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 /* for encoding cft->private value on file */
32 #define BLKIOFILE_PRIVATE(x, val) (((x) << 16) | (val))
33 /* What policy owns the file, proportional or throttle */
34 #define BLKIOFILE_POLICY(val) (((val) >> 16) & 0xffff)
35 #define BLKIOFILE_ATTR(val) ((val) & 0xffff)
37 static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg,
38 struct blkio_policy_node *pn)
40 list_add(&pn->node, &blkcg->policy_list);
43 static inline bool cftype_blkg_same_policy(struct cftype *cft,
44 struct blkio_group *blkg)
46 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
48 if (blkg->plid == plid)
54 /* Determines if policy node matches cgroup file being accessed */
55 static inline bool pn_matches_cftype(struct cftype *cft,
56 struct blkio_policy_node *pn)
58 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
59 int fileid = BLKIOFILE_ATTR(cft->private);
61 return (plid == pn->plid && fileid == pn->fileid);
64 /* Must be called with blkcg->lock held */
65 static inline void blkio_policy_delete_node(struct blkio_policy_node *pn)
70 /* Must be called with blkcg->lock held */
71 static struct blkio_policy_node *
72 blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev,
73 enum blkio_policy_id plid, int fileid)
75 struct blkio_policy_node *pn;
77 list_for_each_entry(pn, &blkcg->policy_list, node) {
78 if (pn->dev == dev && pn->plid == plid && pn->fileid == fileid)
85 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
87 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
88 struct blkio_cgroup, css);
90 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
92 struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
94 return container_of(task_subsys_state(tsk, blkio_subsys_id),
95 struct blkio_cgroup, css);
97 EXPORT_SYMBOL_GPL(task_blkio_cgroup);
100 blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight)
102 struct blkio_policy_type *blkiop;
104 list_for_each_entry(blkiop, &blkio_list, list) {
105 /* If this policy does not own the blkg, do not send updates */
106 if (blkiop->plid != blkg->plid)
108 if (blkiop->ops.blkio_update_group_weight_fn)
109 blkiop->ops.blkio_update_group_weight_fn(blkg->key,
114 static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps,
117 struct blkio_policy_type *blkiop;
119 list_for_each_entry(blkiop, &blkio_list, list) {
121 /* If this policy does not own the blkg, do not send updates */
122 if (blkiop->plid != blkg->plid)
125 if (fileid == BLKIO_THROTL_read_bps_device
126 && blkiop->ops.blkio_update_group_read_bps_fn)
127 blkiop->ops.blkio_update_group_read_bps_fn(blkg->key,
130 if (fileid == BLKIO_THROTL_write_bps_device
131 && blkiop->ops.blkio_update_group_write_bps_fn)
132 blkiop->ops.blkio_update_group_write_bps_fn(blkg->key,
137 static inline void blkio_update_group_iops(struct blkio_group *blkg,
138 unsigned int iops, int fileid)
140 struct blkio_policy_type *blkiop;
142 list_for_each_entry(blkiop, &blkio_list, list) {
144 /* If this policy does not own the blkg, do not send updates */
145 if (blkiop->plid != blkg->plid)
148 if (fileid == BLKIO_THROTL_read_iops_device
149 && blkiop->ops.blkio_update_group_read_iops_fn)
150 blkiop->ops.blkio_update_group_read_iops_fn(blkg->key,
153 if (fileid == BLKIO_THROTL_write_iops_device
154 && blkiop->ops.blkio_update_group_write_iops_fn)
155 blkiop->ops.blkio_update_group_write_iops_fn(blkg->key,
161 * Add to the appropriate stat variable depending on the request type.
162 * This should be called with the blkg->stats_lock held.
164 static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
168 stat[BLKIO_STAT_WRITE] += add;
170 stat[BLKIO_STAT_READ] += add;
172 stat[BLKIO_STAT_SYNC] += add;
174 stat[BLKIO_STAT_ASYNC] += add;
178 * Decrements the appropriate stat variable if non-zero depending on the
179 * request type. Panics on value being zero.
180 * This should be called with the blkg->stats_lock held.
182 static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
185 BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
186 stat[BLKIO_STAT_WRITE]--;
188 BUG_ON(stat[BLKIO_STAT_READ] == 0);
189 stat[BLKIO_STAT_READ]--;
192 BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
193 stat[BLKIO_STAT_SYNC]--;
195 BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
196 stat[BLKIO_STAT_ASYNC]--;
200 #ifdef CONFIG_DEBUG_BLK_CGROUP
201 /* This should be called with the blkg->stats_lock held. */
202 static void blkio_set_start_group_wait_time(struct blkio_group *blkg,
203 struct blkio_group *curr_blkg)
205 if (blkio_blkg_waiting(&blkg->stats))
207 if (blkg == curr_blkg)
209 blkg->stats.start_group_wait_time = sched_clock();
210 blkio_mark_blkg_waiting(&blkg->stats);
213 /* This should be called with the blkg->stats_lock held. */
214 static void blkio_update_group_wait_time(struct blkio_group_stats *stats)
216 unsigned long long now;
218 if (!blkio_blkg_waiting(stats))
222 if (time_after64(now, stats->start_group_wait_time))
223 stats->group_wait_time += now - stats->start_group_wait_time;
224 blkio_clear_blkg_waiting(stats);
227 /* This should be called with the blkg->stats_lock held. */
228 static void blkio_end_empty_time(struct blkio_group_stats *stats)
230 unsigned long long now;
232 if (!blkio_blkg_empty(stats))
236 if (time_after64(now, stats->start_empty_time))
237 stats->empty_time += now - stats->start_empty_time;
238 blkio_clear_blkg_empty(stats);
241 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
245 spin_lock_irqsave(&blkg->stats_lock, flags);
246 BUG_ON(blkio_blkg_idling(&blkg->stats));
247 blkg->stats.start_idle_time = sched_clock();
248 blkio_mark_blkg_idling(&blkg->stats);
249 spin_unlock_irqrestore(&blkg->stats_lock, flags);
251 EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats);
253 void blkiocg_update_idle_time_stats(struct blkio_group *blkg)
256 unsigned long long now;
257 struct blkio_group_stats *stats;
259 spin_lock_irqsave(&blkg->stats_lock, flags);
260 stats = &blkg->stats;
261 if (blkio_blkg_idling(stats)) {
263 if (time_after64(now, stats->start_idle_time))
264 stats->idle_time += now - stats->start_idle_time;
265 blkio_clear_blkg_idling(stats);
267 spin_unlock_irqrestore(&blkg->stats_lock, flags);
269 EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats);
271 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg)
274 struct blkio_group_stats *stats;
276 spin_lock_irqsave(&blkg->stats_lock, flags);
277 stats = &blkg->stats;
278 stats->avg_queue_size_sum +=
279 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
280 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
281 stats->avg_queue_size_samples++;
282 blkio_update_group_wait_time(stats);
283 spin_unlock_irqrestore(&blkg->stats_lock, flags);
285 EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats);
287 void blkiocg_set_start_empty_time(struct blkio_group *blkg)
290 struct blkio_group_stats *stats;
292 spin_lock_irqsave(&blkg->stats_lock, flags);
293 stats = &blkg->stats;
295 if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] ||
296 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) {
297 spin_unlock_irqrestore(&blkg->stats_lock, flags);
302 * group is already marked empty. This can happen if cfqq got new
303 * request in parent group and moved to this group while being added
304 * to service tree. Just ignore the event and move on.
306 if(blkio_blkg_empty(stats)) {
307 spin_unlock_irqrestore(&blkg->stats_lock, flags);
311 stats->start_empty_time = sched_clock();
312 blkio_mark_blkg_empty(stats);
313 spin_unlock_irqrestore(&blkg->stats_lock, flags);
315 EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time);
317 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
318 unsigned long dequeue)
320 blkg->stats.dequeue += dequeue;
322 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
324 static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg,
325 struct blkio_group *curr_blkg) {}
326 static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {}
329 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
330 struct blkio_group *curr_blkg, bool direction,
335 spin_lock_irqsave(&blkg->stats_lock, flags);
336 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction,
338 blkio_end_empty_time(&blkg->stats);
339 blkio_set_start_group_wait_time(blkg, curr_blkg);
340 spin_unlock_irqrestore(&blkg->stats_lock, flags);
342 EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats);
344 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
345 bool direction, bool sync)
349 spin_lock_irqsave(&blkg->stats_lock, flags);
350 blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED],
352 spin_unlock_irqrestore(&blkg->stats_lock, flags);
354 EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats);
356 void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time,
357 unsigned long unaccounted_time)
361 spin_lock_irqsave(&blkg->stats_lock, flags);
362 blkg->stats.time += time;
363 #ifdef CONFIG_DEBUG_BLK_CGROUP
364 blkg->stats.unaccounted_time += unaccounted_time;
366 spin_unlock_irqrestore(&blkg->stats_lock, flags);
368 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
371 * should be called under rcu read lock or queue lock to make sure blkg pointer
374 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
375 uint64_t bytes, bool direction, bool sync)
377 struct blkio_group_stats_cpu *stats_cpu;
381 * Disabling interrupts to provide mutual exclusion between two
382 * writes on same cpu. It probably is not needed for 64bit. Not
383 * optimizing that case yet.
385 local_irq_save(flags);
387 stats_cpu = this_cpu_ptr(blkg->stats_cpu);
389 u64_stats_update_begin(&stats_cpu->syncp);
390 stats_cpu->sectors += bytes >> 9;
391 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
393 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
394 bytes, direction, sync);
395 u64_stats_update_end(&stats_cpu->syncp);
396 local_irq_restore(flags);
398 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
400 void blkiocg_update_completion_stats(struct blkio_group *blkg,
401 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
403 struct blkio_group_stats *stats;
405 unsigned long long now = sched_clock();
407 spin_lock_irqsave(&blkg->stats_lock, flags);
408 stats = &blkg->stats;
409 if (time_after64(now, io_start_time))
410 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
411 now - io_start_time, direction, sync);
412 if (time_after64(io_start_time, start_time))
413 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
414 io_start_time - start_time, direction, sync);
415 spin_unlock_irqrestore(&blkg->stats_lock, flags);
417 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
419 /* Merged stats are per cpu. */
420 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
423 struct blkio_group_stats_cpu *stats_cpu;
427 * Disabling interrupts to provide mutual exclusion between two
428 * writes on same cpu. It probably is not needed for 64bit. Not
429 * optimizing that case yet.
431 local_irq_save(flags);
433 stats_cpu = this_cpu_ptr(blkg->stats_cpu);
435 u64_stats_update_begin(&stats_cpu->syncp);
436 blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1,
438 u64_stats_update_end(&stats_cpu->syncp);
439 local_irq_restore(flags);
441 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
444 * This function allocates the per cpu stats for blkio_group. Should be called
445 * from sleepable context as alloc_per_cpu() requires that.
447 int blkio_alloc_blkg_stats(struct blkio_group *blkg)
449 /* Allocate memory for per cpu stats */
450 blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
451 if (!blkg->stats_cpu)
455 EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
457 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
458 struct blkio_group *blkg, void *key, dev_t dev,
459 enum blkio_policy_id plid)
463 spin_lock_irqsave(&blkcg->lock, flags);
464 spin_lock_init(&blkg->stats_lock);
465 rcu_assign_pointer(blkg->key, key);
466 blkg->blkcg_id = css_id(&blkcg->css);
467 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
469 spin_unlock_irqrestore(&blkcg->lock, flags);
470 /* Need to take css reference ? */
471 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
474 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
476 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
478 hlist_del_init_rcu(&blkg->blkcg_node);
483 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
484 * indicating that blk_group was unhashed by the time we got to it.
486 int blkiocg_del_blkio_group(struct blkio_group *blkg)
488 struct blkio_cgroup *blkcg;
490 struct cgroup_subsys_state *css;
494 css = css_lookup(&blkio_subsys, blkg->blkcg_id);
496 blkcg = container_of(css, struct blkio_cgroup, css);
497 spin_lock_irqsave(&blkcg->lock, flags);
498 if (!hlist_unhashed(&blkg->blkcg_node)) {
499 __blkiocg_del_blkio_group(blkg);
502 spin_unlock_irqrestore(&blkcg->lock, flags);
508 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
510 /* called under rcu_read_lock(). */
511 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
513 struct blkio_group *blkg;
514 struct hlist_node *n;
517 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
525 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
527 static void blkio_reset_stats_cpu(struct blkio_group *blkg)
529 struct blkio_group_stats_cpu *stats_cpu;
532 * Note: On 64 bit arch this should not be an issue. This has the
533 * possibility of returning some inconsistent value on 32bit arch
534 * as 64bit update on 32bit is non atomic. Taking care of this
535 * corner case makes code very complicated, like sending IPIs to
536 * cpus, taking care of stats of offline cpus etc.
538 * reset stats is anyway more of a debug feature and this sounds a
539 * corner case. So I am not complicating the code yet until and
540 * unless this becomes a real issue.
542 for_each_possible_cpu(i) {
543 stats_cpu = per_cpu_ptr(blkg->stats_cpu, i);
544 stats_cpu->sectors = 0;
545 for(j = 0; j < BLKIO_STAT_CPU_NR; j++)
546 for (k = 0; k < BLKIO_STAT_TOTAL; k++)
547 stats_cpu->stat_arr_cpu[j][k] = 0;
552 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
554 struct blkio_cgroup *blkcg;
555 struct blkio_group *blkg;
556 struct blkio_group_stats *stats;
557 struct hlist_node *n;
558 uint64_t queued[BLKIO_STAT_TOTAL];
560 #ifdef CONFIG_DEBUG_BLK_CGROUP
561 bool idling, waiting, empty;
562 unsigned long long now = sched_clock();
565 blkcg = cgroup_to_blkio_cgroup(cgroup);
566 spin_lock_irq(&blkcg->lock);
567 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
568 spin_lock(&blkg->stats_lock);
569 stats = &blkg->stats;
570 #ifdef CONFIG_DEBUG_BLK_CGROUP
571 idling = blkio_blkg_idling(stats);
572 waiting = blkio_blkg_waiting(stats);
573 empty = blkio_blkg_empty(stats);
575 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
576 queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
577 memset(stats, 0, sizeof(struct blkio_group_stats));
578 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
579 stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
580 #ifdef CONFIG_DEBUG_BLK_CGROUP
582 blkio_mark_blkg_idling(stats);
583 stats->start_idle_time = now;
586 blkio_mark_blkg_waiting(stats);
587 stats->start_group_wait_time = now;
590 blkio_mark_blkg_empty(stats);
591 stats->start_empty_time = now;
594 spin_unlock(&blkg->stats_lock);
596 /* Reset Per cpu stats which don't take blkg->stats_lock */
597 blkio_reset_stats_cpu(blkg);
600 spin_unlock_irq(&blkcg->lock);
604 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
605 int chars_left, bool diskname_only)
607 snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
608 chars_left -= strlen(str);
609 if (chars_left <= 0) {
611 "Possibly incorrect cgroup stat display format");
617 case BLKIO_STAT_READ:
618 strlcat(str, " Read", chars_left);
620 case BLKIO_STAT_WRITE:
621 strlcat(str, " Write", chars_left);
623 case BLKIO_STAT_SYNC:
624 strlcat(str, " Sync", chars_left);
626 case BLKIO_STAT_ASYNC:
627 strlcat(str, " Async", chars_left);
629 case BLKIO_STAT_TOTAL:
630 strlcat(str, " Total", chars_left);
633 strlcat(str, " Invalid", chars_left);
637 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
638 struct cgroup_map_cb *cb, dev_t dev)
640 blkio_get_key_name(0, dev, str, chars_left, true);
641 cb->fill(cb, str, val);
646 static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
647 enum stat_type_cpu type, enum stat_sub_type sub_type)
650 struct blkio_group_stats_cpu *stats_cpu;
653 for_each_possible_cpu(cpu) {
655 stats_cpu = per_cpu_ptr(blkg->stats_cpu, cpu);
658 start = u64_stats_fetch_begin(&stats_cpu->syncp);
659 if (type == BLKIO_STAT_CPU_SECTORS)
660 tval = stats_cpu->sectors;
662 tval = stats_cpu->stat_arr_cpu[type][sub_type];
663 } while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
671 static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
672 struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
674 uint64_t disk_total, val;
675 char key_str[MAX_KEY_LEN];
676 enum stat_sub_type sub_type;
678 if (type == BLKIO_STAT_CPU_SECTORS) {
679 val = blkio_read_stat_cpu(blkg, type, 0);
680 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
683 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
685 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
686 val = blkio_read_stat_cpu(blkg, type, sub_type);
687 cb->fill(cb, key_str, val);
690 disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
691 blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
693 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
694 cb->fill(cb, key_str, disk_total);
698 /* This should be called with blkg->stats_lock held */
699 static uint64_t blkio_get_stat(struct blkio_group *blkg,
700 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
703 char key_str[MAX_KEY_LEN];
704 enum stat_sub_type sub_type;
706 if (type == BLKIO_STAT_TIME)
707 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
708 blkg->stats.time, cb, dev);
709 #ifdef CONFIG_DEBUG_BLK_CGROUP
710 if (type == BLKIO_STAT_UNACCOUNTED_TIME)
711 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
712 blkg->stats.unaccounted_time, cb, dev);
713 if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
714 uint64_t sum = blkg->stats.avg_queue_size_sum;
715 uint64_t samples = blkg->stats.avg_queue_size_samples;
717 do_div(sum, samples);
720 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
722 if (type == BLKIO_STAT_GROUP_WAIT_TIME)
723 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
724 blkg->stats.group_wait_time, cb, dev);
725 if (type == BLKIO_STAT_IDLE_TIME)
726 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
727 blkg->stats.idle_time, cb, dev);
728 if (type == BLKIO_STAT_EMPTY_TIME)
729 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
730 blkg->stats.empty_time, cb, dev);
731 if (type == BLKIO_STAT_DEQUEUE)
732 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
733 blkg->stats.dequeue, cb, dev);
736 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
738 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
739 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
741 disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
742 blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
743 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
744 cb->fill(cb, key_str, disk_total);
748 static int blkio_policy_parse_and_set(char *buf,
749 struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
751 struct gendisk *disk = NULL;
752 char *s[4], *p, *major_s = NULL, *minor_s = NULL;
753 unsigned long major, minor;
754 int i = 0, ret = -EINVAL;
759 memset(s, 0, sizeof(s));
761 while ((p = strsep(&buf, " ")) != NULL) {
767 /* Prevent from inputing too many things */
775 p = strsep(&s[0], ":");
785 if (strict_strtoul(major_s, 10, &major))
788 if (strict_strtoul(minor_s, 10, &minor))
791 dev = MKDEV(major, minor);
793 if (strict_strtoull(s[1], 10, &temp))
796 /* For rule removal, do not check for device presence. */
798 disk = get_gendisk(dev, &part);
808 case BLKIO_POLICY_PROP:
809 if ((temp < BLKIO_WEIGHT_MIN && temp > 0) ||
810 temp > BLKIO_WEIGHT_MAX)
814 newpn->fileid = fileid;
815 newpn->val.weight = temp;
817 case BLKIO_POLICY_THROTL:
819 case BLKIO_THROTL_read_bps_device:
820 case BLKIO_THROTL_write_bps_device:
822 newpn->fileid = fileid;
823 newpn->val.bps = temp;
825 case BLKIO_THROTL_read_iops_device:
826 case BLKIO_THROTL_write_iops_device:
827 if (temp > THROTL_IOPS_MAX)
831 newpn->fileid = fileid;
832 newpn->val.iops = (unsigned int)temp;
845 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
848 struct blkio_policy_node *pn;
852 spin_lock_irqsave(&blkcg->lock, flags);
854 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
855 BLKIO_PROP_weight_device);
857 weight = pn->val.weight;
859 weight = blkcg->weight;
861 spin_unlock_irqrestore(&blkcg->lock, flags);
865 EXPORT_SYMBOL_GPL(blkcg_get_weight);
867 uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev)
869 struct blkio_policy_node *pn;
873 spin_lock_irqsave(&blkcg->lock, flags);
874 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
875 BLKIO_THROTL_read_bps_device);
878 spin_unlock_irqrestore(&blkcg->lock, flags);
883 uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
885 struct blkio_policy_node *pn;
889 spin_lock_irqsave(&blkcg->lock, flags);
890 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
891 BLKIO_THROTL_write_bps_device);
894 spin_unlock_irqrestore(&blkcg->lock, flags);
899 unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
901 struct blkio_policy_node *pn;
903 unsigned int iops = -1;
905 spin_lock_irqsave(&blkcg->lock, flags);
906 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
907 BLKIO_THROTL_read_iops_device);
910 spin_unlock_irqrestore(&blkcg->lock, flags);
915 unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
917 struct blkio_policy_node *pn;
919 unsigned int iops = -1;
921 spin_lock_irqsave(&blkcg->lock, flags);
922 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
923 BLKIO_THROTL_write_iops_device);
926 spin_unlock_irqrestore(&blkcg->lock, flags);
931 /* Checks whether user asked for deleting a policy rule */
932 static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
935 case BLKIO_POLICY_PROP:
936 if (pn->val.weight == 0)
939 case BLKIO_POLICY_THROTL:
941 case BLKIO_THROTL_read_bps_device:
942 case BLKIO_THROTL_write_bps_device:
943 if (pn->val.bps == 0)
946 case BLKIO_THROTL_read_iops_device:
947 case BLKIO_THROTL_write_iops_device:
948 if (pn->val.iops == 0)
959 static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
960 struct blkio_policy_node *newpn)
962 switch(oldpn->plid) {
963 case BLKIO_POLICY_PROP:
964 oldpn->val.weight = newpn->val.weight;
966 case BLKIO_POLICY_THROTL:
967 switch(newpn->fileid) {
968 case BLKIO_THROTL_read_bps_device:
969 case BLKIO_THROTL_write_bps_device:
970 oldpn->val.bps = newpn->val.bps;
972 case BLKIO_THROTL_read_iops_device:
973 case BLKIO_THROTL_write_iops_device:
974 oldpn->val.iops = newpn->val.iops;
983 * Some rules/values in blkg have changed. Propagate those to respective
986 static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
987 struct blkio_group *blkg, struct blkio_policy_node *pn)
989 unsigned int weight, iops;
993 case BLKIO_POLICY_PROP:
994 weight = pn->val.weight ? pn->val.weight :
996 blkio_update_group_weight(blkg, weight);
998 case BLKIO_POLICY_THROTL:
1000 case BLKIO_THROTL_read_bps_device:
1001 case BLKIO_THROTL_write_bps_device:
1002 bps = pn->val.bps ? pn->val.bps : (-1);
1003 blkio_update_group_bps(blkg, bps, pn->fileid);
1005 case BLKIO_THROTL_read_iops_device:
1006 case BLKIO_THROTL_write_iops_device:
1007 iops = pn->val.iops ? pn->val.iops : (-1);
1008 blkio_update_group_iops(blkg, iops, pn->fileid);
1018 * A policy node rule has been updated. Propagate this update to all the
1019 * block groups which might be affected by this update.
1021 static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
1022 struct blkio_policy_node *pn)
1024 struct blkio_group *blkg;
1025 struct hlist_node *n;
1027 spin_lock(&blkio_list_lock);
1028 spin_lock_irq(&blkcg->lock);
1030 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1031 if (pn->dev != blkg->dev || pn->plid != blkg->plid)
1033 blkio_update_blkg_policy(blkcg, blkg, pn);
1036 spin_unlock_irq(&blkcg->lock);
1037 spin_unlock(&blkio_list_lock);
1040 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1045 struct blkio_policy_node *newpn, *pn;
1046 struct blkio_cgroup *blkcg;
1048 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1049 int fileid = BLKIOFILE_ATTR(cft->private);
1051 buf = kstrdup(buffer, GFP_KERNEL);
1055 newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
1061 ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
1065 blkcg = cgroup_to_blkio_cgroup(cgrp);
1067 spin_lock_irq(&blkcg->lock);
1069 pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
1071 if (!blkio_delete_rule_command(newpn)) {
1072 blkio_policy_insert_node(blkcg, newpn);
1075 spin_unlock_irq(&blkcg->lock);
1076 goto update_io_group;
1079 if (blkio_delete_rule_command(newpn)) {
1080 blkio_policy_delete_node(pn);
1082 spin_unlock_irq(&blkcg->lock);
1083 goto update_io_group;
1085 spin_unlock_irq(&blkcg->lock);
1087 blkio_update_policy_rule(pn, newpn);
1090 blkio_update_policy_node_blkg(blkcg, newpn);
1101 blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
1104 case BLKIO_POLICY_PROP:
1105 if (pn->fileid == BLKIO_PROP_weight_device)
1106 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1107 MINOR(pn->dev), pn->val.weight);
1109 case BLKIO_POLICY_THROTL:
1110 switch(pn->fileid) {
1111 case BLKIO_THROTL_read_bps_device:
1112 case BLKIO_THROTL_write_bps_device:
1113 seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1114 MINOR(pn->dev), pn->val.bps);
1116 case BLKIO_THROTL_read_iops_device:
1117 case BLKIO_THROTL_write_iops_device:
1118 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1119 MINOR(pn->dev), pn->val.iops);
1128 /* cgroup files which read their data from policy nodes end up here */
1129 static void blkio_read_policy_node_files(struct cftype *cft,
1130 struct blkio_cgroup *blkcg, struct seq_file *m)
1132 struct blkio_policy_node *pn;
1134 if (!list_empty(&blkcg->policy_list)) {
1135 spin_lock_irq(&blkcg->lock);
1136 list_for_each_entry(pn, &blkcg->policy_list, node) {
1137 if (!pn_matches_cftype(cft, pn))
1139 blkio_print_policy_node(m, pn);
1141 spin_unlock_irq(&blkcg->lock);
1145 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1148 struct blkio_cgroup *blkcg;
1149 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1150 int name = BLKIOFILE_ATTR(cft->private);
1152 blkcg = cgroup_to_blkio_cgroup(cgrp);
1155 case BLKIO_POLICY_PROP:
1157 case BLKIO_PROP_weight_device:
1158 blkio_read_policy_node_files(cft, blkcg, m);
1164 case BLKIO_POLICY_THROTL:
1166 case BLKIO_THROTL_read_bps_device:
1167 case BLKIO_THROTL_write_bps_device:
1168 case BLKIO_THROTL_read_iops_device:
1169 case BLKIO_THROTL_write_iops_device:
1170 blkio_read_policy_node_files(cft, blkcg, m);
1183 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1184 struct cftype *cft, struct cgroup_map_cb *cb,
1185 enum stat_type type, bool show_total, bool pcpu)
1187 struct blkio_group *blkg;
1188 struct hlist_node *n;
1189 uint64_t cgroup_total = 0;
1192 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1194 if (!cftype_blkg_same_policy(cft, blkg))
1197 cgroup_total += blkio_get_stat_cpu(blkg, cb,
1200 spin_lock_irq(&blkg->stats_lock);
1201 cgroup_total += blkio_get_stat(blkg, cb,
1203 spin_unlock_irq(&blkg->stats_lock);
1208 cb->fill(cb, "Total", cgroup_total);
1213 /* All map kind of cgroup file get serviced by this function */
1214 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1215 struct cgroup_map_cb *cb)
1217 struct blkio_cgroup *blkcg;
1218 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1219 int name = BLKIOFILE_ATTR(cft->private);
1221 blkcg = cgroup_to_blkio_cgroup(cgrp);
1224 case BLKIO_POLICY_PROP:
1226 case BLKIO_PROP_time:
1227 return blkio_read_blkg_stats(blkcg, cft, cb,
1228 BLKIO_STAT_TIME, 0, 0);
1229 case BLKIO_PROP_sectors:
1230 return blkio_read_blkg_stats(blkcg, cft, cb,
1231 BLKIO_STAT_CPU_SECTORS, 0, 1);
1232 case BLKIO_PROP_io_service_bytes:
1233 return blkio_read_blkg_stats(blkcg, cft, cb,
1234 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1235 case BLKIO_PROP_io_serviced:
1236 return blkio_read_blkg_stats(blkcg, cft, cb,
1237 BLKIO_STAT_CPU_SERVICED, 1, 1);
1238 case BLKIO_PROP_io_service_time:
1239 return blkio_read_blkg_stats(blkcg, cft, cb,
1240 BLKIO_STAT_SERVICE_TIME, 1, 0);
1241 case BLKIO_PROP_io_wait_time:
1242 return blkio_read_blkg_stats(blkcg, cft, cb,
1243 BLKIO_STAT_WAIT_TIME, 1, 0);
1244 case BLKIO_PROP_io_merged:
1245 return blkio_read_blkg_stats(blkcg, cft, cb,
1246 BLKIO_STAT_CPU_MERGED, 1, 1);
1247 case BLKIO_PROP_io_queued:
1248 return blkio_read_blkg_stats(blkcg, cft, cb,
1249 BLKIO_STAT_QUEUED, 1, 0);
1250 #ifdef CONFIG_DEBUG_BLK_CGROUP
1251 case BLKIO_PROP_unaccounted_time:
1252 return blkio_read_blkg_stats(blkcg, cft, cb,
1253 BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
1254 case BLKIO_PROP_dequeue:
1255 return blkio_read_blkg_stats(blkcg, cft, cb,
1256 BLKIO_STAT_DEQUEUE, 0, 0);
1257 case BLKIO_PROP_avg_queue_size:
1258 return blkio_read_blkg_stats(blkcg, cft, cb,
1259 BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
1260 case BLKIO_PROP_group_wait_time:
1261 return blkio_read_blkg_stats(blkcg, cft, cb,
1262 BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
1263 case BLKIO_PROP_idle_time:
1264 return blkio_read_blkg_stats(blkcg, cft, cb,
1265 BLKIO_STAT_IDLE_TIME, 0, 0);
1266 case BLKIO_PROP_empty_time:
1267 return blkio_read_blkg_stats(blkcg, cft, cb,
1268 BLKIO_STAT_EMPTY_TIME, 0, 0);
1274 case BLKIO_POLICY_THROTL:
1276 case BLKIO_THROTL_io_service_bytes:
1277 return blkio_read_blkg_stats(blkcg, cft, cb,
1278 BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1279 case BLKIO_THROTL_io_serviced:
1280 return blkio_read_blkg_stats(blkcg, cft, cb,
1281 BLKIO_STAT_CPU_SERVICED, 1, 1);
1293 static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1295 struct blkio_group *blkg;
1296 struct hlist_node *n;
1297 struct blkio_policy_node *pn;
1299 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1302 spin_lock(&blkio_list_lock);
1303 spin_lock_irq(&blkcg->lock);
1304 blkcg->weight = (unsigned int)val;
1306 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1307 pn = blkio_policy_search_node(blkcg, blkg->dev,
1308 BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1312 blkio_update_group_weight(blkg, blkcg->weight);
1314 spin_unlock_irq(&blkcg->lock);
1315 spin_unlock(&blkio_list_lock);
1319 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1320 struct blkio_cgroup *blkcg;
1321 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1322 int name = BLKIOFILE_ATTR(cft->private);
1324 blkcg = cgroup_to_blkio_cgroup(cgrp);
1327 case BLKIO_POLICY_PROP:
1329 case BLKIO_PROP_weight:
1330 return (u64)blkcg->weight;
1340 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1342 struct blkio_cgroup *blkcg;
1343 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1344 int name = BLKIOFILE_ATTR(cft->private);
1346 blkcg = cgroup_to_blkio_cgroup(cgrp);
1349 case BLKIO_POLICY_PROP:
1351 case BLKIO_PROP_weight:
1352 return blkio_weight_write(blkcg, val);
1362 struct cftype blkio_files[] = {
1364 .name = "weight_device",
1365 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1366 BLKIO_PROP_weight_device),
1367 .read_seq_string = blkiocg_file_read,
1368 .write_string = blkiocg_file_write,
1369 .max_write_len = 256,
1373 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1375 .read_u64 = blkiocg_file_read_u64,
1376 .write_u64 = blkiocg_file_write_u64,
1380 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1382 .read_map = blkiocg_file_read_map,
1386 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1387 BLKIO_PROP_sectors),
1388 .read_map = blkiocg_file_read_map,
1391 .name = "io_service_bytes",
1392 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1393 BLKIO_PROP_io_service_bytes),
1394 .read_map = blkiocg_file_read_map,
1397 .name = "io_serviced",
1398 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1399 BLKIO_PROP_io_serviced),
1400 .read_map = blkiocg_file_read_map,
1403 .name = "io_service_time",
1404 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1405 BLKIO_PROP_io_service_time),
1406 .read_map = blkiocg_file_read_map,
1409 .name = "io_wait_time",
1410 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1411 BLKIO_PROP_io_wait_time),
1412 .read_map = blkiocg_file_read_map,
1415 .name = "io_merged",
1416 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1417 BLKIO_PROP_io_merged),
1418 .read_map = blkiocg_file_read_map,
1421 .name = "io_queued",
1422 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1423 BLKIO_PROP_io_queued),
1424 .read_map = blkiocg_file_read_map,
1427 .name = "reset_stats",
1428 .write_u64 = blkiocg_reset_stats,
1430 #ifdef CONFIG_BLK_DEV_THROTTLING
1432 .name = "throttle.read_bps_device",
1433 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1434 BLKIO_THROTL_read_bps_device),
1435 .read_seq_string = blkiocg_file_read,
1436 .write_string = blkiocg_file_write,
1437 .max_write_len = 256,
1441 .name = "throttle.write_bps_device",
1442 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1443 BLKIO_THROTL_write_bps_device),
1444 .read_seq_string = blkiocg_file_read,
1445 .write_string = blkiocg_file_write,
1446 .max_write_len = 256,
1450 .name = "throttle.read_iops_device",
1451 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1452 BLKIO_THROTL_read_iops_device),
1453 .read_seq_string = blkiocg_file_read,
1454 .write_string = blkiocg_file_write,
1455 .max_write_len = 256,
1459 .name = "throttle.write_iops_device",
1460 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1461 BLKIO_THROTL_write_iops_device),
1462 .read_seq_string = blkiocg_file_read,
1463 .write_string = blkiocg_file_write,
1464 .max_write_len = 256,
1467 .name = "throttle.io_service_bytes",
1468 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1469 BLKIO_THROTL_io_service_bytes),
1470 .read_map = blkiocg_file_read_map,
1473 .name = "throttle.io_serviced",
1474 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1475 BLKIO_THROTL_io_serviced),
1476 .read_map = blkiocg_file_read_map,
1478 #endif /* CONFIG_BLK_DEV_THROTTLING */
1480 #ifdef CONFIG_DEBUG_BLK_CGROUP
1482 .name = "avg_queue_size",
1483 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1484 BLKIO_PROP_avg_queue_size),
1485 .read_map = blkiocg_file_read_map,
1488 .name = "group_wait_time",
1489 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1490 BLKIO_PROP_group_wait_time),
1491 .read_map = blkiocg_file_read_map,
1494 .name = "idle_time",
1495 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1496 BLKIO_PROP_idle_time),
1497 .read_map = blkiocg_file_read_map,
1500 .name = "empty_time",
1501 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1502 BLKIO_PROP_empty_time),
1503 .read_map = blkiocg_file_read_map,
1507 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1508 BLKIO_PROP_dequeue),
1509 .read_map = blkiocg_file_read_map,
1512 .name = "unaccounted_time",
1513 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1514 BLKIO_PROP_unaccounted_time),
1515 .read_map = blkiocg_file_read_map,
1520 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1522 return cgroup_add_files(cgroup, subsys, blkio_files,
1523 ARRAY_SIZE(blkio_files));
1526 static void blkiocg_destroy(struct cgroup *cgroup)
1528 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1529 unsigned long flags;
1530 struct blkio_group *blkg;
1532 struct blkio_policy_type *blkiop;
1533 struct blkio_policy_node *pn, *pntmp;
1537 spin_lock_irqsave(&blkcg->lock, flags);
1539 if (hlist_empty(&blkcg->blkg_list)) {
1540 spin_unlock_irqrestore(&blkcg->lock, flags);
1544 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1546 key = rcu_dereference(blkg->key);
1547 __blkiocg_del_blkio_group(blkg);
1549 spin_unlock_irqrestore(&blkcg->lock, flags);
1552 * This blkio_group is being unlinked as associated cgroup is
1553 * going away. Let all the IO controlling policies know about
1556 spin_lock(&blkio_list_lock);
1557 list_for_each_entry(blkiop, &blkio_list, list) {
1558 if (blkiop->plid != blkg->plid)
1560 blkiop->ops.blkio_unlink_group_fn(key, blkg);
1562 spin_unlock(&blkio_list_lock);
1565 list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1566 blkio_policy_delete_node(pn);
1570 free_css_id(&blkio_subsys, &blkcg->css);
1572 if (blkcg != &blkio_root_cgroup)
1576 static struct cgroup_subsys_state *blkiocg_create(struct cgroup *cgroup)
1578 struct blkio_cgroup *blkcg;
1579 struct cgroup *parent = cgroup->parent;
1582 blkcg = &blkio_root_cgroup;
1586 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1588 return ERR_PTR(-ENOMEM);
1590 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1592 spin_lock_init(&blkcg->lock);
1593 INIT_HLIST_HEAD(&blkcg->blkg_list);
1595 INIT_LIST_HEAD(&blkcg->policy_list);
1600 * We cannot support shared io contexts, as we have no mean to support
1601 * two tasks with the same ioc in two different groups without major rework
1602 * of the main cic data structures. For now we allow a task to change
1603 * its cgroup only if it's the only owner of its ioc.
1605 static int blkiocg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1607 struct task_struct *task;
1608 struct io_context *ioc;
1611 /* task_lock() is needed to avoid races with exit_io_context() */
1612 cgroup_taskset_for_each(task, cgrp, tset) {
1614 ioc = task->io_context;
1615 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1624 static void blkiocg_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
1626 struct task_struct *task;
1627 struct io_context *ioc;
1629 cgroup_taskset_for_each(task, cgrp, tset) {
1630 /* we don't lose anything even if ioc allocation fails */
1631 ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
1633 ioc_cgroup_changed(ioc);
1634 put_io_context(ioc);
1639 struct cgroup_subsys blkio_subsys = {
1641 .create = blkiocg_create,
1642 .can_attach = blkiocg_can_attach,
1643 .attach = blkiocg_attach,
1644 .destroy = blkiocg_destroy,
1645 .populate = blkiocg_populate,
1646 #ifdef CONFIG_BLK_CGROUP
1647 /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */
1648 .subsys_id = blkio_subsys_id,
1651 .module = THIS_MODULE,
1653 EXPORT_SYMBOL_GPL(blkio_subsys);
1655 void blkio_policy_register(struct blkio_policy_type *blkiop)
1657 spin_lock(&blkio_list_lock);
1658 list_add_tail(&blkiop->list, &blkio_list);
1659 spin_unlock(&blkio_list_lock);
1661 EXPORT_SYMBOL_GPL(blkio_policy_register);
1663 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1665 spin_lock(&blkio_list_lock);
1666 list_del_init(&blkiop->list);
1667 spin_unlock(&blkio_list_lock);
1669 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1671 static int __init init_cgroup_blkio(void)
1673 return cgroup_load_subsys(&blkio_subsys);
1676 static void __exit exit_cgroup_blkio(void)
1678 cgroup_unload_subsys(&blkio_subsys);
1681 module_init(init_cgroup_blkio);
1682 module_exit(exit_cgroup_blkio);
1683 MODULE_LICENSE("GPL");