]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU Block backends | |
3 | * | |
4 | * Copyright (C) 2014-2016 Red Hat, Inc. | |
5 | * | |
6 | * Authors: | |
7 | * Markus Armbruster <armbru@redhat.com>, | |
8 | * | |
9 | * This work is licensed under the terms of the GNU LGPL, version 2.1 | |
10 | * or later. See the COPYING.LIB file in the top-level directory. | |
11 | */ | |
12 | ||
13 | #include "qemu/osdep.h" | |
14 | #include "sysemu/block-backend.h" | |
15 | #include "block/block_int.h" | |
16 | #include "block/blockjob.h" | |
17 | #include "block/throttle-groups.h" | |
18 | #include "sysemu/blockdev.h" | |
19 | #include "sysemu/sysemu.h" | |
20 | #include "qapi/error.h" | |
21 | #include "qapi/qapi-events-block.h" | |
22 | #include "qemu/id.h" | |
23 | #include "qemu/option.h" | |
24 | #include "trace.h" | |
25 | #include "migration/misc.h" | |
26 | ||
27 | /* Number of coroutines to reserve per attached device model */ | |
28 | #define COROUTINE_POOL_RESERVATION 64 | |
29 | ||
30 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
31 | ||
32 | static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb); | |
33 | ||
34 | typedef struct BlockBackendAioNotifier { | |
35 | void (*attached_aio_context)(AioContext *new_context, void *opaque); | |
36 | void (*detach_aio_context)(void *opaque); | |
37 | void *opaque; | |
38 | QLIST_ENTRY(BlockBackendAioNotifier) list; | |
39 | } BlockBackendAioNotifier; | |
40 | ||
41 | struct BlockBackend { | |
42 | char *name; | |
43 | int refcnt; | |
44 | BdrvChild *root; | |
45 | AioContext *ctx; | |
46 | DriveInfo *legacy_dinfo; /* null unless created by drive_new() */ | |
47 | QTAILQ_ENTRY(BlockBackend) link; /* for block_backends */ | |
48 | QTAILQ_ENTRY(BlockBackend) monitor_link; /* for monitor_block_backends */ | |
49 | BlockBackendPublic public; | |
50 | ||
51 | DeviceState *dev; /* attached device model, if any */ | |
52 | const BlockDevOps *dev_ops; | |
53 | void *dev_opaque; | |
54 | ||
55 | /* the block size for which the guest device expects atomicity */ | |
56 | int guest_block_size; | |
57 | ||
58 | /* If the BDS tree is removed, some of its options are stored here (which | |
59 | * can be used to restore those options in the new BDS on insert) */ | |
60 | BlockBackendRootState root_state; | |
61 | ||
62 | bool enable_write_cache; | |
63 | ||
64 | /* I/O stats (display with "info blockstats"). */ | |
65 | BlockAcctStats stats; | |
66 | ||
67 | BlockdevOnError on_read_error, on_write_error; | |
68 | bool iostatus_enabled; | |
69 | BlockDeviceIoStatus iostatus; | |
70 | ||
71 | uint64_t perm; | |
72 | uint64_t shared_perm; | |
73 | bool disable_perm; | |
74 | ||
75 | bool allow_aio_context_change; | |
76 | bool allow_write_beyond_eof; | |
77 | ||
78 | NotifierList remove_bs_notifiers, insert_bs_notifiers; | |
79 | QLIST_HEAD(, BlockBackendAioNotifier) aio_notifiers; | |
80 | ||
81 | int quiesce_counter; | |
82 | VMChangeStateEntry *vmsh; | |
83 | bool force_allow_inactivate; | |
84 | ||
85 | /* Number of in-flight aio requests. BlockDriverState also counts | |
86 | * in-flight requests but aio requests can exist even when blk->root is | |
87 | * NULL, so we cannot rely on its counter for that case. | |
88 | * Accessed with atomic ops. | |
89 | */ | |
90 | unsigned int in_flight; | |
91 | }; | |
92 | ||
93 | typedef struct BlockBackendAIOCB { | |
94 | BlockAIOCB common; | |
95 | BlockBackend *blk; | |
96 | int ret; | |
97 | } BlockBackendAIOCB; | |
98 | ||
99 | static const AIOCBInfo block_backend_aiocb_info = { | |
100 | .get_aio_context = blk_aiocb_get_aio_context, | |
101 | .aiocb_size = sizeof(BlockBackendAIOCB), | |
102 | }; | |
103 | ||
104 | static void drive_info_del(DriveInfo *dinfo); | |
105 | static BlockBackend *bdrv_first_blk(BlockDriverState *bs); | |
106 | ||
107 | /* All BlockBackends */ | |
108 | static QTAILQ_HEAD(, BlockBackend) block_backends = | |
109 | QTAILQ_HEAD_INITIALIZER(block_backends); | |
110 | ||
111 | /* All BlockBackends referenced by the monitor and which are iterated through by | |
112 | * blk_next() */ | |
113 | static QTAILQ_HEAD(, BlockBackend) monitor_block_backends = | |
114 | QTAILQ_HEAD_INITIALIZER(monitor_block_backends); | |
115 | ||
116 | static void blk_root_inherit_options(int *child_flags, QDict *child_options, | |
117 | int parent_flags, QDict *parent_options) | |
118 | { | |
119 | /* We're not supposed to call this function for root nodes */ | |
120 | abort(); | |
121 | } | |
122 | static void blk_root_drained_begin(BdrvChild *child); | |
123 | static bool blk_root_drained_poll(BdrvChild *child); | |
124 | static void blk_root_drained_end(BdrvChild *child, int *drained_end_counter); | |
125 | ||
126 | static void blk_root_change_media(BdrvChild *child, bool load); | |
127 | static void blk_root_resize(BdrvChild *child); | |
128 | ||
129 | static bool blk_root_can_set_aio_ctx(BdrvChild *child, AioContext *ctx, | |
130 | GSList **ignore, Error **errp); | |
131 | static void blk_root_set_aio_ctx(BdrvChild *child, AioContext *ctx, | |
132 | GSList **ignore); | |
133 | ||
134 | static char *blk_root_get_parent_desc(BdrvChild *child) | |
135 | { | |
136 | BlockBackend *blk = child->opaque; | |
137 | char *dev_id; | |
138 | ||
139 | if (blk->name) { | |
140 | return g_strdup(blk->name); | |
141 | } | |
142 | ||
143 | dev_id = blk_get_attached_dev_id(blk); | |
144 | if (*dev_id) { | |
145 | return dev_id; | |
146 | } else { | |
147 | /* TODO Callback into the BB owner for something more detailed */ | |
148 | g_free(dev_id); | |
149 | return g_strdup("a block device"); | |
150 | } | |
151 | } | |
152 | ||
153 | static const char *blk_root_get_name(BdrvChild *child) | |
154 | { | |
155 | return blk_name(child->opaque); | |
156 | } | |
157 | ||
158 | static void blk_vm_state_changed(void *opaque, int running, RunState state) | |
159 | { | |
160 | Error *local_err = NULL; | |
161 | BlockBackend *blk = opaque; | |
162 | ||
163 | if (state == RUN_STATE_INMIGRATE) { | |
164 | return; | |
165 | } | |
166 | ||
167 | qemu_del_vm_change_state_handler(blk->vmsh); | |
168 | blk->vmsh = NULL; | |
169 | blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err); | |
170 | if (local_err) { | |
171 | error_report_err(local_err); | |
172 | } | |
173 | } | |
174 | ||
175 | /* | |
176 | * Notifies the user of the BlockBackend that migration has completed. qdev | |
177 | * devices can tighten their permissions in response (specifically revoke | |
178 | * shared write permissions that we needed for storage migration). | |
179 | * | |
180 | * If an error is returned, the VM cannot be allowed to be resumed. | |
181 | */ | |
182 | static void blk_root_activate(BdrvChild *child, Error **errp) | |
183 | { | |
184 | BlockBackend *blk = child->opaque; | |
185 | Error *local_err = NULL; | |
186 | ||
187 | if (!blk->disable_perm) { | |
188 | return; | |
189 | } | |
190 | ||
191 | blk->disable_perm = false; | |
192 | ||
193 | blk_set_perm(blk, blk->perm, BLK_PERM_ALL, &local_err); | |
194 | if (local_err) { | |
195 | error_propagate(errp, local_err); | |
196 | blk->disable_perm = true; | |
197 | return; | |
198 | } | |
199 | ||
200 | if (runstate_check(RUN_STATE_INMIGRATE)) { | |
201 | /* Activation can happen when migration process is still active, for | |
202 | * example when nbd_server_add is called during non-shared storage | |
203 | * migration. Defer the shared_perm update to migration completion. */ | |
204 | if (!blk->vmsh) { | |
205 | blk->vmsh = qemu_add_vm_change_state_handler(blk_vm_state_changed, | |
206 | blk); | |
207 | } | |
208 | return; | |
209 | } | |
210 | ||
211 | blk_set_perm(blk, blk->perm, blk->shared_perm, &local_err); | |
212 | if (local_err) { | |
213 | error_propagate(errp, local_err); | |
214 | blk->disable_perm = true; | |
215 | return; | |
216 | } | |
217 | } | |
218 | ||
219 | void blk_set_force_allow_inactivate(BlockBackend *blk) | |
220 | { | |
221 | blk->force_allow_inactivate = true; | |
222 | } | |
223 | ||
224 | static bool blk_can_inactivate(BlockBackend *blk) | |
225 | { | |
226 | /* If it is a guest device, inactivate is ok. */ | |
227 | if (blk->dev || blk_name(blk)[0]) { | |
228 | return true; | |
229 | } | |
230 | ||
231 | /* Inactivating means no more writes to the image can be done, | |
232 | * even if those writes would be changes invisible to the | |
233 | * guest. For block job BBs that satisfy this, we can just allow | |
234 | * it. This is the case for mirror job source, which is required | |
235 | * by libvirt non-shared block migration. */ | |
236 | if (!(blk->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED))) { | |
237 | return true; | |
238 | } | |
239 | ||
240 | return blk->force_allow_inactivate; | |
241 | } | |
242 | ||
243 | static int blk_root_inactivate(BdrvChild *child) | |
244 | { | |
245 | BlockBackend *blk = child->opaque; | |
246 | ||
247 | if (blk->disable_perm) { | |
248 | return 0; | |
249 | } | |
250 | ||
251 | if (!blk_can_inactivate(blk)) { | |
252 | return -EPERM; | |
253 | } | |
254 | ||
255 | blk->disable_perm = true; | |
256 | if (blk->root) { | |
257 | bdrv_child_try_set_perm(blk->root, 0, BLK_PERM_ALL, &error_abort); | |
258 | } | |
259 | ||
260 | return 0; | |
261 | } | |
262 | ||
263 | static void blk_root_attach(BdrvChild *child) | |
264 | { | |
265 | BlockBackend *blk = child->opaque; | |
266 | BlockBackendAioNotifier *notifier; | |
267 | ||
268 | trace_blk_root_attach(child, blk, child->bs); | |
269 | ||
270 | QLIST_FOREACH(notifier, &blk->aio_notifiers, list) { | |
271 | bdrv_add_aio_context_notifier(child->bs, | |
272 | notifier->attached_aio_context, | |
273 | notifier->detach_aio_context, | |
274 | notifier->opaque); | |
275 | } | |
276 | } | |
277 | ||
278 | static void blk_root_detach(BdrvChild *child) | |
279 | { | |
280 | BlockBackend *blk = child->opaque; | |
281 | BlockBackendAioNotifier *notifier; | |
282 | ||
283 | trace_blk_root_detach(child, blk, child->bs); | |
284 | ||
285 | QLIST_FOREACH(notifier, &blk->aio_notifiers, list) { | |
286 | bdrv_remove_aio_context_notifier(child->bs, | |
287 | notifier->attached_aio_context, | |
288 | notifier->detach_aio_context, | |
289 | notifier->opaque); | |
290 | } | |
291 | } | |
292 | ||
293 | static const BdrvChildRole child_root = { | |
294 | .inherit_options = blk_root_inherit_options, | |
295 | ||
296 | .change_media = blk_root_change_media, | |
297 | .resize = blk_root_resize, | |
298 | .get_name = blk_root_get_name, | |
299 | .get_parent_desc = blk_root_get_parent_desc, | |
300 | ||
301 | .drained_begin = blk_root_drained_begin, | |
302 | .drained_poll = blk_root_drained_poll, | |
303 | .drained_end = blk_root_drained_end, | |
304 | ||
305 | .activate = blk_root_activate, | |
306 | .inactivate = blk_root_inactivate, | |
307 | ||
308 | .attach = blk_root_attach, | |
309 | .detach = blk_root_detach, | |
310 | ||
311 | .can_set_aio_ctx = blk_root_can_set_aio_ctx, | |
312 | .set_aio_ctx = blk_root_set_aio_ctx, | |
313 | }; | |
314 | ||
315 | /* | |
316 | * Create a new BlockBackend with a reference count of one. | |
317 | * | |
318 | * @perm is a bitmasks of BLK_PERM_* constants which describes the permissions | |
319 | * to request for a block driver node that is attached to this BlockBackend. | |
320 | * @shared_perm is a bitmask which describes which permissions may be granted | |
321 | * to other users of the attached node. | |
322 | * Both sets of permissions can be changed later using blk_set_perm(). | |
323 | * | |
324 | * Return the new BlockBackend on success, null on failure. | |
325 | */ | |
326 | BlockBackend *blk_new(AioContext *ctx, uint64_t perm, uint64_t shared_perm) | |
327 | { | |
328 | BlockBackend *blk; | |
329 | ||
330 | blk = g_new0(BlockBackend, 1); | |
331 | blk->refcnt = 1; | |
332 | blk->ctx = ctx; | |
333 | blk->perm = perm; | |
334 | blk->shared_perm = shared_perm; | |
335 | blk_set_enable_write_cache(blk, true); | |
336 | ||
337 | blk->on_read_error = BLOCKDEV_ON_ERROR_REPORT; | |
338 | blk->on_write_error = BLOCKDEV_ON_ERROR_ENOSPC; | |
339 | ||
340 | block_acct_init(&blk->stats); | |
341 | ||
342 | notifier_list_init(&blk->remove_bs_notifiers); | |
343 | notifier_list_init(&blk->insert_bs_notifiers); | |
344 | QLIST_INIT(&blk->aio_notifiers); | |
345 | ||
346 | QTAILQ_INSERT_TAIL(&block_backends, blk, link); | |
347 | return blk; | |
348 | } | |
349 | ||
350 | /* | |
351 | * Creates a new BlockBackend, opens a new BlockDriverState, and connects both. | |
352 | * The new BlockBackend is in the main AioContext. | |
353 | * | |
354 | * Just as with bdrv_open(), after having called this function the reference to | |
355 | * @options belongs to the block layer (even on failure). | |
356 | * | |
357 | * TODO: Remove @filename and @flags; it should be possible to specify a whole | |
358 | * BDS tree just by specifying the @options QDict (or @reference, | |
359 | * alternatively). At the time of adding this function, this is not possible, | |
360 | * though, so callers of this function have to be able to specify @filename and | |
361 | * @flags. | |
362 | */ | |
363 | BlockBackend *blk_new_open(const char *filename, const char *reference, | |
364 | QDict *options, int flags, Error **errp) | |
365 | { | |
366 | BlockBackend *blk; | |
367 | BlockDriverState *bs; | |
368 | uint64_t perm = 0; | |
369 | ||
370 | /* blk_new_open() is mainly used in .bdrv_create implementations and the | |
371 | * tools where sharing isn't a concern because the BDS stays private, so we | |
372 | * just request permission according to the flags. | |
373 | * | |
374 | * The exceptions are xen_disk and blockdev_init(); in these cases, the | |
375 | * caller of blk_new_open() doesn't make use of the permissions, but they | |
376 | * shouldn't hurt either. We can still share everything here because the | |
377 | * guest devices will add their own blockers if they can't share. */ | |
378 | if ((flags & BDRV_O_NO_IO) == 0) { | |
379 | perm |= BLK_PERM_CONSISTENT_READ; | |
380 | if (flags & BDRV_O_RDWR) { | |
381 | perm |= BLK_PERM_WRITE; | |
382 | } | |
383 | } | |
384 | if (flags & BDRV_O_RESIZE) { | |
385 | perm |= BLK_PERM_RESIZE; | |
386 | } | |
387 | ||
388 | blk = blk_new(qemu_get_aio_context(), perm, BLK_PERM_ALL); | |
389 | bs = bdrv_open(filename, reference, options, flags, errp); | |
390 | if (!bs) { | |
391 | blk_unref(blk); | |
392 | return NULL; | |
393 | } | |
394 | ||
395 | blk->root = bdrv_root_attach_child(bs, "root", &child_root, blk->ctx, | |
396 | perm, BLK_PERM_ALL, blk, errp); | |
397 | if (!blk->root) { | |
398 | blk_unref(blk); | |
399 | return NULL; | |
400 | } | |
401 | ||
402 | return blk; | |
403 | } | |
404 | ||
405 | static void blk_delete(BlockBackend *blk) | |
406 | { | |
407 | assert(!blk->refcnt); | |
408 | assert(!blk->name); | |
409 | assert(!blk->dev); | |
410 | if (blk->public.throttle_group_member.throttle_state) { | |
411 | blk_io_limits_disable(blk); | |
412 | } | |
413 | if (blk->root) { | |
414 | blk_remove_bs(blk); | |
415 | } | |
416 | if (blk->vmsh) { | |
417 | qemu_del_vm_change_state_handler(blk->vmsh); | |
418 | blk->vmsh = NULL; | |
419 | } | |
420 | assert(QLIST_EMPTY(&blk->remove_bs_notifiers.notifiers)); | |
421 | assert(QLIST_EMPTY(&blk->insert_bs_notifiers.notifiers)); | |
422 | assert(QLIST_EMPTY(&blk->aio_notifiers)); | |
423 | QTAILQ_REMOVE(&block_backends, blk, link); | |
424 | drive_info_del(blk->legacy_dinfo); | |
425 | block_acct_cleanup(&blk->stats); | |
426 | g_free(blk); | |
427 | } | |
428 | ||
429 | static void drive_info_del(DriveInfo *dinfo) | |
430 | { | |
431 | if (!dinfo) { | |
432 | return; | |
433 | } | |
434 | qemu_opts_del(dinfo->opts); | |
435 | g_free(dinfo); | |
436 | } | |
437 | ||
438 | int blk_get_refcnt(BlockBackend *blk) | |
439 | { | |
440 | return blk ? blk->refcnt : 0; | |
441 | } | |
442 | ||
443 | /* | |
444 | * Increment @blk's reference count. | |
445 | * @blk must not be null. | |
446 | */ | |
447 | void blk_ref(BlockBackend *blk) | |
448 | { | |
449 | assert(blk->refcnt > 0); | |
450 | blk->refcnt++; | |
451 | } | |
452 | ||
453 | /* | |
454 | * Decrement @blk's reference count. | |
455 | * If this drops it to zero, destroy @blk. | |
456 | * For convenience, do nothing if @blk is null. | |
457 | */ | |
458 | void blk_unref(BlockBackend *blk) | |
459 | { | |
460 | if (blk) { | |
461 | assert(blk->refcnt > 0); | |
462 | if (blk->refcnt > 1) { | |
463 | blk->refcnt--; | |
464 | } else { | |
465 | blk_drain(blk); | |
466 | /* blk_drain() cannot resurrect blk, nobody held a reference */ | |
467 | assert(blk->refcnt == 1); | |
468 | blk->refcnt = 0; | |
469 | blk_delete(blk); | |
470 | } | |
471 | } | |
472 | } | |
473 | ||
474 | /* | |
475 | * Behaves similarly to blk_next() but iterates over all BlockBackends, even the | |
476 | * ones which are hidden (i.e. are not referenced by the monitor). | |
477 | */ | |
478 | BlockBackend *blk_all_next(BlockBackend *blk) | |
479 | { | |
480 | return blk ? QTAILQ_NEXT(blk, link) | |
481 | : QTAILQ_FIRST(&block_backends); | |
482 | } | |
483 | ||
484 | void blk_remove_all_bs(void) | |
485 | { | |
486 | BlockBackend *blk = NULL; | |
487 | ||
488 | while ((blk = blk_all_next(blk)) != NULL) { | |
489 | AioContext *ctx = blk_get_aio_context(blk); | |
490 | ||
491 | aio_context_acquire(ctx); | |
492 | if (blk->root) { | |
493 | blk_remove_bs(blk); | |
494 | } | |
495 | aio_context_release(ctx); | |
496 | } | |
497 | } | |
498 | ||
499 | /* | |
500 | * Return the monitor-owned BlockBackend after @blk. | |
501 | * If @blk is null, return the first one. | |
502 | * Else, return @blk's next sibling, which may be null. | |
503 | * | |
504 | * To iterate over all BlockBackends, do | |
505 | * for (blk = blk_next(NULL); blk; blk = blk_next(blk)) { | |
506 | * ... | |
507 | * } | |
508 | */ | |
509 | BlockBackend *blk_next(BlockBackend *blk) | |
510 | { | |
511 | return blk ? QTAILQ_NEXT(blk, monitor_link) | |
512 | : QTAILQ_FIRST(&monitor_block_backends); | |
513 | } | |
514 | ||
515 | /* Iterates over all top-level BlockDriverStates, i.e. BDSs that are owned by | |
516 | * the monitor or attached to a BlockBackend */ | |
517 | BlockDriverState *bdrv_next(BdrvNextIterator *it) | |
518 | { | |
519 | BlockDriverState *bs, *old_bs; | |
520 | ||
521 | /* Must be called from the main loop */ | |
522 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
523 | ||
524 | /* First, return all root nodes of BlockBackends. In order to avoid | |
525 | * returning a BDS twice when multiple BBs refer to it, we only return it | |
526 | * if the BB is the first one in the parent list of the BDS. */ | |
527 | if (it->phase == BDRV_NEXT_BACKEND_ROOTS) { | |
528 | BlockBackend *old_blk = it->blk; | |
529 | ||
530 | old_bs = old_blk ? blk_bs(old_blk) : NULL; | |
531 | ||
532 | do { | |
533 | it->blk = blk_all_next(it->blk); | |
534 | bs = it->blk ? blk_bs(it->blk) : NULL; | |
535 | } while (it->blk && (bs == NULL || bdrv_first_blk(bs) != it->blk)); | |
536 | ||
537 | if (it->blk) { | |
538 | blk_ref(it->blk); | |
539 | } | |
540 | blk_unref(old_blk); | |
541 | ||
542 | if (bs) { | |
543 | bdrv_ref(bs); | |
544 | bdrv_unref(old_bs); | |
545 | return bs; | |
546 | } | |
547 | it->phase = BDRV_NEXT_MONITOR_OWNED; | |
548 | } else { | |
549 | old_bs = it->bs; | |
550 | } | |
551 | ||
552 | /* Then return the monitor-owned BDSes without a BB attached. Ignore all | |
553 | * BDSes that are attached to a BlockBackend here; they have been handled | |
554 | * by the above block already */ | |
555 | do { | |
556 | it->bs = bdrv_next_monitor_owned(it->bs); | |
557 | bs = it->bs; | |
558 | } while (bs && bdrv_has_blk(bs)); | |
559 | ||
560 | if (bs) { | |
561 | bdrv_ref(bs); | |
562 | } | |
563 | bdrv_unref(old_bs); | |
564 | ||
565 | return bs; | |
566 | } | |
567 | ||
568 | static void bdrv_next_reset(BdrvNextIterator *it) | |
569 | { | |
570 | *it = (BdrvNextIterator) { | |
571 | .phase = BDRV_NEXT_BACKEND_ROOTS, | |
572 | }; | |
573 | } | |
574 | ||
575 | BlockDriverState *bdrv_first(BdrvNextIterator *it) | |
576 | { | |
577 | bdrv_next_reset(it); | |
578 | return bdrv_next(it); | |
579 | } | |
580 | ||
581 | /* Must be called when aborting a bdrv_next() iteration before | |
582 | * bdrv_next() returns NULL */ | |
583 | void bdrv_next_cleanup(BdrvNextIterator *it) | |
584 | { | |
585 | /* Must be called from the main loop */ | |
586 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); | |
587 | ||
588 | if (it->phase == BDRV_NEXT_BACKEND_ROOTS) { | |
589 | if (it->blk) { | |
590 | bdrv_unref(blk_bs(it->blk)); | |
591 | blk_unref(it->blk); | |
592 | } | |
593 | } else { | |
594 | bdrv_unref(it->bs); | |
595 | } | |
596 | ||
597 | bdrv_next_reset(it); | |
598 | } | |
599 | ||
600 | /* | |
601 | * Add a BlockBackend into the list of backends referenced by the monitor, with | |
602 | * the given @name acting as the handle for the monitor. | |
603 | * Strictly for use by blockdev.c. | |
604 | * | |
605 | * @name must not be null or empty. | |
606 | * | |
607 | * Returns true on success and false on failure. In the latter case, an Error | |
608 | * object is returned through @errp. | |
609 | */ | |
610 | bool monitor_add_blk(BlockBackend *blk, const char *name, Error **errp) | |
611 | { | |
612 | assert(!blk->name); | |
613 | assert(name && name[0]); | |
614 | ||
615 | if (!id_wellformed(name)) { | |
616 | error_setg(errp, "Invalid device name"); | |
617 | return false; | |
618 | } | |
619 | if (blk_by_name(name)) { | |
620 | error_setg(errp, "Device with id '%s' already exists", name); | |
621 | return false; | |
622 | } | |
623 | if (bdrv_find_node(name)) { | |
624 | error_setg(errp, | |
625 | "Device name '%s' conflicts with an existing node name", | |
626 | name); | |
627 | return false; | |
628 | } | |
629 | ||
630 | blk->name = g_strdup(name); | |
631 | QTAILQ_INSERT_TAIL(&monitor_block_backends, blk, monitor_link); | |
632 | return true; | |
633 | } | |
634 | ||
635 | /* | |
636 | * Remove a BlockBackend from the list of backends referenced by the monitor. | |
637 | * Strictly for use by blockdev.c. | |
638 | */ | |
639 | void monitor_remove_blk(BlockBackend *blk) | |
640 | { | |
641 | if (!blk->name) { | |
642 | return; | |
643 | } | |
644 | ||
645 | QTAILQ_REMOVE(&monitor_block_backends, blk, monitor_link); | |
646 | g_free(blk->name); | |
647 | blk->name = NULL; | |
648 | } | |
649 | ||
650 | /* | |
651 | * Return @blk's name, a non-null string. | |
652 | * Returns an empty string iff @blk is not referenced by the monitor. | |
653 | */ | |
654 | const char *blk_name(const BlockBackend *blk) | |
655 | { | |
656 | return blk->name ?: ""; | |
657 | } | |
658 | ||
659 | /* | |
660 | * Return the BlockBackend with name @name if it exists, else null. | |
661 | * @name must not be null. | |
662 | */ | |
663 | BlockBackend *blk_by_name(const char *name) | |
664 | { | |
665 | BlockBackend *blk = NULL; | |
666 | ||
667 | assert(name); | |
668 | while ((blk = blk_next(blk)) != NULL) { | |
669 | if (!strcmp(name, blk->name)) { | |
670 | return blk; | |
671 | } | |
672 | } | |
673 | return NULL; | |
674 | } | |
675 | ||
676 | /* | |
677 | * Return the BlockDriverState attached to @blk if any, else null. | |
678 | */ | |
679 | BlockDriverState *blk_bs(BlockBackend *blk) | |
680 | { | |
681 | return blk->root ? blk->root->bs : NULL; | |
682 | } | |
683 | ||
684 | static BlockBackend *bdrv_first_blk(BlockDriverState *bs) | |
685 | { | |
686 | BdrvChild *child; | |
687 | QLIST_FOREACH(child, &bs->parents, next_parent) { | |
688 | if (child->role == &child_root) { | |
689 | return child->opaque; | |
690 | } | |
691 | } | |
692 | ||
693 | return NULL; | |
694 | } | |
695 | ||
696 | /* | |
697 | * Returns true if @bs has an associated BlockBackend. | |
698 | */ | |
699 | bool bdrv_has_blk(BlockDriverState *bs) | |
700 | { | |
701 | return bdrv_first_blk(bs) != NULL; | |
702 | } | |
703 | ||
704 | /* | |
705 | * Returns true if @bs has only BlockBackends as parents. | |
706 | */ | |
707 | bool bdrv_is_root_node(BlockDriverState *bs) | |
708 | { | |
709 | BdrvChild *c; | |
710 | ||
711 | QLIST_FOREACH(c, &bs->parents, next_parent) { | |
712 | if (c->role != &child_root) { | |
713 | return false; | |
714 | } | |
715 | } | |
716 | ||
717 | return true; | |
718 | } | |
719 | ||
720 | /* | |
721 | * Return @blk's DriveInfo if any, else null. | |
722 | */ | |
723 | DriveInfo *blk_legacy_dinfo(BlockBackend *blk) | |
724 | { | |
725 | return blk->legacy_dinfo; | |
726 | } | |
727 | ||
728 | /* | |
729 | * Set @blk's DriveInfo to @dinfo, and return it. | |
730 | * @blk must not have a DriveInfo set already. | |
731 | * No other BlockBackend may have the same DriveInfo set. | |
732 | */ | |
733 | DriveInfo *blk_set_legacy_dinfo(BlockBackend *blk, DriveInfo *dinfo) | |
734 | { | |
735 | assert(!blk->legacy_dinfo); | |
736 | return blk->legacy_dinfo = dinfo; | |
737 | } | |
738 | ||
739 | /* | |
740 | * Return the BlockBackend with DriveInfo @dinfo. | |
741 | * It must exist. | |
742 | */ | |
743 | BlockBackend *blk_by_legacy_dinfo(DriveInfo *dinfo) | |
744 | { | |
745 | BlockBackend *blk = NULL; | |
746 | ||
747 | while ((blk = blk_next(blk)) != NULL) { | |
748 | if (blk->legacy_dinfo == dinfo) { | |
749 | return blk; | |
750 | } | |
751 | } | |
752 | abort(); | |
753 | } | |
754 | ||
755 | /* | |
756 | * Returns a pointer to the publicly accessible fields of @blk. | |
757 | */ | |
758 | BlockBackendPublic *blk_get_public(BlockBackend *blk) | |
759 | { | |
760 | return &blk->public; | |
761 | } | |
762 | ||
763 | /* | |
764 | * Returns a BlockBackend given the associated @public fields. | |
765 | */ | |
766 | BlockBackend *blk_by_public(BlockBackendPublic *public) | |
767 | { | |
768 | return container_of(public, BlockBackend, public); | |
769 | } | |
770 | ||
771 | /* | |
772 | * Disassociates the currently associated BlockDriverState from @blk. | |
773 | */ | |
774 | void blk_remove_bs(BlockBackend *blk) | |
775 | { | |
776 | ThrottleGroupMember *tgm = &blk->public.throttle_group_member; | |
777 | BlockDriverState *bs; | |
778 | ||
779 | notifier_list_notify(&blk->remove_bs_notifiers, blk); | |
780 | if (tgm->throttle_state) { | |
781 | bs = blk_bs(blk); | |
782 | bdrv_drained_begin(bs); | |
783 | throttle_group_detach_aio_context(tgm); | |
784 | throttle_group_attach_aio_context(tgm, qemu_get_aio_context()); | |
785 | bdrv_drained_end(bs); | |
786 | } | |
787 | ||
788 | blk_update_root_state(blk); | |
789 | ||
790 | /* bdrv_root_unref_child() will cause blk->root to become stale and may | |
791 | * switch to a completion coroutine later on. Let's drain all I/O here | |
792 | * to avoid that and a potential QEMU crash. | |
793 | */ | |
794 | blk_drain(blk); | |
795 | bdrv_root_unref_child(blk->root); | |
796 | blk->root = NULL; | |
797 | } | |
798 | ||
799 | /* | |
800 | * Associates a new BlockDriverState with @blk. | |
801 | */ | |
802 | int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp) | |
803 | { | |
804 | ThrottleGroupMember *tgm = &blk->public.throttle_group_member; | |
805 | bdrv_ref(bs); | |
806 | blk->root = bdrv_root_attach_child(bs, "root", &child_root, blk->ctx, | |
807 | blk->perm, blk->shared_perm, blk, errp); | |
808 | if (blk->root == NULL) { | |
809 | return -EPERM; | |
810 | } | |
811 | ||
812 | notifier_list_notify(&blk->insert_bs_notifiers, blk); | |
813 | if (tgm->throttle_state) { | |
814 | throttle_group_detach_aio_context(tgm); | |
815 | throttle_group_attach_aio_context(tgm, bdrv_get_aio_context(bs)); | |
816 | } | |
817 | ||
818 | return 0; | |
819 | } | |
820 | ||
821 | /* | |
822 | * Sets the permission bitmasks that the user of the BlockBackend needs. | |
823 | */ | |
824 | int blk_set_perm(BlockBackend *blk, uint64_t perm, uint64_t shared_perm, | |
825 | Error **errp) | |
826 | { | |
827 | int ret; | |
828 | ||
829 | if (blk->root && !blk->disable_perm) { | |
830 | ret = bdrv_child_try_set_perm(blk->root, perm, shared_perm, errp); | |
831 | if (ret < 0) { | |
832 | return ret; | |
833 | } | |
834 | } | |
835 | ||
836 | blk->perm = perm; | |
837 | blk->shared_perm = shared_perm; | |
838 | ||
839 | return 0; | |
840 | } | |
841 | ||
842 | void blk_get_perm(BlockBackend *blk, uint64_t *perm, uint64_t *shared_perm) | |
843 | { | |
844 | *perm = blk->perm; | |
845 | *shared_perm = blk->shared_perm; | |
846 | } | |
847 | ||
848 | /* | |
849 | * Attach device model @dev to @blk. | |
850 | * Return 0 on success, -EBUSY when a device model is attached already. | |
851 | */ | |
852 | int blk_attach_dev(BlockBackend *blk, DeviceState *dev) | |
853 | { | |
854 | if (blk->dev) { | |
855 | return -EBUSY; | |
856 | } | |
857 | ||
858 | /* While migration is still incoming, we don't need to apply the | |
859 | * permissions of guest device BlockBackends. We might still have a block | |
860 | * job or NBD server writing to the image for storage migration. */ | |
861 | if (runstate_check(RUN_STATE_INMIGRATE)) { | |
862 | blk->disable_perm = true; | |
863 | } | |
864 | ||
865 | blk_ref(blk); | |
866 | blk->dev = dev; | |
867 | blk_iostatus_reset(blk); | |
868 | ||
869 | return 0; | |
870 | } | |
871 | ||
872 | /* | |
873 | * Detach device model @dev from @blk. | |
874 | * @dev must be currently attached to @blk. | |
875 | */ | |
876 | void blk_detach_dev(BlockBackend *blk, DeviceState *dev) | |
877 | { | |
878 | assert(blk->dev == dev); | |
879 | blk->dev = NULL; | |
880 | blk->dev_ops = NULL; | |
881 | blk->dev_opaque = NULL; | |
882 | blk->guest_block_size = 512; | |
883 | blk_set_perm(blk, 0, BLK_PERM_ALL, &error_abort); | |
884 | blk_unref(blk); | |
885 | } | |
886 | ||
887 | /* | |
888 | * Return the device model attached to @blk if any, else null. | |
889 | */ | |
890 | DeviceState *blk_get_attached_dev(BlockBackend *blk) | |
891 | { | |
892 | return blk->dev; | |
893 | } | |
894 | ||
895 | /* Return the qdev ID, or if no ID is assigned the QOM path, of the block | |
896 | * device attached to the BlockBackend. */ | |
897 | char *blk_get_attached_dev_id(BlockBackend *blk) | |
898 | { | |
899 | DeviceState *dev = blk->dev; | |
900 | ||
901 | if (!dev) { | |
902 | return g_strdup(""); | |
903 | } else if (dev->id) { | |
904 | return g_strdup(dev->id); | |
905 | } | |
906 | ||
907 | return object_get_canonical_path(OBJECT(dev)) ?: g_strdup(""); | |
908 | } | |
909 | ||
910 | /* | |
911 | * Return the BlockBackend which has the device model @dev attached if it | |
912 | * exists, else null. | |
913 | * | |
914 | * @dev must not be null. | |
915 | */ | |
916 | BlockBackend *blk_by_dev(void *dev) | |
917 | { | |
918 | BlockBackend *blk = NULL; | |
919 | ||
920 | assert(dev != NULL); | |
921 | while ((blk = blk_all_next(blk)) != NULL) { | |
922 | if (blk->dev == dev) { | |
923 | return blk; | |
924 | } | |
925 | } | |
926 | return NULL; | |
927 | } | |
928 | ||
929 | /* | |
930 | * Set @blk's device model callbacks to @ops. | |
931 | * @opaque is the opaque argument to pass to the callbacks. | |
932 | * This is for use by device models. | |
933 | */ | |
934 | void blk_set_dev_ops(BlockBackend *blk, const BlockDevOps *ops, | |
935 | void *opaque) | |
936 | { | |
937 | blk->dev_ops = ops; | |
938 | blk->dev_opaque = opaque; | |
939 | ||
940 | /* Are we currently quiesced? Should we enforce this right now? */ | |
941 | if (blk->quiesce_counter && ops->drained_begin) { | |
942 | ops->drained_begin(opaque); | |
943 | } | |
944 | } | |
945 | ||
946 | /* | |
947 | * Notify @blk's attached device model of media change. | |
948 | * | |
949 | * If @load is true, notify of media load. This action can fail, meaning that | |
950 | * the medium cannot be loaded. @errp is set then. | |
951 | * | |
952 | * If @load is false, notify of media eject. This can never fail. | |
953 | * | |
954 | * Also send DEVICE_TRAY_MOVED events as appropriate. | |
955 | */ | |
956 | void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp) | |
957 | { | |
958 | if (blk->dev_ops && blk->dev_ops->change_media_cb) { | |
959 | bool tray_was_open, tray_is_open; | |
960 | Error *local_err = NULL; | |
961 | ||
962 | tray_was_open = blk_dev_is_tray_open(blk); | |
963 | blk->dev_ops->change_media_cb(blk->dev_opaque, load, &local_err); | |
964 | if (local_err) { | |
965 | assert(load == true); | |
966 | error_propagate(errp, local_err); | |
967 | return; | |
968 | } | |
969 | tray_is_open = blk_dev_is_tray_open(blk); | |
970 | ||
971 | if (tray_was_open != tray_is_open) { | |
972 | char *id = blk_get_attached_dev_id(blk); | |
973 | qapi_event_send_device_tray_moved(blk_name(blk), id, tray_is_open); | |
974 | g_free(id); | |
975 | } | |
976 | } | |
977 | } | |
978 | ||
979 | static void blk_root_change_media(BdrvChild *child, bool load) | |
980 | { | |
981 | blk_dev_change_media_cb(child->opaque, load, NULL); | |
982 | } | |
983 | ||
984 | /* | |
985 | * Does @blk's attached device model have removable media? | |
986 | * %true if no device model is attached. | |
987 | */ | |
988 | bool blk_dev_has_removable_media(BlockBackend *blk) | |
989 | { | |
990 | return !blk->dev || (blk->dev_ops && blk->dev_ops->change_media_cb); | |
991 | } | |
992 | ||
993 | /* | |
994 | * Does @blk's attached device model have a tray? | |
995 | */ | |
996 | bool blk_dev_has_tray(BlockBackend *blk) | |
997 | { | |
998 | return blk->dev_ops && blk->dev_ops->is_tray_open; | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * Notify @blk's attached device model of a media eject request. | |
1003 | * If @force is true, the medium is about to be yanked out forcefully. | |
1004 | */ | |
1005 | void blk_dev_eject_request(BlockBackend *blk, bool force) | |
1006 | { | |
1007 | if (blk->dev_ops && blk->dev_ops->eject_request_cb) { | |
1008 | blk->dev_ops->eject_request_cb(blk->dev_opaque, force); | |
1009 | } | |
1010 | } | |
1011 | ||
1012 | /* | |
1013 | * Does @blk's attached device model have a tray, and is it open? | |
1014 | */ | |
1015 | bool blk_dev_is_tray_open(BlockBackend *blk) | |
1016 | { | |
1017 | if (blk_dev_has_tray(blk)) { | |
1018 | return blk->dev_ops->is_tray_open(blk->dev_opaque); | |
1019 | } | |
1020 | return false; | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * Does @blk's attached device model have the medium locked? | |
1025 | * %false if the device model has no such lock. | |
1026 | */ | |
1027 | bool blk_dev_is_medium_locked(BlockBackend *blk) | |
1028 | { | |
1029 | if (blk->dev_ops && blk->dev_ops->is_medium_locked) { | |
1030 | return blk->dev_ops->is_medium_locked(blk->dev_opaque); | |
1031 | } | |
1032 | return false; | |
1033 | } | |
1034 | ||
1035 | /* | |
1036 | * Notify @blk's attached device model of a backend size change. | |
1037 | */ | |
1038 | static void blk_root_resize(BdrvChild *child) | |
1039 | { | |
1040 | BlockBackend *blk = child->opaque; | |
1041 | ||
1042 | if (blk->dev_ops && blk->dev_ops->resize_cb) { | |
1043 | blk->dev_ops->resize_cb(blk->dev_opaque); | |
1044 | } | |
1045 | } | |
1046 | ||
1047 | void blk_iostatus_enable(BlockBackend *blk) | |
1048 | { | |
1049 | blk->iostatus_enabled = true; | |
1050 | blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK; | |
1051 | } | |
1052 | ||
1053 | /* The I/O status is only enabled if the drive explicitly | |
1054 | * enables it _and_ the VM is configured to stop on errors */ | |
1055 | bool blk_iostatus_is_enabled(const BlockBackend *blk) | |
1056 | { | |
1057 | return (blk->iostatus_enabled && | |
1058 | (blk->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC || | |
1059 | blk->on_write_error == BLOCKDEV_ON_ERROR_STOP || | |
1060 | blk->on_read_error == BLOCKDEV_ON_ERROR_STOP)); | |
1061 | } | |
1062 | ||
1063 | BlockDeviceIoStatus blk_iostatus(const BlockBackend *blk) | |
1064 | { | |
1065 | return blk->iostatus; | |
1066 | } | |
1067 | ||
1068 | void blk_iostatus_disable(BlockBackend *blk) | |
1069 | { | |
1070 | blk->iostatus_enabled = false; | |
1071 | } | |
1072 | ||
1073 | void blk_iostatus_reset(BlockBackend *blk) | |
1074 | { | |
1075 | if (blk_iostatus_is_enabled(blk)) { | |
1076 | blk->iostatus = BLOCK_DEVICE_IO_STATUS_OK; | |
1077 | } | |
1078 | } | |
1079 | ||
1080 | void blk_iostatus_set_err(BlockBackend *blk, int error) | |
1081 | { | |
1082 | assert(blk_iostatus_is_enabled(blk)); | |
1083 | if (blk->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { | |
1084 | blk->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : | |
1085 | BLOCK_DEVICE_IO_STATUS_FAILED; | |
1086 | } | |
1087 | } | |
1088 | ||
1089 | void blk_set_allow_write_beyond_eof(BlockBackend *blk, bool allow) | |
1090 | { | |
1091 | blk->allow_write_beyond_eof = allow; | |
1092 | } | |
1093 | ||
1094 | void blk_set_allow_aio_context_change(BlockBackend *blk, bool allow) | |
1095 | { | |
1096 | blk->allow_aio_context_change = allow; | |
1097 | } | |
1098 | ||
1099 | static int blk_check_byte_request(BlockBackend *blk, int64_t offset, | |
1100 | size_t size) | |
1101 | { | |
1102 | int64_t len; | |
1103 | ||
1104 | if (size > INT_MAX) { | |
1105 | return -EIO; | |
1106 | } | |
1107 | ||
1108 | if (!blk_is_available(blk)) { | |
1109 | return -ENOMEDIUM; | |
1110 | } | |
1111 | ||
1112 | if (offset < 0) { | |
1113 | return -EIO; | |
1114 | } | |
1115 | ||
1116 | if (!blk->allow_write_beyond_eof) { | |
1117 | len = blk_getlength(blk); | |
1118 | if (len < 0) { | |
1119 | return len; | |
1120 | } | |
1121 | ||
1122 | if (offset > len || len - offset < size) { | |
1123 | return -EIO; | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | int coroutine_fn blk_co_preadv(BlockBackend *blk, int64_t offset, | |
1131 | unsigned int bytes, QEMUIOVector *qiov, | |
1132 | BdrvRequestFlags flags) | |
1133 | { | |
1134 | int ret; | |
1135 | BlockDriverState *bs = blk_bs(blk); | |
1136 | ||
1137 | trace_blk_co_preadv(blk, bs, offset, bytes, flags); | |
1138 | ||
1139 | ret = blk_check_byte_request(blk, offset, bytes); | |
1140 | if (ret < 0) { | |
1141 | return ret; | |
1142 | } | |
1143 | ||
1144 | bdrv_inc_in_flight(bs); | |
1145 | ||
1146 | /* throttling disk I/O */ | |
1147 | if (blk->public.throttle_group_member.throttle_state) { | |
1148 | throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member, | |
1149 | bytes, false); | |
1150 | } | |
1151 | ||
1152 | ret = bdrv_co_preadv(blk->root, offset, bytes, qiov, flags); | |
1153 | bdrv_dec_in_flight(bs); | |
1154 | return ret; | |
1155 | } | |
1156 | ||
1157 | int coroutine_fn blk_co_pwritev(BlockBackend *blk, int64_t offset, | |
1158 | unsigned int bytes, QEMUIOVector *qiov, | |
1159 | BdrvRequestFlags flags) | |
1160 | { | |
1161 | int ret; | |
1162 | BlockDriverState *bs = blk_bs(blk); | |
1163 | ||
1164 | trace_blk_co_pwritev(blk, bs, offset, bytes, flags); | |
1165 | ||
1166 | ret = blk_check_byte_request(blk, offset, bytes); | |
1167 | if (ret < 0) { | |
1168 | return ret; | |
1169 | } | |
1170 | ||
1171 | bdrv_inc_in_flight(bs); | |
1172 | /* throttling disk I/O */ | |
1173 | if (blk->public.throttle_group_member.throttle_state) { | |
1174 | throttle_group_co_io_limits_intercept(&blk->public.throttle_group_member, | |
1175 | bytes, true); | |
1176 | } | |
1177 | ||
1178 | if (!blk->enable_write_cache) { | |
1179 | flags |= BDRV_REQ_FUA; | |
1180 | } | |
1181 | ||
1182 | ret = bdrv_co_pwritev(blk->root, offset, bytes, qiov, flags); | |
1183 | bdrv_dec_in_flight(bs); | |
1184 | return ret; | |
1185 | } | |
1186 | ||
1187 | typedef struct BlkRwCo { | |
1188 | BlockBackend *blk; | |
1189 | int64_t offset; | |
1190 | void *iobuf; | |
1191 | int ret; | |
1192 | BdrvRequestFlags flags; | |
1193 | } BlkRwCo; | |
1194 | ||
1195 | static void blk_read_entry(void *opaque) | |
1196 | { | |
1197 | BlkRwCo *rwco = opaque; | |
1198 | QEMUIOVector *qiov = rwco->iobuf; | |
1199 | ||
1200 | rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, qiov->size, | |
1201 | qiov, rwco->flags); | |
1202 | aio_wait_kick(); | |
1203 | } | |
1204 | ||
1205 | static void blk_write_entry(void *opaque) | |
1206 | { | |
1207 | BlkRwCo *rwco = opaque; | |
1208 | QEMUIOVector *qiov = rwco->iobuf; | |
1209 | ||
1210 | rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, qiov->size, | |
1211 | qiov, rwco->flags); | |
1212 | aio_wait_kick(); | |
1213 | } | |
1214 | ||
1215 | static int blk_prw(BlockBackend *blk, int64_t offset, uint8_t *buf, | |
1216 | int64_t bytes, CoroutineEntry co_entry, | |
1217 | BdrvRequestFlags flags) | |
1218 | { | |
1219 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes); | |
1220 | BlkRwCo rwco = { | |
1221 | .blk = blk, | |
1222 | .offset = offset, | |
1223 | .iobuf = &qiov, | |
1224 | .flags = flags, | |
1225 | .ret = NOT_DONE, | |
1226 | }; | |
1227 | ||
1228 | if (qemu_in_coroutine()) { | |
1229 | /* Fast-path if already in coroutine context */ | |
1230 | co_entry(&rwco); | |
1231 | } else { | |
1232 | Coroutine *co = qemu_coroutine_create(co_entry, &rwco); | |
1233 | bdrv_coroutine_enter(blk_bs(blk), co); | |
1234 | BDRV_POLL_WHILE(blk_bs(blk), rwco.ret == NOT_DONE); | |
1235 | } | |
1236 | ||
1237 | return rwco.ret; | |
1238 | } | |
1239 | ||
1240 | int blk_pread_unthrottled(BlockBackend *blk, int64_t offset, uint8_t *buf, | |
1241 | int count) | |
1242 | { | |
1243 | int ret; | |
1244 | ||
1245 | ret = blk_check_byte_request(blk, offset, count); | |
1246 | if (ret < 0) { | |
1247 | return ret; | |
1248 | } | |
1249 | ||
1250 | blk_root_drained_begin(blk->root); | |
1251 | ret = blk_pread(blk, offset, buf, count); | |
1252 | blk_root_drained_end(blk->root, NULL); | |
1253 | return ret; | |
1254 | } | |
1255 | ||
1256 | int blk_pwrite_zeroes(BlockBackend *blk, int64_t offset, | |
1257 | int bytes, BdrvRequestFlags flags) | |
1258 | { | |
1259 | return blk_prw(blk, offset, NULL, bytes, blk_write_entry, | |
1260 | flags | BDRV_REQ_ZERO_WRITE); | |
1261 | } | |
1262 | ||
1263 | int blk_make_zero(BlockBackend *blk, BdrvRequestFlags flags) | |
1264 | { | |
1265 | return bdrv_make_zero(blk->root, flags); | |
1266 | } | |
1267 | ||
1268 | void blk_inc_in_flight(BlockBackend *blk) | |
1269 | { | |
1270 | atomic_inc(&blk->in_flight); | |
1271 | } | |
1272 | ||
1273 | void blk_dec_in_flight(BlockBackend *blk) | |
1274 | { | |
1275 | atomic_dec(&blk->in_flight); | |
1276 | aio_wait_kick(); | |
1277 | } | |
1278 | ||
1279 | static void error_callback_bh(void *opaque) | |
1280 | { | |
1281 | struct BlockBackendAIOCB *acb = opaque; | |
1282 | ||
1283 | blk_dec_in_flight(acb->blk); | |
1284 | acb->common.cb(acb->common.opaque, acb->ret); | |
1285 | qemu_aio_unref(acb); | |
1286 | } | |
1287 | ||
1288 | BlockAIOCB *blk_abort_aio_request(BlockBackend *blk, | |
1289 | BlockCompletionFunc *cb, | |
1290 | void *opaque, int ret) | |
1291 | { | |
1292 | struct BlockBackendAIOCB *acb; | |
1293 | ||
1294 | blk_inc_in_flight(blk); | |
1295 | acb = blk_aio_get(&block_backend_aiocb_info, blk, cb, opaque); | |
1296 | acb->blk = blk; | |
1297 | acb->ret = ret; | |
1298 | ||
1299 | aio_bh_schedule_oneshot(blk_get_aio_context(blk), error_callback_bh, acb); | |
1300 | return &acb->common; | |
1301 | } | |
1302 | ||
1303 | typedef struct BlkAioEmAIOCB { | |
1304 | BlockAIOCB common; | |
1305 | BlkRwCo rwco; | |
1306 | int bytes; | |
1307 | bool has_returned; | |
1308 | } BlkAioEmAIOCB; | |
1309 | ||
1310 | static const AIOCBInfo blk_aio_em_aiocb_info = { | |
1311 | .aiocb_size = sizeof(BlkAioEmAIOCB), | |
1312 | }; | |
1313 | ||
1314 | static void blk_aio_complete(BlkAioEmAIOCB *acb) | |
1315 | { | |
1316 | if (acb->has_returned) { | |
1317 | acb->common.cb(acb->common.opaque, acb->rwco.ret); | |
1318 | blk_dec_in_flight(acb->rwco.blk); | |
1319 | qemu_aio_unref(acb); | |
1320 | } | |
1321 | } | |
1322 | ||
1323 | static void blk_aio_complete_bh(void *opaque) | |
1324 | { | |
1325 | BlkAioEmAIOCB *acb = opaque; | |
1326 | assert(acb->has_returned); | |
1327 | blk_aio_complete(acb); | |
1328 | } | |
1329 | ||
1330 | static BlockAIOCB *blk_aio_prwv(BlockBackend *blk, int64_t offset, int bytes, | |
1331 | void *iobuf, CoroutineEntry co_entry, | |
1332 | BdrvRequestFlags flags, | |
1333 | BlockCompletionFunc *cb, void *opaque) | |
1334 | { | |
1335 | BlkAioEmAIOCB *acb; | |
1336 | Coroutine *co; | |
1337 | ||
1338 | blk_inc_in_flight(blk); | |
1339 | acb = blk_aio_get(&blk_aio_em_aiocb_info, blk, cb, opaque); | |
1340 | acb->rwco = (BlkRwCo) { | |
1341 | .blk = blk, | |
1342 | .offset = offset, | |
1343 | .iobuf = iobuf, | |
1344 | .flags = flags, | |
1345 | .ret = NOT_DONE, | |
1346 | }; | |
1347 | acb->bytes = bytes; | |
1348 | acb->has_returned = false; | |
1349 | ||
1350 | co = qemu_coroutine_create(co_entry, acb); | |
1351 | bdrv_coroutine_enter(blk_bs(blk), co); | |
1352 | ||
1353 | acb->has_returned = true; | |
1354 | if (acb->rwco.ret != NOT_DONE) { | |
1355 | aio_bh_schedule_oneshot(blk_get_aio_context(blk), | |
1356 | blk_aio_complete_bh, acb); | |
1357 | } | |
1358 | ||
1359 | return &acb->common; | |
1360 | } | |
1361 | ||
1362 | static void blk_aio_read_entry(void *opaque) | |
1363 | { | |
1364 | BlkAioEmAIOCB *acb = opaque; | |
1365 | BlkRwCo *rwco = &acb->rwco; | |
1366 | QEMUIOVector *qiov = rwco->iobuf; | |
1367 | ||
1368 | assert(qiov->size == acb->bytes); | |
1369 | rwco->ret = blk_co_preadv(rwco->blk, rwco->offset, acb->bytes, | |
1370 | qiov, rwco->flags); | |
1371 | blk_aio_complete(acb); | |
1372 | } | |
1373 | ||
1374 | static void blk_aio_write_entry(void *opaque) | |
1375 | { | |
1376 | BlkAioEmAIOCB *acb = opaque; | |
1377 | BlkRwCo *rwco = &acb->rwco; | |
1378 | QEMUIOVector *qiov = rwco->iobuf; | |
1379 | ||
1380 | assert(!qiov || qiov->size == acb->bytes); | |
1381 | rwco->ret = blk_co_pwritev(rwco->blk, rwco->offset, acb->bytes, | |
1382 | qiov, rwco->flags); | |
1383 | blk_aio_complete(acb); | |
1384 | } | |
1385 | ||
1386 | BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset, | |
1387 | int count, BdrvRequestFlags flags, | |
1388 | BlockCompletionFunc *cb, void *opaque) | |
1389 | { | |
1390 | return blk_aio_prwv(blk, offset, count, NULL, blk_aio_write_entry, | |
1391 | flags | BDRV_REQ_ZERO_WRITE, cb, opaque); | |
1392 | } | |
1393 | ||
1394 | int blk_pread(BlockBackend *blk, int64_t offset, void *buf, int count) | |
1395 | { | |
1396 | int ret = blk_prw(blk, offset, buf, count, blk_read_entry, 0); | |
1397 | if (ret < 0) { | |
1398 | return ret; | |
1399 | } | |
1400 | return count; | |
1401 | } | |
1402 | ||
1403 | int blk_pwrite(BlockBackend *blk, int64_t offset, const void *buf, int count, | |
1404 | BdrvRequestFlags flags) | |
1405 | { | |
1406 | int ret = blk_prw(blk, offset, (void *) buf, count, blk_write_entry, | |
1407 | flags); | |
1408 | if (ret < 0) { | |
1409 | return ret; | |
1410 | } | |
1411 | return count; | |
1412 | } | |
1413 | ||
1414 | int64_t blk_getlength(BlockBackend *blk) | |
1415 | { | |
1416 | if (!blk_is_available(blk)) { | |
1417 | return -ENOMEDIUM; | |
1418 | } | |
1419 | ||
1420 | return bdrv_getlength(blk_bs(blk)); | |
1421 | } | |
1422 | ||
1423 | void blk_get_geometry(BlockBackend *blk, uint64_t *nb_sectors_ptr) | |
1424 | { | |
1425 | if (!blk_bs(blk)) { | |
1426 | *nb_sectors_ptr = 0; | |
1427 | } else { | |
1428 | bdrv_get_geometry(blk_bs(blk), nb_sectors_ptr); | |
1429 | } | |
1430 | } | |
1431 | ||
1432 | int64_t blk_nb_sectors(BlockBackend *blk) | |
1433 | { | |
1434 | if (!blk_is_available(blk)) { | |
1435 | return -ENOMEDIUM; | |
1436 | } | |
1437 | ||
1438 | return bdrv_nb_sectors(blk_bs(blk)); | |
1439 | } | |
1440 | ||
1441 | BlockAIOCB *blk_aio_preadv(BlockBackend *blk, int64_t offset, | |
1442 | QEMUIOVector *qiov, BdrvRequestFlags flags, | |
1443 | BlockCompletionFunc *cb, void *opaque) | |
1444 | { | |
1445 | return blk_aio_prwv(blk, offset, qiov->size, qiov, | |
1446 | blk_aio_read_entry, flags, cb, opaque); | |
1447 | } | |
1448 | ||
1449 | BlockAIOCB *blk_aio_pwritev(BlockBackend *blk, int64_t offset, | |
1450 | QEMUIOVector *qiov, BdrvRequestFlags flags, | |
1451 | BlockCompletionFunc *cb, void *opaque) | |
1452 | { | |
1453 | return blk_aio_prwv(blk, offset, qiov->size, qiov, | |
1454 | blk_aio_write_entry, flags, cb, opaque); | |
1455 | } | |
1456 | ||
1457 | static void blk_aio_flush_entry(void *opaque) | |
1458 | { | |
1459 | BlkAioEmAIOCB *acb = opaque; | |
1460 | BlkRwCo *rwco = &acb->rwco; | |
1461 | ||
1462 | rwco->ret = blk_co_flush(rwco->blk); | |
1463 | blk_aio_complete(acb); | |
1464 | } | |
1465 | ||
1466 | BlockAIOCB *blk_aio_flush(BlockBackend *blk, | |
1467 | BlockCompletionFunc *cb, void *opaque) | |
1468 | { | |
1469 | return blk_aio_prwv(blk, 0, 0, NULL, blk_aio_flush_entry, 0, cb, opaque); | |
1470 | } | |
1471 | ||
1472 | static void blk_aio_pdiscard_entry(void *opaque) | |
1473 | { | |
1474 | BlkAioEmAIOCB *acb = opaque; | |
1475 | BlkRwCo *rwco = &acb->rwco; | |
1476 | ||
1477 | rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, acb->bytes); | |
1478 | blk_aio_complete(acb); | |
1479 | } | |
1480 | ||
1481 | BlockAIOCB *blk_aio_pdiscard(BlockBackend *blk, | |
1482 | int64_t offset, int bytes, | |
1483 | BlockCompletionFunc *cb, void *opaque) | |
1484 | { | |
1485 | return blk_aio_prwv(blk, offset, bytes, NULL, blk_aio_pdiscard_entry, 0, | |
1486 | cb, opaque); | |
1487 | } | |
1488 | ||
1489 | void blk_aio_cancel(BlockAIOCB *acb) | |
1490 | { | |
1491 | bdrv_aio_cancel(acb); | |
1492 | } | |
1493 | ||
1494 | void blk_aio_cancel_async(BlockAIOCB *acb) | |
1495 | { | |
1496 | bdrv_aio_cancel_async(acb); | |
1497 | } | |
1498 | ||
1499 | int blk_co_ioctl(BlockBackend *blk, unsigned long int req, void *buf) | |
1500 | { | |
1501 | if (!blk_is_available(blk)) { | |
1502 | return -ENOMEDIUM; | |
1503 | } | |
1504 | ||
1505 | return bdrv_co_ioctl(blk_bs(blk), req, buf); | |
1506 | } | |
1507 | ||
1508 | static void blk_ioctl_entry(void *opaque) | |
1509 | { | |
1510 | BlkRwCo *rwco = opaque; | |
1511 | QEMUIOVector *qiov = rwco->iobuf; | |
1512 | ||
1513 | rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset, | |
1514 | qiov->iov[0].iov_base); | |
1515 | aio_wait_kick(); | |
1516 | } | |
1517 | ||
1518 | int blk_ioctl(BlockBackend *blk, unsigned long int req, void *buf) | |
1519 | { | |
1520 | return blk_prw(blk, req, buf, 0, blk_ioctl_entry, 0); | |
1521 | } | |
1522 | ||
1523 | static void blk_aio_ioctl_entry(void *opaque) | |
1524 | { | |
1525 | BlkAioEmAIOCB *acb = opaque; | |
1526 | BlkRwCo *rwco = &acb->rwco; | |
1527 | ||
1528 | rwco->ret = blk_co_ioctl(rwco->blk, rwco->offset, rwco->iobuf); | |
1529 | ||
1530 | blk_aio_complete(acb); | |
1531 | } | |
1532 | ||
1533 | BlockAIOCB *blk_aio_ioctl(BlockBackend *blk, unsigned long int req, void *buf, | |
1534 | BlockCompletionFunc *cb, void *opaque) | |
1535 | { | |
1536 | return blk_aio_prwv(blk, req, 0, buf, blk_aio_ioctl_entry, 0, cb, opaque); | |
1537 | } | |
1538 | ||
1539 | int blk_co_pdiscard(BlockBackend *blk, int64_t offset, int bytes) | |
1540 | { | |
1541 | int ret = blk_check_byte_request(blk, offset, bytes); | |
1542 | if (ret < 0) { | |
1543 | return ret; | |
1544 | } | |
1545 | ||
1546 | return bdrv_co_pdiscard(blk->root, offset, bytes); | |
1547 | } | |
1548 | ||
1549 | int blk_co_flush(BlockBackend *blk) | |
1550 | { | |
1551 | if (!blk_is_available(blk)) { | |
1552 | return -ENOMEDIUM; | |
1553 | } | |
1554 | ||
1555 | return bdrv_co_flush(blk_bs(blk)); | |
1556 | } | |
1557 | ||
1558 | static void blk_flush_entry(void *opaque) | |
1559 | { | |
1560 | BlkRwCo *rwco = opaque; | |
1561 | rwco->ret = blk_co_flush(rwco->blk); | |
1562 | aio_wait_kick(); | |
1563 | } | |
1564 | ||
1565 | int blk_flush(BlockBackend *blk) | |
1566 | { | |
1567 | return blk_prw(blk, 0, NULL, 0, blk_flush_entry, 0); | |
1568 | } | |
1569 | ||
1570 | void blk_drain(BlockBackend *blk) | |
1571 | { | |
1572 | BlockDriverState *bs = blk_bs(blk); | |
1573 | ||
1574 | if (bs) { | |
1575 | bdrv_drained_begin(bs); | |
1576 | } | |
1577 | ||
1578 | /* We may have -ENOMEDIUM completions in flight */ | |
1579 | AIO_WAIT_WHILE(blk_get_aio_context(blk), | |
1580 | atomic_mb_read(&blk->in_flight) > 0); | |
1581 | ||
1582 | if (bs) { | |
1583 | bdrv_drained_end(bs); | |
1584 | } | |
1585 | } | |
1586 | ||
1587 | void blk_drain_all(void) | |
1588 | { | |
1589 | BlockBackend *blk = NULL; | |
1590 | ||
1591 | bdrv_drain_all_begin(); | |
1592 | ||
1593 | while ((blk = blk_all_next(blk)) != NULL) { | |
1594 | AioContext *ctx = blk_get_aio_context(blk); | |
1595 | ||
1596 | aio_context_acquire(ctx); | |
1597 | ||
1598 | /* We may have -ENOMEDIUM completions in flight */ | |
1599 | AIO_WAIT_WHILE(ctx, atomic_mb_read(&blk->in_flight) > 0); | |
1600 | ||
1601 | aio_context_release(ctx); | |
1602 | } | |
1603 | ||
1604 | bdrv_drain_all_end(); | |
1605 | } | |
1606 | ||
1607 | void blk_set_on_error(BlockBackend *blk, BlockdevOnError on_read_error, | |
1608 | BlockdevOnError on_write_error) | |
1609 | { | |
1610 | blk->on_read_error = on_read_error; | |
1611 | blk->on_write_error = on_write_error; | |
1612 | } | |
1613 | ||
1614 | BlockdevOnError blk_get_on_error(BlockBackend *blk, bool is_read) | |
1615 | { | |
1616 | return is_read ? blk->on_read_error : blk->on_write_error; | |
1617 | } | |
1618 | ||
1619 | BlockErrorAction blk_get_error_action(BlockBackend *blk, bool is_read, | |
1620 | int error) | |
1621 | { | |
1622 | BlockdevOnError on_err = blk_get_on_error(blk, is_read); | |
1623 | ||
1624 | switch (on_err) { | |
1625 | case BLOCKDEV_ON_ERROR_ENOSPC: | |
1626 | return (error == ENOSPC) ? | |
1627 | BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT; | |
1628 | case BLOCKDEV_ON_ERROR_STOP: | |
1629 | return BLOCK_ERROR_ACTION_STOP; | |
1630 | case BLOCKDEV_ON_ERROR_REPORT: | |
1631 | return BLOCK_ERROR_ACTION_REPORT; | |
1632 | case BLOCKDEV_ON_ERROR_IGNORE: | |
1633 | return BLOCK_ERROR_ACTION_IGNORE; | |
1634 | case BLOCKDEV_ON_ERROR_AUTO: | |
1635 | default: | |
1636 | abort(); | |
1637 | } | |
1638 | } | |
1639 | ||
1640 | static void send_qmp_error_event(BlockBackend *blk, | |
1641 | BlockErrorAction action, | |
1642 | bool is_read, int error) | |
1643 | { | |
1644 | IoOperationType optype; | |
1645 | BlockDriverState *bs = blk_bs(blk); | |
1646 | ||
1647 | optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE; | |
1648 | qapi_event_send_block_io_error(blk_name(blk), !!bs, | |
1649 | bs ? bdrv_get_node_name(bs) : NULL, optype, | |
1650 | action, blk_iostatus_is_enabled(blk), | |
1651 | error == ENOSPC, strerror(error)); | |
1652 | } | |
1653 | ||
1654 | /* This is done by device models because, while the block layer knows | |
1655 | * about the error, it does not know whether an operation comes from | |
1656 | * the device or the block layer (from a job, for example). | |
1657 | */ | |
1658 | void blk_error_action(BlockBackend *blk, BlockErrorAction action, | |
1659 | bool is_read, int error) | |
1660 | { | |
1661 | assert(error >= 0); | |
1662 | ||
1663 | if (action == BLOCK_ERROR_ACTION_STOP) { | |
1664 | /* First set the iostatus, so that "info block" returns an iostatus | |
1665 | * that matches the events raised so far (an additional error iostatus | |
1666 | * is fine, but not a lost one). | |
1667 | */ | |
1668 | blk_iostatus_set_err(blk, error); | |
1669 | ||
1670 | /* Then raise the request to stop the VM and the event. | |
1671 | * qemu_system_vmstop_request_prepare has two effects. First, | |
1672 | * it ensures that the STOP event always comes after the | |
1673 | * BLOCK_IO_ERROR event. Second, it ensures that even if management | |
1674 | * can observe the STOP event and do a "cont" before the STOP | |
1675 | * event is issued, the VM will not stop. In this case, vm_start() | |
1676 | * also ensures that the STOP/RESUME pair of events is emitted. | |
1677 | */ | |
1678 | qemu_system_vmstop_request_prepare(); | |
1679 | send_qmp_error_event(blk, action, is_read, error); | |
1680 | qemu_system_vmstop_request(RUN_STATE_IO_ERROR); | |
1681 | } else { | |
1682 | send_qmp_error_event(blk, action, is_read, error); | |
1683 | } | |
1684 | } | |
1685 | ||
1686 | bool blk_is_read_only(BlockBackend *blk) | |
1687 | { | |
1688 | BlockDriverState *bs = blk_bs(blk); | |
1689 | ||
1690 | if (bs) { | |
1691 | return bdrv_is_read_only(bs); | |
1692 | } else { | |
1693 | return blk->root_state.read_only; | |
1694 | } | |
1695 | } | |
1696 | ||
1697 | bool blk_is_sg(BlockBackend *blk) | |
1698 | { | |
1699 | BlockDriverState *bs = blk_bs(blk); | |
1700 | ||
1701 | if (!bs) { | |
1702 | return false; | |
1703 | } | |
1704 | ||
1705 | return bdrv_is_sg(bs); | |
1706 | } | |
1707 | ||
1708 | bool blk_enable_write_cache(BlockBackend *blk) | |
1709 | { | |
1710 | return blk->enable_write_cache; | |
1711 | } | |
1712 | ||
1713 | void blk_set_enable_write_cache(BlockBackend *blk, bool wce) | |
1714 | { | |
1715 | blk->enable_write_cache = wce; | |
1716 | } | |
1717 | ||
1718 | void blk_invalidate_cache(BlockBackend *blk, Error **errp) | |
1719 | { | |
1720 | BlockDriverState *bs = blk_bs(blk); | |
1721 | ||
1722 | if (!bs) { | |
1723 | error_setg(errp, "Device '%s' has no medium", blk->name); | |
1724 | return; | |
1725 | } | |
1726 | ||
1727 | bdrv_invalidate_cache(bs, errp); | |
1728 | } | |
1729 | ||
1730 | bool blk_is_inserted(BlockBackend *blk) | |
1731 | { | |
1732 | BlockDriverState *bs = blk_bs(blk); | |
1733 | ||
1734 | return bs && bdrv_is_inserted(bs); | |
1735 | } | |
1736 | ||
1737 | bool blk_is_available(BlockBackend *blk) | |
1738 | { | |
1739 | return blk_is_inserted(blk) && !blk_dev_is_tray_open(blk); | |
1740 | } | |
1741 | ||
1742 | void blk_lock_medium(BlockBackend *blk, bool locked) | |
1743 | { | |
1744 | BlockDriverState *bs = blk_bs(blk); | |
1745 | ||
1746 | if (bs) { | |
1747 | bdrv_lock_medium(bs, locked); | |
1748 | } | |
1749 | } | |
1750 | ||
1751 | void blk_eject(BlockBackend *blk, bool eject_flag) | |
1752 | { | |
1753 | BlockDriverState *bs = blk_bs(blk); | |
1754 | char *id; | |
1755 | ||
1756 | if (bs) { | |
1757 | bdrv_eject(bs, eject_flag); | |
1758 | } | |
1759 | ||
1760 | /* Whether or not we ejected on the backend, | |
1761 | * the frontend experienced a tray event. */ | |
1762 | id = blk_get_attached_dev_id(blk); | |
1763 | qapi_event_send_device_tray_moved(blk_name(blk), id, | |
1764 | eject_flag); | |
1765 | g_free(id); | |
1766 | } | |
1767 | ||
1768 | int blk_get_flags(BlockBackend *blk) | |
1769 | { | |
1770 | BlockDriverState *bs = blk_bs(blk); | |
1771 | ||
1772 | if (bs) { | |
1773 | return bdrv_get_flags(bs); | |
1774 | } else { | |
1775 | return blk->root_state.open_flags; | |
1776 | } | |
1777 | } | |
1778 | ||
1779 | /* Returns the minimum request alignment, in bytes; guaranteed nonzero */ | |
1780 | uint32_t blk_get_request_alignment(BlockBackend *blk) | |
1781 | { | |
1782 | BlockDriverState *bs = blk_bs(blk); | |
1783 | return bs ? bs->bl.request_alignment : BDRV_SECTOR_SIZE; | |
1784 | } | |
1785 | ||
1786 | /* Returns the maximum transfer length, in bytes; guaranteed nonzero */ | |
1787 | uint32_t blk_get_max_transfer(BlockBackend *blk) | |
1788 | { | |
1789 | BlockDriverState *bs = blk_bs(blk); | |
1790 | uint32_t max = 0; | |
1791 | ||
1792 | if (bs) { | |
1793 | max = bs->bl.max_transfer; | |
1794 | } | |
1795 | return MIN_NON_ZERO(max, INT_MAX); | |
1796 | } | |
1797 | ||
1798 | int blk_get_max_iov(BlockBackend *blk) | |
1799 | { | |
1800 | return blk->root->bs->bl.max_iov; | |
1801 | } | |
1802 | ||
1803 | void blk_set_guest_block_size(BlockBackend *blk, int align) | |
1804 | { | |
1805 | blk->guest_block_size = align; | |
1806 | } | |
1807 | ||
1808 | void *blk_try_blockalign(BlockBackend *blk, size_t size) | |
1809 | { | |
1810 | return qemu_try_blockalign(blk ? blk_bs(blk) : NULL, size); | |
1811 | } | |
1812 | ||
1813 | void *blk_blockalign(BlockBackend *blk, size_t size) | |
1814 | { | |
1815 | return qemu_blockalign(blk ? blk_bs(blk) : NULL, size); | |
1816 | } | |
1817 | ||
1818 | bool blk_op_is_blocked(BlockBackend *blk, BlockOpType op, Error **errp) | |
1819 | { | |
1820 | BlockDriverState *bs = blk_bs(blk); | |
1821 | ||
1822 | if (!bs) { | |
1823 | return false; | |
1824 | } | |
1825 | ||
1826 | return bdrv_op_is_blocked(bs, op, errp); | |
1827 | } | |
1828 | ||
1829 | void blk_op_unblock(BlockBackend *blk, BlockOpType op, Error *reason) | |
1830 | { | |
1831 | BlockDriverState *bs = blk_bs(blk); | |
1832 | ||
1833 | if (bs) { | |
1834 | bdrv_op_unblock(bs, op, reason); | |
1835 | } | |
1836 | } | |
1837 | ||
1838 | void blk_op_block_all(BlockBackend *blk, Error *reason) | |
1839 | { | |
1840 | BlockDriverState *bs = blk_bs(blk); | |
1841 | ||
1842 | if (bs) { | |
1843 | bdrv_op_block_all(bs, reason); | |
1844 | } | |
1845 | } | |
1846 | ||
1847 | void blk_op_unblock_all(BlockBackend *blk, Error *reason) | |
1848 | { | |
1849 | BlockDriverState *bs = blk_bs(blk); | |
1850 | ||
1851 | if (bs) { | |
1852 | bdrv_op_unblock_all(bs, reason); | |
1853 | } | |
1854 | } | |
1855 | ||
1856 | AioContext *blk_get_aio_context(BlockBackend *blk) | |
1857 | { | |
1858 | BlockDriverState *bs = blk_bs(blk); | |
1859 | ||
1860 | if (bs) { | |
1861 | AioContext *ctx = bdrv_get_aio_context(blk_bs(blk)); | |
1862 | assert(ctx == blk->ctx); | |
1863 | } | |
1864 | ||
1865 | return blk->ctx; | |
1866 | } | |
1867 | ||
1868 | static AioContext *blk_aiocb_get_aio_context(BlockAIOCB *acb) | |
1869 | { | |
1870 | BlockBackendAIOCB *blk_acb = DO_UPCAST(BlockBackendAIOCB, common, acb); | |
1871 | return blk_get_aio_context(blk_acb->blk); | |
1872 | } | |
1873 | ||
1874 | static int blk_do_set_aio_context(BlockBackend *blk, AioContext *new_context, | |
1875 | bool update_root_node, Error **errp) | |
1876 | { | |
1877 | BlockDriverState *bs = blk_bs(blk); | |
1878 | ThrottleGroupMember *tgm = &blk->public.throttle_group_member; | |
1879 | int ret; | |
1880 | ||
1881 | if (bs) { | |
1882 | if (update_root_node) { | |
1883 | ret = bdrv_child_try_set_aio_context(bs, new_context, blk->root, | |
1884 | errp); | |
1885 | if (ret < 0) { | |
1886 | return ret; | |
1887 | } | |
1888 | } | |
1889 | if (tgm->throttle_state) { | |
1890 | bdrv_drained_begin(bs); | |
1891 | throttle_group_detach_aio_context(tgm); | |
1892 | throttle_group_attach_aio_context(tgm, new_context); | |
1893 | bdrv_drained_end(bs); | |
1894 | } | |
1895 | } | |
1896 | ||
1897 | blk->ctx = new_context; | |
1898 | return 0; | |
1899 | } | |
1900 | ||
1901 | int blk_set_aio_context(BlockBackend *blk, AioContext *new_context, | |
1902 | Error **errp) | |
1903 | { | |
1904 | return blk_do_set_aio_context(blk, new_context, true, errp); | |
1905 | } | |
1906 | ||
1907 | static bool blk_root_can_set_aio_ctx(BdrvChild *child, AioContext *ctx, | |
1908 | GSList **ignore, Error **errp) | |
1909 | { | |
1910 | BlockBackend *blk = child->opaque; | |
1911 | ||
1912 | if (blk->allow_aio_context_change) { | |
1913 | return true; | |
1914 | } | |
1915 | ||
1916 | /* Only manually created BlockBackends that are not attached to anything | |
1917 | * can change their AioContext without updating their user. */ | |
1918 | if (!blk->name || blk->dev) { | |
1919 | /* TODO Add BB name/QOM path */ | |
1920 | error_setg(errp, "Cannot change iothread of active block backend"); | |
1921 | return false; | |
1922 | } | |
1923 | ||
1924 | return true; | |
1925 | } | |
1926 | ||
1927 | static void blk_root_set_aio_ctx(BdrvChild *child, AioContext *ctx, | |
1928 | GSList **ignore) | |
1929 | { | |
1930 | BlockBackend *blk = child->opaque; | |
1931 | blk_do_set_aio_context(blk, ctx, false, &error_abort); | |
1932 | } | |
1933 | ||
1934 | void blk_add_aio_context_notifier(BlockBackend *blk, | |
1935 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
1936 | void (*detach_aio_context)(void *opaque), void *opaque) | |
1937 | { | |
1938 | BlockBackendAioNotifier *notifier; | |
1939 | BlockDriverState *bs = blk_bs(blk); | |
1940 | ||
1941 | notifier = g_new(BlockBackendAioNotifier, 1); | |
1942 | notifier->attached_aio_context = attached_aio_context; | |
1943 | notifier->detach_aio_context = detach_aio_context; | |
1944 | notifier->opaque = opaque; | |
1945 | QLIST_INSERT_HEAD(&blk->aio_notifiers, notifier, list); | |
1946 | ||
1947 | if (bs) { | |
1948 | bdrv_add_aio_context_notifier(bs, attached_aio_context, | |
1949 | detach_aio_context, opaque); | |
1950 | } | |
1951 | } | |
1952 | ||
1953 | void blk_remove_aio_context_notifier(BlockBackend *blk, | |
1954 | void (*attached_aio_context)(AioContext *, | |
1955 | void *), | |
1956 | void (*detach_aio_context)(void *), | |
1957 | void *opaque) | |
1958 | { | |
1959 | BlockBackendAioNotifier *notifier; | |
1960 | BlockDriverState *bs = blk_bs(blk); | |
1961 | ||
1962 | if (bs) { | |
1963 | bdrv_remove_aio_context_notifier(bs, attached_aio_context, | |
1964 | detach_aio_context, opaque); | |
1965 | } | |
1966 | ||
1967 | QLIST_FOREACH(notifier, &blk->aio_notifiers, list) { | |
1968 | if (notifier->attached_aio_context == attached_aio_context && | |
1969 | notifier->detach_aio_context == detach_aio_context && | |
1970 | notifier->opaque == opaque) { | |
1971 | QLIST_REMOVE(notifier, list); | |
1972 | g_free(notifier); | |
1973 | return; | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | abort(); | |
1978 | } | |
1979 | ||
1980 | void blk_add_remove_bs_notifier(BlockBackend *blk, Notifier *notify) | |
1981 | { | |
1982 | notifier_list_add(&blk->remove_bs_notifiers, notify); | |
1983 | } | |
1984 | ||
1985 | void blk_add_insert_bs_notifier(BlockBackend *blk, Notifier *notify) | |
1986 | { | |
1987 | notifier_list_add(&blk->insert_bs_notifiers, notify); | |
1988 | } | |
1989 | ||
1990 | void blk_io_plug(BlockBackend *blk) | |
1991 | { | |
1992 | BlockDriverState *bs = blk_bs(blk); | |
1993 | ||
1994 | if (bs) { | |
1995 | bdrv_io_plug(bs); | |
1996 | } | |
1997 | } | |
1998 | ||
1999 | void blk_io_unplug(BlockBackend *blk) | |
2000 | { | |
2001 | BlockDriverState *bs = blk_bs(blk); | |
2002 | ||
2003 | if (bs) { | |
2004 | bdrv_io_unplug(bs); | |
2005 | } | |
2006 | } | |
2007 | ||
2008 | BlockAcctStats *blk_get_stats(BlockBackend *blk) | |
2009 | { | |
2010 | return &blk->stats; | |
2011 | } | |
2012 | ||
2013 | void *blk_aio_get(const AIOCBInfo *aiocb_info, BlockBackend *blk, | |
2014 | BlockCompletionFunc *cb, void *opaque) | |
2015 | { | |
2016 | return qemu_aio_get(aiocb_info, blk_bs(blk), cb, opaque); | |
2017 | } | |
2018 | ||
2019 | int coroutine_fn blk_co_pwrite_zeroes(BlockBackend *blk, int64_t offset, | |
2020 | int bytes, BdrvRequestFlags flags) | |
2021 | { | |
2022 | return blk_co_pwritev(blk, offset, bytes, NULL, | |
2023 | flags | BDRV_REQ_ZERO_WRITE); | |
2024 | } | |
2025 | ||
2026 | int blk_pwrite_compressed(BlockBackend *blk, int64_t offset, const void *buf, | |
2027 | int count) | |
2028 | { | |
2029 | return blk_prw(blk, offset, (void *) buf, count, blk_write_entry, | |
2030 | BDRV_REQ_WRITE_COMPRESSED); | |
2031 | } | |
2032 | ||
2033 | int blk_truncate(BlockBackend *blk, int64_t offset, PreallocMode prealloc, | |
2034 | Error **errp) | |
2035 | { | |
2036 | if (!blk_is_available(blk)) { | |
2037 | error_setg(errp, "No medium inserted"); | |
2038 | return -ENOMEDIUM; | |
2039 | } | |
2040 | ||
2041 | return bdrv_truncate(blk->root, offset, prealloc, errp); | |
2042 | } | |
2043 | ||
2044 | static void blk_pdiscard_entry(void *opaque) | |
2045 | { | |
2046 | BlkRwCo *rwco = opaque; | |
2047 | QEMUIOVector *qiov = rwco->iobuf; | |
2048 | ||
2049 | rwco->ret = blk_co_pdiscard(rwco->blk, rwco->offset, qiov->size); | |
2050 | aio_wait_kick(); | |
2051 | } | |
2052 | ||
2053 | int blk_pdiscard(BlockBackend *blk, int64_t offset, int bytes) | |
2054 | { | |
2055 | return blk_prw(blk, offset, NULL, bytes, blk_pdiscard_entry, 0); | |
2056 | } | |
2057 | ||
2058 | int blk_save_vmstate(BlockBackend *blk, const uint8_t *buf, | |
2059 | int64_t pos, int size) | |
2060 | { | |
2061 | int ret; | |
2062 | ||
2063 | if (!blk_is_available(blk)) { | |
2064 | return -ENOMEDIUM; | |
2065 | } | |
2066 | ||
2067 | ret = bdrv_save_vmstate(blk_bs(blk), buf, pos, size); | |
2068 | if (ret < 0) { | |
2069 | return ret; | |
2070 | } | |
2071 | ||
2072 | if (ret == size && !blk->enable_write_cache) { | |
2073 | ret = bdrv_flush(blk_bs(blk)); | |
2074 | } | |
2075 | ||
2076 | return ret < 0 ? ret : size; | |
2077 | } | |
2078 | ||
2079 | int blk_load_vmstate(BlockBackend *blk, uint8_t *buf, int64_t pos, int size) | |
2080 | { | |
2081 | if (!blk_is_available(blk)) { | |
2082 | return -ENOMEDIUM; | |
2083 | } | |
2084 | ||
2085 | return bdrv_load_vmstate(blk_bs(blk), buf, pos, size); | |
2086 | } | |
2087 | ||
2088 | int blk_probe_blocksizes(BlockBackend *blk, BlockSizes *bsz) | |
2089 | { | |
2090 | if (!blk_is_available(blk)) { | |
2091 | return -ENOMEDIUM; | |
2092 | } | |
2093 | ||
2094 | return bdrv_probe_blocksizes(blk_bs(blk), bsz); | |
2095 | } | |
2096 | ||
2097 | int blk_probe_geometry(BlockBackend *blk, HDGeometry *geo) | |
2098 | { | |
2099 | if (!blk_is_available(blk)) { | |
2100 | return -ENOMEDIUM; | |
2101 | } | |
2102 | ||
2103 | return bdrv_probe_geometry(blk_bs(blk), geo); | |
2104 | } | |
2105 | ||
2106 | /* | |
2107 | * Updates the BlockBackendRootState object with data from the currently | |
2108 | * attached BlockDriverState. | |
2109 | */ | |
2110 | void blk_update_root_state(BlockBackend *blk) | |
2111 | { | |
2112 | assert(blk->root); | |
2113 | ||
2114 | blk->root_state.open_flags = blk->root->bs->open_flags; | |
2115 | blk->root_state.read_only = blk->root->bs->read_only; | |
2116 | blk->root_state.detect_zeroes = blk->root->bs->detect_zeroes; | |
2117 | } | |
2118 | ||
2119 | /* | |
2120 | * Returns the detect-zeroes setting to be used for bdrv_open() of a | |
2121 | * BlockDriverState which is supposed to inherit the root state. | |
2122 | */ | |
2123 | bool blk_get_detect_zeroes_from_root_state(BlockBackend *blk) | |
2124 | { | |
2125 | return blk->root_state.detect_zeroes; | |
2126 | } | |
2127 | ||
2128 | /* | |
2129 | * Returns the flags to be used for bdrv_open() of a BlockDriverState which is | |
2130 | * supposed to inherit the root state. | |
2131 | */ | |
2132 | int blk_get_open_flags_from_root_state(BlockBackend *blk) | |
2133 | { | |
2134 | int bs_flags; | |
2135 | ||
2136 | bs_flags = blk->root_state.read_only ? 0 : BDRV_O_RDWR; | |
2137 | bs_flags |= blk->root_state.open_flags & ~BDRV_O_RDWR; | |
2138 | ||
2139 | return bs_flags; | |
2140 | } | |
2141 | ||
2142 | BlockBackendRootState *blk_get_root_state(BlockBackend *blk) | |
2143 | { | |
2144 | return &blk->root_state; | |
2145 | } | |
2146 | ||
2147 | int blk_commit_all(void) | |
2148 | { | |
2149 | BlockBackend *blk = NULL; | |
2150 | ||
2151 | while ((blk = blk_all_next(blk)) != NULL) { | |
2152 | AioContext *aio_context = blk_get_aio_context(blk); | |
2153 | ||
2154 | aio_context_acquire(aio_context); | |
2155 | if (blk_is_inserted(blk) && blk->root->bs->backing) { | |
2156 | int ret = bdrv_commit(blk->root->bs); | |
2157 | if (ret < 0) { | |
2158 | aio_context_release(aio_context); | |
2159 | return ret; | |
2160 | } | |
2161 | } | |
2162 | aio_context_release(aio_context); | |
2163 | } | |
2164 | return 0; | |
2165 | } | |
2166 | ||
2167 | ||
2168 | /* throttling disk I/O limits */ | |
2169 | void blk_set_io_limits(BlockBackend *blk, ThrottleConfig *cfg) | |
2170 | { | |
2171 | throttle_group_config(&blk->public.throttle_group_member, cfg); | |
2172 | } | |
2173 | ||
2174 | void blk_io_limits_disable(BlockBackend *blk) | |
2175 | { | |
2176 | BlockDriverState *bs = blk_bs(blk); | |
2177 | ThrottleGroupMember *tgm = &blk->public.throttle_group_member; | |
2178 | assert(tgm->throttle_state); | |
2179 | if (bs) { | |
2180 | bdrv_drained_begin(bs); | |
2181 | } | |
2182 | throttle_group_unregister_tgm(tgm); | |
2183 | if (bs) { | |
2184 | bdrv_drained_end(bs); | |
2185 | } | |
2186 | } | |
2187 | ||
2188 | /* should be called before blk_set_io_limits if a limit is set */ | |
2189 | void blk_io_limits_enable(BlockBackend *blk, const char *group) | |
2190 | { | |
2191 | assert(!blk->public.throttle_group_member.throttle_state); | |
2192 | throttle_group_register_tgm(&blk->public.throttle_group_member, | |
2193 | group, blk_get_aio_context(blk)); | |
2194 | } | |
2195 | ||
2196 | void blk_io_limits_update_group(BlockBackend *blk, const char *group) | |
2197 | { | |
2198 | /* this BB is not part of any group */ | |
2199 | if (!blk->public.throttle_group_member.throttle_state) { | |
2200 | return; | |
2201 | } | |
2202 | ||
2203 | /* this BB is a part of the same group than the one we want */ | |
2204 | if (!g_strcmp0(throttle_group_get_name(&blk->public.throttle_group_member), | |
2205 | group)) { | |
2206 | return; | |
2207 | } | |
2208 | ||
2209 | /* need to change the group this bs belong to */ | |
2210 | blk_io_limits_disable(blk); | |
2211 | blk_io_limits_enable(blk, group); | |
2212 | } | |
2213 | ||
2214 | static void blk_root_drained_begin(BdrvChild *child) | |
2215 | { | |
2216 | BlockBackend *blk = child->opaque; | |
2217 | ||
2218 | if (++blk->quiesce_counter == 1) { | |
2219 | if (blk->dev_ops && blk->dev_ops->drained_begin) { | |
2220 | blk->dev_ops->drained_begin(blk->dev_opaque); | |
2221 | } | |
2222 | } | |
2223 | ||
2224 | /* Note that blk->root may not be accessible here yet if we are just | |
2225 | * attaching to a BlockDriverState that is drained. Use child instead. */ | |
2226 | ||
2227 | if (atomic_fetch_inc(&blk->public.throttle_group_member.io_limits_disabled) == 0) { | |
2228 | throttle_group_restart_tgm(&blk->public.throttle_group_member); | |
2229 | } | |
2230 | } | |
2231 | ||
2232 | static bool blk_root_drained_poll(BdrvChild *child) | |
2233 | { | |
2234 | BlockBackend *blk = child->opaque; | |
2235 | assert(blk->quiesce_counter); | |
2236 | return !!blk->in_flight; | |
2237 | } | |
2238 | ||
2239 | static void blk_root_drained_end(BdrvChild *child, int *drained_end_counter) | |
2240 | { | |
2241 | BlockBackend *blk = child->opaque; | |
2242 | assert(blk->quiesce_counter); | |
2243 | ||
2244 | assert(blk->public.throttle_group_member.io_limits_disabled); | |
2245 | atomic_dec(&blk->public.throttle_group_member.io_limits_disabled); | |
2246 | ||
2247 | if (--blk->quiesce_counter == 0) { | |
2248 | if (blk->dev_ops && blk->dev_ops->drained_end) { | |
2249 | blk->dev_ops->drained_end(blk->dev_opaque); | |
2250 | } | |
2251 | } | |
2252 | } | |
2253 | ||
2254 | void blk_register_buf(BlockBackend *blk, void *host, size_t size) | |
2255 | { | |
2256 | bdrv_register_buf(blk_bs(blk), host, size); | |
2257 | } | |
2258 | ||
2259 | void blk_unregister_buf(BlockBackend *blk, void *host) | |
2260 | { | |
2261 | bdrv_unregister_buf(blk_bs(blk), host); | |
2262 | } | |
2263 | ||
2264 | int coroutine_fn blk_co_copy_range(BlockBackend *blk_in, int64_t off_in, | |
2265 | BlockBackend *blk_out, int64_t off_out, | |
2266 | int bytes, BdrvRequestFlags read_flags, | |
2267 | BdrvRequestFlags write_flags) | |
2268 | { | |
2269 | int r; | |
2270 | r = blk_check_byte_request(blk_in, off_in, bytes); | |
2271 | if (r) { | |
2272 | return r; | |
2273 | } | |
2274 | r = blk_check_byte_request(blk_out, off_out, bytes); | |
2275 | if (r) { | |
2276 | return r; | |
2277 | } | |
2278 | return bdrv_co_copy_range(blk_in->root, off_in, | |
2279 | blk_out->root, off_out, | |
2280 | bytes, read_flags, write_flags); | |
2281 | } | |
2282 | ||
2283 | const BdrvChild *blk_root(BlockBackend *blk) | |
2284 | { | |
2285 | return blk->root; | |
2286 | } |