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 struct cgroup_subsys blkio_subsys = {
42 .create = blkiocg_create,
43 .can_attach = blkiocg_can_attach,
44 .attach = blkiocg_attach,
45 .destroy = blkiocg_destroy,
46 .populate = blkiocg_populate,
47 #ifdef CONFIG_BLK_CGROUP
48 /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */
49 .subsys_id = blkio_subsys_id,
52 .module = THIS_MODULE,
54 EXPORT_SYMBOL_GPL(blkio_subsys);
56 static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg,
57 struct blkio_policy_node *pn)
59 list_add(&pn->node, &blkcg->policy_list);
62 /* Must be called with blkcg->lock held */
63 static inline void blkio_policy_delete_node(struct blkio_policy_node *pn)
68 /* Must be called with blkcg->lock held */
69 static struct blkio_policy_node *
70 blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev)
72 struct blkio_policy_node *pn;
74 list_for_each_entry(pn, &blkcg->policy_list, node) {
82 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
84 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
85 struct blkio_cgroup, css);
87 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
90 * Add to the appropriate stat variable depending on the request type.
91 * This should be called with the blkg->stats_lock held.
93 static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
97 stat[BLKIO_STAT_WRITE] += add;
99 stat[BLKIO_STAT_READ] += add;
101 stat[BLKIO_STAT_SYNC] += add;
103 stat[BLKIO_STAT_ASYNC] += add;
107 * Decrements the appropriate stat variable if non-zero depending on the
108 * request type. Panics on value being zero.
109 * This should be called with the blkg->stats_lock held.
111 static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
114 BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
115 stat[BLKIO_STAT_WRITE]--;
117 BUG_ON(stat[BLKIO_STAT_READ] == 0);
118 stat[BLKIO_STAT_READ]--;
121 BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
122 stat[BLKIO_STAT_SYNC]--;
124 BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
125 stat[BLKIO_STAT_ASYNC]--;
129 #ifdef CONFIG_DEBUG_BLK_CGROUP
130 /* This should be called with the blkg->stats_lock held. */
131 static void blkio_set_start_group_wait_time(struct blkio_group *blkg,
132 struct blkio_group *curr_blkg)
134 if (blkio_blkg_waiting(&blkg->stats))
136 if (blkg == curr_blkg)
138 blkg->stats.start_group_wait_time = sched_clock();
139 blkio_mark_blkg_waiting(&blkg->stats);
142 /* This should be called with the blkg->stats_lock held. */
143 static void blkio_update_group_wait_time(struct blkio_group_stats *stats)
145 unsigned long long now;
147 if (!blkio_blkg_waiting(stats))
151 if (time_after64(now, stats->start_group_wait_time))
152 stats->group_wait_time += now - stats->start_group_wait_time;
153 blkio_clear_blkg_waiting(stats);
156 /* This should be called with the blkg->stats_lock held. */
157 static void blkio_end_empty_time(struct blkio_group_stats *stats)
159 unsigned long long now;
161 if (!blkio_blkg_empty(stats))
165 if (time_after64(now, stats->start_empty_time))
166 stats->empty_time += now - stats->start_empty_time;
167 blkio_clear_blkg_empty(stats);
170 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
174 spin_lock_irqsave(&blkg->stats_lock, flags);
175 BUG_ON(blkio_blkg_idling(&blkg->stats));
176 blkg->stats.start_idle_time = sched_clock();
177 blkio_mark_blkg_idling(&blkg->stats);
178 spin_unlock_irqrestore(&blkg->stats_lock, flags);
180 EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats);
182 void blkiocg_update_idle_time_stats(struct blkio_group *blkg)
185 unsigned long long now;
186 struct blkio_group_stats *stats;
188 spin_lock_irqsave(&blkg->stats_lock, flags);
189 stats = &blkg->stats;
190 if (blkio_blkg_idling(stats)) {
192 if (time_after64(now, stats->start_idle_time))
193 stats->idle_time += now - stats->start_idle_time;
194 blkio_clear_blkg_idling(stats);
196 spin_unlock_irqrestore(&blkg->stats_lock, flags);
198 EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats);
200 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg)
203 struct blkio_group_stats *stats;
205 spin_lock_irqsave(&blkg->stats_lock, flags);
206 stats = &blkg->stats;
207 stats->avg_queue_size_sum +=
208 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
209 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
210 stats->avg_queue_size_samples++;
211 blkio_update_group_wait_time(stats);
212 spin_unlock_irqrestore(&blkg->stats_lock, flags);
214 EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats);
216 void blkiocg_set_start_empty_time(struct blkio_group *blkg)
219 struct blkio_group_stats *stats;
221 spin_lock_irqsave(&blkg->stats_lock, flags);
222 stats = &blkg->stats;
224 if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] ||
225 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) {
226 spin_unlock_irqrestore(&blkg->stats_lock, flags);
231 * group is already marked empty. This can happen if cfqq got new
232 * request in parent group and moved to this group while being added
233 * to service tree. Just ignore the event and move on.
235 if(blkio_blkg_empty(stats)) {
236 spin_unlock_irqrestore(&blkg->stats_lock, flags);
240 stats->start_empty_time = sched_clock();
241 blkio_mark_blkg_empty(stats);
242 spin_unlock_irqrestore(&blkg->stats_lock, flags);
244 EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time);
246 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
247 unsigned long dequeue)
249 blkg->stats.dequeue += dequeue;
251 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
253 static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg,
254 struct blkio_group *curr_blkg) {}
255 static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {}
258 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
259 struct blkio_group *curr_blkg, bool direction,
264 spin_lock_irqsave(&blkg->stats_lock, flags);
265 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction,
267 blkio_end_empty_time(&blkg->stats);
268 blkio_set_start_group_wait_time(blkg, curr_blkg);
269 spin_unlock_irqrestore(&blkg->stats_lock, flags);
271 EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats);
273 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
274 bool direction, bool sync)
278 spin_lock_irqsave(&blkg->stats_lock, flags);
279 blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED],
281 spin_unlock_irqrestore(&blkg->stats_lock, flags);
283 EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats);
285 void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time)
289 spin_lock_irqsave(&blkg->stats_lock, flags);
290 blkg->stats.time += time;
291 spin_unlock_irqrestore(&blkg->stats_lock, flags);
293 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
295 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
296 uint64_t bytes, bool direction, bool sync)
298 struct blkio_group_stats *stats;
301 spin_lock_irqsave(&blkg->stats_lock, flags);
302 stats = &blkg->stats;
303 stats->sectors += bytes >> 9;
304 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICED], 1, direction,
306 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_BYTES], bytes,
308 spin_unlock_irqrestore(&blkg->stats_lock, flags);
310 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
312 void blkiocg_update_completion_stats(struct blkio_group *blkg,
313 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
315 struct blkio_group_stats *stats;
317 unsigned long long now = sched_clock();
319 spin_lock_irqsave(&blkg->stats_lock, flags);
320 stats = &blkg->stats;
321 if (time_after64(now, io_start_time))
322 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
323 now - io_start_time, direction, sync);
324 if (time_after64(io_start_time, start_time))
325 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
326 io_start_time - start_time, direction, sync);
327 spin_unlock_irqrestore(&blkg->stats_lock, flags);
329 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
331 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
336 spin_lock_irqsave(&blkg->stats_lock, flags);
337 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_MERGED], 1, direction,
339 spin_unlock_irqrestore(&blkg->stats_lock, flags);
341 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
343 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
344 struct blkio_group *blkg, void *key, dev_t dev)
348 spin_lock_irqsave(&blkcg->lock, flags);
349 spin_lock_init(&blkg->stats_lock);
350 rcu_assign_pointer(blkg->key, key);
351 blkg->blkcg_id = css_id(&blkcg->css);
352 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
353 spin_unlock_irqrestore(&blkcg->lock, flags);
354 #ifdef CONFIG_DEBUG_BLK_CGROUP
355 /* Need to take css reference ? */
356 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
360 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
362 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
364 hlist_del_init_rcu(&blkg->blkcg_node);
369 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
370 * indicating that blk_group was unhashed by the time we got to it.
372 int blkiocg_del_blkio_group(struct blkio_group *blkg)
374 struct blkio_cgroup *blkcg;
376 struct cgroup_subsys_state *css;
380 css = css_lookup(&blkio_subsys, blkg->blkcg_id);
384 blkcg = container_of(css, struct blkio_cgroup, css);
385 spin_lock_irqsave(&blkcg->lock, flags);
386 if (!hlist_unhashed(&blkg->blkcg_node)) {
387 __blkiocg_del_blkio_group(blkg);
390 spin_unlock_irqrestore(&blkcg->lock, flags);
395 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
397 /* called under rcu_read_lock(). */
398 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
400 struct blkio_group *blkg;
401 struct hlist_node *n;
404 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
412 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
414 #define SHOW_FUNCTION(__VAR) \
415 static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \
416 struct cftype *cftype) \
418 struct blkio_cgroup *blkcg; \
420 blkcg = cgroup_to_blkio_cgroup(cgroup); \
421 return (u64)blkcg->__VAR; \
424 SHOW_FUNCTION(weight);
428 blkiocg_weight_write(struct cgroup *cgroup, struct cftype *cftype, u64 val)
430 struct blkio_cgroup *blkcg;
431 struct blkio_group *blkg;
432 struct hlist_node *n;
433 struct blkio_policy_type *blkiop;
434 struct blkio_policy_node *pn;
436 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
439 blkcg = cgroup_to_blkio_cgroup(cgroup);
440 spin_lock(&blkio_list_lock);
441 spin_lock_irq(&blkcg->lock);
442 blkcg->weight = (unsigned int)val;
444 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
445 pn = blkio_policy_search_node(blkcg, blkg->dev);
450 list_for_each_entry(blkiop, &blkio_list, list)
451 blkiop->ops.blkio_update_group_weight_fn(blkg,
454 spin_unlock_irq(&blkcg->lock);
455 spin_unlock(&blkio_list_lock);
460 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
462 struct blkio_cgroup *blkcg;
463 struct blkio_group *blkg;
464 struct blkio_group_stats *stats;
465 struct hlist_node *n;
466 uint64_t queued[BLKIO_STAT_TOTAL];
468 #ifdef CONFIG_DEBUG_BLK_CGROUP
469 bool idling, waiting, empty;
470 unsigned long long now = sched_clock();
473 blkcg = cgroup_to_blkio_cgroup(cgroup);
474 spin_lock_irq(&blkcg->lock);
475 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
476 spin_lock(&blkg->stats_lock);
477 stats = &blkg->stats;
478 #ifdef CONFIG_DEBUG_BLK_CGROUP
479 idling = blkio_blkg_idling(stats);
480 waiting = blkio_blkg_waiting(stats);
481 empty = blkio_blkg_empty(stats);
483 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
484 queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
485 memset(stats, 0, sizeof(struct blkio_group_stats));
486 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
487 stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
488 #ifdef CONFIG_DEBUG_BLK_CGROUP
490 blkio_mark_blkg_idling(stats);
491 stats->start_idle_time = now;
494 blkio_mark_blkg_waiting(stats);
495 stats->start_group_wait_time = now;
498 blkio_mark_blkg_empty(stats);
499 stats->start_empty_time = now;
502 spin_unlock(&blkg->stats_lock);
504 spin_unlock_irq(&blkcg->lock);
508 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
509 int chars_left, bool diskname_only)
511 snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
512 chars_left -= strlen(str);
513 if (chars_left <= 0) {
515 "Possibly incorrect cgroup stat display format");
521 case BLKIO_STAT_READ:
522 strlcat(str, " Read", chars_left);
524 case BLKIO_STAT_WRITE:
525 strlcat(str, " Write", chars_left);
527 case BLKIO_STAT_SYNC:
528 strlcat(str, " Sync", chars_left);
530 case BLKIO_STAT_ASYNC:
531 strlcat(str, " Async", chars_left);
533 case BLKIO_STAT_TOTAL:
534 strlcat(str, " Total", chars_left);
537 strlcat(str, " Invalid", chars_left);
541 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
542 struct cgroup_map_cb *cb, dev_t dev)
544 blkio_get_key_name(0, dev, str, chars_left, true);
545 cb->fill(cb, str, val);
549 /* This should be called with blkg->stats_lock held */
550 static uint64_t blkio_get_stat(struct blkio_group *blkg,
551 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
554 char key_str[MAX_KEY_LEN];
555 enum stat_sub_type sub_type;
557 if (type == BLKIO_STAT_TIME)
558 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
559 blkg->stats.time, cb, dev);
560 if (type == BLKIO_STAT_SECTORS)
561 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
562 blkg->stats.sectors, cb, dev);
563 #ifdef CONFIG_DEBUG_BLK_CGROUP
564 if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
565 uint64_t sum = blkg->stats.avg_queue_size_sum;
566 uint64_t samples = blkg->stats.avg_queue_size_samples;
568 do_div(sum, samples);
571 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
573 if (type == BLKIO_STAT_GROUP_WAIT_TIME)
574 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
575 blkg->stats.group_wait_time, cb, dev);
576 if (type == BLKIO_STAT_IDLE_TIME)
577 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
578 blkg->stats.idle_time, cb, dev);
579 if (type == BLKIO_STAT_EMPTY_TIME)
580 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
581 blkg->stats.empty_time, cb, dev);
582 if (type == BLKIO_STAT_DEQUEUE)
583 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
584 blkg->stats.dequeue, cb, dev);
587 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
589 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
590 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
592 disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
593 blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
594 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
595 cb->fill(cb, key_str, disk_total);
599 #define SHOW_FUNCTION_PER_GROUP(__VAR, type, show_total) \
600 static int blkiocg_##__VAR##_read(struct cgroup *cgroup, \
601 struct cftype *cftype, struct cgroup_map_cb *cb) \
603 struct blkio_cgroup *blkcg; \
604 struct blkio_group *blkg; \
605 struct hlist_node *n; \
606 uint64_t cgroup_total = 0; \
608 if (!cgroup_lock_live_group(cgroup)) \
611 blkcg = cgroup_to_blkio_cgroup(cgroup); \
613 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\
615 spin_lock_irq(&blkg->stats_lock); \
616 cgroup_total += blkio_get_stat(blkg, cb, \
618 spin_unlock_irq(&blkg->stats_lock); \
622 cb->fill(cb, "Total", cgroup_total); \
628 SHOW_FUNCTION_PER_GROUP(time, BLKIO_STAT_TIME, 0);
629 SHOW_FUNCTION_PER_GROUP(sectors, BLKIO_STAT_SECTORS, 0);
630 SHOW_FUNCTION_PER_GROUP(io_service_bytes, BLKIO_STAT_SERVICE_BYTES, 1);
631 SHOW_FUNCTION_PER_GROUP(io_serviced, BLKIO_STAT_SERVICED, 1);
632 SHOW_FUNCTION_PER_GROUP(io_service_time, BLKIO_STAT_SERVICE_TIME, 1);
633 SHOW_FUNCTION_PER_GROUP(io_wait_time, BLKIO_STAT_WAIT_TIME, 1);
634 SHOW_FUNCTION_PER_GROUP(io_merged, BLKIO_STAT_MERGED, 1);
635 SHOW_FUNCTION_PER_GROUP(io_queued, BLKIO_STAT_QUEUED, 1);
636 #ifdef CONFIG_DEBUG_BLK_CGROUP
637 SHOW_FUNCTION_PER_GROUP(dequeue, BLKIO_STAT_DEQUEUE, 0);
638 SHOW_FUNCTION_PER_GROUP(avg_queue_size, BLKIO_STAT_AVG_QUEUE_SIZE, 0);
639 SHOW_FUNCTION_PER_GROUP(group_wait_time, BLKIO_STAT_GROUP_WAIT_TIME, 0);
640 SHOW_FUNCTION_PER_GROUP(idle_time, BLKIO_STAT_IDLE_TIME, 0);
641 SHOW_FUNCTION_PER_GROUP(empty_time, BLKIO_STAT_EMPTY_TIME, 0);
643 #undef SHOW_FUNCTION_PER_GROUP
645 static int blkio_check_dev_num(dev_t dev)
648 struct gendisk *disk;
650 disk = get_gendisk(dev, &part);
657 static int blkio_policy_parse_and_set(char *buf,
658 struct blkio_policy_node *newpn)
660 char *s[4], *p, *major_s = NULL, *minor_s = NULL;
662 unsigned long major, minor, temp;
666 memset(s, 0, sizeof(s));
668 while ((p = strsep(&buf, " ")) != NULL) {
674 /* Prevent from inputing too many things */
682 p = strsep(&s[0], ":");
692 ret = strict_strtoul(major_s, 10, &major);
696 ret = strict_strtoul(minor_s, 10, &minor);
700 dev = MKDEV(major, minor);
702 ret = blkio_check_dev_num(dev);
711 ret = strict_strtoul(s[1], 10, &temp);
712 if (ret || (temp < BLKIO_WEIGHT_MIN && temp > 0) ||
713 temp > BLKIO_WEIGHT_MAX)
716 newpn->weight = temp;
721 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
724 struct blkio_policy_node *pn;
726 pn = blkio_policy_search_node(blkcg, dev);
730 return blkcg->weight;
732 EXPORT_SYMBOL_GPL(blkcg_get_weight);
735 static int blkiocg_weight_device_write(struct cgroup *cgrp, struct cftype *cft,
740 struct blkio_policy_node *newpn, *pn;
741 struct blkio_cgroup *blkcg;
742 struct blkio_group *blkg;
744 struct hlist_node *n;
745 struct blkio_policy_type *blkiop;
747 buf = kstrdup(buffer, GFP_KERNEL);
751 newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
757 ret = blkio_policy_parse_and_set(buf, newpn);
761 blkcg = cgroup_to_blkio_cgroup(cgrp);
763 spin_lock_irq(&blkcg->lock);
765 pn = blkio_policy_search_node(blkcg, newpn->dev);
767 if (newpn->weight != 0) {
768 blkio_policy_insert_node(blkcg, newpn);
771 spin_unlock_irq(&blkcg->lock);
772 goto update_io_group;
775 if (newpn->weight == 0) {
776 /* weight == 0 means deleteing a specific weight */
777 blkio_policy_delete_node(pn);
778 spin_unlock_irq(&blkcg->lock);
779 goto update_io_group;
781 spin_unlock_irq(&blkcg->lock);
783 pn->weight = newpn->weight;
786 /* update weight for each cfqg */
787 spin_lock(&blkio_list_lock);
788 spin_lock_irq(&blkcg->lock);
790 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
791 if (newpn->dev == blkg->dev) {
792 list_for_each_entry(blkiop, &blkio_list, list)
793 blkiop->ops.blkio_update_group_weight_fn(blkg,
800 spin_unlock_irq(&blkcg->lock);
801 spin_unlock(&blkio_list_lock);
811 static int blkiocg_weight_device_read(struct cgroup *cgrp, struct cftype *cft,
814 struct blkio_cgroup *blkcg;
815 struct blkio_policy_node *pn;
817 seq_printf(m, "dev\tweight\n");
819 blkcg = cgroup_to_blkio_cgroup(cgrp);
820 if (list_empty(&blkcg->policy_list))
823 spin_lock_irq(&blkcg->lock);
824 list_for_each_entry(pn, &blkcg->policy_list, node) {
825 seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
826 MINOR(pn->dev), pn->weight);
828 spin_unlock_irq(&blkcg->lock);
834 struct cftype blkio_files[] = {
836 .name = "weight_device",
837 .read_seq_string = blkiocg_weight_device_read,
838 .write_string = blkiocg_weight_device_write,
839 .max_write_len = 256,
843 .read_u64 = blkiocg_weight_read,
844 .write_u64 = blkiocg_weight_write,
848 .read_map = blkiocg_time_read,
852 .read_map = blkiocg_sectors_read,
855 .name = "io_service_bytes",
856 .read_map = blkiocg_io_service_bytes_read,
859 .name = "io_serviced",
860 .read_map = blkiocg_io_serviced_read,
863 .name = "io_service_time",
864 .read_map = blkiocg_io_service_time_read,
867 .name = "io_wait_time",
868 .read_map = blkiocg_io_wait_time_read,
872 .read_map = blkiocg_io_merged_read,
876 .read_map = blkiocg_io_queued_read,
879 .name = "reset_stats",
880 .write_u64 = blkiocg_reset_stats,
882 #ifdef CONFIG_DEBUG_BLK_CGROUP
884 .name = "avg_queue_size",
885 .read_map = blkiocg_avg_queue_size_read,
888 .name = "group_wait_time",
889 .read_map = blkiocg_group_wait_time_read,
893 .read_map = blkiocg_idle_time_read,
896 .name = "empty_time",
897 .read_map = blkiocg_empty_time_read,
901 .read_map = blkiocg_dequeue_read,
906 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
908 return cgroup_add_files(cgroup, subsys, blkio_files,
909 ARRAY_SIZE(blkio_files));
912 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
914 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
916 struct blkio_group *blkg;
918 struct blkio_policy_type *blkiop;
919 struct blkio_policy_node *pn, *pntmp;
923 spin_lock_irqsave(&blkcg->lock, flags);
925 if (hlist_empty(&blkcg->blkg_list)) {
926 spin_unlock_irqrestore(&blkcg->lock, flags);
930 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
932 key = rcu_dereference(blkg->key);
933 __blkiocg_del_blkio_group(blkg);
935 spin_unlock_irqrestore(&blkcg->lock, flags);
938 * This blkio_group is being unlinked as associated cgroup is going
939 * away. Let all the IO controlling policies know about this event.
941 * Currently this is static call to one io controlling policy. Once
942 * we have more policies in place, we need some dynamic registration
943 * of callback function.
945 spin_lock(&blkio_list_lock);
946 list_for_each_entry(blkiop, &blkio_list, list)
947 blkiop->ops.blkio_unlink_group_fn(key, blkg);
948 spin_unlock(&blkio_list_lock);
952 list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
953 blkio_policy_delete_node(pn);
956 free_css_id(&blkio_subsys, &blkcg->css);
958 if (blkcg != &blkio_root_cgroup)
962 static struct cgroup_subsys_state *
963 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
965 struct blkio_cgroup *blkcg, *parent_blkcg;
967 if (!cgroup->parent) {
968 blkcg = &blkio_root_cgroup;
972 /* Currently we do not support hierarchy deeper than two level (0,1) */
973 parent_blkcg = cgroup_to_blkio_cgroup(cgroup->parent);
974 if (css_depth(&parent_blkcg->css) > 0)
975 return ERR_PTR(-EINVAL);
977 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
979 return ERR_PTR(-ENOMEM);
981 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
983 spin_lock_init(&blkcg->lock);
984 INIT_HLIST_HEAD(&blkcg->blkg_list);
986 INIT_LIST_HEAD(&blkcg->policy_list);
991 * We cannot support shared io contexts, as we have no mean to support
992 * two tasks with the same ioc in two different groups without major rework
993 * of the main cic data structures. For now we allow a task to change
994 * its cgroup only if it's the only owner of its ioc.
996 static int blkiocg_can_attach(struct cgroup_subsys *subsys,
997 struct cgroup *cgroup, struct task_struct *tsk,
1000 struct io_context *ioc;
1003 /* task_lock() is needed to avoid races with exit_io_context() */
1005 ioc = tsk->io_context;
1006 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1013 static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup,
1014 struct cgroup *prev, struct task_struct *tsk,
1017 struct io_context *ioc;
1020 ioc = tsk->io_context;
1022 ioc->cgroup_changed = 1;
1026 void blkio_policy_register(struct blkio_policy_type *blkiop)
1028 spin_lock(&blkio_list_lock);
1029 list_add_tail(&blkiop->list, &blkio_list);
1030 spin_unlock(&blkio_list_lock);
1032 EXPORT_SYMBOL_GPL(blkio_policy_register);
1034 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1036 spin_lock(&blkio_list_lock);
1037 list_del_init(&blkiop->list);
1038 spin_unlock(&blkio_list_lock);
1040 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1042 static int __init init_cgroup_blkio(void)
1044 return cgroup_load_subsys(&blkio_subsys);
1047 static void __exit exit_cgroup_blkio(void)
1049 cgroup_unload_subsys(&blkio_subsys);
1052 module_init(init_cgroup_blkio);
1053 module_exit(exit_cgroup_blkio);
1054 MODULE_LICENSE("GPL");