2 * Common Block IO controller cgroup interface
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
13 #include <linux/ioprio.h>
14 #include <linux/seq_file.h>
15 #include <linux/kdev_t.h>
16 #include <linux/module.h>
17 #include <linux/err.h>
18 #include <linux/blkdev.h>
19 #include <linux/slab.h>
20 #include "blk-cgroup.h"
21 #include <linux/genhd.h>
23 #define MAX_KEY_LEN 100
25 static DEFINE_SPINLOCK(blkio_list_lock);
26 static LIST_HEAD(blkio_list);
28 struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT };
29 EXPORT_SYMBOL_GPL(blkio_root_cgroup);
31 static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *,
33 static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *,
34 struct task_struct *, bool);
35 static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *,
36 struct cgroup *, struct task_struct *, bool);
37 static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *);
38 static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *);
40 /* for encoding cft->private value on file */
41 #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val))
42 /* What policy owns the file, proportional or throttle */
43 #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff)
44 #define BLKIOFILE_ATTR(val) ((val) & 0xffff)
46 struct cgroup_subsys blkio_subsys = {
48 .create = blkiocg_create,
49 .can_attach = blkiocg_can_attach,
50 .attach = blkiocg_attach,
51 .destroy = blkiocg_destroy,
52 .populate = blkiocg_populate,
53 #ifdef CONFIG_BLK_CGROUP
54 /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */
55 .subsys_id = blkio_subsys_id,
58 .module = THIS_MODULE,
60 EXPORT_SYMBOL_GPL(blkio_subsys);
62 static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg,
63 struct blkio_policy_node *pn)
65 list_add(&pn->node, &blkcg->policy_list);
68 static inline bool cftype_blkg_same_policy(struct cftype *cft,
69 struct blkio_group *blkg)
71 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
73 if (blkg->plid == plid)
79 /* Determines if policy node matches cgroup file being accessed */
80 static inline bool pn_matches_cftype(struct cftype *cft,
81 struct blkio_policy_node *pn)
83 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
84 int fileid = BLKIOFILE_ATTR(cft->private);
86 return (plid == pn->plid && fileid == pn->fileid);
89 /* Must be called with blkcg->lock held */
90 static inline void blkio_policy_delete_node(struct blkio_policy_node *pn)
95 /* Must be called with blkcg->lock held */
96 static struct blkio_policy_node *
97 blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev,
98 enum blkio_policy_id plid, int fileid)
100 struct blkio_policy_node *pn;
102 list_for_each_entry(pn, &blkcg->policy_list, node) {
103 if (pn->dev == dev && pn->plid == plid && pn->fileid == fileid)
110 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
112 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
113 struct blkio_cgroup, css);
115 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
117 struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
119 return container_of(task_subsys_state(tsk, blkio_subsys_id),
120 struct blkio_cgroup, css);
122 EXPORT_SYMBOL_GPL(task_blkio_cgroup);
125 blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight)
127 struct blkio_policy_type *blkiop;
129 list_for_each_entry(blkiop, &blkio_list, list) {
130 /* If this policy does not own the blkg, do not send updates */
131 if (blkiop->plid != blkg->plid)
133 if (blkiop->ops.blkio_update_group_weight_fn)
134 blkiop->ops.blkio_update_group_weight_fn(blkg->key,
139 static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps,
142 struct blkio_policy_type *blkiop;
144 list_for_each_entry(blkiop, &blkio_list, list) {
146 /* If this policy does not own the blkg, do not send updates */
147 if (blkiop->plid != blkg->plid)
150 if (fileid == BLKIO_THROTL_read_bps_device
151 && blkiop->ops.blkio_update_group_read_bps_fn)
152 blkiop->ops.blkio_update_group_read_bps_fn(blkg->key,
155 if (fileid == BLKIO_THROTL_write_bps_device
156 && blkiop->ops.blkio_update_group_write_bps_fn)
157 blkiop->ops.blkio_update_group_write_bps_fn(blkg->key,
162 static inline void blkio_update_group_iops(struct blkio_group *blkg,
163 unsigned int iops, int fileid)
165 struct blkio_policy_type *blkiop;
167 list_for_each_entry(blkiop, &blkio_list, list) {
169 /* If this policy does not own the blkg, do not send updates */
170 if (blkiop->plid != blkg->plid)
173 if (fileid == BLKIO_THROTL_read_iops_device
174 && blkiop->ops.blkio_update_group_read_iops_fn)
175 blkiop->ops.blkio_update_group_read_iops_fn(blkg->key,
178 if (fileid == BLKIO_THROTL_write_iops_device
179 && blkiop->ops.blkio_update_group_write_iops_fn)
180 blkiop->ops.blkio_update_group_write_iops_fn(blkg->key,
186 * Add to the appropriate stat variable depending on the request type.
187 * This should be called with the blkg->stats_lock held.
189 static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
193 stat[BLKIO_STAT_WRITE] += add;
195 stat[BLKIO_STAT_READ] += add;
197 stat[BLKIO_STAT_SYNC] += add;
199 stat[BLKIO_STAT_ASYNC] += add;
203 * Decrements the appropriate stat variable if non-zero depending on the
204 * request type. Panics on value being zero.
205 * This should be called with the blkg->stats_lock held.
207 static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
210 BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
211 stat[BLKIO_STAT_WRITE]--;
213 BUG_ON(stat[BLKIO_STAT_READ] == 0);
214 stat[BLKIO_STAT_READ]--;
217 BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
218 stat[BLKIO_STAT_SYNC]--;
220 BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
221 stat[BLKIO_STAT_ASYNC]--;
225 #ifdef CONFIG_DEBUG_BLK_CGROUP
226 /* This should be called with the blkg->stats_lock held. */
227 static void blkio_set_start_group_wait_time(struct blkio_group *blkg,
228 struct blkio_group *curr_blkg)
230 if (blkio_blkg_waiting(&blkg->stats))
232 if (blkg == curr_blkg)
234 blkg->stats.start_group_wait_time = sched_clock();
235 blkio_mark_blkg_waiting(&blkg->stats);
238 /* This should be called with the blkg->stats_lock held. */
239 static void blkio_update_group_wait_time(struct blkio_group_stats *stats)
241 unsigned long long now;
243 if (!blkio_blkg_waiting(stats))
247 if (time_after64(now, stats->start_group_wait_time))
248 stats->group_wait_time += now - stats->start_group_wait_time;
249 blkio_clear_blkg_waiting(stats);
252 /* This should be called with the blkg->stats_lock held. */
253 static void blkio_end_empty_time(struct blkio_group_stats *stats)
255 unsigned long long now;
257 if (!blkio_blkg_empty(stats))
261 if (time_after64(now, stats->start_empty_time))
262 stats->empty_time += now - stats->start_empty_time;
263 blkio_clear_blkg_empty(stats);
266 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
270 spin_lock_irqsave(&blkg->stats_lock, flags);
271 BUG_ON(blkio_blkg_idling(&blkg->stats));
272 blkg->stats.start_idle_time = sched_clock();
273 blkio_mark_blkg_idling(&blkg->stats);
274 spin_unlock_irqrestore(&blkg->stats_lock, flags);
276 EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats);
278 void blkiocg_update_idle_time_stats(struct blkio_group *blkg)
281 unsigned long long now;
282 struct blkio_group_stats *stats;
284 spin_lock_irqsave(&blkg->stats_lock, flags);
285 stats = &blkg->stats;
286 if (blkio_blkg_idling(stats)) {
288 if (time_after64(now, stats->start_idle_time))
289 stats->idle_time += now - stats->start_idle_time;
290 blkio_clear_blkg_idling(stats);
292 spin_unlock_irqrestore(&blkg->stats_lock, flags);
294 EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats);
296 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg)
299 struct blkio_group_stats *stats;
301 spin_lock_irqsave(&blkg->stats_lock, flags);
302 stats = &blkg->stats;
303 stats->avg_queue_size_sum +=
304 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
305 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
306 stats->avg_queue_size_samples++;
307 blkio_update_group_wait_time(stats);
308 spin_unlock_irqrestore(&blkg->stats_lock, flags);
310 EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats);
312 void blkiocg_set_start_empty_time(struct blkio_group *blkg)
315 struct blkio_group_stats *stats;
317 spin_lock_irqsave(&blkg->stats_lock, flags);
318 stats = &blkg->stats;
320 if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] ||
321 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) {
322 spin_unlock_irqrestore(&blkg->stats_lock, flags);
327 * group is already marked empty. This can happen if cfqq got new
328 * request in parent group and moved to this group while being added
329 * to service tree. Just ignore the event and move on.
331 if(blkio_blkg_empty(stats)) {
332 spin_unlock_irqrestore(&blkg->stats_lock, flags);
336 stats->start_empty_time = sched_clock();
337 blkio_mark_blkg_empty(stats);
338 spin_unlock_irqrestore(&blkg->stats_lock, flags);
340 EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time);
342 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
343 unsigned long dequeue)
345 blkg->stats.dequeue += dequeue;
347 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
349 static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg,
350 struct blkio_group *curr_blkg) {}
351 static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {}
354 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
355 struct blkio_group *curr_blkg, bool direction,
360 spin_lock_irqsave(&blkg->stats_lock, flags);
361 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction,
363 blkio_end_empty_time(&blkg->stats);
364 blkio_set_start_group_wait_time(blkg, curr_blkg);
365 spin_unlock_irqrestore(&blkg->stats_lock, flags);
367 EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats);
369 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
370 bool direction, bool sync)
374 spin_lock_irqsave(&blkg->stats_lock, flags);
375 blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED],
377 spin_unlock_irqrestore(&blkg->stats_lock, flags);
379 EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats);
381 void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time,
382 unsigned long unaccounted_time)
386 spin_lock_irqsave(&blkg->stats_lock, flags);
387 blkg->stats.time += time;
388 #ifdef CONFIG_DEBUG_BLK_CGROUP
389 blkg->stats.unaccounted_time += unaccounted_time;
391 spin_unlock_irqrestore(&blkg->stats_lock, flags);
393 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
396 * should be called under rcu read lock or queue lock to make sure blkg pointer
399 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
400 uint64_t bytes, bool direction, bool sync)
402 struct blkio_group_stats_cpu *stats_cpu;
404 stats_cpu = this_cpu_ptr(blkg->stats_cpu);
406 stats_cpu->sectors += bytes >> 9;
407 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
409 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
410 bytes, direction, sync);
412 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
414 void blkiocg_update_completion_stats(struct blkio_group *blkg,
415 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
417 struct blkio_group_stats *stats;
419 unsigned long long now = sched_clock();
421 spin_lock_irqsave(&blkg->stats_lock, flags);
422 stats = &blkg->stats;
423 if (time_after64(now, io_start_time))
424 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
425 now - io_start_time, direction, sync);
426 if (time_after64(io_start_time, start_time))
427 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
428 io_start_time - start_time, direction, sync);
429 spin_unlock_irqrestore(&blkg->stats_lock, flags);
431 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
433 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
438 spin_lock_irqsave(&blkg->stats_lock, flags);
439 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_MERGED], 1, direction,
441 spin_unlock_irqrestore(&blkg->stats_lock, flags);
443 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
446 * This function allocates the per cpu stats for blkio_group. Should be called
447 * from sleepable context as alloc_per_cpu() requires that.
449 int blkio_alloc_blkg_stats(struct blkio_group *blkg)
451 /* Allocate memory for per cpu stats */
452 blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
453 if (!blkg->stats_cpu)
457 EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
459 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
460 struct blkio_group *blkg, void *key, dev_t dev,
461 enum blkio_policy_id plid)
465 spin_lock_irqsave(&blkcg->lock, flags);
466 spin_lock_init(&blkg->stats_lock);
467 rcu_assign_pointer(blkg->key, key);
468 blkg->blkcg_id = css_id(&blkcg->css);
469 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
471 spin_unlock_irqrestore(&blkcg->lock, flags);
472 /* Need to take css reference ? */
473 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
476 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
478 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
480 hlist_del_init_rcu(&blkg->blkcg_node);
485 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
486 * indicating that blk_group was unhashed by the time we got to it.
488 int blkiocg_del_blkio_group(struct blkio_group *blkg)
490 struct blkio_cgroup *blkcg;
492 struct cgroup_subsys_state *css;
496 css = css_lookup(&blkio_subsys, blkg->blkcg_id);
498 blkcg = container_of(css, struct blkio_cgroup, css);
499 spin_lock_irqsave(&blkcg->lock, flags);
500 if (!hlist_unhashed(&blkg->blkcg_node)) {
501 __blkiocg_del_blkio_group(blkg);
504 spin_unlock_irqrestore(&blkcg->lock, flags);
510 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
512 /* called under rcu_read_lock(). */
513 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
515 struct blkio_group *blkg;
516 struct hlist_node *n;
519 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
527 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
530 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
532 struct blkio_cgroup *blkcg;
533 struct blkio_group *blkg;
534 struct blkio_group_stats *stats;
535 struct hlist_node *n;
536 uint64_t queued[BLKIO_STAT_TOTAL];
538 #ifdef CONFIG_DEBUG_BLK_CGROUP
539 bool idling, waiting, empty;
540 unsigned long long now = sched_clock();
543 blkcg = cgroup_to_blkio_cgroup(cgroup);
544 spin_lock_irq(&blkcg->lock);
545 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
546 spin_lock(&blkg->stats_lock);
547 stats = &blkg->stats;
548 #ifdef CONFIG_DEBUG_BLK_CGROUP
549 idling = blkio_blkg_idling(stats);
550 waiting = blkio_blkg_waiting(stats);
551 empty = blkio_blkg_empty(stats);
553 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
554 queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
555 memset(stats, 0, sizeof(struct blkio_group_stats));
556 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
557 stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
558 #ifdef CONFIG_DEBUG_BLK_CGROUP
560 blkio_mark_blkg_idling(stats);
561 stats->start_idle_time = now;
564 blkio_mark_blkg_waiting(stats);
565 stats->start_group_wait_time = now;
568 blkio_mark_blkg_empty(stats);
569 stats->start_empty_time = now;
572 spin_unlock(&blkg->stats_lock);
574 spin_unlock_irq(&blkcg->lock);
578 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
579 int chars_left, bool diskname_only)
581 snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
582 chars_left -= strlen(str);
583 if (chars_left <= 0) {
585 "Possibly incorrect cgroup stat display format");
591 case BLKIO_STAT_READ:
592 strlcat(str, " Read", chars_left);
594 case BLKIO_STAT_WRITE:
595 strlcat(str, " Write", chars_left);
597 case BLKIO_STAT_SYNC:
598 strlcat(str, " Sync", chars_left);
600 case BLKIO_STAT_ASYNC:
601 strlcat(str, " Async", chars_left);
603 case BLKIO_STAT_TOTAL:
604 strlcat(str, " Total", chars_left);
607 strlcat(str, " Invalid", chars_left);
611 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
612 struct cgroup_map_cb *cb, dev_t dev)
614 blkio_get_key_name(0, dev, str, chars_left, true);
615 cb->fill(cb, str, val);
620 static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
621 enum stat_type_cpu type, enum stat_sub_type sub_type)
624 struct blkio_group_stats_cpu *stats_cpu;
627 for_each_possible_cpu(cpu) {
628 stats_cpu = per_cpu_ptr(blkg->stats_cpu, cpu);
630 if (type == BLKIO_STAT_CPU_SECTORS)
631 val += stats_cpu->sectors;
633 val += stats_cpu->stat_arr_cpu[type][sub_type];
639 static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
640 struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
642 uint64_t disk_total, val;
643 char key_str[MAX_KEY_LEN];
644 enum stat_sub_type sub_type;
646 if (type == BLKIO_STAT_CPU_SECTORS) {
647 val = blkio_read_stat_cpu(blkg, type, 0);
648 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
651 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
653 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
654 val = blkio_read_stat_cpu(blkg, type, sub_type);
655 cb->fill(cb, key_str, val);
658 disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
659 blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
661 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
662 cb->fill(cb, key_str, disk_total);
666 /* This should be called with blkg->stats_lock held */
667 static uint64_t blkio_get_stat(struct blkio_group *blkg,
668 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
671 char key_str[MAX_KEY_LEN];
672 enum stat_sub_type sub_type;
674 if (type == BLKIO_STAT_TIME)
675 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
676 blkg->stats.time, cb, dev);
677 #ifdef CONFIG_DEBUG_BLK_CGROUP
678 if (type == BLKIO_STAT_UNACCOUNTED_TIME)
679 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
680 blkg->stats.unaccounted_time, cb, dev);
681 if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
682 uint64_t sum = blkg->stats.avg_queue_size_sum;
683 uint64_t samples = blkg->stats.avg_queue_size_samples;
685 do_div(sum, samples);
688 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
690 if (type == BLKIO_STAT_GROUP_WAIT_TIME)
691 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
692 blkg->stats.group_wait_time, cb, dev);
693 if (type == BLKIO_STAT_IDLE_TIME)
694 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
695 blkg->stats.idle_time, cb, dev);
696 if (type == BLKIO_STAT_EMPTY_TIME)
697 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
698 blkg->stats.empty_time, cb, dev);
699 if (type == BLKIO_STAT_DEQUEUE)
700 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
701 blkg->stats.dequeue, cb, dev);
704 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
706 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
707 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
709 disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
710 blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
711 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
712 cb->fill(cb, key_str, disk_total);
716 static int blkio_check_dev_num(dev_t dev)
719 struct gendisk *disk;
721 disk = get_gendisk(dev, &part);
728 static int blkio_policy_parse_and_set(char *buf,
729 struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
731 char *s[4], *p, *major_s = NULL, *minor_s = NULL;
733 unsigned long major, minor, temp;
738 memset(s, 0, sizeof(s));
740 while ((p = strsep(&buf, " ")) != NULL) {
746 /* Prevent from inputing too many things */
754 p = strsep(&s[0], ":");
764 ret = strict_strtoul(major_s, 10, &major);
768 ret = strict_strtoul(minor_s, 10, &minor);
772 dev = MKDEV(major, minor);
774 ret = blkio_check_dev_num(dev);
784 case BLKIO_POLICY_PROP:
785 ret = strict_strtoul(s[1], 10, &temp);
786 if (ret || (temp < BLKIO_WEIGHT_MIN && temp > 0) ||
787 temp > BLKIO_WEIGHT_MAX)
791 newpn->fileid = fileid;
792 newpn->val.weight = temp;
794 case BLKIO_POLICY_THROTL:
796 case BLKIO_THROTL_read_bps_device:
797 case BLKIO_THROTL_write_bps_device:
798 ret = strict_strtoull(s[1], 10, &bps);
803 newpn->fileid = fileid;
804 newpn->val.bps = bps;
806 case BLKIO_THROTL_read_iops_device:
807 case BLKIO_THROTL_write_iops_device:
808 ret = strict_strtoull(s[1], 10, &iops);
812 if (iops > THROTL_IOPS_MAX)
816 newpn->fileid = fileid;
817 newpn->val.iops = (unsigned int)iops;
828 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
831 struct blkio_policy_node *pn;
833 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
834 BLKIO_PROP_weight_device);
836 return pn->val.weight;
838 return blkcg->weight;
840 EXPORT_SYMBOL_GPL(blkcg_get_weight);
842 uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev)
844 struct blkio_policy_node *pn;
846 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
847 BLKIO_THROTL_read_bps_device);
854 uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
856 struct blkio_policy_node *pn;
857 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
858 BLKIO_THROTL_write_bps_device);
865 unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
867 struct blkio_policy_node *pn;
869 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
870 BLKIO_THROTL_read_iops_device);
877 unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
879 struct blkio_policy_node *pn;
880 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
881 BLKIO_THROTL_write_iops_device);
888 /* Checks whether user asked for deleting a policy rule */
889 static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
892 case BLKIO_POLICY_PROP:
893 if (pn->val.weight == 0)
896 case BLKIO_POLICY_THROTL:
898 case BLKIO_THROTL_read_bps_device:
899 case BLKIO_THROTL_write_bps_device:
900 if (pn->val.bps == 0)
903 case BLKIO_THROTL_read_iops_device:
904 case BLKIO_THROTL_write_iops_device:
905 if (pn->val.iops == 0)
916 static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
917 struct blkio_policy_node *newpn)
919 switch(oldpn->plid) {
920 case BLKIO_POLICY_PROP:
921 oldpn->val.weight = newpn->val.weight;
923 case BLKIO_POLICY_THROTL:
924 switch(newpn->fileid) {
925 case BLKIO_THROTL_read_bps_device:
926 case BLKIO_THROTL_write_bps_device:
927 oldpn->val.bps = newpn->val.bps;
929 case BLKIO_THROTL_read_iops_device:
930 case BLKIO_THROTL_write_iops_device:
931 oldpn->val.iops = newpn->val.iops;
940 * Some rules/values in blkg have changed. Propagate those to respective
943 static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
944 struct blkio_group *blkg, struct blkio_policy_node *pn)
946 unsigned int weight, iops;
950 case BLKIO_POLICY_PROP:
951 weight = pn->val.weight ? pn->val.weight :
953 blkio_update_group_weight(blkg, weight);
955 case BLKIO_POLICY_THROTL:
957 case BLKIO_THROTL_read_bps_device:
958 case BLKIO_THROTL_write_bps_device:
959 bps = pn->val.bps ? pn->val.bps : (-1);
960 blkio_update_group_bps(blkg, bps, pn->fileid);
962 case BLKIO_THROTL_read_iops_device:
963 case BLKIO_THROTL_write_iops_device:
964 iops = pn->val.iops ? pn->val.iops : (-1);
965 blkio_update_group_iops(blkg, iops, pn->fileid);
975 * A policy node rule has been updated. Propagate this update to all the
976 * block groups which might be affected by this update.
978 static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
979 struct blkio_policy_node *pn)
981 struct blkio_group *blkg;
982 struct hlist_node *n;
984 spin_lock(&blkio_list_lock);
985 spin_lock_irq(&blkcg->lock);
987 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
988 if (pn->dev != blkg->dev || pn->plid != blkg->plid)
990 blkio_update_blkg_policy(blkcg, blkg, pn);
993 spin_unlock_irq(&blkcg->lock);
994 spin_unlock(&blkio_list_lock);
997 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1002 struct blkio_policy_node *newpn, *pn;
1003 struct blkio_cgroup *blkcg;
1005 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1006 int fileid = BLKIOFILE_ATTR(cft->private);
1008 buf = kstrdup(buffer, GFP_KERNEL);
1012 newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
1018 ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
1022 blkcg = cgroup_to_blkio_cgroup(cgrp);
1024 spin_lock_irq(&blkcg->lock);
1026 pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
1028 if (!blkio_delete_rule_command(newpn)) {
1029 blkio_policy_insert_node(blkcg, newpn);
1032 spin_unlock_irq(&blkcg->lock);
1033 goto update_io_group;
1036 if (blkio_delete_rule_command(newpn)) {
1037 blkio_policy_delete_node(pn);
1038 spin_unlock_irq(&blkcg->lock);
1039 goto update_io_group;
1041 spin_unlock_irq(&blkcg->lock);
1043 blkio_update_policy_rule(pn, newpn);
1046 blkio_update_policy_node_blkg(blkcg, newpn);
1057 blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
1060 case BLKIO_POLICY_PROP:
1061 if (pn->fileid == BLKIO_PROP_weight_device)
1062 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1063 MINOR(pn->dev), pn->val.weight);
1065 case BLKIO_POLICY_THROTL:
1066 switch(pn->fileid) {
1067 case BLKIO_THROTL_read_bps_device:
1068 case BLKIO_THROTL_write_bps_device:
1069 seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1070 MINOR(pn->dev), pn->val.bps);
1072 case BLKIO_THROTL_read_iops_device:
1073 case BLKIO_THROTL_write_iops_device:
1074 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1075 MINOR(pn->dev), pn->val.iops);
1084 /* cgroup files which read their data from policy nodes end up here */
1085 static void blkio_read_policy_node_files(struct cftype *cft,
1086 struct blkio_cgroup *blkcg, struct seq_file *m)
1088 struct blkio_policy_node *pn;
1090 if (!list_empty(&blkcg->policy_list)) {
1091 spin_lock_irq(&blkcg->lock);
1092 list_for_each_entry(pn, &blkcg->policy_list, node) {
1093 if (!pn_matches_cftype(cft, pn))
1095 blkio_print_policy_node(m, pn);
1097 spin_unlock_irq(&blkcg->lock);
1101 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1104 struct blkio_cgroup *blkcg;
1105 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1106 int name = BLKIOFILE_ATTR(cft->private);
1108 blkcg = cgroup_to_blkio_cgroup(cgrp);
1111 case BLKIO_POLICY_PROP:
1113 case BLKIO_PROP_weight_device:
1114 blkio_read_policy_node_files(cft, blkcg, m);
1120 case BLKIO_POLICY_THROTL:
1122 case BLKIO_THROTL_read_bps_device:
1123 case BLKIO_THROTL_write_bps_device:
1124 case BLKIO_THROTL_read_iops_device:
1125 case BLKIO_THROTL_write_iops_device:
1126 blkio_read_policy_node_files(cft, blkcg, m);
1139 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1140 struct cftype *cft, struct cgroup_map_cb *cb,
1141 enum stat_type type, bool show_total, bool pcpu)
1143 struct blkio_group *blkg;
1144 struct hlist_node *n;
1145 uint64_t cgroup_total = 0;
1148 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1150 if (!cftype_blkg_same_policy(cft, blkg))
1153 cgroup_total += blkio_get_stat_cpu(blkg, cb,
1156 spin_lock_irq(&blkg->stats_lock);
1157 cgroup_total += blkio_get_stat(blkg, cb,
1159 spin_unlock_irq(&blkg->stats_lock);
1164 cb->fill(cb, "Total", cgroup_total);
1169 /* All map kind of cgroup file get serviced by this function */
1170 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1171 struct cgroup_map_cb *cb)
1173 struct blkio_cgroup *blkcg;
1174 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1175 int name = BLKIOFILE_ATTR(cft->private);
1177 blkcg = cgroup_to_blkio_cgroup(cgrp);
1180 case BLKIO_POLICY_PROP:
1182 case BLKIO_PROP_time:
1183 return blkio_read_blkg_stats(blkcg, cft, cb,
1184 BLKIO_STAT_TIME, 0, 0);
1185 case BLKIO_PROP_sectors:
1186 return blkio_read_blkg_stats(blkcg, cft, cb,
1187 BLKIO_STAT_CPU_SECTORS, 0, 1);
1188 case BLKIO_PROP_io_service_bytes:
1189 return blkio_read_blkg_stats(blkcg, cft, cb,
1190 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1191 case BLKIO_PROP_io_serviced:
1192 return blkio_read_blkg_stats(blkcg, cft, cb,
1193 BLKIO_STAT_CPU_SERVICED, 1, 1);
1194 case BLKIO_PROP_io_service_time:
1195 return blkio_read_blkg_stats(blkcg, cft, cb,
1196 BLKIO_STAT_SERVICE_TIME, 1, 0);
1197 case BLKIO_PROP_io_wait_time:
1198 return blkio_read_blkg_stats(blkcg, cft, cb,
1199 BLKIO_STAT_WAIT_TIME, 1, 0);
1200 case BLKIO_PROP_io_merged:
1201 return blkio_read_blkg_stats(blkcg, cft, cb,
1202 BLKIO_STAT_MERGED, 1, 0);
1203 case BLKIO_PROP_io_queued:
1204 return blkio_read_blkg_stats(blkcg, cft, cb,
1205 BLKIO_STAT_QUEUED, 1, 0);
1206 #ifdef CONFIG_DEBUG_BLK_CGROUP
1207 case BLKIO_PROP_unaccounted_time:
1208 return blkio_read_blkg_stats(blkcg, cft, cb,
1209 BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
1210 case BLKIO_PROP_dequeue:
1211 return blkio_read_blkg_stats(blkcg, cft, cb,
1212 BLKIO_STAT_DEQUEUE, 0, 0);
1213 case BLKIO_PROP_avg_queue_size:
1214 return blkio_read_blkg_stats(blkcg, cft, cb,
1215 BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
1216 case BLKIO_PROP_group_wait_time:
1217 return blkio_read_blkg_stats(blkcg, cft, cb,
1218 BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
1219 case BLKIO_PROP_idle_time:
1220 return blkio_read_blkg_stats(blkcg, cft, cb,
1221 BLKIO_STAT_IDLE_TIME, 0, 0);
1222 case BLKIO_PROP_empty_time:
1223 return blkio_read_blkg_stats(blkcg, cft, cb,
1224 BLKIO_STAT_EMPTY_TIME, 0, 0);
1230 case BLKIO_POLICY_THROTL:
1232 case BLKIO_THROTL_io_service_bytes:
1233 return blkio_read_blkg_stats(blkcg, cft, cb,
1234 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1235 case BLKIO_THROTL_io_serviced:
1236 return blkio_read_blkg_stats(blkcg, cft, cb,
1237 BLKIO_STAT_CPU_SERVICED, 1, 1);
1249 static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1251 struct blkio_group *blkg;
1252 struct hlist_node *n;
1253 struct blkio_policy_node *pn;
1255 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1258 spin_lock(&blkio_list_lock);
1259 spin_lock_irq(&blkcg->lock);
1260 blkcg->weight = (unsigned int)val;
1262 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1263 pn = blkio_policy_search_node(blkcg, blkg->dev,
1264 BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1268 blkio_update_group_weight(blkg, blkcg->weight);
1270 spin_unlock_irq(&blkcg->lock);
1271 spin_unlock(&blkio_list_lock);
1275 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1276 struct blkio_cgroup *blkcg;
1277 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1278 int name = BLKIOFILE_ATTR(cft->private);
1280 blkcg = cgroup_to_blkio_cgroup(cgrp);
1283 case BLKIO_POLICY_PROP:
1285 case BLKIO_PROP_weight:
1286 return (u64)blkcg->weight;
1296 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1298 struct blkio_cgroup *blkcg;
1299 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1300 int name = BLKIOFILE_ATTR(cft->private);
1302 blkcg = cgroup_to_blkio_cgroup(cgrp);
1305 case BLKIO_POLICY_PROP:
1307 case BLKIO_PROP_weight:
1308 return blkio_weight_write(blkcg, val);
1318 struct cftype blkio_files[] = {
1320 .name = "weight_device",
1321 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1322 BLKIO_PROP_weight_device),
1323 .read_seq_string = blkiocg_file_read,
1324 .write_string = blkiocg_file_write,
1325 .max_write_len = 256,
1329 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1331 .read_u64 = blkiocg_file_read_u64,
1332 .write_u64 = blkiocg_file_write_u64,
1336 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1338 .read_map = blkiocg_file_read_map,
1342 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1343 BLKIO_PROP_sectors),
1344 .read_map = blkiocg_file_read_map,
1347 .name = "io_service_bytes",
1348 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1349 BLKIO_PROP_io_service_bytes),
1350 .read_map = blkiocg_file_read_map,
1353 .name = "io_serviced",
1354 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1355 BLKIO_PROP_io_serviced),
1356 .read_map = blkiocg_file_read_map,
1359 .name = "io_service_time",
1360 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1361 BLKIO_PROP_io_service_time),
1362 .read_map = blkiocg_file_read_map,
1365 .name = "io_wait_time",
1366 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1367 BLKIO_PROP_io_wait_time),
1368 .read_map = blkiocg_file_read_map,
1371 .name = "io_merged",
1372 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1373 BLKIO_PROP_io_merged),
1374 .read_map = blkiocg_file_read_map,
1377 .name = "io_queued",
1378 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1379 BLKIO_PROP_io_queued),
1380 .read_map = blkiocg_file_read_map,
1383 .name = "reset_stats",
1384 .write_u64 = blkiocg_reset_stats,
1386 #ifdef CONFIG_BLK_DEV_THROTTLING
1388 .name = "throttle.read_bps_device",
1389 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1390 BLKIO_THROTL_read_bps_device),
1391 .read_seq_string = blkiocg_file_read,
1392 .write_string = blkiocg_file_write,
1393 .max_write_len = 256,
1397 .name = "throttle.write_bps_device",
1398 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1399 BLKIO_THROTL_write_bps_device),
1400 .read_seq_string = blkiocg_file_read,
1401 .write_string = blkiocg_file_write,
1402 .max_write_len = 256,
1406 .name = "throttle.read_iops_device",
1407 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1408 BLKIO_THROTL_read_iops_device),
1409 .read_seq_string = blkiocg_file_read,
1410 .write_string = blkiocg_file_write,
1411 .max_write_len = 256,
1415 .name = "throttle.write_iops_device",
1416 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1417 BLKIO_THROTL_write_iops_device),
1418 .read_seq_string = blkiocg_file_read,
1419 .write_string = blkiocg_file_write,
1420 .max_write_len = 256,
1423 .name = "throttle.io_service_bytes",
1424 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1425 BLKIO_THROTL_io_service_bytes),
1426 .read_map = blkiocg_file_read_map,
1429 .name = "throttle.io_serviced",
1430 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1431 BLKIO_THROTL_io_serviced),
1432 .read_map = blkiocg_file_read_map,
1434 #endif /* CONFIG_BLK_DEV_THROTTLING */
1436 #ifdef CONFIG_DEBUG_BLK_CGROUP
1438 .name = "avg_queue_size",
1439 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1440 BLKIO_PROP_avg_queue_size),
1441 .read_map = blkiocg_file_read_map,
1444 .name = "group_wait_time",
1445 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1446 BLKIO_PROP_group_wait_time),
1447 .read_map = blkiocg_file_read_map,
1450 .name = "idle_time",
1451 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1452 BLKIO_PROP_idle_time),
1453 .read_map = blkiocg_file_read_map,
1456 .name = "empty_time",
1457 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1458 BLKIO_PROP_empty_time),
1459 .read_map = blkiocg_file_read_map,
1463 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1464 BLKIO_PROP_dequeue),
1465 .read_map = blkiocg_file_read_map,
1468 .name = "unaccounted_time",
1469 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1470 BLKIO_PROP_unaccounted_time),
1471 .read_map = blkiocg_file_read_map,
1476 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1478 return cgroup_add_files(cgroup, subsys, blkio_files,
1479 ARRAY_SIZE(blkio_files));
1482 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1484 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1485 unsigned long flags;
1486 struct blkio_group *blkg;
1488 struct blkio_policy_type *blkiop;
1489 struct blkio_policy_node *pn, *pntmp;
1493 spin_lock_irqsave(&blkcg->lock, flags);
1495 if (hlist_empty(&blkcg->blkg_list)) {
1496 spin_unlock_irqrestore(&blkcg->lock, flags);
1500 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1502 key = rcu_dereference(blkg->key);
1503 __blkiocg_del_blkio_group(blkg);
1505 spin_unlock_irqrestore(&blkcg->lock, flags);
1508 * This blkio_group is being unlinked as associated cgroup is
1509 * going away. Let all the IO controlling policies know about
1512 spin_lock(&blkio_list_lock);
1513 list_for_each_entry(blkiop, &blkio_list, list) {
1514 if (blkiop->plid != blkg->plid)
1516 blkiop->ops.blkio_unlink_group_fn(key, blkg);
1518 spin_unlock(&blkio_list_lock);
1521 list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1522 blkio_policy_delete_node(pn);
1526 free_css_id(&blkio_subsys, &blkcg->css);
1528 if (blkcg != &blkio_root_cgroup)
1532 static struct cgroup_subsys_state *
1533 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1535 struct blkio_cgroup *blkcg;
1536 struct cgroup *parent = cgroup->parent;
1539 blkcg = &blkio_root_cgroup;
1543 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1545 return ERR_PTR(-ENOMEM);
1547 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1549 spin_lock_init(&blkcg->lock);
1550 INIT_HLIST_HEAD(&blkcg->blkg_list);
1552 INIT_LIST_HEAD(&blkcg->policy_list);
1557 * We cannot support shared io contexts, as we have no mean to support
1558 * two tasks with the same ioc in two different groups without major rework
1559 * of the main cic data structures. For now we allow a task to change
1560 * its cgroup only if it's the only owner of its ioc.
1562 static int blkiocg_can_attach(struct cgroup_subsys *subsys,
1563 struct cgroup *cgroup, struct task_struct *tsk,
1566 struct io_context *ioc;
1569 /* task_lock() is needed to avoid races with exit_io_context() */
1571 ioc = tsk->io_context;
1572 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1579 static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup,
1580 struct cgroup *prev, struct task_struct *tsk,
1583 struct io_context *ioc;
1586 ioc = tsk->io_context;
1588 ioc->cgroup_changed = 1;
1592 void blkio_policy_register(struct blkio_policy_type *blkiop)
1594 spin_lock(&blkio_list_lock);
1595 list_add_tail(&blkiop->list, &blkio_list);
1596 spin_unlock(&blkio_list_lock);
1598 EXPORT_SYMBOL_GPL(blkio_policy_register);
1600 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1602 spin_lock(&blkio_list_lock);
1603 list_del_init(&blkiop->list);
1604 spin_unlock(&blkio_list_lock);
1606 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1608 static int __init init_cgroup_blkio(void)
1610 return cgroup_load_subsys(&blkio_subsys);
1613 static void __exit exit_cgroup_blkio(void)
1615 cgroup_unload_subsys(&blkio_subsys);
1618 module_init(init_cgroup_blkio);
1619 module_exit(exit_cgroup_blkio);
1620 MODULE_LICENSE("GPL");