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_task(struct cgroup *, struct cgroup *,
34 struct task_struct *);
35 static void blkiocg_attach_task(struct cgroup *, struct cgroup *,
36 struct task_struct *);
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_task = blkiocg_can_attach_task,
50 .attach_task = blkiocg_attach_task,
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;
406 * Disabling interrupts to provide mutual exclusion between two
407 * writes on same cpu. It probably is not needed for 64bit. Not
408 * optimizing that case yet.
410 local_irq_save(flags);
412 stats_cpu = this_cpu_ptr(blkg->stats_cpu);
414 u64_stats_update_begin(&stats_cpu->syncp);
415 stats_cpu->sectors += bytes >> 9;
416 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
418 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
419 bytes, direction, sync);
420 u64_stats_update_end(&stats_cpu->syncp);
421 local_irq_restore(flags);
423 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
425 void blkiocg_update_completion_stats(struct blkio_group *blkg,
426 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
428 struct blkio_group_stats *stats;
430 unsigned long long now = sched_clock();
432 spin_lock_irqsave(&blkg->stats_lock, flags);
433 stats = &blkg->stats;
434 if (time_after64(now, io_start_time))
435 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
436 now - io_start_time, direction, sync);
437 if (time_after64(io_start_time, start_time))
438 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
439 io_start_time - start_time, direction, sync);
440 spin_unlock_irqrestore(&blkg->stats_lock, flags);
442 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
444 /* Merged stats are per cpu. */
445 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
448 struct blkio_group_stats_cpu *stats_cpu;
452 * Disabling interrupts to provide mutual exclusion between two
453 * writes on same cpu. It probably is not needed for 64bit. Not
454 * optimizing that case yet.
456 local_irq_save(flags);
458 stats_cpu = this_cpu_ptr(blkg->stats_cpu);
460 u64_stats_update_begin(&stats_cpu->syncp);
461 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1,
463 u64_stats_update_end(&stats_cpu->syncp);
464 local_irq_restore(flags);
466 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
469 * This function allocates the per cpu stats for blkio_group. Should be called
470 * from sleepable context as alloc_per_cpu() requires that.
472 int blkio_alloc_blkg_stats(struct blkio_group *blkg)
474 /* Allocate memory for per cpu stats */
475 blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
476 if (!blkg->stats_cpu)
480 EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
482 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
483 struct blkio_group *blkg, void *key, dev_t dev,
484 enum blkio_policy_id plid)
488 spin_lock_irqsave(&blkcg->lock, flags);
489 spin_lock_init(&blkg->stats_lock);
490 rcu_assign_pointer(blkg->key, key);
491 blkg->blkcg_id = css_id(&blkcg->css);
492 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
494 spin_unlock_irqrestore(&blkcg->lock, flags);
495 /* Need to take css reference ? */
496 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
499 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
501 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
503 hlist_del_init_rcu(&blkg->blkcg_node);
508 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
509 * indicating that blk_group was unhashed by the time we got to it.
511 int blkiocg_del_blkio_group(struct blkio_group *blkg)
513 struct blkio_cgroup *blkcg;
515 struct cgroup_subsys_state *css;
519 css = css_lookup(&blkio_subsys, blkg->blkcg_id);
521 blkcg = container_of(css, struct blkio_cgroup, css);
522 spin_lock_irqsave(&blkcg->lock, flags);
523 if (!hlist_unhashed(&blkg->blkcg_node)) {
524 __blkiocg_del_blkio_group(blkg);
527 spin_unlock_irqrestore(&blkcg->lock, flags);
533 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
535 /* called under rcu_read_lock(). */
536 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
538 struct blkio_group *blkg;
539 struct hlist_node *n;
542 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
550 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
552 static void blkio_reset_stats_cpu(struct blkio_group *blkg)
554 struct blkio_group_stats_cpu *stats_cpu;
557 * Note: On 64 bit arch this should not be an issue. This has the
558 * possibility of returning some inconsistent value on 32bit arch
559 * as 64bit update on 32bit is non atomic. Taking care of this
560 * corner case makes code very complicated, like sending IPIs to
561 * cpus, taking care of stats of offline cpus etc.
563 * reset stats is anyway more of a debug feature and this sounds a
564 * corner case. So I am not complicating the code yet until and
565 * unless this becomes a real issue.
567 for_each_possible_cpu(i) {
568 stats_cpu = per_cpu_ptr(blkg->stats_cpu, i);
569 stats_cpu->sectors = 0;
570 for(j = 0; j < BLKIO_STAT_CPU_NR; j++)
571 for (k = 0; k < BLKIO_STAT_TOTAL; k++)
572 stats_cpu->stat_arr_cpu[j][k] = 0;
577 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
579 struct blkio_cgroup *blkcg;
580 struct blkio_group *blkg;
581 struct blkio_group_stats *stats;
582 struct hlist_node *n;
583 uint64_t queued[BLKIO_STAT_TOTAL];
585 #ifdef CONFIG_DEBUG_BLK_CGROUP
586 bool idling, waiting, empty;
587 unsigned long long now = sched_clock();
590 blkcg = cgroup_to_blkio_cgroup(cgroup);
591 spin_lock_irq(&blkcg->lock);
592 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
593 spin_lock(&blkg->stats_lock);
594 stats = &blkg->stats;
595 #ifdef CONFIG_DEBUG_BLK_CGROUP
596 idling = blkio_blkg_idling(stats);
597 waiting = blkio_blkg_waiting(stats);
598 empty = blkio_blkg_empty(stats);
600 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
601 queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
602 memset(stats, 0, sizeof(struct blkio_group_stats));
603 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
604 stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
605 #ifdef CONFIG_DEBUG_BLK_CGROUP
607 blkio_mark_blkg_idling(stats);
608 stats->start_idle_time = now;
611 blkio_mark_blkg_waiting(stats);
612 stats->start_group_wait_time = now;
615 blkio_mark_blkg_empty(stats);
616 stats->start_empty_time = now;
619 spin_unlock(&blkg->stats_lock);
621 /* Reset Per cpu stats which don't take blkg->stats_lock */
622 blkio_reset_stats_cpu(blkg);
625 spin_unlock_irq(&blkcg->lock);
629 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
630 int chars_left, bool diskname_only)
632 snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
633 chars_left -= strlen(str);
634 if (chars_left <= 0) {
636 "Possibly incorrect cgroup stat display format");
642 case BLKIO_STAT_READ:
643 strlcat(str, " Read", chars_left);
645 case BLKIO_STAT_WRITE:
646 strlcat(str, " Write", chars_left);
648 case BLKIO_STAT_SYNC:
649 strlcat(str, " Sync", chars_left);
651 case BLKIO_STAT_ASYNC:
652 strlcat(str, " Async", chars_left);
654 case BLKIO_STAT_TOTAL:
655 strlcat(str, " Total", chars_left);
658 strlcat(str, " Invalid", chars_left);
662 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
663 struct cgroup_map_cb *cb, dev_t dev)
665 blkio_get_key_name(0, dev, str, chars_left, true);
666 cb->fill(cb, str, val);
671 static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
672 enum stat_type_cpu type, enum stat_sub_type sub_type)
675 struct blkio_group_stats_cpu *stats_cpu;
678 for_each_possible_cpu(cpu) {
680 stats_cpu = per_cpu_ptr(blkg->stats_cpu, cpu);
683 start = u64_stats_fetch_begin(&stats_cpu->syncp);
684 if (type == BLKIO_STAT_CPU_SECTORS)
685 tval = stats_cpu->sectors;
687 tval = stats_cpu->stat_arr_cpu[type][sub_type];
688 } while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
696 static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
697 struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
699 uint64_t disk_total, val;
700 char key_str[MAX_KEY_LEN];
701 enum stat_sub_type sub_type;
703 if (type == BLKIO_STAT_CPU_SECTORS) {
704 val = blkio_read_stat_cpu(blkg, type, 0);
705 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
708 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
710 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
711 val = blkio_read_stat_cpu(blkg, type, sub_type);
712 cb->fill(cb, key_str, val);
715 disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
716 blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
718 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
719 cb->fill(cb, key_str, disk_total);
723 /* This should be called with blkg->stats_lock held */
724 static uint64_t blkio_get_stat(struct blkio_group *blkg,
725 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
728 char key_str[MAX_KEY_LEN];
729 enum stat_sub_type sub_type;
731 if (type == BLKIO_STAT_TIME)
732 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
733 blkg->stats.time, cb, dev);
734 #ifdef CONFIG_DEBUG_BLK_CGROUP
735 if (type == BLKIO_STAT_UNACCOUNTED_TIME)
736 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
737 blkg->stats.unaccounted_time, cb, dev);
738 if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
739 uint64_t sum = blkg->stats.avg_queue_size_sum;
740 uint64_t samples = blkg->stats.avg_queue_size_samples;
742 do_div(sum, samples);
745 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
747 if (type == BLKIO_STAT_GROUP_WAIT_TIME)
748 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
749 blkg->stats.group_wait_time, cb, dev);
750 if (type == BLKIO_STAT_IDLE_TIME)
751 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
752 blkg->stats.idle_time, cb, dev);
753 if (type == BLKIO_STAT_EMPTY_TIME)
754 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
755 blkg->stats.empty_time, cb, dev);
756 if (type == BLKIO_STAT_DEQUEUE)
757 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
758 blkg->stats.dequeue, cb, dev);
761 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
763 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
764 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
766 disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
767 blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
768 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
769 cb->fill(cb, key_str, disk_total);
773 static int blkio_check_dev_num(dev_t dev)
776 struct gendisk *disk;
778 disk = get_gendisk(dev, &part);
785 static int blkio_policy_parse_and_set(char *buf,
786 struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
788 char *s[4], *p, *major_s = NULL, *minor_s = NULL;
790 unsigned long major, minor;
795 memset(s, 0, sizeof(s));
797 while ((p = strsep(&buf, " ")) != NULL) {
803 /* Prevent from inputing too many things */
811 p = strsep(&s[0], ":");
821 ret = strict_strtoul(major_s, 10, &major);
825 ret = strict_strtoul(minor_s, 10, &minor);
829 dev = MKDEV(major, minor);
831 ret = strict_strtoull(s[1], 10, &temp);
835 /* For rule removal, do not check for device presence. */
837 ret = blkio_check_dev_num(dev);
845 case BLKIO_POLICY_PROP:
846 if ((temp < BLKIO_WEIGHT_MIN && temp > 0) ||
847 temp > BLKIO_WEIGHT_MAX)
851 newpn->fileid = fileid;
852 newpn->val.weight = temp;
854 case BLKIO_POLICY_THROTL:
856 case BLKIO_THROTL_read_bps_device:
857 case BLKIO_THROTL_write_bps_device:
859 newpn->fileid = fileid;
860 newpn->val.bps = temp;
862 case BLKIO_THROTL_read_iops_device:
863 case BLKIO_THROTL_write_iops_device:
864 if (temp > THROTL_IOPS_MAX)
868 newpn->fileid = fileid;
869 newpn->val.iops = (unsigned int)temp;
880 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
883 struct blkio_policy_node *pn;
885 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
886 BLKIO_PROP_weight_device);
888 return pn->val.weight;
890 return blkcg->weight;
892 EXPORT_SYMBOL_GPL(blkcg_get_weight);
894 uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev)
896 struct blkio_policy_node *pn;
898 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
899 BLKIO_THROTL_read_bps_device);
906 uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
908 struct blkio_policy_node *pn;
909 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
910 BLKIO_THROTL_write_bps_device);
917 unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
919 struct blkio_policy_node *pn;
921 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
922 BLKIO_THROTL_read_iops_device);
929 unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
931 struct blkio_policy_node *pn;
932 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
933 BLKIO_THROTL_write_iops_device);
940 /* Checks whether user asked for deleting a policy rule */
941 static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
944 case BLKIO_POLICY_PROP:
945 if (pn->val.weight == 0)
948 case BLKIO_POLICY_THROTL:
950 case BLKIO_THROTL_read_bps_device:
951 case BLKIO_THROTL_write_bps_device:
952 if (pn->val.bps == 0)
955 case BLKIO_THROTL_read_iops_device:
956 case BLKIO_THROTL_write_iops_device:
957 if (pn->val.iops == 0)
968 static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
969 struct blkio_policy_node *newpn)
971 switch(oldpn->plid) {
972 case BLKIO_POLICY_PROP:
973 oldpn->val.weight = newpn->val.weight;
975 case BLKIO_POLICY_THROTL:
976 switch(newpn->fileid) {
977 case BLKIO_THROTL_read_bps_device:
978 case BLKIO_THROTL_write_bps_device:
979 oldpn->val.bps = newpn->val.bps;
981 case BLKIO_THROTL_read_iops_device:
982 case BLKIO_THROTL_write_iops_device:
983 oldpn->val.iops = newpn->val.iops;
992 * Some rules/values in blkg have changed. Propagate those to respective
995 static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
996 struct blkio_group *blkg, struct blkio_policy_node *pn)
998 unsigned int weight, iops;
1002 case BLKIO_POLICY_PROP:
1003 weight = pn->val.weight ? pn->val.weight :
1005 blkio_update_group_weight(blkg, weight);
1007 case BLKIO_POLICY_THROTL:
1008 switch(pn->fileid) {
1009 case BLKIO_THROTL_read_bps_device:
1010 case BLKIO_THROTL_write_bps_device:
1011 bps = pn->val.bps ? pn->val.bps : (-1);
1012 blkio_update_group_bps(blkg, bps, pn->fileid);
1014 case BLKIO_THROTL_read_iops_device:
1015 case BLKIO_THROTL_write_iops_device:
1016 iops = pn->val.iops ? pn->val.iops : (-1);
1017 blkio_update_group_iops(blkg, iops, pn->fileid);
1027 * A policy node rule has been updated. Propagate this update to all the
1028 * block groups which might be affected by this update.
1030 static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
1031 struct blkio_policy_node *pn)
1033 struct blkio_group *blkg;
1034 struct hlist_node *n;
1036 spin_lock(&blkio_list_lock);
1037 spin_lock_irq(&blkcg->lock);
1039 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1040 if (pn->dev != blkg->dev || pn->plid != blkg->plid)
1042 blkio_update_blkg_policy(blkcg, blkg, pn);
1045 spin_unlock_irq(&blkcg->lock);
1046 spin_unlock(&blkio_list_lock);
1049 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1054 struct blkio_policy_node *newpn, *pn;
1055 struct blkio_cgroup *blkcg;
1057 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1058 int fileid = BLKIOFILE_ATTR(cft->private);
1060 buf = kstrdup(buffer, GFP_KERNEL);
1064 newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
1070 ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
1074 blkcg = cgroup_to_blkio_cgroup(cgrp);
1076 spin_lock_irq(&blkcg->lock);
1078 pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
1080 if (!blkio_delete_rule_command(newpn)) {
1081 blkio_policy_insert_node(blkcg, newpn);
1084 spin_unlock_irq(&blkcg->lock);
1085 goto update_io_group;
1088 if (blkio_delete_rule_command(newpn)) {
1089 blkio_policy_delete_node(pn);
1090 spin_unlock_irq(&blkcg->lock);
1091 goto update_io_group;
1093 spin_unlock_irq(&blkcg->lock);
1095 blkio_update_policy_rule(pn, newpn);
1098 blkio_update_policy_node_blkg(blkcg, newpn);
1109 blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
1112 case BLKIO_POLICY_PROP:
1113 if (pn->fileid == BLKIO_PROP_weight_device)
1114 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1115 MINOR(pn->dev), pn->val.weight);
1117 case BLKIO_POLICY_THROTL:
1118 switch(pn->fileid) {
1119 case BLKIO_THROTL_read_bps_device:
1120 case BLKIO_THROTL_write_bps_device:
1121 seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1122 MINOR(pn->dev), pn->val.bps);
1124 case BLKIO_THROTL_read_iops_device:
1125 case BLKIO_THROTL_write_iops_device:
1126 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1127 MINOR(pn->dev), pn->val.iops);
1136 /* cgroup files which read their data from policy nodes end up here */
1137 static void blkio_read_policy_node_files(struct cftype *cft,
1138 struct blkio_cgroup *blkcg, struct seq_file *m)
1140 struct blkio_policy_node *pn;
1142 if (!list_empty(&blkcg->policy_list)) {
1143 spin_lock_irq(&blkcg->lock);
1144 list_for_each_entry(pn, &blkcg->policy_list, node) {
1145 if (!pn_matches_cftype(cft, pn))
1147 blkio_print_policy_node(m, pn);
1149 spin_unlock_irq(&blkcg->lock);
1153 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1156 struct blkio_cgroup *blkcg;
1157 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1158 int name = BLKIOFILE_ATTR(cft->private);
1160 blkcg = cgroup_to_blkio_cgroup(cgrp);
1163 case BLKIO_POLICY_PROP:
1165 case BLKIO_PROP_weight_device:
1166 blkio_read_policy_node_files(cft, blkcg, m);
1172 case BLKIO_POLICY_THROTL:
1174 case BLKIO_THROTL_read_bps_device:
1175 case BLKIO_THROTL_write_bps_device:
1176 case BLKIO_THROTL_read_iops_device:
1177 case BLKIO_THROTL_write_iops_device:
1178 blkio_read_policy_node_files(cft, blkcg, m);
1191 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1192 struct cftype *cft, struct cgroup_map_cb *cb,
1193 enum stat_type type, bool show_total, bool pcpu)
1195 struct blkio_group *blkg;
1196 struct hlist_node *n;
1197 uint64_t cgroup_total = 0;
1200 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1202 if (!cftype_blkg_same_policy(cft, blkg))
1205 cgroup_total += blkio_get_stat_cpu(blkg, cb,
1208 spin_lock_irq(&blkg->stats_lock);
1209 cgroup_total += blkio_get_stat(blkg, cb,
1211 spin_unlock_irq(&blkg->stats_lock);
1216 cb->fill(cb, "Total", cgroup_total);
1221 /* All map kind of cgroup file get serviced by this function */
1222 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1223 struct cgroup_map_cb *cb)
1225 struct blkio_cgroup *blkcg;
1226 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1227 int name = BLKIOFILE_ATTR(cft->private);
1229 blkcg = cgroup_to_blkio_cgroup(cgrp);
1232 case BLKIO_POLICY_PROP:
1234 case BLKIO_PROP_time:
1235 return blkio_read_blkg_stats(blkcg, cft, cb,
1236 BLKIO_STAT_TIME, 0, 0);
1237 case BLKIO_PROP_sectors:
1238 return blkio_read_blkg_stats(blkcg, cft, cb,
1239 BLKIO_STAT_CPU_SECTORS, 0, 1);
1240 case BLKIO_PROP_io_service_bytes:
1241 return blkio_read_blkg_stats(blkcg, cft, cb,
1242 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1243 case BLKIO_PROP_io_serviced:
1244 return blkio_read_blkg_stats(blkcg, cft, cb,
1245 BLKIO_STAT_CPU_SERVICED, 1, 1);
1246 case BLKIO_PROP_io_service_time:
1247 return blkio_read_blkg_stats(blkcg, cft, cb,
1248 BLKIO_STAT_SERVICE_TIME, 1, 0);
1249 case BLKIO_PROP_io_wait_time:
1250 return blkio_read_blkg_stats(blkcg, cft, cb,
1251 BLKIO_STAT_WAIT_TIME, 1, 0);
1252 case BLKIO_PROP_io_merged:
1253 return blkio_read_blkg_stats(blkcg, cft, cb,
1254 BLKIO_STAT_CPU_MERGED, 1, 1);
1255 case BLKIO_PROP_io_queued:
1256 return blkio_read_blkg_stats(blkcg, cft, cb,
1257 BLKIO_STAT_QUEUED, 1, 0);
1258 #ifdef CONFIG_DEBUG_BLK_CGROUP
1259 case BLKIO_PROP_unaccounted_time:
1260 return blkio_read_blkg_stats(blkcg, cft, cb,
1261 BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
1262 case BLKIO_PROP_dequeue:
1263 return blkio_read_blkg_stats(blkcg, cft, cb,
1264 BLKIO_STAT_DEQUEUE, 0, 0);
1265 case BLKIO_PROP_avg_queue_size:
1266 return blkio_read_blkg_stats(blkcg, cft, cb,
1267 BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
1268 case BLKIO_PROP_group_wait_time:
1269 return blkio_read_blkg_stats(blkcg, cft, cb,
1270 BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
1271 case BLKIO_PROP_idle_time:
1272 return blkio_read_blkg_stats(blkcg, cft, cb,
1273 BLKIO_STAT_IDLE_TIME, 0, 0);
1274 case BLKIO_PROP_empty_time:
1275 return blkio_read_blkg_stats(blkcg, cft, cb,
1276 BLKIO_STAT_EMPTY_TIME, 0, 0);
1282 case BLKIO_POLICY_THROTL:
1284 case BLKIO_THROTL_io_service_bytes:
1285 return blkio_read_blkg_stats(blkcg, cft, cb,
1286 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1287 case BLKIO_THROTL_io_serviced:
1288 return blkio_read_blkg_stats(blkcg, cft, cb,
1289 BLKIO_STAT_CPU_SERVICED, 1, 1);
1301 static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1303 struct blkio_group *blkg;
1304 struct hlist_node *n;
1305 struct blkio_policy_node *pn;
1307 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1310 spin_lock(&blkio_list_lock);
1311 spin_lock_irq(&blkcg->lock);
1312 blkcg->weight = (unsigned int)val;
1314 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1315 pn = blkio_policy_search_node(blkcg, blkg->dev,
1316 BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1320 blkio_update_group_weight(blkg, blkcg->weight);
1322 spin_unlock_irq(&blkcg->lock);
1323 spin_unlock(&blkio_list_lock);
1327 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1328 struct blkio_cgroup *blkcg;
1329 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1330 int name = BLKIOFILE_ATTR(cft->private);
1332 blkcg = cgroup_to_blkio_cgroup(cgrp);
1335 case BLKIO_POLICY_PROP:
1337 case BLKIO_PROP_weight:
1338 return (u64)blkcg->weight;
1348 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1350 struct blkio_cgroup *blkcg;
1351 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1352 int name = BLKIOFILE_ATTR(cft->private);
1354 blkcg = cgroup_to_blkio_cgroup(cgrp);
1357 case BLKIO_POLICY_PROP:
1359 case BLKIO_PROP_weight:
1360 return blkio_weight_write(blkcg, val);
1370 struct cftype blkio_files[] = {
1372 .name = "weight_device",
1373 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1374 BLKIO_PROP_weight_device),
1375 .read_seq_string = blkiocg_file_read,
1376 .write_string = blkiocg_file_write,
1377 .max_write_len = 256,
1381 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1383 .read_u64 = blkiocg_file_read_u64,
1384 .write_u64 = blkiocg_file_write_u64,
1388 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1390 .read_map = blkiocg_file_read_map,
1394 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1395 BLKIO_PROP_sectors),
1396 .read_map = blkiocg_file_read_map,
1399 .name = "io_service_bytes",
1400 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1401 BLKIO_PROP_io_service_bytes),
1402 .read_map = blkiocg_file_read_map,
1405 .name = "io_serviced",
1406 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1407 BLKIO_PROP_io_serviced),
1408 .read_map = blkiocg_file_read_map,
1411 .name = "io_service_time",
1412 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1413 BLKIO_PROP_io_service_time),
1414 .read_map = blkiocg_file_read_map,
1417 .name = "io_wait_time",
1418 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1419 BLKIO_PROP_io_wait_time),
1420 .read_map = blkiocg_file_read_map,
1423 .name = "io_merged",
1424 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1425 BLKIO_PROP_io_merged),
1426 .read_map = blkiocg_file_read_map,
1429 .name = "io_queued",
1430 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1431 BLKIO_PROP_io_queued),
1432 .read_map = blkiocg_file_read_map,
1435 .name = "reset_stats",
1436 .write_u64 = blkiocg_reset_stats,
1438 #ifdef CONFIG_BLK_DEV_THROTTLING
1440 .name = "throttle.read_bps_device",
1441 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1442 BLKIO_THROTL_read_bps_device),
1443 .read_seq_string = blkiocg_file_read,
1444 .write_string = blkiocg_file_write,
1445 .max_write_len = 256,
1449 .name = "throttle.write_bps_device",
1450 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1451 BLKIO_THROTL_write_bps_device),
1452 .read_seq_string = blkiocg_file_read,
1453 .write_string = blkiocg_file_write,
1454 .max_write_len = 256,
1458 .name = "throttle.read_iops_device",
1459 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1460 BLKIO_THROTL_read_iops_device),
1461 .read_seq_string = blkiocg_file_read,
1462 .write_string = blkiocg_file_write,
1463 .max_write_len = 256,
1467 .name = "throttle.write_iops_device",
1468 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1469 BLKIO_THROTL_write_iops_device),
1470 .read_seq_string = blkiocg_file_read,
1471 .write_string = blkiocg_file_write,
1472 .max_write_len = 256,
1475 .name = "throttle.io_service_bytes",
1476 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1477 BLKIO_THROTL_io_service_bytes),
1478 .read_map = blkiocg_file_read_map,
1481 .name = "throttle.io_serviced",
1482 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1483 BLKIO_THROTL_io_serviced),
1484 .read_map = blkiocg_file_read_map,
1486 #endif /* CONFIG_BLK_DEV_THROTTLING */
1488 #ifdef CONFIG_DEBUG_BLK_CGROUP
1490 .name = "avg_queue_size",
1491 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1492 BLKIO_PROP_avg_queue_size),
1493 .read_map = blkiocg_file_read_map,
1496 .name = "group_wait_time",
1497 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1498 BLKIO_PROP_group_wait_time),
1499 .read_map = blkiocg_file_read_map,
1502 .name = "idle_time",
1503 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1504 BLKIO_PROP_idle_time),
1505 .read_map = blkiocg_file_read_map,
1508 .name = "empty_time",
1509 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1510 BLKIO_PROP_empty_time),
1511 .read_map = blkiocg_file_read_map,
1515 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1516 BLKIO_PROP_dequeue),
1517 .read_map = blkiocg_file_read_map,
1520 .name = "unaccounted_time",
1521 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1522 BLKIO_PROP_unaccounted_time),
1523 .read_map = blkiocg_file_read_map,
1528 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1530 return cgroup_add_files(cgroup, subsys, blkio_files,
1531 ARRAY_SIZE(blkio_files));
1534 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1536 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1537 unsigned long flags;
1538 struct blkio_group *blkg;
1540 struct blkio_policy_type *blkiop;
1541 struct blkio_policy_node *pn, *pntmp;
1545 spin_lock_irqsave(&blkcg->lock, flags);
1547 if (hlist_empty(&blkcg->blkg_list)) {
1548 spin_unlock_irqrestore(&blkcg->lock, flags);
1552 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1554 key = rcu_dereference(blkg->key);
1555 __blkiocg_del_blkio_group(blkg);
1557 spin_unlock_irqrestore(&blkcg->lock, flags);
1560 * This blkio_group is being unlinked as associated cgroup is
1561 * going away. Let all the IO controlling policies know about
1564 spin_lock(&blkio_list_lock);
1565 list_for_each_entry(blkiop, &blkio_list, list) {
1566 if (blkiop->plid != blkg->plid)
1568 blkiop->ops.blkio_unlink_group_fn(key, blkg);
1570 spin_unlock(&blkio_list_lock);
1573 list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1574 blkio_policy_delete_node(pn);
1578 free_css_id(&blkio_subsys, &blkcg->css);
1580 if (blkcg != &blkio_root_cgroup)
1584 static struct cgroup_subsys_state *
1585 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1587 struct blkio_cgroup *blkcg;
1588 struct cgroup *parent = cgroup->parent;
1591 blkcg = &blkio_root_cgroup;
1595 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1597 return ERR_PTR(-ENOMEM);
1599 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1601 spin_lock_init(&blkcg->lock);
1602 INIT_HLIST_HEAD(&blkcg->blkg_list);
1604 INIT_LIST_HEAD(&blkcg->policy_list);
1609 * We cannot support shared io contexts, as we have no mean to support
1610 * two tasks with the same ioc in two different groups without major rework
1611 * of the main cic data structures. For now we allow a task to change
1612 * its cgroup only if it's the only owner of its ioc.
1614 static int blkiocg_can_attach_task(struct cgroup *cgrp, struct cgroup *old_cgrp,
1615 struct task_struct *tsk)
1617 struct io_context *ioc;
1620 /* task_lock() is needed to avoid races with exit_io_context() */
1622 ioc = tsk->io_context;
1623 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1630 static void blkiocg_attach_task(struct cgroup *cgrp, struct cgroup *old_cgrp,
1631 struct task_struct *tsk)
1633 struct io_context *ioc;
1636 ioc = tsk->io_context;
1638 ioc->cgroup_changed = 1;
1642 void blkio_policy_register(struct blkio_policy_type *blkiop)
1644 spin_lock(&blkio_list_lock);
1645 list_add_tail(&blkiop->list, &blkio_list);
1646 spin_unlock(&blkio_list_lock);
1648 EXPORT_SYMBOL_GPL(blkio_policy_register);
1650 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1652 spin_lock(&blkio_list_lock);
1653 list_del_init(&blkiop->list);
1654 spin_unlock(&blkio_list_lock);
1656 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1658 static int __init init_cgroup_blkio(void)
1660 return cgroup_load_subsys(&blkio_subsys);
1663 static void __exit exit_cgroup_blkio(void)
1665 cgroup_unload_subsys(&blkio_subsys);
1668 module_init(init_cgroup_blkio);
1669 module_exit(exit_cgroup_blkio);
1670 MODULE_LICENSE("GPL");