1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28#include <linux/page_counter.h>
29#include <linux/memcontrol.h>
30#include <linux/cgroup.h>
31#include <linux/mm.h>
32#include <linux/hugetlb.h>
33#include <linux/pagemap.h>
34#include <linux/smp.h>
35#include <linux/page-flags.h>
36#include <linux/backing-dev.h>
37#include <linux/bit_spinlock.h>
38#include <linux/rcupdate.h>
39#include <linux/limits.h>
40#include <linux/export.h>
41#include <linux/mutex.h>
42#include <linux/rbtree.h>
43#include <linux/slab.h>
44#include <linux/swap.h>
45#include <linux/swapops.h>
46#include <linux/spinlock.h>
47#include <linux/eventfd.h>
48#include <linux/poll.h>
49#include <linux/sort.h>
50#include <linux/fs.h>
51#include <linux/seq_file.h>
52#include <linux/vmpressure.h>
53#include <linux/mm_inline.h>
54#include <linux/swap_cgroup.h>
55#include <linux/cpu.h>
56#include <linux/oom.h>
57#include <linux/lockdep.h>
58#include <linux/file.h>
59#include "internal.h"
60#include <net/sock.h>
61#include <net/ip.h>
62#include <net/tcp_memcontrol.h>
63#include "slab.h"
64
65#include <asm/uaccess.h>
66
67#include <trace/events/vmscan.h>
68
69struct cgroup_subsys memory_cgrp_subsys __read_mostly;
70EXPORT_SYMBOL(memory_cgrp_subsys);
71
72#define MEM_CGROUP_RECLAIM_RETRIES 5
73static struct mem_cgroup *root_mem_cgroup __read_mostly;
74
75
76#ifdef CONFIG_MEMCG_SWAP
77int do_swap_account __read_mostly;
78#else
79#define do_swap_account 0
80#endif
81
82static const char * const mem_cgroup_stat_names[] = {
83 "cache",
84 "rss",
85 "rss_huge",
86 "mapped_file",
87 "writeback",
88 "swap",
89};
90
91static const char * const mem_cgroup_events_names[] = {
92 "pgpgin",
93 "pgpgout",
94 "pgfault",
95 "pgmajfault",
96};
97
98static const char * const mem_cgroup_lru_names[] = {
99 "inactive_anon",
100 "active_anon",
101 "inactive_file",
102 "active_file",
103 "unevictable",
104};
105
106
107
108
109
110
111
112enum mem_cgroup_events_target {
113 MEM_CGROUP_TARGET_THRESH,
114 MEM_CGROUP_TARGET_SOFTLIMIT,
115 MEM_CGROUP_TARGET_NUMAINFO,
116 MEM_CGROUP_NTARGETS,
117};
118#define THRESHOLDS_EVENTS_TARGET 128
119#define SOFTLIMIT_EVENTS_TARGET 1024
120#define NUMAINFO_EVENTS_TARGET 1024
121
122struct mem_cgroup_stat_cpu {
123 long count[MEM_CGROUP_STAT_NSTATS];
124 unsigned long events[MEMCG_NR_EVENTS];
125 unsigned long nr_page_events;
126 unsigned long targets[MEM_CGROUP_NTARGETS];
127};
128
129struct reclaim_iter {
130 struct mem_cgroup *position;
131
132 unsigned int generation;
133};
134
135
136
137
138struct mem_cgroup_per_zone {
139 struct lruvec lruvec;
140 unsigned long lru_size[NR_LRU_LISTS];
141
142 struct reclaim_iter iter[DEF_PRIORITY + 1];
143
144 struct rb_node tree_node;
145 unsigned long usage_in_excess;
146
147 bool on_tree;
148 struct mem_cgroup *memcg;
149
150};
151
152struct mem_cgroup_per_node {
153 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
154};
155
156
157
158
159
160
161struct mem_cgroup_tree_per_zone {
162 struct rb_root rb_root;
163 spinlock_t lock;
164};
165
166struct mem_cgroup_tree_per_node {
167 struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
168};
169
170struct mem_cgroup_tree {
171 struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
172};
173
174static struct mem_cgroup_tree soft_limit_tree __read_mostly;
175
176struct mem_cgroup_threshold {
177 struct eventfd_ctx *eventfd;
178 unsigned long threshold;
179};
180
181
182struct mem_cgroup_threshold_ary {
183
184 int current_threshold;
185
186 unsigned int size;
187
188 struct mem_cgroup_threshold entries[0];
189};
190
191struct mem_cgroup_thresholds {
192
193 struct mem_cgroup_threshold_ary *primary;
194
195
196
197
198
199 struct mem_cgroup_threshold_ary *spare;
200};
201
202
203struct mem_cgroup_eventfd_list {
204 struct list_head list;
205 struct eventfd_ctx *eventfd;
206};
207
208
209
210
211struct mem_cgroup_event {
212
213
214
215 struct mem_cgroup *memcg;
216
217
218
219 struct eventfd_ctx *eventfd;
220
221
222
223 struct list_head list;
224
225
226
227
228
229 int (*register_event)(struct mem_cgroup *memcg,
230 struct eventfd_ctx *eventfd, const char *args);
231
232
233
234
235
236 void (*unregister_event)(struct mem_cgroup *memcg,
237 struct eventfd_ctx *eventfd);
238
239
240
241
242 poll_table pt;
243 wait_queue_head_t *wqh;
244 wait_queue_t wait;
245 struct work_struct remove;
246};
247
248static void mem_cgroup_threshold(struct mem_cgroup *memcg);
249static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
250
251
252
253
254
255
256
257
258
259
260
261
262struct mem_cgroup {
263 struct cgroup_subsys_state css;
264
265
266 struct page_counter memory;
267 struct page_counter memsw;
268 struct page_counter kmem;
269
270
271 unsigned long low;
272 unsigned long high;
273
274 unsigned long soft_limit;
275
276
277 struct vmpressure vmpressure;
278
279
280 int initialized;
281
282
283
284
285 bool use_hierarchy;
286
287 bool oom_lock;
288 atomic_t under_oom;
289 atomic_t oom_wakeups;
290
291 int swappiness;
292
293 int oom_kill_disable;
294
295
296 struct mutex thresholds_lock;
297
298
299 struct mem_cgroup_thresholds thresholds;
300
301
302 struct mem_cgroup_thresholds memsw_thresholds;
303
304
305 struct list_head oom_notify;
306
307
308
309
310
311 unsigned long move_charge_at_immigrate;
312
313
314
315 atomic_t moving_account;
316
317 spinlock_t move_lock;
318 struct task_struct *move_lock_task;
319 unsigned long move_lock_flags;
320
321
322
323 struct mem_cgroup_stat_cpu __percpu *stat;
324
325
326
327
328 struct mem_cgroup_stat_cpu nocpu_base;
329 spinlock_t pcp_counter_lock;
330
331#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
332 struct cg_proto tcp_mem;
333#endif
334#if defined(CONFIG_MEMCG_KMEM)
335
336 int kmemcg_id;
337 bool kmem_acct_activated;
338 bool kmem_acct_active;
339#endif
340
341 int last_scanned_node;
342#if MAX_NUMNODES > 1
343 nodemask_t scan_nodes;
344 atomic_t numainfo_events;
345 atomic_t numainfo_updating;
346#endif
347
348
349 struct list_head event_list;
350 spinlock_t event_list_lock;
351
352 struct mem_cgroup_per_node *nodeinfo[0];
353
354};
355
356#ifdef CONFIG_MEMCG_KMEM
357bool memcg_kmem_is_active(struct mem_cgroup *memcg)
358{
359 return memcg->kmem_acct_active;
360}
361#endif
362
363
364
365
366
367#define MOVE_ANON 0x1U
368#define MOVE_FILE 0x2U
369#define MOVE_MASK (MOVE_ANON | MOVE_FILE)
370
371
372static struct move_charge_struct {
373 spinlock_t lock;
374 struct mem_cgroup *from;
375 struct mem_cgroup *to;
376 unsigned long flags;
377 unsigned long precharge;
378 unsigned long moved_charge;
379 unsigned long moved_swap;
380 struct task_struct *moving_task;
381 wait_queue_head_t waitq;
382} mc = {
383 .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
384 .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
385};
386
387
388
389
390
391#define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
392#define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
393
394enum charge_type {
395 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
396 MEM_CGROUP_CHARGE_TYPE_ANON,
397 MEM_CGROUP_CHARGE_TYPE_SWAPOUT,
398 MEM_CGROUP_CHARGE_TYPE_DROP,
399 NR_CHARGE_TYPE,
400};
401
402
403enum res_type {
404 _MEM,
405 _MEMSWAP,
406 _OOM_TYPE,
407 _KMEM,
408};
409
410#define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
411#define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
412#define MEMFILE_ATTR(val) ((val) & 0xffff)
413
414#define OOM_CONTROL (0)
415
416
417
418
419
420
421static DEFINE_MUTEX(memcg_create_mutex);
422
423struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
424{
425 return s ? container_of(s, struct mem_cgroup, css) : NULL;
426}
427
428
429struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
430{
431 if (!memcg)
432 memcg = root_mem_cgroup;
433 return &memcg->vmpressure;
434}
435
436struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
437{
438 return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
439}
440
441static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
442{
443 return (memcg == root_mem_cgroup);
444}
445
446
447
448
449
450#define MEM_CGROUP_ID_MAX USHRT_MAX
451
452static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
453{
454 return memcg->css.id;
455}
456
457static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
458{
459 struct cgroup_subsys_state *css;
460
461 css = css_from_id(id, &memory_cgrp_subsys);
462 return mem_cgroup_from_css(css);
463}
464
465
466#if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
467
468void sock_update_memcg(struct sock *sk)
469{
470 if (mem_cgroup_sockets_enabled) {
471 struct mem_cgroup *memcg;
472 struct cg_proto *cg_proto;
473
474 BUG_ON(!sk->sk_prot->proto_cgroup);
475
476
477
478
479
480
481
482
483
484 if (sk->sk_cgrp) {
485 BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
486 css_get(&sk->sk_cgrp->memcg->css);
487 return;
488 }
489
490 rcu_read_lock();
491 memcg = mem_cgroup_from_task(current);
492 cg_proto = sk->sk_prot->proto_cgroup(memcg);
493 if (!mem_cgroup_is_root(memcg) &&
494 memcg_proto_active(cg_proto) &&
495 css_tryget_online(&memcg->css)) {
496 sk->sk_cgrp = cg_proto;
497 }
498 rcu_read_unlock();
499 }
500}
501EXPORT_SYMBOL(sock_update_memcg);
502
503void sock_release_memcg(struct sock *sk)
504{
505 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
506 struct mem_cgroup *memcg;
507 WARN_ON(!sk->sk_cgrp->memcg);
508 memcg = sk->sk_cgrp->memcg;
509 css_put(&sk->sk_cgrp->memcg->css);
510 }
511}
512
513struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
514{
515 if (!memcg || mem_cgroup_is_root(memcg))
516 return NULL;
517
518 return &memcg->tcp_mem;
519}
520EXPORT_SYMBOL(tcp_proto_cgroup);
521
522#endif
523
524#ifdef CONFIG_MEMCG_KMEM
525
526
527
528
529
530
531
532
533
534
535
536static DEFINE_IDA(memcg_cache_ida);
537int memcg_nr_cache_ids;
538
539
540static DECLARE_RWSEM(memcg_cache_ids_sem);
541
542void memcg_get_cache_ids(void)
543{
544 down_read(&memcg_cache_ids_sem);
545}
546
547void memcg_put_cache_ids(void)
548{
549 up_read(&memcg_cache_ids_sem);
550}
551
552
553
554
555
556
557
558
559
560
561
562
563
564#define MEMCG_CACHES_MIN_SIZE 4
565#define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
566
567
568
569
570
571
572
573struct static_key memcg_kmem_enabled_key;
574EXPORT_SYMBOL(memcg_kmem_enabled_key);
575
576#endif
577
578static struct mem_cgroup_per_zone *
579mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
580{
581 int nid = zone_to_nid(zone);
582 int zid = zone_idx(zone);
583
584 return &memcg->nodeinfo[nid]->zoneinfo[zid];
585}
586
587struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
588{
589 return &memcg->css;
590}
591
592static struct mem_cgroup_per_zone *
593mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
594{
595 int nid = page_to_nid(page);
596 int zid = page_zonenum(page);
597
598 return &memcg->nodeinfo[nid]->zoneinfo[zid];
599}
600
601static struct mem_cgroup_tree_per_zone *
602soft_limit_tree_node_zone(int nid, int zid)
603{
604 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
605}
606
607static struct mem_cgroup_tree_per_zone *
608soft_limit_tree_from_page(struct page *page)
609{
610 int nid = page_to_nid(page);
611 int zid = page_zonenum(page);
612
613 return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
614}
615
616static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
617 struct mem_cgroup_tree_per_zone *mctz,
618 unsigned long new_usage_in_excess)
619{
620 struct rb_node **p = &mctz->rb_root.rb_node;
621 struct rb_node *parent = NULL;
622 struct mem_cgroup_per_zone *mz_node;
623
624 if (mz->on_tree)
625 return;
626
627 mz->usage_in_excess = new_usage_in_excess;
628 if (!mz->usage_in_excess)
629 return;
630 while (*p) {
631 parent = *p;
632 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
633 tree_node);
634 if (mz->usage_in_excess < mz_node->usage_in_excess)
635 p = &(*p)->rb_left;
636
637
638
639
640 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
641 p = &(*p)->rb_right;
642 }
643 rb_link_node(&mz->tree_node, parent, p);
644 rb_insert_color(&mz->tree_node, &mctz->rb_root);
645 mz->on_tree = true;
646}
647
648static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
649 struct mem_cgroup_tree_per_zone *mctz)
650{
651 if (!mz->on_tree)
652 return;
653 rb_erase(&mz->tree_node, &mctz->rb_root);
654 mz->on_tree = false;
655}
656
657static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
658 struct mem_cgroup_tree_per_zone *mctz)
659{
660 unsigned long flags;
661
662 spin_lock_irqsave(&mctz->lock, flags);
663 __mem_cgroup_remove_exceeded(mz, mctz);
664 spin_unlock_irqrestore(&mctz->lock, flags);
665}
666
667static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
668{
669 unsigned long nr_pages = page_counter_read(&memcg->memory);
670 unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit);
671 unsigned long excess = 0;
672
673 if (nr_pages > soft_limit)
674 excess = nr_pages - soft_limit;
675
676 return excess;
677}
678
679static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
680{
681 unsigned long excess;
682 struct mem_cgroup_per_zone *mz;
683 struct mem_cgroup_tree_per_zone *mctz;
684
685 mctz = soft_limit_tree_from_page(page);
686
687
688
689
690 for (; memcg; memcg = parent_mem_cgroup(memcg)) {
691 mz = mem_cgroup_page_zoneinfo(memcg, page);
692 excess = soft_limit_excess(memcg);
693
694
695
696
697 if (excess || mz->on_tree) {
698 unsigned long flags;
699
700 spin_lock_irqsave(&mctz->lock, flags);
701
702 if (mz->on_tree)
703 __mem_cgroup_remove_exceeded(mz, mctz);
704
705
706
707
708 __mem_cgroup_insert_exceeded(mz, mctz, excess);
709 spin_unlock_irqrestore(&mctz->lock, flags);
710 }
711 }
712}
713
714static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
715{
716 struct mem_cgroup_tree_per_zone *mctz;
717 struct mem_cgroup_per_zone *mz;
718 int nid, zid;
719
720 for_each_node(nid) {
721 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
722 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
723 mctz = soft_limit_tree_node_zone(nid, zid);
724 mem_cgroup_remove_exceeded(mz, mctz);
725 }
726 }
727}
728
729static struct mem_cgroup_per_zone *
730__mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
731{
732 struct rb_node *rightmost = NULL;
733 struct mem_cgroup_per_zone *mz;
734
735retry:
736 mz = NULL;
737 rightmost = rb_last(&mctz->rb_root);
738 if (!rightmost)
739 goto done;
740
741 mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
742
743
744
745
746
747 __mem_cgroup_remove_exceeded(mz, mctz);
748 if (!soft_limit_excess(mz->memcg) ||
749 !css_tryget_online(&mz->memcg->css))
750 goto retry;
751done:
752 return mz;
753}
754
755static struct mem_cgroup_per_zone *
756mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
757{
758 struct mem_cgroup_per_zone *mz;
759
760 spin_lock_irq(&mctz->lock);
761 mz = __mem_cgroup_largest_soft_limit_node(mctz);
762 spin_unlock_irq(&mctz->lock);
763 return mz;
764}
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
786 enum mem_cgroup_stat_index idx)
787{
788 long val = 0;
789 int cpu;
790
791 get_online_cpus();
792 for_each_online_cpu(cpu)
793 val += per_cpu(memcg->stat->count[idx], cpu);
794#ifdef CONFIG_HOTPLUG_CPU
795 spin_lock(&memcg->pcp_counter_lock);
796 val += memcg->nocpu_base.count[idx];
797 spin_unlock(&memcg->pcp_counter_lock);
798#endif
799 put_online_cpus();
800 return val;
801}
802
803static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
804 enum mem_cgroup_events_index idx)
805{
806 unsigned long val = 0;
807 int cpu;
808
809 get_online_cpus();
810 for_each_online_cpu(cpu)
811 val += per_cpu(memcg->stat->events[idx], cpu);
812#ifdef CONFIG_HOTPLUG_CPU
813 spin_lock(&memcg->pcp_counter_lock);
814 val += memcg->nocpu_base.events[idx];
815 spin_unlock(&memcg->pcp_counter_lock);
816#endif
817 put_online_cpus();
818 return val;
819}
820
821static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
822 struct page *page,
823 int nr_pages)
824{
825
826
827
828
829 if (PageAnon(page))
830 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
831 nr_pages);
832 else
833 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
834 nr_pages);
835
836 if (PageTransHuge(page))
837 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
838 nr_pages);
839
840
841 if (nr_pages > 0)
842 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
843 else {
844 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
845 nr_pages = -nr_pages;
846 }
847
848 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
849}
850
851unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
852{
853 struct mem_cgroup_per_zone *mz;
854
855 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
856 return mz->lru_size[lru];
857}
858
859static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
860 int nid,
861 unsigned int lru_mask)
862{
863 unsigned long nr = 0;
864 int zid;
865
866 VM_BUG_ON((unsigned)nid >= nr_node_ids);
867
868 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
869 struct mem_cgroup_per_zone *mz;
870 enum lru_list lru;
871
872 for_each_lru(lru) {
873 if (!(BIT(lru) & lru_mask))
874 continue;
875 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
876 nr += mz->lru_size[lru];
877 }
878 }
879 return nr;
880}
881
882static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
883 unsigned int lru_mask)
884{
885 unsigned long nr = 0;
886 int nid;
887
888 for_each_node_state(nid, N_MEMORY)
889 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
890 return nr;
891}
892
893static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
894 enum mem_cgroup_events_target target)
895{
896 unsigned long val, next;
897
898 val = __this_cpu_read(memcg->stat->nr_page_events);
899 next = __this_cpu_read(memcg->stat->targets[target]);
900
901 if ((long)next - (long)val < 0) {
902 switch (target) {
903 case MEM_CGROUP_TARGET_THRESH:
904 next = val + THRESHOLDS_EVENTS_TARGET;
905 break;
906 case MEM_CGROUP_TARGET_SOFTLIMIT:
907 next = val + SOFTLIMIT_EVENTS_TARGET;
908 break;
909 case MEM_CGROUP_TARGET_NUMAINFO:
910 next = val + NUMAINFO_EVENTS_TARGET;
911 break;
912 default:
913 break;
914 }
915 __this_cpu_write(memcg->stat->targets[target], next);
916 return true;
917 }
918 return false;
919}
920
921
922
923
924
925static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
926{
927
928 if (unlikely(mem_cgroup_event_ratelimit(memcg,
929 MEM_CGROUP_TARGET_THRESH))) {
930 bool do_softlimit;
931 bool do_numainfo __maybe_unused;
932
933 do_softlimit = mem_cgroup_event_ratelimit(memcg,
934 MEM_CGROUP_TARGET_SOFTLIMIT);
935#if MAX_NUMNODES > 1
936 do_numainfo = mem_cgroup_event_ratelimit(memcg,
937 MEM_CGROUP_TARGET_NUMAINFO);
938#endif
939 mem_cgroup_threshold(memcg);
940 if (unlikely(do_softlimit))
941 mem_cgroup_update_tree(memcg, page);
942#if MAX_NUMNODES > 1
943 if (unlikely(do_numainfo))
944 atomic_inc(&memcg->numainfo_events);
945#endif
946 }
947}
948
949struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
950{
951
952
953
954
955
956 if (unlikely(!p))
957 return NULL;
958
959 return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
960}
961
962static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
963{
964 struct mem_cgroup *memcg = NULL;
965
966 rcu_read_lock();
967 do {
968
969
970
971
972
973 if (unlikely(!mm))
974 memcg = root_mem_cgroup;
975 else {
976 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
977 if (unlikely(!memcg))
978 memcg = root_mem_cgroup;
979 }
980 } while (!css_tryget_online(&memcg->css));
981 rcu_read_unlock();
982 return memcg;
983}
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
1003 struct mem_cgroup *prev,
1004 struct mem_cgroup_reclaim_cookie *reclaim)
1005{
1006 struct reclaim_iter *uninitialized_var(iter);
1007 struct cgroup_subsys_state *css = NULL;
1008 struct mem_cgroup *memcg = NULL;
1009 struct mem_cgroup *pos = NULL;
1010
1011 if (mem_cgroup_disabled())
1012 return NULL;
1013
1014 if (!root)
1015 root = root_mem_cgroup;
1016
1017 if (prev && !reclaim)
1018 pos = prev;
1019
1020 if (!root->use_hierarchy && root != root_mem_cgroup) {
1021 if (prev)
1022 goto out;
1023 return root;
1024 }
1025
1026 rcu_read_lock();
1027
1028 if (reclaim) {
1029 struct mem_cgroup_per_zone *mz;
1030
1031 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
1032 iter = &mz->iter[reclaim->priority];
1033
1034 if (prev && reclaim->generation != iter->generation)
1035 goto out_unlock;
1036
1037 do {
1038 pos = ACCESS_ONCE(iter->position);
1039
1040
1041
1042
1043
1044 } while (pos && !css_tryget(&pos->css));
1045 }
1046
1047 if (pos)
1048 css = &pos->css;
1049
1050 for (;;) {
1051 css = css_next_descendant_pre(css, &root->css);
1052 if (!css) {
1053
1054
1055
1056
1057
1058
1059 if (!prev)
1060 continue;
1061 break;
1062 }
1063
1064
1065
1066
1067
1068
1069 memcg = mem_cgroup_from_css(css);
1070
1071 if (css == &root->css)
1072 break;
1073
1074 if (css_tryget(css)) {
1075
1076
1077
1078
1079
1080 if (smp_load_acquire(&memcg->initialized))
1081 break;
1082
1083 css_put(css);
1084 }
1085
1086 memcg = NULL;
1087 }
1088
1089 if (reclaim) {
1090 if (cmpxchg(&iter->position, pos, memcg) == pos) {
1091 if (memcg)
1092 css_get(&memcg->css);
1093 if (pos)
1094 css_put(&pos->css);
1095 }
1096
1097
1098
1099
1100
1101 if (pos)
1102 css_put(&pos->css);
1103
1104 if (!memcg)
1105 iter->generation++;
1106 else if (!prev)
1107 reclaim->generation = iter->generation;
1108 }
1109
1110out_unlock:
1111 rcu_read_unlock();
1112out:
1113 if (prev && prev != root)
1114 css_put(&prev->css);
1115
1116 return memcg;
1117}
1118
1119
1120
1121
1122
1123
1124void mem_cgroup_iter_break(struct mem_cgroup *root,
1125 struct mem_cgroup *prev)
1126{
1127 if (!root)
1128 root = root_mem_cgroup;
1129 if (prev && prev != root)
1130 css_put(&prev->css);
1131}
1132
1133
1134
1135
1136
1137
1138#define for_each_mem_cgroup_tree(iter, root) \
1139 for (iter = mem_cgroup_iter(root, NULL, NULL); \
1140 iter != NULL; \
1141 iter = mem_cgroup_iter(root, iter, NULL))
1142
1143#define for_each_mem_cgroup(iter) \
1144 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1145 iter != NULL; \
1146 iter = mem_cgroup_iter(NULL, iter, NULL))
1147
1148void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
1149{
1150 struct mem_cgroup *memcg;
1151
1152 rcu_read_lock();
1153 memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1154 if (unlikely(!memcg))
1155 goto out;
1156
1157 switch (idx) {
1158 case PGFAULT:
1159 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1160 break;
1161 case PGMAJFAULT:
1162 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
1163 break;
1164 default:
1165 BUG();
1166 }
1167out:
1168 rcu_read_unlock();
1169}
1170EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1182 struct mem_cgroup *memcg)
1183{
1184 struct mem_cgroup_per_zone *mz;
1185 struct lruvec *lruvec;
1186
1187 if (mem_cgroup_disabled()) {
1188 lruvec = &zone->lruvec;
1189 goto out;
1190 }
1191
1192 mz = mem_cgroup_zone_zoneinfo(memcg, zone);
1193 lruvec = &mz->lruvec;
1194out:
1195
1196
1197
1198
1199
1200 if (unlikely(lruvec->zone != zone))
1201 lruvec->zone = zone;
1202 return lruvec;
1203}
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
1215{
1216 struct mem_cgroup_per_zone *mz;
1217 struct mem_cgroup *memcg;
1218 struct lruvec *lruvec;
1219
1220 if (mem_cgroup_disabled()) {
1221 lruvec = &zone->lruvec;
1222 goto out;
1223 }
1224
1225 memcg = page->mem_cgroup;
1226
1227
1228
1229
1230 if (!memcg)
1231 memcg = root_mem_cgroup;
1232
1233 mz = mem_cgroup_page_zoneinfo(memcg, page);
1234 lruvec = &mz->lruvec;
1235out:
1236
1237
1238
1239
1240
1241 if (unlikely(lruvec->zone != zone))
1242 lruvec->zone = zone;
1243 return lruvec;
1244}
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1256 int nr_pages)
1257{
1258 struct mem_cgroup_per_zone *mz;
1259 unsigned long *lru_size;
1260
1261 if (mem_cgroup_disabled())
1262 return;
1263
1264 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1265 lru_size = mz->lru_size + lru;
1266 *lru_size += nr_pages;
1267 VM_BUG_ON((long)(*lru_size) < 0);
1268}
1269
1270bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root)
1271{
1272 if (root == memcg)
1273 return true;
1274 if (!root->use_hierarchy)
1275 return false;
1276 return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
1277}
1278
1279bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
1280{
1281 struct mem_cgroup *task_memcg;
1282 struct task_struct *p;
1283 bool ret;
1284
1285 p = find_lock_task_mm(task);
1286 if (p) {
1287 task_memcg = get_mem_cgroup_from_mm(p->mm);
1288 task_unlock(p);
1289 } else {
1290
1291
1292
1293
1294
1295 rcu_read_lock();
1296 task_memcg = mem_cgroup_from_task(task);
1297 css_get(&task_memcg->css);
1298 rcu_read_unlock();
1299 }
1300 ret = mem_cgroup_is_descendant(task_memcg, memcg);
1301 css_put(&task_memcg->css);
1302 return ret;
1303}
1304
1305int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
1306{
1307 unsigned long inactive_ratio;
1308 unsigned long inactive;
1309 unsigned long active;
1310 unsigned long gb;
1311
1312 inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1313 active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
1314
1315 gb = (inactive + active) >> (30 - PAGE_SHIFT);
1316 if (gb)
1317 inactive_ratio = int_sqrt(10 * gb);
1318 else
1319 inactive_ratio = 1;
1320
1321 return inactive * inactive_ratio < active;
1322}
1323
1324bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
1325{
1326 struct mem_cgroup_per_zone *mz;
1327 struct mem_cgroup *memcg;
1328
1329 if (mem_cgroup_disabled())
1330 return true;
1331
1332 mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1333 memcg = mz->memcg;
1334
1335 return !!(memcg->css.flags & CSS_ONLINE);
1336}
1337
1338#define mem_cgroup_from_counter(counter, member) \
1339 container_of(counter, struct mem_cgroup, member)
1340
1341
1342
1343
1344
1345
1346
1347
1348static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1349{
1350 unsigned long margin = 0;
1351 unsigned long count;
1352 unsigned long limit;
1353
1354 count = page_counter_read(&memcg->memory);
1355 limit = ACCESS_ONCE(memcg->memory.limit);
1356 if (count < limit)
1357 margin = limit - count;
1358
1359 if (do_swap_account) {
1360 count = page_counter_read(&memcg->memsw);
1361 limit = ACCESS_ONCE(memcg->memsw.limit);
1362 if (count <= limit)
1363 margin = min(margin, limit - count);
1364 }
1365
1366 return margin;
1367}
1368
1369int mem_cgroup_swappiness(struct mem_cgroup *memcg)
1370{
1371
1372 if (mem_cgroup_disabled() || !memcg->css.parent)
1373 return vm_swappiness;
1374
1375 return memcg->swappiness;
1376}
1377
1378
1379
1380
1381
1382
1383
1384
1385static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1386{
1387 struct mem_cgroup *from;
1388 struct mem_cgroup *to;
1389 bool ret = false;
1390
1391
1392
1393
1394 spin_lock(&mc.lock);
1395 from = mc.from;
1396 to = mc.to;
1397 if (!from)
1398 goto unlock;
1399
1400 ret = mem_cgroup_is_descendant(from, memcg) ||
1401 mem_cgroup_is_descendant(to, memcg);
1402unlock:
1403 spin_unlock(&mc.lock);
1404 return ret;
1405}
1406
1407static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1408{
1409 if (mc.moving_task && current != mc.moving_task) {
1410 if (mem_cgroup_under_move(memcg)) {
1411 DEFINE_WAIT(wait);
1412 prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1413
1414 if (mc.moving_task)
1415 schedule();
1416 finish_wait(&mc.waitq, &wait);
1417 return true;
1418 }
1419 }
1420 return false;
1421}
1422
1423#define K(x) ((x) << (PAGE_SHIFT-10))
1424
1425
1426
1427
1428
1429
1430
1431
1432void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1433{
1434
1435 static DEFINE_MUTEX(oom_info_lock);
1436 struct mem_cgroup *iter;
1437 unsigned int i;
1438
1439 if (!p)
1440 return;
1441
1442 mutex_lock(&oom_info_lock);
1443 rcu_read_lock();
1444
1445 pr_info("Task in ");
1446 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1447 pr_cont(" killed as a result of limit of ");
1448 pr_cont_cgroup_path(memcg->css.cgroup);
1449 pr_cont("\n");
1450
1451 rcu_read_unlock();
1452
1453 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1454 K((u64)page_counter_read(&memcg->memory)),
1455 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1456 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1457 K((u64)page_counter_read(&memcg->memsw)),
1458 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1459 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1460 K((u64)page_counter_read(&memcg->kmem)),
1461 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
1462
1463 for_each_mem_cgroup_tree(iter, memcg) {
1464 pr_info("Memory cgroup stats for ");
1465 pr_cont_cgroup_path(iter->css.cgroup);
1466 pr_cont(":");
1467
1468 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1469 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1470 continue;
1471 pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1472 K(mem_cgroup_read_stat(iter, i)));
1473 }
1474
1475 for (i = 0; i < NR_LRU_LISTS; i++)
1476 pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1477 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1478
1479 pr_cont("\n");
1480 }
1481 mutex_unlock(&oom_info_lock);
1482}
1483
1484
1485
1486
1487
1488static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1489{
1490 int num = 0;
1491 struct mem_cgroup *iter;
1492
1493 for_each_mem_cgroup_tree(iter, memcg)
1494 num++;
1495 return num;
1496}
1497
1498
1499
1500
1501static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
1502{
1503 unsigned long limit;
1504
1505 limit = memcg->memory.limit;
1506 if (mem_cgroup_swappiness(memcg)) {
1507 unsigned long memsw_limit;
1508
1509 memsw_limit = memcg->memsw.limit;
1510 limit = min(limit + total_swap_pages, memsw_limit);
1511 }
1512 return limit;
1513}
1514
1515static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1516 int order)
1517{
1518 struct mem_cgroup *iter;
1519 unsigned long chosen_points = 0;
1520 unsigned long totalpages;
1521 unsigned int points = 0;
1522 struct task_struct *chosen = NULL;
1523
1524
1525
1526
1527
1528
1529 if (fatal_signal_pending(current) || task_will_free_mem(current)) {
1530 mark_tsk_oom_victim(current);
1531 return;
1532 }
1533
1534 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
1535 totalpages = mem_cgroup_get_limit(memcg) ? : 1;
1536 for_each_mem_cgroup_tree(iter, memcg) {
1537 struct css_task_iter it;
1538 struct task_struct *task;
1539
1540 css_task_iter_start(&iter->css, &it);
1541 while ((task = css_task_iter_next(&it))) {
1542 switch (oom_scan_process_thread(task, totalpages, NULL,
1543 false)) {
1544 case OOM_SCAN_SELECT:
1545 if (chosen)
1546 put_task_struct(chosen);
1547 chosen = task;
1548 chosen_points = ULONG_MAX;
1549 get_task_struct(chosen);
1550
1551 case OOM_SCAN_CONTINUE:
1552 continue;
1553 case OOM_SCAN_ABORT:
1554 css_task_iter_end(&it);
1555 mem_cgroup_iter_break(memcg, iter);
1556 if (chosen)
1557 put_task_struct(chosen);
1558 return;
1559 case OOM_SCAN_OK:
1560 break;
1561 };
1562 points = oom_badness(task, memcg, NULL, totalpages);
1563 if (!points || points < chosen_points)
1564 continue;
1565
1566 if (points == chosen_points &&
1567 thread_group_leader(chosen))
1568 continue;
1569
1570 if (chosen)
1571 put_task_struct(chosen);
1572 chosen = task;
1573 chosen_points = points;
1574 get_task_struct(chosen);
1575 }
1576 css_task_iter_end(&it);
1577 }
1578
1579 if (!chosen)
1580 return;
1581 points = chosen_points * 1000 / totalpages;
1582 oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1583 NULL, "Memory cgroup out of memory");
1584}
1585
1586#if MAX_NUMNODES > 1
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1599 int nid, bool noswap)
1600{
1601 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1602 return true;
1603 if (noswap || !total_swap_pages)
1604 return false;
1605 if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1606 return true;
1607 return false;
1608
1609}
1610
1611
1612
1613
1614
1615
1616
1617static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1618{
1619 int nid;
1620
1621
1622
1623
1624 if (!atomic_read(&memcg->numainfo_events))
1625 return;
1626 if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1627 return;
1628
1629
1630 memcg->scan_nodes = node_states[N_MEMORY];
1631
1632 for_each_node_mask(nid, node_states[N_MEMORY]) {
1633
1634 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1635 node_clear(nid, memcg->scan_nodes);
1636 }
1637
1638 atomic_set(&memcg->numainfo_events, 0);
1639 atomic_set(&memcg->numainfo_updating, 0);
1640}
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1655{
1656 int node;
1657
1658 mem_cgroup_may_update_nodemask(memcg);
1659 node = memcg->last_scanned_node;
1660
1661 node = next_node(node, memcg->scan_nodes);
1662 if (node == MAX_NUMNODES)
1663 node = first_node(memcg->scan_nodes);
1664
1665
1666
1667
1668
1669
1670 if (unlikely(node == MAX_NUMNODES))
1671 node = numa_node_id();
1672
1673 memcg->last_scanned_node = node;
1674 return node;
1675}
1676#else
1677int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1678{
1679 return 0;
1680}
1681#endif
1682
1683static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1684 struct zone *zone,
1685 gfp_t gfp_mask,
1686 unsigned long *total_scanned)
1687{
1688 struct mem_cgroup *victim = NULL;
1689 int total = 0;
1690 int loop = 0;
1691 unsigned long excess;
1692 unsigned long nr_scanned;
1693 struct mem_cgroup_reclaim_cookie reclaim = {
1694 .zone = zone,
1695 .priority = 0,
1696 };
1697
1698 excess = soft_limit_excess(root_memcg);
1699
1700 while (1) {
1701 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1702 if (!victim) {
1703 loop++;
1704 if (loop >= 2) {
1705
1706
1707
1708
1709
1710 if (!total)
1711 break;
1712
1713
1714
1715
1716
1717
1718 if (total >= (excess >> 2) ||
1719 (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1720 break;
1721 }
1722 continue;
1723 }
1724 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1725 zone, &nr_scanned);
1726 *total_scanned += nr_scanned;
1727 if (!soft_limit_excess(root_memcg))
1728 break;
1729 }
1730 mem_cgroup_iter_break(root_memcg, victim);
1731 return total;
1732}
1733
1734#ifdef CONFIG_LOCKDEP
1735static struct lockdep_map memcg_oom_lock_dep_map = {
1736 .name = "memcg_oom_lock",
1737};
1738#endif
1739
1740static DEFINE_SPINLOCK(memcg_oom_lock);
1741
1742
1743
1744
1745
1746static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
1747{
1748 struct mem_cgroup *iter, *failed = NULL;
1749
1750 spin_lock(&memcg_oom_lock);
1751
1752 for_each_mem_cgroup_tree(iter, memcg) {
1753 if (iter->oom_lock) {
1754
1755
1756
1757
1758 failed = iter;
1759 mem_cgroup_iter_break(memcg, iter);
1760 break;
1761 } else
1762 iter->oom_lock = true;
1763 }
1764
1765 if (failed) {
1766
1767
1768
1769
1770 for_each_mem_cgroup_tree(iter, memcg) {
1771 if (iter == failed) {
1772 mem_cgroup_iter_break(memcg, iter);
1773 break;
1774 }
1775 iter->oom_lock = false;
1776 }
1777 } else
1778 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
1779
1780 spin_unlock(&memcg_oom_lock);
1781
1782 return !failed;
1783}
1784
1785static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1786{
1787 struct mem_cgroup *iter;
1788
1789 spin_lock(&memcg_oom_lock);
1790 mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
1791 for_each_mem_cgroup_tree(iter, memcg)
1792 iter->oom_lock = false;
1793 spin_unlock(&memcg_oom_lock);
1794}
1795
1796static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1797{
1798 struct mem_cgroup *iter;
1799
1800 for_each_mem_cgroup_tree(iter, memcg)
1801 atomic_inc(&iter->under_oom);
1802}
1803
1804static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1805{
1806 struct mem_cgroup *iter;
1807
1808
1809
1810
1811
1812
1813 for_each_mem_cgroup_tree(iter, memcg)
1814 atomic_add_unless(&iter->under_oom, -1, 0);
1815}
1816
1817static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1818
1819struct oom_wait_info {
1820 struct mem_cgroup *memcg;
1821 wait_queue_t wait;
1822};
1823
1824static int memcg_oom_wake_function(wait_queue_t *wait,
1825 unsigned mode, int sync, void *arg)
1826{
1827 struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1828 struct mem_cgroup *oom_wait_memcg;
1829 struct oom_wait_info *oom_wait_info;
1830
1831 oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1832 oom_wait_memcg = oom_wait_info->memcg;
1833
1834 if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1835 !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
1836 return 0;
1837 return autoremove_wake_function(wait, mode, sync, arg);
1838}
1839
1840static void memcg_wakeup_oom(struct mem_cgroup *memcg)
1841{
1842 atomic_inc(&memcg->oom_wakeups);
1843
1844 __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
1845}
1846
1847static void memcg_oom_recover(struct mem_cgroup *memcg)
1848{
1849 if (memcg && atomic_read(&memcg->under_oom))
1850 memcg_wakeup_oom(memcg);
1851}
1852
1853static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
1854{
1855 if (!current->memcg_oom.may_oom)
1856 return;
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871 css_get(&memcg->css);
1872 current->memcg_oom.memcg = memcg;
1873 current->memcg_oom.gfp_mask = mask;
1874 current->memcg_oom.order = order;
1875}
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894bool mem_cgroup_oom_synchronize(bool handle)
1895{
1896 struct mem_cgroup *memcg = current->memcg_oom.memcg;
1897 struct oom_wait_info owait;
1898 bool locked;
1899
1900
1901 if (!memcg)
1902 return false;
1903
1904 if (!handle || oom_killer_disabled)
1905 goto cleanup;
1906
1907 owait.memcg = memcg;
1908 owait.wait.flags = 0;
1909 owait.wait.func = memcg_oom_wake_function;
1910 owait.wait.private = current;
1911 INIT_LIST_HEAD(&owait.wait.task_list);
1912
1913 prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1914 mem_cgroup_mark_under_oom(memcg);
1915
1916 locked = mem_cgroup_oom_trylock(memcg);
1917
1918 if (locked)
1919 mem_cgroup_oom_notify(memcg);
1920
1921 if (locked && !memcg->oom_kill_disable) {
1922 mem_cgroup_unmark_under_oom(memcg);
1923 finish_wait(&memcg_oom_waitq, &owait.wait);
1924 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
1925 current->memcg_oom.order);
1926 } else {
1927 schedule();
1928 mem_cgroup_unmark_under_oom(memcg);
1929 finish_wait(&memcg_oom_waitq, &owait.wait);
1930 }
1931
1932 if (locked) {
1933 mem_cgroup_oom_unlock(memcg);
1934
1935
1936
1937
1938
1939 memcg_oom_recover(memcg);
1940 }
1941cleanup:
1942 current->memcg_oom.memcg = NULL;
1943 css_put(&memcg->css);
1944 return true;
1945}
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
1961{
1962 struct mem_cgroup *memcg;
1963 unsigned long flags;
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977 rcu_read_lock();
1978
1979 if (mem_cgroup_disabled())
1980 return NULL;
1981again:
1982 memcg = page->mem_cgroup;
1983 if (unlikely(!memcg))
1984 return NULL;
1985
1986 if (atomic_read(&memcg->moving_account) <= 0)
1987 return memcg;
1988
1989 spin_lock_irqsave(&memcg->move_lock, flags);
1990 if (memcg != page->mem_cgroup) {
1991 spin_unlock_irqrestore(&memcg->move_lock, flags);
1992 goto again;
1993 }
1994
1995
1996
1997
1998
1999
2000 memcg->move_lock_task = current;
2001 memcg->move_lock_flags = flags;
2002
2003 return memcg;
2004}
2005
2006
2007
2008
2009
2010void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
2011{
2012 if (memcg && memcg->move_lock_task == current) {
2013 unsigned long flags = memcg->move_lock_flags;
2014
2015 memcg->move_lock_task = NULL;
2016 memcg->move_lock_flags = 0;
2017
2018 spin_unlock_irqrestore(&memcg->move_lock, flags);
2019 }
2020
2021 rcu_read_unlock();
2022}
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
2033 enum mem_cgroup_stat_index idx, int val)
2034{
2035 VM_BUG_ON(!rcu_read_lock_held());
2036
2037 if (memcg)
2038 this_cpu_add(memcg->stat->count[idx], val);
2039}
2040
2041
2042
2043
2044
2045#define CHARGE_BATCH 32U
2046struct memcg_stock_pcp {
2047 struct mem_cgroup *cached;
2048 unsigned int nr_pages;
2049 struct work_struct work;
2050 unsigned long flags;
2051#define FLUSHING_CACHED_CHARGE 0
2052};
2053static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
2054static DEFINE_MUTEX(percpu_charge_mutex);
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2068{
2069 struct memcg_stock_pcp *stock;
2070 bool ret = false;
2071
2072 if (nr_pages > CHARGE_BATCH)
2073 return ret;
2074
2075 stock = &get_cpu_var(memcg_stock);
2076 if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
2077 stock->nr_pages -= nr_pages;
2078 ret = true;
2079 }
2080 put_cpu_var(memcg_stock);
2081 return ret;
2082}
2083
2084
2085
2086
2087static void drain_stock(struct memcg_stock_pcp *stock)
2088{
2089 struct mem_cgroup *old = stock->cached;
2090
2091 if (stock->nr_pages) {
2092 page_counter_uncharge(&old->memory, stock->nr_pages);
2093 if (do_swap_account)
2094 page_counter_uncharge(&old->memsw, stock->nr_pages);
2095 css_put_many(&old->css, stock->nr_pages);
2096 stock->nr_pages = 0;
2097 }
2098 stock->cached = NULL;
2099}
2100
2101
2102
2103
2104
2105static void drain_local_stock(struct work_struct *dummy)
2106{
2107 struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
2108 drain_stock(stock);
2109 clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
2110}
2111
2112
2113
2114
2115
2116static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2117{
2118 struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2119
2120 if (stock->cached != memcg) {
2121 drain_stock(stock);
2122 stock->cached = memcg;
2123 }
2124 stock->nr_pages += nr_pages;
2125 put_cpu_var(memcg_stock);
2126}
2127
2128
2129
2130
2131
2132static void drain_all_stock(struct mem_cgroup *root_memcg)
2133{
2134 int cpu, curcpu;
2135
2136
2137 if (!mutex_trylock(&percpu_charge_mutex))
2138 return;
2139
2140 get_online_cpus();
2141 curcpu = get_cpu();
2142 for_each_online_cpu(cpu) {
2143 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2144 struct mem_cgroup *memcg;
2145
2146 memcg = stock->cached;
2147 if (!memcg || !stock->nr_pages)
2148 continue;
2149 if (!mem_cgroup_is_descendant(memcg, root_memcg))
2150 continue;
2151 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2152 if (cpu == curcpu)
2153 drain_local_stock(&stock->work);
2154 else
2155 schedule_work_on(cpu, &stock->work);
2156 }
2157 }
2158 put_cpu();
2159 put_online_cpus();
2160 mutex_unlock(&percpu_charge_mutex);
2161}
2162
2163
2164
2165
2166
2167static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
2168{
2169 int i;
2170
2171 spin_lock(&memcg->pcp_counter_lock);
2172 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
2173 long x = per_cpu(memcg->stat->count[i], cpu);
2174
2175 per_cpu(memcg->stat->count[i], cpu) = 0;
2176 memcg->nocpu_base.count[i] += x;
2177 }
2178 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
2179 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
2180
2181 per_cpu(memcg->stat->events[i], cpu) = 0;
2182 memcg->nocpu_base.events[i] += x;
2183 }
2184 spin_unlock(&memcg->pcp_counter_lock);
2185}
2186
2187static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
2188 unsigned long action,
2189 void *hcpu)
2190{
2191 int cpu = (unsigned long)hcpu;
2192 struct memcg_stock_pcp *stock;
2193 struct mem_cgroup *iter;
2194
2195 if (action == CPU_ONLINE)
2196 return NOTIFY_OK;
2197
2198 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
2199 return NOTIFY_OK;
2200
2201 for_each_mem_cgroup(iter)
2202 mem_cgroup_drain_pcp_counter(iter, cpu);
2203
2204 stock = &per_cpu(memcg_stock, cpu);
2205 drain_stock(stock);
2206 return NOTIFY_OK;
2207}
2208
2209static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2210 unsigned int nr_pages)
2211{
2212 unsigned int batch = max(CHARGE_BATCH, nr_pages);
2213 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2214 struct mem_cgroup *mem_over_limit;
2215 struct page_counter *counter;
2216 unsigned long nr_reclaimed;
2217 bool may_swap = true;
2218 bool drained = false;
2219 int ret = 0;
2220
2221 if (mem_cgroup_is_root(memcg))
2222 goto done;
2223retry:
2224 if (consume_stock(memcg, nr_pages))
2225 goto done;
2226
2227 if (!do_swap_account ||
2228 !page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2229 if (!page_counter_try_charge(&memcg->memory, batch, &counter))
2230 goto done_restock;
2231 if (do_swap_account)
2232 page_counter_uncharge(&memcg->memsw, batch);
2233 mem_over_limit = mem_cgroup_from_counter(counter, memory);
2234 } else {
2235 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
2236 may_swap = false;
2237 }
2238
2239 if (batch > nr_pages) {
2240 batch = nr_pages;
2241 goto retry;
2242 }
2243
2244
2245
2246
2247
2248
2249
2250 if (unlikely(test_thread_flag(TIF_MEMDIE) ||
2251 fatal_signal_pending(current) ||
2252 current->flags & PF_EXITING))
2253 goto bypass;
2254
2255 if (unlikely(task_in_memcg_oom(current)))
2256 goto nomem;
2257
2258 if (!(gfp_mask & __GFP_WAIT))
2259 goto nomem;
2260
2261 mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
2262
2263 nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2264 gfp_mask, may_swap);
2265
2266 if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
2267 goto retry;
2268
2269 if (!drained) {
2270 drain_all_stock(mem_over_limit);
2271 drained = true;
2272 goto retry;
2273 }
2274
2275 if (gfp_mask & __GFP_NORETRY)
2276 goto nomem;
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286 if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
2287 goto retry;
2288
2289
2290
2291
2292 if (mem_cgroup_wait_acct_move(mem_over_limit))
2293 goto retry;
2294
2295 if (nr_retries--)
2296 goto retry;
2297
2298 if (gfp_mask & __GFP_NOFAIL)
2299 goto bypass;
2300
2301 if (fatal_signal_pending(current))
2302 goto bypass;
2303
2304 mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2305
2306 mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages));
2307nomem:
2308 if (!(gfp_mask & __GFP_NOFAIL))
2309 return -ENOMEM;
2310bypass:
2311 return -EINTR;
2312
2313done_restock:
2314 css_get_many(&memcg->css, batch);
2315 if (batch > nr_pages)
2316 refill_stock(memcg, batch - nr_pages);
2317
2318
2319
2320
2321 do {
2322 if (page_counter_read(&memcg->memory) <= memcg->high)
2323 continue;
2324 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
2325 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
2326 } while ((memcg = parent_mem_cgroup(memcg)));
2327done:
2328 return ret;
2329}
2330
2331static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
2332{
2333 if (mem_cgroup_is_root(memcg))
2334 return;
2335
2336 page_counter_uncharge(&memcg->memory, nr_pages);
2337 if (do_swap_account)
2338 page_counter_uncharge(&memcg->memsw, nr_pages);
2339
2340 css_put_many(&memcg->css, nr_pages);
2341}
2342
2343
2344
2345
2346
2347
2348
2349static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
2350{
2351
2352 if (!id)
2353 return NULL;
2354 return mem_cgroup_from_id(id);
2355}
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
2368{
2369 struct mem_cgroup *memcg;
2370 unsigned short id;
2371 swp_entry_t ent;
2372
2373 VM_BUG_ON_PAGE(!PageLocked(page), page);
2374
2375 memcg = page->mem_cgroup;
2376 if (memcg) {
2377 if (!css_tryget_online(&memcg->css))
2378 memcg = NULL;
2379 } else if (PageSwapCache(page)) {
2380 ent.val = page_private(page);
2381 id = lookup_swap_cgroup_id(ent);
2382 rcu_read_lock();
2383 memcg = mem_cgroup_lookup(id);
2384 if (memcg && !css_tryget_online(&memcg->css))
2385 memcg = NULL;
2386 rcu_read_unlock();
2387 }
2388 return memcg;
2389}
2390
2391static void lock_page_lru(struct page *page, int *isolated)
2392{
2393 struct zone *zone = page_zone(page);
2394
2395 spin_lock_irq(&zone->lru_lock);
2396 if (PageLRU(page)) {
2397 struct lruvec *lruvec;
2398
2399 lruvec = mem_cgroup_page_lruvec(page, zone);
2400 ClearPageLRU(page);
2401 del_page_from_lru_list(page, lruvec, page_lru(page));
2402 *isolated = 1;
2403 } else
2404 *isolated = 0;
2405}
2406
2407static void unlock_page_lru(struct page *page, int isolated)
2408{
2409 struct zone *zone = page_zone(page);
2410
2411 if (isolated) {
2412 struct lruvec *lruvec;
2413
2414 lruvec = mem_cgroup_page_lruvec(page, zone);
2415 VM_BUG_ON_PAGE(PageLRU(page), page);
2416 SetPageLRU(page);
2417 add_page_to_lru_list(page, lruvec, page_lru(page));
2418 }
2419 spin_unlock_irq(&zone->lru_lock);
2420}
2421
2422static void commit_charge(struct page *page, struct mem_cgroup *memcg,
2423 bool lrucare)
2424{
2425 int isolated;
2426
2427 VM_BUG_ON_PAGE(page->mem_cgroup, page);
2428
2429
2430
2431
2432
2433 if (lrucare)
2434 lock_page_lru(page, &isolated);
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450 page->mem_cgroup = memcg;
2451
2452 if (lrucare)
2453 unlock_page_lru(page, isolated);
2454}
2455
2456#ifdef CONFIG_MEMCG_KMEM
2457int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
2458 unsigned long nr_pages)
2459{
2460 struct page_counter *counter;
2461 int ret = 0;
2462
2463 ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter);
2464 if (ret < 0)
2465 return ret;
2466
2467 ret = try_charge(memcg, gfp, nr_pages);
2468 if (ret == -EINTR) {
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484 page_counter_charge(&memcg->memory, nr_pages);
2485 if (do_swap_account)
2486 page_counter_charge(&memcg->memsw, nr_pages);
2487 css_get_many(&memcg->css, nr_pages);
2488 ret = 0;
2489 } else if (ret)
2490 page_counter_uncharge(&memcg->kmem, nr_pages);
2491
2492 return ret;
2493}
2494
2495void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages)
2496{
2497 page_counter_uncharge(&memcg->memory, nr_pages);
2498 if (do_swap_account)
2499 page_counter_uncharge(&memcg->memsw, nr_pages);
2500
2501 page_counter_uncharge(&memcg->kmem, nr_pages);
2502
2503 css_put_many(&memcg->css, nr_pages);
2504}
2505
2506
2507
2508
2509
2510
2511int memcg_cache_id(struct mem_cgroup *memcg)
2512{
2513 return memcg ? memcg->kmemcg_id : -1;
2514}
2515
2516static int memcg_alloc_cache_id(void)
2517{
2518 int id, size;
2519 int err;
2520
2521 id = ida_simple_get(&memcg_cache_ida,
2522 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2523 if (id < 0)
2524 return id;
2525
2526 if (id < memcg_nr_cache_ids)
2527 return id;
2528
2529
2530
2531
2532
2533 down_write(&memcg_cache_ids_sem);
2534
2535 size = 2 * (id + 1);
2536 if (size < MEMCG_CACHES_MIN_SIZE)
2537 size = MEMCG_CACHES_MIN_SIZE;
2538 else if (size > MEMCG_CACHES_MAX_SIZE)
2539 size = MEMCG_CACHES_MAX_SIZE;
2540
2541 err = memcg_update_all_caches(size);
2542 if (!err)
2543 err = memcg_update_all_list_lrus(size);
2544 if (!err)
2545 memcg_nr_cache_ids = size;
2546
2547 up_write(&memcg_cache_ids_sem);
2548
2549 if (err) {
2550 ida_simple_remove(&memcg_cache_ida, id);
2551 return err;
2552 }
2553 return id;
2554}
2555
2556static void memcg_free_cache_id(int id)
2557{
2558 ida_simple_remove(&memcg_cache_ida, id);
2559}
2560
2561struct memcg_kmem_cache_create_work {
2562 struct mem_cgroup *memcg;
2563 struct kmem_cache *cachep;
2564 struct work_struct work;
2565};
2566
2567static void memcg_kmem_cache_create_func(struct work_struct *w)
2568{
2569 struct memcg_kmem_cache_create_work *cw =
2570 container_of(w, struct memcg_kmem_cache_create_work, work);
2571 struct mem_cgroup *memcg = cw->memcg;
2572 struct kmem_cache *cachep = cw->cachep;
2573
2574 memcg_create_kmem_cache(memcg, cachep);
2575
2576 css_put(&memcg->css);
2577 kfree(cw);
2578}
2579
2580
2581
2582
2583static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2584 struct kmem_cache *cachep)
2585{
2586 struct memcg_kmem_cache_create_work *cw;
2587
2588 cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
2589 if (!cw)
2590 return;
2591
2592 css_get(&memcg->css);
2593
2594 cw->memcg = memcg;
2595 cw->cachep = cachep;
2596 INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
2597
2598 schedule_work(&cw->work);
2599}
2600
2601static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2602 struct kmem_cache *cachep)
2603{
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615 current->memcg_kmem_skip_account = 1;
2616 __memcg_schedule_kmem_cache_create(memcg, cachep);
2617 current->memcg_kmem_skip_account = 0;
2618}
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep)
2634{
2635 struct mem_cgroup *memcg;
2636 struct kmem_cache *memcg_cachep;
2637 int kmemcg_id;
2638
2639 VM_BUG_ON(!is_root_cache(cachep));
2640
2641 if (current->memcg_kmem_skip_account)
2642 return cachep;
2643
2644 memcg = get_mem_cgroup_from_mm(current->mm);
2645 kmemcg_id = ACCESS_ONCE(memcg->kmemcg_id);
2646 if (kmemcg_id < 0)
2647 goto out;
2648
2649 memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
2650 if (likely(memcg_cachep))
2651 return memcg_cachep;
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665 memcg_schedule_kmem_cache_create(memcg, cachep);
2666out:
2667 css_put(&memcg->css);
2668 return cachep;
2669}
2670
2671void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2672{
2673 if (!is_root_cache(cachep))
2674 css_put(&cachep->memcg_params.memcg->css);
2675}
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691bool
2692__memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
2693{
2694 struct mem_cgroup *memcg;
2695 int ret;
2696
2697 *_memcg = NULL;
2698
2699 memcg = get_mem_cgroup_from_mm(current->mm);
2700
2701 if (!memcg_kmem_is_active(memcg)) {
2702 css_put(&memcg->css);
2703 return true;
2704 }
2705
2706 ret = memcg_charge_kmem(memcg, gfp, 1 << order);
2707 if (!ret)
2708 *_memcg = memcg;
2709
2710 css_put(&memcg->css);
2711 return (ret == 0);
2712}
2713
2714void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
2715 int order)
2716{
2717 VM_BUG_ON(mem_cgroup_is_root(memcg));
2718
2719
2720 if (!page) {
2721 memcg_uncharge_kmem(memcg, 1 << order);
2722 return;
2723 }
2724 page->mem_cgroup = memcg;
2725}
2726
2727void __memcg_kmem_uncharge_pages(struct page *page, int order)
2728{
2729 struct mem_cgroup *memcg = page->mem_cgroup;
2730
2731 if (!memcg)
2732 return;
2733
2734 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
2735
2736 memcg_uncharge_kmem(memcg, 1 << order);
2737 page->mem_cgroup = NULL;
2738}
2739
2740struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr)
2741{
2742 struct mem_cgroup *memcg = NULL;
2743 struct kmem_cache *cachep;
2744 struct page *page;
2745
2746 page = virt_to_head_page(ptr);
2747 if (PageSlab(page)) {
2748 cachep = page->slab_cache;
2749 if (!is_root_cache(cachep))
2750 memcg = cachep->memcg_params.memcg;
2751 } else
2752
2753 memcg = page->mem_cgroup;
2754
2755 return memcg;
2756}
2757#endif
2758
2759#ifdef CONFIG_TRANSPARENT_HUGEPAGE
2760
2761
2762
2763
2764
2765
2766
2767void mem_cgroup_split_huge_fixup(struct page *head)
2768{
2769 int i;
2770
2771 if (mem_cgroup_disabled())
2772 return;
2773
2774 for (i = 1; i < HPAGE_PMD_NR; i++)
2775 head[i].mem_cgroup = head->mem_cgroup;
2776
2777 __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
2778 HPAGE_PMD_NR);
2779}
2780#endif
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796static int mem_cgroup_move_account(struct page *page,
2797 unsigned int nr_pages,
2798 struct mem_cgroup *from,
2799 struct mem_cgroup *to)
2800{
2801 unsigned long flags;
2802 int ret;
2803
2804 VM_BUG_ON(from == to);
2805 VM_BUG_ON_PAGE(PageLRU(page), page);
2806
2807
2808
2809
2810
2811
2812 ret = -EBUSY;
2813 if (nr_pages > 1 && !PageTransHuge(page))
2814 goto out;
2815
2816
2817
2818
2819
2820
2821 if (!trylock_page(page))
2822 goto out;
2823
2824 ret = -EINVAL;
2825 if (page->mem_cgroup != from)
2826 goto out_unlock;
2827
2828 spin_lock_irqsave(&from->move_lock, flags);
2829
2830 if (!PageAnon(page) && page_mapped(page)) {
2831 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
2832 nr_pages);
2833 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
2834 nr_pages);
2835 }
2836
2837 if (PageWriteback(page)) {
2838 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
2839 nr_pages);
2840 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
2841 nr_pages);
2842 }
2843
2844
2845
2846
2847
2848
2849
2850
2851 page->mem_cgroup = to;
2852 spin_unlock_irqrestore(&from->move_lock, flags);
2853
2854 ret = 0;
2855
2856 local_irq_disable();
2857 mem_cgroup_charge_statistics(to, page, nr_pages);
2858 memcg_check_events(to, page);
2859 mem_cgroup_charge_statistics(from, page, -nr_pages);
2860 memcg_check_events(from, page);
2861 local_irq_enable();
2862out_unlock:
2863 unlock_page(page);
2864out:
2865 return ret;
2866}
2867
2868#ifdef CONFIG_MEMCG_SWAP
2869static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2870 bool charge)
2871{
2872 int val = (charge) ? 1 : -1;
2873 this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
2874}
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890static int mem_cgroup_move_swap_account(swp_entry_t entry,
2891 struct mem_cgroup *from, struct mem_cgroup *to)
2892{
2893 unsigned short old_id, new_id;
2894
2895 old_id = mem_cgroup_id(from);
2896 new_id = mem_cgroup_id(to);
2897
2898 if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2899 mem_cgroup_swap_statistics(from, false);
2900 mem_cgroup_swap_statistics(to, true);
2901 return 0;
2902 }
2903 return -EINVAL;
2904}
2905#else
2906static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2907 struct mem_cgroup *from, struct mem_cgroup *to)
2908{
2909 return -EINVAL;
2910}
2911#endif
2912
2913static DEFINE_MUTEX(memcg_limit_mutex);
2914
2915static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2916 unsigned long limit)
2917{
2918 unsigned long curusage;
2919 unsigned long oldusage;
2920 bool enlarge = false;
2921 int retry_count;
2922 int ret;
2923
2924
2925
2926
2927
2928
2929 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2930 mem_cgroup_count_children(memcg);
2931
2932 oldusage = page_counter_read(&memcg->memory);
2933
2934 do {
2935 if (signal_pending(current)) {
2936 ret = -EINTR;
2937 break;
2938 }
2939
2940 mutex_lock(&memcg_limit_mutex);
2941 if (limit > memcg->memsw.limit) {
2942 mutex_unlock(&memcg_limit_mutex);
2943 ret = -EINVAL;
2944 break;
2945 }
2946 if (limit > memcg->memory.limit)
2947 enlarge = true;
2948 ret = page_counter_limit(&memcg->memory, limit);
2949 mutex_unlock(&memcg_limit_mutex);
2950
2951 if (!ret)
2952 break;
2953
2954 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2955
2956 curusage = page_counter_read(&memcg->memory);
2957
2958 if (curusage >= oldusage)
2959 retry_count--;
2960 else
2961 oldusage = curusage;
2962 } while (retry_count);
2963
2964 if (!ret && enlarge)
2965 memcg_oom_recover(memcg);
2966
2967 return ret;
2968}
2969
2970static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2971 unsigned long limit)
2972{
2973 unsigned long curusage;
2974 unsigned long oldusage;
2975 bool enlarge = false;
2976 int retry_count;
2977 int ret;
2978
2979
2980 retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2981 mem_cgroup_count_children(memcg);
2982
2983 oldusage = page_counter_read(&memcg->memsw);
2984
2985 do {
2986 if (signal_pending(current)) {
2987 ret = -EINTR;
2988 break;
2989 }
2990
2991 mutex_lock(&memcg_limit_mutex);
2992 if (limit < memcg->memory.limit) {
2993 mutex_unlock(&memcg_limit_mutex);
2994 ret = -EINVAL;
2995 break;
2996 }
2997 if (limit > memcg->memsw.limit)
2998 enlarge = true;
2999 ret = page_counter_limit(&memcg->memsw, limit);
3000 mutex_unlock(&memcg_limit_mutex);
3001
3002 if (!ret)
3003 break;
3004
3005 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
3006
3007 curusage = page_counter_read(&memcg->memsw);
3008
3009 if (curusage >= oldusage)
3010 retry_count--;
3011 else
3012 oldusage = curusage;
3013 } while (retry_count);
3014
3015 if (!ret && enlarge)
3016 memcg_oom_recover(memcg);
3017
3018 return ret;
3019}
3020
3021unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
3022 gfp_t gfp_mask,
3023 unsigned long *total_scanned)
3024{
3025 unsigned long nr_reclaimed = 0;
3026 struct mem_cgroup_per_zone *mz, *next_mz = NULL;
3027 unsigned long reclaimed;
3028 int loop = 0;
3029 struct mem_cgroup_tree_per_zone *mctz;
3030 unsigned long excess;
3031 unsigned long nr_scanned;
3032
3033 if (order > 0)
3034 return 0;
3035
3036 mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
3037
3038
3039
3040
3041
3042 do {
3043 if (next_mz)
3044 mz = next_mz;
3045 else
3046 mz = mem_cgroup_largest_soft_limit_node(mctz);
3047 if (!mz)
3048 break;
3049
3050 nr_scanned = 0;
3051 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
3052 gfp_mask, &nr_scanned);
3053 nr_reclaimed += reclaimed;
3054 *total_scanned += nr_scanned;
3055 spin_lock_irq(&mctz->lock);
3056 __mem_cgroup_remove_exceeded(mz, mctz);
3057
3058
3059
3060
3061
3062 next_mz = NULL;
3063 if (!reclaimed)
3064 next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
3065
3066 excess = soft_limit_excess(mz->memcg);
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076 __mem_cgroup_insert_exceeded(mz, mctz, excess);
3077 spin_unlock_irq(&mctz->lock);
3078 css_put(&mz->memcg->css);
3079 loop++;
3080
3081
3082
3083
3084
3085 if (!nr_reclaimed &&
3086 (next_mz == NULL ||
3087 loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
3088 break;
3089 } while (!nr_reclaimed);
3090 if (next_mz)
3091 css_put(&next_mz->memcg->css);
3092 return nr_reclaimed;
3093}
3094
3095
3096
3097
3098
3099
3100
3101static inline bool memcg_has_children(struct mem_cgroup *memcg)
3102{
3103 bool ret;
3104
3105
3106
3107
3108
3109
3110
3111 lockdep_assert_held(&memcg_create_mutex);
3112
3113 rcu_read_lock();
3114 ret = css_next_child(NULL, &memcg->css);
3115 rcu_read_unlock();
3116 return ret;
3117}
3118
3119
3120
3121
3122
3123
3124
3125static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
3126{
3127 int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
3128
3129
3130 lru_add_drain_all();
3131
3132 while (nr_retries && page_counter_read(&memcg->memory)) {
3133 int progress;
3134
3135 if (signal_pending(current))
3136 return -EINTR;
3137
3138 progress = try_to_free_mem_cgroup_pages(memcg, 1,
3139 GFP_KERNEL, true);
3140 if (!progress) {
3141 nr_retries--;
3142
3143 congestion_wait(BLK_RW_ASYNC, HZ/10);
3144 }
3145
3146 }
3147
3148 return 0;
3149}
3150
3151static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
3152 char *buf, size_t nbytes,
3153 loff_t off)
3154{
3155 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3156
3157 if (mem_cgroup_is_root(memcg))
3158 return -EINVAL;
3159 return mem_cgroup_force_empty(memcg) ?: nbytes;
3160}
3161
3162static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
3163 struct cftype *cft)
3164{
3165 return mem_cgroup_from_css(css)->use_hierarchy;
3166}
3167
3168static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
3169 struct cftype *cft, u64 val)
3170{
3171 int retval = 0;
3172 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3173 struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
3174
3175 mutex_lock(&memcg_create_mutex);
3176
3177 if (memcg->use_hierarchy == val)
3178 goto out;
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188 if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
3189 (val == 1 || val == 0)) {
3190 if (!memcg_has_children(memcg))
3191 memcg->use_hierarchy = val;
3192 else
3193 retval = -EBUSY;
3194 } else
3195 retval = -EINVAL;
3196
3197out:
3198 mutex_unlock(&memcg_create_mutex);
3199
3200 return retval;
3201}
3202
3203static unsigned long tree_stat(struct mem_cgroup *memcg,
3204 enum mem_cgroup_stat_index idx)
3205{
3206 struct mem_cgroup *iter;
3207 long val = 0;
3208
3209
3210 for_each_mem_cgroup_tree(iter, memcg)
3211 val += mem_cgroup_read_stat(iter, idx);
3212
3213 if (val < 0)
3214 val = 0;
3215 return val;
3216}
3217
3218static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
3219{
3220 u64 val;
3221
3222 if (mem_cgroup_is_root(memcg)) {
3223 val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE);
3224 val += tree_stat(memcg, MEM_CGROUP_STAT_RSS);
3225 if (swap)
3226 val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP);
3227 } else {
3228 if (!swap)
3229 val = page_counter_read(&memcg->memory);
3230 else
3231 val = page_counter_read(&memcg->memsw);
3232 }
3233 return val << PAGE_SHIFT;
3234}
3235
3236enum {
3237 RES_USAGE,
3238 RES_LIMIT,
3239 RES_MAX_USAGE,
3240 RES_FAILCNT,
3241 RES_SOFT_LIMIT,
3242};
3243
3244static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
3245 struct cftype *cft)
3246{
3247 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3248 struct page_counter *counter;
3249
3250 switch (MEMFILE_TYPE(cft->private)) {
3251 case _MEM:
3252 counter = &memcg->memory;
3253 break;
3254 case _MEMSWAP:
3255 counter = &memcg->memsw;
3256 break;
3257 case _KMEM:
3258 counter = &memcg->kmem;
3259 break;
3260 default:
3261 BUG();
3262 }
3263
3264 switch (MEMFILE_ATTR(cft->private)) {
3265 case RES_USAGE:
3266 if (counter == &memcg->memory)
3267 return mem_cgroup_usage(memcg, false);
3268 if (counter == &memcg->memsw)
3269 return mem_cgroup_usage(memcg, true);
3270 return (u64)page_counter_read(counter) * PAGE_SIZE;
3271 case RES_LIMIT:
3272 return (u64)counter->limit * PAGE_SIZE;
3273 case RES_MAX_USAGE:
3274 return (u64)counter->watermark * PAGE_SIZE;
3275 case RES_FAILCNT:
3276 return counter->failcnt;
3277 case RES_SOFT_LIMIT:
3278 return (u64)memcg->soft_limit * PAGE_SIZE;
3279 default:
3280 BUG();
3281 }
3282}
3283
3284#ifdef CONFIG_MEMCG_KMEM
3285static int memcg_activate_kmem(struct mem_cgroup *memcg,
3286 unsigned long nr_pages)
3287{
3288 int err = 0;
3289 int memcg_id;
3290
3291 BUG_ON(memcg->kmemcg_id >= 0);
3292 BUG_ON(memcg->kmem_acct_activated);
3293 BUG_ON(memcg->kmem_acct_active);
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307 mutex_lock(&memcg_create_mutex);
3308 if (cgroup_has_tasks(memcg->css.cgroup) ||
3309 (memcg->use_hierarchy && memcg_has_children(memcg)))
3310 err = -EBUSY;
3311 mutex_unlock(&memcg_create_mutex);
3312 if (err)
3313 goto out;
3314
3315 memcg_id = memcg_alloc_cache_id();
3316 if (memcg_id < 0) {
3317 err = memcg_id;
3318 goto out;
3319 }
3320
3321
3322
3323
3324
3325 err = page_counter_limit(&memcg->kmem, nr_pages);
3326 VM_BUG_ON(err);
3327
3328 static_key_slow_inc(&memcg_kmem_enabled_key);
3329
3330
3331
3332
3333
3334
3335 memcg->kmemcg_id = memcg_id;
3336 memcg->kmem_acct_activated = true;
3337 memcg->kmem_acct_active = true;
3338out:
3339 return err;
3340}
3341
3342static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
3343 unsigned long limit)
3344{
3345 int ret;
3346
3347 mutex_lock(&memcg_limit_mutex);
3348 if (!memcg_kmem_is_active(memcg))
3349 ret = memcg_activate_kmem(memcg, limit);
3350 else
3351 ret = page_counter_limit(&memcg->kmem, limit);
3352 mutex_unlock(&memcg_limit_mutex);
3353 return ret;
3354}
3355
3356static int memcg_propagate_kmem(struct mem_cgroup *memcg)
3357{
3358 int ret = 0;
3359 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
3360
3361 if (!parent)
3362 return 0;
3363
3364 mutex_lock(&memcg_limit_mutex);
3365
3366
3367
3368
3369 if (memcg_kmem_is_active(parent))
3370 ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX);
3371 mutex_unlock(&memcg_limit_mutex);
3372 return ret;
3373}
3374#else
3375static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
3376 unsigned long limit)
3377{
3378 return -EINVAL;
3379}
3380#endif
3381
3382
3383
3384
3385
3386static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3387 char *buf, size_t nbytes, loff_t off)
3388{
3389 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3390 unsigned long nr_pages;
3391 int ret;
3392
3393 buf = strstrip(buf);
3394 ret = page_counter_memparse(buf, "-1", &nr_pages);
3395 if (ret)
3396 return ret;
3397
3398 switch (MEMFILE_ATTR(of_cft(of)->private)) {
3399 case RES_LIMIT:
3400 if (mem_cgroup_is_root(memcg)) {
3401 ret = -EINVAL;
3402 break;
3403 }
3404 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3405 case _MEM:
3406 ret = mem_cgroup_resize_limit(memcg, nr_pages);
3407 break;
3408 case _MEMSWAP:
3409 ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3410 break;
3411 case _KMEM:
3412 ret = memcg_update_kmem_limit(memcg, nr_pages);
3413 break;
3414 }
3415 break;
3416 case RES_SOFT_LIMIT:
3417 memcg->soft_limit = nr_pages;
3418 ret = 0;
3419 break;
3420 }
3421 return ret ?: nbytes;
3422}
3423
3424static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3425 size_t nbytes, loff_t off)
3426{
3427 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3428 struct page_counter *counter;
3429
3430 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3431 case _MEM:
3432 counter = &memcg->memory;
3433 break;
3434 case _MEMSWAP:
3435 counter = &memcg->memsw;
3436 break;
3437 case _KMEM:
3438 counter = &memcg->kmem;
3439 break;
3440 default:
3441 BUG();
3442 }
3443
3444 switch (MEMFILE_ATTR(of_cft(of)->private)) {
3445 case RES_MAX_USAGE:
3446 page_counter_reset_watermark(counter);
3447 break;
3448 case RES_FAILCNT:
3449 counter->failcnt = 0;
3450 break;
3451 default:
3452 BUG();
3453 }
3454
3455 return nbytes;
3456}
3457
3458static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
3459 struct cftype *cft)
3460{
3461 return mem_cgroup_from_css(css)->move_charge_at_immigrate;
3462}
3463
3464#ifdef CONFIG_MMU
3465static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3466 struct cftype *cft, u64 val)
3467{
3468 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3469
3470 if (val & ~MOVE_MASK)
3471 return -EINVAL;
3472
3473
3474
3475
3476
3477
3478
3479 memcg->move_charge_at_immigrate = val;
3480 return 0;
3481}
3482#else
3483static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3484 struct cftype *cft, u64 val)
3485{
3486 return -ENOSYS;
3487}
3488#endif
3489
3490#ifdef CONFIG_NUMA
3491static int memcg_numa_stat_show(struct seq_file *m, void *v)
3492{
3493 struct numa_stat {
3494 const char *name;
3495 unsigned int lru_mask;
3496 };
3497
3498 static const struct numa_stat stats[] = {
3499 { "total", LRU_ALL },
3500 { "file", LRU_ALL_FILE },
3501 { "anon", LRU_ALL_ANON },
3502 { "unevictable", BIT(LRU_UNEVICTABLE) },
3503 };
3504 const struct numa_stat *stat;
3505 int nid;
3506 unsigned long nr;
3507 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3508
3509 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3510 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3511 seq_printf(m, "%s=%lu", stat->name, nr);
3512 for_each_node_state(nid, N_MEMORY) {
3513 nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3514 stat->lru_mask);
3515 seq_printf(m, " N%d=%lu", nid, nr);
3516 }
3517 seq_putc(m, '\n');
3518 }
3519
3520 for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3521 struct mem_cgroup *iter;
3522
3523 nr = 0;
3524 for_each_mem_cgroup_tree(iter, memcg)
3525 nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3526 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3527 for_each_node_state(nid, N_MEMORY) {
3528 nr = 0;
3529 for_each_mem_cgroup_tree(iter, memcg)
3530 nr += mem_cgroup_node_nr_lru_pages(
3531 iter, nid, stat->lru_mask);
3532 seq_printf(m, " N%d=%lu", nid, nr);
3533 }
3534 seq_putc(m, '\n');
3535 }
3536
3537 return 0;
3538}
3539#endif
3540
3541static int memcg_stat_show(struct seq_file *m, void *v)
3542{
3543 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3544 unsigned long memory, memsw;
3545 struct mem_cgroup *mi;
3546 unsigned int i;
3547
3548 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3549 MEM_CGROUP_STAT_NSTATS);
3550 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3551 MEM_CGROUP_EVENTS_NSTATS);
3552 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3553
3554 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3555 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
3556 continue;
3557 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
3558 mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
3559 }
3560
3561 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3562 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3563 mem_cgroup_read_events(memcg, i));
3564
3565 for (i = 0; i < NR_LRU_LISTS; i++)
3566 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3567 mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3568
3569
3570 memory = memsw = PAGE_COUNTER_MAX;
3571 for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3572 memory = min(memory, mi->memory.limit);
3573 memsw = min(memsw, mi->memsw.limit);
3574 }
3575 seq_printf(m, "hierarchical_memory_limit %llu\n",
3576 (u64)memory * PAGE_SIZE);
3577 if (do_swap_account)
3578 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3579 (u64)memsw * PAGE_SIZE);
3580
3581 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3582 long long val = 0;
3583
3584 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
3585 continue;
3586 for_each_mem_cgroup_tree(mi, memcg)
3587 val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
3588 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
3589 }
3590
3591 for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3592 unsigned long long val = 0;
3593
3594 for_each_mem_cgroup_tree(mi, memcg)
3595 val += mem_cgroup_read_events(mi, i);
3596 seq_printf(m, "total_%s %llu\n",
3597 mem_cgroup_events_names[i], val);
3598 }
3599
3600 for (i = 0; i < NR_LRU_LISTS; i++) {
3601 unsigned long long val = 0;
3602
3603 for_each_mem_cgroup_tree(mi, memcg)
3604 val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3605 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
3606 }
3607
3608#ifdef CONFIG_DEBUG_VM
3609 {
3610 int nid, zid;
3611 struct mem_cgroup_per_zone *mz;
3612 struct zone_reclaim_stat *rstat;
3613 unsigned long recent_rotated[2] = {0, 0};
3614 unsigned long recent_scanned[2] = {0, 0};
3615
3616 for_each_online_node(nid)
3617 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3618 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
3619 rstat = &mz->lruvec.reclaim_stat;
3620
3621 recent_rotated[0] += rstat->recent_rotated[0];
3622 recent_rotated[1] += rstat->recent_rotated[1];
3623 recent_scanned[0] += rstat->recent_scanned[0];
3624 recent_scanned[1] += rstat->recent_scanned[1];
3625 }
3626 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3627 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3628 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3629 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
3630 }
3631#endif
3632
3633 return 0;
3634}
3635
3636static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3637 struct cftype *cft)
3638{
3639 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3640
3641 return mem_cgroup_swappiness(memcg);
3642}
3643
3644static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3645 struct cftype *cft, u64 val)
3646{
3647 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3648
3649 if (val > 100)
3650 return -EINVAL;
3651
3652 if (css->parent)
3653 memcg->swappiness = val;
3654 else
3655 vm_swappiness = val;
3656
3657 return 0;
3658}
3659
3660static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3661{
3662 struct mem_cgroup_threshold_ary *t;
3663 unsigned long usage;
3664 int i;
3665
3666 rcu_read_lock();
3667 if (!swap)
3668 t = rcu_dereference(memcg->thresholds.primary);
3669 else
3670 t = rcu_dereference(memcg->memsw_thresholds.primary);
3671
3672 if (!t)
3673 goto unlock;
3674
3675 usage = mem_cgroup_usage(memcg, swap);
3676
3677
3678
3679
3680
3681
3682 i = t->current_threshold;
3683
3684
3685
3686
3687
3688
3689
3690 for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3691 eventfd_signal(t->entries[i].eventfd, 1);
3692
3693
3694 i++;
3695
3696
3697
3698
3699
3700
3701
3702 for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3703 eventfd_signal(t->entries[i].eventfd, 1);
3704
3705
3706 t->current_threshold = i - 1;
3707unlock:
3708 rcu_read_unlock();
3709}
3710
3711static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3712{
3713 while (memcg) {
3714 __mem_cgroup_threshold(memcg, false);
3715 if (do_swap_account)
3716 __mem_cgroup_threshold(memcg, true);
3717
3718 memcg = parent_mem_cgroup(memcg);
3719 }
3720}
3721
3722static int compare_thresholds(const void *a, const void *b)
3723{
3724 const struct mem_cgroup_threshold *_a = a;
3725 const struct mem_cgroup_threshold *_b = b;
3726
3727 if (_a->threshold > _b->threshold)
3728 return 1;
3729
3730 if (_a->threshold < _b->threshold)
3731 return -1;
3732
3733 return 0;
3734}
3735
3736static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
3737{
3738 struct mem_cgroup_eventfd_list *ev;
3739
3740 spin_lock(&memcg_oom_lock);
3741
3742 list_for_each_entry(ev, &memcg->oom_notify, list)
3743 eventfd_signal(ev->eventfd, 1);
3744
3745 spin_unlock(&memcg_oom_lock);
3746 return 0;
3747}
3748
3749static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
3750{
3751 struct mem_cgroup *iter;
3752
3753 for_each_mem_cgroup_tree(iter, memcg)
3754 mem_cgroup_oom_notify_cb(iter);
3755}
3756
3757static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3758 struct eventfd_ctx *eventfd, const char *args, enum res_type type)
3759{
3760 struct mem_cgroup_thresholds *thresholds;
3761 struct mem_cgroup_threshold_ary *new;
3762 unsigned long threshold;
3763 unsigned long usage;
3764 int i, size, ret;
3765
3766 ret = page_counter_memparse(args, "-1", &threshold);
3767 if (ret)
3768 return ret;
3769
3770 mutex_lock(&memcg->thresholds_lock);
3771
3772 if (type == _MEM) {
3773 thresholds = &memcg->thresholds;
3774 usage = mem_cgroup_usage(memcg, false);
3775 } else if (type == _MEMSWAP) {
3776 thresholds = &memcg->memsw_thresholds;
3777 usage = mem_cgroup_usage(memcg, true);
3778 } else
3779 BUG();
3780
3781
3782 if (thresholds->primary)
3783 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3784
3785 size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3786
3787
3788 new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3789 GFP_KERNEL);
3790 if (!new) {
3791 ret = -ENOMEM;
3792 goto unlock;
3793 }
3794 new->size = size;
3795
3796
3797 if (thresholds->primary) {
3798 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3799 sizeof(struct mem_cgroup_threshold));
3800 }
3801
3802
3803 new->entries[size - 1].eventfd = eventfd;
3804 new->entries[size - 1].threshold = threshold;
3805
3806
3807 sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3808 compare_thresholds, NULL);
3809
3810
3811 new->current_threshold = -1;
3812 for (i = 0; i < size; i++) {
3813 if (new->entries[i].threshold <= usage) {
3814
3815
3816
3817
3818
3819 ++new->current_threshold;
3820 } else
3821 break;
3822 }
3823
3824
3825 kfree(thresholds->spare);
3826 thresholds->spare = thresholds->primary;
3827
3828 rcu_assign_pointer(thresholds->primary, new);
3829
3830
3831 synchronize_rcu();
3832
3833unlock:
3834 mutex_unlock(&memcg->thresholds_lock);
3835
3836 return ret;
3837}
3838
3839static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3840 struct eventfd_ctx *eventfd, const char *args)
3841{
3842 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
3843}
3844
3845static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
3846 struct eventfd_ctx *eventfd, const char *args)
3847{
3848 return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
3849}
3850
3851static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3852 struct eventfd_ctx *eventfd, enum res_type type)
3853{
3854 struct mem_cgroup_thresholds *thresholds;
3855 struct mem_cgroup_threshold_ary *new;
3856 unsigned long usage;
3857 int i, j, size;
3858
3859 mutex_lock(&memcg->thresholds_lock);
3860
3861 if (type == _MEM) {
3862 thresholds = &memcg->thresholds;
3863 usage = mem_cgroup_usage(memcg, false);
3864 } else if (type == _MEMSWAP) {
3865 thresholds = &memcg->memsw_thresholds;
3866 usage = mem_cgroup_usage(memcg, true);
3867 } else
3868 BUG();
3869
3870 if (!thresholds->primary)
3871 goto unlock;
3872
3873
3874 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3875
3876
3877 size = 0;
3878 for (i = 0; i < thresholds->primary->size; i++) {
3879 if (thresholds->primary->entries[i].eventfd != eventfd)
3880 size++;
3881 }
3882
3883 new = thresholds->spare;
3884
3885
3886 if (!size) {
3887 kfree(new);
3888 new = NULL;
3889 goto swap_buffers;
3890 }
3891
3892 new->size = size;
3893
3894
3895 new->current_threshold = -1;
3896 for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3897 if (thresholds->primary->entries[i].eventfd == eventfd)
3898 continue;
3899
3900 new->entries[j] = thresholds->primary->entries[i];
3901 if (new->entries[j].threshold <= usage) {
3902
3903
3904
3905
3906
3907 ++new->current_threshold;
3908 }
3909 j++;
3910 }
3911
3912swap_buffers:
3913
3914 thresholds->spare = thresholds->primary;
3915
3916 if (!new) {
3917 kfree(thresholds->spare);
3918 thresholds->spare = NULL;
3919 }
3920
3921 rcu_assign_pointer(thresholds->primary, new);
3922
3923
3924 synchronize_rcu();
3925unlock:
3926 mutex_unlock(&memcg->thresholds_lock);
3927}
3928
3929static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3930 struct eventfd_ctx *eventfd)
3931{
3932 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
3933}
3934
3935static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3936 struct eventfd_ctx *eventfd)
3937{
3938 return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
3939}
3940
3941static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
3942 struct eventfd_ctx *eventfd, const char *args)
3943{
3944 struct mem_cgroup_eventfd_list *event;
3945
3946 event = kmalloc(sizeof(*event), GFP_KERNEL);
3947 if (!event)
3948 return -ENOMEM;
3949
3950 spin_lock(&memcg_oom_lock);
3951
3952 event->eventfd = eventfd;
3953 list_add(&event->list, &memcg->oom_notify);
3954
3955
3956 if (atomic_read(&memcg->under_oom))
3957 eventfd_signal(eventfd, 1);
3958 spin_unlock(&memcg_oom_lock);
3959
3960 return 0;
3961}
3962
3963static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
3964 struct eventfd_ctx *eventfd)
3965{
3966 struct mem_cgroup_eventfd_list *ev, *tmp;
3967
3968 spin_lock(&memcg_oom_lock);
3969
3970 list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
3971 if (ev->eventfd == eventfd) {
3972 list_del(&ev->list);
3973 kfree(ev);
3974 }
3975 }
3976
3977 spin_unlock(&memcg_oom_lock);
3978}
3979
3980static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
3981{
3982 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
3983
3984 seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
3985 seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom));
3986 return 0;
3987}
3988
3989static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
3990 struct cftype *cft, u64 val)
3991{
3992 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3993
3994
3995 if (!css->parent || !((val == 0) || (val == 1)))
3996 return -EINVAL;
3997
3998 memcg->oom_kill_disable = val;
3999 if (!val)
4000 memcg_oom_recover(memcg);
4001
4002 return 0;
4003}
4004
4005#ifdef CONFIG_MEMCG_KMEM
4006static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
4007{
4008 int ret;
4009
4010 ret = memcg_propagate_kmem(memcg);
4011 if (ret)
4012 return ret;
4013
4014 return mem_cgroup_sockets_init(memcg, ss);
4015}
4016
4017static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
4018{
4019 struct cgroup_subsys_state *css;
4020 struct mem_cgroup *parent, *child;
4021 int kmemcg_id;
4022
4023 if (!memcg->kmem_acct_active)
4024 return;
4025
4026
4027
4028
4029
4030
4031
4032 memcg->kmem_acct_active = false;
4033
4034 memcg_deactivate_kmem_caches(memcg);
4035
4036 kmemcg_id = memcg->kmemcg_id;
4037 BUG_ON(kmemcg_id < 0);
4038
4039 parent = parent_mem_cgroup(memcg);
4040 if (!parent)
4041 parent = root_mem_cgroup;
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051 css_for_each_descendant_pre(css, &memcg->css) {
4052 child = mem_cgroup_from_css(css);
4053 BUG_ON(child->kmemcg_id != kmemcg_id);
4054 child->kmemcg_id = parent->kmemcg_id;
4055 if (!memcg->use_hierarchy)
4056 break;
4057 }
4058 memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
4059
4060 memcg_free_cache_id(kmemcg_id);
4061}
4062
4063static void memcg_destroy_kmem(struct mem_cgroup *memcg)
4064{
4065 if (memcg->kmem_acct_activated) {
4066 memcg_destroy_kmem_caches(memcg);
4067 static_key_slow_dec(&memcg_kmem_enabled_key);
4068 WARN_ON(page_counter_read(&memcg->kmem));
4069 }
4070 mem_cgroup_sockets_destroy(memcg);
4071}
4072#else
4073static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
4074{
4075 return 0;
4076}
4077
4078static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
4079{
4080}
4081
4082static void memcg_destroy_kmem(struct mem_cgroup *memcg)
4083{
4084}
4085#endif
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105static void memcg_event_remove(struct work_struct *work)
4106{
4107 struct mem_cgroup_event *event =
4108 container_of(work, struct mem_cgroup_event, remove);
4109 struct mem_cgroup *memcg = event->memcg;
4110
4111 remove_wait_queue(event->wqh, &event->wait);
4112
4113 event->unregister_event(memcg, event->eventfd);
4114
4115
4116 eventfd_signal(event->eventfd, 1);
4117
4118 eventfd_ctx_put(event->eventfd);
4119 kfree(event);
4120 css_put(&memcg->css);
4121}
4122
4123
4124
4125
4126
4127
4128static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
4129 int sync, void *key)
4130{
4131 struct mem_cgroup_event *event =
4132 container_of(wait, struct mem_cgroup_event, wait);
4133 struct mem_cgroup *memcg = event->memcg;
4134 unsigned long flags = (unsigned long)key;
4135
4136 if (flags & POLLHUP) {
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146 spin_lock(&memcg->event_list_lock);
4147 if (!list_empty(&event->list)) {
4148 list_del_init(&event->list);
4149
4150
4151
4152
4153 schedule_work(&event->remove);
4154 }
4155 spin_unlock(&memcg->event_list_lock);
4156 }
4157
4158 return 0;
4159}
4160
4161static void memcg_event_ptable_queue_proc(struct file *file,
4162 wait_queue_head_t *wqh, poll_table *pt)
4163{
4164 struct mem_cgroup_event *event =
4165 container_of(pt, struct mem_cgroup_event, pt);
4166
4167 event->wqh = wqh;
4168 add_wait_queue(wqh, &event->wait);
4169}
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
4180 char *buf, size_t nbytes, loff_t off)
4181{
4182 struct cgroup_subsys_state *css = of_css(of);
4183 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4184 struct mem_cgroup_event *event;
4185 struct cgroup_subsys_state *cfile_css;
4186 unsigned int efd, cfd;
4187 struct fd efile;
4188 struct fd cfile;
4189 const char *name;
4190 char *endp;
4191 int ret;
4192
4193 buf = strstrip(buf);
4194
4195 efd = simple_strtoul(buf, &endp, 10);
4196 if (*endp != ' ')
4197 return -EINVAL;
4198 buf = endp + 1;
4199
4200 cfd = simple_strtoul(buf, &endp, 10);
4201 if ((*endp != ' ') && (*endp != '\0'))
4202 return -EINVAL;
4203 buf = endp + 1;
4204
4205 event = kzalloc(sizeof(*event), GFP_KERNEL);
4206 if (!event)
4207 return -ENOMEM;
4208
4209 event->memcg = memcg;
4210 INIT_LIST_HEAD(&event->list);
4211 init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
4212 init_waitqueue_func_entry(&event->wait, memcg_event_wake);
4213 INIT_WORK(&event->remove, memcg_event_remove);
4214
4215 efile = fdget(efd);
4216 if (!efile.file) {
4217 ret = -EBADF;
4218 goto out_kfree;
4219 }
4220
4221 event->eventfd = eventfd_ctx_fileget(efile.file);
4222 if (IS_ERR(event->eventfd)) {
4223 ret = PTR_ERR(event->eventfd);
4224 goto out_put_efile;
4225 }
4226
4227 cfile = fdget(cfd);
4228 if (!cfile.file) {
4229 ret = -EBADF;
4230 goto out_put_eventfd;
4231 }
4232
4233
4234
4235 ret = inode_permission(file_inode(cfile.file), MAY_READ);
4236 if (ret < 0)
4237 goto out_put_cfile;
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247 name = cfile.file->f_path.dentry->d_name.name;
4248
4249 if (!strcmp(name, "memory.usage_in_bytes")) {
4250 event->register_event = mem_cgroup_usage_register_event;
4251 event->unregister_event = mem_cgroup_usage_unregister_event;
4252 } else if (!strcmp(name, "memory.oom_control")) {
4253 event->register_event = mem_cgroup_oom_register_event;
4254 event->unregister_event = mem_cgroup_oom_unregister_event;
4255 } else if (!strcmp(name, "memory.pressure_level")) {
4256 event->register_event = vmpressure_register_event;
4257 event->unregister_event = vmpressure_unregister_event;
4258 } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
4259 event->register_event = memsw_cgroup_usage_register_event;
4260 event->unregister_event = memsw_cgroup_usage_unregister_event;
4261 } else {
4262 ret = -EINVAL;
4263 goto out_put_cfile;
4264 }
4265
4266
4267
4268
4269
4270
4271 cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
4272 &memory_cgrp_subsys);
4273 ret = -EINVAL;
4274 if (IS_ERR(cfile_css))
4275 goto out_put_cfile;
4276 if (cfile_css != css) {
4277 css_put(cfile_css);
4278 goto out_put_cfile;
4279 }
4280
4281 ret = event->register_event(memcg, event->eventfd, buf);
4282 if (ret)
4283 goto out_put_css;
4284
4285 efile.file->f_op->poll(efile.file, &event->pt);
4286
4287 spin_lock(&memcg->event_list_lock);
4288 list_add(&event->list, &memcg->event_list);
4289 spin_unlock(&memcg->event_list_lock);
4290
4291 fdput(cfile);
4292 fdput(efile);
4293
4294 return nbytes;
4295
4296out_put_css:
4297 css_put(css);
4298out_put_cfile:
4299 fdput(cfile);
4300out_put_eventfd:
4301 eventfd_ctx_put(event->eventfd);
4302out_put_efile:
4303 fdput(efile);
4304out_kfree:
4305 kfree(event);
4306
4307 return ret;
4308}
4309
4310static struct cftype mem_cgroup_legacy_files[] = {
4311 {
4312 .name = "usage_in_bytes",
4313 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
4314 .read_u64 = mem_cgroup_read_u64,
4315 },
4316 {
4317 .name = "max_usage_in_bytes",
4318 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
4319 .write = mem_cgroup_reset,
4320 .read_u64 = mem_cgroup_read_u64,
4321 },
4322 {
4323 .name = "limit_in_bytes",
4324 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
4325 .write = mem_cgroup_write,
4326 .read_u64 = mem_cgroup_read_u64,
4327 },
4328 {
4329 .name = "soft_limit_in_bytes",
4330 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4331 .write = mem_cgroup_write,
4332 .read_u64 = mem_cgroup_read_u64,
4333 },
4334 {
4335 .name = "failcnt",
4336 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
4337 .write = mem_cgroup_reset,
4338 .read_u64 = mem_cgroup_read_u64,
4339 },
4340 {
4341 .name = "stat",
4342 .seq_show = memcg_stat_show,
4343 },
4344 {
4345 .name = "force_empty",
4346 .write = mem_cgroup_force_empty_write,
4347 },
4348 {
4349 .name = "use_hierarchy",
4350 .write_u64 = mem_cgroup_hierarchy_write,
4351 .read_u64 = mem_cgroup_hierarchy_read,
4352 },
4353 {
4354 .name = "cgroup.event_control",
4355 .write = memcg_write_event_control,
4356 .flags = CFTYPE_NO_PREFIX,
4357 .mode = S_IWUGO,
4358 },
4359 {
4360 .name = "swappiness",
4361 .read_u64 = mem_cgroup_swappiness_read,
4362 .write_u64 = mem_cgroup_swappiness_write,
4363 },
4364 {
4365 .name = "move_charge_at_immigrate",
4366 .read_u64 = mem_cgroup_move_charge_read,
4367 .write_u64 = mem_cgroup_move_charge_write,
4368 },
4369 {
4370 .name = "oom_control",
4371 .seq_show = mem_cgroup_oom_control_read,
4372 .write_u64 = mem_cgroup_oom_control_write,
4373 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4374 },
4375 {
4376 .name = "pressure_level",
4377 },
4378#ifdef CONFIG_NUMA
4379 {
4380 .name = "numa_stat",
4381 .seq_show = memcg_numa_stat_show,
4382 },
4383#endif
4384#ifdef CONFIG_MEMCG_KMEM
4385 {
4386 .name = "kmem.limit_in_bytes",
4387 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
4388 .write = mem_cgroup_write,
4389 .read_u64 = mem_cgroup_read_u64,
4390 },
4391 {
4392 .name = "kmem.usage_in_bytes",
4393 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
4394 .read_u64 = mem_cgroup_read_u64,
4395 },
4396 {
4397 .name = "kmem.failcnt",
4398 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
4399 .write = mem_cgroup_reset,
4400 .read_u64 = mem_cgroup_read_u64,
4401 },
4402 {
4403 .name = "kmem.max_usage_in_bytes",
4404 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
4405 .write = mem_cgroup_reset,
4406 .read_u64 = mem_cgroup_read_u64,
4407 },
4408#ifdef CONFIG_SLABINFO
4409 {
4410 .name = "kmem.slabinfo",
4411 .seq_start = slab_start,
4412 .seq_next = slab_next,
4413 .seq_stop = slab_stop,
4414 .seq_show = memcg_slab_show,
4415 },
4416#endif
4417#endif
4418 { },
4419};
4420
4421static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4422{
4423 struct mem_cgroup_per_node *pn;
4424 struct mem_cgroup_per_zone *mz;
4425 int zone, tmp = node;
4426
4427
4428
4429
4430
4431
4432
4433
4434 if (!node_state(node, N_NORMAL_MEMORY))
4435 tmp = -1;
4436 pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4437 if (!pn)
4438 return 1;
4439
4440 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4441 mz = &pn->zoneinfo[zone];
4442 lruvec_init(&mz->lruvec);
4443 mz->usage_in_excess = 0;
4444 mz->on_tree = false;
4445 mz->memcg = memcg;
4446 }
4447 memcg->nodeinfo[node] = pn;
4448 return 0;
4449}
4450
4451static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4452{
4453 kfree(memcg->nodeinfo[node]);
4454}
4455
4456static struct mem_cgroup *mem_cgroup_alloc(void)
4457{
4458 struct mem_cgroup *memcg;
4459 size_t size;
4460
4461 size = sizeof(struct mem_cgroup);
4462 size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
4463
4464 memcg = kzalloc(size, GFP_KERNEL);
4465 if (!memcg)
4466 return NULL;
4467
4468 memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4469 if (!memcg->stat)
4470 goto out_free;
4471 spin_lock_init(&memcg->pcp_counter_lock);
4472 return memcg;
4473
4474out_free:
4475 kfree(memcg);
4476 return NULL;
4477}
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490static void __mem_cgroup_free(struct mem_cgroup *memcg)
4491{
4492 int node;
4493
4494 mem_cgroup_remove_from_trees(memcg);
4495
4496 for_each_node(node)
4497 free_mem_cgroup_per_zone_info(memcg, node);
4498
4499 free_percpu(memcg->stat);
4500 kfree(memcg);
4501}
4502
4503
4504
4505
4506struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
4507{
4508 if (!memcg->memory.parent)
4509 return NULL;
4510 return mem_cgroup_from_counter(memcg->memory.parent, memory);
4511}
4512EXPORT_SYMBOL(parent_mem_cgroup);
4513
4514static struct cgroup_subsys_state * __ref
4515mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
4516{
4517 struct mem_cgroup *memcg;
4518 long error = -ENOMEM;
4519 int node;
4520
4521 memcg = mem_cgroup_alloc();
4522 if (!memcg)
4523 return ERR_PTR(error);
4524
4525 for_each_node(node)
4526 if (alloc_mem_cgroup_per_zone_info(memcg, node))
4527 goto free_out;
4528
4529
4530 if (parent_css == NULL) {
4531 root_mem_cgroup = memcg;
4532 page_counter_init(&memcg->memory, NULL);
4533 memcg->high = PAGE_COUNTER_MAX;
4534 memcg->soft_limit = PAGE_COUNTER_MAX;
4535 page_counter_init(&memcg->memsw, NULL);
4536 page_counter_init(&memcg->kmem, NULL);
4537 }
4538
4539 memcg->last_scanned_node = MAX_NUMNODES;
4540 INIT_LIST_HEAD(&memcg->oom_notify);
4541 memcg->move_charge_at_immigrate = 0;
4542 mutex_init(&memcg->thresholds_lock);
4543 spin_lock_init(&memcg->move_lock);
4544 vmpressure_init(&memcg->vmpressure);
4545 INIT_LIST_HEAD(&memcg->event_list);
4546 spin_lock_init(&memcg->event_list_lock);
4547#ifdef CONFIG_MEMCG_KMEM
4548 memcg->kmemcg_id = -1;
4549#endif
4550
4551 return &memcg->css;
4552
4553free_out:
4554 __mem_cgroup_free(memcg);
4555 return ERR_PTR(error);
4556}
4557
4558static int
4559mem_cgroup_css_online(struct cgroup_subsys_state *css)
4560{
4561 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4562 struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
4563 int ret;
4564
4565 if (css->id > MEM_CGROUP_ID_MAX)
4566 return -ENOSPC;
4567
4568 if (!parent)
4569 return 0;
4570
4571 mutex_lock(&memcg_create_mutex);
4572
4573 memcg->use_hierarchy = parent->use_hierarchy;
4574 memcg->oom_kill_disable = parent->oom_kill_disable;
4575 memcg->swappiness = mem_cgroup_swappiness(parent);
4576
4577 if (parent->use_hierarchy) {
4578 page_counter_init(&memcg->memory, &parent->memory);
4579 memcg->high = PAGE_COUNTER_MAX;
4580 memcg->soft_limit = PAGE_COUNTER_MAX;
4581 page_counter_init(&memcg->memsw, &parent->memsw);
4582 page_counter_init(&memcg->kmem, &parent->kmem);
4583
4584
4585
4586
4587
4588 } else {
4589 page_counter_init(&memcg->memory, NULL);
4590 memcg->high = PAGE_COUNTER_MAX;
4591 memcg->soft_limit = PAGE_COUNTER_MAX;
4592 page_counter_init(&memcg->memsw, NULL);
4593 page_counter_init(&memcg->kmem, NULL);
4594
4595
4596
4597
4598
4599 if (parent != root_mem_cgroup)
4600 memory_cgrp_subsys.broken_hierarchy = true;
4601 }
4602 mutex_unlock(&memcg_create_mutex);
4603
4604 ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
4605 if (ret)
4606 return ret;
4607
4608
4609
4610
4611
4612
4613 smp_store_release(&memcg->initialized, 1);
4614
4615 return 0;
4616}
4617
4618static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
4619{
4620 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4621 struct mem_cgroup_event *event, *tmp;
4622
4623
4624
4625
4626
4627
4628 spin_lock(&memcg->event_list_lock);
4629 list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
4630 list_del_init(&event->list);
4631 schedule_work(&event->remove);
4632 }
4633 spin_unlock(&memcg->event_list_lock);
4634
4635 vmpressure_cleanup(&memcg->vmpressure);
4636
4637 memcg_deactivate_kmem(memcg);
4638}
4639
4640static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
4641{
4642 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4643
4644 memcg_destroy_kmem(memcg);
4645 __mem_cgroup_free(memcg);
4646}
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4662{
4663 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4664
4665 mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX);
4666 mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX);
4667 memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX);
4668 memcg->low = 0;
4669 memcg->high = PAGE_COUNTER_MAX;
4670 memcg->soft_limit = PAGE_COUNTER_MAX;
4671}
4672
4673#ifdef CONFIG_MMU
4674
4675static int mem_cgroup_do_precharge(unsigned long count)
4676{
4677 int ret;
4678
4679
4680 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count);
4681 if (!ret) {
4682 mc.precharge += count;
4683 return ret;
4684 }
4685 if (ret == -EINTR) {
4686 cancel_charge(root_mem_cgroup, count);
4687 return ret;
4688 }
4689
4690
4691 while (count--) {
4692 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
4693
4694
4695
4696
4697
4698
4699 if (ret == -EINTR)
4700 cancel_charge(root_mem_cgroup, 1);
4701 if (ret)
4702 return ret;
4703 mc.precharge++;
4704 cond_resched();
4705 }
4706 return 0;
4707}
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727union mc_target {
4728 struct page *page;
4729 swp_entry_t ent;
4730};
4731
4732enum mc_target_type {
4733 MC_TARGET_NONE = 0,
4734 MC_TARGET_PAGE,
4735 MC_TARGET_SWAP,
4736};
4737
4738static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4739 unsigned long addr, pte_t ptent)
4740{
4741 struct page *page = vm_normal_page(vma, addr, ptent);
4742
4743 if (!page || !page_mapped(page))
4744 return NULL;
4745 if (PageAnon(page)) {
4746 if (!(mc.flags & MOVE_ANON))
4747 return NULL;
4748 } else {
4749 if (!(mc.flags & MOVE_FILE))
4750 return NULL;
4751 }
4752 if (!get_page_unless_zero(page))
4753 return NULL;
4754
4755 return page;
4756}
4757
4758#ifdef CONFIG_SWAP
4759static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4760 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4761{
4762 struct page *page = NULL;
4763 swp_entry_t ent = pte_to_swp_entry(ptent);
4764
4765 if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
4766 return NULL;
4767
4768
4769
4770
4771 page = find_get_page(swap_address_space(ent), ent.val);
4772 if (do_swap_account)
4773 entry->val = ent.val;
4774
4775 return page;
4776}
4777#else
4778static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4779 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4780{
4781 return NULL;
4782}
4783#endif
4784
4785static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4786 unsigned long addr, pte_t ptent, swp_entry_t *entry)
4787{
4788 struct page *page = NULL;
4789 struct address_space *mapping;
4790 pgoff_t pgoff;
4791
4792 if (!vma->vm_file)
4793 return NULL;
4794 if (!(mc.flags & MOVE_FILE))
4795 return NULL;
4796
4797 mapping = vma->vm_file->f_mapping;
4798 pgoff = linear_page_index(vma, addr);
4799
4800
4801#ifdef CONFIG_SWAP
4802
4803 if (shmem_mapping(mapping)) {
4804 page = find_get_entry(mapping, pgoff);
4805 if (radix_tree_exceptional_entry(page)) {
4806 swp_entry_t swp = radix_to_swp_entry(page);
4807 if (do_swap_account)
4808 *entry = swp;
4809 page = find_get_page(swap_address_space(swp), swp.val);
4810 }
4811 } else
4812 page = find_get_page(mapping, pgoff);
4813#else
4814 page = find_get_page(mapping, pgoff);
4815#endif
4816 return page;
4817}
4818
4819static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
4820 unsigned long addr, pte_t ptent, union mc_target *target)
4821{
4822 struct page *page = NULL;
4823 enum mc_target_type ret = MC_TARGET_NONE;
4824 swp_entry_t ent = { .val = 0 };
4825
4826 if (pte_present(ptent))
4827 page = mc_handle_present_pte(vma, addr, ptent);
4828 else if (is_swap_pte(ptent))
4829 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
4830 else if (pte_none(ptent))
4831 page = mc_handle_file_pte(vma, addr, ptent, &ent);
4832
4833 if (!page && !ent.val)
4834 return ret;
4835 if (page) {
4836
4837
4838
4839
4840
4841 if (page->mem_cgroup == mc.from) {
4842 ret = MC_TARGET_PAGE;
4843 if (target)
4844 target->page = page;
4845 }
4846 if (!ret || !target)
4847 put_page(page);
4848 }
4849
4850 if (ent.val && !ret &&
4851 mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
4852 ret = MC_TARGET_SWAP;
4853 if (target)
4854 target->ent = ent;
4855 }
4856 return ret;
4857}
4858
4859#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4860
4861
4862
4863
4864
4865static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4866 unsigned long addr, pmd_t pmd, union mc_target *target)
4867{
4868 struct page *page = NULL;
4869 enum mc_target_type ret = MC_TARGET_NONE;
4870
4871 page = pmd_page(pmd);
4872 VM_BUG_ON_PAGE(!page || !PageHead(page), page);
4873 if (!(mc.flags & MOVE_ANON))
4874 return ret;
4875 if (page->mem_cgroup == mc.from) {
4876 ret = MC_TARGET_PAGE;
4877 if (target) {
4878 get_page(page);
4879 target->page = page;
4880 }
4881 }
4882 return ret;
4883}
4884#else
4885static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4886 unsigned long addr, pmd_t pmd, union mc_target *target)
4887{
4888 return MC_TARGET_NONE;
4889}
4890#endif
4891
4892static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4893 unsigned long addr, unsigned long end,
4894 struct mm_walk *walk)
4895{
4896 struct vm_area_struct *vma = walk->vma;
4897 pte_t *pte;
4898 spinlock_t *ptl;
4899
4900 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
4901 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4902 mc.precharge += HPAGE_PMD_NR;
4903 spin_unlock(ptl);
4904 return 0;
4905 }
4906
4907 if (pmd_trans_unstable(pmd))
4908 return 0;
4909 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4910 for (; addr != end; pte++, addr += PAGE_SIZE)
4911 if (get_mctgt_type(vma, addr, *pte, NULL))
4912 mc.precharge++;
4913 pte_unmap_unlock(pte - 1, ptl);
4914 cond_resched();
4915
4916 return 0;
4917}
4918
4919static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4920{
4921 unsigned long precharge;
4922
4923 struct mm_walk mem_cgroup_count_precharge_walk = {
4924 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4925 .mm = mm,
4926 };
4927 down_read(&mm->mmap_sem);
4928 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
4929 up_read(&mm->mmap_sem);
4930
4931 precharge = mc.precharge;
4932 mc.precharge = 0;
4933
4934 return precharge;
4935}
4936
4937static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4938{
4939 unsigned long precharge = mem_cgroup_count_precharge(mm);
4940
4941 VM_BUG_ON(mc.moving_task);
4942 mc.moving_task = current;
4943 return mem_cgroup_do_precharge(precharge);
4944}
4945
4946
4947static void __mem_cgroup_clear_mc(void)
4948{
4949 struct mem_cgroup *from = mc.from;
4950 struct mem_cgroup *to = mc.to;
4951
4952
4953 if (mc.precharge) {
4954 cancel_charge(mc.to, mc.precharge);
4955 mc.precharge = 0;
4956 }
4957
4958
4959
4960
4961 if (mc.moved_charge) {
4962 cancel_charge(mc.from, mc.moved_charge);
4963 mc.moved_charge = 0;
4964 }
4965
4966 if (mc.moved_swap) {
4967
4968 if (!mem_cgroup_is_root(mc.from))
4969 page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
4970
4971
4972
4973
4974
4975 if (!mem_cgroup_is_root(mc.to))
4976 page_counter_uncharge(&mc.to->memory, mc.moved_swap);
4977
4978 css_put_many(&mc.from->css, mc.moved_swap);
4979
4980
4981 mc.moved_swap = 0;
4982 }
4983 memcg_oom_recover(from);
4984 memcg_oom_recover(to);
4985 wake_up_all(&mc.waitq);
4986}
4987
4988static void mem_cgroup_clear_mc(void)
4989{
4990
4991
4992
4993
4994 mc.moving_task = NULL;
4995 __mem_cgroup_clear_mc();
4996 spin_lock(&mc.lock);
4997 mc.from = NULL;
4998 mc.to = NULL;
4999 spin_unlock(&mc.lock);
5000}
5001
5002static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
5003 struct cgroup_taskset *tset)
5004{
5005 struct task_struct *p = cgroup_taskset_first(tset);
5006 int ret = 0;
5007 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5008 unsigned long move_flags;
5009
5010
5011
5012
5013
5014
5015 move_flags = ACCESS_ONCE(memcg->move_charge_at_immigrate);
5016 if (move_flags) {
5017 struct mm_struct *mm;
5018 struct mem_cgroup *from = mem_cgroup_from_task(p);
5019
5020 VM_BUG_ON(from == memcg);
5021
5022 mm = get_task_mm(p);
5023 if (!mm)
5024 return 0;
5025
5026 if (mm->owner == p) {
5027 VM_BUG_ON(mc.from);
5028 VM_BUG_ON(mc.to);
5029 VM_BUG_ON(mc.precharge);
5030 VM_BUG_ON(mc.moved_charge);
5031 VM_BUG_ON(mc.moved_swap);
5032
5033 spin_lock(&mc.lock);
5034 mc.from = from;
5035 mc.to = memcg;
5036 mc.flags = move_flags;
5037 spin_unlock(&mc.lock);
5038
5039
5040 ret = mem_cgroup_precharge_mc(mm);
5041 if (ret)
5042 mem_cgroup_clear_mc();
5043 }
5044 mmput(mm);
5045 }
5046 return ret;
5047}
5048
5049static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
5050 struct cgroup_taskset *tset)
5051{
5052 if (mc.to)
5053 mem_cgroup_clear_mc();
5054}
5055
5056static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5057 unsigned long addr, unsigned long end,
5058 struct mm_walk *walk)
5059{
5060 int ret = 0;
5061 struct vm_area_struct *vma = walk->vma;
5062 pte_t *pte;
5063 spinlock_t *ptl;
5064 enum mc_target_type target_type;
5065 union mc_target target;
5066 struct page *page;
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
5079 if (mc.precharge < HPAGE_PMD_NR) {
5080 spin_unlock(ptl);
5081 return 0;
5082 }
5083 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5084 if (target_type == MC_TARGET_PAGE) {
5085 page = target.page;
5086 if (!isolate_lru_page(page)) {
5087 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
5088 mc.from, mc.to)) {
5089 mc.precharge -= HPAGE_PMD_NR;
5090 mc.moved_charge += HPAGE_PMD_NR;
5091 }
5092 putback_lru_page(page);
5093 }
5094 put_page(page);
5095 }
5096 spin_unlock(ptl);
5097 return 0;
5098 }
5099
5100 if (pmd_trans_unstable(pmd))
5101 return 0;
5102retry:
5103 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5104 for (; addr != end; addr += PAGE_SIZE) {
5105 pte_t ptent = *(pte++);
5106 swp_entry_t ent;
5107
5108 if (!mc.precharge)
5109 break;
5110
5111 switch (get_mctgt_type(vma, addr, ptent, &target)) {
5112 case MC_TARGET_PAGE:
5113 page = target.page;
5114 if (isolate_lru_page(page))
5115 goto put;
5116 if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) {
5117 mc.precharge--;
5118
5119 mc.moved_charge++;
5120 }
5121 putback_lru_page(page);
5122put:
5123 put_page(page);
5124 break;
5125 case MC_TARGET_SWAP:
5126 ent = target.ent;
5127 if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
5128 mc.precharge--;
5129
5130 mc.moved_swap++;
5131 }
5132 break;
5133 default:
5134 break;
5135 }
5136 }
5137 pte_unmap_unlock(pte - 1, ptl);
5138 cond_resched();
5139
5140 if (addr != end) {
5141
5142
5143
5144
5145
5146
5147 ret = mem_cgroup_do_precharge(1);
5148 if (!ret)
5149 goto retry;
5150 }
5151
5152 return ret;
5153}
5154
5155static void mem_cgroup_move_charge(struct mm_struct *mm)
5156{
5157 struct mm_walk mem_cgroup_move_charge_walk = {
5158 .pmd_entry = mem_cgroup_move_charge_pte_range,
5159 .mm = mm,
5160 };
5161
5162 lru_add_drain_all();
5163
5164
5165
5166
5167
5168 atomic_inc(&mc.from->moving_account);
5169 synchronize_rcu();
5170retry:
5171 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5172
5173
5174
5175
5176
5177
5178
5179 __mem_cgroup_clear_mc();
5180 cond_resched();
5181 goto retry;
5182 }
5183
5184
5185
5186
5187 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
5188 up_read(&mm->mmap_sem);
5189 atomic_dec(&mc.from->moving_account);
5190}
5191
5192static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
5193 struct cgroup_taskset *tset)
5194{
5195 struct task_struct *p = cgroup_taskset_first(tset);
5196 struct mm_struct *mm = get_task_mm(p);
5197
5198 if (mm) {
5199 if (mc.to)
5200 mem_cgroup_move_charge(mm);
5201 mmput(mm);
5202 }
5203 if (mc.to)
5204 mem_cgroup_clear_mc();
5205}
5206#else
5207static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
5208 struct cgroup_taskset *tset)
5209{
5210 return 0;
5211}
5212static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
5213 struct cgroup_taskset *tset)
5214{
5215}
5216static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
5217 struct cgroup_taskset *tset)
5218{
5219}
5220#endif
5221
5222
5223
5224
5225
5226
5227static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
5228{
5229
5230
5231
5232
5233
5234 if (cgroup_on_dfl(root_css->cgroup))
5235 root_mem_cgroup->use_hierarchy = true;
5236 else
5237 root_mem_cgroup->use_hierarchy = false;
5238}
5239
5240static u64 memory_current_read(struct cgroup_subsys_state *css,
5241 struct cftype *cft)
5242{
5243 return mem_cgroup_usage(mem_cgroup_from_css(css), false);
5244}
5245
5246static int memory_low_show(struct seq_file *m, void *v)
5247{
5248 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5249 unsigned long low = ACCESS_ONCE(memcg->low);
5250
5251 if (low == PAGE_COUNTER_MAX)
5252 seq_puts(m, "max\n");
5253 else
5254 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5255
5256 return 0;
5257}
5258
5259static ssize_t memory_low_write(struct kernfs_open_file *of,
5260 char *buf, size_t nbytes, loff_t off)
5261{
5262 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5263 unsigned long low;
5264 int err;
5265
5266 buf = strstrip(buf);
5267 err = page_counter_memparse(buf, "max", &low);
5268 if (err)
5269 return err;
5270
5271 memcg->low = low;
5272
5273 return nbytes;
5274}
5275
5276static int memory_high_show(struct seq_file *m, void *v)
5277{
5278 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5279 unsigned long high = ACCESS_ONCE(memcg->high);
5280
5281 if (high == PAGE_COUNTER_MAX)
5282 seq_puts(m, "max\n");
5283 else
5284 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5285
5286 return 0;
5287}
5288
5289static ssize_t memory_high_write(struct kernfs_open_file *of,
5290 char *buf, size_t nbytes, loff_t off)
5291{
5292 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5293 unsigned long high;
5294 int err;
5295
5296 buf = strstrip(buf);
5297 err = page_counter_memparse(buf, "max", &high);
5298 if (err)
5299 return err;
5300
5301 memcg->high = high;
5302
5303 return nbytes;
5304}
5305
5306static int memory_max_show(struct seq_file *m, void *v)
5307{
5308 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5309 unsigned long max = ACCESS_ONCE(memcg->memory.limit);
5310
5311 if (max == PAGE_COUNTER_MAX)
5312 seq_puts(m, "max\n");
5313 else
5314 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5315
5316 return 0;
5317}
5318
5319static ssize_t memory_max_write(struct kernfs_open_file *of,
5320 char *buf, size_t nbytes, loff_t off)
5321{
5322 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5323 unsigned long max;
5324 int err;
5325
5326 buf = strstrip(buf);
5327 err = page_counter_memparse(buf, "max", &max);
5328 if (err)
5329 return err;
5330
5331 err = mem_cgroup_resize_limit(memcg, max);
5332 if (err)
5333 return err;
5334
5335 return nbytes;
5336}
5337
5338static int memory_events_show(struct seq_file *m, void *v)
5339{
5340 struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5341
5342 seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5343 seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5344 seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5345 seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5346
5347 return 0;
5348}
5349
5350static struct cftype memory_files[] = {
5351 {
5352 .name = "current",
5353 .read_u64 = memory_current_read,
5354 },
5355 {
5356 .name = "low",
5357 .flags = CFTYPE_NOT_ON_ROOT,
5358 .seq_show = memory_low_show,
5359 .write = memory_low_write,
5360 },
5361 {
5362 .name = "high",
5363 .flags = CFTYPE_NOT_ON_ROOT,
5364 .seq_show = memory_high_show,
5365 .write = memory_high_write,
5366 },
5367 {
5368 .name = "max",
5369 .flags = CFTYPE_NOT_ON_ROOT,
5370 .seq_show = memory_max_show,
5371 .write = memory_max_write,
5372 },
5373 {
5374 .name = "events",
5375 .flags = CFTYPE_NOT_ON_ROOT,
5376 .seq_show = memory_events_show,
5377 },
5378 { }
5379};
5380
5381struct cgroup_subsys memory_cgrp_subsys = {
5382 .css_alloc = mem_cgroup_css_alloc,
5383 .css_online = mem_cgroup_css_online,
5384 .css_offline = mem_cgroup_css_offline,
5385 .css_free = mem_cgroup_css_free,
5386 .css_reset = mem_cgroup_css_reset,
5387 .can_attach = mem_cgroup_can_attach,
5388 .cancel_attach = mem_cgroup_cancel_attach,
5389 .attach = mem_cgroup_move_task,
5390 .bind = mem_cgroup_bind,
5391 .dfl_cftypes = memory_files,
5392 .legacy_cftypes = mem_cgroup_legacy_files,
5393 .early_init = 0,
5394};
5395
5396
5397
5398
5399
5400
5401
5402void mem_cgroup_events(struct mem_cgroup *memcg,
5403 enum mem_cgroup_events_index idx,
5404 unsigned int nr)
5405{
5406 this_cpu_add(memcg->stat->events[idx], nr);
5407}
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5418{
5419 if (mem_cgroup_disabled())
5420 return false;
5421
5422
5423
5424
5425
5426
5427
5428 if (memcg == root_mem_cgroup)
5429 return false;
5430
5431 if (page_counter_read(&memcg->memory) >= memcg->low)
5432 return false;
5433
5434 while (memcg != root) {
5435 memcg = parent_mem_cgroup(memcg);
5436
5437 if (memcg == root_mem_cgroup)
5438 break;
5439
5440 if (page_counter_read(&memcg->memory) >= memcg->low)
5441 return false;
5442 }
5443 return true;
5444}
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5464 gfp_t gfp_mask, struct mem_cgroup **memcgp)
5465{
5466 struct mem_cgroup *memcg = NULL;
5467 unsigned int nr_pages = 1;
5468 int ret = 0;
5469
5470 if (mem_cgroup_disabled())
5471 goto out;
5472
5473 if (PageSwapCache(page)) {
5474
5475
5476
5477
5478
5479
5480
5481 if (page->mem_cgroup)
5482 goto out;
5483 }
5484
5485 if (PageTransHuge(page)) {
5486 nr_pages <<= compound_order(page);
5487 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5488 }
5489
5490 if (do_swap_account && PageSwapCache(page))
5491 memcg = try_get_mem_cgroup_from_page(page);
5492 if (!memcg)
5493 memcg = get_mem_cgroup_from_mm(mm);
5494
5495 ret = try_charge(memcg, gfp_mask, nr_pages);
5496
5497 css_put(&memcg->css);
5498
5499 if (ret == -EINTR) {
5500 memcg = root_mem_cgroup;
5501 ret = 0;
5502 }
5503out:
5504 *memcgp = memcg;
5505 return ret;
5506}
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5525 bool lrucare)
5526{
5527 unsigned int nr_pages = 1;
5528
5529 VM_BUG_ON_PAGE(!page->mapping, page);
5530 VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5531
5532 if (mem_cgroup_disabled())
5533 return;
5534
5535
5536
5537
5538
5539 if (!memcg)
5540 return;
5541
5542 commit_charge(page, memcg, lrucare);
5543
5544 if (PageTransHuge(page)) {
5545 nr_pages <<= compound_order(page);
5546 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5547 }
5548
5549 local_irq_disable();
5550 mem_cgroup_charge_statistics(memcg, page, nr_pages);
5551 memcg_check_events(memcg, page);
5552 local_irq_enable();
5553
5554 if (do_swap_account && PageSwapCache(page)) {
5555 swp_entry_t entry = { .val = page_private(page) };
5556
5557
5558
5559
5560
5561 mem_cgroup_uncharge_swap(entry);
5562 }
5563}
5564
5565
5566
5567
5568
5569
5570
5571
5572void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg)
5573{
5574 unsigned int nr_pages = 1;
5575
5576 if (mem_cgroup_disabled())
5577 return;
5578
5579
5580
5581
5582
5583 if (!memcg)
5584 return;
5585
5586 if (PageTransHuge(page)) {
5587 nr_pages <<= compound_order(page);
5588 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5589 }
5590
5591 cancel_charge(memcg, nr_pages);
5592}
5593
5594static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5595 unsigned long nr_anon, unsigned long nr_file,
5596 unsigned long nr_huge, struct page *dummy_page)
5597{
5598 unsigned long nr_pages = nr_anon + nr_file;
5599 unsigned long flags;
5600
5601 if (!mem_cgroup_is_root(memcg)) {
5602 page_counter_uncharge(&memcg->memory, nr_pages);
5603 if (do_swap_account)
5604 page_counter_uncharge(&memcg->memsw, nr_pages);
5605 memcg_oom_recover(memcg);
5606 }
5607
5608 local_irq_save(flags);
5609 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5610 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5611 __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5612 __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
5613 __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
5614 memcg_check_events(memcg, dummy_page);
5615 local_irq_restore(flags);
5616
5617 if (!mem_cgroup_is_root(memcg))
5618 css_put_many(&memcg->css, nr_pages);
5619}
5620
5621static void uncharge_list(struct list_head *page_list)
5622{
5623 struct mem_cgroup *memcg = NULL;
5624 unsigned long nr_anon = 0;
5625 unsigned long nr_file = 0;
5626 unsigned long nr_huge = 0;
5627 unsigned long pgpgout = 0;
5628 struct list_head *next;
5629 struct page *page;
5630
5631 next = page_list->next;
5632 do {
5633 unsigned int nr_pages = 1;
5634
5635 page = list_entry(next, struct page, lru);
5636 next = page->lru.next;
5637
5638 VM_BUG_ON_PAGE(PageLRU(page), page);
5639 VM_BUG_ON_PAGE(page_count(page), page);
5640
5641 if (!page->mem_cgroup)
5642 continue;
5643
5644
5645
5646
5647
5648
5649
5650 if (memcg != page->mem_cgroup) {
5651 if (memcg) {
5652 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5653 nr_huge, page);
5654 pgpgout = nr_anon = nr_file = nr_huge = 0;
5655 }
5656 memcg = page->mem_cgroup;
5657 }
5658
5659 if (PageTransHuge(page)) {
5660 nr_pages <<= compound_order(page);
5661 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5662 nr_huge += nr_pages;
5663 }
5664
5665 if (PageAnon(page))
5666 nr_anon += nr_pages;
5667 else
5668 nr_file += nr_pages;
5669
5670 page->mem_cgroup = NULL;
5671
5672 pgpgout++;
5673 } while (next != page_list);
5674
5675 if (memcg)
5676 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5677 nr_huge, page);
5678}
5679
5680
5681
5682
5683
5684
5685
5686
5687void mem_cgroup_uncharge(struct page *page)
5688{
5689 if (mem_cgroup_disabled())
5690 return;
5691
5692
5693 if (!page->mem_cgroup)
5694 return;
5695
5696 INIT_LIST_HEAD(&page->lru);
5697 uncharge_list(&page->lru);
5698}
5699
5700
5701
5702
5703
5704
5705
5706
5707void mem_cgroup_uncharge_list(struct list_head *page_list)
5708{
5709 if (mem_cgroup_disabled())
5710 return;
5711
5712 if (!list_empty(page_list))
5713 uncharge_list(page_list);
5714}
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
5727 bool lrucare)
5728{
5729 struct mem_cgroup *memcg;
5730 int isolated;
5731
5732 VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5733 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5734 VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage);
5735 VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage);
5736 VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
5737 VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5738 newpage);
5739
5740 if (mem_cgroup_disabled())
5741 return;
5742
5743
5744 if (newpage->mem_cgroup)
5745 return;
5746
5747
5748
5749
5750
5751
5752
5753 memcg = oldpage->mem_cgroup;
5754 if (!memcg)
5755 return;
5756
5757 if (lrucare)
5758 lock_page_lru(oldpage, &isolated);
5759
5760 oldpage->mem_cgroup = NULL;
5761
5762 if (lrucare)
5763 unlock_page_lru(oldpage, isolated);
5764
5765 commit_charge(newpage, memcg, lrucare);
5766}
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776static int __init mem_cgroup_init(void)
5777{
5778 int cpu, node;
5779
5780 hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
5781
5782 for_each_possible_cpu(cpu)
5783 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5784 drain_local_stock);
5785
5786 for_each_node(node) {
5787 struct mem_cgroup_tree_per_node *rtpn;
5788 int zone;
5789
5790 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5791 node_online(node) ? node : NUMA_NO_NODE);
5792
5793 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5794 struct mem_cgroup_tree_per_zone *rtpz;
5795
5796 rtpz = &rtpn->rb_tree_per_zone[zone];
5797 rtpz->rb_root = RB_ROOT;
5798 spin_lock_init(&rtpz->lock);
5799 }
5800 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5801 }
5802
5803 return 0;
5804}
5805subsys_initcall(mem_cgroup_init);
5806
5807#ifdef CONFIG_MEMCG_SWAP
5808
5809
5810
5811
5812
5813
5814
5815void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5816{
5817 struct mem_cgroup *memcg;
5818 unsigned short oldid;
5819
5820 VM_BUG_ON_PAGE(PageLRU(page), page);
5821 VM_BUG_ON_PAGE(page_count(page), page);
5822
5823 if (!do_swap_account)
5824 return;
5825
5826 memcg = page->mem_cgroup;
5827
5828
5829 if (!memcg)
5830 return;
5831
5832 oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5833 VM_BUG_ON_PAGE(oldid, page);
5834 mem_cgroup_swap_statistics(memcg, true);
5835
5836 page->mem_cgroup = NULL;
5837
5838 if (!mem_cgroup_is_root(memcg))
5839 page_counter_uncharge(&memcg->memory, 1);
5840
5841
5842 VM_BUG_ON(!irqs_disabled());
5843
5844 mem_cgroup_charge_statistics(memcg, page, -1);
5845 memcg_check_events(memcg, page);
5846}
5847
5848
5849
5850
5851
5852
5853
5854void mem_cgroup_uncharge_swap(swp_entry_t entry)
5855{
5856 struct mem_cgroup *memcg;
5857 unsigned short id;
5858
5859 if (!do_swap_account)
5860 return;
5861
5862 id = swap_cgroup_record(entry, 0);
5863 rcu_read_lock();
5864 memcg = mem_cgroup_lookup(id);
5865 if (memcg) {
5866 if (!mem_cgroup_is_root(memcg))
5867 page_counter_uncharge(&memcg->memsw, 1);
5868 mem_cgroup_swap_statistics(memcg, false);
5869 css_put(&memcg->css);
5870 }
5871 rcu_read_unlock();
5872}
5873
5874
5875#ifdef CONFIG_MEMCG_SWAP_ENABLED
5876static int really_do_swap_account __initdata = 1;
5877#else
5878static int really_do_swap_account __initdata;
5879#endif
5880
5881static int __init enable_swap_account(char *s)
5882{
5883 if (!strcmp(s, "1"))
5884 really_do_swap_account = 1;
5885 else if (!strcmp(s, "0"))
5886 really_do_swap_account = 0;
5887 return 1;
5888}
5889__setup("swapaccount=", enable_swap_account);
5890
5891static struct cftype memsw_cgroup_files[] = {
5892 {
5893 .name = "memsw.usage_in_bytes",
5894 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5895 .read_u64 = mem_cgroup_read_u64,
5896 },
5897 {
5898 .name = "memsw.max_usage_in_bytes",
5899 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5900 .write = mem_cgroup_reset,
5901 .read_u64 = mem_cgroup_read_u64,
5902 },
5903 {
5904 .name = "memsw.limit_in_bytes",
5905 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5906 .write = mem_cgroup_write,
5907 .read_u64 = mem_cgroup_read_u64,
5908 },
5909 {
5910 .name = "memsw.failcnt",
5911 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5912 .write = mem_cgroup_reset,
5913 .read_u64 = mem_cgroup_read_u64,
5914 },
5915 { },
5916};
5917
5918static int __init mem_cgroup_swap_init(void)
5919{
5920 if (!mem_cgroup_disabled() && really_do_swap_account) {
5921 do_swap_account = 1;
5922 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5923 memsw_cgroup_files));
5924 }
5925 return 0;
5926}
5927subsys_initcall(mem_cgroup_swap_init);
5928
5929#endif
5930