]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU System Emulator block driver | |
3 | * | |
4 | * Copyright (c) 2011 IBM Corp. | |
5 | * Copyright (c) 2012 Red Hat, Inc. | |
6 | * | |
7 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
8 | * of this software and associated documentation files (the "Software"), to deal | |
9 | * in the Software without restriction, including without limitation the rights | |
10 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
11 | * copies of the Software, and to permit persons to whom the Software is | |
12 | * furnished to do so, subject to the following conditions: | |
13 | * | |
14 | * The above copyright notice and this permission notice shall be included in | |
15 | * all copies or substantial portions of the Software. | |
16 | * | |
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
18 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
19 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
20 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
21 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
22 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
23 | * THE SOFTWARE. | |
24 | */ | |
25 | ||
26 | #include "qemu/osdep.h" | |
27 | #include "block/block.h" | |
28 | #include "block/blockjob_int.h" | |
29 | #include "block/block_int.h" | |
30 | #include "block/trace.h" | |
31 | #include "sysemu/block-backend.h" | |
32 | #include "qapi/error.h" | |
33 | #include "qapi/qapi-events-block-core.h" | |
34 | #include "qapi/qmp/qerror.h" | |
35 | #include "qemu/coroutine.h" | |
36 | #include "qemu/timer.h" | |
37 | ||
38 | /* | |
39 | * The block job API is composed of two categories of functions. | |
40 | * | |
41 | * The first includes functions used by the monitor. The monitor is | |
42 | * peculiar in that it accesses the block job list with block_job_get, and | |
43 | * therefore needs consistency across block_job_get and the actual operation | |
44 | * (e.g. block_job_set_speed). The consistency is achieved with | |
45 | * aio_context_acquire/release. These functions are declared in blockjob.h. | |
46 | * | |
47 | * The second includes functions used by the block job drivers and sometimes | |
48 | * by the core block layer. These do not care about locking, because the | |
49 | * whole coroutine runs under the AioContext lock, and are declared in | |
50 | * blockjob_int.h. | |
51 | */ | |
52 | ||
53 | static bool is_block_job(Job *job) | |
54 | { | |
55 | return job_type(job) == JOB_TYPE_BACKUP || | |
56 | job_type(job) == JOB_TYPE_COMMIT || | |
57 | job_type(job) == JOB_TYPE_MIRROR || | |
58 | job_type(job) == JOB_TYPE_STREAM; | |
59 | } | |
60 | ||
61 | BlockJob *block_job_next(BlockJob *bjob) | |
62 | { | |
63 | Job *job = bjob ? &bjob->job : NULL; | |
64 | ||
65 | do { | |
66 | job = job_next(job); | |
67 | } while (job && !is_block_job(job)); | |
68 | ||
69 | return job ? container_of(job, BlockJob, job) : NULL; | |
70 | } | |
71 | ||
72 | BlockJob *block_job_get(const char *id) | |
73 | { | |
74 | Job *job = job_get(id); | |
75 | ||
76 | if (job && is_block_job(job)) { | |
77 | return container_of(job, BlockJob, job); | |
78 | } else { | |
79 | return NULL; | |
80 | } | |
81 | } | |
82 | ||
83 | void block_job_free(Job *job) | |
84 | { | |
85 | BlockJob *bjob = container_of(job, BlockJob, job); | |
86 | ||
87 | block_job_remove_all_bdrv(bjob); | |
88 | blk_unref(bjob->blk); | |
89 | error_free(bjob->blocker); | |
90 | } | |
91 | ||
92 | void block_job_drain(Job *job) | |
93 | { | |
94 | BlockJob *bjob = container_of(job, BlockJob, job); | |
95 | const JobDriver *drv = job->driver; | |
96 | BlockJobDriver *bjdrv = container_of(drv, BlockJobDriver, job_driver); | |
97 | ||
98 | blk_drain(bjob->blk); | |
99 | if (bjdrv->drain) { | |
100 | bjdrv->drain(bjob); | |
101 | } | |
102 | } | |
103 | ||
104 | static char *child_job_get_parent_desc(BdrvChild *c) | |
105 | { | |
106 | BlockJob *job = c->opaque; | |
107 | return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id); | |
108 | } | |
109 | ||
110 | static void child_job_drained_begin(BdrvChild *c) | |
111 | { | |
112 | BlockJob *job = c->opaque; | |
113 | job_pause(&job->job); | |
114 | } | |
115 | ||
116 | static bool child_job_drained_poll(BdrvChild *c) | |
117 | { | |
118 | BlockJob *bjob = c->opaque; | |
119 | Job *job = &bjob->job; | |
120 | const BlockJobDriver *drv = block_job_driver(bjob); | |
121 | ||
122 | /* An inactive or completed job doesn't have any pending requests. Jobs | |
123 | * with !job->busy are either already paused or have a pause point after | |
124 | * being reentered, so no job driver code will run before they pause. */ | |
125 | if (!job->busy || job_is_completed(job)) { | |
126 | return false; | |
127 | } | |
128 | ||
129 | /* Otherwise, assume that it isn't fully stopped yet, but allow the job to | |
130 | * override this assumption. */ | |
131 | if (drv->drained_poll) { | |
132 | return drv->drained_poll(bjob); | |
133 | } else { | |
134 | return true; | |
135 | } | |
136 | } | |
137 | ||
138 | static void child_job_drained_end(BdrvChild *c) | |
139 | { | |
140 | BlockJob *job = c->opaque; | |
141 | job_resume(&job->job); | |
142 | } | |
143 | ||
144 | static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx, | |
145 | GSList **ignore, Error **errp) | |
146 | { | |
147 | BlockJob *job = c->opaque; | |
148 | GSList *l; | |
149 | ||
150 | for (l = job->nodes; l; l = l->next) { | |
151 | BdrvChild *sibling = l->data; | |
152 | if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) { | |
153 | return false; | |
154 | } | |
155 | } | |
156 | return true; | |
157 | } | |
158 | ||
159 | static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx, | |
160 | GSList **ignore) | |
161 | { | |
162 | BlockJob *job = c->opaque; | |
163 | GSList *l; | |
164 | ||
165 | for (l = job->nodes; l; l = l->next) { | |
166 | BdrvChild *sibling = l->data; | |
167 | if (g_slist_find(*ignore, sibling)) { | |
168 | continue; | |
169 | } | |
170 | *ignore = g_slist_prepend(*ignore, sibling); | |
171 | bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore); | |
172 | } | |
173 | ||
174 | job->job.aio_context = ctx; | |
175 | } | |
176 | ||
177 | static const BdrvChildRole child_job = { | |
178 | .get_parent_desc = child_job_get_parent_desc, | |
179 | .drained_begin = child_job_drained_begin, | |
180 | .drained_poll = child_job_drained_poll, | |
181 | .drained_end = child_job_drained_end, | |
182 | .can_set_aio_ctx = child_job_can_set_aio_ctx, | |
183 | .set_aio_ctx = child_job_set_aio_ctx, | |
184 | .stay_at_node = true, | |
185 | }; | |
186 | ||
187 | void block_job_remove_all_bdrv(BlockJob *job) | |
188 | { | |
189 | GSList *l; | |
190 | for (l = job->nodes; l; l = l->next) { | |
191 | BdrvChild *c = l->data; | |
192 | bdrv_op_unblock_all(c->bs, job->blocker); | |
193 | bdrv_root_unref_child(c); | |
194 | } | |
195 | g_slist_free(job->nodes); | |
196 | job->nodes = NULL; | |
197 | } | |
198 | ||
199 | bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs) | |
200 | { | |
201 | GSList *el; | |
202 | ||
203 | for (el = job->nodes; el; el = el->next) { | |
204 | BdrvChild *c = el->data; | |
205 | if (c->bs == bs) { | |
206 | return true; | |
207 | } | |
208 | } | |
209 | ||
210 | return false; | |
211 | } | |
212 | ||
213 | int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs, | |
214 | uint64_t perm, uint64_t shared_perm, Error **errp) | |
215 | { | |
216 | BdrvChild *c; | |
217 | ||
218 | bdrv_ref(bs); | |
219 | if (job->job.aio_context != qemu_get_aio_context()) { | |
220 | aio_context_release(job->job.aio_context); | |
221 | } | |
222 | c = bdrv_root_attach_child(bs, name, &child_job, job->job.aio_context, | |
223 | perm, shared_perm, job, errp); | |
224 | if (job->job.aio_context != qemu_get_aio_context()) { | |
225 | aio_context_acquire(job->job.aio_context); | |
226 | } | |
227 | if (c == NULL) { | |
228 | return -EPERM; | |
229 | } | |
230 | ||
231 | job->nodes = g_slist_prepend(job->nodes, c); | |
232 | bdrv_op_block_all(bs, job->blocker); | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
237 | static void block_job_on_idle(Notifier *n, void *opaque) | |
238 | { | |
239 | aio_wait_kick(); | |
240 | } | |
241 | ||
242 | bool block_job_is_internal(BlockJob *job) | |
243 | { | |
244 | return (job->job.id == NULL); | |
245 | } | |
246 | ||
247 | const BlockJobDriver *block_job_driver(BlockJob *job) | |
248 | { | |
249 | return container_of(job->job.driver, BlockJobDriver, job_driver); | |
250 | } | |
251 | ||
252 | /* Assumes the job_mutex is held */ | |
253 | static bool job_timer_pending(Job *job) | |
254 | { | |
255 | return timer_pending(&job->sleep_timer); | |
256 | } | |
257 | ||
258 | void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp) | |
259 | { | |
260 | int64_t old_speed = job->speed; | |
261 | ||
262 | if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) { | |
263 | return; | |
264 | } | |
265 | if (speed < 0) { | |
266 | error_setg(errp, QERR_INVALID_PARAMETER, "speed"); | |
267 | return; | |
268 | } | |
269 | ||
270 | ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME); | |
271 | ||
272 | job->speed = speed; | |
273 | if (speed && speed <= old_speed) { | |
274 | return; | |
275 | } | |
276 | ||
277 | /* kick only if a timer is pending */ | |
278 | job_enter_cond(&job->job, job_timer_pending); | |
279 | } | |
280 | ||
281 | int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n) | |
282 | { | |
283 | if (!job->speed) { | |
284 | return 0; | |
285 | } | |
286 | ||
287 | return ratelimit_calculate_delay(&job->limit, n); | |
288 | } | |
289 | ||
290 | BlockJobInfo *block_job_query(BlockJob *job, Error **errp) | |
291 | { | |
292 | BlockJobInfo *info; | |
293 | ||
294 | if (block_job_is_internal(job)) { | |
295 | error_setg(errp, "Cannot query QEMU internal jobs"); | |
296 | return NULL; | |
297 | } | |
298 | info = g_new0(BlockJobInfo, 1); | |
299 | info->type = g_strdup(job_type_str(&job->job)); | |
300 | info->device = g_strdup(job->job.id); | |
301 | info->busy = atomic_read(&job->job.busy); | |
302 | info->paused = job->job.pause_count > 0; | |
303 | info->offset = job->job.progress_current; | |
304 | info->len = job->job.progress_total; | |
305 | info->speed = job->speed; | |
306 | info->io_status = job->iostatus; | |
307 | info->ready = job_is_ready(&job->job), | |
308 | info->status = job->job.status; | |
309 | info->auto_finalize = job->job.auto_finalize; | |
310 | info->auto_dismiss = job->job.auto_dismiss; | |
311 | info->has_error = job->job.ret != 0; | |
312 | info->error = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL; | |
313 | return info; | |
314 | } | |
315 | ||
316 | static void block_job_iostatus_set_err(BlockJob *job, int error) | |
317 | { | |
318 | if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { | |
319 | job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : | |
320 | BLOCK_DEVICE_IO_STATUS_FAILED; | |
321 | } | |
322 | } | |
323 | ||
324 | static void block_job_event_cancelled(Notifier *n, void *opaque) | |
325 | { | |
326 | BlockJob *job = opaque; | |
327 | ||
328 | if (block_job_is_internal(job)) { | |
329 | return; | |
330 | } | |
331 | ||
332 | qapi_event_send_block_job_cancelled(job_type(&job->job), | |
333 | job->job.id, | |
334 | job->job.progress_total, | |
335 | job->job.progress_current, | |
336 | job->speed); | |
337 | } | |
338 | ||
339 | static void block_job_event_completed(Notifier *n, void *opaque) | |
340 | { | |
341 | BlockJob *job = opaque; | |
342 | const char *msg = NULL; | |
343 | ||
344 | if (block_job_is_internal(job)) { | |
345 | return; | |
346 | } | |
347 | ||
348 | if (job->job.ret < 0) { | |
349 | msg = strerror(-job->job.ret); | |
350 | } | |
351 | ||
352 | qapi_event_send_block_job_completed(job_type(&job->job), | |
353 | job->job.id, | |
354 | job->job.progress_total, | |
355 | job->job.progress_current, | |
356 | job->speed, | |
357 | !!msg, | |
358 | msg); | |
359 | } | |
360 | ||
361 | static void block_job_event_pending(Notifier *n, void *opaque) | |
362 | { | |
363 | BlockJob *job = opaque; | |
364 | ||
365 | if (block_job_is_internal(job)) { | |
366 | return; | |
367 | } | |
368 | ||
369 | qapi_event_send_block_job_pending(job_type(&job->job), | |
370 | job->job.id); | |
371 | } | |
372 | ||
373 | static void block_job_event_ready(Notifier *n, void *opaque) | |
374 | { | |
375 | BlockJob *job = opaque; | |
376 | ||
377 | if (block_job_is_internal(job)) { | |
378 | return; | |
379 | } | |
380 | ||
381 | qapi_event_send_block_job_ready(job_type(&job->job), | |
382 | job->job.id, | |
383 | job->job.progress_total, | |
384 | job->job.progress_current, | |
385 | job->speed); | |
386 | } | |
387 | ||
388 | ||
389 | /* | |
390 | * API for block job drivers and the block layer. These functions are | |
391 | * declared in blockjob_int.h. | |
392 | */ | |
393 | ||
394 | void *block_job_create(const char *job_id, const BlockJobDriver *driver, | |
395 | JobTxn *txn, BlockDriverState *bs, uint64_t perm, | |
396 | uint64_t shared_perm, int64_t speed, int flags, | |
397 | BlockCompletionFunc *cb, void *opaque, Error **errp) | |
398 | { | |
399 | BlockBackend *blk; | |
400 | BlockJob *job; | |
401 | int ret; | |
402 | ||
403 | if (job_id == NULL && !(flags & JOB_INTERNAL)) { | |
404 | job_id = bdrv_get_device_name(bs); | |
405 | } | |
406 | ||
407 | blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm); | |
408 | ret = blk_insert_bs(blk, bs, errp); | |
409 | if (ret < 0) { | |
410 | blk_unref(blk); | |
411 | return NULL; | |
412 | } | |
413 | ||
414 | job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk), | |
415 | flags, cb, opaque, errp); | |
416 | if (job == NULL) { | |
417 | blk_unref(blk); | |
418 | return NULL; | |
419 | } | |
420 | ||
421 | assert(is_block_job(&job->job)); | |
422 | assert(job->job.driver->free == &block_job_free); | |
423 | assert(job->job.driver->user_resume == &block_job_user_resume); | |
424 | assert(job->job.driver->drain == &block_job_drain); | |
425 | ||
426 | job->blk = blk; | |
427 | ||
428 | job->finalize_cancelled_notifier.notify = block_job_event_cancelled; | |
429 | job->finalize_completed_notifier.notify = block_job_event_completed; | |
430 | job->pending_notifier.notify = block_job_event_pending; | |
431 | job->ready_notifier.notify = block_job_event_ready; | |
432 | job->idle_notifier.notify = block_job_on_idle; | |
433 | ||
434 | notifier_list_add(&job->job.on_finalize_cancelled, | |
435 | &job->finalize_cancelled_notifier); | |
436 | notifier_list_add(&job->job.on_finalize_completed, | |
437 | &job->finalize_completed_notifier); | |
438 | notifier_list_add(&job->job.on_pending, &job->pending_notifier); | |
439 | notifier_list_add(&job->job.on_ready, &job->ready_notifier); | |
440 | notifier_list_add(&job->job.on_idle, &job->idle_notifier); | |
441 | ||
442 | error_setg(&job->blocker, "block device is in use by block job: %s", | |
443 | job_type_str(&job->job)); | |
444 | block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort); | |
445 | ||
446 | bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker); | |
447 | ||
448 | blk_set_allow_aio_context_change(blk, true); | |
449 | ||
450 | /* Only set speed when necessary to avoid NotSupported error */ | |
451 | if (speed != 0) { | |
452 | Error *local_err = NULL; | |
453 | ||
454 | block_job_set_speed(job, speed, &local_err); | |
455 | if (local_err) { | |
456 | job_early_fail(&job->job); | |
457 | error_propagate(errp, local_err); | |
458 | return NULL; | |
459 | } | |
460 | } | |
461 | ||
462 | return job; | |
463 | } | |
464 | ||
465 | void block_job_iostatus_reset(BlockJob *job) | |
466 | { | |
467 | if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { | |
468 | return; | |
469 | } | |
470 | assert(job->job.user_paused && job->job.pause_count > 0); | |
471 | job->iostatus = BLOCK_DEVICE_IO_STATUS_OK; | |
472 | } | |
473 | ||
474 | void block_job_user_resume(Job *job) | |
475 | { | |
476 | BlockJob *bjob = container_of(job, BlockJob, job); | |
477 | block_job_iostatus_reset(bjob); | |
478 | } | |
479 | ||
480 | BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err, | |
481 | int is_read, int error) | |
482 | { | |
483 | BlockErrorAction action; | |
484 | ||
485 | switch (on_err) { | |
486 | case BLOCKDEV_ON_ERROR_ENOSPC: | |
487 | case BLOCKDEV_ON_ERROR_AUTO: | |
488 | action = (error == ENOSPC) ? | |
489 | BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT; | |
490 | break; | |
491 | case BLOCKDEV_ON_ERROR_STOP: | |
492 | action = BLOCK_ERROR_ACTION_STOP; | |
493 | break; | |
494 | case BLOCKDEV_ON_ERROR_REPORT: | |
495 | action = BLOCK_ERROR_ACTION_REPORT; | |
496 | break; | |
497 | case BLOCKDEV_ON_ERROR_IGNORE: | |
498 | action = BLOCK_ERROR_ACTION_IGNORE; | |
499 | break; | |
500 | default: | |
501 | abort(); | |
502 | } | |
503 | if (!block_job_is_internal(job)) { | |
504 | qapi_event_send_block_job_error(job->job.id, | |
505 | is_read ? IO_OPERATION_TYPE_READ : | |
506 | IO_OPERATION_TYPE_WRITE, | |
507 | action); | |
508 | } | |
509 | if (action == BLOCK_ERROR_ACTION_STOP) { | |
510 | if (!job->job.user_paused) { | |
511 | job_pause(&job->job); | |
512 | /* make the pause user visible, which will be resumed from QMP. */ | |
513 | job->job.user_paused = true; | |
514 | } | |
515 | block_job_iostatus_set_err(job, error); | |
516 | } | |
517 | return action; | |
518 | } |