xref: /linux/block/blk-cgroup.c (revision d229807f669ba3dea9f64467ee965051c4366aed)
1 /*
2  * Common Block IO controller cgroup interface
3  *
4  * Based on ideas and code from CFQ, CFS and BFQ:
5  * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
6  *
7  * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8  *		      Paolo Valente <paolo.valente@unimore.it>
9  *
10  * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11  * 	              Nauman Rafique <nauman@google.com>
12  */
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>
22 
23 #define MAX_KEY_LEN 100
24 
25 static DEFINE_SPINLOCK(blkio_list_lock);
26 static LIST_HEAD(blkio_list);
27 
28 struct blkio_cgroup blkio_root_cgroup = { .weight = 2*BLKIO_WEIGHT_DEFAULT };
29 EXPORT_SYMBOL_GPL(blkio_root_cgroup);
30 
31 static struct cgroup_subsys_state *blkiocg_create(struct cgroup_subsys *,
32 						  struct cgroup *);
33 static int blkiocg_can_attach_task(struct cgroup *, struct task_struct *);
34 static void blkiocg_attach_task(struct cgroup *, struct task_struct *);
35 static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *);
36 static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *);
37 
38 /* for encoding cft->private value on file */
39 #define BLKIOFILE_PRIVATE(x, val)	(((x) << 16) | (val))
40 /* What policy owns the file, proportional or throttle */
41 #define BLKIOFILE_POLICY(val)		(((val) >> 16) & 0xffff)
42 #define BLKIOFILE_ATTR(val)		((val) & 0xffff)
43 
44 struct cgroup_subsys blkio_subsys = {
45 	.name = "blkio",
46 	.create = blkiocg_create,
47 	.can_attach_task = blkiocg_can_attach_task,
48 	.attach_task = blkiocg_attach_task,
49 	.destroy = blkiocg_destroy,
50 	.populate = blkiocg_populate,
51 #ifdef CONFIG_BLK_CGROUP
52 	/* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */
53 	.subsys_id = blkio_subsys_id,
54 #endif
55 	.use_id = 1,
56 	.module = THIS_MODULE,
57 };
58 EXPORT_SYMBOL_GPL(blkio_subsys);
59 
60 static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg,
61 					    struct blkio_policy_node *pn)
62 {
63 	list_add(&pn->node, &blkcg->policy_list);
64 }
65 
66 static inline bool cftype_blkg_same_policy(struct cftype *cft,
67 			struct blkio_group *blkg)
68 {
69 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
70 
71 	if (blkg->plid == plid)
72 		return 1;
73 
74 	return 0;
75 }
76 
77 /* Determines if policy node matches cgroup file being accessed */
78 static inline bool pn_matches_cftype(struct cftype *cft,
79 			struct blkio_policy_node *pn)
80 {
81 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
82 	int fileid = BLKIOFILE_ATTR(cft->private);
83 
84 	return (plid == pn->plid && fileid == pn->fileid);
85 }
86 
87 /* Must be called with blkcg->lock held */
88 static inline void blkio_policy_delete_node(struct blkio_policy_node *pn)
89 {
90 	list_del(&pn->node);
91 }
92 
93 /* Must be called with blkcg->lock held */
94 static struct blkio_policy_node *
95 blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev,
96 		enum blkio_policy_id plid, int fileid)
97 {
98 	struct blkio_policy_node *pn;
99 
100 	list_for_each_entry(pn, &blkcg->policy_list, node) {
101 		if (pn->dev == dev && pn->plid == plid && pn->fileid == fileid)
102 			return pn;
103 	}
104 
105 	return NULL;
106 }
107 
108 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
109 {
110 	return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
111 			    struct blkio_cgroup, css);
112 }
113 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
114 
115 struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
116 {
117 	return container_of(task_subsys_state(tsk, blkio_subsys_id),
118 			    struct blkio_cgroup, css);
119 }
120 EXPORT_SYMBOL_GPL(task_blkio_cgroup);
121 
122 static inline void
123 blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight)
124 {
125 	struct blkio_policy_type *blkiop;
126 
127 	list_for_each_entry(blkiop, &blkio_list, list) {
128 		/* If this policy does not own the blkg, do not send updates */
129 		if (blkiop->plid != blkg->plid)
130 			continue;
131 		if (blkiop->ops.blkio_update_group_weight_fn)
132 			blkiop->ops.blkio_update_group_weight_fn(blkg->key,
133 							blkg, weight);
134 	}
135 }
136 
137 static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps,
138 				int fileid)
139 {
140 	struct blkio_policy_type *blkiop;
141 
142 	list_for_each_entry(blkiop, &blkio_list, list) {
143 
144 		/* If this policy does not own the blkg, do not send updates */
145 		if (blkiop->plid != blkg->plid)
146 			continue;
147 
148 		if (fileid == BLKIO_THROTL_read_bps_device
149 		    && blkiop->ops.blkio_update_group_read_bps_fn)
150 			blkiop->ops.blkio_update_group_read_bps_fn(blkg->key,
151 								blkg, bps);
152 
153 		if (fileid == BLKIO_THROTL_write_bps_device
154 		    && blkiop->ops.blkio_update_group_write_bps_fn)
155 			blkiop->ops.blkio_update_group_write_bps_fn(blkg->key,
156 								blkg, bps);
157 	}
158 }
159 
160 static inline void blkio_update_group_iops(struct blkio_group *blkg,
161 			unsigned int iops, int fileid)
162 {
163 	struct blkio_policy_type *blkiop;
164 
165 	list_for_each_entry(blkiop, &blkio_list, list) {
166 
167 		/* If this policy does not own the blkg, do not send updates */
168 		if (blkiop->plid != blkg->plid)
169 			continue;
170 
171 		if (fileid == BLKIO_THROTL_read_iops_device
172 		    && blkiop->ops.blkio_update_group_read_iops_fn)
173 			blkiop->ops.blkio_update_group_read_iops_fn(blkg->key,
174 								blkg, iops);
175 
176 		if (fileid == BLKIO_THROTL_write_iops_device
177 		    && blkiop->ops.blkio_update_group_write_iops_fn)
178 			blkiop->ops.blkio_update_group_write_iops_fn(blkg->key,
179 								blkg,iops);
180 	}
181 }
182 
183 /*
184  * Add to the appropriate stat variable depending on the request type.
185  * This should be called with the blkg->stats_lock held.
186  */
187 static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
188 				bool sync)
189 {
190 	if (direction)
191 		stat[BLKIO_STAT_WRITE] += add;
192 	else
193 		stat[BLKIO_STAT_READ] += add;
194 	if (sync)
195 		stat[BLKIO_STAT_SYNC] += add;
196 	else
197 		stat[BLKIO_STAT_ASYNC] += add;
198 }
199 
200 /*
201  * Decrements the appropriate stat variable if non-zero depending on the
202  * request type. Panics on value being zero.
203  * This should be called with the blkg->stats_lock held.
204  */
205 static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
206 {
207 	if (direction) {
208 		BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
209 		stat[BLKIO_STAT_WRITE]--;
210 	} else {
211 		BUG_ON(stat[BLKIO_STAT_READ] == 0);
212 		stat[BLKIO_STAT_READ]--;
213 	}
214 	if (sync) {
215 		BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
216 		stat[BLKIO_STAT_SYNC]--;
217 	} else {
218 		BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
219 		stat[BLKIO_STAT_ASYNC]--;
220 	}
221 }
222 
223 #ifdef CONFIG_DEBUG_BLK_CGROUP
224 /* This should be called with the blkg->stats_lock held. */
225 static void blkio_set_start_group_wait_time(struct blkio_group *blkg,
226 						struct blkio_group *curr_blkg)
227 {
228 	if (blkio_blkg_waiting(&blkg->stats))
229 		return;
230 	if (blkg == curr_blkg)
231 		return;
232 	blkg->stats.start_group_wait_time = sched_clock();
233 	blkio_mark_blkg_waiting(&blkg->stats);
234 }
235 
236 /* This should be called with the blkg->stats_lock held. */
237 static void blkio_update_group_wait_time(struct blkio_group_stats *stats)
238 {
239 	unsigned long long now;
240 
241 	if (!blkio_blkg_waiting(stats))
242 		return;
243 
244 	now = sched_clock();
245 	if (time_after64(now, stats->start_group_wait_time))
246 		stats->group_wait_time += now - stats->start_group_wait_time;
247 	blkio_clear_blkg_waiting(stats);
248 }
249 
250 /* This should be called with the blkg->stats_lock held. */
251 static void blkio_end_empty_time(struct blkio_group_stats *stats)
252 {
253 	unsigned long long now;
254 
255 	if (!blkio_blkg_empty(stats))
256 		return;
257 
258 	now = sched_clock();
259 	if (time_after64(now, stats->start_empty_time))
260 		stats->empty_time += now - stats->start_empty_time;
261 	blkio_clear_blkg_empty(stats);
262 }
263 
264 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
265 {
266 	unsigned long flags;
267 
268 	spin_lock_irqsave(&blkg->stats_lock, flags);
269 	BUG_ON(blkio_blkg_idling(&blkg->stats));
270 	blkg->stats.start_idle_time = sched_clock();
271 	blkio_mark_blkg_idling(&blkg->stats);
272 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
273 }
274 EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats);
275 
276 void blkiocg_update_idle_time_stats(struct blkio_group *blkg)
277 {
278 	unsigned long flags;
279 	unsigned long long now;
280 	struct blkio_group_stats *stats;
281 
282 	spin_lock_irqsave(&blkg->stats_lock, flags);
283 	stats = &blkg->stats;
284 	if (blkio_blkg_idling(stats)) {
285 		now = sched_clock();
286 		if (time_after64(now, stats->start_idle_time))
287 			stats->idle_time += now - stats->start_idle_time;
288 		blkio_clear_blkg_idling(stats);
289 	}
290 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
291 }
292 EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats);
293 
294 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg)
295 {
296 	unsigned long flags;
297 	struct blkio_group_stats *stats;
298 
299 	spin_lock_irqsave(&blkg->stats_lock, flags);
300 	stats = &blkg->stats;
301 	stats->avg_queue_size_sum +=
302 			stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
303 			stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
304 	stats->avg_queue_size_samples++;
305 	blkio_update_group_wait_time(stats);
306 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
307 }
308 EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats);
309 
310 void blkiocg_set_start_empty_time(struct blkio_group *blkg)
311 {
312 	unsigned long flags;
313 	struct blkio_group_stats *stats;
314 
315 	spin_lock_irqsave(&blkg->stats_lock, flags);
316 	stats = &blkg->stats;
317 
318 	if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] ||
319 			stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) {
320 		spin_unlock_irqrestore(&blkg->stats_lock, flags);
321 		return;
322 	}
323 
324 	/*
325 	 * group is already marked empty. This can happen if cfqq got new
326 	 * request in parent group and moved to this group while being added
327 	 * to service tree. Just ignore the event and move on.
328 	 */
329 	if(blkio_blkg_empty(stats)) {
330 		spin_unlock_irqrestore(&blkg->stats_lock, flags);
331 		return;
332 	}
333 
334 	stats->start_empty_time = sched_clock();
335 	blkio_mark_blkg_empty(stats);
336 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
337 }
338 EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time);
339 
340 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
341 			unsigned long dequeue)
342 {
343 	blkg->stats.dequeue += dequeue;
344 }
345 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
346 #else
347 static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg,
348 					struct blkio_group *curr_blkg) {}
349 static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {}
350 #endif
351 
352 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
353 			struct blkio_group *curr_blkg, bool direction,
354 			bool sync)
355 {
356 	unsigned long flags;
357 
358 	spin_lock_irqsave(&blkg->stats_lock, flags);
359 	blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction,
360 			sync);
361 	blkio_end_empty_time(&blkg->stats);
362 	blkio_set_start_group_wait_time(blkg, curr_blkg);
363 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
364 }
365 EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats);
366 
367 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
368 						bool direction, bool sync)
369 {
370 	unsigned long flags;
371 
372 	spin_lock_irqsave(&blkg->stats_lock, flags);
373 	blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED],
374 					direction, sync);
375 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
376 }
377 EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats);
378 
379 void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time,
380 				unsigned long unaccounted_time)
381 {
382 	unsigned long flags;
383 
384 	spin_lock_irqsave(&blkg->stats_lock, flags);
385 	blkg->stats.time += time;
386 #ifdef CONFIG_DEBUG_BLK_CGROUP
387 	blkg->stats.unaccounted_time += unaccounted_time;
388 #endif
389 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
390 }
391 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
392 
393 /*
394  * should be called under rcu read lock or queue lock to make sure blkg pointer
395  * is valid.
396  */
397 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
398 				uint64_t bytes, bool direction, bool sync)
399 {
400 	struct blkio_group_stats_cpu *stats_cpu;
401 	unsigned long flags;
402 
403 	/*
404 	 * Disabling interrupts to provide mutual exclusion between two
405 	 * writes on same cpu. It probably is not needed for 64bit. Not
406 	 * optimizing that case yet.
407 	 */
408 	local_irq_save(flags);
409 
410 	stats_cpu = this_cpu_ptr(blkg->stats_cpu);
411 
412 	u64_stats_update_begin(&stats_cpu->syncp);
413 	stats_cpu->sectors += bytes >> 9;
414 	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
415 			1, direction, sync);
416 	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
417 			bytes, direction, sync);
418 	u64_stats_update_end(&stats_cpu->syncp);
419 	local_irq_restore(flags);
420 }
421 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
422 
423 void blkiocg_update_completion_stats(struct blkio_group *blkg,
424 	uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
425 {
426 	struct blkio_group_stats *stats;
427 	unsigned long flags;
428 	unsigned long long now = sched_clock();
429 
430 	spin_lock_irqsave(&blkg->stats_lock, flags);
431 	stats = &blkg->stats;
432 	if (time_after64(now, io_start_time))
433 		blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
434 				now - io_start_time, direction, sync);
435 	if (time_after64(io_start_time, start_time))
436 		blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
437 				io_start_time - start_time, direction, sync);
438 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
439 }
440 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
441 
442 /*  Merged stats are per cpu.  */
443 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
444 					bool sync)
445 {
446 	struct blkio_group_stats_cpu *stats_cpu;
447 	unsigned long flags;
448 
449 	/*
450 	 * Disabling interrupts to provide mutual exclusion between two
451 	 * writes on same cpu. It probably is not needed for 64bit. Not
452 	 * optimizing that case yet.
453 	 */
454 	local_irq_save(flags);
455 
456 	stats_cpu = this_cpu_ptr(blkg->stats_cpu);
457 
458 	u64_stats_update_begin(&stats_cpu->syncp);
459 	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1,
460 				direction, sync);
461 	u64_stats_update_end(&stats_cpu->syncp);
462 	local_irq_restore(flags);
463 }
464 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
465 
466 /*
467  * This function allocates the per cpu stats for blkio_group. Should be called
468  * from sleepable context as alloc_per_cpu() requires that.
469  */
470 int blkio_alloc_blkg_stats(struct blkio_group *blkg)
471 {
472 	/* Allocate memory for per cpu stats */
473 	blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
474 	if (!blkg->stats_cpu)
475 		return -ENOMEM;
476 	return 0;
477 }
478 EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
479 
480 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
481 		struct blkio_group *blkg, void *key, dev_t dev,
482 		enum blkio_policy_id plid)
483 {
484 	unsigned long flags;
485 
486 	spin_lock_irqsave(&blkcg->lock, flags);
487 	spin_lock_init(&blkg->stats_lock);
488 	rcu_assign_pointer(blkg->key, key);
489 	blkg->blkcg_id = css_id(&blkcg->css);
490 	hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
491 	blkg->plid = plid;
492 	spin_unlock_irqrestore(&blkcg->lock, flags);
493 	/* Need to take css reference ? */
494 	cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
495 	blkg->dev = dev;
496 }
497 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
498 
499 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
500 {
501 	hlist_del_init_rcu(&blkg->blkcg_node);
502 	blkg->blkcg_id = 0;
503 }
504 
505 /*
506  * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
507  * indicating that blk_group was unhashed by the time we got to it.
508  */
509 int blkiocg_del_blkio_group(struct blkio_group *blkg)
510 {
511 	struct blkio_cgroup *blkcg;
512 	unsigned long flags;
513 	struct cgroup_subsys_state *css;
514 	int ret = 1;
515 
516 	rcu_read_lock();
517 	css = css_lookup(&blkio_subsys, blkg->blkcg_id);
518 	if (css) {
519 		blkcg = container_of(css, struct blkio_cgroup, css);
520 		spin_lock_irqsave(&blkcg->lock, flags);
521 		if (!hlist_unhashed(&blkg->blkcg_node)) {
522 			__blkiocg_del_blkio_group(blkg);
523 			ret = 0;
524 		}
525 		spin_unlock_irqrestore(&blkcg->lock, flags);
526 	}
527 
528 	rcu_read_unlock();
529 	return ret;
530 }
531 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
532 
533 /* called under rcu_read_lock(). */
534 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
535 {
536 	struct blkio_group *blkg;
537 	struct hlist_node *n;
538 	void *__key;
539 
540 	hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
541 		__key = blkg->key;
542 		if (__key == key)
543 			return blkg;
544 	}
545 
546 	return NULL;
547 }
548 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
549 
550 static void blkio_reset_stats_cpu(struct blkio_group *blkg)
551 {
552 	struct blkio_group_stats_cpu *stats_cpu;
553 	int i, j, k;
554 	/*
555 	 * Note: On 64 bit arch this should not be an issue. This has the
556 	 * possibility of returning some inconsistent value on 32bit arch
557 	 * as 64bit update on 32bit is non atomic. Taking care of this
558 	 * corner case makes code very complicated, like sending IPIs to
559 	 * cpus, taking care of stats of offline cpus etc.
560 	 *
561 	 * reset stats is anyway more of a debug feature and this sounds a
562 	 * corner case. So I am not complicating the code yet until and
563 	 * unless this becomes a real issue.
564 	 */
565 	for_each_possible_cpu(i) {
566 		stats_cpu = per_cpu_ptr(blkg->stats_cpu, i);
567 		stats_cpu->sectors = 0;
568 		for(j = 0; j < BLKIO_STAT_CPU_NR; j++)
569 			for (k = 0; k < BLKIO_STAT_TOTAL; k++)
570 				stats_cpu->stat_arr_cpu[j][k] = 0;
571 	}
572 }
573 
574 static int
575 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
576 {
577 	struct blkio_cgroup *blkcg;
578 	struct blkio_group *blkg;
579 	struct blkio_group_stats *stats;
580 	struct hlist_node *n;
581 	uint64_t queued[BLKIO_STAT_TOTAL];
582 	int i;
583 #ifdef CONFIG_DEBUG_BLK_CGROUP
584 	bool idling, waiting, empty;
585 	unsigned long long now = sched_clock();
586 #endif
587 
588 	blkcg = cgroup_to_blkio_cgroup(cgroup);
589 	spin_lock_irq(&blkcg->lock);
590 	hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
591 		spin_lock(&blkg->stats_lock);
592 		stats = &blkg->stats;
593 #ifdef CONFIG_DEBUG_BLK_CGROUP
594 		idling = blkio_blkg_idling(stats);
595 		waiting = blkio_blkg_waiting(stats);
596 		empty = blkio_blkg_empty(stats);
597 #endif
598 		for (i = 0; i < BLKIO_STAT_TOTAL; i++)
599 			queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
600 		memset(stats, 0, sizeof(struct blkio_group_stats));
601 		for (i = 0; i < BLKIO_STAT_TOTAL; i++)
602 			stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
603 #ifdef CONFIG_DEBUG_BLK_CGROUP
604 		if (idling) {
605 			blkio_mark_blkg_idling(stats);
606 			stats->start_idle_time = now;
607 		}
608 		if (waiting) {
609 			blkio_mark_blkg_waiting(stats);
610 			stats->start_group_wait_time = now;
611 		}
612 		if (empty) {
613 			blkio_mark_blkg_empty(stats);
614 			stats->start_empty_time = now;
615 		}
616 #endif
617 		spin_unlock(&blkg->stats_lock);
618 
619 		/* Reset Per cpu stats which don't take blkg->stats_lock */
620 		blkio_reset_stats_cpu(blkg);
621 	}
622 
623 	spin_unlock_irq(&blkcg->lock);
624 	return 0;
625 }
626 
627 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
628 				int chars_left, bool diskname_only)
629 {
630 	snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
631 	chars_left -= strlen(str);
632 	if (chars_left <= 0) {
633 		printk(KERN_WARNING
634 			"Possibly incorrect cgroup stat display format");
635 		return;
636 	}
637 	if (diskname_only)
638 		return;
639 	switch (type) {
640 	case BLKIO_STAT_READ:
641 		strlcat(str, " Read", chars_left);
642 		break;
643 	case BLKIO_STAT_WRITE:
644 		strlcat(str, " Write", chars_left);
645 		break;
646 	case BLKIO_STAT_SYNC:
647 		strlcat(str, " Sync", chars_left);
648 		break;
649 	case BLKIO_STAT_ASYNC:
650 		strlcat(str, " Async", chars_left);
651 		break;
652 	case BLKIO_STAT_TOTAL:
653 		strlcat(str, " Total", chars_left);
654 		break;
655 	default:
656 		strlcat(str, " Invalid", chars_left);
657 	}
658 }
659 
660 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
661 				struct cgroup_map_cb *cb, dev_t dev)
662 {
663 	blkio_get_key_name(0, dev, str, chars_left, true);
664 	cb->fill(cb, str, val);
665 	return val;
666 }
667 
668 
669 static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
670 			enum stat_type_cpu type, enum stat_sub_type sub_type)
671 {
672 	int cpu;
673 	struct blkio_group_stats_cpu *stats_cpu;
674 	u64 val = 0, tval;
675 
676 	for_each_possible_cpu(cpu) {
677 		unsigned int start;
678 		stats_cpu  = per_cpu_ptr(blkg->stats_cpu, cpu);
679 
680 		do {
681 			start = u64_stats_fetch_begin(&stats_cpu->syncp);
682 			if (type == BLKIO_STAT_CPU_SECTORS)
683 				tval = stats_cpu->sectors;
684 			else
685 				tval = stats_cpu->stat_arr_cpu[type][sub_type];
686 		} while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
687 
688 		val += tval;
689 	}
690 
691 	return val;
692 }
693 
694 static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
695 		struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
696 {
697 	uint64_t disk_total, val;
698 	char key_str[MAX_KEY_LEN];
699 	enum stat_sub_type sub_type;
700 
701 	if (type == BLKIO_STAT_CPU_SECTORS) {
702 		val = blkio_read_stat_cpu(blkg, type, 0);
703 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
704 	}
705 
706 	for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
707 			sub_type++) {
708 		blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
709 		val = blkio_read_stat_cpu(blkg, type, sub_type);
710 		cb->fill(cb, key_str, val);
711 	}
712 
713 	disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
714 			blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
715 
716 	blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
717 	cb->fill(cb, key_str, disk_total);
718 	return disk_total;
719 }
720 
721 /* This should be called with blkg->stats_lock held */
722 static uint64_t blkio_get_stat(struct blkio_group *blkg,
723 		struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
724 {
725 	uint64_t disk_total;
726 	char key_str[MAX_KEY_LEN];
727 	enum stat_sub_type sub_type;
728 
729 	if (type == BLKIO_STAT_TIME)
730 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
731 					blkg->stats.time, cb, dev);
732 #ifdef CONFIG_DEBUG_BLK_CGROUP
733 	if (type == BLKIO_STAT_UNACCOUNTED_TIME)
734 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
735 					blkg->stats.unaccounted_time, cb, dev);
736 	if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
737 		uint64_t sum = blkg->stats.avg_queue_size_sum;
738 		uint64_t samples = blkg->stats.avg_queue_size_samples;
739 		if (samples)
740 			do_div(sum, samples);
741 		else
742 			sum = 0;
743 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
744 	}
745 	if (type == BLKIO_STAT_GROUP_WAIT_TIME)
746 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
747 					blkg->stats.group_wait_time, cb, dev);
748 	if (type == BLKIO_STAT_IDLE_TIME)
749 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
750 					blkg->stats.idle_time, cb, dev);
751 	if (type == BLKIO_STAT_EMPTY_TIME)
752 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
753 					blkg->stats.empty_time, cb, dev);
754 	if (type == BLKIO_STAT_DEQUEUE)
755 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
756 					blkg->stats.dequeue, cb, dev);
757 #endif
758 
759 	for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
760 			sub_type++) {
761 		blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
762 		cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
763 	}
764 	disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
765 			blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
766 	blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
767 	cb->fill(cb, key_str, disk_total);
768 	return disk_total;
769 }
770 
771 static int blkio_check_dev_num(dev_t dev)
772 {
773 	int part = 0;
774 	struct gendisk *disk;
775 
776 	disk = get_gendisk(dev, &part);
777 	if (!disk || part)
778 		return -ENODEV;
779 
780 	return 0;
781 }
782 
783 static int blkio_policy_parse_and_set(char *buf,
784 	struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
785 {
786 	char *s[4], *p, *major_s = NULL, *minor_s = NULL;
787 	int ret;
788 	unsigned long major, minor;
789 	int i = 0;
790 	dev_t dev;
791 	u64 temp;
792 
793 	memset(s, 0, sizeof(s));
794 
795 	while ((p = strsep(&buf, " ")) != NULL) {
796 		if (!*p)
797 			continue;
798 
799 		s[i++] = p;
800 
801 		/* Prevent from inputing too many things */
802 		if (i == 3)
803 			break;
804 	}
805 
806 	if (i != 2)
807 		return -EINVAL;
808 
809 	p = strsep(&s[0], ":");
810 	if (p != NULL)
811 		major_s = p;
812 	else
813 		return -EINVAL;
814 
815 	minor_s = s[0];
816 	if (!minor_s)
817 		return -EINVAL;
818 
819 	ret = strict_strtoul(major_s, 10, &major);
820 	if (ret)
821 		return -EINVAL;
822 
823 	ret = strict_strtoul(minor_s, 10, &minor);
824 	if (ret)
825 		return -EINVAL;
826 
827 	dev = MKDEV(major, minor);
828 
829 	ret = strict_strtoull(s[1], 10, &temp);
830 	if (ret)
831 		return -EINVAL;
832 
833 	/* For rule removal, do not check for device presence. */
834 	if (temp) {
835 		ret = blkio_check_dev_num(dev);
836 		if (ret)
837 			return ret;
838 	}
839 
840 	newpn->dev = dev;
841 
842 	switch (plid) {
843 	case BLKIO_POLICY_PROP:
844 		if ((temp < BLKIO_WEIGHT_MIN && temp > 0) ||
845 		     temp > BLKIO_WEIGHT_MAX)
846 			return -EINVAL;
847 
848 		newpn->plid = plid;
849 		newpn->fileid = fileid;
850 		newpn->val.weight = temp;
851 		break;
852 	case BLKIO_POLICY_THROTL:
853 		switch(fileid) {
854 		case BLKIO_THROTL_read_bps_device:
855 		case BLKIO_THROTL_write_bps_device:
856 			newpn->plid = plid;
857 			newpn->fileid = fileid;
858 			newpn->val.bps = temp;
859 			break;
860 		case BLKIO_THROTL_read_iops_device:
861 		case BLKIO_THROTL_write_iops_device:
862 			if (temp > THROTL_IOPS_MAX)
863 				return -EINVAL;
864 
865 			newpn->plid = plid;
866 			newpn->fileid = fileid;
867 			newpn->val.iops = (unsigned int)temp;
868 			break;
869 		}
870 		break;
871 	default:
872 		BUG();
873 	}
874 
875 	return 0;
876 }
877 
878 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
879 			      dev_t dev)
880 {
881 	struct blkio_policy_node *pn;
882 
883 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
884 				BLKIO_PROP_weight_device);
885 	if (pn)
886 		return pn->val.weight;
887 	else
888 		return blkcg->weight;
889 }
890 EXPORT_SYMBOL_GPL(blkcg_get_weight);
891 
892 uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, dev_t dev)
893 {
894 	struct blkio_policy_node *pn;
895 
896 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
897 				BLKIO_THROTL_read_bps_device);
898 	if (pn)
899 		return pn->val.bps;
900 	else
901 		return -1;
902 }
903 
904 uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
905 {
906 	struct blkio_policy_node *pn;
907 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
908 				BLKIO_THROTL_write_bps_device);
909 	if (pn)
910 		return pn->val.bps;
911 	else
912 		return -1;
913 }
914 
915 unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
916 {
917 	struct blkio_policy_node *pn;
918 
919 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
920 				BLKIO_THROTL_read_iops_device);
921 	if (pn)
922 		return pn->val.iops;
923 	else
924 		return -1;
925 }
926 
927 unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
928 {
929 	struct blkio_policy_node *pn;
930 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
931 				BLKIO_THROTL_write_iops_device);
932 	if (pn)
933 		return pn->val.iops;
934 	else
935 		return -1;
936 }
937 
938 /* Checks whether user asked for deleting a policy rule */
939 static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
940 {
941 	switch(pn->plid) {
942 	case BLKIO_POLICY_PROP:
943 		if (pn->val.weight == 0)
944 			return 1;
945 		break;
946 	case BLKIO_POLICY_THROTL:
947 		switch(pn->fileid) {
948 		case BLKIO_THROTL_read_bps_device:
949 		case BLKIO_THROTL_write_bps_device:
950 			if (pn->val.bps == 0)
951 				return 1;
952 			break;
953 		case BLKIO_THROTL_read_iops_device:
954 		case BLKIO_THROTL_write_iops_device:
955 			if (pn->val.iops == 0)
956 				return 1;
957 		}
958 		break;
959 	default:
960 		BUG();
961 	}
962 
963 	return 0;
964 }
965 
966 static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
967 					struct blkio_policy_node *newpn)
968 {
969 	switch(oldpn->plid) {
970 	case BLKIO_POLICY_PROP:
971 		oldpn->val.weight = newpn->val.weight;
972 		break;
973 	case BLKIO_POLICY_THROTL:
974 		switch(newpn->fileid) {
975 		case BLKIO_THROTL_read_bps_device:
976 		case BLKIO_THROTL_write_bps_device:
977 			oldpn->val.bps = newpn->val.bps;
978 			break;
979 		case BLKIO_THROTL_read_iops_device:
980 		case BLKIO_THROTL_write_iops_device:
981 			oldpn->val.iops = newpn->val.iops;
982 		}
983 		break;
984 	default:
985 		BUG();
986 	}
987 }
988 
989 /*
990  * Some rules/values in blkg have changed. Propagate those to respective
991  * policies.
992  */
993 static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
994 		struct blkio_group *blkg, struct blkio_policy_node *pn)
995 {
996 	unsigned int weight, iops;
997 	u64 bps;
998 
999 	switch(pn->plid) {
1000 	case BLKIO_POLICY_PROP:
1001 		weight = pn->val.weight ? pn->val.weight :
1002 				blkcg->weight;
1003 		blkio_update_group_weight(blkg, weight);
1004 		break;
1005 	case BLKIO_POLICY_THROTL:
1006 		switch(pn->fileid) {
1007 		case BLKIO_THROTL_read_bps_device:
1008 		case BLKIO_THROTL_write_bps_device:
1009 			bps = pn->val.bps ? pn->val.bps : (-1);
1010 			blkio_update_group_bps(blkg, bps, pn->fileid);
1011 			break;
1012 		case BLKIO_THROTL_read_iops_device:
1013 		case BLKIO_THROTL_write_iops_device:
1014 			iops = pn->val.iops ? pn->val.iops : (-1);
1015 			blkio_update_group_iops(blkg, iops, pn->fileid);
1016 			break;
1017 		}
1018 		break;
1019 	default:
1020 		BUG();
1021 	}
1022 }
1023 
1024 /*
1025  * A policy node rule has been updated. Propagate this update to all the
1026  * block groups which might be affected by this update.
1027  */
1028 static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
1029 				struct blkio_policy_node *pn)
1030 {
1031 	struct blkio_group *blkg;
1032 	struct hlist_node *n;
1033 
1034 	spin_lock(&blkio_list_lock);
1035 	spin_lock_irq(&blkcg->lock);
1036 
1037 	hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1038 		if (pn->dev != blkg->dev || pn->plid != blkg->plid)
1039 			continue;
1040 		blkio_update_blkg_policy(blkcg, blkg, pn);
1041 	}
1042 
1043 	spin_unlock_irq(&blkcg->lock);
1044 	spin_unlock(&blkio_list_lock);
1045 }
1046 
1047 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1048  				       const char *buffer)
1049 {
1050 	int ret = 0;
1051 	char *buf;
1052 	struct blkio_policy_node *newpn, *pn;
1053 	struct blkio_cgroup *blkcg;
1054 	int keep_newpn = 0;
1055 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1056 	int fileid = BLKIOFILE_ATTR(cft->private);
1057 
1058 	buf = kstrdup(buffer, GFP_KERNEL);
1059 	if (!buf)
1060 		return -ENOMEM;
1061 
1062 	newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
1063 	if (!newpn) {
1064 		ret = -ENOMEM;
1065 		goto free_buf;
1066 	}
1067 
1068 	ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
1069 	if (ret)
1070 		goto free_newpn;
1071 
1072 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1073 
1074 	spin_lock_irq(&blkcg->lock);
1075 
1076 	pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
1077 	if (!pn) {
1078 		if (!blkio_delete_rule_command(newpn)) {
1079 			blkio_policy_insert_node(blkcg, newpn);
1080 			keep_newpn = 1;
1081 		}
1082 		spin_unlock_irq(&blkcg->lock);
1083 		goto update_io_group;
1084 	}
1085 
1086 	if (blkio_delete_rule_command(newpn)) {
1087 		blkio_policy_delete_node(pn);
1088 		spin_unlock_irq(&blkcg->lock);
1089 		goto update_io_group;
1090 	}
1091 	spin_unlock_irq(&blkcg->lock);
1092 
1093 	blkio_update_policy_rule(pn, newpn);
1094 
1095 update_io_group:
1096 	blkio_update_policy_node_blkg(blkcg, newpn);
1097 
1098 free_newpn:
1099 	if (!keep_newpn)
1100 		kfree(newpn);
1101 free_buf:
1102 	kfree(buf);
1103 	return ret;
1104 }
1105 
1106 static void
1107 blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
1108 {
1109 	switch(pn->plid) {
1110 		case BLKIO_POLICY_PROP:
1111 			if (pn->fileid == BLKIO_PROP_weight_device)
1112 				seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1113 					MINOR(pn->dev), pn->val.weight);
1114 			break;
1115 		case BLKIO_POLICY_THROTL:
1116 			switch(pn->fileid) {
1117 			case BLKIO_THROTL_read_bps_device:
1118 			case BLKIO_THROTL_write_bps_device:
1119 				seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1120 					MINOR(pn->dev), pn->val.bps);
1121 				break;
1122 			case BLKIO_THROTL_read_iops_device:
1123 			case BLKIO_THROTL_write_iops_device:
1124 				seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1125 					MINOR(pn->dev), pn->val.iops);
1126 				break;
1127 			}
1128 			break;
1129 		default:
1130 			BUG();
1131 	}
1132 }
1133 
1134 /* cgroup files which read their data from policy nodes end up here */
1135 static void blkio_read_policy_node_files(struct cftype *cft,
1136 			struct blkio_cgroup *blkcg, struct seq_file *m)
1137 {
1138 	struct blkio_policy_node *pn;
1139 
1140 	if (!list_empty(&blkcg->policy_list)) {
1141 		spin_lock_irq(&blkcg->lock);
1142 		list_for_each_entry(pn, &blkcg->policy_list, node) {
1143 			if (!pn_matches_cftype(cft, pn))
1144 				continue;
1145 			blkio_print_policy_node(m, pn);
1146 		}
1147 		spin_unlock_irq(&blkcg->lock);
1148 	}
1149 }
1150 
1151 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1152 				struct seq_file *m)
1153 {
1154 	struct blkio_cgroup *blkcg;
1155 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1156 	int name = BLKIOFILE_ATTR(cft->private);
1157 
1158 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1159 
1160 	switch(plid) {
1161 	case BLKIO_POLICY_PROP:
1162 		switch(name) {
1163 		case BLKIO_PROP_weight_device:
1164 			blkio_read_policy_node_files(cft, blkcg, m);
1165 			return 0;
1166 		default:
1167 			BUG();
1168 		}
1169 		break;
1170 	case BLKIO_POLICY_THROTL:
1171 		switch(name){
1172 		case BLKIO_THROTL_read_bps_device:
1173 		case BLKIO_THROTL_write_bps_device:
1174 		case BLKIO_THROTL_read_iops_device:
1175 		case BLKIO_THROTL_write_iops_device:
1176 			blkio_read_policy_node_files(cft, blkcg, m);
1177 			return 0;
1178 		default:
1179 			BUG();
1180 		}
1181 		break;
1182 	default:
1183 		BUG();
1184 	}
1185 
1186 	return 0;
1187 }
1188 
1189 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1190 		struct cftype *cft, struct cgroup_map_cb *cb,
1191 		enum stat_type type, bool show_total, bool pcpu)
1192 {
1193 	struct blkio_group *blkg;
1194 	struct hlist_node *n;
1195 	uint64_t cgroup_total = 0;
1196 
1197 	rcu_read_lock();
1198 	hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1199 		if (blkg->dev) {
1200 			if (!cftype_blkg_same_policy(cft, blkg))
1201 				continue;
1202 			if (pcpu)
1203 				cgroup_total += blkio_get_stat_cpu(blkg, cb,
1204 						blkg->dev, type);
1205 			else {
1206 				spin_lock_irq(&blkg->stats_lock);
1207 				cgroup_total += blkio_get_stat(blkg, cb,
1208 						blkg->dev, type);
1209 				spin_unlock_irq(&blkg->stats_lock);
1210 			}
1211 		}
1212 	}
1213 	if (show_total)
1214 		cb->fill(cb, "Total", cgroup_total);
1215 	rcu_read_unlock();
1216 	return 0;
1217 }
1218 
1219 /* All map kind of cgroup file get serviced by this function */
1220 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1221 				struct cgroup_map_cb *cb)
1222 {
1223 	struct blkio_cgroup *blkcg;
1224 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1225 	int name = BLKIOFILE_ATTR(cft->private);
1226 
1227 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1228 
1229 	switch(plid) {
1230 	case BLKIO_POLICY_PROP:
1231 		switch(name) {
1232 		case BLKIO_PROP_time:
1233 			return blkio_read_blkg_stats(blkcg, cft, cb,
1234 						BLKIO_STAT_TIME, 0, 0);
1235 		case BLKIO_PROP_sectors:
1236 			return blkio_read_blkg_stats(blkcg, cft, cb,
1237 						BLKIO_STAT_CPU_SECTORS, 0, 1);
1238 		case BLKIO_PROP_io_service_bytes:
1239 			return blkio_read_blkg_stats(blkcg, cft, cb,
1240 					BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1241 		case BLKIO_PROP_io_serviced:
1242 			return blkio_read_blkg_stats(blkcg, cft, cb,
1243 						BLKIO_STAT_CPU_SERVICED, 1, 1);
1244 		case BLKIO_PROP_io_service_time:
1245 			return blkio_read_blkg_stats(blkcg, cft, cb,
1246 						BLKIO_STAT_SERVICE_TIME, 1, 0);
1247 		case BLKIO_PROP_io_wait_time:
1248 			return blkio_read_blkg_stats(blkcg, cft, cb,
1249 						BLKIO_STAT_WAIT_TIME, 1, 0);
1250 		case BLKIO_PROP_io_merged:
1251 			return blkio_read_blkg_stats(blkcg, cft, cb,
1252 						BLKIO_STAT_CPU_MERGED, 1, 1);
1253 		case BLKIO_PROP_io_queued:
1254 			return blkio_read_blkg_stats(blkcg, cft, cb,
1255 						BLKIO_STAT_QUEUED, 1, 0);
1256 #ifdef CONFIG_DEBUG_BLK_CGROUP
1257 		case BLKIO_PROP_unaccounted_time:
1258 			return blkio_read_blkg_stats(blkcg, cft, cb,
1259 					BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
1260 		case BLKIO_PROP_dequeue:
1261 			return blkio_read_blkg_stats(blkcg, cft, cb,
1262 						BLKIO_STAT_DEQUEUE, 0, 0);
1263 		case BLKIO_PROP_avg_queue_size:
1264 			return blkio_read_blkg_stats(blkcg, cft, cb,
1265 					BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
1266 		case BLKIO_PROP_group_wait_time:
1267 			return blkio_read_blkg_stats(blkcg, cft, cb,
1268 					BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
1269 		case BLKIO_PROP_idle_time:
1270 			return blkio_read_blkg_stats(blkcg, cft, cb,
1271 						BLKIO_STAT_IDLE_TIME, 0, 0);
1272 		case BLKIO_PROP_empty_time:
1273 			return blkio_read_blkg_stats(blkcg, cft, cb,
1274 						BLKIO_STAT_EMPTY_TIME, 0, 0);
1275 #endif
1276 		default:
1277 			BUG();
1278 		}
1279 		break;
1280 	case BLKIO_POLICY_THROTL:
1281 		switch(name){
1282 		case BLKIO_THROTL_io_service_bytes:
1283 			return blkio_read_blkg_stats(blkcg, cft, cb,
1284 						BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1285 		case BLKIO_THROTL_io_serviced:
1286 			return blkio_read_blkg_stats(blkcg, cft, cb,
1287 						BLKIO_STAT_CPU_SERVICED, 1, 1);
1288 		default:
1289 			BUG();
1290 		}
1291 		break;
1292 	default:
1293 		BUG();
1294 	}
1295 
1296 	return 0;
1297 }
1298 
1299 static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1300 {
1301 	struct blkio_group *blkg;
1302 	struct hlist_node *n;
1303 	struct blkio_policy_node *pn;
1304 
1305 	if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1306 		return -EINVAL;
1307 
1308 	spin_lock(&blkio_list_lock);
1309 	spin_lock_irq(&blkcg->lock);
1310 	blkcg->weight = (unsigned int)val;
1311 
1312 	hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1313 		pn = blkio_policy_search_node(blkcg, blkg->dev,
1314 				BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1315 		if (pn)
1316 			continue;
1317 
1318 		blkio_update_group_weight(blkg, blkcg->weight);
1319 	}
1320 	spin_unlock_irq(&blkcg->lock);
1321 	spin_unlock(&blkio_list_lock);
1322 	return 0;
1323 }
1324 
1325 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1326 	struct blkio_cgroup *blkcg;
1327 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1328 	int name = BLKIOFILE_ATTR(cft->private);
1329 
1330 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1331 
1332 	switch(plid) {
1333 	case BLKIO_POLICY_PROP:
1334 		switch(name) {
1335 		case BLKIO_PROP_weight:
1336 			return (u64)blkcg->weight;
1337 		}
1338 		break;
1339 	default:
1340 		BUG();
1341 	}
1342 	return 0;
1343 }
1344 
1345 static int
1346 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1347 {
1348 	struct blkio_cgroup *blkcg;
1349 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1350 	int name = BLKIOFILE_ATTR(cft->private);
1351 
1352 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1353 
1354 	switch(plid) {
1355 	case BLKIO_POLICY_PROP:
1356 		switch(name) {
1357 		case BLKIO_PROP_weight:
1358 			return blkio_weight_write(blkcg, val);
1359 		}
1360 		break;
1361 	default:
1362 		BUG();
1363 	}
1364 
1365 	return 0;
1366 }
1367 
1368 struct cftype blkio_files[] = {
1369 	{
1370 		.name = "weight_device",
1371 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1372 				BLKIO_PROP_weight_device),
1373 		.read_seq_string = blkiocg_file_read,
1374 		.write_string = blkiocg_file_write,
1375 		.max_write_len = 256,
1376 	},
1377 	{
1378 		.name = "weight",
1379 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1380 				BLKIO_PROP_weight),
1381 		.read_u64 = blkiocg_file_read_u64,
1382 		.write_u64 = blkiocg_file_write_u64,
1383 	},
1384 	{
1385 		.name = "time",
1386 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1387 				BLKIO_PROP_time),
1388 		.read_map = blkiocg_file_read_map,
1389 	},
1390 	{
1391 		.name = "sectors",
1392 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1393 				BLKIO_PROP_sectors),
1394 		.read_map = blkiocg_file_read_map,
1395 	},
1396 	{
1397 		.name = "io_service_bytes",
1398 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1399 				BLKIO_PROP_io_service_bytes),
1400 		.read_map = blkiocg_file_read_map,
1401 	},
1402 	{
1403 		.name = "io_serviced",
1404 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1405 				BLKIO_PROP_io_serviced),
1406 		.read_map = blkiocg_file_read_map,
1407 	},
1408 	{
1409 		.name = "io_service_time",
1410 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1411 				BLKIO_PROP_io_service_time),
1412 		.read_map = blkiocg_file_read_map,
1413 	},
1414 	{
1415 		.name = "io_wait_time",
1416 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1417 				BLKIO_PROP_io_wait_time),
1418 		.read_map = blkiocg_file_read_map,
1419 	},
1420 	{
1421 		.name = "io_merged",
1422 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1423 				BLKIO_PROP_io_merged),
1424 		.read_map = blkiocg_file_read_map,
1425 	},
1426 	{
1427 		.name = "io_queued",
1428 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1429 				BLKIO_PROP_io_queued),
1430 		.read_map = blkiocg_file_read_map,
1431 	},
1432 	{
1433 		.name = "reset_stats",
1434 		.write_u64 = blkiocg_reset_stats,
1435 	},
1436 #ifdef CONFIG_BLK_DEV_THROTTLING
1437 	{
1438 		.name = "throttle.read_bps_device",
1439 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1440 				BLKIO_THROTL_read_bps_device),
1441 		.read_seq_string = blkiocg_file_read,
1442 		.write_string = blkiocg_file_write,
1443 		.max_write_len = 256,
1444 	},
1445 
1446 	{
1447 		.name = "throttle.write_bps_device",
1448 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1449 				BLKIO_THROTL_write_bps_device),
1450 		.read_seq_string = blkiocg_file_read,
1451 		.write_string = blkiocg_file_write,
1452 		.max_write_len = 256,
1453 	},
1454 
1455 	{
1456 		.name = "throttle.read_iops_device",
1457 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1458 				BLKIO_THROTL_read_iops_device),
1459 		.read_seq_string = blkiocg_file_read,
1460 		.write_string = blkiocg_file_write,
1461 		.max_write_len = 256,
1462 	},
1463 
1464 	{
1465 		.name = "throttle.write_iops_device",
1466 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1467 				BLKIO_THROTL_write_iops_device),
1468 		.read_seq_string = blkiocg_file_read,
1469 		.write_string = blkiocg_file_write,
1470 		.max_write_len = 256,
1471 	},
1472 	{
1473 		.name = "throttle.io_service_bytes",
1474 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1475 				BLKIO_THROTL_io_service_bytes),
1476 		.read_map = blkiocg_file_read_map,
1477 	},
1478 	{
1479 		.name = "throttle.io_serviced",
1480 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1481 				BLKIO_THROTL_io_serviced),
1482 		.read_map = blkiocg_file_read_map,
1483 	},
1484 #endif /* CONFIG_BLK_DEV_THROTTLING */
1485 
1486 #ifdef CONFIG_DEBUG_BLK_CGROUP
1487 	{
1488 		.name = "avg_queue_size",
1489 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1490 				BLKIO_PROP_avg_queue_size),
1491 		.read_map = blkiocg_file_read_map,
1492 	},
1493 	{
1494 		.name = "group_wait_time",
1495 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1496 				BLKIO_PROP_group_wait_time),
1497 		.read_map = blkiocg_file_read_map,
1498 	},
1499 	{
1500 		.name = "idle_time",
1501 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1502 				BLKIO_PROP_idle_time),
1503 		.read_map = blkiocg_file_read_map,
1504 	},
1505 	{
1506 		.name = "empty_time",
1507 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1508 				BLKIO_PROP_empty_time),
1509 		.read_map = blkiocg_file_read_map,
1510 	},
1511 	{
1512 		.name = "dequeue",
1513 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1514 				BLKIO_PROP_dequeue),
1515 		.read_map = blkiocg_file_read_map,
1516 	},
1517 	{
1518 		.name = "unaccounted_time",
1519 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1520 				BLKIO_PROP_unaccounted_time),
1521 		.read_map = blkiocg_file_read_map,
1522 	},
1523 #endif
1524 };
1525 
1526 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1527 {
1528 	return cgroup_add_files(cgroup, subsys, blkio_files,
1529 				ARRAY_SIZE(blkio_files));
1530 }
1531 
1532 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1533 {
1534 	struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1535 	unsigned long flags;
1536 	struct blkio_group *blkg;
1537 	void *key;
1538 	struct blkio_policy_type *blkiop;
1539 	struct blkio_policy_node *pn, *pntmp;
1540 
1541 	rcu_read_lock();
1542 	do {
1543 		spin_lock_irqsave(&blkcg->lock, flags);
1544 
1545 		if (hlist_empty(&blkcg->blkg_list)) {
1546 			spin_unlock_irqrestore(&blkcg->lock, flags);
1547 			break;
1548 		}
1549 
1550 		blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1551 					blkcg_node);
1552 		key = rcu_dereference(blkg->key);
1553 		__blkiocg_del_blkio_group(blkg);
1554 
1555 		spin_unlock_irqrestore(&blkcg->lock, flags);
1556 
1557 		/*
1558 		 * This blkio_group is being unlinked as associated cgroup is
1559 		 * going away. Let all the IO controlling policies know about
1560 		 * this event.
1561 		 */
1562 		spin_lock(&blkio_list_lock);
1563 		list_for_each_entry(blkiop, &blkio_list, list) {
1564 			if (blkiop->plid != blkg->plid)
1565 				continue;
1566 			blkiop->ops.blkio_unlink_group_fn(key, blkg);
1567 		}
1568 		spin_unlock(&blkio_list_lock);
1569 	} while (1);
1570 
1571 	list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1572 		blkio_policy_delete_node(pn);
1573 		kfree(pn);
1574 	}
1575 
1576 	free_css_id(&blkio_subsys, &blkcg->css);
1577 	rcu_read_unlock();
1578 	if (blkcg != &blkio_root_cgroup)
1579 		kfree(blkcg);
1580 }
1581 
1582 static struct cgroup_subsys_state *
1583 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1584 {
1585 	struct blkio_cgroup *blkcg;
1586 	struct cgroup *parent = cgroup->parent;
1587 
1588 	if (!parent) {
1589 		blkcg = &blkio_root_cgroup;
1590 		goto done;
1591 	}
1592 
1593 	blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1594 	if (!blkcg)
1595 		return ERR_PTR(-ENOMEM);
1596 
1597 	blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1598 done:
1599 	spin_lock_init(&blkcg->lock);
1600 	INIT_HLIST_HEAD(&blkcg->blkg_list);
1601 
1602 	INIT_LIST_HEAD(&blkcg->policy_list);
1603 	return &blkcg->css;
1604 }
1605 
1606 /*
1607  * We cannot support shared io contexts, as we have no mean to support
1608  * two tasks with the same ioc in two different groups without major rework
1609  * of the main cic data structures.  For now we allow a task to change
1610  * its cgroup only if it's the only owner of its ioc.
1611  */
1612 static int blkiocg_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1613 {
1614 	struct io_context *ioc;
1615 	int ret = 0;
1616 
1617 	/* task_lock() is needed to avoid races with exit_io_context() */
1618 	task_lock(tsk);
1619 	ioc = tsk->io_context;
1620 	if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1621 		ret = -EINVAL;
1622 	task_unlock(tsk);
1623 
1624 	return ret;
1625 }
1626 
1627 static void blkiocg_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1628 {
1629 	struct io_context *ioc;
1630 
1631 	task_lock(tsk);
1632 	ioc = tsk->io_context;
1633 	if (ioc)
1634 		ioc->cgroup_changed = 1;
1635 	task_unlock(tsk);
1636 }
1637 
1638 void blkio_policy_register(struct blkio_policy_type *blkiop)
1639 {
1640 	spin_lock(&blkio_list_lock);
1641 	list_add_tail(&blkiop->list, &blkio_list);
1642 	spin_unlock(&blkio_list_lock);
1643 }
1644 EXPORT_SYMBOL_GPL(blkio_policy_register);
1645 
1646 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1647 {
1648 	spin_lock(&blkio_list_lock);
1649 	list_del_init(&blkiop->list);
1650 	spin_unlock(&blkio_list_lock);
1651 }
1652 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1653 
1654 static int __init init_cgroup_blkio(void)
1655 {
1656 	return cgroup_load_subsys(&blkio_subsys);
1657 }
1658 
1659 static void __exit exit_cgroup_blkio(void)
1660 {
1661 	cgroup_unload_subsys(&blkio_subsys);
1662 }
1663 
1664 module_init(init_cgroup_blkio);
1665 module_exit(exit_cgroup_blkio);
1666 MODULE_LICENSE("GPL");
1667