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 "blk-cgroup.h"
21 #define MAX_KEY_LEN 100
23 static DEFINE_SPINLOCK(blkio_list_lock);
24 static LIST_HEAD(blkio_list);
26 struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT };
27 EXPORT_SYMBOL_GPL(blkio_root_cgroup);
29 static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *,
31 static int blkiocg_can_attach(struct cgroup_subsys *, struct cgroup *,
32 struct task_struct *, bool);
33 static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *,
34 struct cgroup *, struct task_struct *, bool);
35 static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *);
36 static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *);
38 struct cgroup_subsys blkio_subsys = {
40 .create = blkiocg_create,
41 .can_attach = blkiocg_can_attach,
42 .attach = blkiocg_attach,
43 .destroy = blkiocg_destroy,
44 .populate = blkiocg_populate,
45 #ifdef CONFIG_BLK_CGROUP
46 /* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */
47 .subsys_id = blkio_subsys_id,
50 .module = THIS_MODULE,
52 EXPORT_SYMBOL_GPL(blkio_subsys);
54 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
56 return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
57 struct blkio_cgroup, css);
59 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
61 void blkio_group_init(struct blkio_group *blkg)
63 spin_lock_init(&blkg->stats_lock);
65 EXPORT_SYMBOL_GPL(blkio_group_init);
68 * Add to the appropriate stat variable depending on the request type.
69 * This should be called with the blkg->stats_lock held.
71 static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
75 stat[BLKIO_STAT_WRITE] += add;
77 stat[BLKIO_STAT_READ] += add;
79 stat[BLKIO_STAT_SYNC] += add;
81 stat[BLKIO_STAT_ASYNC] += add;
85 * Decrements the appropriate stat variable if non-zero depending on the
86 * request type. Panics on value being zero.
87 * This should be called with the blkg->stats_lock held.
89 static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
92 BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
93 stat[BLKIO_STAT_WRITE]--;
95 BUG_ON(stat[BLKIO_STAT_READ] == 0);
96 stat[BLKIO_STAT_READ]--;
99 BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
100 stat[BLKIO_STAT_SYNC]--;
102 BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
103 stat[BLKIO_STAT_ASYNC]--;
107 #ifdef CONFIG_DEBUG_BLK_CGROUP
108 void blkiocg_update_set_active_queue_stats(struct blkio_group *blkg)
111 struct blkio_group_stats *stats;
113 spin_lock_irqsave(&blkg->stats_lock, flags);
114 stats = &blkg->stats;
115 stats->avg_queue_size_sum +=
116 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
117 stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
118 stats->avg_queue_size_samples++;
119 spin_unlock_irqrestore(&blkg->stats_lock, flags);
121 EXPORT_SYMBOL_GPL(blkiocg_update_set_active_queue_stats);
124 void blkiocg_update_request_add_stats(struct blkio_group *blkg,
125 struct blkio_group *curr_blkg, bool direction,
130 spin_lock_irqsave(&blkg->stats_lock, flags);
131 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction,
133 spin_unlock_irqrestore(&blkg->stats_lock, flags);
135 EXPORT_SYMBOL_GPL(blkiocg_update_request_add_stats);
137 void blkiocg_update_request_remove_stats(struct blkio_group *blkg,
138 bool direction, bool sync)
142 spin_lock_irqsave(&blkg->stats_lock, flags);
143 blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED],
145 spin_unlock_irqrestore(&blkg->stats_lock, flags);
147 EXPORT_SYMBOL_GPL(blkiocg_update_request_remove_stats);
149 void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time)
153 spin_lock_irqsave(&blkg->stats_lock, flags);
154 blkg->stats.time += time;
155 spin_unlock_irqrestore(&blkg->stats_lock, flags);
157 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
159 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
160 uint64_t bytes, bool direction, bool sync)
162 struct blkio_group_stats *stats;
165 spin_lock_irqsave(&blkg->stats_lock, flags);
166 stats = &blkg->stats;
167 stats->sectors += bytes >> 9;
168 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICED], 1, direction,
170 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_BYTES], bytes,
172 spin_unlock_irqrestore(&blkg->stats_lock, flags);
174 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
176 void blkiocg_update_completion_stats(struct blkio_group *blkg,
177 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
179 struct blkio_group_stats *stats;
181 unsigned long long now = sched_clock();
183 spin_lock_irqsave(&blkg->stats_lock, flags);
184 stats = &blkg->stats;
185 if (time_after64(now, io_start_time))
186 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
187 now - io_start_time, direction, sync);
188 if (time_after64(io_start_time, start_time))
189 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
190 io_start_time - start_time, direction, sync);
191 spin_unlock_irqrestore(&blkg->stats_lock, flags);
193 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
195 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
200 spin_lock_irqsave(&blkg->stats_lock, flags);
201 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_MERGED], 1, direction,
203 spin_unlock_irqrestore(&blkg->stats_lock, flags);
205 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
207 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
208 struct blkio_group *blkg, void *key, dev_t dev)
212 spin_lock_irqsave(&blkcg->lock, flags);
213 rcu_assign_pointer(blkg->key, key);
214 blkg->blkcg_id = css_id(&blkcg->css);
215 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
216 spin_unlock_irqrestore(&blkcg->lock, flags);
217 #ifdef CONFIG_DEBUG_BLK_CGROUP
218 /* Need to take css reference ? */
219 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
223 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
225 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
227 hlist_del_init_rcu(&blkg->blkcg_node);
232 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
233 * indicating that blk_group was unhashed by the time we got to it.
235 int blkiocg_del_blkio_group(struct blkio_group *blkg)
237 struct blkio_cgroup *blkcg;
239 struct cgroup_subsys_state *css;
243 css = css_lookup(&blkio_subsys, blkg->blkcg_id);
247 blkcg = container_of(css, struct blkio_cgroup, css);
248 spin_lock_irqsave(&blkcg->lock, flags);
249 if (!hlist_unhashed(&blkg->blkcg_node)) {
250 __blkiocg_del_blkio_group(blkg);
253 spin_unlock_irqrestore(&blkcg->lock, flags);
258 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
260 /* called under rcu_read_lock(). */
261 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
263 struct blkio_group *blkg;
264 struct hlist_node *n;
267 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
275 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
277 #define SHOW_FUNCTION(__VAR) \
278 static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \
279 struct cftype *cftype) \
281 struct blkio_cgroup *blkcg; \
283 blkcg = cgroup_to_blkio_cgroup(cgroup); \
284 return (u64)blkcg->__VAR; \
287 SHOW_FUNCTION(weight);
291 blkiocg_weight_write(struct cgroup *cgroup, struct cftype *cftype, u64 val)
293 struct blkio_cgroup *blkcg;
294 struct blkio_group *blkg;
295 struct hlist_node *n;
296 struct blkio_policy_type *blkiop;
298 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
301 blkcg = cgroup_to_blkio_cgroup(cgroup);
302 spin_lock(&blkio_list_lock);
303 spin_lock_irq(&blkcg->lock);
304 blkcg->weight = (unsigned int)val;
305 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
306 list_for_each_entry(blkiop, &blkio_list, list)
307 blkiop->ops.blkio_update_group_weight_fn(blkg,
310 spin_unlock_irq(&blkcg->lock);
311 spin_unlock(&blkio_list_lock);
316 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
318 struct blkio_cgroup *blkcg;
319 struct blkio_group *blkg;
320 struct hlist_node *n;
321 uint64_t queued[BLKIO_STAT_TOTAL];
324 blkcg = cgroup_to_blkio_cgroup(cgroup);
325 spin_lock_irq(&blkcg->lock);
326 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
327 spin_lock(&blkg->stats_lock);
328 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
329 queued[i] = blkg->stats.stat_arr[BLKIO_STAT_QUEUED][i];
330 memset(&blkg->stats, 0, sizeof(struct blkio_group_stats));
331 for (i = 0; i < BLKIO_STAT_TOTAL; i++)
332 blkg->stats.stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
333 spin_unlock(&blkg->stats_lock);
335 spin_unlock_irq(&blkcg->lock);
339 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
340 int chars_left, bool diskname_only)
342 snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
343 chars_left -= strlen(str);
344 if (chars_left <= 0) {
346 "Possibly incorrect cgroup stat display format");
352 case BLKIO_STAT_READ:
353 strlcat(str, " Read", chars_left);
355 case BLKIO_STAT_WRITE:
356 strlcat(str, " Write", chars_left);
358 case BLKIO_STAT_SYNC:
359 strlcat(str, " Sync", chars_left);
361 case BLKIO_STAT_ASYNC:
362 strlcat(str, " Async", chars_left);
364 case BLKIO_STAT_TOTAL:
365 strlcat(str, " Total", chars_left);
368 strlcat(str, " Invalid", chars_left);
372 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
373 struct cgroup_map_cb *cb, dev_t dev)
375 blkio_get_key_name(0, dev, str, chars_left, true);
376 cb->fill(cb, str, val);
380 /* This should be called with blkg->stats_lock held */
381 static uint64_t blkio_get_stat(struct blkio_group *blkg,
382 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
385 char key_str[MAX_KEY_LEN];
386 enum stat_sub_type sub_type;
388 if (type == BLKIO_STAT_TIME)
389 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
390 blkg->stats.time, cb, dev);
391 if (type == BLKIO_STAT_SECTORS)
392 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
393 blkg->stats.sectors, cb, dev);
394 #ifdef CONFIG_DEBUG_BLK_CGROUP
395 if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
396 uint64_t sum = blkg->stats.avg_queue_size_sum;
397 uint64_t samples = blkg->stats.avg_queue_size_samples;
399 do_div(sum, samples);
402 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
404 if (type == BLKIO_STAT_DEQUEUE)
405 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
406 blkg->stats.dequeue, cb, dev);
409 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
411 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
412 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
414 disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
415 blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
416 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
417 cb->fill(cb, key_str, disk_total);
421 #define SHOW_FUNCTION_PER_GROUP(__VAR, type, show_total) \
422 static int blkiocg_##__VAR##_read(struct cgroup *cgroup, \
423 struct cftype *cftype, struct cgroup_map_cb *cb) \
425 struct blkio_cgroup *blkcg; \
426 struct blkio_group *blkg; \
427 struct hlist_node *n; \
428 uint64_t cgroup_total = 0; \
430 if (!cgroup_lock_live_group(cgroup)) \
433 blkcg = cgroup_to_blkio_cgroup(cgroup); \
435 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\
437 spin_lock_irq(&blkg->stats_lock); \
438 cgroup_total += blkio_get_stat(blkg, cb, \
440 spin_unlock_irq(&blkg->stats_lock); \
444 cb->fill(cb, "Total", cgroup_total); \
450 SHOW_FUNCTION_PER_GROUP(time, BLKIO_STAT_TIME, 0);
451 SHOW_FUNCTION_PER_GROUP(sectors, BLKIO_STAT_SECTORS, 0);
452 SHOW_FUNCTION_PER_GROUP(io_service_bytes, BLKIO_STAT_SERVICE_BYTES, 1);
453 SHOW_FUNCTION_PER_GROUP(io_serviced, BLKIO_STAT_SERVICED, 1);
454 SHOW_FUNCTION_PER_GROUP(io_service_time, BLKIO_STAT_SERVICE_TIME, 1);
455 SHOW_FUNCTION_PER_GROUP(io_wait_time, BLKIO_STAT_WAIT_TIME, 1);
456 SHOW_FUNCTION_PER_GROUP(io_merged, BLKIO_STAT_MERGED, 1);
457 SHOW_FUNCTION_PER_GROUP(io_queued, BLKIO_STAT_QUEUED, 1);
458 #ifdef CONFIG_DEBUG_BLK_CGROUP
459 SHOW_FUNCTION_PER_GROUP(dequeue, BLKIO_STAT_DEQUEUE, 0);
460 SHOW_FUNCTION_PER_GROUP(avg_queue_size, BLKIO_STAT_AVG_QUEUE_SIZE, 0);
462 #undef SHOW_FUNCTION_PER_GROUP
464 #ifdef CONFIG_DEBUG_BLK_CGROUP
465 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
466 unsigned long dequeue)
468 blkg->stats.dequeue += dequeue;
470 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
473 struct cftype blkio_files[] = {
476 .read_u64 = blkiocg_weight_read,
477 .write_u64 = blkiocg_weight_write,
481 .read_map = blkiocg_time_read,
485 .read_map = blkiocg_sectors_read,
488 .name = "io_service_bytes",
489 .read_map = blkiocg_io_service_bytes_read,
492 .name = "io_serviced",
493 .read_map = blkiocg_io_serviced_read,
496 .name = "io_service_time",
497 .read_map = blkiocg_io_service_time_read,
500 .name = "io_wait_time",
501 .read_map = blkiocg_io_wait_time_read,
505 .read_map = blkiocg_io_merged_read,
509 .read_map = blkiocg_io_queued_read,
512 .name = "reset_stats",
513 .write_u64 = blkiocg_reset_stats,
515 #ifdef CONFIG_DEBUG_BLK_CGROUP
517 .name = "avg_queue_size",
518 .read_map = blkiocg_avg_queue_size_read,
522 .read_map = blkiocg_dequeue_read,
527 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
529 return cgroup_add_files(cgroup, subsys, blkio_files,
530 ARRAY_SIZE(blkio_files));
533 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
535 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
537 struct blkio_group *blkg;
539 struct blkio_policy_type *blkiop;
543 spin_lock_irqsave(&blkcg->lock, flags);
545 if (hlist_empty(&blkcg->blkg_list)) {
546 spin_unlock_irqrestore(&blkcg->lock, flags);
550 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
552 key = rcu_dereference(blkg->key);
553 __blkiocg_del_blkio_group(blkg);
555 spin_unlock_irqrestore(&blkcg->lock, flags);
558 * This blkio_group is being unlinked as associated cgroup is going
559 * away. Let all the IO controlling policies know about this event.
561 * Currently this is static call to one io controlling policy. Once
562 * we have more policies in place, we need some dynamic registration
563 * of callback function.
565 spin_lock(&blkio_list_lock);
566 list_for_each_entry(blkiop, &blkio_list, list)
567 blkiop->ops.blkio_unlink_group_fn(key, blkg);
568 spin_unlock(&blkio_list_lock);
571 free_css_id(&blkio_subsys, &blkcg->css);
573 if (blkcg != &blkio_root_cgroup)
577 static struct cgroup_subsys_state *
578 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
580 struct blkio_cgroup *blkcg, *parent_blkcg;
582 if (!cgroup->parent) {
583 blkcg = &blkio_root_cgroup;
587 /* Currently we do not support hierarchy deeper than two level (0,1) */
588 parent_blkcg = cgroup_to_blkio_cgroup(cgroup->parent);
589 if (css_depth(&parent_blkcg->css) > 0)
590 return ERR_PTR(-EINVAL);
592 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
594 return ERR_PTR(-ENOMEM);
596 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
598 spin_lock_init(&blkcg->lock);
599 INIT_HLIST_HEAD(&blkcg->blkg_list);
605 * We cannot support shared io contexts, as we have no mean to support
606 * two tasks with the same ioc in two different groups without major rework
607 * of the main cic data structures. For now we allow a task to change
608 * its cgroup only if it's the only owner of its ioc.
610 static int blkiocg_can_attach(struct cgroup_subsys *subsys,
611 struct cgroup *cgroup, struct task_struct *tsk,
614 struct io_context *ioc;
617 /* task_lock() is needed to avoid races with exit_io_context() */
619 ioc = tsk->io_context;
620 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
627 static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup,
628 struct cgroup *prev, struct task_struct *tsk,
631 struct io_context *ioc;
634 ioc = tsk->io_context;
636 ioc->cgroup_changed = 1;
640 void blkio_policy_register(struct blkio_policy_type *blkiop)
642 spin_lock(&blkio_list_lock);
643 list_add_tail(&blkiop->list, &blkio_list);
644 spin_unlock(&blkio_list_lock);
646 EXPORT_SYMBOL_GPL(blkio_policy_register);
648 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
650 spin_lock(&blkio_list_lock);
651 list_del_init(&blkiop->list);
652 spin_unlock(&blkio_list_lock);
654 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
656 static int __init init_cgroup_blkio(void)
658 return cgroup_load_subsys(&blkio_subsys);
661 static void __exit exit_cgroup_blkio(void)
663 cgroup_unload_subsys(&blkio_subsys);
666 module_init(init_cgroup_blkio);
667 module_exit(exit_cgroup_blkio);
668 MODULE_LICENSE("GPL");