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;
84 void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time)
88 spin_lock_irqsave(&blkg->stats_lock, flags);
89 blkg->stats.time += time;
90 spin_unlock_irqrestore(&blkg->stats_lock, flags);
92 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
94 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
95 uint64_t bytes, bool direction, bool sync)
97 struct blkio_group_stats *stats;
100 spin_lock_irqsave(&blkg->stats_lock, flags);
101 stats = &blkg->stats;
102 stats->sectors += bytes >> 9;
103 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICED], 1, direction,
105 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_BYTES], bytes,
107 spin_unlock_irqrestore(&blkg->stats_lock, flags);
109 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
111 void blkiocg_update_completion_stats(struct blkio_group *blkg,
112 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
114 struct blkio_group_stats *stats;
116 unsigned long long now = sched_clock();
118 spin_lock_irqsave(&blkg->stats_lock, flags);
119 stats = &blkg->stats;
120 if (time_after64(now, io_start_time))
121 blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
122 now - io_start_time, direction, sync);
123 if (time_after64(io_start_time, start_time))
124 blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
125 io_start_time - start_time, direction, sync);
126 spin_unlock_irqrestore(&blkg->stats_lock, flags);
128 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
130 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
135 spin_lock_irqsave(&blkg->stats_lock, flags);
136 blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_MERGED], 1, direction,
138 spin_unlock_irqrestore(&blkg->stats_lock, flags);
140 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
142 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
143 struct blkio_group *blkg, void *key, dev_t dev)
147 spin_lock_irqsave(&blkcg->lock, flags);
148 rcu_assign_pointer(blkg->key, key);
149 blkg->blkcg_id = css_id(&blkcg->css);
150 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
151 spin_unlock_irqrestore(&blkcg->lock, flags);
152 #ifdef CONFIG_DEBUG_BLK_CGROUP
153 /* Need to take css reference ? */
154 cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
158 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
160 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
162 hlist_del_init_rcu(&blkg->blkcg_node);
167 * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
168 * indicating that blk_group was unhashed by the time we got to it.
170 int blkiocg_del_blkio_group(struct blkio_group *blkg)
172 struct blkio_cgroup *blkcg;
174 struct cgroup_subsys_state *css;
178 css = css_lookup(&blkio_subsys, blkg->blkcg_id);
182 blkcg = container_of(css, struct blkio_cgroup, css);
183 spin_lock_irqsave(&blkcg->lock, flags);
184 if (!hlist_unhashed(&blkg->blkcg_node)) {
185 __blkiocg_del_blkio_group(blkg);
188 spin_unlock_irqrestore(&blkcg->lock, flags);
193 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
195 /* called under rcu_read_lock(). */
196 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
198 struct blkio_group *blkg;
199 struct hlist_node *n;
202 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
210 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
212 #define SHOW_FUNCTION(__VAR) \
213 static u64 blkiocg_##__VAR##_read(struct cgroup *cgroup, \
214 struct cftype *cftype) \
216 struct blkio_cgroup *blkcg; \
218 blkcg = cgroup_to_blkio_cgroup(cgroup); \
219 return (u64)blkcg->__VAR; \
222 SHOW_FUNCTION(weight);
226 blkiocg_weight_write(struct cgroup *cgroup, struct cftype *cftype, u64 val)
228 struct blkio_cgroup *blkcg;
229 struct blkio_group *blkg;
230 struct hlist_node *n;
231 struct blkio_policy_type *blkiop;
233 if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
236 blkcg = cgroup_to_blkio_cgroup(cgroup);
237 spin_lock(&blkio_list_lock);
238 spin_lock_irq(&blkcg->lock);
239 blkcg->weight = (unsigned int)val;
240 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
241 list_for_each_entry(blkiop, &blkio_list, list)
242 blkiop->ops.blkio_update_group_weight_fn(blkg,
245 spin_unlock_irq(&blkcg->lock);
246 spin_unlock(&blkio_list_lock);
251 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
253 struct blkio_cgroup *blkcg;
254 struct blkio_group *blkg;
255 struct hlist_node *n;
256 struct blkio_group_stats *stats;
258 blkcg = cgroup_to_blkio_cgroup(cgroup);
259 spin_lock_irq(&blkcg->lock);
260 hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
261 spin_lock(&blkg->stats_lock);
262 stats = &blkg->stats;
263 memset(stats, 0, sizeof(struct blkio_group_stats));
264 spin_unlock(&blkg->stats_lock);
266 spin_unlock_irq(&blkcg->lock);
270 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
271 int chars_left, bool diskname_only)
273 snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
274 chars_left -= strlen(str);
275 if (chars_left <= 0) {
277 "Possibly incorrect cgroup stat display format");
283 case BLKIO_STAT_READ:
284 strlcat(str, " Read", chars_left);
286 case BLKIO_STAT_WRITE:
287 strlcat(str, " Write", chars_left);
289 case BLKIO_STAT_SYNC:
290 strlcat(str, " Sync", chars_left);
292 case BLKIO_STAT_ASYNC:
293 strlcat(str, " Async", chars_left);
295 case BLKIO_STAT_TOTAL:
296 strlcat(str, " Total", chars_left);
299 strlcat(str, " Invalid", chars_left);
303 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
304 struct cgroup_map_cb *cb, dev_t dev)
306 blkio_get_key_name(0, dev, str, chars_left, true);
307 cb->fill(cb, str, val);
311 /* This should be called with blkg->stats_lock held */
312 static uint64_t blkio_get_stat(struct blkio_group *blkg,
313 struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
316 char key_str[MAX_KEY_LEN];
317 enum stat_sub_type sub_type;
319 if (type == BLKIO_STAT_TIME)
320 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
321 blkg->stats.time, cb, dev);
322 if (type == BLKIO_STAT_SECTORS)
323 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
324 blkg->stats.sectors, cb, dev);
325 #ifdef CONFIG_DEBUG_BLK_CGROUP
326 if (type == BLKIO_STAT_DEQUEUE)
327 return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
328 blkg->stats.dequeue, cb, dev);
331 for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
333 blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
334 cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
336 disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
337 blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
338 blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
339 cb->fill(cb, key_str, disk_total);
343 #define SHOW_FUNCTION_PER_GROUP(__VAR, type, show_total) \
344 static int blkiocg_##__VAR##_read(struct cgroup *cgroup, \
345 struct cftype *cftype, struct cgroup_map_cb *cb) \
347 struct blkio_cgroup *blkcg; \
348 struct blkio_group *blkg; \
349 struct hlist_node *n; \
350 uint64_t cgroup_total = 0; \
352 if (!cgroup_lock_live_group(cgroup)) \
355 blkcg = cgroup_to_blkio_cgroup(cgroup); \
357 hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {\
359 spin_lock_irq(&blkg->stats_lock); \
360 cgroup_total += blkio_get_stat(blkg, cb, \
362 spin_unlock_irq(&blkg->stats_lock); \
366 cb->fill(cb, "Total", cgroup_total); \
372 SHOW_FUNCTION_PER_GROUP(time, BLKIO_STAT_TIME, 0);
373 SHOW_FUNCTION_PER_GROUP(sectors, BLKIO_STAT_SECTORS, 0);
374 SHOW_FUNCTION_PER_GROUP(io_service_bytes, BLKIO_STAT_SERVICE_BYTES, 1);
375 SHOW_FUNCTION_PER_GROUP(io_serviced, BLKIO_STAT_SERVICED, 1);
376 SHOW_FUNCTION_PER_GROUP(io_service_time, BLKIO_STAT_SERVICE_TIME, 1);
377 SHOW_FUNCTION_PER_GROUP(io_wait_time, BLKIO_STAT_WAIT_TIME, 1);
378 SHOW_FUNCTION_PER_GROUP(io_merged, BLKIO_STAT_MERGED, 1);
379 #ifdef CONFIG_DEBUG_BLK_CGROUP
380 SHOW_FUNCTION_PER_GROUP(dequeue, BLKIO_STAT_DEQUEUE, 0);
382 #undef SHOW_FUNCTION_PER_GROUP
384 #ifdef CONFIG_DEBUG_BLK_CGROUP
385 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
386 unsigned long dequeue)
388 blkg->stats.dequeue += dequeue;
390 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
393 struct cftype blkio_files[] = {
396 .read_u64 = blkiocg_weight_read,
397 .write_u64 = blkiocg_weight_write,
401 .read_map = blkiocg_time_read,
405 .read_map = blkiocg_sectors_read,
408 .name = "io_service_bytes",
409 .read_map = blkiocg_io_service_bytes_read,
412 .name = "io_serviced",
413 .read_map = blkiocg_io_serviced_read,
416 .name = "io_service_time",
417 .read_map = blkiocg_io_service_time_read,
420 .name = "io_wait_time",
421 .read_map = blkiocg_io_wait_time_read,
425 .read_map = blkiocg_io_merged_read,
428 .name = "reset_stats",
429 .write_u64 = blkiocg_reset_stats,
431 #ifdef CONFIG_DEBUG_BLK_CGROUP
434 .read_map = blkiocg_dequeue_read,
439 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
441 return cgroup_add_files(cgroup, subsys, blkio_files,
442 ARRAY_SIZE(blkio_files));
445 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
447 struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
449 struct blkio_group *blkg;
451 struct blkio_policy_type *blkiop;
455 spin_lock_irqsave(&blkcg->lock, flags);
457 if (hlist_empty(&blkcg->blkg_list)) {
458 spin_unlock_irqrestore(&blkcg->lock, flags);
462 blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
464 key = rcu_dereference(blkg->key);
465 __blkiocg_del_blkio_group(blkg);
467 spin_unlock_irqrestore(&blkcg->lock, flags);
470 * This blkio_group is being unlinked as associated cgroup is going
471 * away. Let all the IO controlling policies know about this event.
473 * Currently this is static call to one io controlling policy. Once
474 * we have more policies in place, we need some dynamic registration
475 * of callback function.
477 spin_lock(&blkio_list_lock);
478 list_for_each_entry(blkiop, &blkio_list, list)
479 blkiop->ops.blkio_unlink_group_fn(key, blkg);
480 spin_unlock(&blkio_list_lock);
483 free_css_id(&blkio_subsys, &blkcg->css);
485 if (blkcg != &blkio_root_cgroup)
489 static struct cgroup_subsys_state *
490 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
492 struct blkio_cgroup *blkcg, *parent_blkcg;
494 if (!cgroup->parent) {
495 blkcg = &blkio_root_cgroup;
499 /* Currently we do not support hierarchy deeper than two level (0,1) */
500 parent_blkcg = cgroup_to_blkio_cgroup(cgroup->parent);
501 if (css_depth(&parent_blkcg->css) > 0)
502 return ERR_PTR(-EINVAL);
504 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
506 return ERR_PTR(-ENOMEM);
508 blkcg->weight = BLKIO_WEIGHT_DEFAULT;
510 spin_lock_init(&blkcg->lock);
511 INIT_HLIST_HEAD(&blkcg->blkg_list);
517 * We cannot support shared io contexts, as we have no mean to support
518 * two tasks with the same ioc in two different groups without major rework
519 * of the main cic data structures. For now we allow a task to change
520 * its cgroup only if it's the only owner of its ioc.
522 static int blkiocg_can_attach(struct cgroup_subsys *subsys,
523 struct cgroup *cgroup, struct task_struct *tsk,
526 struct io_context *ioc;
529 /* task_lock() is needed to avoid races with exit_io_context() */
531 ioc = tsk->io_context;
532 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
539 static void blkiocg_attach(struct cgroup_subsys *subsys, struct cgroup *cgroup,
540 struct cgroup *prev, struct task_struct *tsk,
543 struct io_context *ioc;
546 ioc = tsk->io_context;
548 ioc->cgroup_changed = 1;
552 void blkio_policy_register(struct blkio_policy_type *blkiop)
554 spin_lock(&blkio_list_lock);
555 list_add_tail(&blkiop->list, &blkio_list);
556 spin_unlock(&blkio_list_lock);
558 EXPORT_SYMBOL_GPL(blkio_policy_register);
560 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
562 spin_lock(&blkio_list_lock);
563 list_del_init(&blkiop->list);
564 spin_unlock(&blkio_list_lock);
566 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
568 static int __init init_cgroup_blkio(void)
570 return cgroup_load_subsys(&blkio_subsys);
573 static void __exit exit_cgroup_blkio(void)
575 cgroup_unload_subsys(&blkio_subsys);
578 module_init(init_cgroup_blkio);
579 module_exit(exit_cgroup_blkio);
580 MODULE_LICENSE("GPL");