xref: /linux/block/blk-cgroup.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
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(struct cgroup_subsys *, struct cgroup *,
34 			      struct cgroup_taskset *);
35 static void blkiocg_attach(struct cgroup_subsys *, struct cgroup *,
36 			   struct cgroup_taskset *);
37 static void blkiocg_destroy(struct cgroup_subsys *, struct cgroup *);
38 static int blkiocg_populate(struct cgroup_subsys *, struct cgroup *);
39 
40 /* for encoding cft->private value on file */
41 #define BLKIOFILE_PRIVATE(x, val)	(((x) << 16) | (val))
42 /* What policy owns the file, proportional or throttle */
43 #define BLKIOFILE_POLICY(val)		(((val) >> 16) & 0xffff)
44 #define BLKIOFILE_ATTR(val)		((val) & 0xffff)
45 
46 struct cgroup_subsys blkio_subsys = {
47 	.name = "blkio",
48 	.create = blkiocg_create,
49 	.can_attach = blkiocg_can_attach,
50 	.attach = blkiocg_attach,
51 	.destroy = blkiocg_destroy,
52 	.populate = blkiocg_populate,
53 #ifdef CONFIG_BLK_CGROUP
54 	/* note: blkio_subsys_id is otherwise defined in blk-cgroup.h */
55 	.subsys_id = blkio_subsys_id,
56 #endif
57 	.use_id = 1,
58 	.module = THIS_MODULE,
59 };
60 EXPORT_SYMBOL_GPL(blkio_subsys);
61 
62 static inline void blkio_policy_insert_node(struct blkio_cgroup *blkcg,
63 					    struct blkio_policy_node *pn)
64 {
65 	list_add(&pn->node, &blkcg->policy_list);
66 }
67 
68 static inline bool cftype_blkg_same_policy(struct cftype *cft,
69 			struct blkio_group *blkg)
70 {
71 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
72 
73 	if (blkg->plid == plid)
74 		return 1;
75 
76 	return 0;
77 }
78 
79 /* Determines if policy node matches cgroup file being accessed */
80 static inline bool pn_matches_cftype(struct cftype *cft,
81 			struct blkio_policy_node *pn)
82 {
83 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
84 	int fileid = BLKIOFILE_ATTR(cft->private);
85 
86 	return (plid == pn->plid && fileid == pn->fileid);
87 }
88 
89 /* Must be called with blkcg->lock held */
90 static inline void blkio_policy_delete_node(struct blkio_policy_node *pn)
91 {
92 	list_del(&pn->node);
93 }
94 
95 /* Must be called with blkcg->lock held */
96 static struct blkio_policy_node *
97 blkio_policy_search_node(const struct blkio_cgroup *blkcg, dev_t dev,
98 		enum blkio_policy_id plid, int fileid)
99 {
100 	struct blkio_policy_node *pn;
101 
102 	list_for_each_entry(pn, &blkcg->policy_list, node) {
103 		if (pn->dev == dev && pn->plid == plid && pn->fileid == fileid)
104 			return pn;
105 	}
106 
107 	return NULL;
108 }
109 
110 struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup)
111 {
112 	return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
113 			    struct blkio_cgroup, css);
114 }
115 EXPORT_SYMBOL_GPL(cgroup_to_blkio_cgroup);
116 
117 struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk)
118 {
119 	return container_of(task_subsys_state(tsk, blkio_subsys_id),
120 			    struct blkio_cgroup, css);
121 }
122 EXPORT_SYMBOL_GPL(task_blkio_cgroup);
123 
124 static inline void
125 blkio_update_group_weight(struct blkio_group *blkg, unsigned int weight)
126 {
127 	struct blkio_policy_type *blkiop;
128 
129 	list_for_each_entry(blkiop, &blkio_list, list) {
130 		/* If this policy does not own the blkg, do not send updates */
131 		if (blkiop->plid != blkg->plid)
132 			continue;
133 		if (blkiop->ops.blkio_update_group_weight_fn)
134 			blkiop->ops.blkio_update_group_weight_fn(blkg->key,
135 							blkg, weight);
136 	}
137 }
138 
139 static inline void blkio_update_group_bps(struct blkio_group *blkg, u64 bps,
140 				int fileid)
141 {
142 	struct blkio_policy_type *blkiop;
143 
144 	list_for_each_entry(blkiop, &blkio_list, list) {
145 
146 		/* If this policy does not own the blkg, do not send updates */
147 		if (blkiop->plid != blkg->plid)
148 			continue;
149 
150 		if (fileid == BLKIO_THROTL_read_bps_device
151 		    && blkiop->ops.blkio_update_group_read_bps_fn)
152 			blkiop->ops.blkio_update_group_read_bps_fn(blkg->key,
153 								blkg, bps);
154 
155 		if (fileid == BLKIO_THROTL_write_bps_device
156 		    && blkiop->ops.blkio_update_group_write_bps_fn)
157 			blkiop->ops.blkio_update_group_write_bps_fn(blkg->key,
158 								blkg, bps);
159 	}
160 }
161 
162 static inline void blkio_update_group_iops(struct blkio_group *blkg,
163 			unsigned int iops, int fileid)
164 {
165 	struct blkio_policy_type *blkiop;
166 
167 	list_for_each_entry(blkiop, &blkio_list, list) {
168 
169 		/* If this policy does not own the blkg, do not send updates */
170 		if (blkiop->plid != blkg->plid)
171 			continue;
172 
173 		if (fileid == BLKIO_THROTL_read_iops_device
174 		    && blkiop->ops.blkio_update_group_read_iops_fn)
175 			blkiop->ops.blkio_update_group_read_iops_fn(blkg->key,
176 								blkg, iops);
177 
178 		if (fileid == BLKIO_THROTL_write_iops_device
179 		    && blkiop->ops.blkio_update_group_write_iops_fn)
180 			blkiop->ops.blkio_update_group_write_iops_fn(blkg->key,
181 								blkg,iops);
182 	}
183 }
184 
185 /*
186  * Add to the appropriate stat variable depending on the request type.
187  * This should be called with the blkg->stats_lock held.
188  */
189 static void blkio_add_stat(uint64_t *stat, uint64_t add, bool direction,
190 				bool sync)
191 {
192 	if (direction)
193 		stat[BLKIO_STAT_WRITE] += add;
194 	else
195 		stat[BLKIO_STAT_READ] += add;
196 	if (sync)
197 		stat[BLKIO_STAT_SYNC] += add;
198 	else
199 		stat[BLKIO_STAT_ASYNC] += add;
200 }
201 
202 /*
203  * Decrements the appropriate stat variable if non-zero depending on the
204  * request type. Panics on value being zero.
205  * This should be called with the blkg->stats_lock held.
206  */
207 static void blkio_check_and_dec_stat(uint64_t *stat, bool direction, bool sync)
208 {
209 	if (direction) {
210 		BUG_ON(stat[BLKIO_STAT_WRITE] == 0);
211 		stat[BLKIO_STAT_WRITE]--;
212 	} else {
213 		BUG_ON(stat[BLKIO_STAT_READ] == 0);
214 		stat[BLKIO_STAT_READ]--;
215 	}
216 	if (sync) {
217 		BUG_ON(stat[BLKIO_STAT_SYNC] == 0);
218 		stat[BLKIO_STAT_SYNC]--;
219 	} else {
220 		BUG_ON(stat[BLKIO_STAT_ASYNC] == 0);
221 		stat[BLKIO_STAT_ASYNC]--;
222 	}
223 }
224 
225 #ifdef CONFIG_DEBUG_BLK_CGROUP
226 /* This should be called with the blkg->stats_lock held. */
227 static void blkio_set_start_group_wait_time(struct blkio_group *blkg,
228 						struct blkio_group *curr_blkg)
229 {
230 	if (blkio_blkg_waiting(&blkg->stats))
231 		return;
232 	if (blkg == curr_blkg)
233 		return;
234 	blkg->stats.start_group_wait_time = sched_clock();
235 	blkio_mark_blkg_waiting(&blkg->stats);
236 }
237 
238 /* This should be called with the blkg->stats_lock held. */
239 static void blkio_update_group_wait_time(struct blkio_group_stats *stats)
240 {
241 	unsigned long long now;
242 
243 	if (!blkio_blkg_waiting(stats))
244 		return;
245 
246 	now = sched_clock();
247 	if (time_after64(now, stats->start_group_wait_time))
248 		stats->group_wait_time += now - stats->start_group_wait_time;
249 	blkio_clear_blkg_waiting(stats);
250 }
251 
252 /* This should be called with the blkg->stats_lock held. */
253 static void blkio_end_empty_time(struct blkio_group_stats *stats)
254 {
255 	unsigned long long now;
256 
257 	if (!blkio_blkg_empty(stats))
258 		return;
259 
260 	now = sched_clock();
261 	if (time_after64(now, stats->start_empty_time))
262 		stats->empty_time += now - stats->start_empty_time;
263 	blkio_clear_blkg_empty(stats);
264 }
265 
266 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
267 {
268 	unsigned long flags;
269 
270 	spin_lock_irqsave(&blkg->stats_lock, flags);
271 	BUG_ON(blkio_blkg_idling(&blkg->stats));
272 	blkg->stats.start_idle_time = sched_clock();
273 	blkio_mark_blkg_idling(&blkg->stats);
274 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
275 }
276 EXPORT_SYMBOL_GPL(blkiocg_update_set_idle_time_stats);
277 
278 void blkiocg_update_idle_time_stats(struct blkio_group *blkg)
279 {
280 	unsigned long flags;
281 	unsigned long long now;
282 	struct blkio_group_stats *stats;
283 
284 	spin_lock_irqsave(&blkg->stats_lock, flags);
285 	stats = &blkg->stats;
286 	if (blkio_blkg_idling(stats)) {
287 		now = sched_clock();
288 		if (time_after64(now, stats->start_idle_time))
289 			stats->idle_time += now - stats->start_idle_time;
290 		blkio_clear_blkg_idling(stats);
291 	}
292 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
293 }
294 EXPORT_SYMBOL_GPL(blkiocg_update_idle_time_stats);
295 
296 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg)
297 {
298 	unsigned long flags;
299 	struct blkio_group_stats *stats;
300 
301 	spin_lock_irqsave(&blkg->stats_lock, flags);
302 	stats = &blkg->stats;
303 	stats->avg_queue_size_sum +=
304 			stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] +
305 			stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE];
306 	stats->avg_queue_size_samples++;
307 	blkio_update_group_wait_time(stats);
308 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
309 }
310 EXPORT_SYMBOL_GPL(blkiocg_update_avg_queue_size_stats);
311 
312 void blkiocg_set_start_empty_time(struct blkio_group *blkg)
313 {
314 	unsigned long flags;
315 	struct blkio_group_stats *stats;
316 
317 	spin_lock_irqsave(&blkg->stats_lock, flags);
318 	stats = &blkg->stats;
319 
320 	if (stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_READ] ||
321 			stats->stat_arr[BLKIO_STAT_QUEUED][BLKIO_STAT_WRITE]) {
322 		spin_unlock_irqrestore(&blkg->stats_lock, flags);
323 		return;
324 	}
325 
326 	/*
327 	 * group is already marked empty. This can happen if cfqq got new
328 	 * request in parent group and moved to this group while being added
329 	 * to service tree. Just ignore the event and move on.
330 	 */
331 	if(blkio_blkg_empty(stats)) {
332 		spin_unlock_irqrestore(&blkg->stats_lock, flags);
333 		return;
334 	}
335 
336 	stats->start_empty_time = sched_clock();
337 	blkio_mark_blkg_empty(stats);
338 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
339 }
340 EXPORT_SYMBOL_GPL(blkiocg_set_start_empty_time);
341 
342 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
343 			unsigned long dequeue)
344 {
345 	blkg->stats.dequeue += dequeue;
346 }
347 EXPORT_SYMBOL_GPL(blkiocg_update_dequeue_stats);
348 #else
349 static inline void blkio_set_start_group_wait_time(struct blkio_group *blkg,
350 					struct blkio_group *curr_blkg) {}
351 static inline void blkio_end_empty_time(struct blkio_group_stats *stats) {}
352 #endif
353 
354 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
355 			struct blkio_group *curr_blkg, bool direction,
356 			bool sync)
357 {
358 	unsigned long flags;
359 
360 	spin_lock_irqsave(&blkg->stats_lock, flags);
361 	blkio_add_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED], 1, direction,
362 			sync);
363 	blkio_end_empty_time(&blkg->stats);
364 	blkio_set_start_group_wait_time(blkg, curr_blkg);
365 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
366 }
367 EXPORT_SYMBOL_GPL(blkiocg_update_io_add_stats);
368 
369 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
370 						bool direction, bool sync)
371 {
372 	unsigned long flags;
373 
374 	spin_lock_irqsave(&blkg->stats_lock, flags);
375 	blkio_check_and_dec_stat(blkg->stats.stat_arr[BLKIO_STAT_QUEUED],
376 					direction, sync);
377 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
378 }
379 EXPORT_SYMBOL_GPL(blkiocg_update_io_remove_stats);
380 
381 void blkiocg_update_timeslice_used(struct blkio_group *blkg, unsigned long time,
382 				unsigned long unaccounted_time)
383 {
384 	unsigned long flags;
385 
386 	spin_lock_irqsave(&blkg->stats_lock, flags);
387 	blkg->stats.time += time;
388 #ifdef CONFIG_DEBUG_BLK_CGROUP
389 	blkg->stats.unaccounted_time += unaccounted_time;
390 #endif
391 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
392 }
393 EXPORT_SYMBOL_GPL(blkiocg_update_timeslice_used);
394 
395 /*
396  * should be called under rcu read lock or queue lock to make sure blkg pointer
397  * is valid.
398  */
399 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
400 				uint64_t bytes, bool direction, bool sync)
401 {
402 	struct blkio_group_stats_cpu *stats_cpu;
403 	unsigned long flags;
404 
405 	/*
406 	 * Disabling interrupts to provide mutual exclusion between two
407 	 * writes on same cpu. It probably is not needed for 64bit. Not
408 	 * optimizing that case yet.
409 	 */
410 	local_irq_save(flags);
411 
412 	stats_cpu = this_cpu_ptr(blkg->stats_cpu);
413 
414 	u64_stats_update_begin(&stats_cpu->syncp);
415 	stats_cpu->sectors += bytes >> 9;
416 	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICED],
417 			1, direction, sync);
418 	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_SERVICE_BYTES],
419 			bytes, direction, sync);
420 	u64_stats_update_end(&stats_cpu->syncp);
421 	local_irq_restore(flags);
422 }
423 EXPORT_SYMBOL_GPL(blkiocg_update_dispatch_stats);
424 
425 void blkiocg_update_completion_stats(struct blkio_group *blkg,
426 	uint64_t start_time, uint64_t io_start_time, bool direction, bool sync)
427 {
428 	struct blkio_group_stats *stats;
429 	unsigned long flags;
430 	unsigned long long now = sched_clock();
431 
432 	spin_lock_irqsave(&blkg->stats_lock, flags);
433 	stats = &blkg->stats;
434 	if (time_after64(now, io_start_time))
435 		blkio_add_stat(stats->stat_arr[BLKIO_STAT_SERVICE_TIME],
436 				now - io_start_time, direction, sync);
437 	if (time_after64(io_start_time, start_time))
438 		blkio_add_stat(stats->stat_arr[BLKIO_STAT_WAIT_TIME],
439 				io_start_time - start_time, direction, sync);
440 	spin_unlock_irqrestore(&blkg->stats_lock, flags);
441 }
442 EXPORT_SYMBOL_GPL(blkiocg_update_completion_stats);
443 
444 /*  Merged stats are per cpu.  */
445 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
446 					bool sync)
447 {
448 	struct blkio_group_stats_cpu *stats_cpu;
449 	unsigned long flags;
450 
451 	/*
452 	 * Disabling interrupts to provide mutual exclusion between two
453 	 * writes on same cpu. It probably is not needed for 64bit. Not
454 	 * optimizing that case yet.
455 	 */
456 	local_irq_save(flags);
457 
458 	stats_cpu = this_cpu_ptr(blkg->stats_cpu);
459 
460 	u64_stats_update_begin(&stats_cpu->syncp);
461 	blkio_add_stat(stats_cpu->stat_arr_cpu[BLKIO_STAT_CPU_MERGED], 1,
462 				direction, sync);
463 	u64_stats_update_end(&stats_cpu->syncp);
464 	local_irq_restore(flags);
465 }
466 EXPORT_SYMBOL_GPL(blkiocg_update_io_merged_stats);
467 
468 /*
469  * This function allocates the per cpu stats for blkio_group. Should be called
470  * from sleepable context as alloc_per_cpu() requires that.
471  */
472 int blkio_alloc_blkg_stats(struct blkio_group *blkg)
473 {
474 	/* Allocate memory for per cpu stats */
475 	blkg->stats_cpu = alloc_percpu(struct blkio_group_stats_cpu);
476 	if (!blkg->stats_cpu)
477 		return -ENOMEM;
478 	return 0;
479 }
480 EXPORT_SYMBOL_GPL(blkio_alloc_blkg_stats);
481 
482 void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
483 		struct blkio_group *blkg, void *key, dev_t dev,
484 		enum blkio_policy_id plid)
485 {
486 	unsigned long flags;
487 
488 	spin_lock_irqsave(&blkcg->lock, flags);
489 	spin_lock_init(&blkg->stats_lock);
490 	rcu_assign_pointer(blkg->key, key);
491 	blkg->blkcg_id = css_id(&blkcg->css);
492 	hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
493 	blkg->plid = plid;
494 	spin_unlock_irqrestore(&blkcg->lock, flags);
495 	/* Need to take css reference ? */
496 	cgroup_path(blkcg->css.cgroup, blkg->path, sizeof(blkg->path));
497 	blkg->dev = dev;
498 }
499 EXPORT_SYMBOL_GPL(blkiocg_add_blkio_group);
500 
501 static void __blkiocg_del_blkio_group(struct blkio_group *blkg)
502 {
503 	hlist_del_init_rcu(&blkg->blkcg_node);
504 	blkg->blkcg_id = 0;
505 }
506 
507 /*
508  * returns 0 if blkio_group was still on cgroup list. Otherwise returns 1
509  * indicating that blk_group was unhashed by the time we got to it.
510  */
511 int blkiocg_del_blkio_group(struct blkio_group *blkg)
512 {
513 	struct blkio_cgroup *blkcg;
514 	unsigned long flags;
515 	struct cgroup_subsys_state *css;
516 	int ret = 1;
517 
518 	rcu_read_lock();
519 	css = css_lookup(&blkio_subsys, blkg->blkcg_id);
520 	if (css) {
521 		blkcg = container_of(css, struct blkio_cgroup, css);
522 		spin_lock_irqsave(&blkcg->lock, flags);
523 		if (!hlist_unhashed(&blkg->blkcg_node)) {
524 			__blkiocg_del_blkio_group(blkg);
525 			ret = 0;
526 		}
527 		spin_unlock_irqrestore(&blkcg->lock, flags);
528 	}
529 
530 	rcu_read_unlock();
531 	return ret;
532 }
533 EXPORT_SYMBOL_GPL(blkiocg_del_blkio_group);
534 
535 /* called under rcu_read_lock(). */
536 struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key)
537 {
538 	struct blkio_group *blkg;
539 	struct hlist_node *n;
540 	void *__key;
541 
542 	hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
543 		__key = blkg->key;
544 		if (__key == key)
545 			return blkg;
546 	}
547 
548 	return NULL;
549 }
550 EXPORT_SYMBOL_GPL(blkiocg_lookup_group);
551 
552 static void blkio_reset_stats_cpu(struct blkio_group *blkg)
553 {
554 	struct blkio_group_stats_cpu *stats_cpu;
555 	int i, j, k;
556 	/*
557 	 * Note: On 64 bit arch this should not be an issue. This has the
558 	 * possibility of returning some inconsistent value on 32bit arch
559 	 * as 64bit update on 32bit is non atomic. Taking care of this
560 	 * corner case makes code very complicated, like sending IPIs to
561 	 * cpus, taking care of stats of offline cpus etc.
562 	 *
563 	 * reset stats is anyway more of a debug feature and this sounds a
564 	 * corner case. So I am not complicating the code yet until and
565 	 * unless this becomes a real issue.
566 	 */
567 	for_each_possible_cpu(i) {
568 		stats_cpu = per_cpu_ptr(blkg->stats_cpu, i);
569 		stats_cpu->sectors = 0;
570 		for(j = 0; j < BLKIO_STAT_CPU_NR; j++)
571 			for (k = 0; k < BLKIO_STAT_TOTAL; k++)
572 				stats_cpu->stat_arr_cpu[j][k] = 0;
573 	}
574 }
575 
576 static int
577 blkiocg_reset_stats(struct cgroup *cgroup, struct cftype *cftype, u64 val)
578 {
579 	struct blkio_cgroup *blkcg;
580 	struct blkio_group *blkg;
581 	struct blkio_group_stats *stats;
582 	struct hlist_node *n;
583 	uint64_t queued[BLKIO_STAT_TOTAL];
584 	int i;
585 #ifdef CONFIG_DEBUG_BLK_CGROUP
586 	bool idling, waiting, empty;
587 	unsigned long long now = sched_clock();
588 #endif
589 
590 	blkcg = cgroup_to_blkio_cgroup(cgroup);
591 	spin_lock_irq(&blkcg->lock);
592 	hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
593 		spin_lock(&blkg->stats_lock);
594 		stats = &blkg->stats;
595 #ifdef CONFIG_DEBUG_BLK_CGROUP
596 		idling = blkio_blkg_idling(stats);
597 		waiting = blkio_blkg_waiting(stats);
598 		empty = blkio_blkg_empty(stats);
599 #endif
600 		for (i = 0; i < BLKIO_STAT_TOTAL; i++)
601 			queued[i] = stats->stat_arr[BLKIO_STAT_QUEUED][i];
602 		memset(stats, 0, sizeof(struct blkio_group_stats));
603 		for (i = 0; i < BLKIO_STAT_TOTAL; i++)
604 			stats->stat_arr[BLKIO_STAT_QUEUED][i] = queued[i];
605 #ifdef CONFIG_DEBUG_BLK_CGROUP
606 		if (idling) {
607 			blkio_mark_blkg_idling(stats);
608 			stats->start_idle_time = now;
609 		}
610 		if (waiting) {
611 			blkio_mark_blkg_waiting(stats);
612 			stats->start_group_wait_time = now;
613 		}
614 		if (empty) {
615 			blkio_mark_blkg_empty(stats);
616 			stats->start_empty_time = now;
617 		}
618 #endif
619 		spin_unlock(&blkg->stats_lock);
620 
621 		/* Reset Per cpu stats which don't take blkg->stats_lock */
622 		blkio_reset_stats_cpu(blkg);
623 	}
624 
625 	spin_unlock_irq(&blkcg->lock);
626 	return 0;
627 }
628 
629 static void blkio_get_key_name(enum stat_sub_type type, dev_t dev, char *str,
630 				int chars_left, bool diskname_only)
631 {
632 	snprintf(str, chars_left, "%d:%d", MAJOR(dev), MINOR(dev));
633 	chars_left -= strlen(str);
634 	if (chars_left <= 0) {
635 		printk(KERN_WARNING
636 			"Possibly incorrect cgroup stat display format");
637 		return;
638 	}
639 	if (diskname_only)
640 		return;
641 	switch (type) {
642 	case BLKIO_STAT_READ:
643 		strlcat(str, " Read", chars_left);
644 		break;
645 	case BLKIO_STAT_WRITE:
646 		strlcat(str, " Write", chars_left);
647 		break;
648 	case BLKIO_STAT_SYNC:
649 		strlcat(str, " Sync", chars_left);
650 		break;
651 	case BLKIO_STAT_ASYNC:
652 		strlcat(str, " Async", chars_left);
653 		break;
654 	case BLKIO_STAT_TOTAL:
655 		strlcat(str, " Total", chars_left);
656 		break;
657 	default:
658 		strlcat(str, " Invalid", chars_left);
659 	}
660 }
661 
662 static uint64_t blkio_fill_stat(char *str, int chars_left, uint64_t val,
663 				struct cgroup_map_cb *cb, dev_t dev)
664 {
665 	blkio_get_key_name(0, dev, str, chars_left, true);
666 	cb->fill(cb, str, val);
667 	return val;
668 }
669 
670 
671 static uint64_t blkio_read_stat_cpu(struct blkio_group *blkg,
672 			enum stat_type_cpu type, enum stat_sub_type sub_type)
673 {
674 	int cpu;
675 	struct blkio_group_stats_cpu *stats_cpu;
676 	u64 val = 0, tval;
677 
678 	for_each_possible_cpu(cpu) {
679 		unsigned int start;
680 		stats_cpu  = per_cpu_ptr(blkg->stats_cpu, cpu);
681 
682 		do {
683 			start = u64_stats_fetch_begin(&stats_cpu->syncp);
684 			if (type == BLKIO_STAT_CPU_SECTORS)
685 				tval = stats_cpu->sectors;
686 			else
687 				tval = stats_cpu->stat_arr_cpu[type][sub_type];
688 		} while(u64_stats_fetch_retry(&stats_cpu->syncp, start));
689 
690 		val += tval;
691 	}
692 
693 	return val;
694 }
695 
696 static uint64_t blkio_get_stat_cpu(struct blkio_group *blkg,
697 		struct cgroup_map_cb *cb, dev_t dev, enum stat_type_cpu type)
698 {
699 	uint64_t disk_total, val;
700 	char key_str[MAX_KEY_LEN];
701 	enum stat_sub_type sub_type;
702 
703 	if (type == BLKIO_STAT_CPU_SECTORS) {
704 		val = blkio_read_stat_cpu(blkg, type, 0);
705 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, val, cb, dev);
706 	}
707 
708 	for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
709 			sub_type++) {
710 		blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
711 		val = blkio_read_stat_cpu(blkg, type, sub_type);
712 		cb->fill(cb, key_str, val);
713 	}
714 
715 	disk_total = blkio_read_stat_cpu(blkg, type, BLKIO_STAT_READ) +
716 			blkio_read_stat_cpu(blkg, type, BLKIO_STAT_WRITE);
717 
718 	blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
719 	cb->fill(cb, key_str, disk_total);
720 	return disk_total;
721 }
722 
723 /* This should be called with blkg->stats_lock held */
724 static uint64_t blkio_get_stat(struct blkio_group *blkg,
725 		struct cgroup_map_cb *cb, dev_t dev, enum stat_type type)
726 {
727 	uint64_t disk_total;
728 	char key_str[MAX_KEY_LEN];
729 	enum stat_sub_type sub_type;
730 
731 	if (type == BLKIO_STAT_TIME)
732 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
733 					blkg->stats.time, cb, dev);
734 #ifdef CONFIG_DEBUG_BLK_CGROUP
735 	if (type == BLKIO_STAT_UNACCOUNTED_TIME)
736 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
737 					blkg->stats.unaccounted_time, cb, dev);
738 	if (type == BLKIO_STAT_AVG_QUEUE_SIZE) {
739 		uint64_t sum = blkg->stats.avg_queue_size_sum;
740 		uint64_t samples = blkg->stats.avg_queue_size_samples;
741 		if (samples)
742 			do_div(sum, samples);
743 		else
744 			sum = 0;
745 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1, sum, cb, dev);
746 	}
747 	if (type == BLKIO_STAT_GROUP_WAIT_TIME)
748 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
749 					blkg->stats.group_wait_time, cb, dev);
750 	if (type == BLKIO_STAT_IDLE_TIME)
751 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
752 					blkg->stats.idle_time, cb, dev);
753 	if (type == BLKIO_STAT_EMPTY_TIME)
754 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
755 					blkg->stats.empty_time, cb, dev);
756 	if (type == BLKIO_STAT_DEQUEUE)
757 		return blkio_fill_stat(key_str, MAX_KEY_LEN - 1,
758 					blkg->stats.dequeue, cb, dev);
759 #endif
760 
761 	for (sub_type = BLKIO_STAT_READ; sub_type < BLKIO_STAT_TOTAL;
762 			sub_type++) {
763 		blkio_get_key_name(sub_type, dev, key_str, MAX_KEY_LEN, false);
764 		cb->fill(cb, key_str, blkg->stats.stat_arr[type][sub_type]);
765 	}
766 	disk_total = blkg->stats.stat_arr[type][BLKIO_STAT_READ] +
767 			blkg->stats.stat_arr[type][BLKIO_STAT_WRITE];
768 	blkio_get_key_name(BLKIO_STAT_TOTAL, dev, key_str, MAX_KEY_LEN, false);
769 	cb->fill(cb, key_str, disk_total);
770 	return disk_total;
771 }
772 
773 static int blkio_policy_parse_and_set(char *buf,
774 	struct blkio_policy_node *newpn, enum blkio_policy_id plid, int fileid)
775 {
776 	struct gendisk *disk = NULL;
777 	char *s[4], *p, *major_s = NULL, *minor_s = NULL;
778 	unsigned long major, minor;
779 	int i = 0, ret = -EINVAL;
780 	int part;
781 	dev_t dev;
782 	u64 temp;
783 
784 	memset(s, 0, sizeof(s));
785 
786 	while ((p = strsep(&buf, " ")) != NULL) {
787 		if (!*p)
788 			continue;
789 
790 		s[i++] = p;
791 
792 		/* Prevent from inputing too many things */
793 		if (i == 3)
794 			break;
795 	}
796 
797 	if (i != 2)
798 		goto out;
799 
800 	p = strsep(&s[0], ":");
801 	if (p != NULL)
802 		major_s = p;
803 	else
804 		goto out;
805 
806 	minor_s = s[0];
807 	if (!minor_s)
808 		goto out;
809 
810 	if (strict_strtoul(major_s, 10, &major))
811 		goto out;
812 
813 	if (strict_strtoul(minor_s, 10, &minor))
814 		goto out;
815 
816 	dev = MKDEV(major, minor);
817 
818 	if (strict_strtoull(s[1], 10, &temp))
819 		goto out;
820 
821 	/* For rule removal, do not check for device presence. */
822 	if (temp) {
823 		disk = get_gendisk(dev, &part);
824 		if (!disk || part) {
825 			ret = -ENODEV;
826 			goto out;
827 		}
828 	}
829 
830 	newpn->dev = dev;
831 
832 	switch (plid) {
833 	case BLKIO_POLICY_PROP:
834 		if ((temp < BLKIO_WEIGHT_MIN && temp > 0) ||
835 		     temp > BLKIO_WEIGHT_MAX)
836 			goto out;
837 
838 		newpn->plid = plid;
839 		newpn->fileid = fileid;
840 		newpn->val.weight = temp;
841 		break;
842 	case BLKIO_POLICY_THROTL:
843 		switch(fileid) {
844 		case BLKIO_THROTL_read_bps_device:
845 		case BLKIO_THROTL_write_bps_device:
846 			newpn->plid = plid;
847 			newpn->fileid = fileid;
848 			newpn->val.bps = temp;
849 			break;
850 		case BLKIO_THROTL_read_iops_device:
851 		case BLKIO_THROTL_write_iops_device:
852 			if (temp > THROTL_IOPS_MAX)
853 				goto out;
854 
855 			newpn->plid = plid;
856 			newpn->fileid = fileid;
857 			newpn->val.iops = (unsigned int)temp;
858 			break;
859 		}
860 		break;
861 	default:
862 		BUG();
863 	}
864 	ret = 0;
865 out:
866 	put_disk(disk);
867 	return ret;
868 }
869 
870 unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
871 			      dev_t dev)
872 {
873 	struct blkio_policy_node *pn;
874 	unsigned long flags;
875 	unsigned int weight;
876 
877 	spin_lock_irqsave(&blkcg->lock, flags);
878 
879 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_PROP,
880 				BLKIO_PROP_weight_device);
881 	if (pn)
882 		weight = pn->val.weight;
883 	else
884 		weight = blkcg->weight;
885 
886 	spin_unlock_irqrestore(&blkcg->lock, flags);
887 
888 	return 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 	unsigned long flags;
896 	uint64_t bps = -1;
897 
898 	spin_lock_irqsave(&blkcg->lock, flags);
899 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
900 				BLKIO_THROTL_read_bps_device);
901 	if (pn)
902 		bps = pn->val.bps;
903 	spin_unlock_irqrestore(&blkcg->lock, flags);
904 
905 	return bps;
906 }
907 
908 uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, dev_t dev)
909 {
910 	struct blkio_policy_node *pn;
911 	unsigned long flags;
912 	uint64_t bps = -1;
913 
914 	spin_lock_irqsave(&blkcg->lock, flags);
915 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
916 				BLKIO_THROTL_write_bps_device);
917 	if (pn)
918 		bps = pn->val.bps;
919 	spin_unlock_irqrestore(&blkcg->lock, flags);
920 
921 	return bps;
922 }
923 
924 unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, dev_t dev)
925 {
926 	struct blkio_policy_node *pn;
927 	unsigned long flags;
928 	unsigned int iops = -1;
929 
930 	spin_lock_irqsave(&blkcg->lock, flags);
931 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
932 				BLKIO_THROTL_read_iops_device);
933 	if (pn)
934 		iops = pn->val.iops;
935 	spin_unlock_irqrestore(&blkcg->lock, flags);
936 
937 	return iops;
938 }
939 
940 unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, dev_t dev)
941 {
942 	struct blkio_policy_node *pn;
943 	unsigned long flags;
944 	unsigned int iops = -1;
945 
946 	spin_lock_irqsave(&blkcg->lock, flags);
947 	pn = blkio_policy_search_node(blkcg, dev, BLKIO_POLICY_THROTL,
948 				BLKIO_THROTL_write_iops_device);
949 	if (pn)
950 		iops = pn->val.iops;
951 	spin_unlock_irqrestore(&blkcg->lock, flags);
952 
953 	return iops;
954 }
955 
956 /* Checks whether user asked for deleting a policy rule */
957 static bool blkio_delete_rule_command(struct blkio_policy_node *pn)
958 {
959 	switch(pn->plid) {
960 	case BLKIO_POLICY_PROP:
961 		if (pn->val.weight == 0)
962 			return 1;
963 		break;
964 	case BLKIO_POLICY_THROTL:
965 		switch(pn->fileid) {
966 		case BLKIO_THROTL_read_bps_device:
967 		case BLKIO_THROTL_write_bps_device:
968 			if (pn->val.bps == 0)
969 				return 1;
970 			break;
971 		case BLKIO_THROTL_read_iops_device:
972 		case BLKIO_THROTL_write_iops_device:
973 			if (pn->val.iops == 0)
974 				return 1;
975 		}
976 		break;
977 	default:
978 		BUG();
979 	}
980 
981 	return 0;
982 }
983 
984 static void blkio_update_policy_rule(struct blkio_policy_node *oldpn,
985 					struct blkio_policy_node *newpn)
986 {
987 	switch(oldpn->plid) {
988 	case BLKIO_POLICY_PROP:
989 		oldpn->val.weight = newpn->val.weight;
990 		break;
991 	case BLKIO_POLICY_THROTL:
992 		switch(newpn->fileid) {
993 		case BLKIO_THROTL_read_bps_device:
994 		case BLKIO_THROTL_write_bps_device:
995 			oldpn->val.bps = newpn->val.bps;
996 			break;
997 		case BLKIO_THROTL_read_iops_device:
998 		case BLKIO_THROTL_write_iops_device:
999 			oldpn->val.iops = newpn->val.iops;
1000 		}
1001 		break;
1002 	default:
1003 		BUG();
1004 	}
1005 }
1006 
1007 /*
1008  * Some rules/values in blkg have changed. Propagate those to respective
1009  * policies.
1010  */
1011 static void blkio_update_blkg_policy(struct blkio_cgroup *blkcg,
1012 		struct blkio_group *blkg, struct blkio_policy_node *pn)
1013 {
1014 	unsigned int weight, iops;
1015 	u64 bps;
1016 
1017 	switch(pn->plid) {
1018 	case BLKIO_POLICY_PROP:
1019 		weight = pn->val.weight ? pn->val.weight :
1020 				blkcg->weight;
1021 		blkio_update_group_weight(blkg, weight);
1022 		break;
1023 	case BLKIO_POLICY_THROTL:
1024 		switch(pn->fileid) {
1025 		case BLKIO_THROTL_read_bps_device:
1026 		case BLKIO_THROTL_write_bps_device:
1027 			bps = pn->val.bps ? pn->val.bps : (-1);
1028 			blkio_update_group_bps(blkg, bps, pn->fileid);
1029 			break;
1030 		case BLKIO_THROTL_read_iops_device:
1031 		case BLKIO_THROTL_write_iops_device:
1032 			iops = pn->val.iops ? pn->val.iops : (-1);
1033 			blkio_update_group_iops(blkg, iops, pn->fileid);
1034 			break;
1035 		}
1036 		break;
1037 	default:
1038 		BUG();
1039 	}
1040 }
1041 
1042 /*
1043  * A policy node rule has been updated. Propagate this update to all the
1044  * block groups which might be affected by this update.
1045  */
1046 static void blkio_update_policy_node_blkg(struct blkio_cgroup *blkcg,
1047 				struct blkio_policy_node *pn)
1048 {
1049 	struct blkio_group *blkg;
1050 	struct hlist_node *n;
1051 
1052 	spin_lock(&blkio_list_lock);
1053 	spin_lock_irq(&blkcg->lock);
1054 
1055 	hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1056 		if (pn->dev != blkg->dev || pn->plid != blkg->plid)
1057 			continue;
1058 		blkio_update_blkg_policy(blkcg, blkg, pn);
1059 	}
1060 
1061 	spin_unlock_irq(&blkcg->lock);
1062 	spin_unlock(&blkio_list_lock);
1063 }
1064 
1065 static int blkiocg_file_write(struct cgroup *cgrp, struct cftype *cft,
1066  				       const char *buffer)
1067 {
1068 	int ret = 0;
1069 	char *buf;
1070 	struct blkio_policy_node *newpn, *pn;
1071 	struct blkio_cgroup *blkcg;
1072 	int keep_newpn = 0;
1073 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1074 	int fileid = BLKIOFILE_ATTR(cft->private);
1075 
1076 	buf = kstrdup(buffer, GFP_KERNEL);
1077 	if (!buf)
1078 		return -ENOMEM;
1079 
1080 	newpn = kzalloc(sizeof(*newpn), GFP_KERNEL);
1081 	if (!newpn) {
1082 		ret = -ENOMEM;
1083 		goto free_buf;
1084 	}
1085 
1086 	ret = blkio_policy_parse_and_set(buf, newpn, plid, fileid);
1087 	if (ret)
1088 		goto free_newpn;
1089 
1090 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1091 
1092 	spin_lock_irq(&blkcg->lock);
1093 
1094 	pn = blkio_policy_search_node(blkcg, newpn->dev, plid, fileid);
1095 	if (!pn) {
1096 		if (!blkio_delete_rule_command(newpn)) {
1097 			blkio_policy_insert_node(blkcg, newpn);
1098 			keep_newpn = 1;
1099 		}
1100 		spin_unlock_irq(&blkcg->lock);
1101 		goto update_io_group;
1102 	}
1103 
1104 	if (blkio_delete_rule_command(newpn)) {
1105 		blkio_policy_delete_node(pn);
1106 		kfree(pn);
1107 		spin_unlock_irq(&blkcg->lock);
1108 		goto update_io_group;
1109 	}
1110 	spin_unlock_irq(&blkcg->lock);
1111 
1112 	blkio_update_policy_rule(pn, newpn);
1113 
1114 update_io_group:
1115 	blkio_update_policy_node_blkg(blkcg, newpn);
1116 
1117 free_newpn:
1118 	if (!keep_newpn)
1119 		kfree(newpn);
1120 free_buf:
1121 	kfree(buf);
1122 	return ret;
1123 }
1124 
1125 static void
1126 blkio_print_policy_node(struct seq_file *m, struct blkio_policy_node *pn)
1127 {
1128 	switch(pn->plid) {
1129 		case BLKIO_POLICY_PROP:
1130 			if (pn->fileid == BLKIO_PROP_weight_device)
1131 				seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1132 					MINOR(pn->dev), pn->val.weight);
1133 			break;
1134 		case BLKIO_POLICY_THROTL:
1135 			switch(pn->fileid) {
1136 			case BLKIO_THROTL_read_bps_device:
1137 			case BLKIO_THROTL_write_bps_device:
1138 				seq_printf(m, "%u:%u\t%llu\n", MAJOR(pn->dev),
1139 					MINOR(pn->dev), pn->val.bps);
1140 				break;
1141 			case BLKIO_THROTL_read_iops_device:
1142 			case BLKIO_THROTL_write_iops_device:
1143 				seq_printf(m, "%u:%u\t%u\n", MAJOR(pn->dev),
1144 					MINOR(pn->dev), pn->val.iops);
1145 				break;
1146 			}
1147 			break;
1148 		default:
1149 			BUG();
1150 	}
1151 }
1152 
1153 /* cgroup files which read their data from policy nodes end up here */
1154 static void blkio_read_policy_node_files(struct cftype *cft,
1155 			struct blkio_cgroup *blkcg, struct seq_file *m)
1156 {
1157 	struct blkio_policy_node *pn;
1158 
1159 	if (!list_empty(&blkcg->policy_list)) {
1160 		spin_lock_irq(&blkcg->lock);
1161 		list_for_each_entry(pn, &blkcg->policy_list, node) {
1162 			if (!pn_matches_cftype(cft, pn))
1163 				continue;
1164 			blkio_print_policy_node(m, pn);
1165 		}
1166 		spin_unlock_irq(&blkcg->lock);
1167 	}
1168 }
1169 
1170 static int blkiocg_file_read(struct cgroup *cgrp, struct cftype *cft,
1171 				struct seq_file *m)
1172 {
1173 	struct blkio_cgroup *blkcg;
1174 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1175 	int name = BLKIOFILE_ATTR(cft->private);
1176 
1177 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1178 
1179 	switch(plid) {
1180 	case BLKIO_POLICY_PROP:
1181 		switch(name) {
1182 		case BLKIO_PROP_weight_device:
1183 			blkio_read_policy_node_files(cft, blkcg, m);
1184 			return 0;
1185 		default:
1186 			BUG();
1187 		}
1188 		break;
1189 	case BLKIO_POLICY_THROTL:
1190 		switch(name){
1191 		case BLKIO_THROTL_read_bps_device:
1192 		case BLKIO_THROTL_write_bps_device:
1193 		case BLKIO_THROTL_read_iops_device:
1194 		case BLKIO_THROTL_write_iops_device:
1195 			blkio_read_policy_node_files(cft, blkcg, m);
1196 			return 0;
1197 		default:
1198 			BUG();
1199 		}
1200 		break;
1201 	default:
1202 		BUG();
1203 	}
1204 
1205 	return 0;
1206 }
1207 
1208 static int blkio_read_blkg_stats(struct blkio_cgroup *blkcg,
1209 		struct cftype *cft, struct cgroup_map_cb *cb,
1210 		enum stat_type type, bool show_total, bool pcpu)
1211 {
1212 	struct blkio_group *blkg;
1213 	struct hlist_node *n;
1214 	uint64_t cgroup_total = 0;
1215 
1216 	rcu_read_lock();
1217 	hlist_for_each_entry_rcu(blkg, n, &blkcg->blkg_list, blkcg_node) {
1218 		if (blkg->dev) {
1219 			if (!cftype_blkg_same_policy(cft, blkg))
1220 				continue;
1221 			if (pcpu)
1222 				cgroup_total += blkio_get_stat_cpu(blkg, cb,
1223 						blkg->dev, type);
1224 			else {
1225 				spin_lock_irq(&blkg->stats_lock);
1226 				cgroup_total += blkio_get_stat(blkg, cb,
1227 						blkg->dev, type);
1228 				spin_unlock_irq(&blkg->stats_lock);
1229 			}
1230 		}
1231 	}
1232 	if (show_total)
1233 		cb->fill(cb, "Total", cgroup_total);
1234 	rcu_read_unlock();
1235 	return 0;
1236 }
1237 
1238 /* All map kind of cgroup file get serviced by this function */
1239 static int blkiocg_file_read_map(struct cgroup *cgrp, struct cftype *cft,
1240 				struct cgroup_map_cb *cb)
1241 {
1242 	struct blkio_cgroup *blkcg;
1243 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1244 	int name = BLKIOFILE_ATTR(cft->private);
1245 
1246 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1247 
1248 	switch(plid) {
1249 	case BLKIO_POLICY_PROP:
1250 		switch(name) {
1251 		case BLKIO_PROP_time:
1252 			return blkio_read_blkg_stats(blkcg, cft, cb,
1253 						BLKIO_STAT_TIME, 0, 0);
1254 		case BLKIO_PROP_sectors:
1255 			return blkio_read_blkg_stats(blkcg, cft, cb,
1256 						BLKIO_STAT_CPU_SECTORS, 0, 1);
1257 		case BLKIO_PROP_io_service_bytes:
1258 			return blkio_read_blkg_stats(blkcg, cft, cb,
1259 					BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1260 		case BLKIO_PROP_io_serviced:
1261 			return blkio_read_blkg_stats(blkcg, cft, cb,
1262 						BLKIO_STAT_CPU_SERVICED, 1, 1);
1263 		case BLKIO_PROP_io_service_time:
1264 			return blkio_read_blkg_stats(blkcg, cft, cb,
1265 						BLKIO_STAT_SERVICE_TIME, 1, 0);
1266 		case BLKIO_PROP_io_wait_time:
1267 			return blkio_read_blkg_stats(blkcg, cft, cb,
1268 						BLKIO_STAT_WAIT_TIME, 1, 0);
1269 		case BLKIO_PROP_io_merged:
1270 			return blkio_read_blkg_stats(blkcg, cft, cb,
1271 						BLKIO_STAT_CPU_MERGED, 1, 1);
1272 		case BLKIO_PROP_io_queued:
1273 			return blkio_read_blkg_stats(blkcg, cft, cb,
1274 						BLKIO_STAT_QUEUED, 1, 0);
1275 #ifdef CONFIG_DEBUG_BLK_CGROUP
1276 		case BLKIO_PROP_unaccounted_time:
1277 			return blkio_read_blkg_stats(blkcg, cft, cb,
1278 					BLKIO_STAT_UNACCOUNTED_TIME, 0, 0);
1279 		case BLKIO_PROP_dequeue:
1280 			return blkio_read_blkg_stats(blkcg, cft, cb,
1281 						BLKIO_STAT_DEQUEUE, 0, 0);
1282 		case BLKIO_PROP_avg_queue_size:
1283 			return blkio_read_blkg_stats(blkcg, cft, cb,
1284 					BLKIO_STAT_AVG_QUEUE_SIZE, 0, 0);
1285 		case BLKIO_PROP_group_wait_time:
1286 			return blkio_read_blkg_stats(blkcg, cft, cb,
1287 					BLKIO_STAT_GROUP_WAIT_TIME, 0, 0);
1288 		case BLKIO_PROP_idle_time:
1289 			return blkio_read_blkg_stats(blkcg, cft, cb,
1290 						BLKIO_STAT_IDLE_TIME, 0, 0);
1291 		case BLKIO_PROP_empty_time:
1292 			return blkio_read_blkg_stats(blkcg, cft, cb,
1293 						BLKIO_STAT_EMPTY_TIME, 0, 0);
1294 #endif
1295 		default:
1296 			BUG();
1297 		}
1298 		break;
1299 	case BLKIO_POLICY_THROTL:
1300 		switch(name){
1301 		case BLKIO_THROTL_io_service_bytes:
1302 			return blkio_read_blkg_stats(blkcg, cft, cb,
1303 						BLKIO_STAT_CPU_SERVICE_BYTES, 1, 1);
1304 		case BLKIO_THROTL_io_serviced:
1305 			return blkio_read_blkg_stats(blkcg, cft, cb,
1306 						BLKIO_STAT_CPU_SERVICED, 1, 1);
1307 		default:
1308 			BUG();
1309 		}
1310 		break;
1311 	default:
1312 		BUG();
1313 	}
1314 
1315 	return 0;
1316 }
1317 
1318 static int blkio_weight_write(struct blkio_cgroup *blkcg, u64 val)
1319 {
1320 	struct blkio_group *blkg;
1321 	struct hlist_node *n;
1322 	struct blkio_policy_node *pn;
1323 
1324 	if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX)
1325 		return -EINVAL;
1326 
1327 	spin_lock(&blkio_list_lock);
1328 	spin_lock_irq(&blkcg->lock);
1329 	blkcg->weight = (unsigned int)val;
1330 
1331 	hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) {
1332 		pn = blkio_policy_search_node(blkcg, blkg->dev,
1333 				BLKIO_POLICY_PROP, BLKIO_PROP_weight_device);
1334 		if (pn)
1335 			continue;
1336 
1337 		blkio_update_group_weight(blkg, blkcg->weight);
1338 	}
1339 	spin_unlock_irq(&blkcg->lock);
1340 	spin_unlock(&blkio_list_lock);
1341 	return 0;
1342 }
1343 
1344 static u64 blkiocg_file_read_u64 (struct cgroup *cgrp, struct cftype *cft) {
1345 	struct blkio_cgroup *blkcg;
1346 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1347 	int name = BLKIOFILE_ATTR(cft->private);
1348 
1349 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1350 
1351 	switch(plid) {
1352 	case BLKIO_POLICY_PROP:
1353 		switch(name) {
1354 		case BLKIO_PROP_weight:
1355 			return (u64)blkcg->weight;
1356 		}
1357 		break;
1358 	default:
1359 		BUG();
1360 	}
1361 	return 0;
1362 }
1363 
1364 static int
1365 blkiocg_file_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1366 {
1367 	struct blkio_cgroup *blkcg;
1368 	enum blkio_policy_id plid = BLKIOFILE_POLICY(cft->private);
1369 	int name = BLKIOFILE_ATTR(cft->private);
1370 
1371 	blkcg = cgroup_to_blkio_cgroup(cgrp);
1372 
1373 	switch(plid) {
1374 	case BLKIO_POLICY_PROP:
1375 		switch(name) {
1376 		case BLKIO_PROP_weight:
1377 			return blkio_weight_write(blkcg, val);
1378 		}
1379 		break;
1380 	default:
1381 		BUG();
1382 	}
1383 
1384 	return 0;
1385 }
1386 
1387 struct cftype blkio_files[] = {
1388 	{
1389 		.name = "weight_device",
1390 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1391 				BLKIO_PROP_weight_device),
1392 		.read_seq_string = blkiocg_file_read,
1393 		.write_string = blkiocg_file_write,
1394 		.max_write_len = 256,
1395 	},
1396 	{
1397 		.name = "weight",
1398 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1399 				BLKIO_PROP_weight),
1400 		.read_u64 = blkiocg_file_read_u64,
1401 		.write_u64 = blkiocg_file_write_u64,
1402 	},
1403 	{
1404 		.name = "time",
1405 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1406 				BLKIO_PROP_time),
1407 		.read_map = blkiocg_file_read_map,
1408 	},
1409 	{
1410 		.name = "sectors",
1411 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1412 				BLKIO_PROP_sectors),
1413 		.read_map = blkiocg_file_read_map,
1414 	},
1415 	{
1416 		.name = "io_service_bytes",
1417 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1418 				BLKIO_PROP_io_service_bytes),
1419 		.read_map = blkiocg_file_read_map,
1420 	},
1421 	{
1422 		.name = "io_serviced",
1423 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1424 				BLKIO_PROP_io_serviced),
1425 		.read_map = blkiocg_file_read_map,
1426 	},
1427 	{
1428 		.name = "io_service_time",
1429 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1430 				BLKIO_PROP_io_service_time),
1431 		.read_map = blkiocg_file_read_map,
1432 	},
1433 	{
1434 		.name = "io_wait_time",
1435 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1436 				BLKIO_PROP_io_wait_time),
1437 		.read_map = blkiocg_file_read_map,
1438 	},
1439 	{
1440 		.name = "io_merged",
1441 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1442 				BLKIO_PROP_io_merged),
1443 		.read_map = blkiocg_file_read_map,
1444 	},
1445 	{
1446 		.name = "io_queued",
1447 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1448 				BLKIO_PROP_io_queued),
1449 		.read_map = blkiocg_file_read_map,
1450 	},
1451 	{
1452 		.name = "reset_stats",
1453 		.write_u64 = blkiocg_reset_stats,
1454 	},
1455 #ifdef CONFIG_BLK_DEV_THROTTLING
1456 	{
1457 		.name = "throttle.read_bps_device",
1458 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1459 				BLKIO_THROTL_read_bps_device),
1460 		.read_seq_string = blkiocg_file_read,
1461 		.write_string = blkiocg_file_write,
1462 		.max_write_len = 256,
1463 	},
1464 
1465 	{
1466 		.name = "throttle.write_bps_device",
1467 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1468 				BLKIO_THROTL_write_bps_device),
1469 		.read_seq_string = blkiocg_file_read,
1470 		.write_string = blkiocg_file_write,
1471 		.max_write_len = 256,
1472 	},
1473 
1474 	{
1475 		.name = "throttle.read_iops_device",
1476 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1477 				BLKIO_THROTL_read_iops_device),
1478 		.read_seq_string = blkiocg_file_read,
1479 		.write_string = blkiocg_file_write,
1480 		.max_write_len = 256,
1481 	},
1482 
1483 	{
1484 		.name = "throttle.write_iops_device",
1485 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1486 				BLKIO_THROTL_write_iops_device),
1487 		.read_seq_string = blkiocg_file_read,
1488 		.write_string = blkiocg_file_write,
1489 		.max_write_len = 256,
1490 	},
1491 	{
1492 		.name = "throttle.io_service_bytes",
1493 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1494 				BLKIO_THROTL_io_service_bytes),
1495 		.read_map = blkiocg_file_read_map,
1496 	},
1497 	{
1498 		.name = "throttle.io_serviced",
1499 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_THROTL,
1500 				BLKIO_THROTL_io_serviced),
1501 		.read_map = blkiocg_file_read_map,
1502 	},
1503 #endif /* CONFIG_BLK_DEV_THROTTLING */
1504 
1505 #ifdef CONFIG_DEBUG_BLK_CGROUP
1506 	{
1507 		.name = "avg_queue_size",
1508 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1509 				BLKIO_PROP_avg_queue_size),
1510 		.read_map = blkiocg_file_read_map,
1511 	},
1512 	{
1513 		.name = "group_wait_time",
1514 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1515 				BLKIO_PROP_group_wait_time),
1516 		.read_map = blkiocg_file_read_map,
1517 	},
1518 	{
1519 		.name = "idle_time",
1520 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1521 				BLKIO_PROP_idle_time),
1522 		.read_map = blkiocg_file_read_map,
1523 	},
1524 	{
1525 		.name = "empty_time",
1526 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1527 				BLKIO_PROP_empty_time),
1528 		.read_map = blkiocg_file_read_map,
1529 	},
1530 	{
1531 		.name = "dequeue",
1532 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1533 				BLKIO_PROP_dequeue),
1534 		.read_map = blkiocg_file_read_map,
1535 	},
1536 	{
1537 		.name = "unaccounted_time",
1538 		.private = BLKIOFILE_PRIVATE(BLKIO_POLICY_PROP,
1539 				BLKIO_PROP_unaccounted_time),
1540 		.read_map = blkiocg_file_read_map,
1541 	},
1542 #endif
1543 };
1544 
1545 static int blkiocg_populate(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1546 {
1547 	return cgroup_add_files(cgroup, subsys, blkio_files,
1548 				ARRAY_SIZE(blkio_files));
1549 }
1550 
1551 static void blkiocg_destroy(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1552 {
1553 	struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup);
1554 	unsigned long flags;
1555 	struct blkio_group *blkg;
1556 	void *key;
1557 	struct blkio_policy_type *blkiop;
1558 	struct blkio_policy_node *pn, *pntmp;
1559 
1560 	rcu_read_lock();
1561 	do {
1562 		spin_lock_irqsave(&blkcg->lock, flags);
1563 
1564 		if (hlist_empty(&blkcg->blkg_list)) {
1565 			spin_unlock_irqrestore(&blkcg->lock, flags);
1566 			break;
1567 		}
1568 
1569 		blkg = hlist_entry(blkcg->blkg_list.first, struct blkio_group,
1570 					blkcg_node);
1571 		key = rcu_dereference(blkg->key);
1572 		__blkiocg_del_blkio_group(blkg);
1573 
1574 		spin_unlock_irqrestore(&blkcg->lock, flags);
1575 
1576 		/*
1577 		 * This blkio_group is being unlinked as associated cgroup is
1578 		 * going away. Let all the IO controlling policies know about
1579 		 * this event.
1580 		 */
1581 		spin_lock(&blkio_list_lock);
1582 		list_for_each_entry(blkiop, &blkio_list, list) {
1583 			if (blkiop->plid != blkg->plid)
1584 				continue;
1585 			blkiop->ops.blkio_unlink_group_fn(key, blkg);
1586 		}
1587 		spin_unlock(&blkio_list_lock);
1588 	} while (1);
1589 
1590 	list_for_each_entry_safe(pn, pntmp, &blkcg->policy_list, node) {
1591 		blkio_policy_delete_node(pn);
1592 		kfree(pn);
1593 	}
1594 
1595 	free_css_id(&blkio_subsys, &blkcg->css);
1596 	rcu_read_unlock();
1597 	if (blkcg != &blkio_root_cgroup)
1598 		kfree(blkcg);
1599 }
1600 
1601 static struct cgroup_subsys_state *
1602 blkiocg_create(struct cgroup_subsys *subsys, struct cgroup *cgroup)
1603 {
1604 	struct blkio_cgroup *blkcg;
1605 	struct cgroup *parent = cgroup->parent;
1606 
1607 	if (!parent) {
1608 		blkcg = &blkio_root_cgroup;
1609 		goto done;
1610 	}
1611 
1612 	blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
1613 	if (!blkcg)
1614 		return ERR_PTR(-ENOMEM);
1615 
1616 	blkcg->weight = BLKIO_WEIGHT_DEFAULT;
1617 done:
1618 	spin_lock_init(&blkcg->lock);
1619 	INIT_HLIST_HEAD(&blkcg->blkg_list);
1620 
1621 	INIT_LIST_HEAD(&blkcg->policy_list);
1622 	return &blkcg->css;
1623 }
1624 
1625 /*
1626  * We cannot support shared io contexts, as we have no mean to support
1627  * two tasks with the same ioc in two different groups without major rework
1628  * of the main cic data structures.  For now we allow a task to change
1629  * its cgroup only if it's the only owner of its ioc.
1630  */
1631 static int blkiocg_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
1632 			      struct cgroup_taskset *tset)
1633 {
1634 	struct task_struct *task;
1635 	struct io_context *ioc;
1636 	int ret = 0;
1637 
1638 	/* task_lock() is needed to avoid races with exit_io_context() */
1639 	cgroup_taskset_for_each(task, cgrp, tset) {
1640 		task_lock(task);
1641 		ioc = task->io_context;
1642 		if (ioc && atomic_read(&ioc->nr_tasks) > 1)
1643 			ret = -EINVAL;
1644 		task_unlock(task);
1645 		if (ret)
1646 			break;
1647 	}
1648 	return ret;
1649 }
1650 
1651 static void blkiocg_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
1652 			   struct cgroup_taskset *tset)
1653 {
1654 	struct task_struct *task;
1655 	struct io_context *ioc;
1656 
1657 	cgroup_taskset_for_each(task, cgrp, tset) {
1658 		/* we don't lose anything even if ioc allocation fails */
1659 		ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
1660 		if (ioc) {
1661 			ioc_cgroup_changed(ioc);
1662 			put_io_context(ioc, NULL);
1663 		}
1664 	}
1665 }
1666 
1667 void blkio_policy_register(struct blkio_policy_type *blkiop)
1668 {
1669 	spin_lock(&blkio_list_lock);
1670 	list_add_tail(&blkiop->list, &blkio_list);
1671 	spin_unlock(&blkio_list_lock);
1672 }
1673 EXPORT_SYMBOL_GPL(blkio_policy_register);
1674 
1675 void blkio_policy_unregister(struct blkio_policy_type *blkiop)
1676 {
1677 	spin_lock(&blkio_list_lock);
1678 	list_del_init(&blkiop->list);
1679 	spin_unlock(&blkio_list_lock);
1680 }
1681 EXPORT_SYMBOL_GPL(blkio_policy_unregister);
1682 
1683 static int __init init_cgroup_blkio(void)
1684 {
1685 	return cgroup_load_subsys(&blkio_subsys);
1686 }
1687 
1688 static void __exit exit_cgroup_blkio(void)
1689 {
1690 	cgroup_unload_subsys(&blkio_subsys);
1691 }
1692 
1693 module_init(init_cgroup_blkio);
1694 module_exit(exit_cgroup_blkio);
1695 MODULE_LICENSE("GPL");
1696