xref: /linux/mm/memcontrol.c (revision a5766f11cfd3a0c03450d99c8fe548c2940be884)
1 /* memcontrol.c - Memory Controller
2  *
3  * Copyright IBM Corporation, 2007
4  * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5  *
6  * Copyright 2007 OpenVZ SWsoft Inc
7  * Author: Pavel Emelianov <xemul@openvz.org>
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License as published by
11  * the Free Software Foundation; either version 2 of the License, or
12  * (at your option) any later version.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  */
19 
20 #include <linux/res_counter.h>
21 #include <linux/memcontrol.h>
22 #include <linux/cgroup.h>
23 #include <linux/mm.h>
24 #include <linux/smp.h>
25 #include <linux/page-flags.h>
26 #include <linux/backing-dev.h>
27 #include <linux/bit_spinlock.h>
28 #include <linux/rcupdate.h>
29 #include <linux/slab.h>
30 #include <linux/swap.h>
31 #include <linux/spinlock.h>
32 #include <linux/fs.h>
33 #include <linux/seq_file.h>
34 #include <linux/vmalloc.h>
35 
36 #include <asm/uaccess.h>
37 
38 struct cgroup_subsys mem_cgroup_subsys __read_mostly;
39 static struct kmem_cache *page_cgroup_cache __read_mostly;
40 #define MEM_CGROUP_RECLAIM_RETRIES	5
41 
42 /*
43  * Statistics for memory cgroup.
44  */
45 enum mem_cgroup_stat_index {
46 	/*
47 	 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
48 	 */
49 	MEM_CGROUP_STAT_CACHE, 	   /* # of pages charged as cache */
50 	MEM_CGROUP_STAT_RSS,	   /* # of pages charged as rss */
51 	MEM_CGROUP_STAT_PGPGIN_COUNT,	/* # of pages paged in */
52 	MEM_CGROUP_STAT_PGPGOUT_COUNT,	/* # of pages paged out */
53 
54 	MEM_CGROUP_STAT_NSTATS,
55 };
56 
57 struct mem_cgroup_stat_cpu {
58 	s64 count[MEM_CGROUP_STAT_NSTATS];
59 } ____cacheline_aligned_in_smp;
60 
61 struct mem_cgroup_stat {
62 	struct mem_cgroup_stat_cpu cpustat[NR_CPUS];
63 };
64 
65 /*
66  * For accounting under irq disable, no need for increment preempt count.
67  */
68 static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat,
69 		enum mem_cgroup_stat_index idx, int val)
70 {
71 	int cpu = smp_processor_id();
72 	stat->cpustat[cpu].count[idx] += val;
73 }
74 
75 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
76 		enum mem_cgroup_stat_index idx)
77 {
78 	int cpu;
79 	s64 ret = 0;
80 	for_each_possible_cpu(cpu)
81 		ret += stat->cpustat[cpu].count[idx];
82 	return ret;
83 }
84 
85 /*
86  * per-zone information in memory controller.
87  */
88 
89 enum mem_cgroup_zstat_index {
90 	MEM_CGROUP_ZSTAT_ACTIVE,
91 	MEM_CGROUP_ZSTAT_INACTIVE,
92 
93 	NR_MEM_CGROUP_ZSTAT,
94 };
95 
96 struct mem_cgroup_per_zone {
97 	/*
98 	 * spin_lock to protect the per cgroup LRU
99 	 */
100 	spinlock_t		lru_lock;
101 	struct list_head	active_list;
102 	struct list_head	inactive_list;
103 	unsigned long count[NR_MEM_CGROUP_ZSTAT];
104 };
105 /* Macro for accessing counter */
106 #define MEM_CGROUP_ZSTAT(mz, idx)	((mz)->count[(idx)])
107 
108 struct mem_cgroup_per_node {
109 	struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
110 };
111 
112 struct mem_cgroup_lru_info {
113 	struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
114 };
115 
116 /*
117  * The memory controller data structure. The memory controller controls both
118  * page cache and RSS per cgroup. We would eventually like to provide
119  * statistics based on the statistics developed by Rik Van Riel for clock-pro,
120  * to help the administrator determine what knobs to tune.
121  *
122  * TODO: Add a water mark for the memory controller. Reclaim will begin when
123  * we hit the water mark. May be even add a low water mark, such that
124  * no reclaim occurs from a cgroup at it's low water mark, this is
125  * a feature that will be implemented much later in the future.
126  */
127 struct mem_cgroup {
128 	struct cgroup_subsys_state css;
129 	/*
130 	 * the counter to account for memory usage
131 	 */
132 	struct res_counter res;
133 	/*
134 	 * Per cgroup active and inactive list, similar to the
135 	 * per zone LRU lists.
136 	 */
137 	struct mem_cgroup_lru_info info;
138 
139 	int	prev_priority;	/* for recording reclaim priority */
140 	/*
141 	 * statistics.
142 	 */
143 	struct mem_cgroup_stat stat;
144 };
145 static struct mem_cgroup init_mem_cgroup;
146 
147 /*
148  * We use the lower bit of the page->page_cgroup pointer as a bit spin
149  * lock.  We need to ensure that page->page_cgroup is at least two
150  * byte aligned (based on comments from Nick Piggin).  But since
151  * bit_spin_lock doesn't actually set that lock bit in a non-debug
152  * uniprocessor kernel, we should avoid setting it here too.
153  */
154 #define PAGE_CGROUP_LOCK_BIT 	0x0
155 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
156 #define PAGE_CGROUP_LOCK 	(1 << PAGE_CGROUP_LOCK_BIT)
157 #else
158 #define PAGE_CGROUP_LOCK	0x0
159 #endif
160 
161 /*
162  * A page_cgroup page is associated with every page descriptor. The
163  * page_cgroup helps us identify information about the cgroup
164  */
165 struct page_cgroup {
166 	struct list_head lru;		/* per cgroup LRU list */
167 	struct page *page;
168 	struct mem_cgroup *mem_cgroup;
169 	int flags;
170 };
171 #define PAGE_CGROUP_FLAG_CACHE	(0x1)	/* charged as cache */
172 #define PAGE_CGROUP_FLAG_ACTIVE (0x2)	/* page is active in this cgroup */
173 
174 static int page_cgroup_nid(struct page_cgroup *pc)
175 {
176 	return page_to_nid(pc->page);
177 }
178 
179 static enum zone_type page_cgroup_zid(struct page_cgroup *pc)
180 {
181 	return page_zonenum(pc->page);
182 }
183 
184 enum charge_type {
185 	MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
186 	MEM_CGROUP_CHARGE_TYPE_MAPPED,
187 	MEM_CGROUP_CHARGE_TYPE_FORCE,	/* used by force_empty */
188 };
189 
190 /*
191  * Always modified under lru lock. Then, not necessary to preempt_disable()
192  */
193 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
194 					bool charge)
195 {
196 	int val = (charge)? 1 : -1;
197 	struct mem_cgroup_stat *stat = &mem->stat;
198 
199 	VM_BUG_ON(!irqs_disabled());
200 	if (flags & PAGE_CGROUP_FLAG_CACHE)
201 		__mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val);
202 	else
203 		__mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val);
204 
205 	if (charge)
206 		__mem_cgroup_stat_add_safe(stat,
207 				MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
208 	else
209 		__mem_cgroup_stat_add_safe(stat,
210 				MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
211 }
212 
213 static struct mem_cgroup_per_zone *
214 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
215 {
216 	return &mem->info.nodeinfo[nid]->zoneinfo[zid];
217 }
218 
219 static struct mem_cgroup_per_zone *
220 page_cgroup_zoneinfo(struct page_cgroup *pc)
221 {
222 	struct mem_cgroup *mem = pc->mem_cgroup;
223 	int nid = page_cgroup_nid(pc);
224 	int zid = page_cgroup_zid(pc);
225 
226 	return mem_cgroup_zoneinfo(mem, nid, zid);
227 }
228 
229 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
230 					enum mem_cgroup_zstat_index idx)
231 {
232 	int nid, zid;
233 	struct mem_cgroup_per_zone *mz;
234 	u64 total = 0;
235 
236 	for_each_online_node(nid)
237 		for (zid = 0; zid < MAX_NR_ZONES; zid++) {
238 			mz = mem_cgroup_zoneinfo(mem, nid, zid);
239 			total += MEM_CGROUP_ZSTAT(mz, idx);
240 		}
241 	return total;
242 }
243 
244 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
245 {
246 	return container_of(cgroup_subsys_state(cont,
247 				mem_cgroup_subsys_id), struct mem_cgroup,
248 				css);
249 }
250 
251 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
252 {
253 	/*
254 	 * mm_update_next_owner() may clear mm->owner to NULL
255 	 * if it races with swapoff, page migration, etc.
256 	 * So this can be called with p == NULL.
257 	 */
258 	if (unlikely(!p))
259 		return NULL;
260 
261 	return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
262 				struct mem_cgroup, css);
263 }
264 
265 static inline int page_cgroup_locked(struct page *page)
266 {
267 	return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
268 }
269 
270 static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
271 {
272 	VM_BUG_ON(!page_cgroup_locked(page));
273 	page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
274 }
275 
276 struct page_cgroup *page_get_page_cgroup(struct page *page)
277 {
278 	return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
279 }
280 
281 static void lock_page_cgroup(struct page *page)
282 {
283 	bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
284 }
285 
286 static int try_lock_page_cgroup(struct page *page)
287 {
288 	return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
289 }
290 
291 static void unlock_page_cgroup(struct page *page)
292 {
293 	bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
294 }
295 
296 static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
297 			struct page_cgroup *pc)
298 {
299 	int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
300 
301 	if (from)
302 		MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
303 	else
304 		MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
305 
306 	mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
307 	list_del(&pc->lru);
308 }
309 
310 static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
311 				struct page_cgroup *pc)
312 {
313 	int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
314 
315 	if (!to) {
316 		MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
317 		list_add(&pc->lru, &mz->inactive_list);
318 	} else {
319 		MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
320 		list_add(&pc->lru, &mz->active_list);
321 	}
322 	mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
323 }
324 
325 static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
326 {
327 	int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
328 	struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
329 
330 	if (from)
331 		MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
332 	else
333 		MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
334 
335 	if (active) {
336 		MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
337 		pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
338 		list_move(&pc->lru, &mz->active_list);
339 	} else {
340 		MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
341 		pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
342 		list_move(&pc->lru, &mz->inactive_list);
343 	}
344 }
345 
346 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
347 {
348 	int ret;
349 
350 	task_lock(task);
351 	ret = task->mm && mm_match_cgroup(task->mm, mem);
352 	task_unlock(task);
353 	return ret;
354 }
355 
356 /*
357  * This routine assumes that the appropriate zone's lru lock is already held
358  */
359 void mem_cgroup_move_lists(struct page *page, bool active)
360 {
361 	struct page_cgroup *pc;
362 	struct mem_cgroup_per_zone *mz;
363 	unsigned long flags;
364 
365 	if (mem_cgroup_subsys.disabled)
366 		return;
367 
368 	/*
369 	 * We cannot lock_page_cgroup while holding zone's lru_lock,
370 	 * because other holders of lock_page_cgroup can be interrupted
371 	 * with an attempt to rotate_reclaimable_page.  But we cannot
372 	 * safely get to page_cgroup without it, so just try_lock it:
373 	 * mem_cgroup_isolate_pages allows for page left on wrong list.
374 	 */
375 	if (!try_lock_page_cgroup(page))
376 		return;
377 
378 	pc = page_get_page_cgroup(page);
379 	if (pc) {
380 		mz = page_cgroup_zoneinfo(pc);
381 		spin_lock_irqsave(&mz->lru_lock, flags);
382 		__mem_cgroup_move_lists(pc, active);
383 		spin_unlock_irqrestore(&mz->lru_lock, flags);
384 	}
385 	unlock_page_cgroup(page);
386 }
387 
388 /*
389  * Calculate mapped_ratio under memory controller. This will be used in
390  * vmscan.c for deteremining we have to reclaim mapped pages.
391  */
392 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
393 {
394 	long total, rss;
395 
396 	/*
397 	 * usage is recorded in bytes. But, here, we assume the number of
398 	 * physical pages can be represented by "long" on any arch.
399 	 */
400 	total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
401 	rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
402 	return (int)((rss * 100L) / total);
403 }
404 
405 /*
406  * This function is called from vmscan.c. In page reclaiming loop. balance
407  * between active and inactive list is calculated. For memory controller
408  * page reclaiming, we should use using mem_cgroup's imbalance rather than
409  * zone's global lru imbalance.
410  */
411 long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem)
412 {
413 	unsigned long active, inactive;
414 	/* active and inactive are the number of pages. 'long' is ok.*/
415 	active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE);
416 	inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE);
417 	return (long) (active / (inactive + 1));
418 }
419 
420 /*
421  * prev_priority control...this will be used in memory reclaim path.
422  */
423 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
424 {
425 	return mem->prev_priority;
426 }
427 
428 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
429 {
430 	if (priority < mem->prev_priority)
431 		mem->prev_priority = priority;
432 }
433 
434 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
435 {
436 	mem->prev_priority = priority;
437 }
438 
439 /*
440  * Calculate # of pages to be scanned in this priority/zone.
441  * See also vmscan.c
442  *
443  * priority starts from "DEF_PRIORITY" and decremented in each loop.
444  * (see include/linux/mmzone.h)
445  */
446 
447 long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem,
448 				   struct zone *zone, int priority)
449 {
450 	long nr_active;
451 	int nid = zone->zone_pgdat->node_id;
452 	int zid = zone_idx(zone);
453 	struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
454 
455 	nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE);
456 	return (nr_active >> priority);
457 }
458 
459 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem,
460 					struct zone *zone, int priority)
461 {
462 	long nr_inactive;
463 	int nid = zone->zone_pgdat->node_id;
464 	int zid = zone_idx(zone);
465 	struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
466 
467 	nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE);
468 	return (nr_inactive >> priority);
469 }
470 
471 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
472 					struct list_head *dst,
473 					unsigned long *scanned, int order,
474 					int mode, struct zone *z,
475 					struct mem_cgroup *mem_cont,
476 					int active)
477 {
478 	unsigned long nr_taken = 0;
479 	struct page *page;
480 	unsigned long scan;
481 	LIST_HEAD(pc_list);
482 	struct list_head *src;
483 	struct page_cgroup *pc, *tmp;
484 	int nid = z->zone_pgdat->node_id;
485 	int zid = zone_idx(z);
486 	struct mem_cgroup_per_zone *mz;
487 
488 	BUG_ON(!mem_cont);
489 	mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
490 	if (active)
491 		src = &mz->active_list;
492 	else
493 		src = &mz->inactive_list;
494 
495 
496 	spin_lock(&mz->lru_lock);
497 	scan = 0;
498 	list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
499 		if (scan >= nr_to_scan)
500 			break;
501 		page = pc->page;
502 
503 		if (unlikely(!PageLRU(page)))
504 			continue;
505 
506 		if (PageActive(page) && !active) {
507 			__mem_cgroup_move_lists(pc, true);
508 			continue;
509 		}
510 		if (!PageActive(page) && active) {
511 			__mem_cgroup_move_lists(pc, false);
512 			continue;
513 		}
514 
515 		scan++;
516 		list_move(&pc->lru, &pc_list);
517 
518 		if (__isolate_lru_page(page, mode) == 0) {
519 			list_move(&page->lru, dst);
520 			nr_taken++;
521 		}
522 	}
523 
524 	list_splice(&pc_list, src);
525 	spin_unlock(&mz->lru_lock);
526 
527 	*scanned = scan;
528 	return nr_taken;
529 }
530 
531 /*
532  * Charge the memory controller for page usage.
533  * Return
534  * 0 if the charge was successful
535  * < 0 if the cgroup is over its limit
536  */
537 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
538 				gfp_t gfp_mask, enum charge_type ctype,
539 				struct mem_cgroup *memcg)
540 {
541 	struct mem_cgroup *mem;
542 	struct page_cgroup *pc;
543 	unsigned long flags;
544 	unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
545 	struct mem_cgroup_per_zone *mz;
546 
547 	pc = kmem_cache_alloc(page_cgroup_cache, gfp_mask);
548 	if (unlikely(pc == NULL))
549 		goto err;
550 
551 	/*
552 	 * We always charge the cgroup the mm_struct belongs to.
553 	 * The mm_struct's mem_cgroup changes on task migration if the
554 	 * thread group leader migrates. It's possible that mm is not
555 	 * set, if so charge the init_mm (happens for pagecache usage).
556 	 */
557 	if (likely(!memcg)) {
558 		rcu_read_lock();
559 		mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
560 		if (unlikely(!mem)) {
561 			rcu_read_unlock();
562 			kmem_cache_free(page_cgroup_cache, pc);
563 			return 0;
564 		}
565 		/*
566 		 * For every charge from the cgroup, increment reference count
567 		 */
568 		css_get(&mem->css);
569 		rcu_read_unlock();
570 	} else {
571 		mem = memcg;
572 		css_get(&memcg->css);
573 	}
574 
575 	while (res_counter_charge(&mem->res, PAGE_SIZE)) {
576 		if (!(gfp_mask & __GFP_WAIT))
577 			goto out;
578 
579 		if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
580 			continue;
581 
582 		/*
583 		 * try_to_free_mem_cgroup_pages() might not give us a full
584 		 * picture of reclaim. Some pages are reclaimed and might be
585 		 * moved to swap cache or just unmapped from the cgroup.
586 		 * Check the limit again to see if the reclaim reduced the
587 		 * current usage of the cgroup before giving up
588 		 */
589 		if (res_counter_check_under_limit(&mem->res))
590 			continue;
591 
592 		if (!nr_retries--) {
593 			mem_cgroup_out_of_memory(mem, gfp_mask);
594 			goto out;
595 		}
596 	}
597 
598 	pc->mem_cgroup = mem;
599 	pc->page = page;
600 	/*
601 	 * If a page is accounted as a page cache, insert to inactive list.
602 	 * If anon, insert to active list.
603 	 */
604 	if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
605 		pc->flags = PAGE_CGROUP_FLAG_CACHE;
606 	else
607 		pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
608 
609 	lock_page_cgroup(page);
610 	if (unlikely(page_get_page_cgroup(page))) {
611 		unlock_page_cgroup(page);
612 		res_counter_uncharge(&mem->res, PAGE_SIZE);
613 		css_put(&mem->css);
614 		kmem_cache_free(page_cgroup_cache, pc);
615 		goto done;
616 	}
617 	page_assign_page_cgroup(page, pc);
618 
619 	mz = page_cgroup_zoneinfo(pc);
620 	spin_lock_irqsave(&mz->lru_lock, flags);
621 	__mem_cgroup_add_list(mz, pc);
622 	spin_unlock_irqrestore(&mz->lru_lock, flags);
623 
624 	unlock_page_cgroup(page);
625 done:
626 	return 0;
627 out:
628 	css_put(&mem->css);
629 	kmem_cache_free(page_cgroup_cache, pc);
630 err:
631 	return -ENOMEM;
632 }
633 
634 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
635 {
636 	if (mem_cgroup_subsys.disabled)
637 		return 0;
638 
639 	/*
640 	 * If already mapped, we don't have to account.
641 	 * If page cache, page->mapping has address_space.
642 	 * But page->mapping may have out-of-use anon_vma pointer,
643 	 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
644 	 * is NULL.
645   	 */
646 	if (page_mapped(page) || (page->mapping && !PageAnon(page)))
647 		return 0;
648 	if (unlikely(!mm))
649 		mm = &init_mm;
650 	return mem_cgroup_charge_common(page, mm, gfp_mask,
651 				MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL);
652 }
653 
654 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
655 				gfp_t gfp_mask)
656 {
657 	if (mem_cgroup_subsys.disabled)
658 		return 0;
659 
660 	/*
661 	 * Corner case handling. This is called from add_to_page_cache()
662 	 * in usual. But some FS (shmem) precharges this page before calling it
663 	 * and call add_to_page_cache() with GFP_NOWAIT.
664 	 *
665 	 * For GFP_NOWAIT case, the page may be pre-charged before calling
666 	 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
667 	 * charge twice. (It works but has to pay a bit larger cost.)
668 	 */
669 	if (!(gfp_mask & __GFP_WAIT)) {
670 		struct page_cgroup *pc;
671 
672 		lock_page_cgroup(page);
673 		pc = page_get_page_cgroup(page);
674 		if (pc) {
675 			VM_BUG_ON(pc->page != page);
676 			VM_BUG_ON(!pc->mem_cgroup);
677 			unlock_page_cgroup(page);
678 			return 0;
679 		}
680 		unlock_page_cgroup(page);
681 	}
682 
683 	if (unlikely(!mm))
684 		mm = &init_mm;
685 
686 	return mem_cgroup_charge_common(page, mm, gfp_mask,
687 				MEM_CGROUP_CHARGE_TYPE_CACHE, NULL);
688 }
689 
690 /*
691  * uncharge if !page_mapped(page)
692  */
693 static void
694 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
695 {
696 	struct page_cgroup *pc;
697 	struct mem_cgroup *mem;
698 	struct mem_cgroup_per_zone *mz;
699 	unsigned long flags;
700 
701 	if (mem_cgroup_subsys.disabled)
702 		return;
703 
704 	/*
705 	 * Check if our page_cgroup is valid
706 	 */
707 	lock_page_cgroup(page);
708 	pc = page_get_page_cgroup(page);
709 	if (unlikely(!pc))
710 		goto unlock;
711 
712 	VM_BUG_ON(pc->page != page);
713 
714 	if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED)
715 	    && ((pc->flags & PAGE_CGROUP_FLAG_CACHE)
716 		|| page_mapped(page)))
717 		goto unlock;
718 
719 	mz = page_cgroup_zoneinfo(pc);
720 	spin_lock_irqsave(&mz->lru_lock, flags);
721 	__mem_cgroup_remove_list(mz, pc);
722 	spin_unlock_irqrestore(&mz->lru_lock, flags);
723 
724 	page_assign_page_cgroup(page, NULL);
725 	unlock_page_cgroup(page);
726 
727 	mem = pc->mem_cgroup;
728 	res_counter_uncharge(&mem->res, PAGE_SIZE);
729 	css_put(&mem->css);
730 
731 	kmem_cache_free(page_cgroup_cache, pc);
732 	return;
733 unlock:
734 	unlock_page_cgroup(page);
735 }
736 
737 void mem_cgroup_uncharge_page(struct page *page)
738 {
739 	__mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED);
740 }
741 
742 void mem_cgroup_uncharge_cache_page(struct page *page)
743 {
744 	VM_BUG_ON(page_mapped(page));
745 	__mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
746 }
747 
748 /*
749  * Before starting migration, account against new page.
750  */
751 int mem_cgroup_prepare_migration(struct page *page, struct page *newpage)
752 {
753 	struct page_cgroup *pc;
754 	struct mem_cgroup *mem = NULL;
755 	enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED;
756 	int ret = 0;
757 
758 	if (mem_cgroup_subsys.disabled)
759 		return 0;
760 
761 	lock_page_cgroup(page);
762 	pc = page_get_page_cgroup(page);
763 	if (pc) {
764 		mem = pc->mem_cgroup;
765 		css_get(&mem->css);
766 		if (pc->flags & PAGE_CGROUP_FLAG_CACHE)
767 			ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
768 	}
769 	unlock_page_cgroup(page);
770 	if (mem) {
771 		ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL,
772 			ctype, mem);
773 		css_put(&mem->css);
774 	}
775 	return ret;
776 }
777 
778 /* remove redundant charge if migration failed*/
779 void mem_cgroup_end_migration(struct page *newpage)
780 {
781 	/*
782 	 * At success, page->mapping is not NULL.
783 	 * special rollback care is necessary when
784 	 * 1. at migration failure. (newpage->mapping is cleared in this case)
785 	 * 2. the newpage was moved but not remapped again because the task
786 	 *    exits and the newpage is obsolete. In this case, the new page
787 	 *    may be a swapcache. So, we just call mem_cgroup_uncharge_page()
788 	 *    always for avoiding mess. The  page_cgroup will be removed if
789 	 *    unnecessary. File cache pages is still on radix-tree. Don't
790 	 *    care it.
791 	 */
792 	if (!newpage->mapping)
793 		__mem_cgroup_uncharge_common(newpage,
794 					 MEM_CGROUP_CHARGE_TYPE_FORCE);
795 	else if (PageAnon(newpage))
796 		mem_cgroup_uncharge_page(newpage);
797 }
798 
799 /*
800  * A call to try to shrink memory usage under specified resource controller.
801  * This is typically used for page reclaiming for shmem for reducing side
802  * effect of page allocation from shmem, which is used by some mem_cgroup.
803  */
804 int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask)
805 {
806 	struct mem_cgroup *mem;
807 	int progress = 0;
808 	int retry = MEM_CGROUP_RECLAIM_RETRIES;
809 
810 	if (mem_cgroup_subsys.disabled)
811 		return 0;
812 	if (!mm)
813 		return 0;
814 
815 	rcu_read_lock();
816 	mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
817 	if (unlikely(!mem)) {
818 		rcu_read_unlock();
819 		return 0;
820 	}
821 	css_get(&mem->css);
822 	rcu_read_unlock();
823 
824 	do {
825 		progress = try_to_free_mem_cgroup_pages(mem, gfp_mask);
826 		progress += res_counter_check_under_limit(&mem->res);
827 	} while (!progress && --retry);
828 
829 	css_put(&mem->css);
830 	if (!retry)
831 		return -ENOMEM;
832 	return 0;
833 }
834 
835 int mem_cgroup_resize_limit(struct mem_cgroup *memcg, unsigned long long val)
836 {
837 
838 	int retry_count = MEM_CGROUP_RECLAIM_RETRIES;
839 	int progress;
840 	int ret = 0;
841 
842 	while (res_counter_set_limit(&memcg->res, val)) {
843 		if (signal_pending(current)) {
844 			ret = -EINTR;
845 			break;
846 		}
847 		if (!retry_count) {
848 			ret = -EBUSY;
849 			break;
850 		}
851 		progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL);
852 		if (!progress)
853 			retry_count--;
854 	}
855 	return ret;
856 }
857 
858 
859 /*
860  * This routine traverse page_cgroup in given list and drop them all.
861  * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
862  */
863 #define FORCE_UNCHARGE_BATCH	(128)
864 static void mem_cgroup_force_empty_list(struct mem_cgroup *mem,
865 			    struct mem_cgroup_per_zone *mz,
866 			    int active)
867 {
868 	struct page_cgroup *pc;
869 	struct page *page;
870 	int count = FORCE_UNCHARGE_BATCH;
871 	unsigned long flags;
872 	struct list_head *list;
873 
874 	if (active)
875 		list = &mz->active_list;
876 	else
877 		list = &mz->inactive_list;
878 
879 	spin_lock_irqsave(&mz->lru_lock, flags);
880 	while (!list_empty(list)) {
881 		pc = list_entry(list->prev, struct page_cgroup, lru);
882 		page = pc->page;
883 		get_page(page);
884 		spin_unlock_irqrestore(&mz->lru_lock, flags);
885 		/*
886 		 * Check if this page is on LRU. !LRU page can be found
887 		 * if it's under page migration.
888 		 */
889 		if (PageLRU(page)) {
890 			__mem_cgroup_uncharge_common(page,
891 					MEM_CGROUP_CHARGE_TYPE_FORCE);
892 			put_page(page);
893 			if (--count <= 0) {
894 				count = FORCE_UNCHARGE_BATCH;
895 				cond_resched();
896 			}
897 		} else
898 			cond_resched();
899 		spin_lock_irqsave(&mz->lru_lock, flags);
900 	}
901 	spin_unlock_irqrestore(&mz->lru_lock, flags);
902 }
903 
904 /*
905  * make mem_cgroup's charge to be 0 if there is no task.
906  * This enables deleting this mem_cgroup.
907  */
908 static int mem_cgroup_force_empty(struct mem_cgroup *mem)
909 {
910 	int ret = -EBUSY;
911 	int node, zid;
912 
913 	css_get(&mem->css);
914 	/*
915 	 * page reclaim code (kswapd etc..) will move pages between
916 	 * active_list <-> inactive_list while we don't take a lock.
917 	 * So, we have to do loop here until all lists are empty.
918 	 */
919 	while (mem->res.usage > 0) {
920 		if (atomic_read(&mem->css.cgroup->count) > 0)
921 			goto out;
922 		for_each_node_state(node, N_POSSIBLE)
923 			for (zid = 0; zid < MAX_NR_ZONES; zid++) {
924 				struct mem_cgroup_per_zone *mz;
925 				mz = mem_cgroup_zoneinfo(mem, node, zid);
926 				/* drop all page_cgroup in active_list */
927 				mem_cgroup_force_empty_list(mem, mz, 1);
928 				/* drop all page_cgroup in inactive_list */
929 				mem_cgroup_force_empty_list(mem, mz, 0);
930 			}
931 	}
932 	ret = 0;
933 out:
934 	css_put(&mem->css);
935 	return ret;
936 }
937 
938 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
939 {
940 	return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
941 				    cft->private);
942 }
943 /*
944  * The user of this function is...
945  * RES_LIMIT.
946  */
947 static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
948 			    const char *buffer)
949 {
950 	struct mem_cgroup *memcg = mem_cgroup_from_cont(cont);
951 	unsigned long long val;
952 	int ret;
953 
954 	switch (cft->private) {
955 	case RES_LIMIT:
956 		/* This function does all necessary parse...reuse it */
957 		ret = res_counter_memparse_write_strategy(buffer, &val);
958 		if (!ret)
959 			ret = mem_cgroup_resize_limit(memcg, val);
960 		break;
961 	default:
962 		ret = -EINVAL; /* should be BUG() ? */
963 		break;
964 	}
965 	return ret;
966 }
967 
968 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
969 {
970 	struct mem_cgroup *mem;
971 
972 	mem = mem_cgroup_from_cont(cont);
973 	switch (event) {
974 	case RES_MAX_USAGE:
975 		res_counter_reset_max(&mem->res);
976 		break;
977 	case RES_FAILCNT:
978 		res_counter_reset_failcnt(&mem->res);
979 		break;
980 	}
981 	return 0;
982 }
983 
984 static int mem_force_empty_write(struct cgroup *cont, unsigned int event)
985 {
986 	return mem_cgroup_force_empty(mem_cgroup_from_cont(cont));
987 }
988 
989 static const struct mem_cgroup_stat_desc {
990 	const char *msg;
991 	u64 unit;
992 } mem_cgroup_stat_desc[] = {
993 	[MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
994 	[MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
995 	[MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
996 	[MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
997 };
998 
999 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
1000 				 struct cgroup_map_cb *cb)
1001 {
1002 	struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
1003 	struct mem_cgroup_stat *stat = &mem_cont->stat;
1004 	int i;
1005 
1006 	for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
1007 		s64 val;
1008 
1009 		val = mem_cgroup_read_stat(stat, i);
1010 		val *= mem_cgroup_stat_desc[i].unit;
1011 		cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
1012 	}
1013 	/* showing # of active pages */
1014 	{
1015 		unsigned long active, inactive;
1016 
1017 		inactive = mem_cgroup_get_all_zonestat(mem_cont,
1018 						MEM_CGROUP_ZSTAT_INACTIVE);
1019 		active = mem_cgroup_get_all_zonestat(mem_cont,
1020 						MEM_CGROUP_ZSTAT_ACTIVE);
1021 		cb->fill(cb, "active", (active) * PAGE_SIZE);
1022 		cb->fill(cb, "inactive", (inactive) * PAGE_SIZE);
1023 	}
1024 	return 0;
1025 }
1026 
1027 static struct cftype mem_cgroup_files[] = {
1028 	{
1029 		.name = "usage_in_bytes",
1030 		.private = RES_USAGE,
1031 		.read_u64 = mem_cgroup_read,
1032 	},
1033 	{
1034 		.name = "max_usage_in_bytes",
1035 		.private = RES_MAX_USAGE,
1036 		.trigger = mem_cgroup_reset,
1037 		.read_u64 = mem_cgroup_read,
1038 	},
1039 	{
1040 		.name = "limit_in_bytes",
1041 		.private = RES_LIMIT,
1042 		.write_string = mem_cgroup_write,
1043 		.read_u64 = mem_cgroup_read,
1044 	},
1045 	{
1046 		.name = "failcnt",
1047 		.private = RES_FAILCNT,
1048 		.trigger = mem_cgroup_reset,
1049 		.read_u64 = mem_cgroup_read,
1050 	},
1051 	{
1052 		.name = "force_empty",
1053 		.trigger = mem_force_empty_write,
1054 	},
1055 	{
1056 		.name = "stat",
1057 		.read_map = mem_control_stat_show,
1058 	},
1059 };
1060 
1061 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1062 {
1063 	struct mem_cgroup_per_node *pn;
1064 	struct mem_cgroup_per_zone *mz;
1065 	int zone, tmp = node;
1066 	/*
1067 	 * This routine is called against possible nodes.
1068 	 * But it's BUG to call kmalloc() against offline node.
1069 	 *
1070 	 * TODO: this routine can waste much memory for nodes which will
1071 	 *       never be onlined. It's better to use memory hotplug callback
1072 	 *       function.
1073 	 */
1074 	if (!node_state(node, N_NORMAL_MEMORY))
1075 		tmp = -1;
1076 	pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
1077 	if (!pn)
1078 		return 1;
1079 
1080 	mem->info.nodeinfo[node] = pn;
1081 	memset(pn, 0, sizeof(*pn));
1082 
1083 	for (zone = 0; zone < MAX_NR_ZONES; zone++) {
1084 		mz = &pn->zoneinfo[zone];
1085 		INIT_LIST_HEAD(&mz->active_list);
1086 		INIT_LIST_HEAD(&mz->inactive_list);
1087 		spin_lock_init(&mz->lru_lock);
1088 	}
1089 	return 0;
1090 }
1091 
1092 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
1093 {
1094 	kfree(mem->info.nodeinfo[node]);
1095 }
1096 
1097 static struct mem_cgroup *mem_cgroup_alloc(void)
1098 {
1099 	struct mem_cgroup *mem;
1100 
1101 	if (sizeof(*mem) < PAGE_SIZE)
1102 		mem = kmalloc(sizeof(*mem), GFP_KERNEL);
1103 	else
1104 		mem = vmalloc(sizeof(*mem));
1105 
1106 	if (mem)
1107 		memset(mem, 0, sizeof(*mem));
1108 	return mem;
1109 }
1110 
1111 static void mem_cgroup_free(struct mem_cgroup *mem)
1112 {
1113 	if (sizeof(*mem) < PAGE_SIZE)
1114 		kfree(mem);
1115 	else
1116 		vfree(mem);
1117 }
1118 
1119 
1120 static struct cgroup_subsys_state *
1121 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1122 {
1123 	struct mem_cgroup *mem;
1124 	int node;
1125 
1126 	if (unlikely((cont->parent) == NULL)) {
1127 		mem = &init_mem_cgroup;
1128 		page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC);
1129 	} else {
1130 		mem = mem_cgroup_alloc();
1131 		if (!mem)
1132 			return ERR_PTR(-ENOMEM);
1133 	}
1134 
1135 	res_counter_init(&mem->res);
1136 
1137 	for_each_node_state(node, N_POSSIBLE)
1138 		if (alloc_mem_cgroup_per_zone_info(mem, node))
1139 			goto free_out;
1140 
1141 	return &mem->css;
1142 free_out:
1143 	for_each_node_state(node, N_POSSIBLE)
1144 		free_mem_cgroup_per_zone_info(mem, node);
1145 	if (cont->parent != NULL)
1146 		mem_cgroup_free(mem);
1147 	return ERR_PTR(-ENOMEM);
1148 }
1149 
1150 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1151 					struct cgroup *cont)
1152 {
1153 	struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1154 	mem_cgroup_force_empty(mem);
1155 }
1156 
1157 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1158 				struct cgroup *cont)
1159 {
1160 	int node;
1161 	struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1162 
1163 	for_each_node_state(node, N_POSSIBLE)
1164 		free_mem_cgroup_per_zone_info(mem, node);
1165 
1166 	mem_cgroup_free(mem_cgroup_from_cont(cont));
1167 }
1168 
1169 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1170 				struct cgroup *cont)
1171 {
1172 	return cgroup_add_files(cont, ss, mem_cgroup_files,
1173 					ARRAY_SIZE(mem_cgroup_files));
1174 }
1175 
1176 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1177 				struct cgroup *cont,
1178 				struct cgroup *old_cont,
1179 				struct task_struct *p)
1180 {
1181 	struct mm_struct *mm;
1182 	struct mem_cgroup *mem, *old_mem;
1183 
1184 	mm = get_task_mm(p);
1185 	if (mm == NULL)
1186 		return;
1187 
1188 	mem = mem_cgroup_from_cont(cont);
1189 	old_mem = mem_cgroup_from_cont(old_cont);
1190 
1191 	/*
1192 	 * Only thread group leaders are allowed to migrate, the mm_struct is
1193 	 * in effect owned by the leader
1194 	 */
1195 	if (!thread_group_leader(p))
1196 		goto out;
1197 
1198 out:
1199 	mmput(mm);
1200 }
1201 
1202 struct cgroup_subsys mem_cgroup_subsys = {
1203 	.name = "memory",
1204 	.subsys_id = mem_cgroup_subsys_id,
1205 	.create = mem_cgroup_create,
1206 	.pre_destroy = mem_cgroup_pre_destroy,
1207 	.destroy = mem_cgroup_destroy,
1208 	.populate = mem_cgroup_populate,
1209 	.attach = mem_cgroup_move_task,
1210 	.early_init = 0,
1211 };
1212