blkcg: kill blkio_list and replace blkio_list_lock with a mutex
[deliverable/linux.git] / block / blk-cgroup.c
CommitLineData
31e4c28d
VG
1/*
2 * Common Block IO controller cgroup interface
3 *
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
6 *
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
9 *
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
12 */
13#include <linux/ioprio.h>
22084190 14#include <linux/kdev_t.h>
9d6a986c 15#include <linux/module.h>
accee785 16#include <linux/err.h>
9195291e 17#include <linux/blkdev.h>
5a0e3ad6 18#include <linux/slab.h>
34d0f179 19#include <linux/genhd.h>
72e06c25 20#include <linux/delay.h>
9a9e8a26 21#include <linux/atomic.h>
72e06c25 22#include "blk-cgroup.h"
5efd6113 23#include "blk.h"
3e252066 24
84c124da
DS
25#define MAX_KEY_LEN 100
26
bc0d6501 27static DEFINE_MUTEX(blkcg_pol_mutex);
923adde1
TH
28static DEFINE_MUTEX(all_q_mutex);
29static LIST_HEAD(all_q_list);
30
3381cb8d 31struct blkio_cgroup blkio_root_cgroup = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT };
9d6a986c
VG
32EXPORT_SYMBOL_GPL(blkio_root_cgroup);
33
035d10b2
TH
34static struct blkio_policy_type *blkio_policy[BLKIO_NR_POLICIES];
35
31e4c28d
VG
36struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
37{
38 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
39 struct blkio_cgroup, css);
40}
9d6a986c 41EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
31e4c28d 42
4f85cb96 43static struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
70087dc3
VG
44{
45 return container_of(task_subsys_state(tsk, blkio_subsys_id),
46 struct blkio_cgroup, css);
47}
4f85cb96
TH
48
49struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio)
50{
51 if (bio && bio->bi_css)
52 return container_of(bio->bi_css, struct blkio_cgroup, css);
53 return task_blkio_cgroup(current);
54}
55EXPORT_SYMBOL_GPL(bio_blkio_cgroup);
70087dc3 56
0381411e
TH
57/**
58 * blkg_free - free a blkg
59 * @blkg: blkg to free
60 *
61 * Free @blkg which may be partially allocated.
62 */
63static void blkg_free(struct blkio_group *blkg)
64{
e8989fae 65 int i;
549d3aa8
TH
66
67 if (!blkg)
68 return;
69
e8989fae 70 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
9ade5ea4 71 struct blkio_policy_type *pol = blkio_policy[i];
e8989fae
TH
72 struct blkg_policy_data *pd = blkg->pd[i];
73
9ade5ea4
TH
74 if (!pd)
75 continue;
76
77 if (pol && pol->ops.blkio_exit_group_fn)
78 pol->ops.blkio_exit_group_fn(blkg);
79
9ade5ea4 80 kfree(pd);
0381411e 81 }
e8989fae 82
549d3aa8 83 kfree(blkg);
0381411e
TH
84}
85
86/**
87 * blkg_alloc - allocate a blkg
88 * @blkcg: block cgroup the new blkg is associated with
89 * @q: request_queue the new blkg is associated with
0381411e 90 *
e8989fae 91 * Allocate a new blkg assocating @blkcg and @q.
0381411e
TH
92 */
93static struct blkio_group *blkg_alloc(struct blkio_cgroup *blkcg,
e8989fae 94 struct request_queue *q)
0381411e
TH
95{
96 struct blkio_group *blkg;
e8989fae 97 int i;
0381411e
TH
98
99 /* alloc and init base part */
100 blkg = kzalloc_node(sizeof(*blkg), GFP_ATOMIC, q->node);
101 if (!blkg)
102 return NULL;
103
c875f4d0 104 blkg->q = q;
e8989fae 105 INIT_LIST_HEAD(&blkg->q_node);
0381411e 106 blkg->blkcg = blkcg;
1adaf3dd 107 blkg->refcnt = 1;
0381411e
TH
108 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
109
e8989fae
TH
110 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
111 struct blkio_policy_type *pol = blkio_policy[i];
112 struct blkg_policy_data *pd;
0381411e 113
e8989fae
TH
114 if (!pol)
115 continue;
116
117 /* alloc per-policy data and attach it to blkg */
118 pd = kzalloc_node(sizeof(*pd) + pol->pdata_size, GFP_ATOMIC,
119 q->node);
120 if (!pd) {
121 blkg_free(blkg);
122 return NULL;
123 }
549d3aa8 124
e8989fae
TH
125 blkg->pd[i] = pd;
126 pd->blkg = blkg;
0381411e
TH
127 }
128
549d3aa8 129 /* invoke per-policy init */
e8989fae
TH
130 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
131 struct blkio_policy_type *pol = blkio_policy[i];
132
133 if (pol)
134 pol->ops.blkio_init_group_fn(blkg);
135 }
136
0381411e
TH
137 return blkg;
138}
139
cd1604fa
TH
140struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
141 struct request_queue *q,
cd1604fa
TH
142 bool for_root)
143 __releases(q->queue_lock) __acquires(q->queue_lock)
5624a4e4 144{
1cd9e039 145 struct blkio_group *blkg;
5624a4e4 146
cd1604fa
TH
147 WARN_ON_ONCE(!rcu_read_lock_held());
148 lockdep_assert_held(q->queue_lock);
149
150 /*
151 * This could be the first entry point of blkcg implementation and
152 * we shouldn't allow anything to go through for a bypassing queue.
153 * The following can be removed if blkg lookup is guaranteed to
154 * fail on a bypassing queue.
155 */
156 if (unlikely(blk_queue_bypass(q)) && !for_root)
157 return ERR_PTR(blk_queue_dead(q) ? -EINVAL : -EBUSY);
158
e8989fae 159 blkg = blkg_lookup(blkcg, q);
cd1604fa
TH
160 if (blkg)
161 return blkg;
162
7ee9c562 163 /* blkg holds a reference to blkcg */
cd1604fa
TH
164 if (!css_tryget(&blkcg->css))
165 return ERR_PTR(-EINVAL);
166
167 /*
168 * Allocate and initialize.
cd1604fa 169 */
1cd9e039 170 blkg = blkg_alloc(blkcg, q);
cd1604fa
TH
171
172 /* did alloc fail? */
1cd9e039 173 if (unlikely(!blkg)) {
cd1604fa
TH
174 blkg = ERR_PTR(-ENOMEM);
175 goto out;
176 }
177
178 /* insert */
179 spin_lock(&blkcg->lock);
31e4c28d 180 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
e8989fae 181 list_add(&blkg->q_node, &q->blkg_list);
cd1604fa
TH
182 spin_unlock(&blkcg->lock);
183out:
cd1604fa 184 return blkg;
31e4c28d 185}
cd1604fa 186EXPORT_SYMBOL_GPL(blkg_lookup_create);
31e4c28d 187
31e4c28d 188/* called under rcu_read_lock(). */
cd1604fa 189struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
e8989fae 190 struct request_queue *q)
31e4c28d
VG
191{
192 struct blkio_group *blkg;
193 struct hlist_node *n;
31e4c28d 194
ca32aefc 195 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node)
e8989fae 196 if (blkg->q == q)
31e4c28d 197 return blkg;
31e4c28d
VG
198 return NULL;
199}
cd1604fa 200EXPORT_SYMBOL_GPL(blkg_lookup);
31e4c28d 201
e8989fae 202static void blkg_destroy(struct blkio_group *blkg)
03aa264a
TH
203{
204 struct request_queue *q = blkg->q;
9f13ef67 205 struct blkio_cgroup *blkcg = blkg->blkcg;
03aa264a
TH
206
207 lockdep_assert_held(q->queue_lock);
9f13ef67 208 lockdep_assert_held(&blkcg->lock);
03aa264a
TH
209
210 /* Something wrong if we are trying to remove same group twice */
e8989fae 211 WARN_ON_ONCE(list_empty(&blkg->q_node));
9f13ef67 212 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
e8989fae 213 list_del_init(&blkg->q_node);
9f13ef67 214 hlist_del_init_rcu(&blkg->blkcg_node);
03aa264a 215
03aa264a
TH
216 /*
217 * Put the reference taken at the time of creation so that when all
218 * queues are gone, group can be destroyed.
219 */
220 blkg_put(blkg);
221}
222
e8989fae
TH
223/*
224 * XXX: This updates blkg policy data in-place for root blkg, which is
225 * necessary across elevator switch and policy registration as root blkgs
226 * aren't shot down. This broken and racy implementation is temporary.
227 * Eventually, blkg shoot down will be replaced by proper in-place update.
228 */
229void update_root_blkg_pd(struct request_queue *q, enum blkio_policy_id plid)
230{
231 struct blkio_policy_type *pol = blkio_policy[plid];
232 struct blkio_group *blkg = blkg_lookup(&blkio_root_cgroup, q);
233 struct blkg_policy_data *pd;
234
235 if (!blkg)
236 return;
237
238 kfree(blkg->pd[plid]);
239 blkg->pd[plid] = NULL;
240
241 if (!pol)
242 return;
243
244 pd = kzalloc(sizeof(*pd) + pol->pdata_size, GFP_KERNEL);
245 WARN_ON_ONCE(!pd);
246
e8989fae
TH
247 blkg->pd[plid] = pd;
248 pd->blkg = blkg;
249 pol->ops.blkio_init_group_fn(blkg);
250}
251EXPORT_SYMBOL_GPL(update_root_blkg_pd);
252
9f13ef67
TH
253/**
254 * blkg_destroy_all - destroy all blkgs associated with a request_queue
255 * @q: request_queue of interest
256 * @destroy_root: whether to destroy root blkg or not
257 *
258 * Destroy blkgs associated with @q. If @destroy_root is %true, all are
259 * destroyed; otherwise, root blkg is left alone.
260 */
e8989fae 261void blkg_destroy_all(struct request_queue *q, bool destroy_root)
72e06c25 262{
03aa264a 263 struct blkio_group *blkg, *n;
72e06c25 264
9f13ef67 265 spin_lock_irq(q->queue_lock);
72e06c25 266
9f13ef67
TH
267 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
268 struct blkio_cgroup *blkcg = blkg->blkcg;
72e06c25 269
9f13ef67
TH
270 /* skip root? */
271 if (!destroy_root && blkg->blkcg == &blkio_root_cgroup)
272 continue;
72e06c25 273
9f13ef67
TH
274 spin_lock(&blkcg->lock);
275 blkg_destroy(blkg);
276 spin_unlock(&blkcg->lock);
72e06c25 277 }
9f13ef67
TH
278
279 spin_unlock_irq(q->queue_lock);
72e06c25 280}
03aa264a 281EXPORT_SYMBOL_GPL(blkg_destroy_all);
72e06c25 282
1adaf3dd
TH
283static void blkg_rcu_free(struct rcu_head *rcu_head)
284{
285 blkg_free(container_of(rcu_head, struct blkio_group, rcu_head));
286}
287
288void __blkg_release(struct blkio_group *blkg)
289{
290 /* release the extra blkcg reference this blkg has been holding */
291 css_put(&blkg->blkcg->css);
292
293 /*
294 * A group is freed in rcu manner. But having an rcu lock does not
295 * mean that one can access all the fields of blkg and assume these
296 * are valid. For example, don't try to follow throtl_data and
297 * request queue links.
298 *
299 * Having a reference to blkg under an rcu allows acess to only
300 * values local to groups like group stats and group rate limits
301 */
302 call_rcu(&blkg->rcu_head, blkg_rcu_free);
303}
304EXPORT_SYMBOL_GPL(__blkg_release);
305
303a3acb 306static int
84c124da 307blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
303a3acb 308{
997a026c 309 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
303a3acb
DS
310 struct blkio_group *blkg;
311 struct hlist_node *n;
bc0d6501 312 int i;
303a3acb 313
bc0d6501 314 mutex_lock(&blkcg_pol_mutex);
303a3acb 315 spin_lock_irq(&blkcg->lock);
997a026c
TH
316
317 /*
318 * Note that stat reset is racy - it doesn't synchronize against
319 * stat updates. This is a debug feature which shouldn't exist
320 * anyway. If you get hit by a race, retry.
321 */
303a3acb 322 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
bc0d6501
TH
323 for (i = 0; i < BLKIO_NR_POLICIES; i++) {
324 struct blkio_policy_type *pol = blkio_policy[i];
549d3aa8 325
bc0d6501 326 if (pol && pol->ops.blkio_reset_group_stats_fn)
9ade5ea4 327 pol->ops.blkio_reset_group_stats_fn(blkg);
bc0d6501 328 }
303a3acb 329 }
f0bdc8cd 330
303a3acb 331 spin_unlock_irq(&blkcg->lock);
bc0d6501 332 mutex_unlock(&blkcg_pol_mutex);
303a3acb
DS
333 return 0;
334}
335
d3d32e69 336static const char *blkg_dev_name(struct blkio_group *blkg)
303a3acb 337{
d3d32e69
TH
338 /* some drivers (floppy) instantiate a queue w/o disk registered */
339 if (blkg->q->backing_dev_info.dev)
340 return dev_name(blkg->q->backing_dev_info.dev);
341 return NULL;
303a3acb
DS
342}
343
d3d32e69
TH
344/**
345 * blkcg_print_blkgs - helper for printing per-blkg data
346 * @sf: seq_file to print to
347 * @blkcg: blkcg of interest
348 * @prfill: fill function to print out a blkg
349 * @pol: policy in question
350 * @data: data to be passed to @prfill
351 * @show_total: to print out sum of prfill return values or not
352 *
353 * This function invokes @prfill on each blkg of @blkcg if pd for the
354 * policy specified by @pol exists. @prfill is invoked with @sf, the
355 * policy data and @data. If @show_total is %true, the sum of the return
356 * values from @prfill is printed with "Total" label at the end.
357 *
358 * This is to be used to construct print functions for
359 * cftype->read_seq_string method.
360 */
829fdb50 361void blkcg_print_blkgs(struct seq_file *sf, struct blkio_cgroup *blkcg,
d366e7ec 362 u64 (*prfill)(struct seq_file *, void *, int),
829fdb50 363 int pol, int data, bool show_total)
5624a4e4 364{
d3d32e69
TH
365 struct blkio_group *blkg;
366 struct hlist_node *n;
367 u64 total = 0;
5624a4e4 368
d3d32e69
TH
369 spin_lock_irq(&blkcg->lock);
370 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node)
371 if (blkg->pd[pol])
d366e7ec 372 total += prfill(sf, blkg->pd[pol]->pdata, data);
d3d32e69
TH
373 spin_unlock_irq(&blkcg->lock);
374
375 if (show_total)
376 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
377}
829fdb50 378EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
d3d32e69
TH
379
380/**
381 * __blkg_prfill_u64 - prfill helper for a single u64 value
382 * @sf: seq_file to print to
d366e7ec 383 * @pdata: policy private data of interest
d3d32e69
TH
384 * @v: value to print
385 *
d366e7ec 386 * Print @v to @sf for the device assocaited with @pdata.
d3d32e69 387 */
d366e7ec 388u64 __blkg_prfill_u64(struct seq_file *sf, void *pdata, u64 v)
d3d32e69 389{
d366e7ec 390 const char *dname = blkg_dev_name(pdata_to_blkg(pdata));
d3d32e69
TH
391
392 if (!dname)
393 return 0;
394
395 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
396 return v;
397}
829fdb50 398EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
d3d32e69
TH
399
400/**
401 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
402 * @sf: seq_file to print to
d366e7ec 403 * @pdata: policy private data of interest
d3d32e69
TH
404 * @rwstat: rwstat to print
405 *
d366e7ec 406 * Print @rwstat to @sf for the device assocaited with @pdata.
d3d32e69 407 */
d366e7ec 408u64 __blkg_prfill_rwstat(struct seq_file *sf, void *pdata,
829fdb50 409 const struct blkg_rwstat *rwstat)
d3d32e69
TH
410{
411 static const char *rwstr[] = {
412 [BLKG_RWSTAT_READ] = "Read",
413 [BLKG_RWSTAT_WRITE] = "Write",
414 [BLKG_RWSTAT_SYNC] = "Sync",
415 [BLKG_RWSTAT_ASYNC] = "Async",
416 };
d366e7ec 417 const char *dname = blkg_dev_name(pdata_to_blkg(pdata));
d3d32e69
TH
418 u64 v;
419 int i;
420
421 if (!dname)
422 return 0;
423
424 for (i = 0; i < BLKG_RWSTAT_NR; i++)
425 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
426 (unsigned long long)rwstat->cnt[i]);
427
428 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
429 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
430 return v;
431}
432
5bc4afb1
TH
433/**
434 * blkg_prfill_stat - prfill callback for blkg_stat
435 * @sf: seq_file to print to
436 * @pdata: policy private data of interest
437 * @off: offset to the blkg_stat in @pdata
438 *
439 * prfill callback for printing a blkg_stat.
440 */
441u64 blkg_prfill_stat(struct seq_file *sf, void *pdata, int off)
d3d32e69 442{
d366e7ec 443 return __blkg_prfill_u64(sf, pdata, blkg_stat_read(pdata + off));
d3d32e69 444}
5bc4afb1 445EXPORT_SYMBOL_GPL(blkg_prfill_stat);
d3d32e69 446
5bc4afb1
TH
447/**
448 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
449 * @sf: seq_file to print to
450 * @pdata: policy private data of interest
451 * @off: offset to the blkg_rwstat in @pdata
452 *
453 * prfill callback for printing a blkg_rwstat.
454 */
455u64 blkg_prfill_rwstat(struct seq_file *sf, void *pdata, int off)
d3d32e69 456{
d366e7ec 457 struct blkg_rwstat rwstat = blkg_rwstat_read(pdata + off);
d3d32e69 458
d366e7ec 459 return __blkg_prfill_rwstat(sf, pdata, &rwstat);
d3d32e69 460}
5bc4afb1 461EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
d3d32e69 462
3a8b31d3
TH
463/**
464 * blkg_conf_prep - parse and prepare for per-blkg config update
465 * @blkcg: target block cgroup
466 * @input: input string
467 * @ctx: blkg_conf_ctx to be filled
468 *
469 * Parse per-blkg config update from @input and initialize @ctx with the
470 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
471 * value. This function returns with RCU read locked and must be paired
472 * with blkg_conf_finish().
473 */
829fdb50
TH
474int blkg_conf_prep(struct blkio_cgroup *blkcg, const char *input,
475 struct blkg_conf_ctx *ctx)
3a8b31d3 476 __acquires(rcu)
34d0f179 477{
3a8b31d3
TH
478 struct gendisk *disk;
479 struct blkio_group *blkg;
726fa694
TH
480 unsigned int major, minor;
481 unsigned long long v;
482 int part, ret;
34d0f179 483
726fa694
TH
484 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
485 return -EINVAL;
3a8b31d3 486
726fa694 487 disk = get_gendisk(MKDEV(major, minor), &part);
4bfd482e 488 if (!disk || part)
726fa694 489 return -EINVAL;
e56da7e2
TH
490
491 rcu_read_lock();
492
4bfd482e 493 spin_lock_irq(disk->queue->queue_lock);
aaec55a0 494 blkg = blkg_lookup_create(blkcg, disk->queue, false);
4bfd482e 495 spin_unlock_irq(disk->queue->queue_lock);
e56da7e2 496
4bfd482e
TH
497 if (IS_ERR(blkg)) {
498 ret = PTR_ERR(blkg);
3a8b31d3
TH
499 rcu_read_unlock();
500 put_disk(disk);
501 /*
502 * If queue was bypassing, we should retry. Do so after a
503 * short msleep(). It isn't strictly necessary but queue
504 * can be bypassing for some time and it's always nice to
505 * avoid busy looping.
506 */
507 if (ret == -EBUSY) {
508 msleep(10);
509 ret = restart_syscall();
7702e8f4 510 }
726fa694 511 return ret;
062a644d 512 }
3a8b31d3
TH
513
514 ctx->disk = disk;
515 ctx->blkg = blkg;
726fa694
TH
516 ctx->v = v;
517 return 0;
34d0f179 518}
829fdb50 519EXPORT_SYMBOL_GPL(blkg_conf_prep);
34d0f179 520
3a8b31d3
TH
521/**
522 * blkg_conf_finish - finish up per-blkg config update
523 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
524 *
525 * Finish up after per-blkg config update. This function must be paired
526 * with blkg_conf_prep().
527 */
829fdb50 528void blkg_conf_finish(struct blkg_conf_ctx *ctx)
3a8b31d3 529 __releases(rcu)
34d0f179 530{
3a8b31d3
TH
531 rcu_read_unlock();
532 put_disk(ctx->disk);
34d0f179 533}
829fdb50 534EXPORT_SYMBOL_GPL(blkg_conf_finish);
34d0f179 535
31e4c28d 536struct cftype blkio_files[] = {
84c124da
DS
537 {
538 .name = "reset_stats",
539 .write_u64 = blkiocg_reset_stats,
22084190 540 },
4baf6e33 541 { } /* terminate */
31e4c28d
VG
542};
543
9f13ef67
TH
544/**
545 * blkiocg_pre_destroy - cgroup pre_destroy callback
9f13ef67
TH
546 * @cgroup: cgroup of interest
547 *
548 * This function is called when @cgroup is about to go away and responsible
549 * for shooting down all blkgs associated with @cgroup. blkgs should be
550 * removed while holding both q and blkcg locks. As blkcg lock is nested
551 * inside q lock, this function performs reverse double lock dancing.
552 *
553 * This is the blkcg counterpart of ioc_release_fn().
554 */
959d851c 555static int blkiocg_pre_destroy(struct cgroup *cgroup)
31e4c28d
VG
556{
557 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
b1c35769 558
9f13ef67 559 spin_lock_irq(&blkcg->lock);
7ee9c562 560
9f13ef67
TH
561 while (!hlist_empty(&blkcg->blkg_list)) {
562 struct blkio_group *blkg = hlist_entry(blkcg->blkg_list.first,
563 struct blkio_group, blkcg_node);
c875f4d0 564 struct request_queue *q = blkg->q;
b1c35769 565
9f13ef67
TH
566 if (spin_trylock(q->queue_lock)) {
567 blkg_destroy(blkg);
568 spin_unlock(q->queue_lock);
569 } else {
570 spin_unlock_irq(&blkcg->lock);
9f13ef67 571 cpu_relax();
a5567932 572 spin_lock_irq(&blkcg->lock);
0f3942a3 573 }
9f13ef67 574 }
b1c35769 575
9f13ef67 576 spin_unlock_irq(&blkcg->lock);
7ee9c562
TH
577 return 0;
578}
579
959d851c 580static void blkiocg_destroy(struct cgroup *cgroup)
7ee9c562
TH
581{
582 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
583
67523c48
BB
584 if (blkcg != &blkio_root_cgroup)
585 kfree(blkcg);
31e4c28d
VG
586}
587
761b3ef5 588static struct cgroup_subsys_state *blkiocg_create(struct cgroup *cgroup)
31e4c28d 589{
9a9e8a26 590 static atomic64_t id_seq = ATOMIC64_INIT(0);
0341509f
LZ
591 struct blkio_cgroup *blkcg;
592 struct cgroup *parent = cgroup->parent;
31e4c28d 593
0341509f 594 if (!parent) {
31e4c28d
VG
595 blkcg = &blkio_root_cgroup;
596 goto done;
597 }
598
31e4c28d
VG
599 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
600 if (!blkcg)
601 return ERR_PTR(-ENOMEM);
602
3381cb8d 603 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
9a9e8a26 604 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
31e4c28d
VG
605done:
606 spin_lock_init(&blkcg->lock);
607 INIT_HLIST_HEAD(&blkcg->blkg_list);
608
609 return &blkcg->css;
610}
611
5efd6113
TH
612/**
613 * blkcg_init_queue - initialize blkcg part of request queue
614 * @q: request_queue to initialize
615 *
616 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
617 * part of new request_queue @q.
618 *
619 * RETURNS:
620 * 0 on success, -errno on failure.
621 */
622int blkcg_init_queue(struct request_queue *q)
623{
923adde1
TH
624 int ret;
625
5efd6113
TH
626 might_sleep();
627
923adde1
TH
628 ret = blk_throtl_init(q);
629 if (ret)
630 return ret;
631
632 mutex_lock(&all_q_mutex);
633 INIT_LIST_HEAD(&q->all_q_node);
634 list_add_tail(&q->all_q_node, &all_q_list);
635 mutex_unlock(&all_q_mutex);
636
637 return 0;
5efd6113
TH
638}
639
640/**
641 * blkcg_drain_queue - drain blkcg part of request_queue
642 * @q: request_queue to drain
643 *
644 * Called from blk_drain_queue(). Responsible for draining blkcg part.
645 */
646void blkcg_drain_queue(struct request_queue *q)
647{
648 lockdep_assert_held(q->queue_lock);
649
650 blk_throtl_drain(q);
651}
652
653/**
654 * blkcg_exit_queue - exit and release blkcg part of request_queue
655 * @q: request_queue being released
656 *
657 * Called from blk_release_queue(). Responsible for exiting blkcg part.
658 */
659void blkcg_exit_queue(struct request_queue *q)
660{
923adde1
TH
661 mutex_lock(&all_q_mutex);
662 list_del_init(&q->all_q_node);
663 mutex_unlock(&all_q_mutex);
664
e8989fae
TH
665 blkg_destroy_all(q, true);
666
5efd6113
TH
667 blk_throtl_exit(q);
668}
669
31e4c28d
VG
670/*
671 * We cannot support shared io contexts, as we have no mean to support
672 * two tasks with the same ioc in two different groups without major rework
673 * of the main cic data structures. For now we allow a task to change
674 * its cgroup only if it's the only owner of its ioc.
675 */
761b3ef5 676static int blkiocg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
31e4c28d 677{
bb9d97b6 678 struct task_struct *task;
31e4c28d
VG
679 struct io_context *ioc;
680 int ret = 0;
681
682 /* task_lock() is needed to avoid races with exit_io_context() */
bb9d97b6
TH
683 cgroup_taskset_for_each(task, cgrp, tset) {
684 task_lock(task);
685 ioc = task->io_context;
686 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
687 ret = -EINVAL;
688 task_unlock(task);
689 if (ret)
690 break;
691 }
31e4c28d
VG
692 return ret;
693}
694
923adde1
TH
695static void blkcg_bypass_start(void)
696 __acquires(&all_q_mutex)
697{
698 struct request_queue *q;
699
700 mutex_lock(&all_q_mutex);
701
702 list_for_each_entry(q, &all_q_list, all_q_node) {
703 blk_queue_bypass_start(q);
e8989fae 704 blkg_destroy_all(q, false);
923adde1
TH
705 }
706}
707
708static void blkcg_bypass_end(void)
709 __releases(&all_q_mutex)
710{
711 struct request_queue *q;
712
713 list_for_each_entry(q, &all_q_list, all_q_node)
714 blk_queue_bypass_end(q);
715
716 mutex_unlock(&all_q_mutex);
717}
718
676f7c8f
TH
719struct cgroup_subsys blkio_subsys = {
720 .name = "blkio",
721 .create = blkiocg_create,
722 .can_attach = blkiocg_can_attach,
959d851c 723 .pre_destroy = blkiocg_pre_destroy,
676f7c8f 724 .destroy = blkiocg_destroy,
676f7c8f 725 .subsys_id = blkio_subsys_id,
4baf6e33 726 .base_cftypes = blkio_files,
676f7c8f
TH
727 .module = THIS_MODULE,
728};
729EXPORT_SYMBOL_GPL(blkio_subsys);
730
3e252066
VG
731void blkio_policy_register(struct blkio_policy_type *blkiop)
732{
e8989fae
TH
733 struct request_queue *q;
734
bc0d6501
TH
735 mutex_lock(&blkcg_pol_mutex);
736
923adde1 737 blkcg_bypass_start();
035d10b2
TH
738
739 BUG_ON(blkio_policy[blkiop->plid]);
740 blkio_policy[blkiop->plid] = blkiop;
e8989fae
TH
741 list_for_each_entry(q, &all_q_list, all_q_node)
742 update_root_blkg_pd(q, blkiop->plid);
bc0d6501 743
923adde1 744 blkcg_bypass_end();
44ea53de
TH
745
746 if (blkiop->cftypes)
747 WARN_ON(cgroup_add_cftypes(&blkio_subsys, blkiop->cftypes));
bc0d6501
TH
748
749 mutex_unlock(&blkcg_pol_mutex);
3e252066
VG
750}
751EXPORT_SYMBOL_GPL(blkio_policy_register);
752
753void blkio_policy_unregister(struct blkio_policy_type *blkiop)
754{
e8989fae
TH
755 struct request_queue *q;
756
bc0d6501
TH
757 mutex_lock(&blkcg_pol_mutex);
758
44ea53de
TH
759 if (blkiop->cftypes)
760 cgroup_rm_cftypes(&blkio_subsys, blkiop->cftypes);
761
923adde1 762 blkcg_bypass_start();
035d10b2
TH
763
764 BUG_ON(blkio_policy[blkiop->plid] != blkiop);
765 blkio_policy[blkiop->plid] = NULL;
035d10b2 766
e8989fae
TH
767 list_for_each_entry(q, &all_q_list, all_q_node)
768 update_root_blkg_pd(q, blkiop->plid);
923adde1 769 blkcg_bypass_end();
bc0d6501
TH
770
771 mutex_unlock(&blkcg_pol_mutex);
3e252066
VG
772}
773EXPORT_SYMBOL_GPL(blkio_policy_unregister);
This page took 0.159302 seconds and 5 git commands to generate.