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);
395 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
396 uint64_t bytes, bool direction, bool sync)
398 struct blkio_group_stats *stats;
401 spin_lock_irqsave(&blkg->stats_lock, flags);
402 stats = &blkg->stats;
403 stats->sectors += bytes >> 9;
404 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICED], 1, direction,
406 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_BYTES], bytes,
408 spin_unlock_irqrestore(&blkg->stats_lock, flags);
410 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
412 void blkiocg_update_completion_stats(struct blkio_group *blkg,
413 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
415 struct blkio_group_stats *stats;
417 unsigned long long now = sched_clock();
419 spin_lock_irqsave(&blkg->stats_lock, flags);
420 stats = &blkg->stats;
421 if (time_after64(now, io_start_time))
422 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
423 now - io_start_time, direction, sync);
424 if (time_after64(io_start_time, start_time))
425 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
426 io_start_time - start_time, direction, sync);
427 spin_unlock_irqrestore(&blkg->stats_lock, flags);
429 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
431 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
436 spin_lock_irqsave(&blkg->stats_lock, flags);
437 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_MERGED], 1, direction,
439 spin_unlock_irqrestore(&blkg->stats_lock, flags);
441 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
443 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
444 struct blkio_group *blkg, void *key, dev_t dev,
445 enum blkio_policy_id plid)
449 spin_lock_irqsave(&blkcg->lock, flags);
450 spin_lock_init(&blkg->stats_lock);
451 rcu_assign_pointer(blkg->key, key);
452 blkg->blkcg_id = css_id(&blkcg->css);
453 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
455 spin_unlock_irqrestore(&blkcg->lock, flags);
456 /* Need to take css reference ? */
457 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
460 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
462 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
464 hlist_del_init_rcu(&blkg->blkcg_node);
469 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
470 * indicating that blk_group was unhashed by the time we got to it.
472 int blkiocg_del_blkio_group(struct blkio_group *blkg)
474 struct blkio_cgroup *blkcg;
476 struct cgroup_subsys_state *css;
480 css = css_lookup(&blkio_subsys, blkg->blkcg_id);
482 blkcg = container_of(css, struct blkio_cgroup, css);
483 spin_lock_irqsave(&blkcg->lock, flags);
484 if (!hlist_unhashed(&blkg->blkcg_node)) {
485 __blkiocg_del_blkio_group(blkg);
488 spin_unlock_irqrestore(&blkcg->lock, flags);
494 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
496 /* called under rcu_read_lock(). */
497 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
499 struct blkio_group *blkg;
500 struct hlist_node *n;
503 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
511 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
514 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
516 struct blkio_cgroup *blkcg;
517 struct blkio_group *blkg;
518 struct blkio_group_stats *stats;
519 struct hlist_node *n;
520 uint64_t queued[BLKIO_STAT_TOTAL];
522 #ifdef CONFIG_DEBUG_BLK_CGROUP
523 bool idling, waiting, empty;
524 unsigned long long now = sched_clock();
527 blkcg = cgroup_to_blkio_cgroup(cgroup);
528 spin_lock_irq(&blkcg->lock);
529 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
530 spin_lock(&blkg->stats_lock);
531 stats = &blkg->stats;
532 #ifdef CONFIG_DEBUG_BLK_CGROUP
533 idling = blkio_blkg_idling(stats);
534 waiting = blkio_blkg_waiting(stats);
535 empty = blkio_blkg_empty(stats);
537 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
538 queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
539 memset(stats, 0, sizeof(struct blkio_group_stats));
540 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
541 stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
542 #ifdef CONFIG_DEBUG_BLK_CGROUP
544 blkio_mark_blkg_idling(stats);
545 stats->start_idle_time = now;
548 blkio_mark_blkg_waiting(stats);
549 stats->start_group_wait_time = now;
552 blkio_mark_blkg_empty(stats);
553 stats->start_empty_time = now;
556 spin_unlock(&blkg->stats_lock);
558 spin_unlock_irq(&blkcg->lock);
562 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
563 int chars_left, bool diskname_only)
565 snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
566 chars_left -= strlen(str);
567 if (chars_left <= 0) {
569 "Possibly incorrect cgroup stat display format");
575 case BLKIO_STAT_READ:
576 strlcat(str, " Read", chars_left);
578 case BLKIO_STAT_WRITE:
579 strlcat(str, " Write", chars_left);
581 case BLKIO_STAT_SYNC:
582 strlcat(str, " Sync", chars_left);
584 case BLKIO_STAT_ASYNC:
585 strlcat(str, " Async", chars_left);
587 case BLKIO_STAT_TOTAL:
588 strlcat(str, " Total", chars_left);
591 strlcat(str, " Invalid", chars_left);
595 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
596 struct cgroup_map_cb *cb, dev_t dev)
598 blkio_get_key_name(0, dev, str, chars_left, true);
599 cb->fill(cb, str, val);
603 /* This should be called with blkg->stats_lock held */
604 static uint64_t blkio_get_stat(struct blkio_group *blkg,
605 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
608 char key_str[MAX_KEY_LEN];
609 enum stat_sub_type sub_type;
611 if (type == BLKIO_STAT_TIME)
612 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
613 blkg->stats.time, cb, dev);
614 if (type == BLKIO_STAT_SECTORS)
615 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
616 blkg->stats.sectors, cb, dev);
617 #ifdef CONFIG_DEBUG_BLK_CGROUP
618 if (type == BLKIO_STAT_UNACCOUNTED_TIME)
619 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
620 blkg->stats.unaccounted_time, cb, dev);
621 if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
622 uint64_t sum = blkg->stats.avg_queue_size_sum;
623 uint64_t samples = blkg->stats.avg_queue_size_samples;
625 do_div(sum, samples);
628 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
630 if (type == BLKIO_STAT_GROUP_WAIT_TIME)
631 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
632 blkg->stats.group_wait_time, cb, dev);
633 if (type == BLKIO_STAT_IDLE_TIME)
634 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
635 blkg->stats.idle_time, cb, dev);
636 if (type == BLKIO_STAT_EMPTY_TIME)
637 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
638 blkg->stats.empty_time, cb, dev);
639 if (type == BLKIO_STAT_DEQUEUE)
640 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
641 blkg->stats.dequeue, cb, dev);
644 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
646 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
647 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
649 disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
650 blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
651 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
652 cb->fill(cb, key_str, disk_total);
656 static int blkio_check_dev_num(dev_t dev)
659 struct gendisk *disk;
661 disk = get_gendisk(dev, &part);
668 static int blkio_policy_parse_and_set(char *buf,
669 struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
671 char *s[4], *p, *major_s = NULL, *minor_s = NULL;
673 unsigned long major, minor, temp;
678 memset(s, 0, sizeof(s));
680 while ((p = strsep(&buf, " ")) != NULL) {
686 /* Prevent from inputing too many things */
694 p = strsep(&s[0], ":");
704 ret = strict_strtoul(major_s, 10, &major);
708 ret = strict_strtoul(minor_s, 10, &minor);
712 dev = MKDEV(major, minor);
714 ret = blkio_check_dev_num(dev);
724 case BLKIO_POLICY_PROP:
725 ret = strict_strtoul(s[1], 10, &temp);
726 if (ret || (temp < BLKIO_WEIGHT_MIN && temp > 0) ||
727 temp > BLKIO_WEIGHT_MAX)
731 newpn->fileid = fileid;
732 newpn->val.weight = temp;
734 case BLKIO_POLICY_THROTL:
736 case BLKIO_THROTL_read_bps_device:
737 case BLKIO_THROTL_write_bps_device:
738 ret = strict_strtoull(s[1], 10, &bps);
743 newpn->fileid = fileid;
744 newpn->val.bps = bps;
746 case BLKIO_THROTL_read_iops_device:
747 case BLKIO_THROTL_write_iops_device:
748 ret = strict_strtoull(s[1], 10, &iops);
752 if (iops > THROTL_IOPS_MAX)
756 newpn->fileid = fileid;
757 newpn->val.iops = (unsigned int)iops;
768 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
771 struct blkio_policy_node *pn;
773 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
774 BLKIO_PROP_weight_device);
776 return pn->val.weight;
778 return blkcg->weight;
780 EXPORT_SYMBOL_GPL(blkcg_get_weight);
782 uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev)
784 struct blkio_policy_node *pn;
786 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
787 BLKIO_THROTL_read_bps_device);
794 uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
796 struct blkio_policy_node *pn;
797 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
798 BLKIO_THROTL_write_bps_device);
805 unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
807 struct blkio_policy_node *pn;
809 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
810 BLKIO_THROTL_read_iops_device);
817 unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
819 struct blkio_policy_node *pn;
820 pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
821 BLKIO_THROTL_write_iops_device);
828 /* Checks whether user asked for deleting a policy rule */
829 static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
832 case BLKIO_POLICY_PROP:
833 if (pn->val.weight == 0)
836 case BLKIO_POLICY_THROTL:
838 case BLKIO_THROTL_read_bps_device:
839 case BLKIO_THROTL_write_bps_device:
840 if (pn->val.bps == 0)
843 case BLKIO_THROTL_read_iops_device:
844 case BLKIO_THROTL_write_iops_device:
845 if (pn->val.iops == 0)
856 static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
857 struct blkio_policy_node *newpn)
859 switch(oldpn->plid) {
860 case BLKIO_POLICY_PROP:
861 oldpn->val.weight = newpn->val.weight;
863 case BLKIO_POLICY_THROTL:
864 switch(newpn->fileid) {
865 case BLKIO_THROTL_read_bps_device:
866 case BLKIO_THROTL_write_bps_device:
867 oldpn->val.bps = newpn->val.bps;
869 case BLKIO_THROTL_read_iops_device:
870 case BLKIO_THROTL_write_iops_device:
871 oldpn->val.iops = newpn->val.iops;
880 * Some rules/values in blkg have changed. Propagate those to respective
883 static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
884 struct blkio_group *blkg, struct blkio_policy_node *pn)
886 unsigned int weight, iops;
890 case BLKIO_POLICY_PROP:
891 weight = pn->val.weight ? pn->val.weight :
893 blkio_update_group_weight(blkg, weight);
895 case BLKIO_POLICY_THROTL:
897 case BLKIO_THROTL_read_bps_device:
898 case BLKIO_THROTL_write_bps_device:
899 bps = pn->val.bps ? pn->val.bps : (-1);
900 blkio_update_group_bps(blkg, bps, pn->fileid);
902 case BLKIO_THROTL_read_iops_device:
903 case BLKIO_THROTL_write_iops_device:
904 iops = pn->val.iops ? pn->val.iops : (-1);
905 blkio_update_group_iops(blkg, iops, pn->fileid);
915 * A policy node rule has been updated. Propagate this update to all the
916 * block groups which might be affected by this update.
918 static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
919 struct blkio_policy_node *pn)
921 struct blkio_group *blkg;
922 struct hlist_node *n;
924 spin_lock(&blkio_list_lock);
925 spin_lock_irq(&blkcg->lock);
927 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
928 if (pn->dev != blkg->dev || pn->plid != blkg->plid)
930 blkio_update_blkg_policy(blkcg, blkg, pn);
933 spin_unlock_irq(&blkcg->lock);
934 spin_unlock(&blkio_list_lock);
937 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
942 struct blkio_policy_node *newpn, *pn;
943 struct blkio_cgroup *blkcg;
945 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
946 int fileid = BLKIOFILE_ATTR(cft->private);
948 buf = kstrdup(buffer, GFP_KERNEL);
952 newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
958 ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
962 blkcg = cgroup_to_blkio_cgroup(cgrp);
964 spin_lock_irq(&blkcg->lock);
966 pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
968 if (!blkio_delete_rule_command(newpn)) {
969 blkio_policy_insert_node(blkcg, newpn);
972 spin_unlock_irq(&blkcg->lock);
973 goto update_io_group;
976 if (blkio_delete_rule_command(newpn)) {
977 blkio_policy_delete_node(pn);
978 spin_unlock_irq(&blkcg->lock);
979 goto update_io_group;
981 spin_unlock_irq(&blkcg->lock);
983 blkio_update_policy_rule(pn, newpn);
986 blkio_update_policy_node_blkg(blkcg, newpn);
997 blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
1000 case BLKIO_POLICY_PROP:
1001 if (pn->fileid == BLKIO_PROP_weight_device)
1002 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1003 MINOR(pn->dev), pn->val.weight);
1005 case BLKIO_POLICY_THROTL:
1006 switch(pn->fileid) {
1007 case BLKIO_THROTL_read_bps_device:
1008 case BLKIO_THROTL_write_bps_device:
1009 seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1010 MINOR(pn->dev), pn->val.bps);
1012 case BLKIO_THROTL_read_iops_device:
1013 case BLKIO_THROTL_write_iops_device:
1014 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1015 MINOR(pn->dev), pn->val.iops);
1024 /* cgroup files which read their data from policy nodes end up here */
1025 static void blkio_read_policy_node_files(struct cftype *cft,
1026 struct blkio_cgroup *blkcg, struct seq_file *m)
1028 struct blkio_policy_node *pn;
1030 if (!list_empty(&blkcg->policy_list)) {
1031 spin_lock_irq(&blkcg->lock);
1032 list_for_each_entry(pn, &blkcg->policy_list, node) {
1033 if (!pn_matches_cftype(cft, pn))
1035 blkio_print_policy_node(m, pn);
1037 spin_unlock_irq(&blkcg->lock);
1041 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1044 struct blkio_cgroup *blkcg;
1045 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1046 int name = BLKIOFILE_ATTR(cft->private);
1048 blkcg = cgroup_to_blkio_cgroup(cgrp);
1051 case BLKIO_POLICY_PROP:
1053 case BLKIO_PROP_weight_device:
1054 blkio_read_policy_node_files(cft, blkcg, m);
1060 case BLKIO_POLICY_THROTL:
1062 case BLKIO_THROTL_read_bps_device:
1063 case BLKIO_THROTL_write_bps_device:
1064 case BLKIO_THROTL_read_iops_device:
1065 case BLKIO_THROTL_write_iops_device:
1066 blkio_read_policy_node_files(cft, blkcg, m);
1079 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1080 struct cftype *cft, struct cgroup_map_cb *cb, enum stat_type type,
1083 struct blkio_group *blkg;
1084 struct hlist_node *n;
1085 uint64_t cgroup_total = 0;
1088 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1090 if (!cftype_blkg_same_policy(cft, blkg))
1092 spin_lock_irq(&blkg->stats_lock);
1093 cgroup_total += blkio_get_stat(blkg, cb, blkg->dev,
1095 spin_unlock_irq(&blkg->stats_lock);
1099 cb->fill(cb, "Total", cgroup_total);
1104 /* All map kind of cgroup file get serviced by this function */
1105 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1106 struct cgroup_map_cb *cb)
1108 struct blkio_cgroup *blkcg;
1109 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1110 int name = BLKIOFILE_ATTR(cft->private);
1112 blkcg = cgroup_to_blkio_cgroup(cgrp);
1115 case BLKIO_POLICY_PROP:
1117 case BLKIO_PROP_time:
1118 return blkio_read_blkg_stats(blkcg, cft, cb,
1119 BLKIO_STAT_TIME, 0);
1120 case BLKIO_PROP_sectors:
1121 return blkio_read_blkg_stats(blkcg, cft, cb,
1122 BLKIO_STAT_SECTORS, 0);
1123 case BLKIO_PROP_io_service_bytes:
1124 return blkio_read_blkg_stats(blkcg, cft, cb,
1125 BLKIO_STAT_SERVICE_BYTES, 1);
1126 case BLKIO_PROP_io_serviced:
1127 return blkio_read_blkg_stats(blkcg, cft, cb,
1128 BLKIO_STAT_SERVICED, 1);
1129 case BLKIO_PROP_io_service_time:
1130 return blkio_read_blkg_stats(blkcg, cft, cb,
1131 BLKIO_STAT_SERVICE_TIME, 1);
1132 case BLKIO_PROP_io_wait_time:
1133 return blkio_read_blkg_stats(blkcg, cft, cb,
1134 BLKIO_STAT_WAIT_TIME, 1);
1135 case BLKIO_PROP_io_merged:
1136 return blkio_read_blkg_stats(blkcg, cft, cb,
1137 BLKIO_STAT_MERGED, 1);
1138 case BLKIO_PROP_io_queued:
1139 return blkio_read_blkg_stats(blkcg, cft, cb,
1140 BLKIO_STAT_QUEUED, 1);
1141 #ifdef CONFIG_DEBUG_BLK_CGROUP
1142 case BLKIO_PROP_unaccounted_time:
1143 return blkio_read_blkg_stats(blkcg, cft, cb,
1144 BLKIO_STAT_UNACCOUNTED_TIME, 0);
1145 case BLKIO_PROP_dequeue:
1146 return blkio_read_blkg_stats(blkcg, cft, cb,
1147 BLKIO_STAT_DEQUEUE, 0);
1148 case BLKIO_PROP_avg_queue_size:
1149 return blkio_read_blkg_stats(blkcg, cft, cb,
1150 BLKIO_STAT_AVG_QUEUE_SIZE, 0);
1151 case BLKIO_PROP_group_wait_time:
1152 return blkio_read_blkg_stats(blkcg, cft, cb,
1153 BLKIO_STAT_GROUP_WAIT_TIME, 0);
1154 case BLKIO_PROP_idle_time:
1155 return blkio_read_blkg_stats(blkcg, cft, cb,
1156 BLKIO_STAT_IDLE_TIME, 0);
1157 case BLKIO_PROP_empty_time:
1158 return blkio_read_blkg_stats(blkcg, cft, cb,
1159 BLKIO_STAT_EMPTY_TIME, 0);
1165 case BLKIO_POLICY_THROTL:
1167 case BLKIO_THROTL_io_service_bytes:
1168 return blkio_read_blkg_stats(blkcg, cft, cb,
1169 BLKIO_STAT_SERVICE_BYTES, 1);
1170 case BLKIO_THROTL_io_serviced:
1171 return blkio_read_blkg_stats(blkcg, cft, cb,
1172 BLKIO_STAT_SERVICED, 1);
1184 static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1186 struct blkio_group *blkg;
1187 struct hlist_node *n;
1188 struct blkio_policy_node *pn;
1190 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1193 spin_lock(&blkio_list_lock);
1194 spin_lock_irq(&blkcg->lock);
1195 blkcg->weight = (unsigned int)val;
1197 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1198 pn = blkio_policy_search_node(blkcg, blkg->dev,
1199 BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1203 blkio_update_group_weight(blkg, blkcg->weight);
1205 spin_unlock_irq(&blkcg->lock);
1206 spin_unlock(&blkio_list_lock);
1210 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1211 struct blkio_cgroup *blkcg;
1212 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1213 int name = BLKIOFILE_ATTR(cft->private);
1215 blkcg = cgroup_to_blkio_cgroup(cgrp);
1218 case BLKIO_POLICY_PROP:
1220 case BLKIO_PROP_weight:
1221 return (u64)blkcg->weight;
1231 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1233 struct blkio_cgroup *blkcg;
1234 enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1235 int name = BLKIOFILE_ATTR(cft->private);
1237 blkcg = cgroup_to_blkio_cgroup(cgrp);
1240 case BLKIO_POLICY_PROP:
1242 case BLKIO_PROP_weight:
1243 return blkio_weight_write(blkcg, val);
1253 struct cftype blkio_files[] = {
1255 .name = "weight_device",
1256 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1257 BLKIO_PROP_weight_device),
1258 .read_seq_string = blkiocg_file_read,
1259 .write_string = blkiocg_file_write,
1260 .max_write_len = 256,
1264 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1266 .read_u64 = blkiocg_file_read_u64,
1267 .write_u64 = blkiocg_file_write_u64,
1271 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1273 .read_map = blkiocg_file_read_map,
1277 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1278 BLKIO_PROP_sectors),
1279 .read_map = blkiocg_file_read_map,
1282 .name = "io_service_bytes",
1283 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1284 BLKIO_PROP_io_service_bytes),
1285 .read_map = blkiocg_file_read_map,
1288 .name = "io_serviced",
1289 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1290 BLKIO_PROP_io_serviced),
1291 .read_map = blkiocg_file_read_map,
1294 .name = "io_service_time",
1295 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1296 BLKIO_PROP_io_service_time),
1297 .read_map = blkiocg_file_read_map,
1300 .name = "io_wait_time",
1301 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1302 BLKIO_PROP_io_wait_time),
1303 .read_map = blkiocg_file_read_map,
1306 .name = "io_merged",
1307 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1308 BLKIO_PROP_io_merged),
1309 .read_map = blkiocg_file_read_map,
1312 .name = "io_queued",
1313 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1314 BLKIO_PROP_io_queued),
1315 .read_map = blkiocg_file_read_map,
1318 .name = "reset_stats",
1319 .write_u64 = blkiocg_reset_stats,
1321 #ifdef CONFIG_BLK_DEV_THROTTLING
1323 .name = "throttle.read_bps_device",
1324 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1325 BLKIO_THROTL_read_bps_device),
1326 .read_seq_string = blkiocg_file_read,
1327 .write_string = blkiocg_file_write,
1328 .max_write_len = 256,
1332 .name = "throttle.write_bps_device",
1333 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1334 BLKIO_THROTL_write_bps_device),
1335 .read_seq_string = blkiocg_file_read,
1336 .write_string = blkiocg_file_write,
1337 .max_write_len = 256,
1341 .name = "throttle.read_iops_device",
1342 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1343 BLKIO_THROTL_read_iops_device),
1344 .read_seq_string = blkiocg_file_read,
1345 .write_string = blkiocg_file_write,
1346 .max_write_len = 256,
1350 .name = "throttle.write_iops_device",
1351 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1352 BLKIO_THROTL_write_iops_device),
1353 .read_seq_string = blkiocg_file_read,
1354 .write_string = blkiocg_file_write,
1355 .max_write_len = 256,
1358 .name = "throttle.io_service_bytes",
1359 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1360 BLKIO_THROTL_io_service_bytes),
1361 .read_map = blkiocg_file_read_map,
1364 .name = "throttle.io_serviced",
1365 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1366 BLKIO_THROTL_io_serviced),
1367 .read_map = blkiocg_file_read_map,
1369 #endif /* CONFIG_BLK_DEV_THROTTLING */
1371 #ifdef CONFIG_DEBUG_BLK_CGROUP
1373 .name = "avg_queue_size",
1374 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1375 BLKIO_PROP_avg_queue_size),
1376 .read_map = blkiocg_file_read_map,
1379 .name = "group_wait_time",
1380 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1381 BLKIO_PROP_group_wait_time),
1382 .read_map = blkiocg_file_read_map,
1385 .name = "idle_time",
1386 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1387 BLKIO_PROP_idle_time),
1388 .read_map = blkiocg_file_read_map,
1391 .name = "empty_time",
1392 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1393 BLKIO_PROP_empty_time),
1394 .read_map = blkiocg_file_read_map,
1398 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1399 BLKIO_PROP_dequeue),
1400 .read_map = blkiocg_file_read_map,
1403 .name = "unaccounted_time",
1404 .private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1405 BLKIO_PROP_unaccounted_time),
1406 .read_map = blkiocg_file_read_map,
1411 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1413 return cgroup_add_files(cgroup, subsys, blkio_files,
1414 ARRAY_SIZE(blkio_files));
1417 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1419 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1420 unsigned long flags;
1421 struct blkio_group *blkg;
1423 struct blkio_policy_type *blkiop;
1424 struct blkio_policy_node *pn, *pntmp;
1428 spin_lock_irqsave(&blkcg->lock, flags);
1430 if (hlist_empty(&blkcg->blkg_list)) {
1431 spin_unlock_irqrestore(&blkcg->lock, flags);
1435 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1437 key = rcu_dereference(blkg->key);
1438 __blkiocg_del_blkio_group(blkg);
1440 spin_unlock_irqrestore(&blkcg->lock, flags);
1443 * This blkio_group is being unlinked as associated cgroup is
1444 * going away. Let all the IO controlling policies know about
1447 spin_lock(&blkio_list_lock);
1448 list_for_each_entry(blkiop, &blkio_list, list) {
1449 if (blkiop->plid != blkg->plid)
1451 blkiop->ops.blkio_unlink_group_fn(key, blkg);
1453 spin_unlock(&blkio_list_lock);
1456 list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1457 blkio_policy_delete_node(pn);
1461 free_css_id(&blkio_subsys, &blkcg->css);
1463 if (blkcg != &blkio_root_cgroup)
1467 static struct cgroup_subsys_state *
1468 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1470 struct blkio_cgroup *blkcg;
1471 struct cgroup *parent = cgroup->parent;
1474 blkcg = &blkio_root_cgroup;
1478 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1480 return ERR_PTR(-ENOMEM);
1482 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1484 spin_lock_init(&blkcg->lock);
1485 INIT_HLIST_HEAD(&blkcg->blkg_list);
1487 INIT_LIST_HEAD(&blkcg->policy_list);
1492 * We cannot support shared io contexts, as we have no mean to support
1493 * two tasks with the same ioc in two different groups without major rework
1494 * of the main cic data structures. For now we allow a task to change
1495 * its cgroup only if it's the only owner of its ioc.
1497 static int blkiocg_can_attach(struct cgroup_subsys *subsys,
1498 struct cgroup *cgroup, struct task_struct *tsk,
1501 struct io_context *ioc;
1504 /* task_lock() is needed to avoid races with exit_io_context() */
1506 ioc = tsk->io_context;
1507 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1514 static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup,
1515 struct cgroup *prev, struct task_struct *tsk,
1518 struct io_context *ioc;
1521 ioc = tsk->io_context;
1523 ioc->cgroup_changed = 1;
1527 void blkio_policy_register(struct blkio_policy_type *blkiop)
1529 spin_lock(&blkio_list_lock);
1530 list_add_tail(&blkiop->list, &blkio_list);
1531 spin_unlock(&blkio_list_lock);
1533 EXPORT_SYMBOL_GPL(blkio_policy_register);
1535 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1537 spin_lock(&blkio_list_lock);
1538 list_del_init(&blkiop->list);
1539 spin_unlock(&blkio_list_lock);
1541 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1543 static int __init init_cgroup_blkio(void)
1545 return cgroup_load_subsys(&blkio_subsys);
1548 static void __exit exit_cgroup_blkio(void)
1550 cgroup_unload_subsys(&blkio_subsys);
1553 module_init(init_cgroup_blkio);
1554 module_exit(exit_cgroup_blkio);
1555 MODULE_LICENSE("GPL");