]>
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" |
61007b31 SH |
35 | |
36 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
37 | ||
cb2e2878 EB |
38 | /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */ |
39 | #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS) | |
40 | ||
7f8f03ef | 41 | static void bdrv_parent_cb_resize(BlockDriverState *bs); |
d05aa8bb | 42 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 43 | int64_t offset, int bytes, BdrvRequestFlags flags); |
61007b31 | 44 | |
6cd5c9d7 KW |
45 | void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore, |
46 | bool ignore_bds_parents) | |
61007b31 | 47 | { |
02d21300 | 48 | BdrvChild *c, *next; |
27ccdd52 | 49 | |
02d21300 | 50 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { |
6cd5c9d7 | 51 | if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { |
0152bf40 KW |
52 | continue; |
53 | } | |
4be6a6d1 | 54 | bdrv_parent_drained_begin_single(c, false); |
ce0f1412 PB |
55 | } |
56 | } | |
61007b31 | 57 | |
6cd5c9d7 KW |
58 | void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore, |
59 | bool ignore_bds_parents) | |
ce0f1412 | 60 | { |
02d21300 | 61 | BdrvChild *c, *next; |
27ccdd52 | 62 | |
02d21300 | 63 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { |
6cd5c9d7 | 64 | if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { |
0152bf40 KW |
65 | continue; |
66 | } | |
c2066af0 KW |
67 | if (c->role->drained_end) { |
68 | c->role->drained_end(c); | |
69 | } | |
27ccdd52 | 70 | } |
61007b31 SH |
71 | } |
72 | ||
4be6a6d1 KW |
73 | static bool bdrv_parent_drained_poll_single(BdrvChild *c) |
74 | { | |
75 | if (c->role->drained_poll) { | |
76 | return c->role->drained_poll(c); | |
77 | } | |
78 | return false; | |
79 | } | |
80 | ||
6cd5c9d7 KW |
81 | static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore, |
82 | bool ignore_bds_parents) | |
89bd0305 KW |
83 | { |
84 | BdrvChild *c, *next; | |
85 | bool busy = false; | |
86 | ||
87 | QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) { | |
6cd5c9d7 | 88 | if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) { |
89bd0305 KW |
89 | continue; |
90 | } | |
4be6a6d1 | 91 | busy |= bdrv_parent_drained_poll_single(c); |
89bd0305 KW |
92 | } |
93 | ||
94 | return busy; | |
95 | } | |
96 | ||
4be6a6d1 KW |
97 | void bdrv_parent_drained_begin_single(BdrvChild *c, bool poll) |
98 | { | |
99 | if (c->role->drained_begin) { | |
100 | c->role->drained_begin(c); | |
101 | } | |
102 | if (poll) { | |
103 | BDRV_POLL_WHILE(c->bs, bdrv_parent_drained_poll_single(c)); | |
104 | } | |
105 | } | |
106 | ||
d9e0dfa2 EB |
107 | static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src) |
108 | { | |
109 | dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer); | |
110 | dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer); | |
111 | dst->opt_mem_alignment = MAX(dst->opt_mem_alignment, | |
112 | src->opt_mem_alignment); | |
113 | dst->min_mem_alignment = MAX(dst->min_mem_alignment, | |
114 | src->min_mem_alignment); | |
115 | dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov); | |
116 | } | |
117 | ||
61007b31 SH |
118 | void bdrv_refresh_limits(BlockDriverState *bs, Error **errp) |
119 | { | |
120 | BlockDriver *drv = bs->drv; | |
121 | Error *local_err = NULL; | |
122 | ||
123 | memset(&bs->bl, 0, sizeof(bs->bl)); | |
124 | ||
125 | if (!drv) { | |
126 | return; | |
127 | } | |
128 | ||
79ba8c98 | 129 | /* Default alignment based on whether driver has byte interface */ |
e31f6864 EB |
130 | bs->bl.request_alignment = (drv->bdrv_co_preadv || |
131 | drv->bdrv_aio_preadv) ? 1 : 512; | |
79ba8c98 | 132 | |
61007b31 SH |
133 | /* Take some limits from the children as a default */ |
134 | if (bs->file) { | |
9a4f4c31 | 135 | bdrv_refresh_limits(bs->file->bs, &local_err); |
61007b31 SH |
136 | if (local_err) { |
137 | error_propagate(errp, local_err); | |
138 | return; | |
139 | } | |
d9e0dfa2 | 140 | bdrv_merge_limits(&bs->bl, &bs->file->bs->bl); |
61007b31 | 141 | } else { |
4196d2f0 | 142 | bs->bl.min_mem_alignment = 512; |
459b4e66 | 143 | bs->bl.opt_mem_alignment = getpagesize(); |
bd44feb7 SH |
144 | |
145 | /* Safe default since most protocols use readv()/writev()/etc */ | |
146 | bs->bl.max_iov = IOV_MAX; | |
61007b31 SH |
147 | } |
148 | ||
760e0063 KW |
149 | if (bs->backing) { |
150 | bdrv_refresh_limits(bs->backing->bs, &local_err); | |
61007b31 SH |
151 | if (local_err) { |
152 | error_propagate(errp, local_err); | |
153 | return; | |
154 | } | |
d9e0dfa2 | 155 | bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl); |
61007b31 SH |
156 | } |
157 | ||
158 | /* Then let the driver override it */ | |
159 | if (drv->bdrv_refresh_limits) { | |
160 | drv->bdrv_refresh_limits(bs, errp); | |
161 | } | |
162 | } | |
163 | ||
164 | /** | |
165 | * The copy-on-read flag is actually a reference count so multiple users may | |
166 | * use the feature without worrying about clobbering its previous state. | |
167 | * Copy-on-read stays enabled until all users have called to disable it. | |
168 | */ | |
169 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
170 | { | |
d3faa13e | 171 | atomic_inc(&bs->copy_on_read); |
61007b31 SH |
172 | } |
173 | ||
174 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
175 | { | |
d3faa13e PB |
176 | int old = atomic_fetch_dec(&bs->copy_on_read); |
177 | assert(old >= 1); | |
61007b31 SH |
178 | } |
179 | ||
61124f03 PB |
180 | typedef struct { |
181 | Coroutine *co; | |
182 | BlockDriverState *bs; | |
183 | bool done; | |
481cad48 | 184 | bool begin; |
b0165585 | 185 | bool recursive; |
fe4f0614 | 186 | bool poll; |
0152bf40 | 187 | BdrvChild *parent; |
6cd5c9d7 | 188 | bool ignore_bds_parents; |
61124f03 PB |
189 | } BdrvCoDrainData; |
190 | ||
191 | static void coroutine_fn bdrv_drain_invoke_entry(void *opaque) | |
192 | { | |
193 | BdrvCoDrainData *data = opaque; | |
194 | BlockDriverState *bs = data->bs; | |
195 | ||
481cad48 | 196 | if (data->begin) { |
f8ea8dac | 197 | bs->drv->bdrv_co_drain_begin(bs); |
481cad48 MP |
198 | } else { |
199 | bs->drv->bdrv_co_drain_end(bs); | |
200 | } | |
61124f03 PB |
201 | |
202 | /* Set data->done before reading bs->wakeup. */ | |
203 | atomic_mb_set(&data->done, true); | |
0109e7e6 KW |
204 | bdrv_dec_in_flight(bs); |
205 | ||
206 | if (data->begin) { | |
207 | g_free(data); | |
208 | } | |
61124f03 PB |
209 | } |
210 | ||
db0289b9 | 211 | /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */ |
7d40d9ef | 212 | static void bdrv_drain_invoke(BlockDriverState *bs, bool begin) |
61124f03 | 213 | { |
0109e7e6 | 214 | BdrvCoDrainData *data; |
61124f03 | 215 | |
f8ea8dac | 216 | if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) || |
481cad48 | 217 | (!begin && !bs->drv->bdrv_co_drain_end)) { |
61124f03 PB |
218 | return; |
219 | } | |
220 | ||
0109e7e6 KW |
221 | data = g_new(BdrvCoDrainData, 1); |
222 | *data = (BdrvCoDrainData) { | |
223 | .bs = bs, | |
224 | .done = false, | |
225 | .begin = begin | |
226 | }; | |
227 | ||
228 | /* Make sure the driver callback completes during the polling phase for | |
229 | * drain_begin. */ | |
230 | bdrv_inc_in_flight(bs); | |
231 | data->co = qemu_coroutine_create(bdrv_drain_invoke_entry, data); | |
232 | aio_co_schedule(bdrv_get_aio_context(bs), data->co); | |
233 | ||
234 | if (!begin) { | |
235 | BDRV_POLL_WHILE(bs, !data->done); | |
236 | g_free(data); | |
237 | } | |
61124f03 PB |
238 | } |
239 | ||
1cc8e54a | 240 | /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */ |
fe4f0614 | 241 | bool bdrv_drain_poll(BlockDriverState *bs, bool recursive, |
6cd5c9d7 | 242 | BdrvChild *ignore_parent, bool ignore_bds_parents) |
89bd0305 | 243 | { |
fe4f0614 KW |
244 | BdrvChild *child, *next; |
245 | ||
6cd5c9d7 | 246 | if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) { |
89bd0305 KW |
247 | return true; |
248 | } | |
249 | ||
fe4f0614 KW |
250 | if (atomic_read(&bs->in_flight)) { |
251 | return true; | |
252 | } | |
253 | ||
254 | if (recursive) { | |
6cd5c9d7 | 255 | assert(!ignore_bds_parents); |
fe4f0614 | 256 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
6cd5c9d7 | 257 | if (bdrv_drain_poll(child->bs, recursive, child, false)) { |
fe4f0614 KW |
258 | return true; |
259 | } | |
260 | } | |
261 | } | |
262 | ||
263 | return false; | |
89bd0305 KW |
264 | } |
265 | ||
fe4f0614 | 266 | static bool bdrv_drain_poll_top_level(BlockDriverState *bs, bool recursive, |
89bd0305 | 267 | BdrvChild *ignore_parent) |
1cc8e54a | 268 | { |
6cd5c9d7 | 269 | return bdrv_drain_poll(bs, recursive, ignore_parent, false); |
1cc8e54a KW |
270 | } |
271 | ||
b0165585 | 272 | static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, |
6cd5c9d7 KW |
273 | BdrvChild *parent, bool ignore_bds_parents, |
274 | bool poll); | |
b0165585 | 275 | static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, |
6cd5c9d7 | 276 | BdrvChild *parent, bool ignore_bds_parents); |
0152bf40 | 277 | |
a77fd4bb FZ |
278 | static void bdrv_co_drain_bh_cb(void *opaque) |
279 | { | |
280 | BdrvCoDrainData *data = opaque; | |
281 | Coroutine *co = data->co; | |
99723548 | 282 | BlockDriverState *bs = data->bs; |
a77fd4bb | 283 | |
c8ca33d0 | 284 | if (bs) { |
aa1361d5 KW |
285 | AioContext *ctx = bdrv_get_aio_context(bs); |
286 | AioContext *co_ctx = qemu_coroutine_get_aio_context(co); | |
287 | ||
288 | /* | |
289 | * When the coroutine yielded, the lock for its home context was | |
290 | * released, so we need to re-acquire it here. If it explicitly | |
291 | * acquired a different context, the lock is still held and we don't | |
292 | * want to lock it a second time (or AIO_WAIT_WHILE() would hang). | |
293 | */ | |
294 | if (ctx == co_ctx) { | |
295 | aio_context_acquire(ctx); | |
296 | } | |
c8ca33d0 KW |
297 | bdrv_dec_in_flight(bs); |
298 | if (data->begin) { | |
6cd5c9d7 KW |
299 | bdrv_do_drained_begin(bs, data->recursive, data->parent, |
300 | data->ignore_bds_parents, data->poll); | |
c8ca33d0 | 301 | } else { |
6cd5c9d7 KW |
302 | bdrv_do_drained_end(bs, data->recursive, data->parent, |
303 | data->ignore_bds_parents); | |
c8ca33d0 | 304 | } |
aa1361d5 KW |
305 | if (ctx == co_ctx) { |
306 | aio_context_release(ctx); | |
307 | } | |
481cad48 | 308 | } else { |
c8ca33d0 KW |
309 | assert(data->begin); |
310 | bdrv_drain_all_begin(); | |
481cad48 MP |
311 | } |
312 | ||
a77fd4bb | 313 | data->done = true; |
1919631e | 314 | aio_co_wake(co); |
a77fd4bb FZ |
315 | } |
316 | ||
481cad48 | 317 | static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs, |
b0165585 | 318 | bool begin, bool recursive, |
6cd5c9d7 KW |
319 | BdrvChild *parent, |
320 | bool ignore_bds_parents, | |
321 | bool poll) | |
a77fd4bb FZ |
322 | { |
323 | BdrvCoDrainData data; | |
324 | ||
325 | /* Calling bdrv_drain() from a BH ensures the current coroutine yields and | |
c40a2545 | 326 | * other coroutines run if they were queued by aio_co_enter(). */ |
a77fd4bb FZ |
327 | |
328 | assert(qemu_in_coroutine()); | |
329 | data = (BdrvCoDrainData) { | |
330 | .co = qemu_coroutine_self(), | |
331 | .bs = bs, | |
332 | .done = false, | |
481cad48 | 333 | .begin = begin, |
b0165585 | 334 | .recursive = recursive, |
0152bf40 | 335 | .parent = parent, |
6cd5c9d7 | 336 | .ignore_bds_parents = ignore_bds_parents, |
fe4f0614 | 337 | .poll = poll, |
a77fd4bb | 338 | }; |
c8ca33d0 KW |
339 | if (bs) { |
340 | bdrv_inc_in_flight(bs); | |
341 | } | |
fffb6e12 PB |
342 | aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), |
343 | bdrv_co_drain_bh_cb, &data); | |
a77fd4bb FZ |
344 | |
345 | qemu_coroutine_yield(); | |
346 | /* If we are resumed from some other event (such as an aio completion or a | |
347 | * timer callback), it is a bug in the caller that should be fixed. */ | |
348 | assert(data.done); | |
349 | } | |
350 | ||
dcf94a23 | 351 | void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, |
6cd5c9d7 | 352 | BdrvChild *parent, bool ignore_bds_parents) |
6820643f | 353 | { |
dcf94a23 | 354 | assert(!qemu_in_coroutine()); |
d42cf288 | 355 | |
60369b86 | 356 | /* Stop things in parent-to-child order */ |
414c2ec3 | 357 | if (atomic_fetch_inc(&bs->quiesce_counter) == 0) { |
6820643f | 358 | aio_disable_external(bdrv_get_aio_context(bs)); |
6820643f KW |
359 | } |
360 | ||
6cd5c9d7 | 361 | bdrv_parent_drained_begin(bs, parent, ignore_bds_parents); |
7d40d9ef | 362 | bdrv_drain_invoke(bs, true); |
dcf94a23 KW |
363 | } |
364 | ||
365 | static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive, | |
6cd5c9d7 KW |
366 | BdrvChild *parent, bool ignore_bds_parents, |
367 | bool poll) | |
dcf94a23 KW |
368 | { |
369 | BdrvChild *child, *next; | |
370 | ||
371 | if (qemu_in_coroutine()) { | |
6cd5c9d7 KW |
372 | bdrv_co_yield_to_drain(bs, true, recursive, parent, ignore_bds_parents, |
373 | poll); | |
dcf94a23 KW |
374 | return; |
375 | } | |
376 | ||
6cd5c9d7 | 377 | bdrv_do_drained_begin_quiesce(bs, parent, ignore_bds_parents); |
d30b8e64 | 378 | |
b0165585 | 379 | if (recursive) { |
6cd5c9d7 | 380 | assert(!ignore_bds_parents); |
d736f119 | 381 | bs->recursive_quiesce_counter++; |
b0165585 | 382 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
6cd5c9d7 KW |
383 | bdrv_do_drained_begin(child->bs, true, child, ignore_bds_parents, |
384 | false); | |
b0165585 KW |
385 | } |
386 | } | |
fe4f0614 KW |
387 | |
388 | /* | |
389 | * Wait for drained requests to finish. | |
390 | * | |
391 | * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The | |
392 | * call is needed so things in this AioContext can make progress even | |
393 | * though we don't return to the main AioContext loop - this automatically | |
394 | * includes other nodes in the same AioContext and therefore all child | |
395 | * nodes. | |
396 | */ | |
397 | if (poll) { | |
6cd5c9d7 | 398 | assert(!ignore_bds_parents); |
fe4f0614 KW |
399 | BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, recursive, parent)); |
400 | } | |
6820643f KW |
401 | } |
402 | ||
0152bf40 KW |
403 | void bdrv_drained_begin(BlockDriverState *bs) |
404 | { | |
6cd5c9d7 | 405 | bdrv_do_drained_begin(bs, false, NULL, false, true); |
b0165585 KW |
406 | } |
407 | ||
408 | void bdrv_subtree_drained_begin(BlockDriverState *bs) | |
409 | { | |
6cd5c9d7 | 410 | bdrv_do_drained_begin(bs, true, NULL, false, true); |
0152bf40 KW |
411 | } |
412 | ||
6cd5c9d7 KW |
413 | static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive, |
414 | BdrvChild *parent, bool ignore_bds_parents) | |
6820643f | 415 | { |
b0165585 | 416 | BdrvChild *child, *next; |
0f115168 KW |
417 | int old_quiesce_counter; |
418 | ||
481cad48 | 419 | if (qemu_in_coroutine()) { |
6cd5c9d7 KW |
420 | bdrv_co_yield_to_drain(bs, false, recursive, parent, ignore_bds_parents, |
421 | false); | |
481cad48 MP |
422 | return; |
423 | } | |
6820643f | 424 | assert(bs->quiesce_counter > 0); |
0f115168 | 425 | old_quiesce_counter = atomic_fetch_dec(&bs->quiesce_counter); |
6820643f | 426 | |
60369b86 | 427 | /* Re-enable things in child-to-parent order */ |
7d40d9ef | 428 | bdrv_drain_invoke(bs, false); |
6cd5c9d7 | 429 | bdrv_parent_drained_end(bs, parent, ignore_bds_parents); |
0f115168 | 430 | if (old_quiesce_counter == 1) { |
0f115168 KW |
431 | aio_enable_external(bdrv_get_aio_context(bs)); |
432 | } | |
b0165585 KW |
433 | |
434 | if (recursive) { | |
6cd5c9d7 | 435 | assert(!ignore_bds_parents); |
d736f119 | 436 | bs->recursive_quiesce_counter--; |
b0165585 | 437 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { |
6cd5c9d7 | 438 | bdrv_do_drained_end(child->bs, true, child, ignore_bds_parents); |
b0165585 KW |
439 | } |
440 | } | |
6820643f KW |
441 | } |
442 | ||
0152bf40 KW |
443 | void bdrv_drained_end(BlockDriverState *bs) |
444 | { | |
6cd5c9d7 | 445 | bdrv_do_drained_end(bs, false, NULL, false); |
b0165585 KW |
446 | } |
447 | ||
448 | void bdrv_subtree_drained_end(BlockDriverState *bs) | |
449 | { | |
6cd5c9d7 | 450 | bdrv_do_drained_end(bs, true, NULL, false); |
0152bf40 KW |
451 | } |
452 | ||
d736f119 KW |
453 | void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent) |
454 | { | |
455 | int i; | |
456 | ||
457 | for (i = 0; i < new_parent->recursive_quiesce_counter; i++) { | |
6cd5c9d7 | 458 | bdrv_do_drained_begin(child->bs, true, child, false, true); |
d736f119 KW |
459 | } |
460 | } | |
461 | ||
462 | void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent) | |
463 | { | |
464 | int i; | |
465 | ||
466 | for (i = 0; i < old_parent->recursive_quiesce_counter; i++) { | |
6cd5c9d7 | 467 | bdrv_do_drained_end(child->bs, true, child, false); |
d736f119 KW |
468 | } |
469 | } | |
470 | ||
61007b31 | 471 | /* |
67da1dc5 FZ |
472 | * Wait for pending requests to complete on a single BlockDriverState subtree, |
473 | * and suspend block driver's internal I/O until next request arrives. | |
61007b31 | 474 | * |
61007b31 SH |
475 | * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState |
476 | * AioContext. | |
477 | */ | |
b6e84c97 | 478 | void coroutine_fn bdrv_co_drain(BlockDriverState *bs) |
61007b31 | 479 | { |
6820643f KW |
480 | assert(qemu_in_coroutine()); |
481 | bdrv_drained_begin(bs); | |
482 | bdrv_drained_end(bs); | |
b6e84c97 | 483 | } |
f406c03c | 484 | |
b6e84c97 PB |
485 | void bdrv_drain(BlockDriverState *bs) |
486 | { | |
6820643f KW |
487 | bdrv_drained_begin(bs); |
488 | bdrv_drained_end(bs); | |
61007b31 SH |
489 | } |
490 | ||
c13ad59f KW |
491 | static void bdrv_drain_assert_idle(BlockDriverState *bs) |
492 | { | |
493 | BdrvChild *child, *next; | |
494 | ||
495 | assert(atomic_read(&bs->in_flight) == 0); | |
496 | QLIST_FOREACH_SAFE(child, &bs->children, next, next) { | |
497 | bdrv_drain_assert_idle(child->bs); | |
498 | } | |
499 | } | |
500 | ||
0f12264e KW |
501 | unsigned int bdrv_drain_all_count = 0; |
502 | ||
503 | static bool bdrv_drain_all_poll(void) | |
504 | { | |
505 | BlockDriverState *bs = NULL; | |
506 | bool result = false; | |
507 | ||
0f12264e KW |
508 | /* bdrv_drain_poll() can't make changes to the graph and we are holding the |
509 | * main AioContext lock, so iterating bdrv_next_all_states() is safe. */ | |
510 | while ((bs = bdrv_next_all_states(bs))) { | |
511 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
512 | aio_context_acquire(aio_context); | |
513 | result |= bdrv_drain_poll(bs, false, NULL, true); | |
514 | aio_context_release(aio_context); | |
515 | } | |
516 | ||
517 | return result; | |
518 | } | |
519 | ||
61007b31 SH |
520 | /* |
521 | * Wait for pending requests to complete across all BlockDriverStates | |
522 | * | |
523 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
524 | * after calling this function. | |
c0778f66 AG |
525 | * |
526 | * This pauses all block jobs and disables external clients. It must | |
527 | * be paired with bdrv_drain_all_end(). | |
528 | * | |
529 | * NOTE: no new block jobs or BlockDriverStates can be created between | |
530 | * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls. | |
61007b31 | 531 | */ |
c0778f66 | 532 | void bdrv_drain_all_begin(void) |
61007b31 | 533 | { |
0f12264e | 534 | BlockDriverState *bs = NULL; |
61007b31 | 535 | |
c8ca33d0 | 536 | if (qemu_in_coroutine()) { |
0f12264e | 537 | bdrv_co_yield_to_drain(NULL, true, false, NULL, true, true); |
c8ca33d0 KW |
538 | return; |
539 | } | |
540 | ||
0f12264e KW |
541 | /* AIO_WAIT_WHILE() with a NULL context can only be called from the main |
542 | * loop AioContext, so make sure we're in the main context. */ | |
9a7e86c8 | 543 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
0f12264e KW |
544 | assert(bdrv_drain_all_count < INT_MAX); |
545 | bdrv_drain_all_count++; | |
9a7e86c8 | 546 | |
0f12264e KW |
547 | /* Quiesce all nodes, without polling in-flight requests yet. The graph |
548 | * cannot change during this loop. */ | |
549 | while ((bs = bdrv_next_all_states(bs))) { | |
61007b31 SH |
550 | AioContext *aio_context = bdrv_get_aio_context(bs); |
551 | ||
552 | aio_context_acquire(aio_context); | |
0f12264e | 553 | bdrv_do_drained_begin(bs, false, NULL, true, false); |
61007b31 SH |
554 | aio_context_release(aio_context); |
555 | } | |
556 | ||
0f12264e | 557 | /* Now poll the in-flight requests */ |
cfe29d82 | 558 | AIO_WAIT_WHILE(NULL, bdrv_drain_all_poll()); |
0f12264e KW |
559 | |
560 | while ((bs = bdrv_next_all_states(bs))) { | |
c13ad59f | 561 | bdrv_drain_assert_idle(bs); |
61007b31 | 562 | } |
c0778f66 AG |
563 | } |
564 | ||
565 | void bdrv_drain_all_end(void) | |
566 | { | |
0f12264e | 567 | BlockDriverState *bs = NULL; |
c0778f66 | 568 | |
0f12264e | 569 | while ((bs = bdrv_next_all_states(bs))) { |
61007b31 SH |
570 | AioContext *aio_context = bdrv_get_aio_context(bs); |
571 | ||
572 | aio_context_acquire(aio_context); | |
0f12264e | 573 | bdrv_do_drained_end(bs, false, NULL, true); |
61007b31 SH |
574 | aio_context_release(aio_context); |
575 | } | |
0f12264e KW |
576 | |
577 | assert(bdrv_drain_all_count > 0); | |
578 | bdrv_drain_all_count--; | |
61007b31 SH |
579 | } |
580 | ||
c0778f66 AG |
581 | void bdrv_drain_all(void) |
582 | { | |
583 | bdrv_drain_all_begin(); | |
584 | bdrv_drain_all_end(); | |
585 | } | |
586 | ||
61007b31 SH |
587 | /** |
588 | * Remove an active request from the tracked requests list | |
589 | * | |
590 | * This function should be called when a tracked request is completing. | |
591 | */ | |
592 | static void tracked_request_end(BdrvTrackedRequest *req) | |
593 | { | |
594 | if (req->serialising) { | |
20fc71b2 | 595 | atomic_dec(&req->bs->serialising_in_flight); |
61007b31 SH |
596 | } |
597 | ||
3783fa3d | 598 | qemu_co_mutex_lock(&req->bs->reqs_lock); |
61007b31 SH |
599 | QLIST_REMOVE(req, list); |
600 | qemu_co_queue_restart_all(&req->wait_queue); | |
3783fa3d | 601 | qemu_co_mutex_unlock(&req->bs->reqs_lock); |
61007b31 SH |
602 | } |
603 | ||
604 | /** | |
605 | * Add an active request to the tracked requests list | |
606 | */ | |
607 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
608 | BlockDriverState *bs, | |
609 | int64_t offset, | |
22931a15 | 610 | uint64_t bytes, |
ebde595c | 611 | enum BdrvTrackedRequestType type) |
61007b31 | 612 | { |
22931a15 FZ |
613 | assert(bytes <= INT64_MAX && offset <= INT64_MAX - bytes); |
614 | ||
61007b31 SH |
615 | *req = (BdrvTrackedRequest){ |
616 | .bs = bs, | |
617 | .offset = offset, | |
618 | .bytes = bytes, | |
ebde595c | 619 | .type = type, |
61007b31 SH |
620 | .co = qemu_coroutine_self(), |
621 | .serialising = false, | |
622 | .overlap_offset = offset, | |
623 | .overlap_bytes = bytes, | |
624 | }; | |
625 | ||
626 | qemu_co_queue_init(&req->wait_queue); | |
627 | ||
3783fa3d | 628 | qemu_co_mutex_lock(&bs->reqs_lock); |
61007b31 | 629 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); |
3783fa3d | 630 | qemu_co_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
631 | } |
632 | ||
633 | static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align) | |
634 | { | |
635 | int64_t overlap_offset = req->offset & ~(align - 1); | |
22931a15 | 636 | uint64_t overlap_bytes = ROUND_UP(req->offset + req->bytes, align) |
61007b31 SH |
637 | - overlap_offset; |
638 | ||
639 | if (!req->serialising) { | |
20fc71b2 | 640 | atomic_inc(&req->bs->serialising_in_flight); |
61007b31 SH |
641 | req->serialising = true; |
642 | } | |
643 | ||
644 | req->overlap_offset = MIN(req->overlap_offset, overlap_offset); | |
645 | req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes); | |
646 | } | |
647 | ||
09d2f948 VSO |
648 | static bool is_request_serialising_and_aligned(BdrvTrackedRequest *req) |
649 | { | |
650 | /* | |
651 | * If the request is serialising, overlap_offset and overlap_bytes are set, | |
652 | * so we can check if the request is aligned. Otherwise, don't care and | |
653 | * return false. | |
654 | */ | |
655 | ||
656 | return req->serialising && (req->offset == req->overlap_offset) && | |
657 | (req->bytes == req->overlap_bytes); | |
658 | } | |
659 | ||
244483e6 KW |
660 | /** |
661 | * Round a region to cluster boundaries | |
662 | */ | |
663 | void bdrv_round_to_clusters(BlockDriverState *bs, | |
7cfd5275 | 664 | int64_t offset, int64_t bytes, |
244483e6 | 665 | int64_t *cluster_offset, |
7cfd5275 | 666 | int64_t *cluster_bytes) |
244483e6 KW |
667 | { |
668 | BlockDriverInfo bdi; | |
669 | ||
670 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
671 | *cluster_offset = offset; | |
672 | *cluster_bytes = bytes; | |
673 | } else { | |
674 | int64_t c = bdi.cluster_size; | |
675 | *cluster_offset = QEMU_ALIGN_DOWN(offset, c); | |
676 | *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c); | |
677 | } | |
678 | } | |
679 | ||
61007b31 SH |
680 | static int bdrv_get_cluster_size(BlockDriverState *bs) |
681 | { | |
682 | BlockDriverInfo bdi; | |
683 | int ret; | |
684 | ||
685 | ret = bdrv_get_info(bs, &bdi); | |
686 | if (ret < 0 || bdi.cluster_size == 0) { | |
a5b8dd2c | 687 | return bs->bl.request_alignment; |
61007b31 SH |
688 | } else { |
689 | return bdi.cluster_size; | |
690 | } | |
691 | } | |
692 | ||
693 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
22931a15 | 694 | int64_t offset, uint64_t bytes) |
61007b31 SH |
695 | { |
696 | /* aaaa bbbb */ | |
697 | if (offset >= req->overlap_offset + req->overlap_bytes) { | |
698 | return false; | |
699 | } | |
700 | /* bbbb aaaa */ | |
701 | if (req->overlap_offset >= offset + bytes) { | |
702 | return false; | |
703 | } | |
704 | return true; | |
705 | } | |
706 | ||
99723548 PB |
707 | void bdrv_inc_in_flight(BlockDriverState *bs) |
708 | { | |
709 | atomic_inc(&bs->in_flight); | |
710 | } | |
711 | ||
c9d1a561 PB |
712 | void bdrv_wakeup(BlockDriverState *bs) |
713 | { | |
cfe29d82 | 714 | aio_wait_kick(); |
c9d1a561 PB |
715 | } |
716 | ||
99723548 PB |
717 | void bdrv_dec_in_flight(BlockDriverState *bs) |
718 | { | |
719 | atomic_dec(&bs->in_flight); | |
c9d1a561 | 720 | bdrv_wakeup(bs); |
99723548 PB |
721 | } |
722 | ||
61007b31 SH |
723 | static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self) |
724 | { | |
725 | BlockDriverState *bs = self->bs; | |
726 | BdrvTrackedRequest *req; | |
727 | bool retry; | |
728 | bool waited = false; | |
729 | ||
20fc71b2 | 730 | if (!atomic_read(&bs->serialising_in_flight)) { |
61007b31 SH |
731 | return false; |
732 | } | |
733 | ||
734 | do { | |
735 | retry = false; | |
3783fa3d | 736 | qemu_co_mutex_lock(&bs->reqs_lock); |
61007b31 SH |
737 | QLIST_FOREACH(req, &bs->tracked_requests, list) { |
738 | if (req == self || (!req->serialising && !self->serialising)) { | |
739 | continue; | |
740 | } | |
741 | if (tracked_request_overlaps(req, self->overlap_offset, | |
742 | self->overlap_bytes)) | |
743 | { | |
744 | /* Hitting this means there was a reentrant request, for | |
745 | * example, a block driver issuing nested requests. This must | |
746 | * never happen since it means deadlock. | |
747 | */ | |
748 | assert(qemu_coroutine_self() != req->co); | |
749 | ||
750 | /* If the request is already (indirectly) waiting for us, or | |
751 | * will wait for us as soon as it wakes up, then just go on | |
752 | * (instead of producing a deadlock in the former case). */ | |
753 | if (!req->waiting_for) { | |
754 | self->waiting_for = req; | |
3783fa3d | 755 | qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock); |
61007b31 SH |
756 | self->waiting_for = NULL; |
757 | retry = true; | |
758 | waited = true; | |
759 | break; | |
760 | } | |
761 | } | |
762 | } | |
3783fa3d | 763 | qemu_co_mutex_unlock(&bs->reqs_lock); |
61007b31 SH |
764 | } while (retry); |
765 | ||
766 | return waited; | |
767 | } | |
768 | ||
769 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
770 | size_t size) | |
771 | { | |
772 | if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) { | |
773 | return -EIO; | |
774 | } | |
775 | ||
776 | if (!bdrv_is_inserted(bs)) { | |
777 | return -ENOMEDIUM; | |
778 | } | |
779 | ||
780 | if (offset < 0) { | |
781 | return -EIO; | |
782 | } | |
783 | ||
784 | return 0; | |
785 | } | |
786 | ||
61007b31 | 787 | typedef struct RwCo { |
e293b7a3 | 788 | BdrvChild *child; |
61007b31 SH |
789 | int64_t offset; |
790 | QEMUIOVector *qiov; | |
791 | bool is_write; | |
792 | int ret; | |
793 | BdrvRequestFlags flags; | |
794 | } RwCo; | |
795 | ||
796 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
797 | { | |
798 | RwCo *rwco = opaque; | |
799 | ||
800 | if (!rwco->is_write) { | |
a03ef88f | 801 | rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset, |
cab3a356 KW |
802 | rwco->qiov->size, rwco->qiov, |
803 | rwco->flags); | |
61007b31 | 804 | } else { |
a03ef88f | 805 | rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset, |
cab3a356 KW |
806 | rwco->qiov->size, rwco->qiov, |
807 | rwco->flags); | |
61007b31 | 808 | } |
4720cbee | 809 | aio_wait_kick(); |
61007b31 SH |
810 | } |
811 | ||
812 | /* | |
813 | * Process a vectored synchronous request using coroutines | |
814 | */ | |
e293b7a3 | 815 | static int bdrv_prwv_co(BdrvChild *child, int64_t offset, |
61007b31 SH |
816 | QEMUIOVector *qiov, bool is_write, |
817 | BdrvRequestFlags flags) | |
818 | { | |
819 | Coroutine *co; | |
820 | RwCo rwco = { | |
e293b7a3 | 821 | .child = child, |
61007b31 SH |
822 | .offset = offset, |
823 | .qiov = qiov, | |
824 | .is_write = is_write, | |
825 | .ret = NOT_DONE, | |
826 | .flags = flags, | |
827 | }; | |
828 | ||
61007b31 SH |
829 | if (qemu_in_coroutine()) { |
830 | /* Fast-path if already in coroutine context */ | |
831 | bdrv_rw_co_entry(&rwco); | |
832 | } else { | |
0b8b8753 | 833 | co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco); |
e92f0e19 | 834 | bdrv_coroutine_enter(child->bs, co); |
88b062c2 | 835 | BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE); |
61007b31 SH |
836 | } |
837 | return rwco.ret; | |
838 | } | |
839 | ||
840 | /* | |
841 | * Process a synchronous request using coroutines | |
842 | */ | |
e293b7a3 | 843 | static int bdrv_rw_co(BdrvChild *child, int64_t sector_num, uint8_t *buf, |
61007b31 SH |
844 | int nb_sectors, bool is_write, BdrvRequestFlags flags) |
845 | { | |
0d93ed08 VSO |
846 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, |
847 | nb_sectors * BDRV_SECTOR_SIZE); | |
61007b31 SH |
848 | |
849 | if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) { | |
850 | return -EINVAL; | |
851 | } | |
852 | ||
e293b7a3 | 853 | return bdrv_prwv_co(child, sector_num << BDRV_SECTOR_BITS, |
61007b31 SH |
854 | &qiov, is_write, flags); |
855 | } | |
856 | ||
857 | /* return < 0 if error. See bdrv_write() for the return codes */ | |
fbcbbf4e | 858 | int bdrv_read(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
859 | uint8_t *buf, int nb_sectors) |
860 | { | |
e293b7a3 | 861 | return bdrv_rw_co(child, sector_num, buf, nb_sectors, false, 0); |
61007b31 SH |
862 | } |
863 | ||
61007b31 SH |
864 | /* Return < 0 if error. Important errors are: |
865 | -EIO generic I/O error (may happen for all errors) | |
866 | -ENOMEDIUM No media inserted. | |
867 | -EINVAL Invalid sector number or nb_sectors | |
868 | -EACCES Trying to write a read-only device | |
869 | */ | |
18d51c4b | 870 | int bdrv_write(BdrvChild *child, int64_t sector_num, |
61007b31 SH |
871 | const uint8_t *buf, int nb_sectors) |
872 | { | |
e293b7a3 | 873 | return bdrv_rw_co(child, sector_num, (uint8_t *)buf, nb_sectors, true, 0); |
61007b31 SH |
874 | } |
875 | ||
720ff280 | 876 | int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset, |
f5a5ca79 | 877 | int bytes, BdrvRequestFlags flags) |
61007b31 | 878 | { |
0d93ed08 | 879 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, NULL, bytes); |
74021bc4 | 880 | |
e293b7a3 | 881 | return bdrv_prwv_co(child, offset, &qiov, true, |
74021bc4 | 882 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
883 | } |
884 | ||
885 | /* | |
74021bc4 | 886 | * Completely zero out a block device with the help of bdrv_pwrite_zeroes. |
61007b31 SH |
887 | * The operation is sped up by checking the block status and only writing |
888 | * zeroes to the device if they currently do not return zeroes. Optional | |
74021bc4 | 889 | * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP, |
465fe887 | 890 | * BDRV_REQ_FUA). |
61007b31 SH |
891 | * |
892 | * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). | |
893 | */ | |
720ff280 | 894 | int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags) |
61007b31 | 895 | { |
237d78f8 EB |
896 | int ret; |
897 | int64_t target_size, bytes, offset = 0; | |
720ff280 | 898 | BlockDriverState *bs = child->bs; |
61007b31 | 899 | |
7286d610 EB |
900 | target_size = bdrv_getlength(bs); |
901 | if (target_size < 0) { | |
902 | return target_size; | |
61007b31 SH |
903 | } |
904 | ||
905 | for (;;) { | |
7286d610 EB |
906 | bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES); |
907 | if (bytes <= 0) { | |
61007b31 SH |
908 | return 0; |
909 | } | |
237d78f8 | 910 | ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL); |
61007b31 | 911 | if (ret < 0) { |
7286d610 EB |
912 | error_report("error getting block status at offset %" PRId64 ": %s", |
913 | offset, strerror(-ret)); | |
61007b31 SH |
914 | return ret; |
915 | } | |
916 | if (ret & BDRV_BLOCK_ZERO) { | |
237d78f8 | 917 | offset += bytes; |
61007b31 SH |
918 | continue; |
919 | } | |
237d78f8 | 920 | ret = bdrv_pwrite_zeroes(child, offset, bytes, flags); |
61007b31 | 921 | if (ret < 0) { |
7286d610 EB |
922 | error_report("error writing zeroes at offset %" PRId64 ": %s", |
923 | offset, strerror(-ret)); | |
61007b31 SH |
924 | return ret; |
925 | } | |
237d78f8 | 926 | offset += bytes; |
61007b31 SH |
927 | } |
928 | } | |
929 | ||
cf2ab8fc | 930 | int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
f1e84741 KW |
931 | { |
932 | int ret; | |
933 | ||
e293b7a3 | 934 | ret = bdrv_prwv_co(child, offset, qiov, false, 0); |
f1e84741 KW |
935 | if (ret < 0) { |
936 | return ret; | |
937 | } | |
938 | ||
939 | return qiov->size; | |
940 | } | |
941 | ||
cf2ab8fc | 942 | int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes) |
61007b31 | 943 | { |
0d93ed08 | 944 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes); |
61007b31 SH |
945 | |
946 | if (bytes < 0) { | |
947 | return -EINVAL; | |
948 | } | |
949 | ||
cf2ab8fc | 950 | return bdrv_preadv(child, offset, &qiov); |
61007b31 SH |
951 | } |
952 | ||
d9ca2ea2 | 953 | int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov) |
61007b31 SH |
954 | { |
955 | int ret; | |
956 | ||
e293b7a3 | 957 | ret = bdrv_prwv_co(child, offset, qiov, true, 0); |
61007b31 SH |
958 | if (ret < 0) { |
959 | return ret; | |
960 | } | |
961 | ||
962 | return qiov->size; | |
963 | } | |
964 | ||
d9ca2ea2 | 965 | int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes) |
61007b31 | 966 | { |
0d93ed08 | 967 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes); |
61007b31 SH |
968 | |
969 | if (bytes < 0) { | |
970 | return -EINVAL; | |
971 | } | |
972 | ||
d9ca2ea2 | 973 | return bdrv_pwritev(child, offset, &qiov); |
61007b31 SH |
974 | } |
975 | ||
976 | /* | |
977 | * Writes to the file and ensures that no writes are reordered across this | |
978 | * request (acts as a barrier) | |
979 | * | |
980 | * Returns 0 on success, -errno in error cases. | |
981 | */ | |
d9ca2ea2 KW |
982 | int bdrv_pwrite_sync(BdrvChild *child, int64_t offset, |
983 | const void *buf, int count) | |
61007b31 SH |
984 | { |
985 | int ret; | |
986 | ||
d9ca2ea2 | 987 | ret = bdrv_pwrite(child, offset, buf, count); |
61007b31 SH |
988 | if (ret < 0) { |
989 | return ret; | |
990 | } | |
991 | ||
d9ca2ea2 | 992 | ret = bdrv_flush(child->bs); |
855a6a93 KW |
993 | if (ret < 0) { |
994 | return ret; | |
61007b31 SH |
995 | } |
996 | ||
997 | return 0; | |
998 | } | |
999 | ||
08844473 KW |
1000 | typedef struct CoroutineIOCompletion { |
1001 | Coroutine *coroutine; | |
1002 | int ret; | |
1003 | } CoroutineIOCompletion; | |
1004 | ||
1005 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
1006 | { | |
1007 | CoroutineIOCompletion *co = opaque; | |
1008 | ||
1009 | co->ret = ret; | |
b9e413dd | 1010 | aio_co_wake(co->coroutine); |
08844473 KW |
1011 | } |
1012 | ||
166fe960 KW |
1013 | static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs, |
1014 | uint64_t offset, uint64_t bytes, | |
1015 | QEMUIOVector *qiov, int flags) | |
1016 | { | |
1017 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
1018 | int64_t sector_num; |
1019 | unsigned int nb_sectors; | |
1020 | ||
fa166538 EB |
1021 | assert(!(flags & ~BDRV_REQ_MASK)); |
1022 | ||
d470ad42 HR |
1023 | if (!drv) { |
1024 | return -ENOMEDIUM; | |
1025 | } | |
1026 | ||
3fb06697 KW |
1027 | if (drv->bdrv_co_preadv) { |
1028 | return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags); | |
1029 | } | |
1030 | ||
edfab6a0 | 1031 | if (drv->bdrv_aio_preadv) { |
08844473 KW |
1032 | BlockAIOCB *acb; |
1033 | CoroutineIOCompletion co = { | |
1034 | .coroutine = qemu_coroutine_self(), | |
1035 | }; | |
1036 | ||
edfab6a0 EB |
1037 | acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags, |
1038 | bdrv_co_io_em_complete, &co); | |
08844473 KW |
1039 | if (acb == NULL) { |
1040 | return -EIO; | |
1041 | } else { | |
1042 | qemu_coroutine_yield(); | |
1043 | return co.ret; | |
1044 | } | |
1045 | } | |
edfab6a0 EB |
1046 | |
1047 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1048 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1049 | ||
1050 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1051 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1052 | assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS); | |
1053 | assert(drv->bdrv_co_readv); | |
1054 | ||
1055 | return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
166fe960 KW |
1056 | } |
1057 | ||
78a07294 KW |
1058 | static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs, |
1059 | uint64_t offset, uint64_t bytes, | |
1060 | QEMUIOVector *qiov, int flags) | |
1061 | { | |
1062 | BlockDriver *drv = bs->drv; | |
3fb06697 KW |
1063 | int64_t sector_num; |
1064 | unsigned int nb_sectors; | |
78a07294 KW |
1065 | int ret; |
1066 | ||
fa166538 EB |
1067 | assert(!(flags & ~BDRV_REQ_MASK)); |
1068 | ||
d470ad42 HR |
1069 | if (!drv) { |
1070 | return -ENOMEDIUM; | |
1071 | } | |
1072 | ||
3fb06697 | 1073 | if (drv->bdrv_co_pwritev) { |
515c2f43 KW |
1074 | ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov, |
1075 | flags & bs->supported_write_flags); | |
1076 | flags &= ~bs->supported_write_flags; | |
3fb06697 KW |
1077 | goto emulate_flags; |
1078 | } | |
1079 | ||
edfab6a0 | 1080 | if (drv->bdrv_aio_pwritev) { |
08844473 KW |
1081 | BlockAIOCB *acb; |
1082 | CoroutineIOCompletion co = { | |
1083 | .coroutine = qemu_coroutine_self(), | |
1084 | }; | |
1085 | ||
edfab6a0 EB |
1086 | acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov, |
1087 | flags & bs->supported_write_flags, | |
1088 | bdrv_co_io_em_complete, &co); | |
1089 | flags &= ~bs->supported_write_flags; | |
08844473 | 1090 | if (acb == NULL) { |
3fb06697 | 1091 | ret = -EIO; |
08844473 KW |
1092 | } else { |
1093 | qemu_coroutine_yield(); | |
3fb06697 | 1094 | ret = co.ret; |
08844473 | 1095 | } |
edfab6a0 EB |
1096 | goto emulate_flags; |
1097 | } | |
1098 | ||
1099 | sector_num = offset >> BDRV_SECTOR_BITS; | |
1100 | nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
1101 | ||
1102 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1103 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
1104 | assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS); | |
1105 | ||
e18a58b4 EB |
1106 | assert(drv->bdrv_co_writev); |
1107 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov, | |
1108 | flags & bs->supported_write_flags); | |
1109 | flags &= ~bs->supported_write_flags; | |
78a07294 | 1110 | |
3fb06697 | 1111 | emulate_flags: |
4df863f3 | 1112 | if (ret == 0 && (flags & BDRV_REQ_FUA)) { |
78a07294 KW |
1113 | ret = bdrv_co_flush(bs); |
1114 | } | |
1115 | ||
1116 | return ret; | |
1117 | } | |
1118 | ||
29a298af PB |
1119 | static int coroutine_fn |
1120 | bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset, | |
1121 | uint64_t bytes, QEMUIOVector *qiov) | |
1122 | { | |
1123 | BlockDriver *drv = bs->drv; | |
1124 | ||
d470ad42 HR |
1125 | if (!drv) { |
1126 | return -ENOMEDIUM; | |
1127 | } | |
1128 | ||
29a298af PB |
1129 | if (!drv->bdrv_co_pwritev_compressed) { |
1130 | return -ENOTSUP; | |
1131 | } | |
1132 | ||
29a298af PB |
1133 | return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov); |
1134 | } | |
1135 | ||
85c97ca7 | 1136 | static int coroutine_fn bdrv_co_do_copy_on_readv(BdrvChild *child, |
244483e6 | 1137 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov) |
61007b31 | 1138 | { |
85c97ca7 KW |
1139 | BlockDriverState *bs = child->bs; |
1140 | ||
61007b31 SH |
1141 | /* Perform I/O through a temporary buffer so that users who scribble over |
1142 | * their read buffer while the operation is in progress do not end up | |
1143 | * modifying the image file. This is critical for zero-copy guest I/O | |
1144 | * where anything might happen inside guest memory. | |
1145 | */ | |
1146 | void *bounce_buffer; | |
1147 | ||
1148 | BlockDriver *drv = bs->drv; | |
cb2e2878 | 1149 | QEMUIOVector local_qiov; |
244483e6 | 1150 | int64_t cluster_offset; |
7cfd5275 | 1151 | int64_t cluster_bytes; |
61007b31 SH |
1152 | size_t skip_bytes; |
1153 | int ret; | |
cb2e2878 EB |
1154 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, |
1155 | BDRV_REQUEST_MAX_BYTES); | |
1156 | unsigned int progress = 0; | |
61007b31 | 1157 | |
d470ad42 HR |
1158 | if (!drv) { |
1159 | return -ENOMEDIUM; | |
1160 | } | |
1161 | ||
1bf03e66 KW |
1162 | /* FIXME We cannot require callers to have write permissions when all they |
1163 | * are doing is a read request. If we did things right, write permissions | |
1164 | * would be obtained anyway, but internally by the copy-on-read code. As | |
765d9df9 | 1165 | * long as it is implemented here rather than in a separate filter driver, |
1bf03e66 KW |
1166 | * the copy-on-read code doesn't have its own BdrvChild, however, for which |
1167 | * it could request permissions. Therefore we have to bypass the permission | |
1168 | * system for the moment. */ | |
1169 | // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
afa4b293 | 1170 | |
61007b31 | 1171 | /* Cover entire cluster so no additional backing file I/O is required when |
cb2e2878 EB |
1172 | * allocating cluster in the image file. Note that this value may exceed |
1173 | * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which | |
1174 | * is one reason we loop rather than doing it all at once. | |
61007b31 | 1175 | */ |
244483e6 | 1176 | bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes); |
cb2e2878 | 1177 | skip_bytes = offset - cluster_offset; |
61007b31 | 1178 | |
244483e6 KW |
1179 | trace_bdrv_co_do_copy_on_readv(bs, offset, bytes, |
1180 | cluster_offset, cluster_bytes); | |
61007b31 | 1181 | |
cb2e2878 EB |
1182 | bounce_buffer = qemu_try_blockalign(bs, |
1183 | MIN(MIN(max_transfer, cluster_bytes), | |
1184 | MAX_BOUNCE_BUFFER)); | |
61007b31 SH |
1185 | if (bounce_buffer == NULL) { |
1186 | ret = -ENOMEM; | |
1187 | goto err; | |
1188 | } | |
1189 | ||
cb2e2878 EB |
1190 | while (cluster_bytes) { |
1191 | int64_t pnum; | |
61007b31 | 1192 | |
cb2e2878 EB |
1193 | ret = bdrv_is_allocated(bs, cluster_offset, |
1194 | MIN(cluster_bytes, max_transfer), &pnum); | |
1195 | if (ret < 0) { | |
1196 | /* Safe to treat errors in querying allocation as if | |
1197 | * unallocated; we'll probably fail again soon on the | |
1198 | * read, but at least that will set a decent errno. | |
1199 | */ | |
1200 | pnum = MIN(cluster_bytes, max_transfer); | |
1201 | } | |
61007b31 | 1202 | |
b0ddcbbb KW |
1203 | /* Stop at EOF if the image ends in the middle of the cluster */ |
1204 | if (ret == 0 && pnum == 0) { | |
1205 | assert(progress >= bytes); | |
1206 | break; | |
1207 | } | |
1208 | ||
cb2e2878 | 1209 | assert(skip_bytes < pnum); |
61007b31 | 1210 | |
cb2e2878 EB |
1211 | if (ret <= 0) { |
1212 | /* Must copy-on-read; use the bounce buffer */ | |
0d93ed08 VSO |
1213 | pnum = MIN(pnum, MAX_BOUNCE_BUFFER); |
1214 | qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum); | |
61007b31 | 1215 | |
cb2e2878 EB |
1216 | ret = bdrv_driver_preadv(bs, cluster_offset, pnum, |
1217 | &local_qiov, 0); | |
1218 | if (ret < 0) { | |
1219 | goto err; | |
1220 | } | |
1221 | ||
1222 | bdrv_debug_event(bs, BLKDBG_COR_WRITE); | |
1223 | if (drv->bdrv_co_pwrite_zeroes && | |
1224 | buffer_is_zero(bounce_buffer, pnum)) { | |
1225 | /* FIXME: Should we (perhaps conditionally) be setting | |
1226 | * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy | |
1227 | * that still correctly reads as zero? */ | |
7adcf59f HR |
1228 | ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, pnum, |
1229 | BDRV_REQ_WRITE_UNCHANGED); | |
cb2e2878 EB |
1230 | } else { |
1231 | /* This does not change the data on the disk, it is not | |
1232 | * necessary to flush even in cache=writethrough mode. | |
1233 | */ | |
1234 | ret = bdrv_driver_pwritev(bs, cluster_offset, pnum, | |
7adcf59f HR |
1235 | &local_qiov, |
1236 | BDRV_REQ_WRITE_UNCHANGED); | |
cb2e2878 EB |
1237 | } |
1238 | ||
1239 | if (ret < 0) { | |
1240 | /* It might be okay to ignore write errors for guest | |
1241 | * requests. If this is a deliberate copy-on-read | |
1242 | * then we don't want to ignore the error. Simply | |
1243 | * report it in all cases. | |
1244 | */ | |
1245 | goto err; | |
1246 | } | |
1247 | ||
1248 | qemu_iovec_from_buf(qiov, progress, bounce_buffer + skip_bytes, | |
1249 | pnum - skip_bytes); | |
1250 | } else { | |
1251 | /* Read directly into the destination */ | |
1252 | qemu_iovec_init(&local_qiov, qiov->niov); | |
1253 | qemu_iovec_concat(&local_qiov, qiov, progress, pnum - skip_bytes); | |
1254 | ret = bdrv_driver_preadv(bs, offset + progress, local_qiov.size, | |
1255 | &local_qiov, 0); | |
1256 | qemu_iovec_destroy(&local_qiov); | |
1257 | if (ret < 0) { | |
1258 | goto err; | |
1259 | } | |
1260 | } | |
1261 | ||
1262 | cluster_offset += pnum; | |
1263 | cluster_bytes -= pnum; | |
1264 | progress += pnum - skip_bytes; | |
1265 | skip_bytes = 0; | |
1266 | } | |
1267 | ret = 0; | |
61007b31 SH |
1268 | |
1269 | err: | |
1270 | qemu_vfree(bounce_buffer); | |
1271 | return ret; | |
1272 | } | |
1273 | ||
1274 | /* | |
1275 | * Forwards an already correctly aligned request to the BlockDriver. This | |
1a62d0ac EB |
1276 | * handles copy on read, zeroing after EOF, and fragmentation of large |
1277 | * reads; any other features must be implemented by the caller. | |
61007b31 | 1278 | */ |
85c97ca7 | 1279 | static int coroutine_fn bdrv_aligned_preadv(BdrvChild *child, |
61007b31 SH |
1280 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, |
1281 | int64_t align, QEMUIOVector *qiov, int flags) | |
1282 | { | |
85c97ca7 | 1283 | BlockDriverState *bs = child->bs; |
c9d20029 | 1284 | int64_t total_bytes, max_bytes; |
1a62d0ac EB |
1285 | int ret = 0; |
1286 | uint64_t bytes_remaining = bytes; | |
1287 | int max_transfer; | |
61007b31 | 1288 | |
49c07526 KW |
1289 | assert(is_power_of_2(align)); |
1290 | assert((offset & (align - 1)) == 0); | |
1291 | assert((bytes & (align - 1)) == 0); | |
61007b31 | 1292 | assert(!qiov || bytes == qiov->size); |
abb06c5a | 1293 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); |
1a62d0ac EB |
1294 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1295 | align); | |
a604fa2b EB |
1296 | |
1297 | /* TODO: We would need a per-BDS .supported_read_flags and | |
1298 | * potential fallback support, if we ever implement any read flags | |
1299 | * to pass through to drivers. For now, there aren't any | |
1300 | * passthrough flags. */ | |
1301 | assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ))); | |
61007b31 SH |
1302 | |
1303 | /* Handle Copy on Read and associated serialisation */ | |
1304 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
1305 | /* If we touch the same cluster it counts as an overlap. This | |
1306 | * guarantees that allocating writes will be serialized and not race | |
1307 | * with each other for the same cluster. For example, in copy-on-read | |
1308 | * it ensures that the CoR read and write operations are atomic and | |
1309 | * guest writes cannot interleave between them. */ | |
1310 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
1311 | } | |
1312 | ||
09d2f948 VSO |
1313 | /* BDRV_REQ_SERIALISING is only for write operation */ |
1314 | assert(!(flags & BDRV_REQ_SERIALISING)); | |
1315 | ||
61408b25 FZ |
1316 | if (!(flags & BDRV_REQ_NO_SERIALISING)) { |
1317 | wait_serialising_requests(req); | |
1318 | } | |
61007b31 SH |
1319 | |
1320 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
d6a644bb | 1321 | int64_t pnum; |
61007b31 | 1322 | |
88e63df2 | 1323 | ret = bdrv_is_allocated(bs, offset, bytes, &pnum); |
61007b31 SH |
1324 | if (ret < 0) { |
1325 | goto out; | |
1326 | } | |
1327 | ||
88e63df2 | 1328 | if (!ret || pnum != bytes) { |
85c97ca7 | 1329 | ret = bdrv_co_do_copy_on_readv(child, offset, bytes, qiov); |
61007b31 SH |
1330 | goto out; |
1331 | } | |
1332 | } | |
1333 | ||
1a62d0ac | 1334 | /* Forward the request to the BlockDriver, possibly fragmenting it */ |
c9d20029 KW |
1335 | total_bytes = bdrv_getlength(bs); |
1336 | if (total_bytes < 0) { | |
1337 | ret = total_bytes; | |
1338 | goto out; | |
1339 | } | |
61007b31 | 1340 | |
c9d20029 | 1341 | max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align); |
1a62d0ac | 1342 | if (bytes <= max_bytes && bytes <= max_transfer) { |
c9d20029 | 1343 | ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0); |
1a62d0ac EB |
1344 | goto out; |
1345 | } | |
61007b31 | 1346 | |
1a62d0ac EB |
1347 | while (bytes_remaining) { |
1348 | int num; | |
61007b31 | 1349 | |
1a62d0ac EB |
1350 | if (max_bytes) { |
1351 | QEMUIOVector local_qiov; | |
61007b31 | 1352 | |
1a62d0ac EB |
1353 | num = MIN(bytes_remaining, MIN(max_bytes, max_transfer)); |
1354 | assert(num); | |
1355 | qemu_iovec_init(&local_qiov, qiov->niov); | |
1356 | qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num); | |
61007b31 | 1357 | |
1a62d0ac EB |
1358 | ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining, |
1359 | num, &local_qiov, 0); | |
1360 | max_bytes -= num; | |
1361 | qemu_iovec_destroy(&local_qiov); | |
1362 | } else { | |
1363 | num = bytes_remaining; | |
1364 | ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0, | |
1365 | bytes_remaining); | |
1366 | } | |
1367 | if (ret < 0) { | |
1368 | goto out; | |
1369 | } | |
1370 | bytes_remaining -= num; | |
61007b31 SH |
1371 | } |
1372 | ||
1373 | out: | |
1a62d0ac | 1374 | return ret < 0 ? ret : 0; |
61007b31 SH |
1375 | } |
1376 | ||
61007b31 SH |
1377 | /* |
1378 | * Handle a read request in coroutine context | |
1379 | */ | |
a03ef88f | 1380 | int coroutine_fn bdrv_co_preadv(BdrvChild *child, |
61007b31 SH |
1381 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1382 | BdrvRequestFlags flags) | |
1383 | { | |
a03ef88f | 1384 | BlockDriverState *bs = child->bs; |
61007b31 SH |
1385 | BlockDriver *drv = bs->drv; |
1386 | BdrvTrackedRequest req; | |
1387 | ||
a5b8dd2c | 1388 | uint64_t align = bs->bl.request_alignment; |
61007b31 SH |
1389 | uint8_t *head_buf = NULL; |
1390 | uint8_t *tail_buf = NULL; | |
1391 | QEMUIOVector local_qiov; | |
1392 | bool use_local_qiov = false; | |
1393 | int ret; | |
1394 | ||
f42cf447 DB |
1395 | trace_bdrv_co_preadv(child->bs, offset, bytes, flags); |
1396 | ||
61007b31 SH |
1397 | if (!drv) { |
1398 | return -ENOMEDIUM; | |
1399 | } | |
1400 | ||
1401 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1402 | if (ret < 0) { | |
1403 | return ret; | |
1404 | } | |
1405 | ||
99723548 PB |
1406 | bdrv_inc_in_flight(bs); |
1407 | ||
9568b511 | 1408 | /* Don't do copy-on-read if we read data before write operation */ |
d3faa13e | 1409 | if (atomic_read(&bs->copy_on_read) && !(flags & BDRV_REQ_NO_SERIALISING)) { |
61007b31 SH |
1410 | flags |= BDRV_REQ_COPY_ON_READ; |
1411 | } | |
1412 | ||
61007b31 SH |
1413 | /* Align read if necessary by padding qiov */ |
1414 | if (offset & (align - 1)) { | |
1415 | head_buf = qemu_blockalign(bs, align); | |
1416 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1417 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1418 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1419 | use_local_qiov = true; | |
1420 | ||
1421 | bytes += offset & (align - 1); | |
1422 | offset = offset & ~(align - 1); | |
1423 | } | |
1424 | ||
1425 | if ((offset + bytes) & (align - 1)) { | |
1426 | if (!use_local_qiov) { | |
1427 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1428 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1429 | use_local_qiov = true; | |
1430 | } | |
1431 | tail_buf = qemu_blockalign(bs, align); | |
1432 | qemu_iovec_add(&local_qiov, tail_buf, | |
1433 | align - ((offset + bytes) & (align - 1))); | |
1434 | ||
1435 | bytes = ROUND_UP(bytes, align); | |
1436 | } | |
1437 | ||
ebde595c | 1438 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ); |
85c97ca7 | 1439 | ret = bdrv_aligned_preadv(child, &req, offset, bytes, align, |
61007b31 SH |
1440 | use_local_qiov ? &local_qiov : qiov, |
1441 | flags); | |
1442 | tracked_request_end(&req); | |
99723548 | 1443 | bdrv_dec_in_flight(bs); |
61007b31 SH |
1444 | |
1445 | if (use_local_qiov) { | |
1446 | qemu_iovec_destroy(&local_qiov); | |
1447 | qemu_vfree(head_buf); | |
1448 | qemu_vfree(tail_buf); | |
1449 | } | |
1450 | ||
1451 | return ret; | |
1452 | } | |
1453 | ||
d05aa8bb | 1454 | static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 1455 | int64_t offset, int bytes, BdrvRequestFlags flags) |
61007b31 SH |
1456 | { |
1457 | BlockDriver *drv = bs->drv; | |
1458 | QEMUIOVector qiov; | |
0d93ed08 | 1459 | void *buf = NULL; |
61007b31 | 1460 | int ret = 0; |
465fe887 | 1461 | bool need_flush = false; |
443668ca DL |
1462 | int head = 0; |
1463 | int tail = 0; | |
61007b31 | 1464 | |
cf081fca | 1465 | int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX); |
a5b8dd2c EB |
1466 | int alignment = MAX(bs->bl.pwrite_zeroes_alignment, |
1467 | bs->bl.request_alignment); | |
cb2e2878 | 1468 | int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER); |
d05aa8bb | 1469 | |
d470ad42 HR |
1470 | if (!drv) { |
1471 | return -ENOMEDIUM; | |
1472 | } | |
1473 | ||
b8d0a980 EB |
1474 | assert(alignment % bs->bl.request_alignment == 0); |
1475 | head = offset % alignment; | |
f5a5ca79 | 1476 | tail = (offset + bytes) % alignment; |
b8d0a980 EB |
1477 | max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment); |
1478 | assert(max_write_zeroes >= bs->bl.request_alignment); | |
61007b31 | 1479 | |
f5a5ca79 MP |
1480 | while (bytes > 0 && !ret) { |
1481 | int num = bytes; | |
61007b31 SH |
1482 | |
1483 | /* Align request. Block drivers can expect the "bulk" of the request | |
443668ca DL |
1484 | * to be aligned, and that unaligned requests do not cross cluster |
1485 | * boundaries. | |
61007b31 | 1486 | */ |
443668ca | 1487 | if (head) { |
b2f95fee EB |
1488 | /* Make a small request up to the first aligned sector. For |
1489 | * convenience, limit this request to max_transfer even if | |
1490 | * we don't need to fall back to writes. */ | |
f5a5ca79 | 1491 | num = MIN(MIN(bytes, max_transfer), alignment - head); |
b2f95fee EB |
1492 | head = (head + num) % alignment; |
1493 | assert(num < max_write_zeroes); | |
d05aa8bb | 1494 | } else if (tail && num > alignment) { |
443668ca DL |
1495 | /* Shorten the request to the last aligned sector. */ |
1496 | num -= tail; | |
61007b31 SH |
1497 | } |
1498 | ||
1499 | /* limit request size */ | |
1500 | if (num > max_write_zeroes) { | |
1501 | num = max_write_zeroes; | |
1502 | } | |
1503 | ||
1504 | ret = -ENOTSUP; | |
1505 | /* First try the efficient write zeroes operation */ | |
d05aa8bb EB |
1506 | if (drv->bdrv_co_pwrite_zeroes) { |
1507 | ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num, | |
1508 | flags & bs->supported_zero_flags); | |
1509 | if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) && | |
1510 | !(bs->supported_zero_flags & BDRV_REQ_FUA)) { | |
1511 | need_flush = true; | |
1512 | } | |
465fe887 EB |
1513 | } else { |
1514 | assert(!bs->supported_zero_flags); | |
61007b31 SH |
1515 | } |
1516 | ||
1517 | if (ret == -ENOTSUP) { | |
1518 | /* Fall back to bounce buffer if write zeroes is unsupported */ | |
465fe887 EB |
1519 | BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE; |
1520 | ||
1521 | if ((flags & BDRV_REQ_FUA) && | |
1522 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1523 | /* No need for bdrv_driver_pwrite() to do a fallback | |
1524 | * flush on each chunk; use just one at the end */ | |
1525 | write_flags &= ~BDRV_REQ_FUA; | |
1526 | need_flush = true; | |
1527 | } | |
5def6b80 | 1528 | num = MIN(num, max_transfer); |
0d93ed08 VSO |
1529 | if (buf == NULL) { |
1530 | buf = qemu_try_blockalign0(bs, num); | |
1531 | if (buf == NULL) { | |
61007b31 SH |
1532 | ret = -ENOMEM; |
1533 | goto fail; | |
1534 | } | |
61007b31 | 1535 | } |
0d93ed08 | 1536 | qemu_iovec_init_buf(&qiov, buf, num); |
61007b31 | 1537 | |
d05aa8bb | 1538 | ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags); |
61007b31 SH |
1539 | |
1540 | /* Keep bounce buffer around if it is big enough for all | |
1541 | * all future requests. | |
1542 | */ | |
5def6b80 | 1543 | if (num < max_transfer) { |
0d93ed08 VSO |
1544 | qemu_vfree(buf); |
1545 | buf = NULL; | |
61007b31 SH |
1546 | } |
1547 | } | |
1548 | ||
d05aa8bb | 1549 | offset += num; |
f5a5ca79 | 1550 | bytes -= num; |
61007b31 SH |
1551 | } |
1552 | ||
1553 | fail: | |
465fe887 EB |
1554 | if (ret == 0 && need_flush) { |
1555 | ret = bdrv_co_flush(bs); | |
1556 | } | |
0d93ed08 | 1557 | qemu_vfree(buf); |
61007b31 SH |
1558 | return ret; |
1559 | } | |
1560 | ||
85fe2479 FZ |
1561 | static inline int coroutine_fn |
1562 | bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes, | |
1563 | BdrvTrackedRequest *req, int flags) | |
1564 | { | |
1565 | BlockDriverState *bs = child->bs; | |
1566 | bool waited; | |
1567 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1568 | ||
1569 | if (bs->read_only) { | |
1570 | return -EPERM; | |
1571 | } | |
1572 | ||
1573 | /* BDRV_REQ_NO_SERIALISING is only for read operation */ | |
1574 | assert(!(flags & BDRV_REQ_NO_SERIALISING)); | |
1575 | assert(!(bs->open_flags & BDRV_O_INACTIVE)); | |
1576 | assert((bs->open_flags & BDRV_O_NO_IO) == 0); | |
1577 | assert(!(flags & ~BDRV_REQ_MASK)); | |
1578 | ||
1579 | if (flags & BDRV_REQ_SERIALISING) { | |
1580 | mark_request_serialising(req, bdrv_get_cluster_size(bs)); | |
1581 | } | |
1582 | ||
1583 | waited = wait_serialising_requests(req); | |
1584 | ||
1585 | assert(!waited || !req->serialising || | |
1586 | is_request_serialising_and_aligned(req)); | |
1587 | assert(req->overlap_offset <= offset); | |
1588 | assert(offset + bytes <= req->overlap_offset + req->overlap_bytes); | |
cd47d792 | 1589 | assert(end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE); |
85fe2479 | 1590 | |
cd47d792 FZ |
1591 | switch (req->type) { |
1592 | case BDRV_TRACKED_WRITE: | |
1593 | case BDRV_TRACKED_DISCARD: | |
1594 | if (flags & BDRV_REQ_WRITE_UNCHANGED) { | |
1595 | assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE)); | |
1596 | } else { | |
1597 | assert(child->perm & BLK_PERM_WRITE); | |
1598 | } | |
1599 | return notifier_with_return_list_notify(&bs->before_write_notifiers, | |
1600 | req); | |
1601 | case BDRV_TRACKED_TRUNCATE: | |
1602 | assert(child->perm & BLK_PERM_RESIZE); | |
1603 | return 0; | |
1604 | default: | |
1605 | abort(); | |
85fe2479 | 1606 | } |
85fe2479 FZ |
1607 | } |
1608 | ||
1609 | static inline void coroutine_fn | |
1610 | bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, uint64_t bytes, | |
1611 | BdrvTrackedRequest *req, int ret) | |
1612 | { | |
1613 | int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE); | |
1614 | BlockDriverState *bs = child->bs; | |
1615 | ||
1616 | atomic_inc(&bs->write_gen); | |
85fe2479 | 1617 | |
00695c27 FZ |
1618 | /* |
1619 | * Discard cannot extend the image, but in error handling cases, such as | |
1620 | * when reverting a qcow2 cluster allocation, the discarded range can pass | |
1621 | * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD | |
1622 | * here. Instead, just skip it, since semantically a discard request | |
1623 | * beyond EOF cannot expand the image anyway. | |
1624 | */ | |
7f8f03ef | 1625 | if (ret == 0 && |
cd47d792 FZ |
1626 | (req->type == BDRV_TRACKED_TRUNCATE || |
1627 | end_sector > bs->total_sectors) && | |
1628 | req->type != BDRV_TRACKED_DISCARD) { | |
7f8f03ef FZ |
1629 | bs->total_sectors = end_sector; |
1630 | bdrv_parent_cb_resize(bs); | |
1631 | bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS); | |
85fe2479 | 1632 | } |
00695c27 FZ |
1633 | if (req->bytes) { |
1634 | switch (req->type) { | |
1635 | case BDRV_TRACKED_WRITE: | |
1636 | stat64_max(&bs->wr_highest_offset, offset + bytes); | |
1637 | /* fall through, to set dirty bits */ | |
1638 | case BDRV_TRACKED_DISCARD: | |
1639 | bdrv_set_dirty(bs, offset, bytes); | |
1640 | break; | |
1641 | default: | |
1642 | break; | |
1643 | } | |
1644 | } | |
85fe2479 FZ |
1645 | } |
1646 | ||
61007b31 | 1647 | /* |
04ed95f4 EB |
1648 | * Forwards an already correctly aligned write request to the BlockDriver, |
1649 | * after possibly fragmenting it. | |
61007b31 | 1650 | */ |
85c97ca7 | 1651 | static int coroutine_fn bdrv_aligned_pwritev(BdrvChild *child, |
61007b31 | 1652 | BdrvTrackedRequest *req, int64_t offset, unsigned int bytes, |
cff86b38 | 1653 | int64_t align, QEMUIOVector *qiov, int flags) |
61007b31 | 1654 | { |
85c97ca7 | 1655 | BlockDriverState *bs = child->bs; |
61007b31 | 1656 | BlockDriver *drv = bs->drv; |
61007b31 SH |
1657 | int ret; |
1658 | ||
04ed95f4 EB |
1659 | uint64_t bytes_remaining = bytes; |
1660 | int max_transfer; | |
61007b31 | 1661 | |
d470ad42 HR |
1662 | if (!drv) { |
1663 | return -ENOMEDIUM; | |
1664 | } | |
1665 | ||
d6883bc9 VSO |
1666 | if (bdrv_has_readonly_bitmaps(bs)) { |
1667 | return -EPERM; | |
1668 | } | |
1669 | ||
cff86b38 EB |
1670 | assert(is_power_of_2(align)); |
1671 | assert((offset & (align - 1)) == 0); | |
1672 | assert((bytes & (align - 1)) == 0); | |
61007b31 | 1673 | assert(!qiov || bytes == qiov->size); |
04ed95f4 EB |
1674 | max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX), |
1675 | align); | |
61007b31 | 1676 | |
85fe2479 | 1677 | ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags); |
61007b31 SH |
1678 | |
1679 | if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF && | |
c1499a5e | 1680 | !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes && |
61007b31 SH |
1681 | qemu_iovec_is_zero(qiov)) { |
1682 | flags |= BDRV_REQ_ZERO_WRITE; | |
1683 | if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) { | |
1684 | flags |= BDRV_REQ_MAY_UNMAP; | |
1685 | } | |
1686 | } | |
1687 | ||
1688 | if (ret < 0) { | |
1689 | /* Do nothing, write notifier decided to fail this request */ | |
1690 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
9a4f4c31 | 1691 | bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO); |
9896c876 | 1692 | ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags); |
3ea1a091 PB |
1693 | } else if (flags & BDRV_REQ_WRITE_COMPRESSED) { |
1694 | ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, qiov); | |
04ed95f4 | 1695 | } else if (bytes <= max_transfer) { |
9a4f4c31 | 1696 | bdrv_debug_event(bs, BLKDBG_PWRITEV); |
78a07294 | 1697 | ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags); |
04ed95f4 EB |
1698 | } else { |
1699 | bdrv_debug_event(bs, BLKDBG_PWRITEV); | |
1700 | while (bytes_remaining) { | |
1701 | int num = MIN(bytes_remaining, max_transfer); | |
1702 | QEMUIOVector local_qiov; | |
1703 | int local_flags = flags; | |
1704 | ||
1705 | assert(num); | |
1706 | if (num < bytes_remaining && (flags & BDRV_REQ_FUA) && | |
1707 | !(bs->supported_write_flags & BDRV_REQ_FUA)) { | |
1708 | /* If FUA is going to be emulated by flush, we only | |
1709 | * need to flush on the last iteration */ | |
1710 | local_flags &= ~BDRV_REQ_FUA; | |
1711 | } | |
1712 | qemu_iovec_init(&local_qiov, qiov->niov); | |
1713 | qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num); | |
1714 | ||
1715 | ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining, | |
1716 | num, &local_qiov, local_flags); | |
1717 | qemu_iovec_destroy(&local_qiov); | |
1718 | if (ret < 0) { | |
1719 | break; | |
1720 | } | |
1721 | bytes_remaining -= num; | |
1722 | } | |
61007b31 | 1723 | } |
9a4f4c31 | 1724 | bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE); |
61007b31 | 1725 | |
61007b31 | 1726 | if (ret >= 0) { |
04ed95f4 | 1727 | ret = 0; |
61007b31 | 1728 | } |
85fe2479 | 1729 | bdrv_co_write_req_finish(child, offset, bytes, req, ret); |
61007b31 SH |
1730 | |
1731 | return ret; | |
1732 | } | |
1733 | ||
85c97ca7 | 1734 | static int coroutine_fn bdrv_co_do_zero_pwritev(BdrvChild *child, |
9eeb6dd1 FZ |
1735 | int64_t offset, |
1736 | unsigned int bytes, | |
1737 | BdrvRequestFlags flags, | |
1738 | BdrvTrackedRequest *req) | |
1739 | { | |
85c97ca7 | 1740 | BlockDriverState *bs = child->bs; |
9eeb6dd1 FZ |
1741 | uint8_t *buf = NULL; |
1742 | QEMUIOVector local_qiov; | |
a5b8dd2c | 1743 | uint64_t align = bs->bl.request_alignment; |
9eeb6dd1 FZ |
1744 | unsigned int head_padding_bytes, tail_padding_bytes; |
1745 | int ret = 0; | |
1746 | ||
1747 | head_padding_bytes = offset & (align - 1); | |
f13ce1be | 1748 | tail_padding_bytes = (align - (offset + bytes)) & (align - 1); |
9eeb6dd1 FZ |
1749 | |
1750 | ||
1751 | assert(flags & BDRV_REQ_ZERO_WRITE); | |
1752 | if (head_padding_bytes || tail_padding_bytes) { | |
1753 | buf = qemu_blockalign(bs, align); | |
0d93ed08 | 1754 | qemu_iovec_init_buf(&local_qiov, buf, align); |
9eeb6dd1 FZ |
1755 | } |
1756 | if (head_padding_bytes) { | |
1757 | uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes); | |
1758 | ||
1759 | /* RMW the unaligned part before head. */ | |
1760 | mark_request_serialising(req, align); | |
1761 | wait_serialising_requests(req); | |
9a4f4c31 | 1762 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
85c97ca7 | 1763 | ret = bdrv_aligned_preadv(child, req, offset & ~(align - 1), align, |
9eeb6dd1 FZ |
1764 | align, &local_qiov, 0); |
1765 | if (ret < 0) { | |
1766 | goto fail; | |
1767 | } | |
9a4f4c31 | 1768 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
9eeb6dd1 FZ |
1769 | |
1770 | memset(buf + head_padding_bytes, 0, zero_bytes); | |
85c97ca7 | 1771 | ret = bdrv_aligned_pwritev(child, req, offset & ~(align - 1), align, |
cff86b38 | 1772 | align, &local_qiov, |
9eeb6dd1 FZ |
1773 | flags & ~BDRV_REQ_ZERO_WRITE); |
1774 | if (ret < 0) { | |
1775 | goto fail; | |
1776 | } | |
1777 | offset += zero_bytes; | |
1778 | bytes -= zero_bytes; | |
1779 | } | |
1780 | ||
1781 | assert(!bytes || (offset & (align - 1)) == 0); | |
1782 | if (bytes >= align) { | |
1783 | /* Write the aligned part in the middle. */ | |
1784 | uint64_t aligned_bytes = bytes & ~(align - 1); | |
85c97ca7 | 1785 | ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align, |
9eeb6dd1 FZ |
1786 | NULL, flags); |
1787 | if (ret < 0) { | |
1788 | goto fail; | |
1789 | } | |
1790 | bytes -= aligned_bytes; | |
1791 | offset += aligned_bytes; | |
1792 | } | |
1793 | ||
1794 | assert(!bytes || (offset & (align - 1)) == 0); | |
1795 | if (bytes) { | |
1796 | assert(align == tail_padding_bytes + bytes); | |
1797 | /* RMW the unaligned part after tail. */ | |
1798 | mark_request_serialising(req, align); | |
1799 | wait_serialising_requests(req); | |
9a4f4c31 | 1800 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
85c97ca7 | 1801 | ret = bdrv_aligned_preadv(child, req, offset, align, |
9eeb6dd1 FZ |
1802 | align, &local_qiov, 0); |
1803 | if (ret < 0) { | |
1804 | goto fail; | |
1805 | } | |
9a4f4c31 | 1806 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
9eeb6dd1 FZ |
1807 | |
1808 | memset(buf, 0, bytes); | |
85c97ca7 | 1809 | ret = bdrv_aligned_pwritev(child, req, offset, align, align, |
9eeb6dd1 FZ |
1810 | &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE); |
1811 | } | |
1812 | fail: | |
1813 | qemu_vfree(buf); | |
1814 | return ret; | |
1815 | ||
1816 | } | |
1817 | ||
61007b31 SH |
1818 | /* |
1819 | * Handle a write request in coroutine context | |
1820 | */ | |
a03ef88f | 1821 | int coroutine_fn bdrv_co_pwritev(BdrvChild *child, |
61007b31 SH |
1822 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, |
1823 | BdrvRequestFlags flags) | |
1824 | { | |
a03ef88f | 1825 | BlockDriverState *bs = child->bs; |
61007b31 | 1826 | BdrvTrackedRequest req; |
a5b8dd2c | 1827 | uint64_t align = bs->bl.request_alignment; |
61007b31 SH |
1828 | uint8_t *head_buf = NULL; |
1829 | uint8_t *tail_buf = NULL; | |
1830 | QEMUIOVector local_qiov; | |
1831 | bool use_local_qiov = false; | |
1832 | int ret; | |
1833 | ||
f42cf447 DB |
1834 | trace_bdrv_co_pwritev(child->bs, offset, bytes, flags); |
1835 | ||
61007b31 SH |
1836 | if (!bs->drv) { |
1837 | return -ENOMEDIUM; | |
1838 | } | |
61007b31 SH |
1839 | |
1840 | ret = bdrv_check_byte_request(bs, offset, bytes); | |
1841 | if (ret < 0) { | |
1842 | return ret; | |
1843 | } | |
1844 | ||
99723548 | 1845 | bdrv_inc_in_flight(bs); |
61007b31 SH |
1846 | /* |
1847 | * Align write if necessary by performing a read-modify-write cycle. | |
1848 | * Pad qiov with the read parts and be sure to have a tracked request not | |
1849 | * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle. | |
1850 | */ | |
ebde595c | 1851 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE); |
61007b31 | 1852 | |
18a59f03 | 1853 | if (flags & BDRV_REQ_ZERO_WRITE) { |
85c97ca7 | 1854 | ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req); |
9eeb6dd1 FZ |
1855 | goto out; |
1856 | } | |
1857 | ||
61007b31 SH |
1858 | if (offset & (align - 1)) { |
1859 | QEMUIOVector head_qiov; | |
61007b31 SH |
1860 | |
1861 | mark_request_serialising(&req, align); | |
1862 | wait_serialising_requests(&req); | |
1863 | ||
1864 | head_buf = qemu_blockalign(bs, align); | |
0d93ed08 | 1865 | qemu_iovec_init_buf(&head_qiov, head_buf, align); |
61007b31 | 1866 | |
9a4f4c31 | 1867 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD); |
85c97ca7 | 1868 | ret = bdrv_aligned_preadv(child, &req, offset & ~(align - 1), align, |
61007b31 SH |
1869 | align, &head_qiov, 0); |
1870 | if (ret < 0) { | |
1871 | goto fail; | |
1872 | } | |
9a4f4c31 | 1873 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD); |
61007b31 SH |
1874 | |
1875 | qemu_iovec_init(&local_qiov, qiov->niov + 2); | |
1876 | qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1)); | |
1877 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1878 | use_local_qiov = true; | |
1879 | ||
1880 | bytes += offset & (align - 1); | |
1881 | offset = offset & ~(align - 1); | |
117bc3fa PL |
1882 | |
1883 | /* We have read the tail already if the request is smaller | |
1884 | * than one aligned block. | |
1885 | */ | |
1886 | if (bytes < align) { | |
1887 | qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes); | |
1888 | bytes = align; | |
1889 | } | |
61007b31 SH |
1890 | } |
1891 | ||
1892 | if ((offset + bytes) & (align - 1)) { | |
1893 | QEMUIOVector tail_qiov; | |
61007b31 SH |
1894 | size_t tail_bytes; |
1895 | bool waited; | |
1896 | ||
1897 | mark_request_serialising(&req, align); | |
1898 | waited = wait_serialising_requests(&req); | |
1899 | assert(!waited || !use_local_qiov); | |
1900 | ||
1901 | tail_buf = qemu_blockalign(bs, align); | |
0d93ed08 | 1902 | qemu_iovec_init_buf(&tail_qiov, tail_buf, align); |
61007b31 | 1903 | |
9a4f4c31 | 1904 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL); |
85c97ca7 KW |
1905 | ret = bdrv_aligned_preadv(child, &req, (offset + bytes) & ~(align - 1), |
1906 | align, align, &tail_qiov, 0); | |
61007b31 SH |
1907 | if (ret < 0) { |
1908 | goto fail; | |
1909 | } | |
9a4f4c31 | 1910 | bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL); |
61007b31 SH |
1911 | |
1912 | if (!use_local_qiov) { | |
1913 | qemu_iovec_init(&local_qiov, qiov->niov + 1); | |
1914 | qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size); | |
1915 | use_local_qiov = true; | |
1916 | } | |
1917 | ||
1918 | tail_bytes = (offset + bytes) & (align - 1); | |
1919 | qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes); | |
1920 | ||
1921 | bytes = ROUND_UP(bytes, align); | |
1922 | } | |
1923 | ||
85c97ca7 | 1924 | ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align, |
3ea1a091 PB |
1925 | use_local_qiov ? &local_qiov : qiov, |
1926 | flags); | |
61007b31 SH |
1927 | |
1928 | fail: | |
61007b31 SH |
1929 | |
1930 | if (use_local_qiov) { | |
1931 | qemu_iovec_destroy(&local_qiov); | |
1932 | } | |
1933 | qemu_vfree(head_buf); | |
1934 | qemu_vfree(tail_buf); | |
9eeb6dd1 FZ |
1935 | out: |
1936 | tracked_request_end(&req); | |
99723548 | 1937 | bdrv_dec_in_flight(bs); |
61007b31 SH |
1938 | return ret; |
1939 | } | |
1940 | ||
a03ef88f | 1941 | int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, |
f5a5ca79 | 1942 | int bytes, BdrvRequestFlags flags) |
61007b31 | 1943 | { |
f5a5ca79 | 1944 | trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags); |
61007b31 | 1945 | |
a03ef88f | 1946 | if (!(child->bs->open_flags & BDRV_O_UNMAP)) { |
61007b31 SH |
1947 | flags &= ~BDRV_REQ_MAY_UNMAP; |
1948 | } | |
61007b31 | 1949 | |
f5a5ca79 | 1950 | return bdrv_co_pwritev(child, offset, bytes, NULL, |
74021bc4 | 1951 | BDRV_REQ_ZERO_WRITE | flags); |
61007b31 SH |
1952 | } |
1953 | ||
4085f5c7 JS |
1954 | /* |
1955 | * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not. | |
1956 | */ | |
1957 | int bdrv_flush_all(void) | |
1958 | { | |
1959 | BdrvNextIterator it; | |
1960 | BlockDriverState *bs = NULL; | |
1961 | int result = 0; | |
1962 | ||
1963 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { | |
1964 | AioContext *aio_context = bdrv_get_aio_context(bs); | |
1965 | int ret; | |
1966 | ||
1967 | aio_context_acquire(aio_context); | |
1968 | ret = bdrv_flush(bs); | |
1969 | if (ret < 0 && !result) { | |
1970 | result = ret; | |
1971 | } | |
1972 | aio_context_release(aio_context); | |
1973 | } | |
1974 | ||
1975 | return result; | |
1976 | } | |
1977 | ||
1978 | ||
4bcd936e | 1979 | typedef struct BdrvCoBlockStatusData { |
61007b31 SH |
1980 | BlockDriverState *bs; |
1981 | BlockDriverState *base; | |
c9ce8c4d | 1982 | bool want_zero; |
4bcd936e EB |
1983 | int64_t offset; |
1984 | int64_t bytes; | |
1985 | int64_t *pnum; | |
1986 | int64_t *map; | |
c9ce8c4d | 1987 | BlockDriverState **file; |
4bcd936e | 1988 | int ret; |
61007b31 | 1989 | bool done; |
4bcd936e | 1990 | } BdrvCoBlockStatusData; |
61007b31 | 1991 | |
3e4d0e72 EB |
1992 | int coroutine_fn bdrv_co_block_status_from_file(BlockDriverState *bs, |
1993 | bool want_zero, | |
1994 | int64_t offset, | |
1995 | int64_t bytes, | |
1996 | int64_t *pnum, | |
1997 | int64_t *map, | |
1998 | BlockDriverState **file) | |
f7cc69b3 MP |
1999 | { |
2000 | assert(bs->file && bs->file->bs); | |
3e4d0e72 EB |
2001 | *pnum = bytes; |
2002 | *map = offset; | |
f7cc69b3 | 2003 | *file = bs->file->bs; |
3e4d0e72 | 2004 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; |
f7cc69b3 MP |
2005 | } |
2006 | ||
3e4d0e72 EB |
2007 | int coroutine_fn bdrv_co_block_status_from_backing(BlockDriverState *bs, |
2008 | bool want_zero, | |
2009 | int64_t offset, | |
2010 | int64_t bytes, | |
2011 | int64_t *pnum, | |
2012 | int64_t *map, | |
2013 | BlockDriverState **file) | |
f7cc69b3 MP |
2014 | { |
2015 | assert(bs->backing && bs->backing->bs); | |
3e4d0e72 EB |
2016 | *pnum = bytes; |
2017 | *map = offset; | |
f7cc69b3 | 2018 | *file = bs->backing->bs; |
3e4d0e72 | 2019 | return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID; |
f7cc69b3 MP |
2020 | } |
2021 | ||
61007b31 SH |
2022 | /* |
2023 | * Returns the allocation status of the specified sectors. | |
2024 | * Drivers not implementing the functionality are assumed to not support | |
2025 | * backing files, hence all their sectors are reported as allocated. | |
2026 | * | |
86a3d5c6 EB |
2027 | * If 'want_zero' is true, the caller is querying for mapping |
2028 | * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and | |
2029 | * _ZERO where possible; otherwise, the result favors larger 'pnum', | |
2030 | * with a focus on accurate BDRV_BLOCK_ALLOCATED. | |
c9ce8c4d | 2031 | * |
2e8bc787 | 2032 | * If 'offset' is beyond the end of the disk image the return value is |
fb0d8654 | 2033 | * BDRV_BLOCK_EOF and 'pnum' is set to 0. |
61007b31 | 2034 | * |
2e8bc787 | 2035 | * 'bytes' is the max value 'pnum' should be set to. If bytes goes |
fb0d8654 EB |
2036 | * beyond the end of the disk image it will be clamped; if 'pnum' is set to |
2037 | * the end of the image, then the returned value will include BDRV_BLOCK_EOF. | |
67a0fd2a | 2038 | * |
2e8bc787 EB |
2039 | * 'pnum' is set to the number of bytes (including and immediately |
2040 | * following the specified offset) that are easily known to be in the | |
2041 | * same allocated/unallocated state. Note that a second call starting | |
2042 | * at the original offset plus returned pnum may have the same status. | |
2043 | * The returned value is non-zero on success except at end-of-file. | |
2044 | * | |
2045 | * Returns negative errno on failure. Otherwise, if the | |
2046 | * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are | |
2047 | * set to the host mapping and BDS corresponding to the guest offset. | |
61007b31 | 2048 | */ |
2e8bc787 EB |
2049 | static int coroutine_fn bdrv_co_block_status(BlockDriverState *bs, |
2050 | bool want_zero, | |
2051 | int64_t offset, int64_t bytes, | |
2052 | int64_t *pnum, int64_t *map, | |
2053 | BlockDriverState **file) | |
2054 | { | |
2055 | int64_t total_size; | |
2056 | int64_t n; /* bytes */ | |
efa6e2ed | 2057 | int ret; |
2e8bc787 | 2058 | int64_t local_map = 0; |
298a1665 | 2059 | BlockDriverState *local_file = NULL; |
efa6e2ed EB |
2060 | int64_t aligned_offset, aligned_bytes; |
2061 | uint32_t align; | |
61007b31 | 2062 | |
298a1665 EB |
2063 | assert(pnum); |
2064 | *pnum = 0; | |
2e8bc787 EB |
2065 | total_size = bdrv_getlength(bs); |
2066 | if (total_size < 0) { | |
2067 | ret = total_size; | |
298a1665 | 2068 | goto early_out; |
61007b31 SH |
2069 | } |
2070 | ||
2e8bc787 | 2071 | if (offset >= total_size) { |
298a1665 EB |
2072 | ret = BDRV_BLOCK_EOF; |
2073 | goto early_out; | |
61007b31 | 2074 | } |
2e8bc787 | 2075 | if (!bytes) { |
298a1665 EB |
2076 | ret = 0; |
2077 | goto early_out; | |
9cdcfd9f | 2078 | } |
61007b31 | 2079 | |
2e8bc787 EB |
2080 | n = total_size - offset; |
2081 | if (n < bytes) { | |
2082 | bytes = n; | |
61007b31 SH |
2083 | } |
2084 | ||
d470ad42 HR |
2085 | /* Must be non-NULL or bdrv_getlength() would have failed */ |
2086 | assert(bs->drv); | |
636cb512 | 2087 | if (!bs->drv->bdrv_co_block_status) { |
2e8bc787 | 2088 | *pnum = bytes; |
61007b31 | 2089 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED; |
2e8bc787 | 2090 | if (offset + bytes == total_size) { |
fb0d8654 EB |
2091 | ret |= BDRV_BLOCK_EOF; |
2092 | } | |
61007b31 | 2093 | if (bs->drv->protocol_name) { |
2e8bc787 EB |
2094 | ret |= BDRV_BLOCK_OFFSET_VALID; |
2095 | local_map = offset; | |
298a1665 | 2096 | local_file = bs; |
61007b31 | 2097 | } |
298a1665 | 2098 | goto early_out; |
61007b31 SH |
2099 | } |
2100 | ||
99723548 | 2101 | bdrv_inc_in_flight(bs); |
efa6e2ed EB |
2102 | |
2103 | /* Round out to request_alignment boundaries */ | |
86a3d5c6 | 2104 | align = bs->bl.request_alignment; |
efa6e2ed EB |
2105 | aligned_offset = QEMU_ALIGN_DOWN(offset, align); |
2106 | aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset; | |
2107 | ||
636cb512 EB |
2108 | ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset, |
2109 | aligned_bytes, pnum, &local_map, | |
2110 | &local_file); | |
2111 | if (ret < 0) { | |
2112 | *pnum = 0; | |
2113 | goto out; | |
efa6e2ed EB |
2114 | } |
2115 | ||
2e8bc787 | 2116 | /* |
636cb512 | 2117 | * The driver's result must be a non-zero multiple of request_alignment. |
efa6e2ed | 2118 | * Clamp pnum and adjust map to original request. |
2e8bc787 | 2119 | */ |
636cb512 EB |
2120 | assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) && |
2121 | align > offset - aligned_offset); | |
efa6e2ed EB |
2122 | *pnum -= offset - aligned_offset; |
2123 | if (*pnum > bytes) { | |
2124 | *pnum = bytes; | |
61007b31 | 2125 | } |
2e8bc787 | 2126 | if (ret & BDRV_BLOCK_OFFSET_VALID) { |
efa6e2ed | 2127 | local_map += offset - aligned_offset; |
2e8bc787 | 2128 | } |
61007b31 SH |
2129 | |
2130 | if (ret & BDRV_BLOCK_RAW) { | |
298a1665 | 2131 | assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file); |
2e8bc787 EB |
2132 | ret = bdrv_co_block_status(local_file, want_zero, local_map, |
2133 | *pnum, pnum, &local_map, &local_file); | |
99723548 | 2134 | goto out; |
61007b31 SH |
2135 | } |
2136 | ||
2137 | if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) { | |
2138 | ret |= BDRV_BLOCK_ALLOCATED; | |
c9ce8c4d | 2139 | } else if (want_zero) { |
61007b31 SH |
2140 | if (bdrv_unallocated_blocks_are_zero(bs)) { |
2141 | ret |= BDRV_BLOCK_ZERO; | |
760e0063 KW |
2142 | } else if (bs->backing) { |
2143 | BlockDriverState *bs2 = bs->backing->bs; | |
2e8bc787 | 2144 | int64_t size2 = bdrv_getlength(bs2); |
c9ce8c4d | 2145 | |
2e8bc787 | 2146 | if (size2 >= 0 && offset >= size2) { |
61007b31 SH |
2147 | ret |= BDRV_BLOCK_ZERO; |
2148 | } | |
2149 | } | |
2150 | } | |
2151 | ||
c9ce8c4d | 2152 | if (want_zero && local_file && local_file != bs && |
61007b31 SH |
2153 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && |
2154 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
2e8bc787 EB |
2155 | int64_t file_pnum; |
2156 | int ret2; | |
61007b31 | 2157 | |
2e8bc787 EB |
2158 | ret2 = bdrv_co_block_status(local_file, want_zero, local_map, |
2159 | *pnum, &file_pnum, NULL, NULL); | |
61007b31 SH |
2160 | if (ret2 >= 0) { |
2161 | /* Ignore errors. This is just providing extra information, it | |
2162 | * is useful but not necessary. | |
2163 | */ | |
c61e684e EB |
2164 | if (ret2 & BDRV_BLOCK_EOF && |
2165 | (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) { | |
2166 | /* | |
2167 | * It is valid for the format block driver to read | |
2168 | * beyond the end of the underlying file's current | |
2169 | * size; such areas read as zero. | |
2170 | */ | |
61007b31 SH |
2171 | ret |= BDRV_BLOCK_ZERO; |
2172 | } else { | |
2173 | /* Limit request to the range reported by the protocol driver */ | |
2174 | *pnum = file_pnum; | |
2175 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
2176 | } | |
2177 | } | |
2178 | } | |
2179 | ||
99723548 PB |
2180 | out: |
2181 | bdrv_dec_in_flight(bs); | |
2e8bc787 | 2182 | if (ret >= 0 && offset + *pnum == total_size) { |
fb0d8654 EB |
2183 | ret |= BDRV_BLOCK_EOF; |
2184 | } | |
298a1665 EB |
2185 | early_out: |
2186 | if (file) { | |
2187 | *file = local_file; | |
2188 | } | |
2e8bc787 EB |
2189 | if (map) { |
2190 | *map = local_map; | |
2191 | } | |
61007b31 SH |
2192 | return ret; |
2193 | } | |
2194 | ||
5b648c67 EB |
2195 | static int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs, |
2196 | BlockDriverState *base, | |
2197 | bool want_zero, | |
2198 | int64_t offset, | |
2199 | int64_t bytes, | |
2200 | int64_t *pnum, | |
2201 | int64_t *map, | |
2202 | BlockDriverState **file) | |
ba3f0e25 FZ |
2203 | { |
2204 | BlockDriverState *p; | |
5b648c67 | 2205 | int ret = 0; |
c61e684e | 2206 | bool first = true; |
ba3f0e25 FZ |
2207 | |
2208 | assert(bs != base); | |
760e0063 | 2209 | for (p = bs; p != base; p = backing_bs(p)) { |
5b648c67 EB |
2210 | ret = bdrv_co_block_status(p, want_zero, offset, bytes, pnum, map, |
2211 | file); | |
c61e684e EB |
2212 | if (ret < 0) { |
2213 | break; | |
2214 | } | |
2215 | if (ret & BDRV_BLOCK_ZERO && ret & BDRV_BLOCK_EOF && !first) { | |
2216 | /* | |
2217 | * Reading beyond the end of the file continues to read | |
2218 | * zeroes, but we can only widen the result to the | |
2219 | * unallocated length we learned from an earlier | |
2220 | * iteration. | |
2221 | */ | |
5b648c67 | 2222 | *pnum = bytes; |
c61e684e EB |
2223 | } |
2224 | if (ret & (BDRV_BLOCK_ZERO | BDRV_BLOCK_DATA)) { | |
ba3f0e25 FZ |
2225 | break; |
2226 | } | |
5b648c67 EB |
2227 | /* [offset, pnum] unallocated on this layer, which could be only |
2228 | * the first part of [offset, bytes]. */ | |
2229 | bytes = MIN(bytes, *pnum); | |
c61e684e | 2230 | first = false; |
ba3f0e25 FZ |
2231 | } |
2232 | return ret; | |
2233 | } | |
2234 | ||
31826642 | 2235 | /* Coroutine wrapper for bdrv_block_status_above() */ |
5b648c67 | 2236 | static void coroutine_fn bdrv_block_status_above_co_entry(void *opaque) |
61007b31 | 2237 | { |
4bcd936e | 2238 | BdrvCoBlockStatusData *data = opaque; |
61007b31 | 2239 | |
5b648c67 EB |
2240 | data->ret = bdrv_co_block_status_above(data->bs, data->base, |
2241 | data->want_zero, | |
2242 | data->offset, data->bytes, | |
2243 | data->pnum, data->map, data->file); | |
61007b31 | 2244 | data->done = true; |
4720cbee | 2245 | aio_wait_kick(); |
61007b31 SH |
2246 | } |
2247 | ||
2248 | /* | |
5b648c67 | 2249 | * Synchronous wrapper around bdrv_co_block_status_above(). |
61007b31 | 2250 | * |
5b648c67 | 2251 | * See bdrv_co_block_status_above() for details. |
61007b31 | 2252 | */ |
7ddb99b9 EB |
2253 | static int bdrv_common_block_status_above(BlockDriverState *bs, |
2254 | BlockDriverState *base, | |
2255 | bool want_zero, int64_t offset, | |
2256 | int64_t bytes, int64_t *pnum, | |
2257 | int64_t *map, | |
2258 | BlockDriverState **file) | |
61007b31 SH |
2259 | { |
2260 | Coroutine *co; | |
4bcd936e | 2261 | BdrvCoBlockStatusData data = { |
61007b31 | 2262 | .bs = bs, |
ba3f0e25 | 2263 | .base = base, |
c9ce8c4d | 2264 | .want_zero = want_zero, |
7ddb99b9 EB |
2265 | .offset = offset, |
2266 | .bytes = bytes, | |
2267 | .pnum = pnum, | |
2268 | .map = map, | |
c9ce8c4d | 2269 | .file = file, |
61007b31 SH |
2270 | .done = false, |
2271 | }; | |
2272 | ||
2273 | if (qemu_in_coroutine()) { | |
2274 | /* Fast-path if already in coroutine context */ | |
5b648c67 | 2275 | bdrv_block_status_above_co_entry(&data); |
61007b31 | 2276 | } else { |
5b648c67 | 2277 | co = qemu_coroutine_create(bdrv_block_status_above_co_entry, &data); |
e92f0e19 | 2278 | bdrv_coroutine_enter(bs, co); |
88b062c2 | 2279 | BDRV_POLL_WHILE(bs, !data.done); |
61007b31 | 2280 | } |
7ddb99b9 | 2281 | return data.ret; |
61007b31 SH |
2282 | } |
2283 | ||
31826642 EB |
2284 | int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base, |
2285 | int64_t offset, int64_t bytes, int64_t *pnum, | |
2286 | int64_t *map, BlockDriverState **file) | |
c9ce8c4d | 2287 | { |
31826642 EB |
2288 | return bdrv_common_block_status_above(bs, base, true, offset, bytes, |
2289 | pnum, map, file); | |
c9ce8c4d EB |
2290 | } |
2291 | ||
237d78f8 EB |
2292 | int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2293 | int64_t *pnum, int64_t *map, BlockDriverState **file) | |
ba3f0e25 | 2294 | { |
31826642 EB |
2295 | return bdrv_block_status_above(bs, backing_bs(bs), |
2296 | offset, bytes, pnum, map, file); | |
ba3f0e25 FZ |
2297 | } |
2298 | ||
d6a644bb EB |
2299 | int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t offset, |
2300 | int64_t bytes, int64_t *pnum) | |
61007b31 | 2301 | { |
7ddb99b9 EB |
2302 | int ret; |
2303 | int64_t dummy; | |
d6a644bb | 2304 | |
7ddb99b9 EB |
2305 | ret = bdrv_common_block_status_above(bs, backing_bs(bs), false, offset, |
2306 | bytes, pnum ? pnum : &dummy, NULL, | |
c9ce8c4d | 2307 | NULL); |
61007b31 SH |
2308 | if (ret < 0) { |
2309 | return ret; | |
2310 | } | |
2311 | return !!(ret & BDRV_BLOCK_ALLOCATED); | |
2312 | } | |
2313 | ||
2314 | /* | |
2315 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
2316 | * | |
51b0a488 EB |
2317 | * Return true if (a prefix of) the given range is allocated in any image |
2318 | * between BASE and TOP (inclusive). BASE can be NULL to check if the given | |
2319 | * offset is allocated in any image of the chain. Return false otherwise, | |
d6a644bb | 2320 | * or negative errno on failure. |
61007b31 | 2321 | * |
51b0a488 EB |
2322 | * 'pnum' is set to the number of bytes (including and immediately |
2323 | * following the specified offset) that are known to be in the same | |
2324 | * allocated/unallocated state. Note that a subsequent call starting | |
2325 | * at 'offset + *pnum' may return the same allocation status (in other | |
2326 | * words, the result is not necessarily the maximum possible range); | |
2327 | * but 'pnum' will only be 0 when end of file is reached. | |
61007b31 SH |
2328 | * |
2329 | */ | |
2330 | int bdrv_is_allocated_above(BlockDriverState *top, | |
2331 | BlockDriverState *base, | |
51b0a488 | 2332 | int64_t offset, int64_t bytes, int64_t *pnum) |
61007b31 SH |
2333 | { |
2334 | BlockDriverState *intermediate; | |
51b0a488 EB |
2335 | int ret; |
2336 | int64_t n = bytes; | |
61007b31 SH |
2337 | |
2338 | intermediate = top; | |
2339 | while (intermediate && intermediate != base) { | |
d6a644bb | 2340 | int64_t pnum_inter; |
c00716be | 2341 | int64_t size_inter; |
d6a644bb | 2342 | |
51b0a488 | 2343 | ret = bdrv_is_allocated(intermediate, offset, bytes, &pnum_inter); |
61007b31 SH |
2344 | if (ret < 0) { |
2345 | return ret; | |
d6a644bb | 2346 | } |
d6a644bb | 2347 | if (ret) { |
51b0a488 | 2348 | *pnum = pnum_inter; |
61007b31 SH |
2349 | return 1; |
2350 | } | |
2351 | ||
51b0a488 | 2352 | size_inter = bdrv_getlength(intermediate); |
c00716be EB |
2353 | if (size_inter < 0) { |
2354 | return size_inter; | |
2355 | } | |
51b0a488 EB |
2356 | if (n > pnum_inter && |
2357 | (intermediate == top || offset + pnum_inter < size_inter)) { | |
2358 | n = pnum_inter; | |
61007b31 SH |
2359 | } |
2360 | ||
760e0063 | 2361 | intermediate = backing_bs(intermediate); |
61007b31 SH |
2362 | } |
2363 | ||
2364 | *pnum = n; | |
2365 | return 0; | |
2366 | } | |
2367 | ||
1a8ae822 KW |
2368 | typedef struct BdrvVmstateCo { |
2369 | BlockDriverState *bs; | |
2370 | QEMUIOVector *qiov; | |
2371 | int64_t pos; | |
2372 | bool is_read; | |
2373 | int ret; | |
2374 | } BdrvVmstateCo; | |
2375 | ||
2376 | static int coroutine_fn | |
2377 | bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
2378 | bool is_read) | |
2379 | { | |
2380 | BlockDriver *drv = bs->drv; | |
dc88a467 SH |
2381 | int ret = -ENOTSUP; |
2382 | ||
2383 | bdrv_inc_in_flight(bs); | |
1a8ae822 KW |
2384 | |
2385 | if (!drv) { | |
dc88a467 | 2386 | ret = -ENOMEDIUM; |
1a8ae822 | 2387 | } else if (drv->bdrv_load_vmstate) { |
dc88a467 SH |
2388 | if (is_read) { |
2389 | ret = drv->bdrv_load_vmstate(bs, qiov, pos); | |
2390 | } else { | |
2391 | ret = drv->bdrv_save_vmstate(bs, qiov, pos); | |
2392 | } | |
1a8ae822 | 2393 | } else if (bs->file) { |
dc88a467 | 2394 | ret = bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read); |
1a8ae822 KW |
2395 | } |
2396 | ||
dc88a467 SH |
2397 | bdrv_dec_in_flight(bs); |
2398 | return ret; | |
1a8ae822 KW |
2399 | } |
2400 | ||
2401 | static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque) | |
2402 | { | |
2403 | BdrvVmstateCo *co = opaque; | |
2404 | co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read); | |
4720cbee | 2405 | aio_wait_kick(); |
1a8ae822 KW |
2406 | } |
2407 | ||
2408 | static inline int | |
2409 | bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos, | |
2410 | bool is_read) | |
2411 | { | |
2412 | if (qemu_in_coroutine()) { | |
2413 | return bdrv_co_rw_vmstate(bs, qiov, pos, is_read); | |
2414 | } else { | |
2415 | BdrvVmstateCo data = { | |
2416 | .bs = bs, | |
2417 | .qiov = qiov, | |
2418 | .pos = pos, | |
2419 | .is_read = is_read, | |
2420 | .ret = -EINPROGRESS, | |
2421 | }; | |
0b8b8753 | 2422 | Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data); |
1a8ae822 | 2423 | |
e92f0e19 | 2424 | bdrv_coroutine_enter(bs, co); |
ea17c9d2 | 2425 | BDRV_POLL_WHILE(bs, data.ret == -EINPROGRESS); |
1a8ae822 KW |
2426 | return data.ret; |
2427 | } | |
2428 | } | |
2429 | ||
61007b31 SH |
2430 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, |
2431 | int64_t pos, int size) | |
2432 | { | |
0d93ed08 | 2433 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b433d942 | 2434 | int ret; |
61007b31 | 2435 | |
b433d942 KW |
2436 | ret = bdrv_writev_vmstate(bs, &qiov, pos); |
2437 | if (ret < 0) { | |
2438 | return ret; | |
2439 | } | |
2440 | ||
2441 | return size; | |
61007b31 SH |
2442 | } |
2443 | ||
2444 | int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
2445 | { | |
1a8ae822 | 2446 | return bdrv_rw_vmstate(bs, qiov, pos, false); |
61007b31 SH |
2447 | } |
2448 | ||
2449 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
2450 | int64_t pos, int size) | |
5ddda0b8 | 2451 | { |
0d93ed08 | 2452 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size); |
b433d942 | 2453 | int ret; |
5ddda0b8 | 2454 | |
b433d942 KW |
2455 | ret = bdrv_readv_vmstate(bs, &qiov, pos); |
2456 | if (ret < 0) { | |
2457 | return ret; | |
2458 | } | |
2459 | ||
2460 | return size; | |
5ddda0b8 KW |
2461 | } |
2462 | ||
2463 | int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
61007b31 | 2464 | { |
1a8ae822 | 2465 | return bdrv_rw_vmstate(bs, qiov, pos, true); |
61007b31 SH |
2466 | } |
2467 | ||
2468 | /**************************************************************/ | |
2469 | /* async I/Os */ | |
2470 | ||
61007b31 SH |
2471 | void bdrv_aio_cancel(BlockAIOCB *acb) |
2472 | { | |
2473 | qemu_aio_ref(acb); | |
2474 | bdrv_aio_cancel_async(acb); | |
2475 | while (acb->refcnt > 1) { | |
2476 | if (acb->aiocb_info->get_aio_context) { | |
2477 | aio_poll(acb->aiocb_info->get_aio_context(acb), true); | |
2478 | } else if (acb->bs) { | |
2f47da5f PB |
2479 | /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so |
2480 | * assert that we're not using an I/O thread. Thread-safe | |
2481 | * code should use bdrv_aio_cancel_async exclusively. | |
2482 | */ | |
2483 | assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context()); | |
61007b31 SH |
2484 | aio_poll(bdrv_get_aio_context(acb->bs), true); |
2485 | } else { | |
2486 | abort(); | |
2487 | } | |
2488 | } | |
2489 | qemu_aio_unref(acb); | |
2490 | } | |
2491 | ||
2492 | /* Async version of aio cancel. The caller is not blocked if the acb implements | |
2493 | * cancel_async, otherwise we do nothing and let the request normally complete. | |
2494 | * In either case the completion callback must be called. */ | |
2495 | void bdrv_aio_cancel_async(BlockAIOCB *acb) | |
2496 | { | |
2497 | if (acb->aiocb_info->cancel_async) { | |
2498 | acb->aiocb_info->cancel_async(acb); | |
2499 | } | |
2500 | } | |
2501 | ||
61007b31 SH |
2502 | /**************************************************************/ |
2503 | /* Coroutine block device emulation */ | |
2504 | ||
e293b7a3 KW |
2505 | typedef struct FlushCo { |
2506 | BlockDriverState *bs; | |
2507 | int ret; | |
2508 | } FlushCo; | |
2509 | ||
2510 | ||
61007b31 SH |
2511 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) |
2512 | { | |
e293b7a3 | 2513 | FlushCo *rwco = opaque; |
61007b31 SH |
2514 | |
2515 | rwco->ret = bdrv_co_flush(rwco->bs); | |
4720cbee | 2516 | aio_wait_kick(); |
61007b31 SH |
2517 | } |
2518 | ||
2519 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
2520 | { | |
49ca6259 FZ |
2521 | int current_gen; |
2522 | int ret = 0; | |
2523 | ||
2524 | bdrv_inc_in_flight(bs); | |
61007b31 | 2525 | |
e914404e | 2526 | if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs) || |
1b6bc94d | 2527 | bdrv_is_sg(bs)) { |
49ca6259 | 2528 | goto early_exit; |
61007b31 SH |
2529 | } |
2530 | ||
3783fa3d | 2531 | qemu_co_mutex_lock(&bs->reqs_lock); |
47fec599 | 2532 | current_gen = atomic_read(&bs->write_gen); |
3ff2f67a EY |
2533 | |
2534 | /* Wait until any previous flushes are completed */ | |
99723548 | 2535 | while (bs->active_flush_req) { |
3783fa3d | 2536 | qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock); |
3ff2f67a EY |
2537 | } |
2538 | ||
3783fa3d | 2539 | /* Flushes reach this point in nondecreasing current_gen order. */ |
99723548 | 2540 | bs->active_flush_req = true; |
3783fa3d | 2541 | qemu_co_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2542 | |
c32b82af PD |
2543 | /* Write back all layers by calling one driver function */ |
2544 | if (bs->drv->bdrv_co_flush) { | |
2545 | ret = bs->drv->bdrv_co_flush(bs); | |
2546 | goto out; | |
2547 | } | |
2548 | ||
61007b31 SH |
2549 | /* Write back cached data to the OS even with cache=unsafe */ |
2550 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); | |
2551 | if (bs->drv->bdrv_co_flush_to_os) { | |
2552 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
2553 | if (ret < 0) { | |
cdb5e315 | 2554 | goto out; |
61007b31 SH |
2555 | } |
2556 | } | |
2557 | ||
2558 | /* But don't actually force it to the disk with cache=unsafe */ | |
2559 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
2560 | goto flush_parent; | |
2561 | } | |
2562 | ||
3ff2f67a EY |
2563 | /* Check if we really need to flush anything */ |
2564 | if (bs->flushed_gen == current_gen) { | |
2565 | goto flush_parent; | |
2566 | } | |
2567 | ||
61007b31 | 2568 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); |
d470ad42 HR |
2569 | if (!bs->drv) { |
2570 | /* bs->drv->bdrv_co_flush() might have ejected the BDS | |
2571 | * (even in case of apparent success) */ | |
2572 | ret = -ENOMEDIUM; | |
2573 | goto out; | |
2574 | } | |
61007b31 SH |
2575 | if (bs->drv->bdrv_co_flush_to_disk) { |
2576 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
2577 | } else if (bs->drv->bdrv_aio_flush) { | |
2578 | BlockAIOCB *acb; | |
2579 | CoroutineIOCompletion co = { | |
2580 | .coroutine = qemu_coroutine_self(), | |
2581 | }; | |
2582 | ||
2583 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
2584 | if (acb == NULL) { | |
2585 | ret = -EIO; | |
2586 | } else { | |
2587 | qemu_coroutine_yield(); | |
2588 | ret = co.ret; | |
2589 | } | |
2590 | } else { | |
2591 | /* | |
2592 | * Some block drivers always operate in either writethrough or unsafe | |
2593 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
2594 | * know how the server works (because the behaviour is hardcoded or | |
2595 | * depends on server-side configuration), so we can't ensure that | |
2596 | * everything is safe on disk. Returning an error doesn't work because | |
2597 | * that would break guests even if the server operates in writethrough | |
2598 | * mode. | |
2599 | * | |
2600 | * Let's hope the user knows what he's doing. | |
2601 | */ | |
2602 | ret = 0; | |
2603 | } | |
3ff2f67a | 2604 | |
61007b31 | 2605 | if (ret < 0) { |
cdb5e315 | 2606 | goto out; |
61007b31 SH |
2607 | } |
2608 | ||
2609 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
2610 | * in the case of cache=unsafe, so there are no useless flushes. | |
2611 | */ | |
2612 | flush_parent: | |
cdb5e315 FZ |
2613 | ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0; |
2614 | out: | |
3ff2f67a | 2615 | /* Notify any pending flushes that we have completed */ |
e6af1e08 KW |
2616 | if (ret == 0) { |
2617 | bs->flushed_gen = current_gen; | |
2618 | } | |
3783fa3d PB |
2619 | |
2620 | qemu_co_mutex_lock(&bs->reqs_lock); | |
99723548 | 2621 | bs->active_flush_req = false; |
156af3ac DL |
2622 | /* Return value is ignored - it's ok if wait queue is empty */ |
2623 | qemu_co_queue_next(&bs->flush_queue); | |
3783fa3d | 2624 | qemu_co_mutex_unlock(&bs->reqs_lock); |
3ff2f67a | 2625 | |
49ca6259 | 2626 | early_exit: |
99723548 | 2627 | bdrv_dec_in_flight(bs); |
cdb5e315 | 2628 | return ret; |
61007b31 SH |
2629 | } |
2630 | ||
2631 | int bdrv_flush(BlockDriverState *bs) | |
2632 | { | |
2633 | Coroutine *co; | |
e293b7a3 | 2634 | FlushCo flush_co = { |
61007b31 SH |
2635 | .bs = bs, |
2636 | .ret = NOT_DONE, | |
2637 | }; | |
2638 | ||
2639 | if (qemu_in_coroutine()) { | |
2640 | /* Fast-path if already in coroutine context */ | |
e293b7a3 | 2641 | bdrv_flush_co_entry(&flush_co); |
61007b31 | 2642 | } else { |
0b8b8753 | 2643 | co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co); |
e92f0e19 | 2644 | bdrv_coroutine_enter(bs, co); |
88b062c2 | 2645 | BDRV_POLL_WHILE(bs, flush_co.ret == NOT_DONE); |
61007b31 SH |
2646 | } |
2647 | ||
e293b7a3 | 2648 | return flush_co.ret; |
61007b31 SH |
2649 | } |
2650 | ||
2651 | typedef struct DiscardCo { | |
0b9fd3f4 | 2652 | BdrvChild *child; |
0c51a893 | 2653 | int64_t offset; |
f5a5ca79 | 2654 | int bytes; |
61007b31 SH |
2655 | int ret; |
2656 | } DiscardCo; | |
0c51a893 | 2657 | static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque) |
61007b31 SH |
2658 | { |
2659 | DiscardCo *rwco = opaque; | |
2660 | ||
0b9fd3f4 | 2661 | rwco->ret = bdrv_co_pdiscard(rwco->child, rwco->offset, rwco->bytes); |
4720cbee | 2662 | aio_wait_kick(); |
61007b31 SH |
2663 | } |
2664 | ||
0b9fd3f4 | 2665 | int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, int bytes) |
61007b31 | 2666 | { |
b1066c87 | 2667 | BdrvTrackedRequest req; |
9f1963b3 | 2668 | int max_pdiscard, ret; |
3482b9bc | 2669 | int head, tail, align; |
0b9fd3f4 | 2670 | BlockDriverState *bs = child->bs; |
61007b31 | 2671 | |
0b9fd3f4 | 2672 | if (!bs || !bs->drv) { |
61007b31 SH |
2673 | return -ENOMEDIUM; |
2674 | } | |
2675 | ||
d6883bc9 VSO |
2676 | if (bdrv_has_readonly_bitmaps(bs)) { |
2677 | return -EPERM; | |
2678 | } | |
2679 | ||
f5a5ca79 | 2680 | ret = bdrv_check_byte_request(bs, offset, bytes); |
61007b31 SH |
2681 | if (ret < 0) { |
2682 | return ret; | |
61007b31 SH |
2683 | } |
2684 | ||
61007b31 SH |
2685 | /* Do nothing if disabled. */ |
2686 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
2687 | return 0; | |
2688 | } | |
2689 | ||
02aefe43 | 2690 | if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) { |
61007b31 SH |
2691 | return 0; |
2692 | } | |
2693 | ||
3482b9bc EB |
2694 | /* Discard is advisory, but some devices track and coalesce |
2695 | * unaligned requests, so we must pass everything down rather than | |
2696 | * round here. Still, most devices will just silently ignore | |
2697 | * unaligned requests (by returning -ENOTSUP), so we must fragment | |
2698 | * the request accordingly. */ | |
02aefe43 | 2699 | align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment); |
b8d0a980 EB |
2700 | assert(align % bs->bl.request_alignment == 0); |
2701 | head = offset % align; | |
f5a5ca79 | 2702 | tail = (offset + bytes) % align; |
9f1963b3 | 2703 | |
99723548 | 2704 | bdrv_inc_in_flight(bs); |
f5a5ca79 | 2705 | tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD); |
50824995 | 2706 | |
00695c27 | 2707 | ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0); |
ec050f77 DL |
2708 | if (ret < 0) { |
2709 | goto out; | |
2710 | } | |
2711 | ||
9f1963b3 EB |
2712 | max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX), |
2713 | align); | |
3482b9bc | 2714 | assert(max_pdiscard >= bs->bl.request_alignment); |
61007b31 | 2715 | |
f5a5ca79 | 2716 | while (bytes > 0) { |
f5a5ca79 | 2717 | int num = bytes; |
3482b9bc EB |
2718 | |
2719 | if (head) { | |
2720 | /* Make small requests to get to alignment boundaries. */ | |
f5a5ca79 | 2721 | num = MIN(bytes, align - head); |
3482b9bc EB |
2722 | if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) { |
2723 | num %= bs->bl.request_alignment; | |
2724 | } | |
2725 | head = (head + num) % align; | |
2726 | assert(num < max_pdiscard); | |
2727 | } else if (tail) { | |
2728 | if (num > align) { | |
2729 | /* Shorten the request to the last aligned cluster. */ | |
2730 | num -= tail; | |
2731 | } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) && | |
2732 | tail > bs->bl.request_alignment) { | |
2733 | tail %= bs->bl.request_alignment; | |
2734 | num -= tail; | |
2735 | } | |
2736 | } | |
2737 | /* limit request size */ | |
2738 | if (num > max_pdiscard) { | |
2739 | num = max_pdiscard; | |
2740 | } | |
61007b31 | 2741 | |
d470ad42 HR |
2742 | if (!bs->drv) { |
2743 | ret = -ENOMEDIUM; | |
2744 | goto out; | |
2745 | } | |
47a5486d EB |
2746 | if (bs->drv->bdrv_co_pdiscard) { |
2747 | ret = bs->drv->bdrv_co_pdiscard(bs, offset, num); | |
61007b31 SH |
2748 | } else { |
2749 | BlockAIOCB *acb; | |
2750 | CoroutineIOCompletion co = { | |
2751 | .coroutine = qemu_coroutine_self(), | |
2752 | }; | |
2753 | ||
4da444a0 EB |
2754 | acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num, |
2755 | bdrv_co_io_em_complete, &co); | |
61007b31 | 2756 | if (acb == NULL) { |
b1066c87 FZ |
2757 | ret = -EIO; |
2758 | goto out; | |
61007b31 SH |
2759 | } else { |
2760 | qemu_coroutine_yield(); | |
2761 | ret = co.ret; | |
2762 | } | |
2763 | } | |
2764 | if (ret && ret != -ENOTSUP) { | |
b1066c87 | 2765 | goto out; |
61007b31 SH |
2766 | } |
2767 | ||
9f1963b3 | 2768 | offset += num; |
f5a5ca79 | 2769 | bytes -= num; |
61007b31 | 2770 | } |
b1066c87 FZ |
2771 | ret = 0; |
2772 | out: | |
00695c27 | 2773 | bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret); |
b1066c87 | 2774 | tracked_request_end(&req); |
99723548 | 2775 | bdrv_dec_in_flight(bs); |
b1066c87 | 2776 | return ret; |
61007b31 SH |
2777 | } |
2778 | ||
0b9fd3f4 | 2779 | int bdrv_pdiscard(BdrvChild *child, int64_t offset, int bytes) |
61007b31 SH |
2780 | { |
2781 | Coroutine *co; | |
2782 | DiscardCo rwco = { | |
0b9fd3f4 | 2783 | .child = child, |
0c51a893 | 2784 | .offset = offset, |
f5a5ca79 | 2785 | .bytes = bytes, |
61007b31 SH |
2786 | .ret = NOT_DONE, |
2787 | }; | |
2788 | ||
2789 | if (qemu_in_coroutine()) { | |
2790 | /* Fast-path if already in coroutine context */ | |
0c51a893 | 2791 | bdrv_pdiscard_co_entry(&rwco); |
61007b31 | 2792 | } else { |
0c51a893 | 2793 | co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco); |
0b9fd3f4 FZ |
2794 | bdrv_coroutine_enter(child->bs, co); |
2795 | BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE); | |
61007b31 SH |
2796 | } |
2797 | ||
2798 | return rwco.ret; | |
2799 | } | |
2800 | ||
48af776a | 2801 | int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf) |
61007b31 SH |
2802 | { |
2803 | BlockDriver *drv = bs->drv; | |
5c5ae76a FZ |
2804 | CoroutineIOCompletion co = { |
2805 | .coroutine = qemu_coroutine_self(), | |
2806 | }; | |
2807 | BlockAIOCB *acb; | |
61007b31 | 2808 | |
99723548 | 2809 | bdrv_inc_in_flight(bs); |
16a389dc | 2810 | if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) { |
5c5ae76a FZ |
2811 | co.ret = -ENOTSUP; |
2812 | goto out; | |
2813 | } | |
2814 | ||
16a389dc KW |
2815 | if (drv->bdrv_co_ioctl) { |
2816 | co.ret = drv->bdrv_co_ioctl(bs, req, buf); | |
2817 | } else { | |
2818 | acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co); | |
2819 | if (!acb) { | |
2820 | co.ret = -ENOTSUP; | |
2821 | goto out; | |
2822 | } | |
2823 | qemu_coroutine_yield(); | |
5c5ae76a | 2824 | } |
5c5ae76a | 2825 | out: |
99723548 | 2826 | bdrv_dec_in_flight(bs); |
5c5ae76a FZ |
2827 | return co.ret; |
2828 | } | |
2829 | ||
61007b31 SH |
2830 | void *qemu_blockalign(BlockDriverState *bs, size_t size) |
2831 | { | |
2832 | return qemu_memalign(bdrv_opt_mem_align(bs), size); | |
2833 | } | |
2834 | ||
2835 | void *qemu_blockalign0(BlockDriverState *bs, size_t size) | |
2836 | { | |
2837 | return memset(qemu_blockalign(bs, size), 0, size); | |
2838 | } | |
2839 | ||
2840 | void *qemu_try_blockalign(BlockDriverState *bs, size_t size) | |
2841 | { | |
2842 | size_t align = bdrv_opt_mem_align(bs); | |
2843 | ||
2844 | /* Ensure that NULL is never returned on success */ | |
2845 | assert(align > 0); | |
2846 | if (size == 0) { | |
2847 | size = align; | |
2848 | } | |
2849 | ||
2850 | return qemu_try_memalign(align, size); | |
2851 | } | |
2852 | ||
2853 | void *qemu_try_blockalign0(BlockDriverState *bs, size_t size) | |
2854 | { | |
2855 | void *mem = qemu_try_blockalign(bs, size); | |
2856 | ||
2857 | if (mem) { | |
2858 | memset(mem, 0, size); | |
2859 | } | |
2860 | ||
2861 | return mem; | |
2862 | } | |
2863 | ||
2864 | /* | |
2865 | * Check if all memory in this vector is sector aligned. | |
2866 | */ | |
2867 | bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
2868 | { | |
2869 | int i; | |
4196d2f0 | 2870 | size_t alignment = bdrv_min_mem_align(bs); |
61007b31 SH |
2871 | |
2872 | for (i = 0; i < qiov->niov; i++) { | |
2873 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
2874 | return false; | |
2875 | } | |
2876 | if (qiov->iov[i].iov_len % alignment) { | |
2877 | return false; | |
2878 | } | |
2879 | } | |
2880 | ||
2881 | return true; | |
2882 | } | |
2883 | ||
2884 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
2885 | NotifierWithReturn *notifier) | |
2886 | { | |
2887 | notifier_with_return_list_add(&bs->before_write_notifiers, notifier); | |
2888 | } | |
2889 | ||
2890 | void bdrv_io_plug(BlockDriverState *bs) | |
2891 | { | |
6b98bd64 PB |
2892 | BdrvChild *child; |
2893 | ||
2894 | QLIST_FOREACH(child, &bs->children, next) { | |
2895 | bdrv_io_plug(child->bs); | |
2896 | } | |
2897 | ||
850d54a2 | 2898 | if (atomic_fetch_inc(&bs->io_plugged) == 0) { |
6b98bd64 PB |
2899 | BlockDriver *drv = bs->drv; |
2900 | if (drv && drv->bdrv_io_plug) { | |
2901 | drv->bdrv_io_plug(bs); | |
2902 | } | |
61007b31 SH |
2903 | } |
2904 | } | |
2905 | ||
2906 | void bdrv_io_unplug(BlockDriverState *bs) | |
2907 | { | |
6b98bd64 PB |
2908 | BdrvChild *child; |
2909 | ||
2910 | assert(bs->io_plugged); | |
850d54a2 | 2911 | if (atomic_fetch_dec(&bs->io_plugged) == 1) { |
6b98bd64 PB |
2912 | BlockDriver *drv = bs->drv; |
2913 | if (drv && drv->bdrv_io_unplug) { | |
2914 | drv->bdrv_io_unplug(bs); | |
2915 | } | |
2916 | } | |
2917 | ||
2918 | QLIST_FOREACH(child, &bs->children, next) { | |
2919 | bdrv_io_unplug(child->bs); | |
61007b31 SH |
2920 | } |
2921 | } | |
23d0ba93 FZ |
2922 | |
2923 | void bdrv_register_buf(BlockDriverState *bs, void *host, size_t size) | |
2924 | { | |
2925 | BdrvChild *child; | |
2926 | ||
2927 | if (bs->drv && bs->drv->bdrv_register_buf) { | |
2928 | bs->drv->bdrv_register_buf(bs, host, size); | |
2929 | } | |
2930 | QLIST_FOREACH(child, &bs->children, next) { | |
2931 | bdrv_register_buf(child->bs, host, size); | |
2932 | } | |
2933 | } | |
2934 | ||
2935 | void bdrv_unregister_buf(BlockDriverState *bs, void *host) | |
2936 | { | |
2937 | BdrvChild *child; | |
2938 | ||
2939 | if (bs->drv && bs->drv->bdrv_unregister_buf) { | |
2940 | bs->drv->bdrv_unregister_buf(bs, host); | |
2941 | } | |
2942 | QLIST_FOREACH(child, &bs->children, next) { | |
2943 | bdrv_unregister_buf(child->bs, host); | |
2944 | } | |
2945 | } | |
fcc67678 | 2946 | |
67b51fb9 VSO |
2947 | static int coroutine_fn bdrv_co_copy_range_internal( |
2948 | BdrvChild *src, uint64_t src_offset, BdrvChild *dst, | |
2949 | uint64_t dst_offset, uint64_t bytes, | |
2950 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags, | |
2951 | bool recurse_src) | |
fcc67678 | 2952 | { |
999658a0 | 2953 | BdrvTrackedRequest req; |
fcc67678 FZ |
2954 | int ret; |
2955 | ||
d4d3e5a0 | 2956 | if (!dst || !dst->bs) { |
fcc67678 FZ |
2957 | return -ENOMEDIUM; |
2958 | } | |
fcc67678 FZ |
2959 | ret = bdrv_check_byte_request(dst->bs, dst_offset, bytes); |
2960 | if (ret) { | |
2961 | return ret; | |
2962 | } | |
67b51fb9 VSO |
2963 | if (write_flags & BDRV_REQ_ZERO_WRITE) { |
2964 | return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags); | |
fcc67678 FZ |
2965 | } |
2966 | ||
d4d3e5a0 FZ |
2967 | if (!src || !src->bs) { |
2968 | return -ENOMEDIUM; | |
2969 | } | |
2970 | ret = bdrv_check_byte_request(src->bs, src_offset, bytes); | |
2971 | if (ret) { | |
2972 | return ret; | |
2973 | } | |
2974 | ||
fcc67678 FZ |
2975 | if (!src->bs->drv->bdrv_co_copy_range_from |
2976 | || !dst->bs->drv->bdrv_co_copy_range_to | |
2977 | || src->bs->encrypted || dst->bs->encrypted) { | |
2978 | return -ENOTSUP; | |
2979 | } | |
37aec7d7 | 2980 | |
fcc67678 | 2981 | if (recurse_src) { |
999658a0 VSO |
2982 | bdrv_inc_in_flight(src->bs); |
2983 | tracked_request_begin(&req, src->bs, src_offset, bytes, | |
2984 | BDRV_TRACKED_READ); | |
2985 | ||
09d2f948 VSO |
2986 | /* BDRV_REQ_SERIALISING is only for write operation */ |
2987 | assert(!(read_flags & BDRV_REQ_SERIALISING)); | |
67b51fb9 | 2988 | if (!(read_flags & BDRV_REQ_NO_SERIALISING)) { |
999658a0 VSO |
2989 | wait_serialising_requests(&req); |
2990 | } | |
2991 | ||
37aec7d7 FZ |
2992 | ret = src->bs->drv->bdrv_co_copy_range_from(src->bs, |
2993 | src, src_offset, | |
2994 | dst, dst_offset, | |
67b51fb9 VSO |
2995 | bytes, |
2996 | read_flags, write_flags); | |
999658a0 VSO |
2997 | |
2998 | tracked_request_end(&req); | |
2999 | bdrv_dec_in_flight(src->bs); | |
fcc67678 | 3000 | } else { |
999658a0 VSO |
3001 | bdrv_inc_in_flight(dst->bs); |
3002 | tracked_request_begin(&req, dst->bs, dst_offset, bytes, | |
3003 | BDRV_TRACKED_WRITE); | |
0eb1e891 FZ |
3004 | ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req, |
3005 | write_flags); | |
3006 | if (!ret) { | |
3007 | ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs, | |
3008 | src, src_offset, | |
3009 | dst, dst_offset, | |
3010 | bytes, | |
3011 | read_flags, write_flags); | |
3012 | } | |
3013 | bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret); | |
999658a0 VSO |
3014 | tracked_request_end(&req); |
3015 | bdrv_dec_in_flight(dst->bs); | |
fcc67678 | 3016 | } |
999658a0 | 3017 | |
37aec7d7 | 3018 | return ret; |
fcc67678 FZ |
3019 | } |
3020 | ||
3021 | /* Copy range from @src to @dst. | |
3022 | * | |
3023 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3024 | * semantics. */ | |
3025 | int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, uint64_t src_offset, | |
3026 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3027 | uint64_t bytes, |
3028 | BdrvRequestFlags read_flags, | |
3029 | BdrvRequestFlags write_flags) | |
fcc67678 | 3030 | { |
ecc983a5 FZ |
3031 | trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes, |
3032 | read_flags, write_flags); | |
fcc67678 | 3033 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3034 | bytes, read_flags, write_flags, true); |
fcc67678 FZ |
3035 | } |
3036 | ||
3037 | /* Copy range from @src to @dst. | |
3038 | * | |
3039 | * See the comment of bdrv_co_copy_range for the parameter and return value | |
3040 | * semantics. */ | |
3041 | int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, uint64_t src_offset, | |
3042 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3043 | uint64_t bytes, |
3044 | BdrvRequestFlags read_flags, | |
3045 | BdrvRequestFlags write_flags) | |
fcc67678 | 3046 | { |
ecc983a5 FZ |
3047 | trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3048 | read_flags, write_flags); | |
fcc67678 | 3049 | return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset, |
67b51fb9 | 3050 | bytes, read_flags, write_flags, false); |
fcc67678 FZ |
3051 | } |
3052 | ||
3053 | int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset, | |
3054 | BdrvChild *dst, uint64_t dst_offset, | |
67b51fb9 VSO |
3055 | uint64_t bytes, BdrvRequestFlags read_flags, |
3056 | BdrvRequestFlags write_flags) | |
fcc67678 | 3057 | { |
37aec7d7 FZ |
3058 | return bdrv_co_copy_range_from(src, src_offset, |
3059 | dst, dst_offset, | |
67b51fb9 | 3060 | bytes, read_flags, write_flags); |
fcc67678 | 3061 | } |
3d9f2d2a KW |
3062 | |
3063 | static void bdrv_parent_cb_resize(BlockDriverState *bs) | |
3064 | { | |
3065 | BdrvChild *c; | |
3066 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
3067 | if (c->role->resize) { | |
3068 | c->role->resize(c); | |
3069 | } | |
3070 | } | |
3071 | } | |
3072 | ||
3073 | /** | |
3074 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
3075 | */ | |
3076 | int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset, | |
3077 | PreallocMode prealloc, Error **errp) | |
3078 | { | |
3079 | BlockDriverState *bs = child->bs; | |
3080 | BlockDriver *drv = bs->drv; | |
1bc5f09f KW |
3081 | BdrvTrackedRequest req; |
3082 | int64_t old_size, new_bytes; | |
3d9f2d2a KW |
3083 | int ret; |
3084 | ||
3d9f2d2a KW |
3085 | |
3086 | /* if bs->drv == NULL, bs is closed, so there's nothing to do here */ | |
3087 | if (!drv) { | |
3088 | error_setg(errp, "No medium inserted"); | |
3089 | return -ENOMEDIUM; | |
3090 | } | |
3091 | if (offset < 0) { | |
3092 | error_setg(errp, "Image size cannot be negative"); | |
3093 | return -EINVAL; | |
3094 | } | |
3095 | ||
1bc5f09f KW |
3096 | old_size = bdrv_getlength(bs); |
3097 | if (old_size < 0) { | |
3098 | error_setg_errno(errp, -old_size, "Failed to get old image size"); | |
3099 | return old_size; | |
3100 | } | |
3101 | ||
3102 | if (offset > old_size) { | |
3103 | new_bytes = offset - old_size; | |
3104 | } else { | |
3105 | new_bytes = 0; | |
3106 | } | |
3107 | ||
3d9f2d2a | 3108 | bdrv_inc_in_flight(bs); |
5416a11e FZ |
3109 | tracked_request_begin(&req, bs, offset - new_bytes, new_bytes, |
3110 | BDRV_TRACKED_TRUNCATE); | |
1bc5f09f KW |
3111 | |
3112 | /* If we are growing the image and potentially using preallocation for the | |
3113 | * new area, we need to make sure that no write requests are made to it | |
3114 | * concurrently or they might be overwritten by preallocation. */ | |
3115 | if (new_bytes) { | |
3116 | mark_request_serialising(&req, 1); | |
cd47d792 FZ |
3117 | } |
3118 | if (bs->read_only) { | |
3119 | error_setg(errp, "Image is read-only"); | |
3120 | ret = -EACCES; | |
3121 | goto out; | |
3122 | } | |
3123 | ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req, | |
3124 | 0); | |
3125 | if (ret < 0) { | |
3126 | error_setg_errno(errp, -ret, | |
3127 | "Failed to prepare request for truncation"); | |
3128 | goto out; | |
1bc5f09f | 3129 | } |
3d9f2d2a KW |
3130 | |
3131 | if (!drv->bdrv_co_truncate) { | |
3132 | if (bs->file && drv->is_filter) { | |
3133 | ret = bdrv_co_truncate(bs->file, offset, prealloc, errp); | |
3134 | goto out; | |
3135 | } | |
3136 | error_setg(errp, "Image format driver does not support resize"); | |
3137 | ret = -ENOTSUP; | |
3138 | goto out; | |
3139 | } | |
3d9f2d2a KW |
3140 | |
3141 | ret = drv->bdrv_co_truncate(bs, offset, prealloc, errp); | |
3142 | if (ret < 0) { | |
3143 | goto out; | |
3144 | } | |
3145 | ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); | |
3146 | if (ret < 0) { | |
3147 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
3148 | } else { | |
3149 | offset = bs->total_sectors * BDRV_SECTOR_SIZE; | |
3150 | } | |
cd47d792 FZ |
3151 | /* It's possible that truncation succeeded but refresh_total_sectors |
3152 | * failed, but the latter doesn't affect how we should finish the request. | |
3153 | * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. */ | |
3154 | bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0); | |
3d9f2d2a KW |
3155 | |
3156 | out: | |
1bc5f09f | 3157 | tracked_request_end(&req); |
3d9f2d2a | 3158 | bdrv_dec_in_flight(bs); |
1bc5f09f | 3159 | |
3d9f2d2a KW |
3160 | return ret; |
3161 | } | |
3162 | ||
3163 | typedef struct TruncateCo { | |
3164 | BdrvChild *child; | |
3165 | int64_t offset; | |
3166 | PreallocMode prealloc; | |
3167 | Error **errp; | |
3168 | int ret; | |
3169 | } TruncateCo; | |
3170 | ||
3171 | static void coroutine_fn bdrv_truncate_co_entry(void *opaque) | |
3172 | { | |
3173 | TruncateCo *tco = opaque; | |
3174 | tco->ret = bdrv_co_truncate(tco->child, tco->offset, tco->prealloc, | |
3175 | tco->errp); | |
4720cbee | 3176 | aio_wait_kick(); |
3d9f2d2a KW |
3177 | } |
3178 | ||
3179 | int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc, | |
3180 | Error **errp) | |
3181 | { | |
3182 | Coroutine *co; | |
3183 | TruncateCo tco = { | |
3184 | .child = child, | |
3185 | .offset = offset, | |
3186 | .prealloc = prealloc, | |
3187 | .errp = errp, | |
3188 | .ret = NOT_DONE, | |
3189 | }; | |
3190 | ||
3191 | if (qemu_in_coroutine()) { | |
3192 | /* Fast-path if already in coroutine context */ | |
3193 | bdrv_truncate_co_entry(&tco); | |
3194 | } else { | |
3195 | co = qemu_coroutine_create(bdrv_truncate_co_entry, &tco); | |
4720cbee | 3196 | bdrv_coroutine_enter(child->bs, co); |
3d9f2d2a KW |
3197 | BDRV_POLL_WHILE(child->bs, tco.ret == NOT_DONE); |
3198 | } | |
3199 | ||
3200 | return tco.ret; | |
3201 | } |