer even if | |
1656 | * we don't need to fall back to writes. */ | |
f5a5ca79 | 1657 | num = MIN(MIN(bytes, max_transfer), alignment - head); |
b2f95fee EB |
1658 | head = (head + num) % alignment; |
1659 | assert(num < max_write_zeroes); | |
d05aa8bb | 1660 | } else if (tail && num > alignment) { |
443668ca DL |
1661 | /* Shorten the request to the last aligned sector. */ |
1662 | num -= tail; | |
61007b31 SH |
1663 | } |
1664 | ||
1665 | /* limit request size */ | |
1666 | if (num > max_write_zeroes) { | |
1667 | num = max_write_zeroes; | |
1668 | } | |
1669 | ||
1670 | ret = -ENOTSUP; | |
1671 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1672 | if (drv->bdrv_co_pwrite_zeroes) { |
1673 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1674 | flags & bs->supported_zero_flags); | |
1675 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1676 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1677 | need_flush = true; | |
1678 | } | |
465fe887 EB |
1679 | } else { |
1680 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1681 | } |
1682 | ||
118f9944 | 1683 | if (ret < 0 && !(flags & BDRV_REQ_NO_FALLBACK)) { |
61007b31 | 1684 | /* Fall back to bounce buffer if write zeroes is unsupported */ |
465fe887 EB |
1685 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1686 | ||
1687 | if ((flags & BDRV_REQ_FUA) && | |
1688 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1689 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1690 | * flush on each chunk; use just one at the end */ | |
1691 | write_flags &= ~BDRV_REQ_FUA; | |
1692 | need_flush = true; | |
1693 | } | |
5def6b80 | 1694 | num = MIN(num, max_transfer); |
0d93ed08 VSO |
1695 | if (buf == NULL) { |
1696 | buf = qemu_try_blockalign0(bs, num); | |
1697 | if (buf == NULL) { | |
61007b31 SH |
1698 | ret = -ENOMEM; |
1699 | goto fail; | |
1700 | } | |
61007b31 | 1701 | } |
0d93ed08 | 1702 | qemu_iovec_init_buf(&qiov, buf, num); |
61007b31 | 1703 | |
d05aa8bb | 1704 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags); |
61007b31 SH |
1705 | |
1706 | /* Keep bounce buffer around if it is big enough for all | |
1707 | * all future requests. | |
1708 | */ | |
5def6b80 | 1709 | if (num < max_transfer) { |
0d93ed08 VSO |
1710 | qemu_vfree(buf); |
1711 | buf = NULL; | |
61007b31 SH |
1712 | } |
1713 | } | |
1714 | ||
d05aa8bb | 1715 | offset += num; |
f5a5ca79 | 1716 | bytes -= num; |
61007b31 SH |
1717 | } |
1718 | ||
1719 | fail: | |
465fe887 EB |
1720 | if (ret == 0 && need_flush) { |
1721 | ret = bdrv_co_flush(bs); | |
1722 | } | |
0d93ed08 | 1723 | qemu_vfree(buf); |
61007b31 SH |
1724 | return ret; |
1725 | } | |
1726 | ||
85fe2479 FZ |
1727 | static inline int coroutine_fn |
1728 | bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes, | |
1729 | BdrvTrackedRequest *req, int flags) | |
1730 | { | |
1731 | BlockDriverState *bs = child->bs; | |
1732 | bool waited; | |
1733 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1734 | ||
1735 | if (bs->read_only) { | |
1736 | return -EPERM; | |
1737 | } | |
1738 | ||
1739 | /* BDRV_REQ_NO_SERIALISING is only for read operation */ | |
1740 | assert(!(flags & BDRV_REQ_NO_SERIALISING)); | |
1741 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
1742 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); | |
1743 | assert(!(flags & ~BDRV_REQ_MASK)); | |
1744 | ||
1745 | if (flags & BDRV_REQ_SERIALISING) { | |
1746 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
1747 | } | |
1748 | ||
1749 | waited = wait_serialising_requests(req); | |
1750 | ||
1751 | assert(!waited || !req->serialising || | |
1752 | is_request_serialising_and_aligned(req)); | |
1753 | assert(req->overlap_offset <= offset); | |
1754 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
cd47d792 | 1755 | assert(end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE); |
85fe2479 | 1756 | |
cd47d792 FZ |
1757 | switch (req->type) { |
1758 | case BDRV_TRACKED_WRITE: | |
1759 | case BDRV_TRACKED_DISCARD: | |
1760 | if (flags & BDRV_REQ_WRITE_UNCHANGED) { | |
1761 | assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
1762 | } else { | |
1763 | assert(child->perm & BLK_PERM_WRITE); | |
1764 | } | |
1765 | return notifier_with_return_list_notify(&bs->before_write_notifiers, | |
1766 | req); | |
1767 | case BDRV_TRACKED_TRUNCATE: | |
1768 | assert(child->perm & BLK_PERM_RESIZE); | |
1769 | return 0; | |
1770 | default: | |
1771 | abort(); | |
85fe2479 | 1772 | } |
85fe2479 FZ |
1773 | } |
1774 | ||
1775 | static inline void coroutine_fn | |
1776 | bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, uint64_t bytes, | |
1777 | BdrvTrackedRequest *req, int ret) | |
1778 | { | |
1779 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1780 | BlockDriverState *bs = child->bs; | |
1781 | ||
1782 | atomic_inc(&bs->write_gen); | |
85fe2479 | 1783 | |
00695c27 FZ |
1784 | /* |
1785 | * Discard cannot extend the image, but in error handling cases, such as | |
1786 | * when reverting a qcow2 cluster allocation, the discarded range can pass | |
1787 | * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD | |
1788 | * here. Instead, just skip it, since semantically a discard request | |
1789 | * beyond EOF cannot expand the image anyway. | |
1790 | */ | |
7f8f03ef | 1791 | if (ret == 0 && |
cd47d792 FZ |
1792 | (req->type == BDRV_TRACKED_TRUNCATE || |
1793 | end_sector > bs->total_sectors) && | |
1794 | req->type != BDRV_TRACKED_DISCARD) { | |
7f8f03ef FZ |
1795 | bs->total_sectors = end_sector; |
1796 | bdrv_parent_cb_resize(bs); | |
1797 | bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS); | |
85fe2479 | 1798 | } |
00695c27 FZ |
1799 | if (req->bytes) { |
1800 | switch (req->type) { | |
1801 | case BDRV_TRACKED_WRITE: | |
1802 | stat64_max(&bs->wr_highest_offset, offset + bytes); | |
1803 | /* fall through, to set dirty bits */ | |
1804 | case BDRV_TRACKED_DISCARD: | |
1805 | bdrv_set_dirty(bs, offset, bytes); | |
1806 | break; | |
1807 | default: | |
1808 | break; | |
1809 | } | |
1810 | } | |
85fe2479 FZ |
1811 | } |
1812 | ||
61007b31 | 1813 | /* |
04ed95f4 EB |
1814 | * Forwards an already correctly aligned write request to the BlockDriver, |
1815 | * after possibly fragmenting it. | |
61007b31 | 1816 | */ |
85c97ca7 | 1817 | static int coroutine_fn bdrv_aligned_pwritev(BdrvChild *child, |
61007b31 | 1818 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, |
cff86b38 | 1819 | int64_t align, QEMUIOVector *qiov, int flags) |
61007b31 | 1820 | { |
85c97ca7 | 1821 | BlockDriverState *bs = child->bs; |
61007b31 | 1822 | BlockDriver *drv = bs->drv; |
61007b31 SH |
1823 | int ret; |
1824 | ||
04ed95f4 EB |
1825 | uint64_t bytes_remaining = bytes; |
1826 | int max_transfer; | |
61007b31 | 1827 | |
d470ad42 HR |
1828 | if (!drv) { |
1829 | return -ENOMEDIUM; | |
1830 | } | |
1831 | ||
d6883bc9 VSO |
1832 | if (bdrv_has_readonly_bitmaps(bs)) { |
1833 | return -EPERM; | |
1834 | } | |
1835 | ||
cff86b38 EB |
1836 | assert(is_power_of_2(align)); |
1837 | assert((offset & (align - 1)) == 0); | |
1838 | assert((bytes & (align - 1)) == 0); | |
61007b31 | 1839 | assert(!qiov || bytes == qiov->size); |
04ed95f4 EB |
1840 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1841 | align); | |
61007b31 | 1842 | |
85fe2479 | 1843 | ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags); |
61007b31 SH |
1844 | |
1845 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 1846 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
f76889e7 | 1847 | qemu_iovec_is_zero(qiov, 0, qiov->size)) { |
61007b31 SH |
1848 | flags |= BDRV_REQ_ZERO_WRITE; |
1849 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
1850 | flags |= BDRV_REQ_MAY_UNMAP; | |
1851 | } | |
1852 | } | |
1853 | ||
1854 | if (ret < 0) { | |
1855 | /* Do nothing, write notifier decided to fail this request */ | |
1856 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
9a4f4c31 | 1857 | bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 1858 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
3ea1a091 PB |
1859 | } else if (flags & BDRV_REQ_WRITE_COMPRESSED) { |
1860 | ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, qiov); | |
04ed95f4 | 1861 | } else if (bytes <= max_transfer) { |
9a4f4c31 | 1862 | bdrv_debug_event(bs, BLKDBG_PWRITEV); |
78a07294 | 1863 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags); |
04ed95f4 EB |
1864 | } else { |
1865 | bdrv_debug_event(bs, BLKDBG_PWRITEV); | |
1866 | while (bytes_remaining) { | |
1867 | int num = MIN(bytes_remaining, max_transfer); | |
1868 | QEMUIOVector local_qiov; | |
1869 | int local_flags = flags; | |
1870 | ||
1871 | assert(num); | |
1872 | if (num < bytes_remaining && (flags & BDRV_REQ_FUA) && | |
1873 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1874 | /* If FUA is going to be emulated by flush, we only | |
1875 | * need to flush on the last iteration */ | |
1876 | local_flags &= ~BDRV_REQ_FUA; | |
1877 | } | |
1878 | qemu_iovec_init(&local_qiov, qiov->niov); | |
1879 | qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num); | |
1880 | ||
1881 | ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining, | |
1882 | num, &local_qiov, local_flags); | |
1883 | qemu_iovec_destroy(&local_qiov); | |
1884 | if (ret < 0) { | |
1885 | break; | |
1886 | } | |
1887 | bytes_remaining -= num; | |
1888 | } | |
61007b31 | 1889 | } |
9a4f4c31 | 1890 | bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 1891 | |
61007b31 | 1892 | if (ret >= 0) { |
04ed95f4 | 1893 | ret = 0; |
61007b31 | 1894 | } |
85fe2479 | 1895 | bdrv_co_write_req_finish(child, offset, bytes, req, ret); |
61007b31 SH |
1896 | |
1897 | return ret; | |
1898 | } | |
1899 | ||
85c97ca7 | 1900 | static int coroutine_fn bdrv_co_do_zero_pwritev(BdrvChild *child, |
9eeb6dd1 FZ |
1901 | int64_t offset, |
1902 | unsigned int bytes, | |
1903 | BdrvRequestFlags flags, | |
1904 | BdrvTrackedRequest *req) | |
1905 | { | |
85c97ca7 | 1906 | BlockDriverState *bs = child->bs; |
9eeb6dd1 | 1907 | QEMUIOVector local_qiov; |
a5b8dd2c | 1908 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 | 1909 | int ret = 0; |
7a3f542f VSO |
1910 | bool padding; |
1911 | BdrvRequestPadding pad; | |
9eeb6dd1 | 1912 | |
7a3f542f VSO |
1913 | padding = bdrv_init_padding(bs, offset, bytes, &pad); |
1914 | if (padding) { | |
9eeb6dd1 FZ |
1915 | mark_request_serialising(req, align); |
1916 | wait_serialising_requests(req); | |
9eeb6dd1 | 1917 | |
7a3f542f VSO |
1918 | bdrv_padding_rmw_read(child, req, &pad, true); |
1919 | ||
1920 | if (pad.head || pad.merge_reads) { | |
1921 | int64_t aligned_offset = offset & ~(align - 1); | |
1922 | int64_t write_bytes = pad.merge_reads ? pad.buf_len : align; | |
1923 | ||
1924 | qemu_iovec_init_buf(&local_qiov, pad.buf, write_bytes); | |
1925 | ret = bdrv_aligned_pwritev(child, req, aligned_offset, write_bytes, | |
1926 | align, &local_qiov, | |
1927 | flags & ~BDRV_REQ_ZERO_WRITE); | |
1928 | if (ret < 0 || pad.merge_reads) { | |
1929 | /* Error or all work is done */ | |
1930 | goto out; | |
1931 | } | |
1932 | offset += write_bytes - pad.head; | |
1933 | bytes -= write_bytes - pad.head; | |
9eeb6dd1 | 1934 | } |
9eeb6dd1 FZ |
1935 | } |
1936 | ||
1937 | assert(!bytes || (offset & (align - 1)) == 0); | |
1938 | if (bytes >= align) { | |
1939 | /* Write the aligned part in the middle. */ | |
1940 | uint64_t aligned_bytes = bytes & ~(align - 1); | |
85c97ca7 | 1941 | ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align, |
9eeb6dd1 FZ |
1942 | NULL, flags); |
1943 | if (ret < 0) { | |
7a3f542f | 1944 | goto out; |
9eeb6dd1 FZ |
1945 | } |
1946 | bytes -= aligned_bytes; | |
1947 | offset += aligned_bytes; | |
1948 | } | |
1949 | ||
1950 | assert(!bytes || (offset & (align - 1)) == 0); | |
1951 | if (bytes) { | |
7a3f542f | 1952 | assert(align == pad.tail + bytes); |
9eeb6dd1 | 1953 | |
7a3f542f | 1954 | qemu_iovec_init_buf(&local_qiov, pad.tail_buf, align); |
85c97ca7 | 1955 | ret = bdrv_aligned_pwritev(child, req, offset, align, align, |
9eeb6dd1 FZ |
1956 | &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE); |
1957 | } | |
9eeb6dd1 | 1958 | |
7a3f542f VSO |
1959 | out: |
1960 | bdrv_padding_destroy(&pad); | |
1961 | ||
1962 | return ret; | |
9eeb6dd1 FZ |
1963 | } |
1964 | ||
61007b31 SH |
1965 | /* |
1966 | * Handle a write request in coroutine context | |
1967 | */ | |
a03ef88f | 1968 | int coroutine_fn bdrv_co_pwritev(BdrvChild *child, |
61007b31 SH |
1969 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1970 | BdrvRequestFlags flags) | |
1971 | { | |
a03ef88f | 1972 | BlockDriverState *bs = child->bs; |
61007b31 | 1973 | BdrvTrackedRequest req; |
a5b8dd2c | 1974 | uint64_t align = bs->bl.request_alignment; |
7a3f542f | 1975 | BdrvRequestPadding pad; |
61007b31 SH |
1976 | int ret; |
1977 | ||
f42cf447 DB |
1978 | trace_bdrv_co_pwritev(child->bs, offset, bytes, flags); |
1979 | ||
61007b31 SH |
1980 | if (!bs->drv) { |
1981 | return -ENOMEDIUM; | |
1982 | } | |
61007b31 SH |
1983 | |
1984 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1985 | if (ret < 0) { | |
1986 | return ret; | |
1987 | } | |
1988 | ||
99723548 | 1989 | bdrv_inc_in_flight(bs); |
61007b31 SH |
1990 | /* |
1991 | * Align write if necessary by performing a read-modify-write cycle. | |
1992 | * Pad qiov with the read parts and be sure to have a tracked request not | |
1993 | * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle. | |
1994 | */ | |
ebde595c | 1995 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 1996 | |
18a59f03 | 1997 | if (flags & BDRV_REQ_ZERO_WRITE) { |
85c97ca7 | 1998 | ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req); |
9eeb6dd1 FZ |
1999 | goto out; |
2000 | } | |
2001 | ||
7a3f542f | 2002 | if (bdrv_pad_request(bs, &qiov, &offset, &bytes, &pad)) { |
61007b31 SH |
2003 | mark_request_serialising(&req, align); |
2004 | wait_serialising_requests(&req); | |
7a3f542f | 2005 | bdrv_padding_rmw_read(child, &req, &pad, false); |
61007b31 SH |
2006 | } |
2007 | ||
85c97ca7 | 2008 | ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align, |
7a3f542f | 2009 | qiov, flags); |
61007b31 | 2010 | |
7a3f542f | 2011 | bdrv_padding_destroy(&pad); |
61007b31 | 2012 | |
9eeb6dd1 FZ |
2013 | out: |
2014 | tracked_request_end(&req); | |
99723548 | 2015 | bdrv_dec_in_flight(bs); |
7a3f542f | 2016 | |
61007b31 SH |
2017 | return ret; |
2018 | } | |
2019 | ||
a03ef88f | 2020 | int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, |
f5a5ca79 | 2021 | int bytes, BdrvRequestFlags flags) |
61007b31 | 2022 | { |
f5a5ca79 | 2023 | trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags); |
61007b31 | 2024 | |
a03ef88f | 2025 | if (!(child->bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
2026 | flags &= ~BDRV_REQ_MAY_UNMAP; |
2027 | } | |
61007b31 | 2028 | |
f5a5ca79 | 2029 | return bdrv_co_pwritev(child, offset, bytes, NULL, |
74021bc4 | 2030 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
2031 | } |
2032 | ||
4085f5c7 JS |
2033 | /* |
2034 | * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not. | |
2035 | */ | |
2036 | int bdrv_flush_all(void) | |
2037 | { | |
2038 | BdrvNextIterator it; | |
2039 | BlockDriverState *bs = NULL; | |
2040 | int result = 0; | |
2041 | ||
2042 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
2043 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
2044 | int ret; | |
2045 | ||
2046 | aio_context_acquire(aio_context); | |
2047 | ret = bdrv_flush(bs); | |
2048 | if (ret < 0 && !result) { | |
2049 | result = ret; | |
2050 | } | |
2051 | aio_context_release(aio_context); | |
2052 | } | |
2053 | ||
2054 | return result; | |
2055 | } | |
2056 | ||
2057 | ||
4bcd936e | 2058 | typedef struct BdrvCoBlockStatusData { |
61007b31 SH |
2059 | BlockDriverState *bs; |
2060 | BlockDriverState *base; | |
c9ce8c4d | 2061 | bool want_zero; |
4bcd936e EB |
2062 | int64_t offset; |
2063 | int64_t bytes; | |
2064 | int64_t *pnum; | |
2065 | int64_t *map; | |
c9ce8c4d | 2066 | BlockDriverState **file; |
4bcd936e | 2067 | int ret; |
61007b31 | 2068 | bool done; |
4bcd936e | 2069 | } BdrvCoBlockStatusData; |
61007b31 | 2070 | |
3e4d0e72 EB |
2071 | int coroutine_fn bdrv_co_block_status_from_file(BlockDriverState *bs, |
2072 | bool want_zero, | |
2073 | int64_t offset, | |
2074 | int64_t bytes, | |
2075 | int64_t *pnum, | |
2076 | int64_t *map, | |
2077 | BlockDriverState **file) | |
f7cc69b3 MP |
2078 | { |
2079 | assert(bs->file && bs->file->bs); | |
3e4d0e72 EB |
2080 | *pnum = bytes; |
2081 | *map = offset; | |
f7cc69b3 | 2082 | *file = bs->file->bs; |
3e4d0e72 | 2083 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; |
f7cc69b3 MP |
2084 | } |
2085 | ||
3e4d0e72 EB |
2086 | int coroutine_fn bdrv_co_block_status_from_backing(BlockDriverState *bs, |
2087 | bool want_zero, | |
2088 | int64_t offset, | |
2089 | int64_t bytes, | |
2090 | int64_t *pnum, | |
2091 | int64_t *map, | |
2092 | BlockDriverState **file) | |
f7cc69b3 MP |
2093 | { |
2094 | assert(bs->backing && bs->backing->bs); | |
3e4d0e72 EB |
2095 | *pnum = bytes; |
2096 | *map = offset; | |
f7cc69b3 | 2097 | *file = bs->backing->bs; |
3e4d0e72 | 2098 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; |
f7cc69b3 MP |
2099 | } |
2100 | ||
61007b31 SH |
2101 | /* |
2102 | * Returns the allocation status of the specified sectors. | |
2103 | * Drivers not implementing the functionality are assumed to not support | |
2104 | * backing files, hence all their sectors are reported as allocated. | |
2105 | * | |
86a3d5c6 EB |
2106 | * If 'want_zero' is true, the caller is querying for mapping |
2107 | * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and | |
2108 | * _ZERO where possible; otherwise, the result favors larger 'pnum', | |
2109 | * with a focus on accurate BDRV_BLOCK_ALLOCATED. | |
c9ce8c4d | 2110 | * |
2e8bc787 | 2111 | * If 'offset' is beyond the end of the disk image the return value is |
fb0d8654 | 2112 | * BDRV_BLOCK_EOF and 'pnum' is set to 0. |
61007b31 | 2113 | * |
2e8bc787 | 2114 | * 'bytes' is the max value 'pnum' should be set to. If bytes goes |
fb0d8654 EB |
2115 | * beyond the end of the disk image it will be clamped; if 'pnum' is set to |
2116 | * the end of the image, then the returned value will include BDRV_BLOCK_EOF. | |
67a0fd2a | 2117 | * |
2e8bc787 EB |
2118 | * 'pnum' is set to the number of bytes (including and immediately |
2119 | * following the specified offset) that are easily known to be in the | |
2120 | * same allocated/unallocated state. Note that a second call starting | |
2121 | * at the original offset plus returned pnum may have the same status. | |
2122 | * The returned value is non-zero on success except at end-of-file. | |
2123 | * | |
2124 | * Returns negative errno on failure. Otherwise, if the | |
2125 | * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are | |
2126 | * set to the host mapping and BDS corresponding to the guest offset. | |
61007b31 | 2127 | */ |
2e8bc787 EB |
2128 | static int coroutine_fn bdrv_co_block_status(BlockDriverState *bs, |
2129 | bool want_zero, | |
2130 | int64_t offset, int64_t bytes, | |
2131 | int64_t *pnum, int64_t *map, | |
2132 | BlockDriverState **file) | |
2133 | { | |
2134 | int64_t total_size; | |
2135 | int64_t n; /* bytes */ | |
efa6e2ed | 2136 | int ret; |
2e8bc787 | 2137 | int64_t local_map = 0; |
298a1665 | 2138 | BlockDriverState *local_file = NULL; |
efa6e2ed EB |
2139 | int64_t aligned_offset, aligned_bytes; |
2140 | uint32_t align; | |
61007b31 | 2141 | |
298a1665 EB |
2142 | assert(pnum); |
2143 | *pnum = 0; | |
2e8bc787 EB |
2144 | total_size = bdrv_getlength(bs); |
2145 | if (total_size < 0) { | |
2146 | ret = total_size; | |
298a1665 | 2147 | goto early_out; |
61007b31 SH |
2148 | } |
2149 | ||
2e8bc787 | 2150 | if (offset >= total_size) { |
298a1665 EB |
2151 | ret = BDRV_BLOCK_EOF; |
2152 | goto early_out; | |
61007b31 | 2153 | } |
2e8bc787 | 2154 | if (!bytes) { |
298a1665 EB |
2155 | ret = 0; |
2156 | goto early_out; | |
9cdcfd9f | 2157 | } |
61007b31 | 2158 | |
2e8bc787 EB |
2159 | n = total_size - offset; |
2160 | if (n < bytes) { | |
2161 | bytes = n; | |
61007b31 SH |
2162 | } |
2163 | ||
d470ad42 HR |
2164 | /* Must be non-NULL or bdrv_getlength() would have failed */ |
2165 | assert(bs->drv); | |
636cb512 | 2166 | if (!bs->drv->bdrv_co_block_status) { |
2e8bc787 | 2167 | *pnum = bytes; |
61007b31 | 2168 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; |
2e8bc787 | 2169 | if (offset + bytes == total_size) { |
fb0d8654 EB |
2170 | ret |= BDRV_BLOCK_EOF; |
2171 | } | |
61007b31 | 2172 | if (bs->drv->protocol_name) { |
2e8bc787 EB |
2173 | ret |= BDRV_BLOCK_OFFSET_VALID; |
2174 | local_map = offset; | |
298a1665 | 2175 | local_file = bs; |
61007b31 | 2176 | } |
298a1665 | 2177 | goto early_out; |
61007b31 SH |
2178 | } |
2179 | ||
99723548 | 2180 | bdrv_inc_in_flight(bs); |
efa6e2ed EB |
2181 | |
2182 | /* Round out to request_alignment boundaries */ | |
86a3d5c6 | 2183 | align = bs->bl.request_alignment; |
efa6e2ed EB |
2184 | aligned_offset = QEMU_ALIGN_DOWN(offset, align); |
2185 | aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset; | |
2186 | ||
636cb512 EB |
2187 | ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset, |
2188 | aligned_bytes, pnum, &local_map, | |
2189 | &local_file); | |
2190 | if (ret < 0) { | |
2191 | *pnum = 0; | |
2192 | goto out; | |
efa6e2ed EB |
2193 | } |
2194 | ||
2e8bc787 | 2195 | /* |
636cb512 | 2196 | * The driver's result must be a non-zero multiple of request_alignment. |
efa6e2ed | 2197 | * Clamp pnum and adjust map to original request. |
2e8bc787 | 2198 | */ |
636cb512 EB |
2199 | assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) && |
2200 | align > offset - aligned_offset); | |
69f47505 VSO |
2201 | if (ret & BDRV_BLOCK_RECURSE) { |
2202 | assert(ret & BDRV_BLOCK_DATA); | |
2203 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
2204 | assert(!(ret & BDRV_BLOCK_ZERO)); | |
2205 | } | |
2206 | ||
efa6e2ed EB |
2207 | *pnum -= offset - aligned_offset; |
2208 | if (*pnum > bytes) { | |
2209 | *pnum = bytes; | |
61007b31 | 2210 | } |
2e8bc787 | 2211 | if (ret & BDRV_BLOCK_OFFSET_VALID) { |
efa6e2ed | 2212 | local_map += offset - aligned_offset; |
2e8bc787 | 2213 | } |
61007b31 SH |
2214 | |
2215 | if (ret & BDRV_BLOCK_RAW) { | |
298a1665 | 2216 | assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file); |
2e8bc787 EB |
2217 | ret = bdrv_co_block_status(local_file, want_zero, local_map, |
2218 | *pnum, pnum, &local_map, &local_file); | |
99723548 | 2219 | goto out; |
61007b31 SH |
2220 | } |
2221 | ||
2222 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
2223 | ret |= BDRV_BLOCK_ALLOCATED; | |
c9ce8c4d | 2224 | } else if (want_zero) { |
61007b31 SH |
2225 | if (bdrv_unallocated_blocks_are_zero(bs)) { |
2226 | ret |= BDRV_BLOCK_ZERO; | |
760e0063 KW |
2227 | } else if (bs->backing) { |
2228 | BlockDriverState *bs2 = bs->backing->bs; | |
2e8bc787 | 2229 | int64_t size2 = bdrv_getlength(bs2); |
c9ce8c4d | 2230 | |
2e8bc787 | 2231 | if (size2 >= 0 && offset >= size2) { |
61007b31 SH |
2232 | ret |= BDRV_BLOCK_ZERO; |
2233 | } | |
2234 | } | |
2235 | } | |
2236 | ||
69f47505 VSO |
2237 | if (want_zero && ret & BDRV_BLOCK_RECURSE && |
2238 | local_file && local_file != bs && | |
61007b31 SH |
2239 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
2240 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
2e8bc787 EB |
2241 | int64_t file_pnum; |
2242 | int ret2; | |
61007b31 | 2243 | |
2e8bc787 EB |
2244 | ret2 = bdrv_co_block_status(local_file, want_zero, local_map, |
2245 | *pnum, &file_pnum, NULL, NULL); | |
61007b31 SH |
2246 | if (ret2 >= 0) { |
2247 | /* Ignore errors. This is just providing extra information, it | |
2248 | * is useful but not necessary. | |
2249 | */ | |
c61e684e EB |
2250 | if (ret2 & BDRV_BLOCK_EOF && |
2251 | (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) { | |
2252 | /* | |
2253 | * It is valid for the format block driver to read | |
2254 | * beyond the end of the underlying file's current | |
2255 | * size; such areas read as zero. | |
2256 | */ | |
61007b31 SH |
2257 | ret |= BDRV_BLOCK_ZERO; |
2258 | } else { | |
2259 | /* Limit request to the range reported by the protocol driver */ | |
2260 | *pnum = file_pnum; | |
2261 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
2262 | } | |
2263 | } | |
2264 | } | |
2265 | ||
99723548 PB |
2266 | out: |
2267 | bdrv_dec_in_flight(bs); | |
2e8bc787 | 2268 | if (ret >= 0 && offset + *pnum == total_size) { |
fb0d8654 EB |
2269 | ret |= BDRV_BLOCK_EOF; |
2270 | } | |
298a1665 EB |
2271 | early_out: |
2272 | if (file) { | |
2273 | *file = local_file; | |
2274 | } | |
2e8bc787 EB |
2275 | if (map) { |
2276 | *map = local_map; | |
2277 | } | |
61007b31 SH |
2278 | return ret; |
2279 | } | |
2280 | ||
5b648c67 EB |
2281 | static int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs, |
2282 | BlockDriverState *base, | |
2283 | bool want_zero, | |
2284 | int64_t offset, | |
2285 | int64_t bytes, | |
2286 | int64_t *pnum, | |
2287 | int64_t *map, | |
2288 | BlockDriverState **file) | |
ba3f0e25 FZ |
2289 | { |
2290 | BlockDriverState *p; | |
5b648c67 | 2291 | int ret = 0; |
c61e684e | 2292 | bool first = true; |
ba3f0e25 FZ |
2293 | |
2294 | assert(bs != base); | |
760e0063 | 2295 | for (p = bs; p != base; p = backing_bs(p)) { |
5b648c67 EB |
2296 | ret = bdrv_co_block_status(p, want_zero, offset, bytes, pnum, map, |
2297 | file); | |
c61e684e EB |
2298 | if (ret < 0) { |
2299 | break; | |
2300 | } | |
2301 | if (ret & BDRV_BLOCK_ZERO && ret & BDRV_BLOCK_EOF && !first) { | |
2302 | /* | |
2303 | * Reading beyond the end of the file continues to read | |
2304 | * zeroes, but we can only widen the result to the | |
2305 | * unallocated length we learned from an earlier | |
2306 | * iteration. | |
2307 | */ | |
5b648c67 | 2308 | *pnum = bytes; |
c61e684e EB |
2309 | } |
2310 | if (ret & (BDRV_BLOCK_ZERO | BDRV_BLOCK_DATA)) { | |
ba3f0e25 FZ |
2311 | break; |
2312 | } | |
5b648c67 EB |
2313 | /* [offset, pnum] unallocated on this layer, which could be only |
2314 | * the first part of [offset, bytes]. */ | |
2315 | bytes = MIN(bytes, *pnum); | |
c61e684e | 2316 | first = false; |
ba3f0e25 FZ |
2317 | } |
2318 | return ret; | |
2319 | } | |
2320 | ||
31826642 | 2321 | /* Coroutine wrapper for bdrv_block_status_above() */ |
5b648c67 | 2322 | static void coroutine_fn bdrv_block_status_above_co_entry(void *opaque) |
61007b31 | 2323 | { |
4bcd936e | 2324 | BdrvCoBlockStatusData *data = opaque; |
61007b31 | 2325 | |
5b648c67 EB |
2326 | data->ret = bdrv_co_block_status_above(data->bs, data->base, |
2327 | data->want_zero, | |
2328 | data->offset, data->bytes, | |
2329 | data->pnum, data->map, data->file); | |
61007b31 | 2330 | data->done = true; |
4720cbee | 2331 | aio_wait_kick(); |
61007b31 SH |
2332 | } |
2333 | ||
2334 | /* | |
5b648c67 | 2335 | * Synchronous wrapper around bdrv_co_block_status_above(). |
61007b31 | 2336 | * |
5b648c67 | 2337 | * See bdrv_co_block_status_above() for details. |
61007b31 | 2338 | */ |
7ddb99b9 EB |
2339 | static int bdrv_common_block_status_above(BlockDriverState *bs, |
2340 | BlockDriverState *base, | |
2341 | bool want_zero, int64_t offset, | |
2342 | int64_t bytes, int64_t *pnum, | |
2343 | int64_t *map, | |
2344 | BlockDriverState **file) | |
61007b31 SH |
2345 | { |
2346 | Coroutine *co; | |
4bcd936e | 2347 | BdrvCoBlockStatusData data = { |
61007b31 | 2348 | .bs = bs, |
ba3f0e25 | 2349 | .base = base, |
c9ce8c4d | 2350 | .want_zero = want_zero, |
7ddb99b9 EB |
2351 | .offset = offset, |
2352 | .bytes = bytes, | |
2353 | .pnum = pnum, | |
2354 | .map = map, | |
c9ce8c4d | 2355 | .file = file, |
61007b31 SH |
2356 | .done = false, |
2357 | }; | |
2358 | ||
2359 | if (qemu_in_coroutine()) { | |
2360 | /* Fast-path if already in coroutine context */ | |
5b648c67 | 2361 | bdrv_block_status_above_co_entry(&data); |
61007b31 | 2362 | } else { |
5b648c67 | 2363 | co = qemu_coroutine_create(bdrv_block_status_above_co_entry, &data); |
e92f0e19 | 2364 | bdrv_coroutine_enter(bs, co); |
88b062c2 | 2365 | BDRV_POLL_WHILE(bs, !data.done); |
61007b31 | 2366 | } |
7ddb99b9 | 2367 | return data.ret; |
61007b31 SH |
2368 | } |
2369 | ||
31826642 EB |
2370 | int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base, |
2371 | int64_t offset, int64_t bytes, int64_t *pnum, | |
2372 | int64_t *map, BlockDriverState **file) | |
c9ce8c4d | 2373 | { |
31826642 EB |
2374 | return bdrv_common_block_status_above(bs, base, true, offset, bytes, |
2375 | pnum, map, file); | |
c9ce8c4d EB |
2376 | } |
2377 | ||
237d78f8 EB |
2378 | int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2379 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
ba3f0e25 | 2380 | { |
31826642 EB |
2381 | return bdrv_block_status_above(bs, backing_bs(bs), |
2382 | offset, bytes, pnum, map, file); | |
ba3f0e25 FZ |
2383 | } |
2384 | ||
d6a644bb EB |
2385 | int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t offset, |
2386 | int64_t bytes, int64_t *pnum) | |
61007b31 | 2387 | { |
7ddb99b9 EB |
2388 | int ret; |
2389 | int64_t dummy; | |
d6a644bb | 2390 | |
7ddb99b9 EB |
2391 | ret = bdrv_common_block_status_above(bs, backing_bs(bs), false, offset, |
2392 | bytes, pnum ? pnum : &dummy, NULL, | |
c9ce8c4d | 2393 | NULL); |
61007b31 SH |
2394 | if (ret < 0) { |
2395 | return ret; | |
2396 | } | |
2397 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2398 | } | |
2399 | ||
2400 | /* | |
2401 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
2402 | * | |
170d3bd3 AS |
2403 | * Return 1 if (a prefix of) the given range is allocated in any image |
2404 | * between BASE and TOP (BASE is only included if include_base is set). | |
2405 | * BASE can be NULL to check if the given offset is allocated in any | |
2406 | * image of the chain. Return 0 otherwise, or negative errno on | |
2407 | * failure. | |
61007b31 | 2408 | * |
51b0a488 EB |
2409 | * 'pnum' is set to the number of bytes (including and immediately |
2410 | * following the specified offset) that are known to be in the same | |
2411 | * allocated/unallocated state. Note that a subsequent call starting | |
2412 | * at 'offset + *pnum' may return the same allocation status (in other | |
2413 | * words, the result is not necessarily the maximum possible range); | |
2414 | * but 'pnum' will only be 0 when end of file is reached. | |
61007b31 SH |
2415 | * |
2416 | */ | |
2417 | int bdrv_is_allocated_above(BlockDriverState *top, | |
2418 | BlockDriverState *base, | |
170d3bd3 AS |
2419 | bool include_base, int64_t offset, |
2420 | int64_t bytes, int64_t *pnum) | |
61007b31 SH |
2421 | { |
2422 | BlockDriverState *intermediate; | |
51b0a488 EB |
2423 | int ret; |
2424 | int64_t n = bytes; | |
61007b31 | 2425 | |
170d3bd3 AS |
2426 | assert(base || !include_base); |
2427 | ||
61007b31 | 2428 | intermediate = top; |
170d3bd3 | 2429 | while (include_base || intermediate != base) { |
d6a644bb | 2430 | int64_t pnum_inter; |
c00716be | 2431 | int64_t size_inter; |
d6a644bb | 2432 | |
170d3bd3 | 2433 | assert(intermediate); |
51b0a488 | 2434 | ret = bdrv_is_allocated(intermediate, offset, bytes, &pnum_inter); |
61007b31 SH |
2435 | if (ret < 0) { |
2436 | return ret; | |
d6a644bb | 2437 | } |
d6a644bb | 2438 | if (ret) { |
51b0a488 | 2439 | *pnum = pnum_inter; |
61007b31 SH |
2440 | return 1; |
2441 | } | |
2442 | ||
51b0a488 | 2443 | size_inter = bdrv_getlength(intermediate); |
c00716be EB |
2444 | if (size_inter < 0) { |
2445 | return size_inter; | |
2446 | } | |
51b0a488 EB |
2447 | if (n > pnum_inter && |
2448 | (intermediate == top || offset + pnum_inter < size_inter)) { | |
2449 | n = pnum_inter; | |
61007b31 SH |
2450 | } |
2451 | ||
170d3bd3 AS |
2452 | if (intermediate == base) { |
2453 | break; | |
2454 | } | |
2455 | ||
760e0063 | 2456 | intermediate = backing_bs(intermediate); |
61007b31 SH |
2457 | } |
2458 | ||
2459 | *pnum = n; | |
2460 | return 0; | |
2461 | } | |
2462 | ||
1a8ae822 KW |
2463 | typedef struct BdrvVmstateCo { |
2464 | BlockDriverState *bs; | |
2465 | QEMUIOVector *qiov; | |
2466 | int64_t pos; | |
2467 | bool is_read; | |
2468 | int ret; | |
2469 | } BdrvVmstateCo; | |
2470 | ||
2471 | static int coroutine_fn | |
2472 | bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
2473 | bool is_read) | |
2474 | { | |
2475 | BlockDriver *drv = bs->drv; | |
dc88a467 SH |
2476 | int ret = -ENOTSUP; |
2477 | ||
2478 | bdrv_inc_in_flight(bs); | |
1a8ae822 KW |
2479 | |
2480 | if (!drv) { | |
dc88a467 | 2481 | ret = -ENOMEDIUM; |
1a8ae822 | 2482 | } else if (drv->bdrv_load_vmstate) { |
dc88a467 SH |
2483 | if (is_read) { |
2484 | ret = drv->bdrv_load_vmstate(bs, qiov, pos); | |
2485 | } else { | |
2486 | ret = drv->bdrv_save_vmstate(bs, qiov, pos); | |
2487 | } | |
1a8ae822 | 2488 | } else if (bs->file) { |
dc88a467 | 2489 | ret = bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read); |
1a8ae822 KW |
2490 | } |
2491 | ||
dc88a467 SH |
2492 | bdrv_dec_in_flight(bs); |
2493 | return ret; | |
1a8ae822 KW |
2494 | } |
2495 | ||
2496 | static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque) | |
2497 | { | |
2498 | BdrvVmstateCo *co = opaque; | |
2499 | co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read); | |
4720cbee | 2500 | aio_wait_kick(); |
1a8ae822 KW |
2501 | } |
2502 | ||
2503 | static inline int | |
2504 | bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
2505 | bool is_read) | |
2506 | { | |
2507 | if (qemu_in_coroutine()) { | |
2508 | return bdrv_co_rw_vmstate(bs, qiov, pos, is_read); | |
2509 | } else { | |
2510 | BdrvVmstateCo data = { | |
2511 | .bs = bs, | |
2512 | .qiov = qiov, | |
2513 | .pos = pos, | |
2514 | .is_read = is_read, | |
2515 | .ret = -EINPROGRESS, | |
2516 | }; | |
0b8b8753 | 2517 | Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data); |
1a8ae822 | 2518 | |
e92f0e19 | 2519 | bdrv_coroutine_enter(bs, co); |
ea17c9d2 | 2520 | BDRV_POLL_WHILE(bs, data.ret == -EINPROGRESS); |
1a8ae822 KW |
2521 | return data.ret; |
2522 | } | |
2523 | } | |
2524 | ||
61007b31 SH |
2525 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
2526 | int64_t pos, int size) | |
2527 | { | |
0d93ed08 | 2528 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b433d942 | 2529 | int ret; |
61007b31 | 2530 | |
b433d942 KW |
2531 | ret = bdrv_writev_vmstate(bs, &qiov, pos); |
2532 | if (ret < 0) { | |
2533 | return ret; | |
2534 | } | |
2535 | ||
2536 | return size; | |
61007b31 SH |
2537 | } |
2538 | ||
2539 | int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
2540 | { | |
1a8ae822 | 2541 | return bdrv_rw_vmstate(bs, qiov, pos, false); |
61007b31 SH |
2542 | } |
2543 | ||
2544 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
2545 | int64_t pos, int size) | |
5ddda0b8 | 2546 | { |
0d93ed08 | 2547 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b433d942 | 2548 | int ret; |
5ddda0b8 | 2549 | |
b433d942 KW |
2550 | ret = bdrv_readv_vmstate(bs, &qiov, pos); |
2551 | if (ret < 0) { | |
2552 | return ret; | |
2553 | } | |
2554 | ||
2555 | return size; | |
5ddda0b8 KW |
2556 | } |
2557 | ||
2558 | int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 2559 | { |
1a8ae822 | 2560 | return bdrv_rw_vmstate(bs, qiov, pos, true); |
61007b31 SH |
2561 | } |
2562 | ||
2563 | /**************************************************************/ | |
2564 | /* async I/Os */ | |
2565 | ||
61007b31 SH |
2566 | void bdrv_aio_cancel(BlockAIOCB *acb) |
2567 | { | |
2568 | qemu_aio_ref(acb); | |
2569 | bdrv_aio_cancel_async(acb); | |
2570 | while (acb->refcnt > 1) { | |
2571 | if (acb->aiocb_info->get_aio_context) { | |
2572 | aio_poll(acb->aiocb_info->get_aio_context(acb), true); | |
2573 | } else if (acb->bs) { | |
2f47da5f PB |
2574 | /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so |
2575 | * assert that we're not using an I/O thread. Thread-safe | |
2576 | * code should use bdrv_aio_cancel_async exclusively. | |
2577 | */ | |
2578 | assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context()); | |
61007b31 SH |
2579 | aio_poll(bdrv_get_aio_context(acb->bs), true); |
2580 | } else { | |
2581 | abort(); | |
2582 | } | |
2583 | } | |
2584 | qemu_aio_unref(acb); | |
2585 | } | |
2586 | ||
2587 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2588 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2589 | * In either case the completion callback must be called. */ | |
2590 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2591 | { | |
2592 | if (acb->aiocb_info->cancel_async) { | |
2593 | acb->aiocb_info->cancel_async(acb); | |
2594 | } | |
2595 | } | |
2596 | ||
61007b31 SH |
2597 | /**************************************************************/ |
2598 | /* Coroutine block device emulation */ | |
2599 | ||
e293b7a3 KW |
2600 | typedef struct FlushCo { |
2601 | BlockDriverState *bs; | |
2602 | int ret; | |
2603 | } FlushCo; | |
2604 | ||
2605 | ||
61007b31 SH |
2606 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) |
2607 | { | |
e293b7a3 | 2608 | FlushCo *rwco = opaque; |
61007b31 SH |
2609 | |
2610 | rwco->ret = bdrv_co_flush(rwco->bs); | |
4720cbee | 2611 | aio_wait_kick(); |
61007b31 SH |
2612 | } |
2613 | ||
2614 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
2615 | { | |
49ca6259 FZ |
2616 | int current_gen; |
2617 | int ret = 0; | |
2618 | ||
2619 | bdrv_inc_in_flight(bs); | |
61007b31 | 2620 | |
e914404e | 2621 | if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs) || |
1b6bc94d | 2622 | bdrv_is_sg(bs)) { |
49ca6259 | 2623 | goto early_exit; |
61007b31 SH |
2624 | } |
2625 | ||
3783fa3d | 2626 | qemu_co_mutex_lock(&bs->reqs_lock); |
47fec599 | 2627 | current_gen = atomic_read(&bs->write_gen); |
3ff2f67a EY |
2628 | |
2629 | /* Wait until any previous flushes are completed */ | |
99723548 | 2630 | while (bs->active_flush_req) { |
3783fa3d | 2631 | qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock); |
3ff2f67a EY |
2632 | } |
2633 | ||
3783fa3d | 2634 | /* Flushes reach this point in nondecreasing current_gen order. */ |
99723548 | 2635 | bs->active_flush_req = true; |
3783fa3d | 2636 | qemu_co_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2637 | |
c32b82af PD |
2638 | /* Write back all layers by calling one driver function */ |
2639 | if (bs->drv->bdrv_co_flush) { | |
2640 | ret = bs->drv->bdrv_co_flush(bs); | |
2641 | goto out; | |
2642 | } | |
2643 | ||
61007b31 SH |
2644 | /* Write back cached data to the OS even with cache=unsafe */ |
2645 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); | |
2646 | if (bs->drv->bdrv_co_flush_to_os) { | |
2647 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
2648 | if (ret < 0) { | |
cdb5e315 | 2649 | goto out; |
61007b31 SH |
2650 | } |
2651 | } | |
2652 | ||
2653 | /* But don't actually force it to the disk with cache=unsafe */ | |
2654 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
2655 | goto flush_parent; | |
2656 | } | |
2657 | ||
3ff2f67a EY |
2658 | /* Check if we really need to flush anything */ |
2659 | if (bs->flushed_gen == current_gen) { | |
2660 | goto flush_parent; | |
2661 | } | |
2662 | ||
61007b31 | 2663 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); |
d470ad42 HR |
2664 | if (!bs->drv) { |
2665 | /* bs->drv->bdrv_co_flush() might have ejected the BDS | |
2666 | * (even in case of apparent success) */ | |
2667 | ret = -ENOMEDIUM; | |
2668 | goto out; | |
2669 | } | |
61007b31 SH |
2670 | if (bs->drv->bdrv_co_flush_to_disk) { |
2671 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
2672 | } else if (bs->drv->bdrv_aio_flush) { | |
2673 | BlockAIOCB *acb; | |
2674 | CoroutineIOCompletion co = { | |
2675 | .coroutine = qemu_coroutine_self(), | |
2676 | }; | |
2677 | ||
2678 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
2679 | if (acb == NULL) { | |
2680 | ret = -EIO; | |
2681 | } else { | |
2682 | qemu_coroutine_yield(); | |
2683 | ret = co.ret; | |
2684 | } | |
2685 | } else { | |
2686 | /* | |
2687 | * Some block drivers always operate in either writethrough or unsafe | |
2688 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
2689 | * know how the server works (because the behaviour is hardcoded or | |
2690 | * depends on server-side configuration), so we can't ensure that | |
2691 | * everything is safe on disk. Returning an error doesn't work because | |
2692 | * that would break guests even if the server operates in writethrough | |
2693 | * mode. | |
2694 | * | |
2695 | * Let's hope the user knows what he's doing. | |
2696 | */ | |
2697 | ret = 0; | |
2698 | } | |
3ff2f67a | 2699 | |
61007b31 | 2700 | if (ret < 0) { |
cdb5e315 | 2701 | goto out; |
61007b31 SH |
2702 | } |
2703 | ||
2704 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
2705 | * in the case of cache=unsafe, so there are no useless flushes. | |
2706 | */ | |
2707 | flush_parent: | |
cdb5e315 FZ |
2708 | ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0; |
2709 | out: | |
3ff2f67a | 2710 | /* Notify any pending flushes that we have completed */ |
e6af1e08 KW |
2711 | if (ret == 0) { |
2712 | bs->flushed_gen = current_gen; | |
2713 | } | |
3783fa3d PB |
2714 | |
2715 | qemu_co_mutex_lock(&bs->reqs_lock); | |
99723548 | 2716 | bs->active_flush_req = false; |
156af3ac DL |
2717 | /* Return value is ignored - it's ok if wait queue is empty */ |
2718 | qemu_co_queue_next(&bs->flush_queue); | |
3783fa3d | 2719 | qemu_co_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2720 | |
49ca6259 | 2721 | early_exit: |
99723548 | 2722 | bdrv_dec_in_flight(bs); |
cdb5e315 | 2723 | return ret; |
61007b31 SH |
2724 | } |
2725 | ||
2726 | int bdrv_flush(BlockDriverState *bs) | |
2727 | { | |
2728 | Coroutine *co; | |
e293b7a3 | 2729 | FlushCo flush_co = { |
61007b31 SH |
2730 | .bs = bs, |
2731 | .ret = NOT_DONE, | |
2732 | }; | |
2733 | ||
2734 | if (qemu_in_coroutine()) { | |
2735 | /* Fast-path if already in coroutine context */ | |
e293b7a3 | 2736 | bdrv_flush_co_entry(&flush_co); |
61007b31 | 2737 | } else { |
0b8b8753 | 2738 | co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co); |
e92f0e19 | 2739 | bdrv_coroutine_enter(bs, co); |
88b062c2 | 2740 | BDRV_POLL_WHILE(bs, flush_co.ret == NOT_DONE); |
61007b31 SH |
2741 | } |
2742 | ||
e293b7a3 | 2743 | return flush_co.ret; |
61007b31 SH |
2744 | } |
2745 | ||
2746 | typedef struct DiscardCo { | |
0b9fd3f4 | 2747 | BdrvChild *child; |
0c51a893 | 2748 | int64_t offset; |
d93e5726 | 2749 | int64_t bytes; |
61007b31 SH |
2750 | int ret; |
2751 | } DiscardCo; | |
0c51a893 | 2752 | static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque) |
61007b31 SH |
2753 | { |
2754 | DiscardCo *rwco = opaque; | |
2755 | ||
0b9fd3f4 | 2756 | rwco->ret = bdrv_co_pdiscard(rwco->child, rwco->offset, rwco->bytes); |
4720cbee | 2757 | aio_wait_kick(); |
61007b31 SH |
2758 | } |
2759 | ||
d93e5726 VSO |
2760 | int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, |
2761 | int64_t bytes) | |
61007b31 | 2762 | { |
b1066c87 | 2763 | BdrvTrackedRequest req; |
9f1963b3 | 2764 | int max_pdiscard, ret; |
3482b9bc | 2765 | int head, tail, align; |
0b9fd3f4 | 2766 | BlockDriverState *bs = child->bs; |
61007b31 | 2767 | |
d93e5726 | 2768 | if (!bs || !bs->drv || !bdrv_is_inserted(bs)) { |
61007b31 SH |
2769 | return -ENOMEDIUM; |
2770 | } | |
2771 | ||
d6883bc9 VSO |
2772 | if (bdrv_has_readonly_bitmaps(bs)) { |
2773 | return -EPERM; | |
2774 | } | |
2775 | ||
d93e5726 VSO |
2776 | if (offset < 0 || bytes < 0 || bytes > INT64_MAX - offset) { |
2777 | return -EIO; | |
61007b31 SH |
2778 | } |
2779 | ||
61007b31 SH |
2780 | /* Do nothing if disabled. */ |
2781 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
2782 | return 0; | |
2783 | } | |
2784 | ||
02aefe43 | 2785 | if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) { |
61007b31 SH |
2786 | return 0; |
2787 | } | |
2788 | ||
3482b9bc EB |
2789 | /* Discard is advisory, but some devices track and coalesce |
2790 | * unaligned requests, so we must pass everything down rather than | |
2791 | * round here. Still, most devices will just silently ignore | |
2792 | * unaligned requests (by returning -ENOTSUP), so we must fragment | |
2793 | * the request accordingly. */ | |
02aefe43 | 2794 | align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment); |
b8d0a980 EB |
2795 | assert(align % bs->bl.request_alignment == 0); |
2796 | head = offset % align; | |
f5a5ca79 | 2797 | tail = (offset + bytes) % align; |
9f1963b3 | 2798 | |
99723548 | 2799 | bdrv_inc_in_flight(bs); |
f5a5ca79 | 2800 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD); |
50824995 | 2801 | |
00695c27 | 2802 | ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0); |
ec050f77 DL |
2803 | if (ret < 0) { |
2804 | goto out; | |
2805 | } | |
2806 | ||
9f1963b3 EB |
2807 | max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX), |
2808 | align); | |
3482b9bc | 2809 | assert(max_pdiscard >= bs->bl.request_alignment); |
61007b31 | 2810 | |
f5a5ca79 | 2811 | while (bytes > 0) { |
d93e5726 | 2812 | int64_t num = bytes; |
3482b9bc EB |
2813 | |
2814 | if (head) { | |
2815 | /* Make small requests to get to alignment boundaries. */ | |
f5a5ca79 | 2816 | num = MIN(bytes, align - head); |
3482b9bc EB |
2817 | if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) { |
2818 | num %= bs->bl.request_alignment; | |
2819 | } | |
2820 | head = (head + num) % align; | |
2821 | assert(num < max_pdiscard); | |
2822 | } else if (tail) { | |
2823 | if (num > align) { | |
2824 | /* Shorten the request to the last aligned cluster. */ | |
2825 | num -= tail; | |
2826 | } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) && | |
2827 | tail > bs->bl.request_alignment) { | |
2828 | tail %= bs->bl.request_alignment; | |
2829 | num -= tail; | |
2830 | } | |
2831 | } | |
2832 | /* limit request size */ | |
2833 | if (num > max_pdiscard) { | |
2834 | num = max_pdiscard; | |
2835 | } | |
61007b31 | 2836 | |
d470ad42 HR |
2837 | if (!bs->drv) { |
2838 | ret = -ENOMEDIUM; | |
2839 | goto out; | |
2840 | } | |
47a5486d EB |
2841 | if (bs->drv->bdrv_co_pdiscard) { |
2842 | ret = bs->drv->bdrv_co_pdiscard(bs, offset, num); | |
61007b31 SH |
2843 | } else { |
2844 | BlockAIOCB *acb; | |
2845 | CoroutineIOCompletion co = { | |
2846 | .coroutine = qemu_coroutine_self(), | |
2847 | }; | |
2848 | ||
4da444a0 EB |
2849 | acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num, |
2850 | bdrv_co_io_em_complete, &co); | |
61007b31 | 2851 | if (acb == NULL) { |
b1066c87 FZ |
2852 | ret = -EIO; |
2853 | goto out; | |
61007b31 SH |
2854 | } else { |
2855 | qemu_coroutine_yield(); | |
2856 | ret = co.ret; | |
2857 | } | |
2858 | } | |
2859 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 2860 | goto out; |
61007b31 SH |
2861 | } |
2862 | ||
9f1963b3 | 2863 | offset += num; |
f5a5ca79 | 2864 | bytes -= num; |
61007b31 | 2865 | } |
b1066c87 FZ |
2866 | ret = 0; |
2867 | out: | |
00695c27 | 2868 | bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret); |
b1066c87 | 2869 | tracked_request_end(&req); |
99723548 | 2870 | bdrv_dec_in_flight(bs); |
b1066c87 | 2871 | return ret; |
61007b31 SH |
2872 | } |
2873 | ||
d93e5726 | 2874 | int bdrv_pdiscard(BdrvChild *child, int64_t offset, int64_t bytes) |
61007b31 SH |
2875 | { |
2876 | Coroutine *co; | |
2877 | DiscardCo rwco = { | |
0b9fd3f4 | 2878 | .child = child, |
0c51a893 | 2879 | .offset = offset, |
f5a5ca79 | 2880 | .bytes = bytes, |
61007b31 SH |
2881 | .ret = NOT_DONE, |
2882 | }; | |
2883 | ||
2884 | if (qemu_in_coroutine()) { | |
2885 | /* Fast-path if already in coroutine context */ | |
0c51a893 | 2886 | bdrv_pdiscard_co_entry(&rwco); |
61007b31 | 2887 | } else { |
0c51a893 | 2888 | co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco); |
0b9fd3f4 FZ |
2889 | bdrv_coroutine_enter(child->bs, co); |
2890 | BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE); | |
61007b31 SH |
2891 | } |
2892 | ||
2893 | return rwco.ret; | |
2894 | } | |
2895 | ||
48af776a | 2896 | int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
2897 | { |
2898 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
2899 | CoroutineIOCompletion co = { |
2900 | .coroutine = qemu_coroutine_self(), | |
2901 | }; | |
2902 | BlockAIOCB *acb; | |
61007b31 | 2903 | |
99723548 | 2904 | bdrv_inc_in_flight(bs); |
16a389dc | 2905 | if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) { |
5c5ae76a FZ |
2906 | co.ret = -ENOTSUP; |
2907 | goto out; | |
2908 | } | |
2909 | ||
16a389dc KW |
2910 | if (drv->bdrv_co_ioctl) { |
2911 | co.ret = drv->bdrv_co_ioctl(bs, req, buf); | |
2912 | } else { | |
2913 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
2914 | if (!acb) { | |
2915 | co.ret = -ENOTSUP; | |
2916 | goto out; | |
2917 | } | |
2918 | qemu_coroutine_yield(); | |
5c5ae76a | 2919 | } |
5c5ae76a | 2920 | out: |
99723548 | 2921 | bdrv_dec_in_flight(bs); |
5c5ae76a FZ |
2922 | return co.ret; |
2923 | } | |
2924 | ||
61007b31 SH |
2925 | void *qemu_blockalign(BlockDriverState *bs, size_t size) |
2926 | { | |
2927 | return qemu_memalign(bdrv_opt_mem_align(bs), size); | |
2928 | } | |
2929 | ||
2930 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
2931 | { | |
2932 | return memset(qemu_blockalign(bs, size), 0, size); | |
2933 | } | |
2934 | ||
2935 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
2936 | { | |
2937 | size_t align = bdrv_opt_mem_align(bs); | |
2938 | ||
2939 | /* Ensure that NULL is never returned on success */ | |
2940 | assert(align > 0); | |
2941 | if (size == 0) { | |
2942 | size = align; | |
2943 | } | |
2944 | ||
2945 | return qemu_try_memalign(align, size); | |
2946 | } | |
2947 | ||
2948 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
2949 | { | |
2950 | void *mem = qemu_try_blockalign(bs, size); | |
2951 | ||
2952 | if (mem) { | |
2953 | memset(mem, 0, size); | |
2954 | } | |
2955 | ||
2956 | return mem; | |
2957 | } | |
2958 | ||
2959 | /* | |
2960 | * Check if all memory in this vector is sector aligned. | |
2961 | */ | |
2962 | bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
2963 | { | |
2964 | int i; | |
4196d2f0 | 2965 | size_t alignment = bdrv_min_mem_align(bs); |
61007b31 SH |
2966 | |
2967 | for (i = 0; i < qiov->niov; i++) { | |
2968 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
2969 | return false; | |
2970 | } | |
2971 | if (qiov->iov[i].iov_len % alignment) { | |
2972 | return false; | |
2973 | } | |
2974 | } | |
2975 | ||
2976 | return true; | |
2977 | } | |
2978 | ||
2979 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
2980 | NotifierWithReturn *notifier) | |
2981 | { | |
2982 | notifier_with_return_list_add(&bs->before_write_notifiers, notifier); | |
2983 | } | |
2984 | ||
2985 | void bdrv_io_plug(BlockDriverState *bs) | |
2986 | { | |
6b98bd64 PB |
2987 | BdrvChild *child; |
2988 | ||
2989 | QLIST_FOREACH(child, &bs->children, next) { | |
2990 | bdrv_io_plug(child->bs); | |
2991 | } | |
2992 | ||
850d54a2 | 2993 | if (atomic_fetch_inc(&bs->io_plugged) == 0) { |
6b98bd64 PB |
2994 | BlockDriver *drv = bs->drv; |
2995 | if (drv && drv->bdrv_io_plug) { | |
2996 | drv->bdrv_io_plug(bs); | |
2997 | } | |
61007b31 SH |
2998 | } |
2999 | } | |
3000 | ||
3001 | void bdrv_io_unplug(BlockDriverState *bs) | |
3002 | { | |
6b98bd64 PB |
3003 | BdrvChild *child; |
3004 | ||
3005 | assert(bs->io_plugged); | |
850d54a2 | 3006 | if (atomic_fetch_dec(&bs->io_plugged) == 1) { |
6b98bd64 PB |
3007 | BlockDriver *drv = bs->drv; |
3008 | if (drv && drv->bdrv_io_unplug) { | |
3009 | drv->bdrv_io_unplug(bs); | |
3010 | } | |
3011 | } | |
3012 | ||
3013 | QLIST_FOREACH(child, &bs->children, next) { | |
3014 | bdrv_io_unplug(child->bs); | |
61007b31 SH |
3015 | } |
3016 | } | |
23d0ba93 FZ |
3017 | |
3018 | void bdrv_register_buf(BlockDriverState *bs, void *host, size_t size) | |
3019 | { | |
3020 | BdrvChild *child; | |
3021 | ||
3022 | if (bs->drv && bs->drv->bdrv_register_buf) { | |
3023 | bs->drv->bdrv_register_buf(bs, host, size); | |
3024 | } | |
3025 | QLIST_FOREACH(child, &bs->children, next) { | |
3026 | bdrv_register_buf(child->bs, host, size); | |
3027 | } | |
3028 | } | |
3029 | ||
3030 | void bdrv_unregister_buf(BlockDriverState *bs, void *host) | |
3031 | { | |
3032 | BdrvChild *child; | |
3033 | ||
3034 | if (bs->drv && bs->drv->bdrv_unregister_buf) { | |
3035 | bs->drv->bdrv_unregister_buf(bs, host); | |
3036 | } | |
3037 | QLIST_FOREACH(child, &bs->children, next) { | |
3038 | bdrv_unregister_buf(child->bs, host); | |
3039 | } | |
3040 | } | |
fcc67678 | 3041 | |
67b51fb9 VSO |
3042 | static int coroutine_fn bdrv_co_copy_range_internal( |
3043 | BdrvChild *src, uint64_t src_offset, BdrvChild *dst, | |
3044 | uint64_t dst_offset, uint64_t bytes, | |
3045 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags, | |
3046 | bool recurse_src) | |
fcc67678 | 3047 | { |
999658a0 | 3048 | BdrvTrackedRequest req; |
fcc67678 FZ |
3049 | int ret; |
3050 | ||
fe0480d6 KW |
3051 | /* TODO We can support BDRV_REQ_NO_FALLBACK here */ |
3052 | assert(!(read_flags & BDRV_REQ_NO_FALLBACK)); | |
3053 | assert(!(write_flags & BDRV_REQ_NO_FALLBACK)); | |
3054 | ||
d4d3e5a0 | 3055 | if (!dst || !dst->bs) { |
fcc67678 FZ |
3056 | return -ENOMEDIUM; |
3057 | } | |
fcc67678 FZ |
3058 | ret = bdrv_check_byte_request(dst->bs, dst_offset, bytes); |
3059 | if (ret) { | |
3060 | return ret; | |
3061 | } | |
67b51fb9 VSO |
3062 | if (write_flags & BDRV_REQ_ZERO_WRITE) { |
3063 | return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags); | |
fcc67678 FZ |
3064 | } |
3065 | ||
d4d3e5a0 FZ |
3066 | if (!src || !src->bs) { |
3067 | return -ENOMEDIUM; | |
3068 | } | |
3069 | ret = bdrv_check_byte_request(src->bs, src_offset, bytes); | |
3070 | if (ret) { | |
3071 | return ret; | |
3072 | } | |
3073 | ||
fcc67678 FZ |
3074 | if (!src->bs->drv->bdrv_co_copy_range_from |
3075 | || !dst->bs->drv->bdrv_co_copy_range_to | |
3076 | || src->bs->encrypted || dst->bs->encrypted) { | |
3077 | return -ENOTSUP; | |
3078 | } | |
37aec7d7 | 3079 | |
fcc67678 | 3080 | if (recurse_src) { |
999658a0 VSO |
3081 | bdrv_inc_in_flight(src->bs); |
3082 | tracked_request_begin(&req, src->bs, src_offset, bytes, | |
3083 | BDRV_TRACKED_READ); | |
3084 | ||
09d2f948 VSO |
3085 | /* BDRV_REQ_SERIALISING is only for write operation */ |
3086 | assert(!(read_flags & BDRV_REQ_SERIALISING)); | |
67b51fb9 | 3087 | if (!(read_flags & BDRV_REQ_NO_SERIALISING)) { |
999658a0 VSO |
3088 | wait_serialising_requests(&req); |
3089 | } | |
3090 | ||
37aec7d7 FZ |
3091 | ret = src->bs->drv->bdrv_co_copy_range_from(src->bs, |
3092 | src, src_offset, | |
3093 | dst, dst_offset, | |
67b51fb9 VSO |
3094 | bytes, |
3095 | read_flags, write_flags); | |
999658a0 VSO |
3096 | |
3097 | tracked_request_end(&req); | |
3098 | bdrv_dec_in_flight(src->bs); | |
fcc67678 | 3099 | } else { |
999658a0 VSO |
3100 | bdrv_inc_in_flight(dst->bs); |
3101 | tracked_request_begin(&req, dst->bs, dst_offset, bytes, | |
3102 | BDRV_TRACKED_WRITE); | |
0eb1e891 FZ |
3103 | ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req, |
3104 | write_flags); | |
3105 | if (!ret) { | |
3106 | ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs, | |
3107 | src, src_offset, | |
3108 | dst, dst_offset, | |
3109 | bytes, | |
3110 | read_flags, write_flags); | |
3111 | } | |
3112 | bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret); | |
999658a0 VSO |
3113 | tracked_request_end(&req); |
3114 | bdrv_dec_in_flight(dst->bs); | |
fcc67678 | 3115 | } |
999658a0 | 3116 | |
37aec7d7 | 3117 | return ret; |
fcc67678 FZ |
3118 | } |
3119 | ||
3120 | /* Copy range from @src to @dst. | |
3121 | * | |
3122 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3123 | * semantics. */ | |
3124 | int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, uint64_t src_offset, | |
3125 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3126 | uint64_t bytes, |
3127 | BdrvRequestFlags read_flags, | |
3128 | BdrvRequestFlags write_flags) | |
fcc67678 | 3129 | { |
ecc983a5 FZ |
3130 | trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes, |
3131 | read_flags, write_flags); | |
fcc67678 | 3132 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3133 | bytes, read_flags, write_flags, true); |
fcc67678 FZ |
3134 | } |
3135 | ||
3136 | /* Copy range from @src to @dst. | |
3137 | * | |
3138 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3139 | * semantics. */ | |
3140 | int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, uint64_t src_offset, | |
3141 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3142 | uint64_t bytes, |
3143 | BdrvRequestFlags read_flags, | |
3144 | BdrvRequestFlags write_flags) | |
fcc67678 | 3145 | { |
ecc983a5 FZ |
3146 | trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3147 | read_flags, write_flags); | |
fcc67678 | 3148 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3149 | bytes, read_flags, write_flags, false); |
fcc67678 FZ |
3150 | } |
3151 | ||
3152 | int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset, | |
3153 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3154 | uint64_t bytes, BdrvRequestFlags read_flags, |
3155 | BdrvRequestFlags write_flags) | |
fcc67678 | 3156 | { |
37aec7d7 FZ |
3157 | return bdrv_co_copy_range_from(src, src_offset, |
3158 | dst, dst_offset, | |
67b51fb9 | 3159 | bytes, read_flags, write_flags); |
fcc67678 | 3160 | } |
3d9f2d2a KW |
3161 | |
3162 | static void bdrv_parent_cb_resize(BlockDriverState *bs) | |
3163 | { | |
3164 | BdrvChild *c; | |
3165 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
3166 | if (c->role->resize) { | |
3167 | c->role->resize(c); | |
3168 | } | |
3169 | } | |
3170 | } | |
3171 | ||
3172 | /** | |
3173 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
3174 | */ | |
3175 | int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, | |
3176 | PreallocMode prealloc, Error **errp) | |
3177 | { | |
3178 | BlockDriverState *bs = child->bs; | |
3179 | BlockDriver *drv = bs->drv; | |
1bc5f09f KW |
3180 | BdrvTrackedRequest req; |
3181 | int64_t old_size, new_bytes; | |
3d9f2d2a KW |
3182 | int ret; |
3183 | ||
3d9f2d2a KW |
3184 | |
3185 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ | |
3186 | if (!drv) { | |
3187 | error_setg(errp, "No medium inserted"); | |
3188 | return -ENOMEDIUM; | |
3189 | } | |
3190 | if (offset < 0) { | |
3191 | error_setg(errp, "Image size cannot be negative"); | |
3192 | return -EINVAL; | |
3193 | } | |
3194 | ||
1bc5f09f KW |
3195 | old_size = bdrv_getlength(bs); |
3196 | if (old_size < 0) { | |
3197 | error_setg_errno(errp, -old_size, "Failed to get old image size"); | |
3198 | return old_size; | |
3199 | } | |
3200 | ||
3201 | if (offset > old_size) { | |
3202 | new_bytes = offset - old_size; | |
3203 | } else { | |
3204 | new_bytes = 0; | |
3205 | } | |
3206 | ||
3d9f2d2a | 3207 | bdrv_inc_in_flight(bs); |
5416a11e FZ |
3208 | tracked_request_begin(&req, bs, offset - new_bytes, new_bytes, |
3209 | BDRV_TRACKED_TRUNCATE); | |
1bc5f09f KW |
3210 | |
3211 | /* If we are growing the image and potentially using preallocation for the | |
3212 | * new area, we need to make sure that no write requests are made to it | |
3213 | * concurrently or they might be overwritten by preallocation. */ | |
3214 | if (new_bytes) { | |
3215 | mark_request_serialising(&req, 1); | |
cd47d792 FZ |
3216 | } |
3217 | if (bs->read_only) { | |
3218 | error_setg(errp, "Image is read-only"); | |
3219 | ret = -EACCES; | |
3220 | goto out; | |
3221 | } | |
3222 | ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req, | |
3223 | 0); | |
3224 | if (ret < 0) { | |
3225 | error_setg_errno(errp, -ret, | |
3226 | "Failed to prepare request for truncation"); | |
3227 | goto out; | |
1bc5f09f | 3228 | } |
3d9f2d2a KW |
3229 | |
3230 | if (!drv->bdrv_co_truncate) { | |
3231 | if (bs->file && drv->is_filter) { | |
3232 | ret = bdrv_co_truncate(bs->file, offset, prealloc, errp); | |
3233 | goto out; | |
3234 | } | |
3235 | error_setg(errp, "Image format driver does not support resize"); | |
3236 | ret = -ENOTSUP; | |
3237 | goto out; | |
3238 | } | |
3d9f2d2a KW |
3239 | |
3240 | ret = drv->bdrv_co_truncate(bs, offset, prealloc, errp); | |
3241 | if (ret < 0) { | |
3242 | goto out; | |
3243 | } | |
3244 | ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); | |
3245 | if (ret < 0) { | |
3246 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
3247 | } else { | |
3248 | offset = bs->total_sectors * BDRV_SECTOR_SIZE; | |
3249 | } | |
cd47d792 FZ |
3250 | /* It's possible that truncation succeeded but refresh_total_sectors |
3251 | * failed, but the latter doesn't affect how we should finish the request. | |
3252 | * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. */ | |
3253 | bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0); | |
3d9f2d2a KW |
3254 | |
3255 | out: | |
1bc5f09f | 3256 | tracked_request_end(&req); |
3d9f2d2a | 3257 | bdrv_dec_in_flight(bs); |
1bc5f09f | 3258 | |
3d9f2d2a KW |
3259 | return ret; |
3260 | } | |
3261 | ||
3262 | typedef struct TruncateCo { | |
3263 | BdrvChild *child; | |
3264 | int64_t offset; | |
3265 | PreallocMode prealloc; | |
3266 | Error **errp; | |
3267 | int ret; | |
3268 | } TruncateCo; | |
3269 | ||
3270 | static void coroutine_fn bdrv_truncate_co_entry(void *opaque) | |
3271 | { | |
3272 | TruncateCo *tco = opaque; | |
3273 | tco->ret = bdrv_co_truncate(tco->child, tco->offset, tco->prealloc, | |
3274 | tco->errp); | |
4720cbee | 3275 | aio_wait_kick(); |
3d9f2d2a KW |
3276 | } |
3277 | ||
3278 | int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc, | |
3279 | Error **errp) | |
3280 | { | |
3281 | Coroutine *co; | |
3282 | TruncateCo tco = { | |
3283 | .child = child, | |
3284 | .offset = offset, | |
3285 | .prealloc = prealloc, | |
3286 | .errp = errp, | |
3287 | .ret = NOT_DONE, | |
3288 | }; | |
3289 | ||
3290 | if (qemu_in_coroutine()) { | |
3291 | /* Fast-path if already in coroutine context */ | |
3292 | bdrv_truncate_co_entry(&tco); | |
3293 | } else { | |
3294 | co = qemu_coroutine_create(bdrv_truncate_co_entry, &tco); | |
4720cbee | 3295 | bdrv_coroutine_enter(child->bs, co); |
3d9f2d2a KW |
3296 | BDRV_POLL_WHILE(child->bs, tco.ret == NOT_DONE); |
3297 | } | |
3298 | ||
3299 | return tco.ret; | |
3300 | } |