Merge remote-tracking branch 'remotes/ehabkost/tags/machine-next-pull-request' into...
[qemu.git] / blockjob.c
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/main-loop.h"
37 #include "qemu/timer.h"
38
39 /*
40 * The block job API is composed of two categories of functions.
41 *
42 * The first includes functions used by the monitor. The monitor is
43 * peculiar in that it accesses the block job list with block_job_get, and
44 * therefore needs consistency across block_job_get and the actual operation
45 * (e.g. block_job_set_speed). The consistency is achieved with
46 * aio_context_acquire/release. These functions are declared in blockjob.h.
47 *
48 * The second includes functions used by the block job drivers and sometimes
49 * by the core block layer. These do not care about locking, because the
50 * whole coroutine runs under the AioContext lock, and are declared in
51 * blockjob_int.h.
52 */
53
54 static bool is_block_job(Job *job)
55 {
56 return job_type(job) == JOB_TYPE_BACKUP ||
57 job_type(job) == JOB_TYPE_COMMIT ||
58 job_type(job) == JOB_TYPE_MIRROR ||
59 job_type(job) == JOB_TYPE_STREAM;
60 }
61
62 BlockJob *block_job_next(BlockJob *bjob)
63 {
64 Job *job = bjob ? &bjob->job : NULL;
65
66 do {
67 job = job_next(job);
68 } while (job && !is_block_job(job));
69
70 return job ? container_of(job, BlockJob, job) : NULL;
71 }
72
73 BlockJob *block_job_get(const char *id)
74 {
75 Job *job = job_get(id);
76
77 if (job && is_block_job(job)) {
78 return container_of(job, BlockJob, job);
79 } else {
80 return NULL;
81 }
82 }
83
84 void block_job_free(Job *job)
85 {
86 BlockJob *bjob = container_of(job, BlockJob, job);
87
88 block_job_remove_all_bdrv(bjob);
89 blk_unref(bjob->blk);
90 error_free(bjob->blocker);
91 }
92
93 static char *child_job_get_parent_desc(BdrvChild *c)
94 {
95 BlockJob *job = c->opaque;
96 return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
97 }
98
99 static void child_job_drained_begin(BdrvChild *c)
100 {
101 BlockJob *job = c->opaque;
102 job_pause(&job->job);
103 }
104
105 static bool child_job_drained_poll(BdrvChild *c)
106 {
107 BlockJob *bjob = c->opaque;
108 Job *job = &bjob->job;
109 const BlockJobDriver *drv = block_job_driver(bjob);
110
111 /* An inactive or completed job doesn't have any pending requests. Jobs
112 * with !job->busy are either already paused or have a pause point after
113 * being reentered, so no job driver code will run before they pause. */
114 if (!job->busy || job_is_completed(job)) {
115 return false;
116 }
117
118 /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
119 * override this assumption. */
120 if (drv->drained_poll) {
121 return drv->drained_poll(bjob);
122 } else {
123 return true;
124 }
125 }
126
127 static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
128 {
129 BlockJob *job = c->opaque;
130 job_resume(&job->job);
131 }
132
133 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
134 GSList **ignore, Error **errp)
135 {
136 BlockJob *job = c->opaque;
137 GSList *l;
138
139 for (l = job->nodes; l; l = l->next) {
140 BdrvChild *sibling = l->data;
141 if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
142 return false;
143 }
144 }
145 return true;
146 }
147
148 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
149 GSList **ignore)
150 {
151 BlockJob *job = c->opaque;
152 GSList *l;
153
154 for (l = job->nodes; l; l = l->next) {
155 BdrvChild *sibling = l->data;
156 if (g_slist_find(*ignore, sibling)) {
157 continue;
158 }
159 *ignore = g_slist_prepend(*ignore, sibling);
160 bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
161 }
162
163 job->job.aio_context = ctx;
164 }
165
166 static const BdrvChildRole child_job = {
167 .get_parent_desc = child_job_get_parent_desc,
168 .drained_begin = child_job_drained_begin,
169 .drained_poll = child_job_drained_poll,
170 .drained_end = child_job_drained_end,
171 .can_set_aio_ctx = child_job_can_set_aio_ctx,
172 .set_aio_ctx = child_job_set_aio_ctx,
173 .stay_at_node = true,
174 };
175
176 void block_job_remove_all_bdrv(BlockJob *job)
177 {
178 /*
179 * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
180 * which will also traverse job->nodes, so consume the list one by
181 * one to make sure that such a concurrent access does not attempt
182 * to process an already freed BdrvChild.
183 */
184 while (job->nodes) {
185 GSList *l = job->nodes;
186 BdrvChild *c = l->data;
187
188 job->nodes = l->next;
189
190 bdrv_op_unblock_all(c->bs, job->blocker);
191 bdrv_root_unref_child(c);
192
193 g_slist_free_1(l);
194 }
195 }
196
197 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
198 {
199 GSList *el;
200
201 for (el = job->nodes; el; el = el->next) {
202 BdrvChild *c = el->data;
203 if (c->bs == bs) {
204 return true;
205 }
206 }
207
208 return false;
209 }
210
211 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
212 uint64_t perm, uint64_t shared_perm, Error **errp)
213 {
214 BdrvChild *c;
215
216 bdrv_ref(bs);
217 if (job->job.aio_context != qemu_get_aio_context()) {
218 aio_context_release(job->job.aio_context);
219 }
220 c = bdrv_root_attach_child(bs, name, &child_job, job->job.aio_context,
221 perm, shared_perm, job, errp);
222 if (job->job.aio_context != qemu_get_aio_context()) {
223 aio_context_acquire(job->job.aio_context);
224 }
225 if (c == NULL) {
226 return -EPERM;
227 }
228
229 job->nodes = g_slist_prepend(job->nodes, c);
230 bdrv_op_block_all(bs, job->blocker);
231
232 return 0;
233 }
234
235 static void block_job_on_idle(Notifier *n, void *opaque)
236 {
237 aio_wait_kick();
238 }
239
240 bool block_job_is_internal(BlockJob *job)
241 {
242 return (job->job.id == NULL);
243 }
244
245 const BlockJobDriver *block_job_driver(BlockJob *job)
246 {
247 return container_of(job->job.driver, BlockJobDriver, job_driver);
248 }
249
250 /* Assumes the job_mutex is held */
251 static bool job_timer_pending(Job *job)
252 {
253 return timer_pending(&job->sleep_timer);
254 }
255
256 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
257 {
258 int64_t old_speed = job->speed;
259
260 if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
261 return;
262 }
263 if (speed < 0) {
264 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
265 "a non-negative value");
266 return;
267 }
268
269 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
270
271 job->speed = speed;
272 if (speed && speed <= old_speed) {
273 return;
274 }
275
276 /* kick only if a timer is pending */
277 job_enter_cond(&job->job, job_timer_pending);
278 }
279
280 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
281 {
282 if (!job->speed) {
283 return 0;
284 }
285
286 return ratelimit_calculate_delay(&job->limit, n);
287 }
288
289 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
290 {
291 BlockJobInfo *info;
292
293 if (block_job_is_internal(job)) {
294 error_setg(errp, "Cannot query QEMU internal jobs");
295 return NULL;
296 }
297 info = g_new0(BlockJobInfo, 1);
298 info->type = g_strdup(job_type_str(&job->job));
299 info->device = g_strdup(job->job.id);
300 info->busy = atomic_read(&job->job.busy);
301 info->paused = job->job.pause_count > 0;
302 info->offset = job->job.progress_current;
303 info->len = job->job.progress_total;
304 info->speed = job->speed;
305 info->io_status = job->iostatus;
306 info->ready = job_is_ready(&job->job),
307 info->status = job->job.status;
308 info->auto_finalize = job->job.auto_finalize;
309 info->auto_dismiss = job->job.auto_dismiss;
310 info->has_error = job->job.ret != 0;
311 info->error = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
312 return info;
313 }
314
315 static void block_job_iostatus_set_err(BlockJob *job, int error)
316 {
317 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
318 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
319 BLOCK_DEVICE_IO_STATUS_FAILED;
320 }
321 }
322
323 static void block_job_event_cancelled(Notifier *n, void *opaque)
324 {
325 BlockJob *job = opaque;
326
327 if (block_job_is_internal(job)) {
328 return;
329 }
330
331 qapi_event_send_block_job_cancelled(job_type(&job->job),
332 job->job.id,
333 job->job.progress_total,
334 job->job.progress_current,
335 job->speed);
336 }
337
338 static void block_job_event_completed(Notifier *n, void *opaque)
339 {
340 BlockJob *job = opaque;
341 const char *msg = NULL;
342
343 if (block_job_is_internal(job)) {
344 return;
345 }
346
347 if (job->job.ret < 0) {
348 msg = strerror(-job->job.ret);
349 }
350
351 qapi_event_send_block_job_completed(job_type(&job->job),
352 job->job.id,
353 job->job.progress_total,
354 job->job.progress_current,
355 job->speed,
356 !!msg,
357 msg);
358 }
359
360 static void block_job_event_pending(Notifier *n, void *opaque)
361 {
362 BlockJob *job = opaque;
363
364 if (block_job_is_internal(job)) {
365 return;
366 }
367
368 qapi_event_send_block_job_pending(job_type(&job->job),
369 job->job.id);
370 }
371
372 static void block_job_event_ready(Notifier *n, void *opaque)
373 {
374 BlockJob *job = opaque;
375
376 if (block_job_is_internal(job)) {
377 return;
378 }
379
380 qapi_event_send_block_job_ready(job_type(&job->job),
381 job->job.id,
382 job->job.progress_total,
383 job->job.progress_current,
384 job->speed);
385 }
386
387
388 /*
389 * API for block job drivers and the block layer. These functions are
390 * declared in blockjob_int.h.
391 */
392
393 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
394 JobTxn *txn, BlockDriverState *bs, uint64_t perm,
395 uint64_t shared_perm, int64_t speed, int flags,
396 BlockCompletionFunc *cb, void *opaque, Error **errp)
397 {
398 BlockBackend *blk;
399 BlockJob *job;
400 int ret;
401
402 if (job_id == NULL && !(flags & JOB_INTERNAL)) {
403 job_id = bdrv_get_device_name(bs);
404 }
405
406 blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
407 ret = blk_insert_bs(blk, bs, errp);
408 if (ret < 0) {
409 blk_unref(blk);
410 return NULL;
411 }
412
413 job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
414 flags, cb, opaque, errp);
415 if (job == NULL) {
416 blk_unref(blk);
417 return NULL;
418 }
419
420 assert(is_block_job(&job->job));
421 assert(job->job.driver->free == &block_job_free);
422 assert(job->job.driver->user_resume == &block_job_user_resume);
423
424 job->blk = blk;
425
426 job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
427 job->finalize_completed_notifier.notify = block_job_event_completed;
428 job->pending_notifier.notify = block_job_event_pending;
429 job->ready_notifier.notify = block_job_event_ready;
430 job->idle_notifier.notify = block_job_on_idle;
431
432 notifier_list_add(&job->job.on_finalize_cancelled,
433 &job->finalize_cancelled_notifier);
434 notifier_list_add(&job->job.on_finalize_completed,
435 &job->finalize_completed_notifier);
436 notifier_list_add(&job->job.on_pending, &job->pending_notifier);
437 notifier_list_add(&job->job.on_ready, &job->ready_notifier);
438 notifier_list_add(&job->job.on_idle, &job->idle_notifier);
439
440 error_setg(&job->blocker, "block device is in use by block job: %s",
441 job_type_str(&job->job));
442 block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
443
444 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
445
446 /* Disable request queuing in the BlockBackend to avoid deadlocks on drain:
447 * The job reports that it's busy until it reaches a pause point. */
448 blk_set_disable_request_queuing(blk, true);
449 blk_set_allow_aio_context_change(blk, true);
450
451 /* Only set speed when necessary to avoid NotSupported error */
452 if (speed != 0) {
453 Error *local_err = NULL;
454
455 block_job_set_speed(job, speed, &local_err);
456 if (local_err) {
457 job_early_fail(&job->job);
458 error_propagate(errp, local_err);
459 return NULL;
460 }
461 }
462
463 return job;
464 }
465
466 void block_job_iostatus_reset(BlockJob *job)
467 {
468 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
469 return;
470 }
471 assert(job->job.user_paused && job->job.pause_count > 0);
472 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
473 }
474
475 void block_job_user_resume(Job *job)
476 {
477 BlockJob *bjob = container_of(job, BlockJob, job);
478 block_job_iostatus_reset(bjob);
479 }
480
481 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
482 int is_read, int error)
483 {
484 BlockErrorAction action;
485
486 switch (on_err) {
487 case BLOCKDEV_ON_ERROR_ENOSPC:
488 case BLOCKDEV_ON_ERROR_AUTO:
489 action = (error == ENOSPC) ?
490 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
491 break;
492 case BLOCKDEV_ON_ERROR_STOP:
493 action = BLOCK_ERROR_ACTION_STOP;
494 break;
495 case BLOCKDEV_ON_ERROR_REPORT:
496 action = BLOCK_ERROR_ACTION_REPORT;
497 break;
498 case BLOCKDEV_ON_ERROR_IGNORE:
499 action = BLOCK_ERROR_ACTION_IGNORE;
500 break;
501 default:
502 abort();
503 }
504 if (!block_job_is_internal(job)) {
505 qapi_event_send_block_job_error(job->job.id,
506 is_read ? IO_OPERATION_TYPE_READ :
507 IO_OPERATION_TYPE_WRITE,
508 action);
509 }
510 if (action == BLOCK_ERROR_ACTION_STOP) {
511 if (!job->job.user_paused) {
512 job_pause(&job->job);
513 /* make the pause user visible, which will be resumed from QMP. */
514 job->job.user_paused = true;
515 }
516 block_job_iostatus_set_err(job, error);
517 }
518 return action;
519 }