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
54static 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
62BlockJob *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
73BlockJob *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
84void 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
93void block_job_drain(Job *job)
94{
95 BlockJob *bjob = container_of(job, BlockJob, job);
96 const JobDriver *drv = job->driver;
97 BlockJobDriver *bjdrv = container_of(drv, BlockJobDriver, job_driver);
98
99 blk_drain(bjob->blk);
100 if (bjdrv->drain) {
101 bjdrv->drain(bjob);
102 }
103}
104
105static char *child_job_get_parent_desc(BdrvChild *c)
106{
107 BlockJob *job = c->opaque;
108 return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
109}
110
111static void child_job_drained_begin(BdrvChild *c)
112{
113 BlockJob *job = c->opaque;
114 job_pause(&job->job);
115}
116
117static bool child_job_drained_poll(BdrvChild *c)
118{
119 BlockJob *bjob = c->opaque;
120 Job *job = &bjob->job;
121 const BlockJobDriver *drv = block_job_driver(bjob);
122
123 /* An inactive or completed job doesn't have any pending requests. Jobs
124 * with !job->busy are either already paused or have a pause point after
125 * being reentered, so no job driver code will run before they pause. */
126 if (!job->busy || job_is_completed(job)) {
127 return false;
128 }
129
130 /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
131 * override this assumption. */
132 if (drv->drained_poll) {
133 return drv->drained_poll(bjob);
134 } else {
135 return true;
136 }
137}
138
139static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
140{
141 BlockJob *job = c->opaque;
142 job_resume(&job->job);
143}
144
145static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
146 GSList **ignore, Error **errp)
147{
148 BlockJob *job = c->opaque;
149 GSList *l;
150
151 for (l = job->nodes; l; l = l->next) {
152 BdrvChild *sibling = l->data;
153 if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
154 return false;
155 }
156 }
157 return true;
158}
159
160static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
161 GSList **ignore)
162{
163 BlockJob *job = c->opaque;
164 GSList *l;
165
166 for (l = job->nodes; l; l = l->next) {
167 BdrvChild *sibling = l->data;
168 if (g_slist_find(*ignore, sibling)) {
169 continue;
170 }
171 *ignore = g_slist_prepend(*ignore, sibling);
172 bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
173 }
174
175 job->job.aio_context = ctx;
176}
177
178static const BdrvChildRole child_job = {
179 .get_parent_desc = child_job_get_parent_desc,
180 .drained_begin = child_job_drained_begin,
181 .drained_poll = child_job_drained_poll,
182 .drained_end = child_job_drained_end,
183 .can_set_aio_ctx = child_job_can_set_aio_ctx,
184 .set_aio_ctx = child_job_set_aio_ctx,
185 .stay_at_node = true,
186};
187
188void block_job_remove_all_bdrv(BlockJob *job)
189{
190 GSList *l;
191 for (l = job->nodes; l; l = l->next) {
192 BdrvChild *c = l->data;
193 bdrv_op_unblock_all(c->bs, job->blocker);
194 bdrv_root_unref_child(c);
195 }
196 g_slist_free(job->nodes);
197 job->nodes = NULL;
198}
199
200bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
201{
202 GSList *el;
203
204 for (el = job->nodes; el; el = el->next) {
205 BdrvChild *c = el->data;
206 if (c->bs == bs) {
207 return true;
208 }
209 }
210
211 return false;
212}
213
214int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
215 uint64_t perm, uint64_t shared_perm, Error **errp)
216{
217 BdrvChild *c;
218
219 bdrv_ref(bs);
220 if (job->job.aio_context != qemu_get_aio_context()) {
221 aio_context_release(job->job.aio_context);
222 }
223 c = bdrv_root_attach_child(bs, name, &child_job, job->job.aio_context,
224 perm, shared_perm, job, errp);
225 if (job->job.aio_context != qemu_get_aio_context()) {
226 aio_context_acquire(job->job.aio_context);
227 }
228 if (c == NULL) {
229 return -EPERM;
230 }
231
232 job->nodes = g_slist_prepend(job->nodes, c);
233 bdrv_op_block_all(bs, job->blocker);
234
235 return 0;
236}
237
238static void block_job_on_idle(Notifier *n, void *opaque)
239{
240 aio_wait_kick();
241}
242
243bool block_job_is_internal(BlockJob *job)
244{
245 return (job->job.id == NULL);
246}
247
248const BlockJobDriver *block_job_driver(BlockJob *job)
249{
250 return container_of(job->job.driver, BlockJobDriver, job_driver);
251}
252
253/* Assumes the job_mutex is held */
254static bool job_timer_pending(Job *job)
255{
256 return timer_pending(&job->sleep_timer);
257}
258
259void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
260{
261 int64_t old_speed = job->speed;
262
263 if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
264 return;
265 }
266 if (speed < 0) {
267 error_setg(errp, QERR_INVALID_PARAMETER, "speed");
268 return;
269 }
270
271 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
272
273 job->speed = speed;
274 if (speed && speed <= old_speed) {
275 return;
276 }
277
278 /* kick only if a timer is pending */
279 job_enter_cond(&job->job, job_timer_pending);
280}
281
282int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
283{
284 if (!job->speed) {
285 return 0;
286 }
287
288 return ratelimit_calculate_delay(&job->limit, n);
289}
290
291BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
292{
293 BlockJobInfo *info;
294
295 if (block_job_is_internal(job)) {
296 error_setg(errp, "Cannot query QEMU internal jobs");
297 return NULL;
298 }
299 info = g_new0(BlockJobInfo, 1);
300 info->type = g_strdup(job_type_str(&job->job));
301 info->device = g_strdup(job->job.id);
302 info->busy = atomic_read(&job->job.busy);
303 info->paused = job->job.pause_count > 0;
304 info->offset = job->job.progress_current;
305 info->len = job->job.progress_total;
306 info->speed = job->speed;
307 info->io_status = job->iostatus;
308 info->ready = job_is_ready(&job->job),
309 info->status = job->job.status;
310 info->auto_finalize = job->job.auto_finalize;
311 info->auto_dismiss = job->job.auto_dismiss;
312 info->has_error = job->job.ret != 0;
313 info->error = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
314 return info;
315}
316
317static void block_job_iostatus_set_err(BlockJob *job, int error)
318{
319 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
320 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
321 BLOCK_DEVICE_IO_STATUS_FAILED;
322 }
323}
324
325static void block_job_event_cancelled(Notifier *n, void *opaque)
326{
327 BlockJob *job = opaque;
328
329 if (block_job_is_internal(job)) {
330 return;
331 }
332
333 qapi_event_send_block_job_cancelled(job_type(&job->job),
334 job->job.id,
335 job->job.progress_total,
336 job->job.progress_current,
337 job->speed);
338}
339
340static void block_job_event_completed(Notifier *n, void *opaque)
341{
342 BlockJob *job = opaque;
343 const char *msg = NULL;
344
345 if (block_job_is_internal(job)) {
346 return;
347 }
348
349 if (job->job.ret < 0) {
350 msg = strerror(-job->job.ret);
351 }
352
353 qapi_event_send_block_job_completed(job_type(&job->job),
354 job->job.id,
355 job->job.progress_total,
356 job->job.progress_current,
357 job->speed,
358 !!msg,
359 msg);
360}
361
362static void block_job_event_pending(Notifier *n, void *opaque)
363{
364 BlockJob *job = opaque;
365
366 if (block_job_is_internal(job)) {
367 return;
368 }
369
370 qapi_event_send_block_job_pending(job_type(&job->job),
371 job->job.id);
372}
373
374static void block_job_event_ready(Notifier *n, void *opaque)
375{
376 BlockJob *job = opaque;
377
378 if (block_job_is_internal(job)) {
379 return;
380 }
381
382 qapi_event_send_block_job_ready(job_type(&job->job),
383 job->job.id,
384 job->job.progress_total,
385 job->job.progress_current,
386 job->speed);
387}
388
389
390/*
391 * API for block job drivers and the block layer. These functions are
392 * declared in blockjob_int.h.
393 */
394
395void *block_job_create(const char *job_id, const BlockJobDriver *driver,
396 JobTxn *txn, BlockDriverState *bs, uint64_t perm,
397 uint64_t shared_perm, int64_t speed, int flags,
398 BlockCompletionFunc *cb, void *opaque, Error **errp)
399{
400 BlockBackend *blk;
401 BlockJob *job;
402 int ret;
403
404 if (job_id == NULL && !(flags & JOB_INTERNAL)) {
405 job_id = bdrv_get_device_name(bs);
406 }
407
408 blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
409 ret = blk_insert_bs(blk, bs, errp);
410 if (ret < 0) {
411 blk_unref(blk);
412 return NULL;
413 }
414
415 job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
416 flags, cb, opaque, errp);
417 if (job == NULL) {
418 blk_unref(blk);
419 return NULL;
420 }
421
422 assert(is_block_job(&job->job));
423 assert(job->job.driver->free == &block_job_free);
424 assert(job->job.driver->user_resume == &block_job_user_resume);
425 assert(job->job.driver->drain == &block_job_drain);
426
427 job->blk = blk;
428
429 job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
430 job->finalize_completed_notifier.notify = block_job_event_completed;
431 job->pending_notifier.notify = block_job_event_pending;
432 job->ready_notifier.notify = block_job_event_ready;
433 job->idle_notifier.notify = block_job_on_idle;
434
435 notifier_list_add(&job->job.on_finalize_cancelled,
436 &job->finalize_cancelled_notifier);
437 notifier_list_add(&job->job.on_finalize_completed,
438 &job->finalize_completed_notifier);
439 notifier_list_add(&job->job.on_pending, &job->pending_notifier);
440 notifier_list_add(&job->job.on_ready, &job->ready_notifier);
441 notifier_list_add(&job->job.on_idle, &job->idle_notifier);
442
443 error_setg(&job->blocker, "block device is in use by block job: %s",
444 job_type_str(&job->job));
445 block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
446
447 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
448
449 /* Disable request queuing in the BlockBackend to avoid deadlocks on drain:
450 * The job reports that it's busy until it reaches a pause point. */
451 blk_set_disable_request_queuing(blk, true);
452 blk_set_allow_aio_context_change(blk, true);
453
454 /* Only set speed when necessary to avoid NotSupported error */
455 if (speed != 0) {
456 Error *local_err = NULL;
457
458 block_job_set_speed(job, speed, &local_err);
459 if (local_err) {
460 job_early_fail(&job->job);
461 error_propagate(errp, local_err);
462 return NULL;
463 }
464 }
465
466 return job;
467}
468
469void block_job_iostatus_reset(BlockJob *job)
470{
471 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
472 return;
473 }
474 assert(job->job.user_paused && job->job.pause_count > 0);
475 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
476}
477
478void block_job_user_resume(Job *job)
479{
480 BlockJob *bjob = container_of(job, BlockJob, job);
481 block_job_iostatus_reset(bjob);
482}
483
484BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
485 int is_read, int error)
486{
487 BlockErrorAction action;
488
489 switch (on_err) {
490 case BLOCKDEV_ON_ERROR_ENOSPC:
491 case BLOCKDEV_ON_ERROR_AUTO:
492 action = (error == ENOSPC) ?
493 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
494 break;
495 case BLOCKDEV_ON_ERROR_STOP:
496 action = BLOCK_ERROR_ACTION_STOP;
497 break;
498 case BLOCKDEV_ON_ERROR_REPORT:
499 action = BLOCK_ERROR_ACTION_REPORT;
500 break;
501 case BLOCKDEV_ON_ERROR_IGNORE:
502 action = BLOCK_ERROR_ACTION_IGNORE;
503 break;
504 default:
505 abort();
506 }
507 if (!block_job_is_internal(job)) {
508 qapi_event_send_block_job_error(job->job.id,
509 is_read ? IO_OPERATION_TYPE_READ :
510 IO_OPERATION_TYPE_WRITE,
511 action);
512 }
513 if (action == BLOCK_ERROR_ACTION_STOP) {
514 if (!job->job.user_paused) {
515 job_pause(&job->job);
516 /* make the pause user visible, which will be resumed from QMP. */
517 job->job.user_paused = true;
518 }
519 block_job_iostatus_set_err(job, error);
520 }
521 return action;
522}
523