~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/memcontrol.c

Version: ~ [ linux-6.0 ] ~ [ linux-5.19.12 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.71 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.146 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.215 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.260 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.295 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.330 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /* memcontrol.c - Memory Controller
  2  *
  3  * Copyright IBM Corporation, 2007
  4  * Author Balbir Singh <balbir@linux.vnet.ibm.com>
  5  *
  6  * Copyright 2007 OpenVZ SWsoft Inc
  7  * Author: Pavel Emelianov <xemul@openvz.org>
  8  *
  9  * Memory thresholds
 10  * Copyright (C) 2009 Nokia Corporation
 11  * Author: Kirill A. Shutemov
 12  *
 13  * Kernel Memory Controller
 14  * Copyright (C) 2012 Parallels Inc. and Google Inc.
 15  * Authors: Glauber Costa and Suleiman Souhlal
 16  *
 17  * Native page reclaim
 18  * Charge lifetime sanitation
 19  * Lockless page tracking & accounting
 20  * Unified hierarchy configuration model
 21  * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
 22  *
 23  * This program is free software; you can redistribute it and/or modify
 24  * it under the terms of the GNU General Public License as published by
 25  * the Free Software Foundation; either version 2 of the License, or
 26  * (at your option) any later version.
 27  *
 28  * This program is distributed in the hope that it will be useful,
 29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 31  * GNU General Public License for more details.
 32  */
 33 
 34 #include <linux/page_counter.h>
 35 #include <linux/memcontrol.h>
 36 #include <linux/cgroup.h>
 37 #include <linux/mm.h>
 38 #include <linux/hugetlb.h>
 39 #include <linux/pagemap.h>
 40 #include <linux/smp.h>
 41 #include <linux/page-flags.h>
 42 #include <linux/backing-dev.h>
 43 #include <linux/bit_spinlock.h>
 44 #include <linux/rcupdate.h>
 45 #include <linux/limits.h>
 46 #include <linux/export.h>
 47 #include <linux/mutex.h>
 48 #include <linux/rbtree.h>
 49 #include <linux/slab.h>
 50 #include <linux/swap.h>
 51 #include <linux/swapops.h>
 52 #include <linux/spinlock.h>
 53 #include <linux/eventfd.h>
 54 #include <linux/poll.h>
 55 #include <linux/sort.h>
 56 #include <linux/fs.h>
 57 #include <linux/seq_file.h>
 58 #include <linux/vmpressure.h>
 59 #include <linux/mm_inline.h>
 60 #include <linux/swap_cgroup.h>
 61 #include <linux/cpu.h>
 62 #include <linux/oom.h>
 63 #include <linux/lockdep.h>
 64 #include <linux/file.h>
 65 #include "internal.h"
 66 #include <net/sock.h>
 67 #include <net/ip.h>
 68 #include <net/tcp_memcontrol.h>
 69 #include "slab.h"
 70 
 71 #include <asm/uaccess.h>
 72 
 73 #include <trace/events/vmscan.h>
 74 
 75 struct cgroup_subsys memory_cgrp_subsys __read_mostly;
 76 EXPORT_SYMBOL(memory_cgrp_subsys);
 77 
 78 #define MEM_CGROUP_RECLAIM_RETRIES      5
 79 static struct mem_cgroup *root_mem_cgroup __read_mostly;
 80 
 81 /* Whether the swap controller is active */
 82 #ifdef CONFIG_MEMCG_SWAP
 83 int do_swap_account __read_mostly;
 84 #else
 85 #define do_swap_account         0
 86 #endif
 87 
 88 static const char * const mem_cgroup_stat_names[] = {
 89         "cache",
 90         "rss",
 91         "rss_huge",
 92         "mapped_file",
 93         "writeback",
 94         "swap",
 95 };
 96 
 97 static const char * const mem_cgroup_events_names[] = {
 98         "pgpgin",
 99         "pgpgout",
100         "pgfault",
101         "pgmajfault",
102 };
103 
104 static const char * const mem_cgroup_lru_names[] = {
105         "inactive_anon",
106         "active_anon",
107         "inactive_file",
108         "active_file",
109         "unevictable",
110 };
111 
112 /*
113  * Per memcg event counter is incremented at every pagein/pageout. With THP,
114  * it will be incremated by the number of pages. This counter is used for
115  * for trigger some periodic events. This is straightforward and better
116  * than using jiffies etc. to handle periodic memcg event.
117  */
118 enum mem_cgroup_events_target {
119         MEM_CGROUP_TARGET_THRESH,
120         MEM_CGROUP_TARGET_SOFTLIMIT,
121         MEM_CGROUP_TARGET_NUMAINFO,
122         MEM_CGROUP_NTARGETS,
123 };
124 #define THRESHOLDS_EVENTS_TARGET 128
125 #define SOFTLIMIT_EVENTS_TARGET 1024
126 #define NUMAINFO_EVENTS_TARGET  1024
127 
128 struct mem_cgroup_stat_cpu {
129         long count[MEM_CGROUP_STAT_NSTATS];
130         unsigned long events[MEMCG_NR_EVENTS];
131         unsigned long nr_page_events;
132         unsigned long targets[MEM_CGROUP_NTARGETS];
133 };
134 
135 struct reclaim_iter {
136         struct mem_cgroup *position;
137         /* scan generation, increased every round-trip */
138         unsigned int generation;
139 };
140 
141 /*
142  * per-zone information in memory controller.
143  */
144 struct mem_cgroup_per_zone {
145         struct lruvec           lruvec;
146         unsigned long           lru_size[NR_LRU_LISTS];
147 
148         struct reclaim_iter     iter[DEF_PRIORITY + 1];
149 
150         struct rb_node          tree_node;      /* RB tree node */
151         unsigned long           usage_in_excess;/* Set to the value by which */
152                                                 /* the soft limit is exceeded*/
153         bool                    on_tree;
154         struct mem_cgroup       *memcg;         /* Back pointer, we cannot */
155                                                 /* use container_of        */
156 };
157 
158 struct mem_cgroup_per_node {
159         struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
160 };
161 
162 /*
163  * Cgroups above their limits are maintained in a RB-Tree, independent of
164  * their hierarchy representation
165  */
166 
167 struct mem_cgroup_tree_per_zone {
168         struct rb_root rb_root;
169         spinlock_t lock;
170 };
171 
172 struct mem_cgroup_tree_per_node {
173         struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES];
174 };
175 
176 struct mem_cgroup_tree {
177         struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES];
178 };
179 
180 static struct mem_cgroup_tree soft_limit_tree __read_mostly;
181 
182 struct mem_cgroup_threshold {
183         struct eventfd_ctx *eventfd;
184         unsigned long threshold;
185 };
186 
187 /* For threshold */
188 struct mem_cgroup_threshold_ary {
189         /* An array index points to threshold just below or equal to usage. */
190         int current_threshold;
191         /* Size of entries[] */
192         unsigned int size;
193         /* Array of thresholds */
194         struct mem_cgroup_threshold entries[0];
195 };
196 
197 struct mem_cgroup_thresholds {
198         /* Primary thresholds array */
199         struct mem_cgroup_threshold_ary *primary;
200         /*
201          * Spare threshold array.
202          * This is needed to make mem_cgroup_unregister_event() "never fail".
203          * It must be able to store at least primary->size - 1 entries.
204          */
205         struct mem_cgroup_threshold_ary *spare;
206 };
207 
208 /* for OOM */
209 struct mem_cgroup_eventfd_list {
210         struct list_head list;
211         struct eventfd_ctx *eventfd;
212 };
213 
214 /*
215  * cgroup_event represents events which userspace want to receive.
216  */
217 struct mem_cgroup_event {
218         /*
219          * memcg which the event belongs to.
220          */
221         struct mem_cgroup *memcg;
222         /*
223          * eventfd to signal userspace about the event.
224          */
225         struct eventfd_ctx *eventfd;
226         /*
227          * Each of these stored in a list by the cgroup.
228          */
229         struct list_head list;
230         /*
231          * register_event() callback will be used to add new userspace
232          * waiter for changes related to this event.  Use eventfd_signal()
233          * on eventfd to send notification to userspace.
234          */
235         int (*register_event)(struct mem_cgroup *memcg,
236                               struct eventfd_ctx *eventfd, const char *args);
237         /*
238          * unregister_event() callback will be called when userspace closes
239          * the eventfd or on cgroup removing.  This callback must be set,
240          * if you want provide notification functionality.
241          */
242         void (*unregister_event)(struct mem_cgroup *memcg,
243                                  struct eventfd_ctx *eventfd);
244         /*
245          * All fields below needed to unregister event when
246          * userspace closes eventfd.
247          */
248         poll_table pt;
249         wait_queue_head_t *wqh;
250         wait_queue_t wait;
251         struct work_struct remove;
252 };
253 
254 static void mem_cgroup_threshold(struct mem_cgroup *memcg);
255 static void mem_cgroup_oom_notify(struct mem_cgroup *memcg);
256 
257 /*
258  * The memory controller data structure. The memory controller controls both
259  * page cache and RSS per cgroup. We would eventually like to provide
260  * statistics based on the statistics developed by Rik Van Riel for clock-pro,
261  * to help the administrator determine what knobs to tune.
262  */
263 struct mem_cgroup {
264         struct cgroup_subsys_state css;
265 
266         /* Accounted resources */
267         struct page_counter memory;
268         struct page_counter memsw;
269         struct page_counter kmem;
270 
271         /* Normal memory consumption range */
272         unsigned long low;
273         unsigned long high;
274 
275         unsigned long soft_limit;
276 
277         /* vmpressure notifications */
278         struct vmpressure vmpressure;
279 
280         /* css_online() has been completed */
281         int initialized;
282 
283         /*
284          * Should the accounting and control be hierarchical, per subtree?
285          */
286         bool use_hierarchy;
287 
288         bool            oom_lock;
289         atomic_t        under_oom;
290         atomic_t        oom_wakeups;
291 
292         int     swappiness;
293         /* OOM-Killer disable */
294         int             oom_kill_disable;
295 
296         /* protect arrays of thresholds */
297         struct mutex thresholds_lock;
298 
299         /* thresholds for memory usage. RCU-protected */
300         struct mem_cgroup_thresholds thresholds;
301 
302         /* thresholds for mem+swap usage. RCU-protected */
303         struct mem_cgroup_thresholds memsw_thresholds;
304 
305         /* For oom notifier event fd */
306         struct list_head oom_notify;
307 
308         /*
309          * Should we move charges of a task when a task is moved into this
310          * mem_cgroup ? And what type of charges should we move ?
311          */
312         unsigned long move_charge_at_immigrate;
313         /*
314          * set > 0 if pages under this cgroup are moving to other cgroup.
315          */
316         atomic_t                moving_account;
317         /* taken only while moving_account > 0 */
318         spinlock_t              move_lock;
319         struct task_struct      *move_lock_task;
320         unsigned long           move_lock_flags;
321         /*
322          * percpu counter.
323          */
324         struct mem_cgroup_stat_cpu __percpu *stat;
325         /*
326          * used when a cpu is offlined or other synchronizations
327          * See mem_cgroup_read_stat().
328          */
329         struct mem_cgroup_stat_cpu nocpu_base;
330         spinlock_t pcp_counter_lock;
331 
332 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
333         struct cg_proto tcp_mem;
334 #endif
335 #if defined(CONFIG_MEMCG_KMEM)
336         /* Index in the kmem_cache->memcg_params.memcg_caches array */
337         int kmemcg_id;
338         bool kmem_acct_activated;
339         bool kmem_acct_active;
340 #endif
341 
342         int last_scanned_node;
343 #if MAX_NUMNODES > 1
344         nodemask_t      scan_nodes;
345         atomic_t        numainfo_events;
346         atomic_t        numainfo_updating;
347 #endif
348 
349         /* List of events which userspace want to receive */
350         struct list_head event_list;
351         spinlock_t event_list_lock;
352 
353         struct mem_cgroup_per_node *nodeinfo[0];
354         /* WARNING: nodeinfo must be the last member here */
355 };
356 
357 #ifdef CONFIG_MEMCG_KMEM
358 bool memcg_kmem_is_active(struct mem_cgroup *memcg)
359 {
360         return memcg->kmem_acct_active;
361 }
362 #endif
363 
364 /* Stuffs for move charges at task migration. */
365 /*
366  * Types of charges to be moved.
367  */
368 #define MOVE_ANON       0x1U
369 #define MOVE_FILE       0x2U
370 #define MOVE_MASK       (MOVE_ANON | MOVE_FILE)
371 
372 /* "mc" and its members are protected by cgroup_mutex */
373 static struct move_charge_struct {
374         spinlock_t        lock; /* for from, to */
375         struct mem_cgroup *from;
376         struct mem_cgroup *to;
377         unsigned long flags;
378         unsigned long precharge;
379         unsigned long moved_charge;
380         unsigned long moved_swap;
381         struct task_struct *moving_task;        /* a task moving charges */
382         wait_queue_head_t waitq;                /* a waitq for other context */
383 } mc = {
384         .lock = __SPIN_LOCK_UNLOCKED(mc.lock),
385         .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq),
386 };
387 
388 /*
389  * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
390  * limit reclaim to prevent infinite loops, if they ever occur.
391  */
392 #define MEM_CGROUP_MAX_RECLAIM_LOOPS            100
393 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
394 
395 enum charge_type {
396         MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
397         MEM_CGROUP_CHARGE_TYPE_ANON,
398         MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
399         MEM_CGROUP_CHARGE_TYPE_DROP,    /* a page was unused swap cache */
400         NR_CHARGE_TYPE,
401 };
402 
403 /* for encoding cft->private value on file */
404 enum res_type {
405         _MEM,
406         _MEMSWAP,
407         _OOM_TYPE,
408         _KMEM,
409 };
410 
411 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
412 #define MEMFILE_TYPE(val)       ((val) >> 16 & 0xffff)
413 #define MEMFILE_ATTR(val)       ((val) & 0xffff)
414 /* Used for OOM nofiier */
415 #define OOM_CONTROL             (0)
416 
417 /*
418  * The memcg_create_mutex will be held whenever a new cgroup is created.
419  * As a consequence, any change that needs to protect against new child cgroups
420  * appearing has to hold it as well.
421  */
422 static DEFINE_MUTEX(memcg_create_mutex);
423 
424 struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
425 {
426         return s ? container_of(s, struct mem_cgroup, css) : NULL;
427 }
428 
429 /* Some nice accessors for the vmpressure. */
430 struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg)
431 {
432         if (!memcg)
433                 memcg = root_mem_cgroup;
434         return &memcg->vmpressure;
435 }
436 
437 struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr)
438 {
439         return &container_of(vmpr, struct mem_cgroup, vmpressure)->css;
440 }
441 
442 static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg)
443 {
444         return (memcg == root_mem_cgroup);
445 }
446 
447 /*
448  * We restrict the id in the range of [1, 65535], so it can fit into
449  * an unsigned short.
450  */
451 #define MEM_CGROUP_ID_MAX       USHRT_MAX
452 
453 static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg)
454 {
455         return memcg->css.id;
456 }
457 
458 /*
459  * A helper function to get mem_cgroup from ID. must be called under
460  * rcu_read_lock().  The caller is responsible for calling
461  * css_tryget_online() if the mem_cgroup is used for charging. (dropping
462  * refcnt from swap can be called against removed memcg.)
463  */
464 static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
465 {
466         struct cgroup_subsys_state *css;
467 
468         css = css_from_id(id, &memory_cgrp_subsys);
469         return mem_cgroup_from_css(css);
470 }
471 
472 /* Writing them here to avoid exposing memcg's inner layout */
473 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
474 
475 void sock_update_memcg(struct sock *sk)
476 {
477         if (mem_cgroup_sockets_enabled) {
478                 struct mem_cgroup *memcg;
479                 struct cg_proto *cg_proto;
480 
481                 BUG_ON(!sk->sk_prot->proto_cgroup);
482 
483                 /* Socket cloning can throw us here with sk_cgrp already
484                  * filled. It won't however, necessarily happen from
485                  * process context. So the test for root memcg given
486                  * the current task's memcg won't help us in this case.
487                  *
488                  * Respecting the original socket's memcg is a better
489                  * decision in this case.
490                  */
491                 if (sk->sk_cgrp) {
492                         BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg));
493                         css_get(&sk->sk_cgrp->memcg->css);
494                         return;
495                 }
496 
497                 rcu_read_lock();
498                 memcg = mem_cgroup_from_task(current);
499                 cg_proto = sk->sk_prot->proto_cgroup(memcg);
500                 if (!mem_cgroup_is_root(memcg) &&
501                     memcg_proto_active(cg_proto) &&
502                     css_tryget_online(&memcg->css)) {
503                         sk->sk_cgrp = cg_proto;
504                 }
505                 rcu_read_unlock();
506         }
507 }
508 EXPORT_SYMBOL(sock_update_memcg);
509 
510 void sock_release_memcg(struct sock *sk)
511 {
512         if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
513                 struct mem_cgroup *memcg;
514                 WARN_ON(!sk->sk_cgrp->memcg);
515                 memcg = sk->sk_cgrp->memcg;
516                 css_put(&sk->sk_cgrp->memcg->css);
517         }
518 }
519 
520 struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg)
521 {
522         if (!memcg || mem_cgroup_is_root(memcg))
523                 return NULL;
524 
525         return &memcg->tcp_mem;
526 }
527 EXPORT_SYMBOL(tcp_proto_cgroup);
528 
529 #endif
530 
531 #ifdef CONFIG_MEMCG_KMEM
532 /*
533  * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
534  * The main reason for not using cgroup id for this:
535  *  this works better in sparse environments, where we have a lot of memcgs,
536  *  but only a few kmem-limited. Or also, if we have, for instance, 200
537  *  memcgs, and none but the 200th is kmem-limited, we'd have to have a
538  *  200 entry array for that.
539  *
540  * The current size of the caches array is stored in memcg_nr_cache_ids. It
541  * will double each time we have to increase it.
542  */
543 static DEFINE_IDA(memcg_cache_ida);
544 int memcg_nr_cache_ids;
545 
546 /* Protects memcg_nr_cache_ids */
547 static DECLARE_RWSEM(memcg_cache_ids_sem);
548 
549 void memcg_get_cache_ids(void)
550 {
551         down_read(&memcg_cache_ids_sem);
552 }
553 
554 void memcg_put_cache_ids(void)
555 {
556         up_read(&memcg_cache_ids_sem);
557 }
558 
559 /*
560  * MIN_SIZE is different than 1, because we would like to avoid going through
561  * the alloc/free process all the time. In a small machine, 4 kmem-limited
562  * cgroups is a reasonable guess. In the future, it could be a parameter or
563  * tunable, but that is strictly not necessary.
564  *
565  * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
566  * this constant directly from cgroup, but it is understandable that this is
567  * better kept as an internal representation in cgroup.c. In any case, the
568  * cgrp_id space is not getting any smaller, and we don't have to necessarily
569  * increase ours as well if it increases.
570  */
571 #define MEMCG_CACHES_MIN_SIZE 4
572 #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
573 
574 /*
575  * A lot of the calls to the cache allocation functions are expected to be
576  * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
577  * conditional to this static branch, we'll have to allow modules that does
578  * kmem_cache_alloc and the such to see this symbol as well
579  */
580 struct static_key memcg_kmem_enabled_key;
581 EXPORT_SYMBOL(memcg_kmem_enabled_key);
582 
583 #endif /* CONFIG_MEMCG_KMEM */
584 
585 static struct mem_cgroup_per_zone *
586 mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone)
587 {
588         int nid = zone_to_nid(zone);
589         int zid = zone_idx(zone);
590 
591         return &memcg->nodeinfo[nid]->zoneinfo[zid];
592 }
593 
594 struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg)
595 {
596         return &memcg->css;
597 }
598 
599 static struct mem_cgroup_per_zone *
600 mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page)
601 {
602         int nid = page_to_nid(page);
603         int zid = page_zonenum(page);
604 
605         return &memcg->nodeinfo[nid]->zoneinfo[zid];
606 }
607 
608 static struct mem_cgroup_tree_per_zone *
609 soft_limit_tree_node_zone(int nid, int zid)
610 {
611         return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
612 }
613 
614 static struct mem_cgroup_tree_per_zone *
615 soft_limit_tree_from_page(struct page *page)
616 {
617         int nid = page_to_nid(page);
618         int zid = page_zonenum(page);
619 
620         return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid];
621 }
622 
623 static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz,
624                                          struct mem_cgroup_tree_per_zone *mctz,
625                                          unsigned long new_usage_in_excess)
626 {
627         struct rb_node **p = &mctz->rb_root.rb_node;
628         struct rb_node *parent = NULL;
629         struct mem_cgroup_per_zone *mz_node;
630 
631         if (mz->on_tree)
632                 return;
633 
634         mz->usage_in_excess = new_usage_in_excess;
635         if (!mz->usage_in_excess)
636                 return;
637         while (*p) {
638                 parent = *p;
639                 mz_node = rb_entry(parent, struct mem_cgroup_per_zone,
640                                         tree_node);
641                 if (mz->usage_in_excess < mz_node->usage_in_excess)
642                         p = &(*p)->rb_left;
643                 /*
644                  * We can't avoid mem cgroups that are over their soft
645                  * limit by the same amount
646                  */
647                 else if (mz->usage_in_excess >= mz_node->usage_in_excess)
648                         p = &(*p)->rb_right;
649         }
650         rb_link_node(&mz->tree_node, parent, p);
651         rb_insert_color(&mz->tree_node, &mctz->rb_root);
652         mz->on_tree = true;
653 }
654 
655 static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
656                                          struct mem_cgroup_tree_per_zone *mctz)
657 {
658         if (!mz->on_tree)
659                 return;
660         rb_erase(&mz->tree_node, &mctz->rb_root);
661         mz->on_tree = false;
662 }
663 
664 static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz,
665                                        struct mem_cgroup_tree_per_zone *mctz)
666 {
667         unsigned long flags;
668 
669         spin_lock_irqsave(&mctz->lock, flags);
670         __mem_cgroup_remove_exceeded(mz, mctz);
671         spin_unlock_irqrestore(&mctz->lock, flags);
672 }
673 
674 static unsigned long soft_limit_excess(struct mem_cgroup *memcg)
675 {
676         unsigned long nr_pages = page_counter_read(&memcg->memory);
677         unsigned long soft_limit = READ_ONCE(memcg->soft_limit);
678         unsigned long excess = 0;
679 
680         if (nr_pages > soft_limit)
681                 excess = nr_pages - soft_limit;
682 
683         return excess;
684 }
685 
686 static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page)
687 {
688         unsigned long excess;
689         struct mem_cgroup_per_zone *mz;
690         struct mem_cgroup_tree_per_zone *mctz;
691 
692         mctz = soft_limit_tree_from_page(page);
693         /*
694          * Necessary to update all ancestors when hierarchy is used.
695          * because their event counter is not touched.
696          */
697         for (; memcg; memcg = parent_mem_cgroup(memcg)) {
698                 mz = mem_cgroup_page_zoneinfo(memcg, page);
699                 excess = soft_limit_excess(memcg);
700                 /*
701                  * We have to update the tree if mz is on RB-tree or
702                  * mem is over its softlimit.
703                  */
704                 if (excess || mz->on_tree) {
705                         unsigned long flags;
706 
707                         spin_lock_irqsave(&mctz->lock, flags);
708                         /* if on-tree, remove it */
709                         if (mz->on_tree)
710                                 __mem_cgroup_remove_exceeded(mz, mctz);
711                         /*
712                          * Insert again. mz->usage_in_excess will be updated.
713                          * If excess is 0, no tree ops.
714                          */
715                         __mem_cgroup_insert_exceeded(mz, mctz, excess);
716                         spin_unlock_irqrestore(&mctz->lock, flags);
717                 }
718         }
719 }
720 
721 static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg)
722 {
723         struct mem_cgroup_tree_per_zone *mctz;
724         struct mem_cgroup_per_zone *mz;
725         int nid, zid;
726 
727         for_each_node(nid) {
728                 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
729                         mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
730                         mctz = soft_limit_tree_node_zone(nid, zid);
731                         mem_cgroup_remove_exceeded(mz, mctz);
732                 }
733         }
734 }
735 
736 static struct mem_cgroup_per_zone *
737 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
738 {
739         struct rb_node *rightmost = NULL;
740         struct mem_cgroup_per_zone *mz;
741 
742 retry:
743         mz = NULL;
744         rightmost = rb_last(&mctz->rb_root);
745         if (!rightmost)
746                 goto done;              /* Nothing to reclaim from */
747 
748         mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node);
749         /*
750          * Remove the node now but someone else can add it back,
751          * we will to add it back at the end of reclaim to its correct
752          * position in the tree.
753          */
754         __mem_cgroup_remove_exceeded(mz, mctz);
755         if (!soft_limit_excess(mz->memcg) ||
756             !css_tryget_online(&mz->memcg->css))
757                 goto retry;
758 done:
759         return mz;
760 }
761 
762 static struct mem_cgroup_per_zone *
763 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
764 {
765         struct mem_cgroup_per_zone *mz;
766 
767         spin_lock_irq(&mctz->lock);
768         mz = __mem_cgroup_largest_soft_limit_node(mctz);
769         spin_unlock_irq(&mctz->lock);
770         return mz;
771 }
772 
773 /*
774  * Implementation Note: reading percpu statistics for memcg.
775  *
776  * Both of vmstat[] and percpu_counter has threshold and do periodic
777  * synchronization to implement "quick" read. There are trade-off between
778  * reading cost and precision of value. Then, we may have a chance to implement
779  * a periodic synchronizion of counter in memcg's counter.
780  *
781  * But this _read() function is used for user interface now. The user accounts
782  * memory usage by memory cgroup and he _always_ requires exact value because
783  * he accounts memory. Even if we provide quick-and-fuzzy read, we always
784  * have to visit all online cpus and make sum. So, for now, unnecessary
785  * synchronization is not implemented. (just implemented for cpu hotplug)
786  *
787  * If there are kernel internal actions which can make use of some not-exact
788  * value, and reading all cpu value can be performance bottleneck in some
789  * common workload, threashold and synchonization as vmstat[] should be
790  * implemented.
791  */
792 static long mem_cgroup_read_stat(struct mem_cgroup *memcg,
793                                  enum mem_cgroup_stat_index idx)
794 {
795         long val = 0;
796         int cpu;
797 
798         get_online_cpus();
799         for_each_online_cpu(cpu)
800                 val += per_cpu(memcg->stat->count[idx], cpu);
801 #ifdef CONFIG_HOTPLUG_CPU
802         spin_lock(&memcg->pcp_counter_lock);
803         val += memcg->nocpu_base.count[idx];
804         spin_unlock(&memcg->pcp_counter_lock);
805 #endif
806         put_online_cpus();
807         return val;
808 }
809 
810 static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg,
811                                             enum mem_cgroup_events_index idx)
812 {
813         unsigned long val = 0;
814         int cpu;
815 
816         get_online_cpus();
817         for_each_online_cpu(cpu)
818                 val += per_cpu(memcg->stat->events[idx], cpu);
819 #ifdef CONFIG_HOTPLUG_CPU
820         spin_lock(&memcg->pcp_counter_lock);
821         val += memcg->nocpu_base.events[idx];
822         spin_unlock(&memcg->pcp_counter_lock);
823 #endif
824         put_online_cpus();
825         return val;
826 }
827 
828 static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
829                                          struct page *page,
830                                          int nr_pages)
831 {
832         /*
833          * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
834          * counted as CACHE even if it's on ANON LRU.
835          */
836         if (PageAnon(page))
837                 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS],
838                                 nr_pages);
839         else
840                 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE],
841                                 nr_pages);
842 
843         if (PageTransHuge(page))
844                 __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
845                                 nr_pages);
846 
847         /* pagein of a big page is an event. So, ignore page size */
848         if (nr_pages > 0)
849                 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]);
850         else {
851                 __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]);
852                 nr_pages = -nr_pages; /* for event */
853         }
854 
855         __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
856 }
857 
858 unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru)
859 {
860         struct mem_cgroup_per_zone *mz;
861 
862         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
863         return mz->lru_size[lru];
864 }
865 
866 static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg,
867                                                   int nid,
868                                                   unsigned int lru_mask)
869 {
870         unsigned long nr = 0;
871         int zid;
872 
873         VM_BUG_ON((unsigned)nid >= nr_node_ids);
874 
875         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
876                 struct mem_cgroup_per_zone *mz;
877                 enum lru_list lru;
878 
879                 for_each_lru(lru) {
880                         if (!(BIT(lru) & lru_mask))
881                                 continue;
882                         mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
883                         nr += mz->lru_size[lru];
884                 }
885         }
886         return nr;
887 }
888 
889 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg,
890                         unsigned int lru_mask)
891 {
892         unsigned long nr = 0;
893         int nid;
894 
895         for_each_node_state(nid, N_MEMORY)
896                 nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask);
897         return nr;
898 }
899 
900 static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg,
901                                        enum mem_cgroup_events_target target)
902 {
903         unsigned long val, next;
904 
905         val = __this_cpu_read(memcg->stat->nr_page_events);
906         next = __this_cpu_read(memcg->stat->targets[target]);
907         /* from time_after() in jiffies.h */
908         if ((long)next - (long)val < 0) {
909                 switch (target) {
910                 case MEM_CGROUP_TARGET_THRESH:
911                         next = val + THRESHOLDS_EVENTS_TARGET;
912                         break;
913                 case MEM_CGROUP_TARGET_SOFTLIMIT:
914                         next = val + SOFTLIMIT_EVENTS_TARGET;
915                         break;
916                 case MEM_CGROUP_TARGET_NUMAINFO:
917                         next = val + NUMAINFO_EVENTS_TARGET;
918                         break;
919                 default:
920                         break;
921                 }
922                 __this_cpu_write(memcg->stat->targets[target], next);
923                 return true;
924         }
925         return false;
926 }
927 
928 /*
929  * Check events in order.
930  *
931  */
932 static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
933 {
934         /* threshold event is triggered in finer grain than soft limit */
935         if (unlikely(mem_cgroup_event_ratelimit(memcg,
936                                                 MEM_CGROUP_TARGET_THRESH))) {
937                 bool do_softlimit;
938                 bool do_numainfo __maybe_unused;
939 
940                 do_softlimit = mem_cgroup_event_ratelimit(memcg,
941                                                 MEM_CGROUP_TARGET_SOFTLIMIT);
942 #if MAX_NUMNODES > 1
943                 do_numainfo = mem_cgroup_event_ratelimit(memcg,
944                                                 MEM_CGROUP_TARGET_NUMAINFO);
945 #endif
946                 mem_cgroup_threshold(memcg);
947                 if (unlikely(do_softlimit))
948                         mem_cgroup_update_tree(memcg, page);
949 #if MAX_NUMNODES > 1
950                 if (unlikely(do_numainfo))
951                         atomic_inc(&memcg->numainfo_events);
952 #endif
953         }
954 }
955 
956 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
957 {
958         /*
959          * mm_update_next_owner() may clear mm->owner to NULL
960          * if it races with swapoff, page migration, etc.
961          * So this can be called with p == NULL.
962          */
963         if (unlikely(!p))
964                 return NULL;
965 
966         return mem_cgroup_from_css(task_css(p, memory_cgrp_id));
967 }
968 
969 static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm)
970 {
971         struct mem_cgroup *memcg = NULL;
972 
973         rcu_read_lock();
974         do {
975                 /*
976                  * Page cache insertions can happen withou an
977                  * actual mm context, e.g. during disk probing
978                  * on boot, loopback IO, acct() writes etc.
979                  */
980                 if (unlikely(!mm))
981                         memcg = root_mem_cgroup;
982                 else {
983                         memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
984                         if (unlikely(!memcg))
985                                 memcg = root_mem_cgroup;
986                 }
987         } while (!css_tryget_online(&memcg->css));
988         rcu_read_unlock();
989         return memcg;
990 }
991 
992 /**
993  * mem_cgroup_iter - iterate over memory cgroup hierarchy
994  * @root: hierarchy root
995  * @prev: previously returned memcg, NULL on first invocation
996  * @reclaim: cookie for shared reclaim walks, NULL for full walks
997  *
998  * Returns references to children of the hierarchy below @root, or
999  * @root itself, or %NULL after a full round-trip.
1000  *
1001  * Caller must pass the return value in @prev on subsequent
1002  * invocations for reference counting, or use mem_cgroup_iter_break()
1003  * to cancel a hierarchy walk before the round-trip is complete.
1004  *
1005  * Reclaimers can specify a zone and a priority level in @reclaim to
1006  * divide up the memcgs in the hierarchy among all concurrent
1007  * reclaimers operating on the same zone and priority.
1008  */
1009 struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
1010                                    struct mem_cgroup *prev,
1011                                    struct mem_cgroup_reclaim_cookie *reclaim)
1012 {
1013         struct reclaim_iter *uninitialized_var(iter);
1014         struct cgroup_subsys_state *css = NULL;
1015         struct mem_cgroup *memcg = NULL;
1016         struct mem_cgroup *pos = NULL;
1017 
1018         if (mem_cgroup_disabled())
1019                 return NULL;
1020 
1021         if (!root)
1022                 root = root_mem_cgroup;
1023 
1024         if (prev && !reclaim)
1025                 pos = prev;
1026 
1027         if (!root->use_hierarchy && root != root_mem_cgroup) {
1028                 if (prev)
1029                         goto out;
1030                 return root;
1031         }
1032 
1033         rcu_read_lock();
1034 
1035         if (reclaim) {
1036                 struct mem_cgroup_per_zone *mz;
1037 
1038                 mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone);
1039                 iter = &mz->iter[reclaim->priority];
1040 
1041                 if (prev && reclaim->generation != iter->generation)
1042                         goto out_unlock;
1043 
1044                 do {
1045                         pos = READ_ONCE(iter->position);
1046                         /*
1047                          * A racing update may change the position and
1048                          * put the last reference, hence css_tryget(),
1049                          * or retry to see the updated position.
1050                          */
1051                 } while (pos && !css_tryget(&pos->css));
1052         }
1053 
1054         if (pos)
1055                 css = &pos->css;
1056 
1057         for (;;) {
1058                 css = css_next_descendant_pre(css, &root->css);
1059                 if (!css) {
1060                         /*
1061                          * Reclaimers share the hierarchy walk, and a
1062                          * new one might jump in right at the end of
1063                          * the hierarchy - make sure they see at least
1064                          * one group and restart from the beginning.
1065                          */
1066                         if (!prev)
1067                                 continue;
1068                         break;
1069                 }
1070 
1071                 /*
1072                  * Verify the css and acquire a reference.  The root
1073                  * is provided by the caller, so we know it's alive
1074                  * and kicking, and don't take an extra reference.
1075                  */
1076                 memcg = mem_cgroup_from_css(css);
1077 
1078                 if (css == &root->css)
1079                         break;
1080 
1081                 if (css_tryget(css)) {
1082                         /*
1083                          * Make sure the memcg is initialized:
1084                          * mem_cgroup_css_online() orders the the
1085                          * initialization against setting the flag.
1086                          */
1087                         if (smp_load_acquire(&memcg->initialized))
1088                                 break;
1089 
1090                         css_put(css);
1091                 }
1092 
1093                 memcg = NULL;
1094         }
1095 
1096         if (reclaim) {
1097                 if (cmpxchg(&iter->position, pos, memcg) == pos) {
1098                         if (memcg)
1099                                 css_get(&memcg->css);
1100                         if (pos)
1101                                 css_put(&pos->css);
1102                 }
1103 
1104                 /*
1105                  * pairs with css_tryget when dereferencing iter->position
1106                  * above.
1107                  */
1108                 if (pos)
1109                         css_put(&pos->css);
1110 
1111                 if (!memcg)
1112                         iter->generation++;
1113                 else if (!prev)
1114                         reclaim->generation = iter->generation;
1115         }
1116 
1117 out_unlock:
1118         rcu_read_unlock();
1119 out:
1120         if (prev && prev != root)
1121                 css_put(&prev->css);
1122 
1123         return memcg;
1124 }
1125 
1126 /**
1127  * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1128  * @root: hierarchy root
1129  * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1130  */
1131 void mem_cgroup_iter_break(struct mem_cgroup *root,
1132                            struct mem_cgroup *prev)
1133 {
1134         if (!root)
1135                 root = root_mem_cgroup;
1136         if (prev && prev != root)
1137                 css_put(&prev->css);
1138 }
1139 
1140 /*
1141  * Iteration constructs for visiting all cgroups (under a tree).  If
1142  * loops are exited prematurely (break), mem_cgroup_iter_break() must
1143  * be used for reference counting.
1144  */
1145 #define for_each_mem_cgroup_tree(iter, root)            \
1146         for (iter = mem_cgroup_iter(root, NULL, NULL);  \
1147              iter != NULL;                              \
1148              iter = mem_cgroup_iter(root, iter, NULL))
1149 
1150 #define for_each_mem_cgroup(iter)                       \
1151         for (iter = mem_cgroup_iter(NULL, NULL, NULL);  \
1152              iter != NULL;                              \
1153              iter = mem_cgroup_iter(NULL, iter, NULL))
1154 
1155 void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx)
1156 {
1157         struct mem_cgroup *memcg;
1158 
1159         rcu_read_lock();
1160         memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
1161         if (unlikely(!memcg))
1162                 goto out;
1163 
1164         switch (idx) {
1165         case PGFAULT:
1166                 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]);
1167                 break;
1168         case PGMAJFAULT:
1169                 this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]);
1170                 break;
1171         default:
1172                 BUG();
1173         }
1174 out:
1175         rcu_read_unlock();
1176 }
1177 EXPORT_SYMBOL(__mem_cgroup_count_vm_event);
1178 
1179 /**
1180  * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1181  * @zone: zone of the wanted lruvec
1182  * @memcg: memcg of the wanted lruvec
1183  *
1184  * Returns the lru list vector holding pages for the given @zone and
1185  * @mem.  This can be the global zone lruvec, if the memory controller
1186  * is disabled.
1187  */
1188 struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone,
1189                                       struct mem_cgroup *memcg)
1190 {
1191         struct mem_cgroup_per_zone *mz;
1192         struct lruvec *lruvec;
1193 
1194         if (mem_cgroup_disabled()) {
1195                 lruvec = &zone->lruvec;
1196                 goto out;
1197         }
1198 
1199         mz = mem_cgroup_zone_zoneinfo(memcg, zone);
1200         lruvec = &mz->lruvec;
1201 out:
1202         /*
1203          * Since a node can be onlined after the mem_cgroup was created,
1204          * we have to be prepared to initialize lruvec->zone here;
1205          * and if offlined then reonlined, we need to reinitialize it.
1206          */
1207         if (unlikely(lruvec->zone != zone))
1208                 lruvec->zone = zone;
1209         return lruvec;
1210 }
1211 
1212 /**
1213  * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
1214  * @page: the page
1215  * @zone: zone of the page
1216  *
1217  * This function is only safe when following the LRU page isolation
1218  * and putback protocol: the LRU lock must be held, and the page must
1219  * either be PageLRU() or the caller must have isolated/allocated it.
1220  */
1221 struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone)
1222 {
1223         struct mem_cgroup_per_zone *mz;
1224         struct mem_cgroup *memcg;
1225         struct lruvec *lruvec;
1226 
1227         if (mem_cgroup_disabled()) {
1228                 lruvec = &zone->lruvec;
1229                 goto out;
1230         }
1231 
1232         memcg = page->mem_cgroup;
1233         /*
1234          * Swapcache readahead pages are added to the LRU - and
1235          * possibly migrated - before they are charged.
1236          */
1237         if (!memcg)
1238                 memcg = root_mem_cgroup;
1239 
1240         mz = mem_cgroup_page_zoneinfo(memcg, page);
1241         lruvec = &mz->lruvec;
1242 out:
1243         /*
1244          * Since a node can be onlined after the mem_cgroup was created,
1245          * we have to be prepared to initialize lruvec->zone here;
1246          * and if offlined then reonlined, we need to reinitialize it.
1247          */
1248         if (unlikely(lruvec->zone != zone))
1249                 lruvec->zone = zone;
1250         return lruvec;
1251 }
1252 
1253 /**
1254  * mem_cgroup_update_lru_size - account for adding or removing an lru page
1255  * @lruvec: mem_cgroup per zone lru vector
1256  * @lru: index of lru list the page is sitting on
1257  * @nr_pages: positive when adding or negative when removing
1258  *
1259  * This function must be called when a page is added to or removed from an
1260  * lru list.
1261  */
1262 void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru,
1263                                 int nr_pages)
1264 {
1265         struct mem_cgroup_per_zone *mz;
1266         unsigned long *lru_size;
1267 
1268         if (mem_cgroup_disabled())
1269                 return;
1270 
1271         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1272         lru_size = mz->lru_size + lru;
1273         *lru_size += nr_pages;
1274         VM_BUG_ON((long)(*lru_size) < 0);
1275 }
1276 
1277 bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root)
1278 {
1279         if (root == memcg)
1280                 return true;
1281         if (!root->use_hierarchy)
1282                 return false;
1283         return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup);
1284 }
1285 
1286 bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg)
1287 {
1288         struct mem_cgroup *task_memcg;
1289         struct task_struct *p;
1290         bool ret;
1291 
1292         p = find_lock_task_mm(task);
1293         if (p) {
1294                 task_memcg = get_mem_cgroup_from_mm(p->mm);
1295                 task_unlock(p);
1296         } else {
1297                 /*
1298                  * All threads may have already detached their mm's, but the oom
1299                  * killer still needs to detect if they have already been oom
1300                  * killed to prevent needlessly killing additional tasks.
1301                  */
1302                 rcu_read_lock();
1303                 task_memcg = mem_cgroup_from_task(task);
1304                 css_get(&task_memcg->css);
1305                 rcu_read_unlock();
1306         }
1307         ret = mem_cgroup_is_descendant(task_memcg, memcg);
1308         css_put(&task_memcg->css);
1309         return ret;
1310 }
1311 
1312 int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec)
1313 {
1314         unsigned long inactive_ratio;
1315         unsigned long inactive;
1316         unsigned long active;
1317         unsigned long gb;
1318 
1319         inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON);
1320         active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON);
1321 
1322         gb = (inactive + active) >> (30 - PAGE_SHIFT);
1323         if (gb)
1324                 inactive_ratio = int_sqrt(10 * gb);
1325         else
1326                 inactive_ratio = 1;
1327 
1328         return inactive * inactive_ratio < active;
1329 }
1330 
1331 bool mem_cgroup_lruvec_online(struct lruvec *lruvec)
1332 {
1333         struct mem_cgroup_per_zone *mz;
1334         struct mem_cgroup *memcg;
1335 
1336         if (mem_cgroup_disabled())
1337                 return true;
1338 
1339         mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec);
1340         memcg = mz->memcg;
1341 
1342         return !!(memcg->css.flags & CSS_ONLINE);
1343 }
1344 
1345 #define mem_cgroup_from_counter(counter, member)        \
1346         container_of(counter, struct mem_cgroup, member)
1347 
1348 /**
1349  * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1350  * @memcg: the memory cgroup
1351  *
1352  * Returns the maximum amount of memory @mem can be charged with, in
1353  * pages.
1354  */
1355 static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg)
1356 {
1357         unsigned long margin = 0;
1358         unsigned long count;
1359         unsigned long limit;
1360 
1361         count = page_counter_read(&memcg->memory);
1362         limit = READ_ONCE(memcg->memory.limit);
1363         if (count < limit)
1364                 margin = limit - count;
1365 
1366         if (do_swap_account) {
1367                 count = page_counter_read(&memcg->memsw);
1368                 limit = READ_ONCE(memcg->memsw.limit);
1369                 if (count <= limit)
1370                         margin = min(margin, limit - count);
1371         }
1372 
1373         return margin;
1374 }
1375 
1376 int mem_cgroup_swappiness(struct mem_cgroup *memcg)
1377 {
1378         /* root ? */
1379         if (mem_cgroup_disabled() || !memcg->css.parent)
1380                 return vm_swappiness;
1381 
1382         return memcg->swappiness;
1383 }
1384 
1385 /*
1386  * A routine for checking "mem" is under move_account() or not.
1387  *
1388  * Checking a cgroup is mc.from or mc.to or under hierarchy of
1389  * moving cgroups. This is for waiting at high-memory pressure
1390  * caused by "move".
1391  */
1392 static bool mem_cgroup_under_move(struct mem_cgroup *memcg)
1393 {
1394         struct mem_cgroup *from;
1395         struct mem_cgroup *to;
1396         bool ret = false;
1397         /*
1398          * Unlike task_move routines, we access mc.to, mc.from not under
1399          * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1400          */
1401         spin_lock(&mc.lock);
1402         from = mc.from;
1403         to = mc.to;
1404         if (!from)
1405                 goto unlock;
1406 
1407         ret = mem_cgroup_is_descendant(from, memcg) ||
1408                 mem_cgroup_is_descendant(to, memcg);
1409 unlock:
1410         spin_unlock(&mc.lock);
1411         return ret;
1412 }
1413 
1414 static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
1415 {
1416         if (mc.moving_task && current != mc.moving_task) {
1417                 if (mem_cgroup_under_move(memcg)) {
1418                         DEFINE_WAIT(wait);
1419                         prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE);
1420                         /* moving charge context might have finished. */
1421                         if (mc.moving_task)
1422                                 schedule();
1423                         finish_wait(&mc.waitq, &wait);
1424                         return true;
1425                 }
1426         }
1427         return false;
1428 }
1429 
1430 #define K(x) ((x) << (PAGE_SHIFT-10))
1431 /**
1432  * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
1433  * @memcg: The memory cgroup that went over limit
1434  * @p: Task that is going to be killed
1435  *
1436  * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1437  * enabled
1438  */
1439 void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p)
1440 {
1441         /* oom_info_lock ensures that parallel ooms do not interleave */
1442         static DEFINE_MUTEX(oom_info_lock);
1443         struct mem_cgroup *iter;
1444         unsigned int i;
1445 
1446         mutex_lock(&oom_info_lock);
1447         rcu_read_lock();
1448 
1449         if (p) {
1450                 pr_info("Task in ");
1451                 pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id));
1452                 pr_cont(" killed as a result of limit of ");
1453         } else {
1454                 pr_info("Memory limit reached of cgroup ");
1455         }
1456 
1457         pr_cont_cgroup_path(memcg->css.cgroup);
1458         pr_cont("\n");
1459 
1460         rcu_read_unlock();
1461 
1462         pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1463                 K((u64)page_counter_read(&memcg->memory)),
1464                 K((u64)memcg->memory.limit), memcg->memory.failcnt);
1465         pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1466                 K((u64)page_counter_read(&memcg->memsw)),
1467                 K((u64)memcg->memsw.limit), memcg->memsw.failcnt);
1468         pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1469                 K((u64)page_counter_read(&memcg->kmem)),
1470                 K((u64)memcg->kmem.limit), memcg->kmem.failcnt);
1471 
1472         for_each_mem_cgroup_tree(iter, memcg) {
1473                 pr_info("Memory cgroup stats for ");
1474                 pr_cont_cgroup_path(iter->css.cgroup);
1475                 pr_cont(":");
1476 
1477                 for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
1478                         if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
1479                                 continue;
1480                         pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i],
1481                                 K(mem_cgroup_read_stat(iter, i)));
1482                 }
1483 
1484                 for (i = 0; i < NR_LRU_LISTS; i++)
1485                         pr_cont(" %s:%luKB", mem_cgroup_lru_names[i],
1486                                 K(mem_cgroup_nr_lru_pages(iter, BIT(i))));
1487 
1488                 pr_cont("\n");
1489         }
1490         mutex_unlock(&oom_info_lock);
1491 }
1492 
1493 /*
1494  * This function returns the number of memcg under hierarchy tree. Returns
1495  * 1(self count) if no children.
1496  */
1497 static int mem_cgroup_count_children(struct mem_cgroup *memcg)
1498 {
1499         int num = 0;
1500         struct mem_cgroup *iter;
1501 
1502         for_each_mem_cgroup_tree(iter, memcg)
1503                 num++;
1504         return num;
1505 }
1506 
1507 /*
1508  * Return the memory (and swap, if configured) limit for a memcg.
1509  */
1510 static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg)
1511 {
1512         unsigned long limit;
1513 
1514         limit = memcg->memory.limit;
1515         if (mem_cgroup_swappiness(memcg)) {
1516                 unsigned long memsw_limit;
1517 
1518                 memsw_limit = memcg->memsw.limit;
1519                 limit = min(limit + total_swap_pages, memsw_limit);
1520         }
1521         return limit;
1522 }
1523 
1524 static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
1525                                      int order)
1526 {
1527         struct mem_cgroup *iter;
1528         unsigned long chosen_points = 0;
1529         unsigned long totalpages;
1530         unsigned int points = 0;
1531         struct task_struct *chosen = NULL;
1532 
1533         /*
1534          * If current has a pending SIGKILL or is exiting, then automatically
1535          * select it.  The goal is to allow it to allocate so that it may
1536          * quickly exit and free its memory.
1537          */
1538         if (fatal_signal_pending(current) || task_will_free_mem(current)) {
1539                 mark_tsk_oom_victim(current);
1540                 return;
1541         }
1542 
1543         check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL, memcg);
1544         totalpages = mem_cgroup_get_limit(memcg) ? : 1;
1545         for_each_mem_cgroup_tree(iter, memcg) {
1546                 struct css_task_iter it;
1547                 struct task_struct *task;
1548 
1549                 css_task_iter_start(&iter->css, &it);
1550                 while ((task = css_task_iter_next(&it))) {
1551                         switch (oom_scan_process_thread(task, totalpages, NULL,
1552                                                         false)) {
1553                         case OOM_SCAN_SELECT:
1554                                 if (chosen)
1555                                         put_task_struct(chosen);
1556                                 chosen = task;
1557                                 chosen_points = ULONG_MAX;
1558                                 get_task_struct(chosen);
1559                                 /* fall through */
1560                         case OOM_SCAN_CONTINUE:
1561                                 continue;
1562                         case OOM_SCAN_ABORT:
1563                                 css_task_iter_end(&it);
1564                                 mem_cgroup_iter_break(memcg, iter);
1565                                 if (chosen)
1566                                         put_task_struct(chosen);
1567                                 return;
1568                         case OOM_SCAN_OK:
1569                                 break;
1570                         };
1571                         points = oom_badness(task, memcg, NULL, totalpages);
1572                         if (!points || points < chosen_points)
1573                                 continue;
1574                         /* Prefer thread group leaders for display purposes */
1575                         if (points == chosen_points &&
1576                             thread_group_leader(chosen))
1577                                 continue;
1578 
1579                         if (chosen)
1580                                 put_task_struct(chosen);
1581                         chosen = task;
1582                         chosen_points = points;
1583                         get_task_struct(chosen);
1584                 }
1585                 css_task_iter_end(&it);
1586         }
1587 
1588         if (!chosen)
1589                 return;
1590         points = chosen_points * 1000 / totalpages;
1591         oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg,
1592                          NULL, "Memory cgroup out of memory");
1593 }
1594 
1595 #if MAX_NUMNODES > 1
1596 
1597 /**
1598  * test_mem_cgroup_node_reclaimable
1599  * @memcg: the target memcg
1600  * @nid: the node ID to be checked.
1601  * @noswap : specify true here if the user wants flle only information.
1602  *
1603  * This function returns whether the specified memcg contains any
1604  * reclaimable pages on a node. Returns true if there are any reclaimable
1605  * pages in the node.
1606  */
1607 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
1608                 int nid, bool noswap)
1609 {
1610         if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE))
1611                 return true;
1612         if (noswap || !total_swap_pages)
1613                 return false;
1614         if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON))
1615                 return true;
1616         return false;
1617 
1618 }
1619 
1620 /*
1621  * Always updating the nodemask is not very good - even if we have an empty
1622  * list or the wrong list here, we can start from some node and traverse all
1623  * nodes based on the zonelist. So update the list loosely once per 10 secs.
1624  *
1625  */
1626 static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg)
1627 {
1628         int nid;
1629         /*
1630          * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1631          * pagein/pageout changes since the last update.
1632          */
1633         if (!atomic_read(&memcg->numainfo_events))
1634                 return;
1635         if (atomic_inc_return(&memcg->numainfo_updating) > 1)
1636                 return;
1637 
1638         /* make a nodemask where this memcg uses memory from */
1639         memcg->scan_nodes = node_states[N_MEMORY];
1640 
1641         for_each_node_mask(nid, node_states[N_MEMORY]) {
1642 
1643                 if (!test_mem_cgroup_node_reclaimable(memcg, nid, false))
1644                         node_clear(nid, memcg->scan_nodes);
1645         }
1646 
1647         atomic_set(&memcg->numainfo_events, 0);
1648         atomic_set(&memcg->numainfo_updating, 0);
1649 }
1650 
1651 /*
1652  * Selecting a node where we start reclaim from. Because what we need is just
1653  * reducing usage counter, start from anywhere is O,K. Considering
1654  * memory reclaim from current node, there are pros. and cons.
1655  *
1656  * Freeing memory from current node means freeing memory from a node which
1657  * we'll use or we've used. So, it may make LRU bad. And if several threads
1658  * hit limits, it will see a contention on a node. But freeing from remote
1659  * node means more costs for memory reclaim because of memory latency.
1660  *
1661  * Now, we use round-robin. Better algorithm is welcomed.
1662  */
1663 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1664 {
1665         int node;
1666 
1667         mem_cgroup_may_update_nodemask(memcg);
1668         node = memcg->last_scanned_node;
1669 
1670         node = next_node(node, memcg->scan_nodes);
1671         if (node == MAX_NUMNODES)
1672                 node = first_node(memcg->scan_nodes);
1673         /*
1674          * We call this when we hit limit, not when pages are added to LRU.
1675          * No LRU may hold pages because all pages are UNEVICTABLE or
1676          * memcg is too small and all pages are not on LRU. In that case,
1677          * we use curret node.
1678          */
1679         if (unlikely(node == MAX_NUMNODES))
1680                 node = numa_node_id();
1681 
1682         memcg->last_scanned_node = node;
1683         return node;
1684 }
1685 #else
1686 int mem_cgroup_select_victim_node(struct mem_cgroup *memcg)
1687 {
1688         return 0;
1689 }
1690 #endif
1691 
1692 static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg,
1693                                    struct zone *zone,
1694                                    gfp_t gfp_mask,
1695                                    unsigned long *total_scanned)
1696 {
1697         struct mem_cgroup *victim = NULL;
1698         int total = 0;
1699         int loop = 0;
1700         unsigned long excess;
1701         unsigned long nr_scanned;
1702         struct mem_cgroup_reclaim_cookie reclaim = {
1703                 .zone = zone,
1704                 .priority = 0,
1705         };
1706 
1707         excess = soft_limit_excess(root_memcg);
1708 
1709         while (1) {
1710                 victim = mem_cgroup_iter(root_memcg, victim, &reclaim);
1711                 if (!victim) {
1712                         loop++;
1713                         if (loop >= 2) {
1714                                 /*
1715                                  * If we have not been able to reclaim
1716                                  * anything, it might because there are
1717                                  * no reclaimable pages under this hierarchy
1718                                  */
1719                                 if (!total)
1720                                         break;
1721                                 /*
1722                                  * We want to do more targeted reclaim.
1723                                  * excess >> 2 is not to excessive so as to
1724                                  * reclaim too much, nor too less that we keep
1725                                  * coming back to reclaim from this cgroup
1726                                  */
1727                                 if (total >= (excess >> 2) ||
1728                                         (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS))
1729                                         break;
1730                         }
1731                         continue;
1732                 }
1733                 total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false,
1734                                                      zone, &nr_scanned);
1735                 *total_scanned += nr_scanned;
1736                 if (!soft_limit_excess(root_memcg))
1737                         break;
1738         }
1739         mem_cgroup_iter_break(root_memcg, victim);
1740         return total;
1741 }
1742 
1743 #ifdef CONFIG_LOCKDEP
1744 static struct lockdep_map memcg_oom_lock_dep_map = {
1745         .name = "memcg_oom_lock",
1746 };
1747 #endif
1748 
1749 static DEFINE_SPINLOCK(memcg_oom_lock);
1750 
1751 /*
1752  * Check OOM-Killer is already running under our hierarchy.
1753  * If someone is running, return false.
1754  */
1755 static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg)
1756 {
1757         struct mem_cgroup *iter, *failed = NULL;
1758 
1759         spin_lock(&memcg_oom_lock);
1760 
1761         for_each_mem_cgroup_tree(iter, memcg) {
1762                 if (iter->oom_lock) {
1763                         /*
1764                          * this subtree of our hierarchy is already locked
1765                          * so we cannot give a lock.
1766                          */
1767                         failed = iter;
1768                         mem_cgroup_iter_break(memcg, iter);
1769                         break;
1770                 } else
1771                         iter->oom_lock = true;
1772         }
1773 
1774         if (failed) {
1775                 /*
1776                  * OK, we failed to lock the whole subtree so we have
1777                  * to clean up what we set up to the failing subtree
1778                  */
1779                 for_each_mem_cgroup_tree(iter, memcg) {
1780                         if (iter == failed) {
1781                                 mem_cgroup_iter_break(memcg, iter);
1782                                 break;
1783                         }
1784                         iter->oom_lock = false;
1785                 }
1786         } else
1787                 mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_);
1788 
1789         spin_unlock(&memcg_oom_lock);
1790 
1791         return !failed;
1792 }
1793 
1794 static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg)
1795 {
1796         struct mem_cgroup *iter;
1797 
1798         spin_lock(&memcg_oom_lock);
1799         mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_);
1800         for_each_mem_cgroup_tree(iter, memcg)
1801                 iter->oom_lock = false;
1802         spin_unlock(&memcg_oom_lock);
1803 }
1804 
1805 static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg)
1806 {
1807         struct mem_cgroup *iter;
1808 
1809         for_each_mem_cgroup_tree(iter, memcg)
1810                 atomic_inc(&iter->under_oom);
1811 }
1812 
1813 static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg)
1814 {
1815         struct mem_cgroup *iter;
1816 
1817         /*
1818          * When a new child is created while the hierarchy is under oom,
1819          * mem_cgroup_oom_lock() may not be called. We have to use
1820          * atomic_add_unless() here.
1821          */
1822         for_each_mem_cgroup_tree(iter, memcg)
1823                 atomic_add_unless(&iter->under_oom, -1, 0);
1824 }
1825 
1826 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq);
1827 
1828 struct oom_wait_info {
1829         struct mem_cgroup *memcg;
1830         wait_queue_t    wait;
1831 };
1832 
1833 static int memcg_oom_wake_function(wait_queue_t *wait,
1834         unsigned mode, int sync, void *arg)
1835 {
1836         struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg;
1837         struct mem_cgroup *oom_wait_memcg;
1838         struct oom_wait_info *oom_wait_info;
1839 
1840         oom_wait_info = container_of(wait, struct oom_wait_info, wait);
1841         oom_wait_memcg = oom_wait_info->memcg;
1842 
1843         if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) &&
1844             !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg))
1845                 return 0;
1846         return autoremove_wake_function(wait, mode, sync, arg);
1847 }
1848 
1849 static void memcg_wakeup_oom(struct mem_cgroup *memcg)
1850 {
1851         atomic_inc(&memcg->oom_wakeups);
1852         /* for filtering, pass "memcg" as argument. */
1853         __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg);
1854 }
1855 
1856 static void memcg_oom_recover(struct mem_cgroup *memcg)
1857 {
1858         if (memcg && atomic_read(&memcg->under_oom))
1859                 memcg_wakeup_oom(memcg);
1860 }
1861 
1862 static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order)
1863 {
1864         if (!current->memcg_oom.may_oom)
1865                 return;
1866         /*
1867          * We are in the middle of the charge context here, so we
1868          * don't want to block when potentially sitting on a callstack
1869          * that holds all kinds of filesystem and mm locks.
1870          *
1871          * Also, the caller may handle a failed allocation gracefully
1872          * (like optional page cache readahead) and so an OOM killer
1873          * invocation might not even be necessary.
1874          *
1875          * That's why we don't do anything here except remember the
1876          * OOM context and then deal with it at the end of the page
1877          * fault when the stack is unwound, the locks are released,
1878          * and when we know whether the fault was overall successful.
1879          */
1880         css_get(&memcg->css);
1881         current->memcg_oom.memcg = memcg;
1882         current->memcg_oom.gfp_mask = mask;
1883         current->memcg_oom.order = order;
1884 }
1885 
1886 /**
1887  * mem_cgroup_oom_synchronize - complete memcg OOM handling
1888  * @handle: actually kill/wait or just clean up the OOM state
1889  *
1890  * This has to be called at the end of a page fault if the memcg OOM
1891  * handler was enabled.
1892  *
1893  * Memcg supports userspace OOM handling where failed allocations must
1894  * sleep on a waitqueue until the userspace task resolves the
1895  * situation.  Sleeping directly in the charge context with all kinds
1896  * of locks held is not a good idea, instead we remember an OOM state
1897  * in the task and mem_cgroup_oom_synchronize() has to be called at
1898  * the end of the page fault to complete the OOM handling.
1899  *
1900  * Returns %true if an ongoing memcg OOM situation was detected and
1901  * completed, %false otherwise.
1902  */
1903 bool mem_cgroup_oom_synchronize(bool handle)
1904 {
1905         struct mem_cgroup *memcg = current->memcg_oom.memcg;
1906         struct oom_wait_info owait;
1907         bool locked;
1908 
1909         /* OOM is global, do not handle */
1910         if (!memcg)
1911                 return false;
1912 
1913         if (!handle || oom_killer_disabled)
1914                 goto cleanup;
1915 
1916         owait.memcg = memcg;
1917         owait.wait.flags = 0;
1918         owait.wait.func = memcg_oom_wake_function;
1919         owait.wait.private = current;
1920         INIT_LIST_HEAD(&owait.wait.task_list);
1921 
1922         prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE);
1923         mem_cgroup_mark_under_oom(memcg);
1924 
1925         locked = mem_cgroup_oom_trylock(memcg);
1926 
1927         if (locked)
1928                 mem_cgroup_oom_notify(memcg);
1929 
1930         if (locked && !memcg->oom_kill_disable) {
1931                 mem_cgroup_unmark_under_oom(memcg);
1932                 finish_wait(&memcg_oom_waitq, &owait.wait);
1933                 mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask,
1934                                          current->memcg_oom.order);
1935         } else {
1936                 schedule();
1937                 mem_cgroup_unmark_under_oom(memcg);
1938                 finish_wait(&memcg_oom_waitq, &owait.wait);
1939         }
1940 
1941         if (locked) {
1942                 mem_cgroup_oom_unlock(memcg);
1943                 /*
1944                  * There is no guarantee that an OOM-lock contender
1945                  * sees the wakeups triggered by the OOM kill
1946                  * uncharges.  Wake any sleepers explicitely.
1947                  */
1948                 memcg_oom_recover(memcg);
1949         }
1950 cleanup:
1951         current->memcg_oom.memcg = NULL;
1952         css_put(&memcg->css);
1953         return true;
1954 }
1955 
1956 /**
1957  * mem_cgroup_begin_page_stat - begin a page state statistics transaction
1958  * @page: page that is going to change accounted state
1959  *
1960  * This function must mark the beginning of an accounted page state
1961  * change to prevent double accounting when the page is concurrently
1962  * being moved to another memcg:
1963  *
1964  *   memcg = mem_cgroup_begin_page_stat(page);
1965  *   if (TestClearPageState(page))
1966  *     mem_cgroup_update_page_stat(memcg, state, -1);
1967  *   mem_cgroup_end_page_stat(memcg);
1968  */
1969 struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page)
1970 {
1971         struct mem_cgroup *memcg;
1972         unsigned long flags;
1973 
1974         /*
1975          * The RCU lock is held throughout the transaction.  The fast
1976          * path can get away without acquiring the memcg->move_lock
1977          * because page moving starts with an RCU grace period.
1978          *
1979          * The RCU lock also protects the memcg from being freed when
1980          * the page state that is going to change is the only thing
1981          * preventing the page from being uncharged.
1982          * E.g. end-writeback clearing PageWriteback(), which allows
1983          * migration to go ahead and uncharge the page before the
1984          * account transaction might be complete.
1985          */
1986         rcu_read_lock();
1987 
1988         if (mem_cgroup_disabled())
1989                 return NULL;
1990 again:
1991         memcg = page->mem_cgroup;
1992         if (unlikely(!memcg))
1993                 return NULL;
1994 
1995         if (atomic_read(&memcg->moving_account) <= 0)
1996                 return memcg;
1997 
1998         spin_lock_irqsave(&memcg->move_lock, flags);
1999         if (memcg != page->mem_cgroup) {
2000                 spin_unlock_irqrestore(&memcg->move_lock, flags);
2001                 goto again;
2002         }
2003 
2004         /*
2005          * When charge migration first begins, we can have locked and
2006          * unlocked page stat updates happening concurrently.  Track
2007          * the task who has the lock for mem_cgroup_end_page_stat().
2008          */
2009         memcg->move_lock_task = current;
2010         memcg->move_lock_flags = flags;
2011 
2012         return memcg;
2013 }
2014 
2015 /**
2016  * mem_cgroup_end_page_stat - finish a page state statistics transaction
2017  * @memcg: the memcg that was accounted against
2018  */
2019 void mem_cgroup_end_page_stat(struct mem_cgroup *memcg)
2020 {
2021         if (memcg && memcg->move_lock_task == current) {
2022                 unsigned long flags = memcg->move_lock_flags;
2023 
2024                 memcg->move_lock_task = NULL;
2025                 memcg->move_lock_flags = 0;
2026 
2027                 spin_unlock_irqrestore(&memcg->move_lock, flags);
2028         }
2029 
2030         rcu_read_unlock();
2031 }
2032 
2033 /**
2034  * mem_cgroup_update_page_stat - update page state statistics
2035  * @memcg: memcg to account against
2036  * @idx: page state item to account
2037  * @val: number of pages (positive or negative)
2038  *
2039  * See mem_cgroup_begin_page_stat() for locking requirements.
2040  */
2041 void mem_cgroup_update_page_stat(struct mem_cgroup *memcg,
2042                                  enum mem_cgroup_stat_index idx, int val)
2043 {
2044         VM_BUG_ON(!rcu_read_lock_held());
2045 
2046         if (memcg)
2047                 this_cpu_add(memcg->stat->count[idx], val);
2048 }
2049 
2050 /*
2051  * size of first charge trial. "32" comes from vmscan.c's magic value.
2052  * TODO: maybe necessary to use big numbers in big irons.
2053  */
2054 #define CHARGE_BATCH    32U
2055 struct memcg_stock_pcp {
2056         struct mem_cgroup *cached; /* this never be root cgroup */
2057         unsigned int nr_pages;
2058         struct work_struct work;
2059         unsigned long flags;
2060 #define FLUSHING_CACHED_CHARGE  0
2061 };
2062 static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock);
2063 static DEFINE_MUTEX(percpu_charge_mutex);
2064 
2065 /**
2066  * consume_stock: Try to consume stocked charge on this cpu.
2067  * @memcg: memcg to consume from.
2068  * @nr_pages: how many pages to charge.
2069  *
2070  * The charges will only happen if @memcg matches the current cpu's memcg
2071  * stock, and at least @nr_pages are available in that stock.  Failure to
2072  * service an allocation will refill the stock.
2073  *
2074  * returns true if successful, false otherwise.
2075  */
2076 static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2077 {
2078         struct memcg_stock_pcp *stock;
2079         bool ret = false;
2080 
2081         if (nr_pages > CHARGE_BATCH)
2082                 return ret;
2083 
2084         stock = &get_cpu_var(memcg_stock);
2085         if (memcg == stock->cached && stock->nr_pages >= nr_pages) {
2086                 stock->nr_pages -= nr_pages;
2087                 ret = true;
2088         }
2089         put_cpu_var(memcg_stock);
2090         return ret;
2091 }
2092 
2093 /*
2094  * Returns stocks cached in percpu and reset cached information.
2095  */
2096 static void drain_stock(struct memcg_stock_pcp *stock)
2097 {
2098         struct mem_cgroup *old = stock->cached;
2099 
2100         if (stock->nr_pages) {
2101                 page_counter_uncharge(&old->memory, stock->nr_pages);
2102                 if (do_swap_account)
2103                         page_counter_uncharge(&old->memsw, stock->nr_pages);
2104                 css_put_many(&old->css, stock->nr_pages);
2105                 stock->nr_pages = 0;
2106         }
2107         stock->cached = NULL;
2108 }
2109 
2110 /*
2111  * This must be called under preempt disabled or must be called by
2112  * a thread which is pinned to local cpu.
2113  */
2114 static void drain_local_stock(struct work_struct *dummy)
2115 {
2116         struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
2117         drain_stock(stock);
2118         clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
2119 }
2120 
2121 /*
2122  * Cache charges(val) to local per_cpu area.
2123  * This will be consumed by consume_stock() function, later.
2124  */
2125 static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
2126 {
2127         struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
2128 
2129         if (stock->cached != memcg) { /* reset if necessary */
2130                 drain_stock(stock);
2131                 stock->cached = memcg;
2132         }
2133         stock->nr_pages += nr_pages;
2134         put_cpu_var(memcg_stock);
2135 }
2136 
2137 /*
2138  * Drains all per-CPU charge caches for given root_memcg resp. subtree
2139  * of the hierarchy under it.
2140  */
2141 static void drain_all_stock(struct mem_cgroup *root_memcg)
2142 {
2143         int cpu, curcpu;
2144 
2145         /* If someone's already draining, avoid adding running more workers. */
2146         if (!mutex_trylock(&percpu_charge_mutex))
2147                 return;
2148         /* Notify other cpus that system-wide "drain" is running */
2149         get_online_cpus();
2150         curcpu = get_cpu();
2151         for_each_online_cpu(cpu) {
2152                 struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
2153                 struct mem_cgroup *memcg;
2154 
2155                 memcg = stock->cached;
2156                 if (!memcg || !stock->nr_pages)
2157                         continue;
2158                 if (!mem_cgroup_is_descendant(memcg, root_memcg))
2159                         continue;
2160                 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) {
2161                         if (cpu == curcpu)
2162                                 drain_local_stock(&stock->work);
2163                         else
2164                                 schedule_work_on(cpu, &stock->work);
2165                 }
2166         }
2167         put_cpu();
2168         put_online_cpus();
2169         mutex_unlock(&percpu_charge_mutex);
2170 }
2171 
2172 /*
2173  * This function drains percpu counter value from DEAD cpu and
2174  * move it to local cpu. Note that this function can be preempted.
2175  */
2176 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu)
2177 {
2178         int i;
2179 
2180         spin_lock(&memcg->pcp_counter_lock);
2181         for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
2182                 long x = per_cpu(memcg->stat->count[i], cpu);
2183 
2184                 per_cpu(memcg->stat->count[i], cpu) = 0;
2185                 memcg->nocpu_base.count[i] += x;
2186         }
2187         for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
2188                 unsigned long x = per_cpu(memcg->stat->events[i], cpu);
2189 
2190                 per_cpu(memcg->stat->events[i], cpu) = 0;
2191                 memcg->nocpu_base.events[i] += x;
2192         }
2193         spin_unlock(&memcg->pcp_counter_lock);
2194 }
2195 
2196 static int memcg_cpu_hotplug_callback(struct notifier_block *nb,
2197                                         unsigned long action,
2198                                         void *hcpu)
2199 {
2200         int cpu = (unsigned long)hcpu;
2201         struct memcg_stock_pcp *stock;
2202         struct mem_cgroup *iter;
2203 
2204         if (action == CPU_ONLINE)
2205                 return NOTIFY_OK;
2206 
2207         if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
2208                 return NOTIFY_OK;
2209 
2210         for_each_mem_cgroup(iter)
2211                 mem_cgroup_drain_pcp_counter(iter, cpu);
2212 
2213         stock = &per_cpu(memcg_stock, cpu);
2214         drain_stock(stock);
2215         return NOTIFY_OK;
2216 }
2217 
2218 static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
2219                       unsigned int nr_pages)
2220 {
2221         unsigned int batch = max(CHARGE_BATCH, nr_pages);
2222         int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
2223         struct mem_cgroup *mem_over_limit;
2224         struct page_counter *counter;
2225         unsigned long nr_reclaimed;
2226         bool may_swap = true;
2227         bool drained = false;
2228         int ret = 0;
2229 
2230         if (mem_cgroup_is_root(memcg))
2231                 goto done;
2232 retry:
2233         if (consume_stock(memcg, nr_pages))
2234                 goto done;
2235 
2236         if (!do_swap_account ||
2237             !page_counter_try_charge(&memcg->memsw, batch, &counter)) {
2238                 if (!page_counter_try_charge(&memcg->memory, batch, &counter))
2239                         goto done_restock;
2240                 if (do_swap_account)
2241                         page_counter_uncharge(&memcg->memsw, batch);
2242                 mem_over_limit = mem_cgroup_from_counter(counter, memory);
2243         } else {
2244                 mem_over_limit = mem_cgroup_from_counter(counter, memsw);
2245                 may_swap = false;
2246         }
2247 
2248         if (batch > nr_pages) {
2249                 batch = nr_pages;
2250                 goto retry;
2251         }
2252 
2253         /*
2254          * Unlike in global OOM situations, memcg is not in a physical
2255          * memory shortage.  Allow dying and OOM-killed tasks to
2256          * bypass the last charges so that they can exit quickly and
2257          * free their memory.
2258          */
2259         if (unlikely(test_thread_flag(TIF_MEMDIE) ||
2260                      fatal_signal_pending(current) ||
2261                      current->flags & PF_EXITING))
2262                 goto bypass;
2263 
2264         if (unlikely(task_in_memcg_oom(current)))
2265                 goto nomem;
2266 
2267         if (!(gfp_mask & __GFP_WAIT))
2268                 goto nomem;
2269 
2270         mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1);
2271 
2272         nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages,
2273                                                     gfp_mask, may_swap);
2274 
2275         if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
2276                 goto retry;
2277 
2278         if (!drained) {
2279                 drain_all_stock(mem_over_limit);
2280                 drained = true;
2281                 goto retry;
2282         }
2283 
2284         if (gfp_mask & __GFP_NORETRY)
2285                 goto nomem;
2286         /*
2287          * Even though the limit is exceeded at this point, reclaim
2288          * may have been able to free some pages.  Retry the charge
2289          * before killing the task.
2290          *
2291          * Only for regular pages, though: huge pages are rather
2292          * unlikely to succeed so close to the limit, and we fall back
2293          * to regular pages anyway in case of failure.
2294          */
2295         if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER))
2296                 goto retry;
2297         /*
2298          * At task move, charge accounts can be doubly counted. So, it's
2299          * better to wait until the end of task_move if something is going on.
2300          */
2301         if (mem_cgroup_wait_acct_move(mem_over_limit))
2302                 goto retry;
2303 
2304         if (nr_retries--)
2305                 goto retry;
2306 
2307         if (gfp_mask & __GFP_NOFAIL)
2308                 goto bypass;
2309 
2310         if (fatal_signal_pending(current))
2311                 goto bypass;
2312 
2313         mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1);
2314 
2315         mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages));
2316 nomem:
2317         if (!(gfp_mask & __GFP_NOFAIL))
2318                 return -ENOMEM;
2319 bypass:
2320         return -EINTR;
2321 
2322 done_restock:
2323         css_get_many(&memcg->css, batch);
2324         if (batch > nr_pages)
2325                 refill_stock(memcg, batch - nr_pages);
2326         if (!(gfp_mask & __GFP_WAIT))
2327                 goto done;
2328         /*
2329          * If the hierarchy is above the normal consumption range,
2330          * make the charging task trim their excess contribution.
2331          */
2332         do {
2333                 if (page_counter_read(&memcg->memory) <= memcg->high)
2334                         continue;
2335                 mem_cgroup_events(memcg, MEMCG_HIGH, 1);
2336                 try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true);
2337         } while ((memcg = parent_mem_cgroup(memcg)));
2338 done:
2339         return ret;
2340 }
2341 
2342 static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages)
2343 {
2344         if (mem_cgroup_is_root(memcg))
2345                 return;
2346 
2347         page_counter_uncharge(&memcg->memory, nr_pages);
2348         if (do_swap_account)
2349                 page_counter_uncharge(&memcg->memsw, nr_pages);
2350 
2351         css_put_many(&memcg->css, nr_pages);
2352 }
2353 
2354 /*
2355  * try_get_mem_cgroup_from_page - look up page's memcg association
2356  * @page: the page
2357  *
2358  * Look up, get a css reference, and return the memcg that owns @page.
2359  *
2360  * The page must be locked to prevent racing with swap-in and page
2361  * cache charges.  If coming from an unlocked page table, the caller
2362  * must ensure the page is on the LRU or this can race with charging.
2363  */
2364 struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
2365 {
2366         struct mem_cgroup *memcg;
2367         unsigned short id;
2368         swp_entry_t ent;
2369 
2370         VM_BUG_ON_PAGE(!PageLocked(page), page);
2371 
2372         memcg = page->mem_cgroup;
2373         if (memcg) {
2374                 if (!css_tryget_online(&memcg->css))
2375                         memcg = NULL;
2376         } else if (PageSwapCache(page)) {
2377                 ent.val = page_private(page);
2378                 id = lookup_swap_cgroup_id(ent);
2379                 rcu_read_lock();
2380                 memcg = mem_cgroup_from_id(id);
2381                 if (memcg && !css_tryget_online(&memcg->css))
2382                         memcg = NULL;
2383                 rcu_read_unlock();
2384         }
2385         return memcg;
2386 }
2387 
2388 static void lock_page_lru(struct page *page, int *isolated)
2389 {
2390         struct zone *zone = page_zone(page);
2391 
2392         spin_lock_irq(&zone->lru_lock);
2393         if (PageLRU(page)) {
2394                 struct lruvec *lruvec;
2395 
2396                 lruvec = mem_cgroup_page_lruvec(page, zone);
2397                 ClearPageLRU(page);
2398                 del_page_from_lru_list(page, lruvec, page_lru(page));
2399                 *isolated = 1;
2400         } else
2401                 *isolated = 0;
2402 }
2403 
2404 static void unlock_page_lru(struct page *page, int isolated)
2405 {
2406         struct zone *zone = page_zone(page);
2407 
2408         if (isolated) {
2409                 struct lruvec *lruvec;
2410 
2411                 lruvec = mem_cgroup_page_lruvec(page, zone);
2412                 VM_BUG_ON_PAGE(PageLRU(page), page);
2413                 SetPageLRU(page);
2414                 add_page_to_lru_list(page, lruvec, page_lru(page));
2415         }
2416         spin_unlock_irq(&zone->lru_lock);
2417 }
2418 
2419 static void commit_charge(struct page *page, struct mem_cgroup *memcg,
2420                           bool lrucare)
2421 {
2422         int isolated;
2423 
2424         VM_BUG_ON_PAGE(page->mem_cgroup, page);
2425 
2426         /*
2427          * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2428          * may already be on some other mem_cgroup's LRU.  Take care of it.
2429          */
2430         if (lrucare)
2431                 lock_page_lru(page, &isolated);
2432 
2433         /*
2434          * Nobody should be changing or seriously looking at
2435          * page->mem_cgroup at this point:
2436          *
2437          * - the page is uncharged
2438          *
2439          * - the page is off-LRU
2440          *
2441          * - an anonymous fault has exclusive page access, except for
2442          *   a locked page table
2443          *
2444          * - a page cache insertion, a swapin fault, or a migration
2445          *   have the page locked
2446          */
2447         page->mem_cgroup = memcg;
2448 
2449         if (lrucare)
2450                 unlock_page_lru(page, isolated);
2451 }
2452 
2453 #ifdef CONFIG_MEMCG_KMEM
2454 int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp,
2455                       unsigned long nr_pages)
2456 {
2457         struct page_counter *counter;
2458         int ret = 0;
2459 
2460         ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter);
2461         if (ret < 0)
2462                 return ret;
2463 
2464         ret = try_charge(memcg, gfp, nr_pages);
2465         if (ret == -EINTR)  {
2466                 /*
2467                  * try_charge() chose to bypass to root due to OOM kill or
2468                  * fatal signal.  Since our only options are to either fail
2469                  * the allocation or charge it to this cgroup, do it as a
2470                  * temporary condition. But we can't fail. From a kmem/slab
2471                  * perspective, the cache has already been selected, by
2472                  * mem_cgroup_kmem_get_cache(), so it is too late to change
2473                  * our minds.
2474                  *
2475                  * This condition will only trigger if the task entered
2476                  * memcg_charge_kmem in a sane state, but was OOM-killed
2477                  * during try_charge() above. Tasks that were already dying
2478                  * when the allocation triggers should have been already
2479                  * directed to the root cgroup in memcontrol.h
2480                  */
2481                 page_counter_charge(&memcg->memory, nr_pages);
2482                 if (do_swap_account)
2483                         page_counter_charge(&memcg->memsw, nr_pages);
2484                 css_get_many(&memcg->css, nr_pages);
2485                 ret = 0;
2486         } else if (ret)
2487                 page_counter_uncharge(&memcg->kmem, nr_pages);
2488 
2489         return ret;
2490 }
2491 
2492 void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages)
2493 {
2494         page_counter_uncharge(&memcg->memory, nr_pages);
2495         if (do_swap_account)
2496                 page_counter_uncharge(&memcg->memsw, nr_pages);
2497 
2498         page_counter_uncharge(&memcg->kmem, nr_pages);
2499 
2500         css_put_many(&memcg->css, nr_pages);
2501 }
2502 
2503 /*
2504  * helper for acessing a memcg's index. It will be used as an index in the
2505  * child cache array in kmem_cache, and also to derive its name. This function
2506  * will return -1 when this is not a kmem-limited memcg.
2507  */
2508 int memcg_cache_id(struct mem_cgroup *memcg)
2509 {
2510         return memcg ? memcg->kmemcg_id : -1;
2511 }
2512 
2513 static int memcg_alloc_cache_id(void)
2514 {
2515         int id, size;
2516         int err;
2517 
2518         id = ida_simple_get(&memcg_cache_ida,
2519                             0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL);
2520         if (id < 0)
2521                 return id;
2522 
2523         if (id < memcg_nr_cache_ids)
2524                 return id;
2525 
2526         /*
2527          * There's no space for the new id in memcg_caches arrays,
2528          * so we have to grow them.
2529          */
2530         down_write(&memcg_cache_ids_sem);
2531 
2532         size = 2 * (id + 1);
2533         if (size < MEMCG_CACHES_MIN_SIZE)
2534                 size = MEMCG_CACHES_MIN_SIZE;
2535         else if (size > MEMCG_CACHES_MAX_SIZE)
2536                 size = MEMCG_CACHES_MAX_SIZE;
2537 
2538         err = memcg_update_all_caches(size);
2539         if (!err)
2540                 err = memcg_update_all_list_lrus(size);
2541         if (!err)
2542                 memcg_nr_cache_ids = size;
2543 
2544         up_write(&memcg_cache_ids_sem);
2545 
2546         if (err) {
2547                 ida_simple_remove(&memcg_cache_ida, id);
2548                 return err;
2549         }
2550         return id;
2551 }
2552 
2553 static void memcg_free_cache_id(int id)
2554 {
2555         ida_simple_remove(&memcg_cache_ida, id);
2556 }
2557 
2558 struct memcg_kmem_cache_create_work {
2559         struct mem_cgroup *memcg;
2560         struct kmem_cache *cachep;
2561         struct work_struct work;
2562 };
2563 
2564 static void memcg_kmem_cache_create_func(struct work_struct *w)
2565 {
2566         struct memcg_kmem_cache_create_work *cw =
2567                 container_of(w, struct memcg_kmem_cache_create_work, work);
2568         struct mem_cgroup *memcg = cw->memcg;
2569         struct kmem_cache *cachep = cw->cachep;
2570 
2571         memcg_create_kmem_cache(memcg, cachep);
2572 
2573         css_put(&memcg->css);
2574         kfree(cw);
2575 }
2576 
2577 /*
2578  * Enqueue the creation of a per-memcg kmem_cache.
2579  */
2580 static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2581                                                struct kmem_cache *cachep)
2582 {
2583         struct memcg_kmem_cache_create_work *cw;
2584 
2585         cw = kmalloc(sizeof(*cw), GFP_NOWAIT);
2586         if (!cw)
2587                 return;
2588 
2589         css_get(&memcg->css);
2590 
2591         cw->memcg = memcg;
2592         cw->cachep = cachep;
2593         INIT_WORK(&cw->work, memcg_kmem_cache_create_func);
2594 
2595         schedule_work(&cw->work);
2596 }
2597 
2598 static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg,
2599                                              struct kmem_cache *cachep)
2600 {
2601         /*
2602          * We need to stop accounting when we kmalloc, because if the
2603          * corresponding kmalloc cache is not yet created, the first allocation
2604          * in __memcg_schedule_kmem_cache_create will recurse.
2605          *
2606          * However, it is better to enclose the whole function. Depending on
2607          * the debugging options enabled, INIT_WORK(), for instance, can
2608          * trigger an allocation. This too, will make us recurse. Because at
2609          * this point we can't allow ourselves back into memcg_kmem_get_cache,
2610          * the safest choice is to do it like this, wrapping the whole function.
2611          */
2612         current->memcg_kmem_skip_account = 1;
2613         __memcg_schedule_kmem_cache_create(memcg, cachep);
2614         current->memcg_kmem_skip_account = 0;
2615 }
2616 
2617 /*
2618  * Return the kmem_cache we're supposed to use for a slab allocation.
2619  * We try to use the current memcg's version of the cache.
2620  *
2621  * If the cache does not exist yet, if we are the first user of it,
2622  * we either create it immediately, if possible, or create it asynchronously
2623  * in a workqueue.
2624  * In the latter case, we will let the current allocation go through with
2625  * the original cache.
2626  *
2627  * Can't be called in interrupt context or from kernel threads.
2628  * This function needs to be called with rcu_read_lock() held.
2629  */
2630 struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep)
2631 {
2632         struct mem_cgroup *memcg;
2633         struct kmem_cache *memcg_cachep;
2634         int kmemcg_id;
2635 
2636         VM_BUG_ON(!is_root_cache(cachep));
2637 
2638         if (current->memcg_kmem_skip_account)
2639                 return cachep;
2640 
2641         memcg = get_mem_cgroup_from_mm(current->mm);
2642         kmemcg_id = READ_ONCE(memcg->kmemcg_id);
2643         if (kmemcg_id < 0)
2644                 goto out;
2645 
2646         memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id);
2647         if (likely(memcg_cachep))
2648                 return memcg_cachep;
2649 
2650         /*
2651          * If we are in a safe context (can wait, and not in interrupt
2652          * context), we could be be predictable and return right away.
2653          * This would guarantee that the allocation being performed
2654          * already belongs in the new cache.
2655          *
2656          * However, there are some clashes that can arrive from locking.
2657          * For instance, because we acquire the slab_mutex while doing
2658          * memcg_create_kmem_cache, this means no further allocation
2659          * could happen with the slab_mutex held. So it's better to
2660          * defer everything.
2661          */
2662         memcg_schedule_kmem_cache_create(memcg, cachep);
2663 out:
2664         css_put(&memcg->css);
2665         return cachep;
2666 }
2667 
2668 void __memcg_kmem_put_cache(struct kmem_cache *cachep)
2669 {
2670         if (!is_root_cache(cachep))
2671                 css_put(&cachep->memcg_params.memcg->css);
2672 }
2673 
2674 /*
2675  * We need to verify if the allocation against current->mm->owner's memcg is
2676  * possible for the given order. But the page is not allocated yet, so we'll
2677  * need a further commit step to do the final arrangements.
2678  *
2679  * It is possible for the task to switch cgroups in this mean time, so at
2680  * commit time, we can't rely on task conversion any longer.  We'll then use
2681  * the handle argument to return to the caller which cgroup we should commit
2682  * against. We could also return the memcg directly and avoid the pointer
2683  * passing, but a boolean return value gives better semantics considering
2684  * the compiled-out case as well.
2685  *
2686  * Returning true means the allocation is possible.
2687  */
2688 bool
2689 __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order)
2690 {
2691         struct mem_cgroup *memcg;
2692         int ret;
2693 
2694         *_memcg = NULL;
2695 
2696         memcg = get_mem_cgroup_from_mm(current->mm);
2697 
2698         if (!memcg_kmem_is_active(memcg)) {
2699                 css_put(&memcg->css);
2700                 return true;
2701         }
2702 
2703         ret = memcg_charge_kmem(memcg, gfp, 1 << order);
2704         if (!ret)
2705                 *_memcg = memcg;
2706 
2707         css_put(&memcg->css);
2708         return (ret == 0);
2709 }
2710 
2711 void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
2712                               int order)
2713 {
2714         VM_BUG_ON(mem_cgroup_is_root(memcg));
2715 
2716         /* The page allocation failed. Revert */
2717         if (!page) {
2718                 memcg_uncharge_kmem(memcg, 1 << order);
2719                 return;
2720         }
2721         page->mem_cgroup = memcg;
2722 }
2723 
2724 void __memcg_kmem_uncharge_pages(struct page *page, int order)
2725 {
2726         struct mem_cgroup *memcg = page->mem_cgroup;
2727 
2728         if (!memcg)
2729                 return;
2730 
2731         VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page);
2732 
2733         memcg_uncharge_kmem(memcg, 1 << order);
2734         page->mem_cgroup = NULL;
2735 }
2736 
2737 struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr)
2738 {
2739         struct mem_cgroup *memcg = NULL;
2740         struct kmem_cache *cachep;
2741         struct page *page;
2742 
2743         page = virt_to_head_page(ptr);
2744         if (PageSlab(page)) {
2745                 cachep = page->slab_cache;
2746                 if (!is_root_cache(cachep))
2747                         memcg = cachep->memcg_params.memcg;
2748         } else
2749                 /* page allocated by alloc_kmem_pages */
2750                 memcg = page->mem_cgroup;
2751 
2752         return memcg;
2753 }
2754 #endif /* CONFIG_MEMCG_KMEM */
2755 
2756 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2757 
2758 /*
2759  * Because tail pages are not marked as "used", set it. We're under
2760  * zone->lru_lock, 'splitting on pmd' and compound_lock.
2761  * charge/uncharge will be never happen and move_account() is done under
2762  * compound_lock(), so we don't have to take care of races.
2763  */
2764 void mem_cgroup_split_huge_fixup(struct page *head)
2765 {
2766         int i;
2767 
2768         if (mem_cgroup_disabled())
2769                 return;
2770 
2771         for (i = 1; i < HPAGE_PMD_NR; i++)
2772                 head[i].mem_cgroup = head->mem_cgroup;
2773 
2774         __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE],
2775                        HPAGE_PMD_NR);
2776 }
2777 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2778 
2779 #ifdef CONFIG_MEMCG_SWAP
2780 static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
2781                                          bool charge)
2782 {
2783         int val = (charge) ? 1 : -1;
2784         this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val);
2785 }
2786 
2787 /**
2788  * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2789  * @entry: swap entry to be moved
2790  * @from:  mem_cgroup which the entry is moved from
2791  * @to:  mem_cgroup which the entry is moved to
2792  *
2793  * It succeeds only when the swap_cgroup's record for this entry is the same
2794  * as the mem_cgroup's id of @from.
2795  *
2796  * Returns 0 on success, -EINVAL on failure.
2797  *
2798  * The caller must have charged to @to, IOW, called page_counter_charge() about
2799  * both res and memsw, and called css_get().
2800  */
2801 static int mem_cgroup_move_swap_account(swp_entry_t entry,
2802                                 struct mem_cgroup *from, struct mem_cgroup *to)
2803 {
2804         unsigned short old_id, new_id;
2805 
2806         old_id = mem_cgroup_id(from);
2807         new_id = mem_cgroup_id(to);
2808 
2809         if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
2810                 mem_cgroup_swap_statistics(from, false);
2811                 mem_cgroup_swap_statistics(to, true);
2812                 return 0;
2813         }
2814         return -EINVAL;
2815 }
2816 #else
2817 static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
2818                                 struct mem_cgroup *from, struct mem_cgroup *to)
2819 {
2820         return -EINVAL;
2821 }
2822 #endif
2823 
2824 static DEFINE_MUTEX(memcg_limit_mutex);
2825 
2826 static int mem_cgroup_resize_limit(struct mem_cgroup *memcg,
2827                                    unsigned long limit)
2828 {
2829         unsigned long curusage;
2830         unsigned long oldusage;
2831         bool enlarge = false;
2832         int retry_count;
2833         int ret;
2834 
2835         /*
2836          * For keeping hierarchical_reclaim simple, how long we should retry
2837          * is depends on callers. We set our retry-count to be function
2838          * of # of children which we should visit in this loop.
2839          */
2840         retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2841                       mem_cgroup_count_children(memcg);
2842 
2843         oldusage = page_counter_read(&memcg->memory);
2844 
2845         do {
2846                 if (signal_pending(current)) {
2847                         ret = -EINTR;
2848                         break;
2849                 }
2850 
2851                 mutex_lock(&memcg_limit_mutex);
2852                 if (limit > memcg->memsw.limit) {
2853                         mutex_unlock(&memcg_limit_mutex);
2854                         ret = -EINVAL;
2855                         break;
2856                 }
2857                 if (limit > memcg->memory.limit)
2858                         enlarge = true;
2859                 ret = page_counter_limit(&memcg->memory, limit);
2860                 mutex_unlock(&memcg_limit_mutex);
2861 
2862                 if (!ret)
2863                         break;
2864 
2865                 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true);
2866 
2867                 curusage = page_counter_read(&memcg->memory);
2868                 /* Usage is reduced ? */
2869                 if (curusage >= oldusage)
2870                         retry_count--;
2871                 else
2872                         oldusage = curusage;
2873         } while (retry_count);
2874 
2875         if (!ret && enlarge)
2876                 memcg_oom_recover(memcg);
2877 
2878         return ret;
2879 }
2880 
2881 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg,
2882                                          unsigned long limit)
2883 {
2884         unsigned long curusage;
2885         unsigned long oldusage;
2886         bool enlarge = false;
2887         int retry_count;
2888         int ret;
2889 
2890         /* see mem_cgroup_resize_res_limit */
2891         retry_count = MEM_CGROUP_RECLAIM_RETRIES *
2892                       mem_cgroup_count_children(memcg);
2893 
2894         oldusage = page_counter_read(&memcg->memsw);
2895 
2896         do {
2897                 if (signal_pending(current)) {
2898                         ret = -EINTR;
2899                         break;
2900                 }
2901 
2902                 mutex_lock(&memcg_limit_mutex);
2903                 if (limit < memcg->memory.limit) {
2904                         mutex_unlock(&memcg_limit_mutex);
2905                         ret = -EINVAL;
2906                         break;
2907                 }
2908                 if (limit > memcg->memsw.limit)
2909                         enlarge = true;
2910                 ret = page_counter_limit(&memcg->memsw, limit);
2911                 mutex_unlock(&memcg_limit_mutex);
2912 
2913                 if (!ret)
2914                         break;
2915 
2916                 try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false);
2917 
2918                 curusage = page_counter_read(&memcg->memsw);
2919                 /* Usage is reduced ? */
2920                 if (curusage >= oldusage)
2921                         retry_count--;
2922                 else
2923                         oldusage = curusage;
2924         } while (retry_count);
2925 
2926         if (!ret && enlarge)
2927                 memcg_oom_recover(memcg);
2928 
2929         return ret;
2930 }
2931 
2932 unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order,
2933                                             gfp_t gfp_mask,
2934                                             unsigned long *total_scanned)
2935 {
2936         unsigned long nr_reclaimed = 0;
2937         struct mem_cgroup_per_zone *mz, *next_mz = NULL;
2938         unsigned long reclaimed;
2939         int loop = 0;
2940         struct mem_cgroup_tree_per_zone *mctz;
2941         unsigned long excess;
2942         unsigned long nr_scanned;
2943 
2944         if (order > 0)
2945                 return 0;
2946 
2947         mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone));
2948         /*
2949          * This loop can run a while, specially if mem_cgroup's continuously
2950          * keep exceeding their soft limit and putting the system under
2951          * pressure
2952          */
2953         do {
2954                 if (next_mz)
2955                         mz = next_mz;
2956                 else
2957                         mz = mem_cgroup_largest_soft_limit_node(mctz);
2958                 if (!mz)
2959                         break;
2960 
2961                 nr_scanned = 0;
2962                 reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone,
2963                                                     gfp_mask, &nr_scanned);
2964                 nr_reclaimed += reclaimed;
2965                 *total_scanned += nr_scanned;
2966                 spin_lock_irq(&mctz->lock);
2967                 __mem_cgroup_remove_exceeded(mz, mctz);
2968 
2969                 /*
2970                  * If we failed to reclaim anything from this memory cgroup
2971                  * it is time to move on to the next cgroup
2972                  */
2973                 next_mz = NULL;
2974                 if (!reclaimed)
2975                         next_mz = __mem_cgroup_largest_soft_limit_node(mctz);
2976 
2977                 excess = soft_limit_excess(mz->memcg);
2978                 /*
2979                  * One school of thought says that we should not add
2980                  * back the node to the tree if reclaim returns 0.
2981                  * But our reclaim could return 0, simply because due
2982                  * to priority we are exposing a smaller subset of
2983                  * memory to reclaim from. Consider this as a longer
2984                  * term TODO.
2985                  */
2986                 /* If excess == 0, no tree ops */
2987                 __mem_cgroup_insert_exceeded(mz, mctz, excess);
2988                 spin_unlock_irq(&mctz->lock);
2989                 css_put(&mz->memcg->css);
2990                 loop++;
2991                 /*
2992                  * Could not reclaim anything and there are no more
2993                  * mem cgroups to try or we seem to be looping without
2994                  * reclaiming anything.
2995                  */
2996                 if (!nr_reclaimed &&
2997                         (next_mz == NULL ||
2998                         loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS))
2999                         break;
3000         } while (!nr_reclaimed);
3001         if (next_mz)
3002                 css_put(&next_mz->memcg->css);
3003         return nr_reclaimed;
3004 }
3005 
3006 /*
3007  * Test whether @memcg has children, dead or alive.  Note that this
3008  * function doesn't care whether @memcg has use_hierarchy enabled and
3009  * returns %true if there are child csses according to the cgroup
3010  * hierarchy.  Testing use_hierarchy is the caller's responsiblity.
3011  */
3012 static inline bool memcg_has_children(struct mem_cgroup *memcg)
3013 {
3014         bool ret;
3015 
3016         /*
3017          * The lock does not prevent addition or deletion of children, but
3018          * it prevents a new child from being initialized based on this
3019          * parent in css_online(), so it's enough to decide whether
3020          * hierarchically inherited attributes can still be changed or not.
3021          */
3022         lockdep_assert_held(&memcg_create_mutex);
3023 
3024         rcu_read_lock();
3025         ret = css_next_child(NULL, &memcg->css);
3026         rcu_read_unlock();
3027         return ret;
3028 }
3029 
3030 /*
3031  * Reclaims as many pages from the given memcg as possible and moves
3032  * the rest to the parent.
3033  *
3034  * Caller is responsible for holding css reference for memcg.
3035  */
3036 static int mem_cgroup_force_empty(struct mem_cgroup *memcg)
3037 {
3038         int nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
3039 
3040         /* we call try-to-free pages for make this cgroup empty */
3041         lru_add_drain_all();
3042         /* try to free all pages in this cgroup */
3043         while (nr_retries && page_counter_read(&memcg->memory)) {
3044                 int progress;
3045 
3046                 if (signal_pending(current))
3047                         return -EINTR;
3048 
3049                 progress = try_to_free_mem_cgroup_pages(memcg, 1,
3050                                                         GFP_KERNEL, true);
3051                 if (!progress) {
3052                         nr_retries--;
3053                         /* maybe some writeback is necessary */
3054                         congestion_wait(BLK_RW_ASYNC, HZ/10);
3055                 }
3056 
3057         }
3058 
3059         return 0;
3060 }
3061 
3062 static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of,
3063                                             char *buf, size_t nbytes,
3064                                             loff_t off)
3065 {
3066         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3067 
3068         if (mem_cgroup_is_root(memcg))
3069                 return -EINVAL;
3070         return mem_cgroup_force_empty(memcg) ?: nbytes;
3071 }
3072 
3073 static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css,
3074                                      struct cftype *cft)
3075 {
3076         return mem_cgroup_from_css(css)->use_hierarchy;
3077 }
3078 
3079 static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css,
3080                                       struct cftype *cft, u64 val)
3081 {
3082         int retval = 0;
3083         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3084         struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent);
3085 
3086         mutex_lock(&memcg_create_mutex);
3087 
3088         if (memcg->use_hierarchy == val)
3089                 goto out;
3090 
3091         /*
3092          * If parent's use_hierarchy is set, we can't make any modifications
3093          * in the child subtrees. If it is unset, then the change can
3094          * occur, provided the current cgroup has no children.
3095          *
3096          * For the root cgroup, parent_mem is NULL, we allow value to be
3097          * set if there are no children.
3098          */
3099         if ((!parent_memcg || !parent_memcg->use_hierarchy) &&
3100                                 (val == 1 || val == 0)) {
3101                 if (!memcg_has_children(memcg))
3102                         memcg->use_hierarchy = val;
3103                 else
3104                         retval = -EBUSY;
3105         } else
3106                 retval = -EINVAL;
3107 
3108 out:
3109         mutex_unlock(&memcg_create_mutex);
3110 
3111         return retval;
3112 }
3113 
3114 static unsigned long tree_stat(struct mem_cgroup *memcg,
3115                                enum mem_cgroup_stat_index idx)
3116 {
3117         struct mem_cgroup *iter;
3118         long val = 0;
3119 
3120         /* Per-cpu values can be negative, use a signed accumulator */
3121         for_each_mem_cgroup_tree(iter, memcg)
3122                 val += mem_cgroup_read_stat(iter, idx);
3123 
3124         if (val < 0) /* race ? */
3125                 val = 0;
3126         return val;
3127 }
3128 
3129 static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap)
3130 {
3131         u64 val;
3132 
3133         if (mem_cgroup_is_root(memcg)) {
3134                 val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE);
3135                 val += tree_stat(memcg, MEM_CGROUP_STAT_RSS);
3136                 if (swap)
3137                         val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP);
3138         } else {
3139                 if (!swap)
3140                         val = page_counter_read(&memcg->memory);
3141                 else
3142                         val = page_counter_read(&memcg->memsw);
3143         }
3144         return val << PAGE_SHIFT;
3145 }
3146 
3147 enum {
3148         RES_USAGE,
3149         RES_LIMIT,
3150         RES_MAX_USAGE,
3151         RES_FAILCNT,
3152         RES_SOFT_LIMIT,
3153 };
3154 
3155 static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css,
3156                                struct cftype *cft)
3157 {
3158         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3159         struct page_counter *counter;
3160 
3161         switch (MEMFILE_TYPE(cft->private)) {
3162         case _MEM:
3163                 counter = &memcg->memory;
3164                 break;
3165         case _MEMSWAP:
3166                 counter = &memcg->memsw;
3167                 break;
3168         case _KMEM:
3169                 counter = &memcg->kmem;
3170                 break;
3171         default:
3172                 BUG();
3173         }
3174 
3175         switch (MEMFILE_ATTR(cft->private)) {
3176         case RES_USAGE:
3177                 if (counter == &memcg->memory)
3178                         return mem_cgroup_usage(memcg, false);
3179                 if (counter == &memcg->memsw)
3180                         return mem_cgroup_usage(memcg, true);
3181                 return (u64)page_counter_read(counter) * PAGE_SIZE;
3182         case RES_LIMIT:
3183                 return (u64)counter->limit * PAGE_SIZE;
3184         case RES_MAX_USAGE:
3185                 return (u64)counter->watermark * PAGE_SIZE;
3186         case RES_FAILCNT:
3187                 return counter->failcnt;
3188         case RES_SOFT_LIMIT:
3189                 return (u64)memcg->soft_limit * PAGE_SIZE;
3190         default:
3191                 BUG();
3192         }
3193 }
3194 
3195 #ifdef CONFIG_MEMCG_KMEM
3196 static int memcg_activate_kmem(struct mem_cgroup *memcg,
3197                                unsigned long nr_pages)
3198 {
3199         int err = 0;
3200         int memcg_id;
3201 
3202         BUG_ON(memcg->kmemcg_id >= 0);
3203         BUG_ON(memcg->kmem_acct_activated);
3204         BUG_ON(memcg->kmem_acct_active);
3205 
3206         /*
3207          * For simplicity, we won't allow this to be disabled.  It also can't
3208          * be changed if the cgroup has children already, or if tasks had
3209          * already joined.
3210          *
3211          * If tasks join before we set the limit, a person looking at
3212          * kmem.usage_in_bytes will have no way to determine when it took
3213          * place, which makes the value quite meaningless.
3214          *
3215          * After it first became limited, changes in the value of the limit are
3216          * of course permitted.
3217          */
3218         mutex_lock(&memcg_create_mutex);
3219         if (cgroup_has_tasks(memcg->css.cgroup) ||
3220             (memcg->use_hierarchy && memcg_has_children(memcg)))
3221                 err = -EBUSY;
3222         mutex_unlock(&memcg_create_mutex);
3223         if (err)
3224                 goto out;
3225 
3226         memcg_id = memcg_alloc_cache_id();
3227         if (memcg_id < 0) {
3228                 err = memcg_id;
3229                 goto out;
3230         }
3231 
3232         /*
3233          * We couldn't have accounted to this cgroup, because it hasn't got
3234          * activated yet, so this should succeed.
3235          */
3236         err = page_counter_limit(&memcg->kmem, nr_pages);
3237         VM_BUG_ON(err);
3238 
3239         static_key_slow_inc(&memcg_kmem_enabled_key);
3240         /*
3241          * A memory cgroup is considered kmem-active as soon as it gets
3242          * kmemcg_id. Setting the id after enabling static branching will
3243          * guarantee no one starts accounting before all call sites are
3244          * patched.
3245          */
3246         memcg->kmemcg_id = memcg_id;
3247         memcg->kmem_acct_activated = true;
3248         memcg->kmem_acct_active = true;
3249 out:
3250         return err;
3251 }
3252 
3253 static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
3254                                    unsigned long limit)
3255 {
3256         int ret;
3257 
3258         mutex_lock(&memcg_limit_mutex);
3259         if (!memcg_kmem_is_active(memcg))
3260                 ret = memcg_activate_kmem(memcg, limit);
3261         else
3262                 ret = page_counter_limit(&memcg->kmem, limit);
3263         mutex_unlock(&memcg_limit_mutex);
3264         return ret;
3265 }
3266 
3267 static int memcg_propagate_kmem(struct mem_cgroup *memcg)
3268 {
3269         int ret = 0;
3270         struct mem_cgroup *parent = parent_mem_cgroup(memcg);
3271 
3272         if (!parent)
3273                 return 0;
3274 
3275         mutex_lock(&memcg_limit_mutex);
3276         /*
3277          * If the parent cgroup is not kmem-active now, it cannot be activated
3278          * after this point, because it has at least one child already.
3279          */
3280         if (memcg_kmem_is_active(parent))
3281                 ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX);
3282         mutex_unlock(&memcg_limit_mutex);
3283         return ret;
3284 }
3285 #else
3286 static int memcg_update_kmem_limit(struct mem_cgroup *memcg,
3287                                    unsigned long limit)
3288 {
3289         return -EINVAL;
3290 }
3291 #endif /* CONFIG_MEMCG_KMEM */
3292 
3293 /*
3294  * The user of this function is...
3295  * RES_LIMIT.
3296  */
3297 static ssize_t mem_cgroup_write(struct kernfs_open_file *of,
3298                                 char *buf, size_t nbytes, loff_t off)
3299 {
3300         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3301         unsigned long nr_pages;
3302         int ret;
3303 
3304         buf = strstrip(buf);
3305         ret = page_counter_memparse(buf, "-1", &nr_pages);
3306         if (ret)
3307                 return ret;
3308 
3309         switch (MEMFILE_ATTR(of_cft(of)->private)) {
3310         case RES_LIMIT:
3311                 if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */
3312                         ret = -EINVAL;
3313                         break;
3314                 }
3315                 switch (MEMFILE_TYPE(of_cft(of)->private)) {
3316                 case _MEM:
3317                         ret = mem_cgroup_resize_limit(memcg, nr_pages);
3318                         break;
3319                 case _MEMSWAP:
3320                         ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages);
3321                         break;
3322                 case _KMEM:
3323                         ret = memcg_update_kmem_limit(memcg, nr_pages);
3324                         break;
3325                 }
3326                 break;
3327         case RES_SOFT_LIMIT:
3328                 memcg->soft_limit = nr_pages;
3329                 ret = 0;
3330                 break;
3331         }
3332         return ret ?: nbytes;
3333 }
3334 
3335 static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf,
3336                                 size_t nbytes, loff_t off)
3337 {
3338         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
3339         struct page_counter *counter;
3340 
3341         switch (MEMFILE_TYPE(of_cft(of)->private)) {
3342         case _MEM:
3343                 counter = &memcg->memory;
3344                 break;
3345         case _MEMSWAP:
3346                 counter = &memcg->memsw;
3347                 break;
3348         case _KMEM:
3349                 counter = &memcg->kmem;
3350                 break;
3351         default:
3352                 BUG();
3353         }
3354 
3355         switch (MEMFILE_ATTR(of_cft(of)->private)) {
3356         case RES_MAX_USAGE:
3357                 page_counter_reset_watermark(counter);
3358                 break;
3359         case RES_FAILCNT:
3360                 counter->failcnt = 0;
3361                 break;
3362         default:
3363                 BUG();
3364         }
3365 
3366         return nbytes;
3367 }
3368 
3369 static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css,
3370                                         struct cftype *cft)
3371 {
3372         return mem_cgroup_from_css(css)->move_charge_at_immigrate;
3373 }
3374 
3375 #ifdef CONFIG_MMU
3376 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3377                                         struct cftype *cft, u64 val)
3378 {
3379         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3380 
3381         if (val & ~MOVE_MASK)
3382                 return -EINVAL;
3383 
3384         /*
3385          * No kind of locking is needed in here, because ->can_attach() will
3386          * check this value once in the beginning of the process, and then carry
3387          * on with stale data. This means that changes to this value will only
3388          * affect task migrations starting after the change.
3389          */
3390         memcg->move_charge_at_immigrate = val;
3391         return 0;
3392 }
3393 #else
3394 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
3395                                         struct cftype *cft, u64 val)
3396 {
3397         return -ENOSYS;
3398 }
3399 #endif
3400 
3401 #ifdef CONFIG_NUMA
3402 static int memcg_numa_stat_show(struct seq_file *m, void *v)
3403 {
3404         struct numa_stat {
3405                 const char *name;
3406                 unsigned int lru_mask;
3407         };
3408 
3409         static const struct numa_stat stats[] = {
3410                 { "total", LRU_ALL },
3411                 { "file", LRU_ALL_FILE },
3412                 { "anon", LRU_ALL_ANON },
3413                 { "unevictable", BIT(LRU_UNEVICTABLE) },
3414         };
3415         const struct numa_stat *stat;
3416         int nid;
3417         unsigned long nr;
3418         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3419 
3420         for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3421                 nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask);
3422                 seq_printf(m, "%s=%lu", stat->name, nr);
3423                 for_each_node_state(nid, N_MEMORY) {
3424                         nr = mem_cgroup_node_nr_lru_pages(memcg, nid,
3425                                                           stat->lru_mask);
3426                         seq_printf(m, " N%d=%lu", nid, nr);
3427                 }
3428                 seq_putc(m, '\n');
3429         }
3430 
3431         for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) {
3432                 struct mem_cgroup *iter;
3433 
3434                 nr = 0;
3435                 for_each_mem_cgroup_tree(iter, memcg)
3436                         nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask);
3437                 seq_printf(m, "hierarchical_%s=%lu", stat->name, nr);
3438                 for_each_node_state(nid, N_MEMORY) {
3439                         nr = 0;
3440                         for_each_mem_cgroup_tree(iter, memcg)
3441                                 nr += mem_cgroup_node_nr_lru_pages(
3442                                         iter, nid, stat->lru_mask);
3443                         seq_printf(m, " N%d=%lu", nid, nr);
3444                 }
3445                 seq_putc(m, '\n');
3446         }
3447 
3448         return 0;
3449 }
3450 #endif /* CONFIG_NUMA */
3451 
3452 static int memcg_stat_show(struct seq_file *m, void *v)
3453 {
3454         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
3455         unsigned long memory, memsw;
3456         struct mem_cgroup *mi;
3457         unsigned int i;
3458 
3459         BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) !=
3460                      MEM_CGROUP_STAT_NSTATS);
3461         BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) !=
3462                      MEM_CGROUP_EVENTS_NSTATS);
3463         BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS);
3464 
3465         for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3466                 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
3467                         continue;
3468                 seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i],
3469                            mem_cgroup_read_stat(memcg, i) * PAGE_SIZE);
3470         }
3471 
3472         for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++)
3473                 seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i],
3474                            mem_cgroup_read_events(memcg, i));
3475 
3476         for (i = 0; i < NR_LRU_LISTS; i++)
3477                 seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i],
3478                            mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE);
3479 
3480         /* Hierarchical information */
3481         memory = memsw = PAGE_COUNTER_MAX;
3482         for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) {
3483                 memory = min(memory, mi->memory.limit);
3484                 memsw = min(memsw, mi->memsw.limit);
3485         }
3486         seq_printf(m, "hierarchical_memory_limit %llu\n",
3487                    (u64)memory * PAGE_SIZE);
3488         if (do_swap_account)
3489                 seq_printf(m, "hierarchical_memsw_limit %llu\n",
3490                            (u64)memsw * PAGE_SIZE);
3491 
3492         for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) {
3493                 long long val = 0;
3494 
3495                 if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account)
3496                         continue;
3497                 for_each_mem_cgroup_tree(mi, memcg)
3498                         val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE;
3499                 seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val);
3500         }
3501 
3502         for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) {
3503                 unsigned long long val = 0;
3504 
3505                 for_each_mem_cgroup_tree(mi, memcg)
3506                         val += mem_cgroup_read_events(mi, i);
3507                 seq_printf(m, "total_%s %llu\n",
3508                            mem_cgroup_events_names[i], val);
3509         }
3510 
3511         for (i = 0; i < NR_LRU_LISTS; i++) {
3512                 unsigned long long val = 0;
3513 
3514                 for_each_mem_cgroup_tree(mi, memcg)
3515                         val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE;
3516                 seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val);
3517         }
3518 
3519 #ifdef CONFIG_DEBUG_VM
3520         {
3521                 int nid, zid;
3522                 struct mem_cgroup_per_zone *mz;
3523                 struct zone_reclaim_stat *rstat;
3524                 unsigned long recent_rotated[2] = {0, 0};
3525                 unsigned long recent_scanned[2] = {0, 0};
3526 
3527                 for_each_online_node(nid)
3528                         for (zid = 0; zid < MAX_NR_ZONES; zid++) {
3529                                 mz = &memcg->nodeinfo[nid]->zoneinfo[zid];
3530                                 rstat = &mz->lruvec.reclaim_stat;
3531 
3532                                 recent_rotated[0] += rstat->recent_rotated[0];
3533                                 recent_rotated[1] += rstat->recent_rotated[1];
3534                                 recent_scanned[0] += rstat->recent_scanned[0];
3535                                 recent_scanned[1] += rstat->recent_scanned[1];
3536                         }
3537                 seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]);
3538                 seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]);
3539                 seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]);
3540                 seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]);
3541         }
3542 #endif
3543 
3544         return 0;
3545 }
3546 
3547 static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css,
3548                                       struct cftype *cft)
3549 {
3550         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3551 
3552         return mem_cgroup_swappiness(memcg);
3553 }
3554 
3555 static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css,
3556                                        struct cftype *cft, u64 val)
3557 {
3558         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3559 
3560         if (val > 100)
3561                 return -EINVAL;
3562 
3563         if (css->parent)
3564                 memcg->swappiness = val;
3565         else
3566                 vm_swappiness = val;
3567 
3568         return 0;
3569 }
3570 
3571 static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap)
3572 {
3573         struct mem_cgroup_threshold_ary *t;
3574         unsigned long usage;
3575         int i;
3576 
3577         rcu_read_lock();
3578         if (!swap)
3579                 t = rcu_dereference(memcg->thresholds.primary);
3580         else
3581                 t = rcu_dereference(memcg->memsw_thresholds.primary);
3582 
3583         if (!t)
3584                 goto unlock;
3585 
3586         usage = mem_cgroup_usage(memcg, swap);
3587 
3588         /*
3589          * current_threshold points to threshold just below or equal to usage.
3590          * If it's not true, a threshold was crossed after last
3591          * call of __mem_cgroup_threshold().
3592          */
3593         i = t->current_threshold;
3594 
3595         /*
3596          * Iterate backward over array of thresholds starting from
3597          * current_threshold and check if a threshold is crossed.
3598          * If none of thresholds below usage is crossed, we read
3599          * only one element of the array here.
3600          */
3601         for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--)
3602                 eventfd_signal(t->entries[i].eventfd, 1);
3603 
3604         /* i = current_threshold + 1 */
3605         i++;
3606 
3607         /*
3608          * Iterate forward over array of thresholds starting from
3609          * current_threshold+1 and check if a threshold is crossed.
3610          * If none of thresholds above usage is crossed, we read
3611          * only one element of the array here.
3612          */
3613         for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++)
3614                 eventfd_signal(t->entries[i].eventfd, 1);
3615 
3616         /* Update current_threshold */
3617         t->current_threshold = i - 1;
3618 unlock:
3619         rcu_read_unlock();
3620 }
3621 
3622 static void mem_cgroup_threshold(struct mem_cgroup *memcg)
3623 {
3624         while (memcg) {
3625                 __mem_cgroup_threshold(memcg, false);
3626                 if (do_swap_account)
3627                         __mem_cgroup_threshold(memcg, true);
3628 
3629                 memcg = parent_mem_cgroup(memcg);
3630         }
3631 }
3632 
3633 static int compare_thresholds(const void *a, const void *b)
3634 {
3635         const struct mem_cgroup_threshold *_a = a;
3636         const struct mem_cgroup_threshold *_b = b;
3637 
3638         if (_a->threshold > _b->threshold)
3639                 return 1;
3640 
3641         if (_a->threshold < _b->threshold)
3642                 return -1;
3643 
3644         return 0;
3645 }
3646 
3647 static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg)
3648 {
3649         struct mem_cgroup_eventfd_list *ev;
3650 
3651         spin_lock(&memcg_oom_lock);
3652 
3653         list_for_each_entry(ev, &memcg->oom_notify, list)
3654                 eventfd_signal(ev->eventfd, 1);
3655 
3656         spin_unlock(&memcg_oom_lock);
3657         return 0;
3658 }
3659 
3660 static void mem_cgroup_oom_notify(struct mem_cgroup *memcg)
3661 {
3662         struct mem_cgroup *iter;
3663 
3664         for_each_mem_cgroup_tree(iter, memcg)
3665                 mem_cgroup_oom_notify_cb(iter);
3666 }
3667 
3668 static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3669         struct eventfd_ctx *eventfd, const char *args, enum res_type type)
3670 {
3671         struct mem_cgroup_thresholds *thresholds;
3672         struct mem_cgroup_threshold_ary *new;
3673         unsigned long threshold;
3674         unsigned long usage;
3675         int i, size, ret;
3676 
3677         ret = page_counter_memparse(args, "-1", &threshold);
3678         if (ret)
3679                 return ret;
3680         threshold <<= PAGE_SHIFT;
3681 
3682         mutex_lock(&memcg->thresholds_lock);
3683 
3684         if (type == _MEM) {
3685                 thresholds = &memcg->thresholds;
3686                 usage = mem_cgroup_usage(memcg, false);
3687         } else if (type == _MEMSWAP) {
3688                 thresholds = &memcg->memsw_thresholds;
3689                 usage = mem_cgroup_usage(memcg, true);
3690         } else
3691                 BUG();
3692 
3693         /* Check if a threshold crossed before adding a new one */
3694         if (thresholds->primary)
3695                 __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3696 
3697         size = thresholds->primary ? thresholds->primary->size + 1 : 1;
3698 
3699         /* Allocate memory for new array of thresholds */
3700         new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold),
3701                         GFP_KERNEL);
3702         if (!new) {
3703                 ret = -ENOMEM;
3704                 goto unlock;
3705         }
3706         new->size = size;
3707 
3708         /* Copy thresholds (if any) to new array */
3709         if (thresholds->primary) {
3710                 memcpy(new->entries, thresholds->primary->entries, (size - 1) *
3711                                 sizeof(struct mem_cgroup_threshold));
3712         }
3713 
3714         /* Add new threshold */
3715         new->entries[size - 1].eventfd = eventfd;
3716         new->entries[size - 1].threshold = threshold;
3717 
3718         /* Sort thresholds. Registering of new threshold isn't time-critical */
3719         sort(new->entries, size, sizeof(struct mem_cgroup_threshold),
3720                         compare_thresholds, NULL);
3721 
3722         /* Find current threshold */
3723         new->current_threshold = -1;
3724         for (i = 0; i < size; i++) {
3725                 if (new->entries[i].threshold <= usage) {
3726                         /*
3727                          * new->current_threshold will not be used until
3728                          * rcu_assign_pointer(), so it's safe to increment
3729                          * it here.
3730                          */
3731                         ++new->current_threshold;
3732                 } else
3733                         break;
3734         }
3735 
3736         /* Free old spare buffer and save old primary buffer as spare */
3737         kfree(thresholds->spare);
3738         thresholds->spare = thresholds->primary;
3739 
3740         rcu_assign_pointer(thresholds->primary, new);
3741 
3742         /* To be sure that nobody uses thresholds */
3743         synchronize_rcu();
3744 
3745 unlock:
3746         mutex_unlock(&memcg->thresholds_lock);
3747 
3748         return ret;
3749 }
3750 
3751 static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg,
3752         struct eventfd_ctx *eventfd, const char *args)
3753 {
3754         return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM);
3755 }
3756 
3757 static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg,
3758         struct eventfd_ctx *eventfd, const char *args)
3759 {
3760         return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP);
3761 }
3762 
3763 static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3764         struct eventfd_ctx *eventfd, enum res_type type)
3765 {
3766         struct mem_cgroup_thresholds *thresholds;
3767         struct mem_cgroup_threshold_ary *new;
3768         unsigned long usage;
3769         int i, j, size;
3770 
3771         mutex_lock(&memcg->thresholds_lock);
3772 
3773         if (type == _MEM) {
3774                 thresholds = &memcg->thresholds;
3775                 usage = mem_cgroup_usage(memcg, false);
3776         } else if (type == _MEMSWAP) {
3777                 thresholds = &memcg->memsw_thresholds;
3778                 usage = mem_cgroup_usage(memcg, true);
3779         } else
3780                 BUG();
3781 
3782         if (!thresholds->primary)
3783                 goto unlock;
3784 
3785         /* Check if a threshold crossed before removing */
3786         __mem_cgroup_threshold(memcg, type == _MEMSWAP);
3787 
3788         /* Calculate new number of threshold */
3789         size = 0;
3790         for (i = 0; i < thresholds->primary->size; i++) {
3791                 if (thresholds->primary->entries[i].eventfd != eventfd)
3792                         size++;
3793         }
3794 
3795         new = thresholds->spare;
3796 
3797         /* Set thresholds array to NULL if we don't have thresholds */
3798         if (!size) {
3799                 kfree(new);
3800                 new = NULL;
3801                 goto swap_buffers;
3802         }
3803 
3804         new->size = size;
3805 
3806         /* Copy thresholds and find current threshold */
3807         new->current_threshold = -1;
3808         for (i = 0, j = 0; i < thresholds->primary->size; i++) {
3809                 if (thresholds->primary->entries[i].eventfd == eventfd)
3810                         continue;
3811 
3812                 new->entries[j] = thresholds->primary->entries[i];
3813                 if (new->entries[j].threshold <= usage) {
3814                         /*
3815                          * new->current_threshold will not be used
3816                          * until rcu_assign_pointer(), so it's safe to increment
3817                          * it here.
3818                          */
3819                         ++new->current_threshold;
3820                 }
3821                 j++;
3822         }
3823 
3824 swap_buffers:
3825         /* Swap primary and spare array */
3826         thresholds->spare = thresholds->primary;
3827 
3828         rcu_assign_pointer(thresholds->primary, new);
3829 
3830         /* To be sure that nobody uses thresholds */
3831         synchronize_rcu();
3832 
3833         /* If all events are unregistered, free the spare array */
3834         if (!new) {
3835                 kfree(thresholds->spare);
3836                 thresholds->spare = NULL;
3837         }
3838 unlock:
3839         mutex_unlock(&memcg->thresholds_lock);
3840 }
3841 
3842 static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3843         struct eventfd_ctx *eventfd)
3844 {
3845         return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM);
3846 }
3847 
3848 static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg,
3849         struct eventfd_ctx *eventfd)
3850 {
3851         return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP);
3852 }
3853 
3854 static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg,
3855         struct eventfd_ctx *eventfd, const char *args)
3856 {
3857         struct mem_cgroup_eventfd_list *event;
3858 
3859         event = kmalloc(sizeof(*event), GFP_KERNEL);
3860         if (!event)
3861                 return -ENOMEM;
3862 
3863         spin_lock(&memcg_oom_lock);
3864 
3865         event->eventfd = eventfd;
3866         list_add(&event->list, &memcg->oom_notify);
3867 
3868         /* already in OOM ? */
3869         if (atomic_read(&memcg->under_oom))
3870                 eventfd_signal(eventfd, 1);
3871         spin_unlock(&memcg_oom_lock);
3872 
3873         return 0;
3874 }
3875 
3876 static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg,
3877         struct eventfd_ctx *eventfd)
3878 {
3879         struct mem_cgroup_eventfd_list *ev, *tmp;
3880 
3881         spin_lock(&memcg_oom_lock);
3882 
3883         list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) {
3884                 if (ev->eventfd == eventfd) {
3885                         list_del(&ev->list);
3886                         kfree(ev);
3887                 }
3888         }
3889 
3890         spin_unlock(&memcg_oom_lock);
3891 }
3892 
3893 static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v)
3894 {
3895         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf));
3896 
3897         seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable);
3898         seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom));
3899         return 0;
3900 }
3901 
3902 static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css,
3903         struct cftype *cft, u64 val)
3904 {
3905         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
3906 
3907         /* cannot set to root cgroup and only 0 and 1 are allowed */
3908         if (!css->parent || !((val == 0) || (val == 1)))
3909                 return -EINVAL;
3910 
3911         memcg->oom_kill_disable = val;
3912         if (!val)
3913                 memcg_oom_recover(memcg);
3914 
3915         return 0;
3916 }
3917 
3918 #ifdef CONFIG_MEMCG_KMEM
3919 static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
3920 {
3921         int ret;
3922 
3923         ret = memcg_propagate_kmem(memcg);
3924         if (ret)
3925                 return ret;
3926 
3927         return mem_cgroup_sockets_init(memcg, ss);
3928 }
3929 
3930 static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3931 {
3932         struct cgroup_subsys_state *css;
3933         struct mem_cgroup *parent, *child;
3934         int kmemcg_id;
3935 
3936         if (!memcg->kmem_acct_active)
3937                 return;
3938 
3939         /*
3940          * Clear the 'active' flag before clearing memcg_caches arrays entries.
3941          * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it
3942          * guarantees no cache will be created for this cgroup after we are
3943          * done (see memcg_create_kmem_cache()).
3944          */
3945         memcg->kmem_acct_active = false;
3946 
3947         memcg_deactivate_kmem_caches(memcg);
3948 
3949         kmemcg_id = memcg->kmemcg_id;
3950         BUG_ON(kmemcg_id < 0);
3951 
3952         parent = parent_mem_cgroup(memcg);
3953         if (!parent)
3954                 parent = root_mem_cgroup;
3955 
3956         /*
3957          * Change kmemcg_id of this cgroup and all its descendants to the
3958          * parent's id, and then move all entries from this cgroup's list_lrus
3959          * to ones of the parent. After we have finished, all list_lrus
3960          * corresponding to this cgroup are guaranteed to remain empty. The
3961          * ordering is imposed by list_lru_node->lock taken by
3962          * memcg_drain_all_list_lrus().
3963          */
3964         css_for_each_descendant_pre(css, &memcg->css) {
3965                 child = mem_cgroup_from_css(css);
3966                 BUG_ON(child->kmemcg_id != kmemcg_id);
3967                 child->kmemcg_id = parent->kmemcg_id;
3968                 if (!memcg->use_hierarchy)
3969                         break;
3970         }
3971         memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id);
3972 
3973         memcg_free_cache_id(kmemcg_id);
3974 }
3975 
3976 static void memcg_destroy_kmem(struct mem_cgroup *memcg)
3977 {
3978         if (memcg->kmem_acct_activated) {
3979                 memcg_destroy_kmem_caches(memcg);
3980                 static_key_slow_dec(&memcg_kmem_enabled_key);
3981                 WARN_ON(page_counter_read(&memcg->kmem));
3982         }
3983         mem_cgroup_sockets_destroy(memcg);
3984 }
3985 #else
3986 static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
3987 {
3988         return 0;
3989 }
3990 
3991 static void memcg_deactivate_kmem(struct mem_cgroup *memcg)
3992 {
3993 }
3994 
3995 static void memcg_destroy_kmem(struct mem_cgroup *memcg)
3996 {
3997 }
3998 #endif
3999 
4000 /*
4001  * DO NOT USE IN NEW FILES.
4002  *
4003  * "cgroup.event_control" implementation.
4004  *
4005  * This is way over-engineered.  It tries to support fully configurable
4006  * events for each user.  Such level of flexibility is completely
4007  * unnecessary especially in the light of the planned unified hierarchy.
4008  *
4009  * Please deprecate this and replace with something simpler if at all
4010  * possible.
4011  */
4012 
4013 /*
4014  * Unregister event and free resources.
4015  *
4016  * Gets called from workqueue.
4017  */
4018 static void memcg_event_remove(struct work_struct *work)
4019 {
4020         struct mem_cgroup_event *event =
4021                 container_of(work, struct mem_cgroup_event, remove);
4022         struct mem_cgroup *memcg = event->memcg;
4023 
4024         remove_wait_queue(event->wqh, &event->wait);
4025 
4026         event->unregister_event(memcg, event->eventfd);
4027 
4028         /* Notify userspace the event is going away. */
4029         eventfd_signal(event->eventfd, 1);
4030 
4031         eventfd_ctx_put(event->eventfd);
4032         kfree(event);
4033         css_put(&memcg->css);
4034 }
4035 
4036 /*
4037  * Gets called on POLLHUP on eventfd when user closes it.
4038  *
4039  * Called with wqh->lock held and interrupts disabled.
4040  */
4041 static int memcg_event_wake(wait_queue_t *wait, unsigned mode,
4042                             int sync, void *key)
4043 {
4044         struct mem_cgroup_event *event =
4045                 container_of(wait, struct mem_cgroup_event, wait);
4046         struct mem_cgroup *memcg = event->memcg;
4047         unsigned long flags = (unsigned long)key;
4048 
4049         if (flags & POLLHUP) {
4050                 /*
4051                  * If the event has been detached at cgroup removal, we
4052                  * can simply return knowing the other side will cleanup
4053                  * for us.
4054                  *
4055                  * We can't race against event freeing since the other
4056                  * side will require wqh->lock via remove_wait_queue(),
4057                  * which we hold.
4058                  */
4059                 spin_lock(&memcg->event_list_lock);
4060                 if (!list_empty(&event->list)) {
4061                         list_del_init(&event->list);
4062                         /*
4063                          * We are in atomic context, but cgroup_event_remove()
4064                          * may sleep, so we have to call it in workqueue.
4065                          */
4066                         schedule_work(&event->remove);
4067                 }
4068                 spin_unlock(&memcg->event_list_lock);
4069         }
4070 
4071         return 0;
4072 }
4073 
4074 static void memcg_event_ptable_queue_proc(struct file *file,
4075                 wait_queue_head_t *wqh, poll_table *pt)
4076 {
4077         struct mem_cgroup_event *event =
4078                 container_of(pt, struct mem_cgroup_event, pt);
4079 
4080         event->wqh = wqh;
4081         add_wait_queue(wqh, &event->wait);
4082 }
4083 
4084 /*
4085  * DO NOT USE IN NEW FILES.
4086  *
4087  * Parse input and register new cgroup event handler.
4088  *
4089  * Input must be in format '<event_fd> <control_fd> <args>'.
4090  * Interpretation of args is defined by control file implementation.
4091  */
4092 static ssize_t memcg_write_event_control(struct kernfs_open_file *of,
4093                                          char *buf, size_t nbytes, loff_t off)
4094 {
4095         struct cgroup_subsys_state *css = of_css(of);
4096         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4097         struct mem_cgroup_event *event;
4098         struct cgroup_subsys_state *cfile_css;
4099         unsigned int efd, cfd;
4100         struct fd efile;
4101         struct fd cfile;
4102         const char *name;
4103         char *endp;
4104         int ret;
4105 
4106         buf = strstrip(buf);
4107 
4108         efd = simple_strtoul(buf, &endp, 10);
4109         if (*endp != ' ')
4110                 return -EINVAL;
4111         buf = endp + 1;
4112 
4113         cfd = simple_strtoul(buf, &endp, 10);
4114         if ((*endp != ' ') && (*endp != '\0'))
4115                 return -EINVAL;
4116         buf = endp + 1;
4117 
4118         event = kzalloc(sizeof(*event), GFP_KERNEL);
4119         if (!event)
4120                 return -ENOMEM;
4121 
4122         event->memcg = memcg;
4123         INIT_LIST_HEAD(&event->list);
4124         init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc);
4125         init_waitqueue_func_entry(&event->wait, memcg_event_wake);
4126         INIT_WORK(&event->remove, memcg_event_remove);
4127 
4128         efile = fdget(efd);
4129         if (!efile.file) {
4130                 ret = -EBADF;
4131                 goto out_kfree;
4132         }
4133 
4134         event->eventfd = eventfd_ctx_fileget(efile.file);
4135         if (IS_ERR(event->eventfd)) {
4136                 ret = PTR_ERR(event->eventfd);
4137                 goto out_put_efile;
4138         }
4139 
4140         cfile = fdget(cfd);
4141         if (!cfile.file) {
4142                 ret = -EBADF;
4143                 goto out_put_eventfd;
4144         }
4145 
4146         /* the process need read permission on control file */
4147         /* AV: shouldn't we check that it's been opened for read instead? */
4148         ret = inode_permission(file_inode(cfile.file), MAY_READ);
4149         if (ret < 0)
4150                 goto out_put_cfile;
4151 
4152         /*
4153          * Determine the event callbacks and set them in @event.  This used
4154          * to be done via struct cftype but cgroup core no longer knows
4155          * about these events.  The following is crude but the whole thing
4156          * is for compatibility anyway.
4157          *
4158          * DO NOT ADD NEW FILES.
4159          */
4160         name = cfile.file->f_path.dentry->d_name.name;
4161 
4162         if (!strcmp(name, "memory.usage_in_bytes")) {
4163                 event->register_event = mem_cgroup_usage_register_event;
4164                 event->unregister_event = mem_cgroup_usage_unregister_event;
4165         } else if (!strcmp(name, "memory.oom_control")) {
4166                 event->register_event = mem_cgroup_oom_register_event;
4167                 event->unregister_event = mem_cgroup_oom_unregister_event;
4168         } else if (!strcmp(name, "memory.pressure_level")) {
4169                 event->register_event = vmpressure_register_event;
4170                 event->unregister_event = vmpressure_unregister_event;
4171         } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) {
4172                 event->register_event = memsw_cgroup_usage_register_event;
4173                 event->unregister_event = memsw_cgroup_usage_unregister_event;
4174         } else {
4175                 ret = -EINVAL;
4176                 goto out_put_cfile;
4177         }
4178 
4179         /*
4180          * Verify @cfile should belong to @css.  Also, remaining events are
4181          * automatically removed on cgroup destruction but the removal is
4182          * asynchronous, so take an extra ref on @css.
4183          */
4184         cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent,
4185                                                &memory_cgrp_subsys);
4186         ret = -EINVAL;
4187         if (IS_ERR(cfile_css))
4188                 goto out_put_cfile;
4189         if (cfile_css != css) {
4190                 css_put(cfile_css);
4191                 goto out_put_cfile;
4192         }
4193 
4194         ret = event->register_event(memcg, event->eventfd, buf);
4195         if (ret)
4196                 goto out_put_css;
4197 
4198         efile.file->f_op->poll(efile.file, &event->pt);
4199 
4200         spin_lock(&memcg->event_list_lock);
4201         list_add(&event->list, &memcg->event_list);
4202         spin_unlock(&memcg->event_list_lock);
4203 
4204         fdput(cfile);
4205         fdput(efile);
4206 
4207         return nbytes;
4208 
4209 out_put_css:
4210         css_put(css);
4211 out_put_cfile:
4212         fdput(cfile);
4213 out_put_eventfd:
4214         eventfd_ctx_put(event->eventfd);
4215 out_put_efile:
4216         fdput(efile);
4217 out_kfree:
4218         kfree(event);
4219 
4220         return ret;
4221 }
4222 
4223 static struct cftype mem_cgroup_legacy_files[] = {
4224         {
4225                 .name = "usage_in_bytes",
4226                 .private = MEMFILE_PRIVATE(_MEM, RES_USAGE),
4227                 .read_u64 = mem_cgroup_read_u64,
4228         },
4229         {
4230                 .name = "max_usage_in_bytes",
4231                 .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE),
4232                 .write = mem_cgroup_reset,
4233                 .read_u64 = mem_cgroup_read_u64,
4234         },
4235         {
4236                 .name = "limit_in_bytes",
4237                 .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT),
4238                 .write = mem_cgroup_write,
4239                 .read_u64 = mem_cgroup_read_u64,
4240         },
4241         {
4242                 .name = "soft_limit_in_bytes",
4243                 .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT),
4244                 .write = mem_cgroup_write,
4245                 .read_u64 = mem_cgroup_read_u64,
4246         },
4247         {
4248                 .name = "failcnt",
4249                 .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT),
4250                 .write = mem_cgroup_reset,
4251                 .read_u64 = mem_cgroup_read_u64,
4252         },
4253         {
4254                 .name = "stat",
4255                 .seq_show = memcg_stat_show,
4256         },
4257         {
4258                 .name = "force_empty",
4259                 .write = mem_cgroup_force_empty_write,
4260         },
4261         {
4262                 .name = "use_hierarchy",
4263                 .write_u64 = mem_cgroup_hierarchy_write,
4264                 .read_u64 = mem_cgroup_hierarchy_read,
4265         },
4266         {
4267                 .name = "cgroup.event_control",         /* XXX: for compat */
4268                 .write = memcg_write_event_control,
4269                 .flags = CFTYPE_NO_PREFIX,
4270                 .mode = S_IWUGO,
4271         },
4272         {
4273                 .name = "swappiness",
4274                 .read_u64 = mem_cgroup_swappiness_read,
4275                 .write_u64 = mem_cgroup_swappiness_write,
4276         },
4277         {
4278                 .name = "move_charge_at_immigrate",
4279                 .read_u64 = mem_cgroup_move_charge_read,
4280                 .write_u64 = mem_cgroup_move_charge_write,
4281         },
4282         {
4283                 .name = "oom_control",
4284                 .seq_show = mem_cgroup_oom_control_read,
4285                 .write_u64 = mem_cgroup_oom_control_write,
4286                 .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL),
4287         },
4288         {
4289                 .name = "pressure_level",
4290         },
4291 #ifdef CONFIG_NUMA
4292         {
4293                 .name = "numa_stat",
4294                 .seq_show = memcg_numa_stat_show,
4295         },
4296 #endif
4297 #ifdef CONFIG_MEMCG_KMEM
4298         {
4299                 .name = "kmem.limit_in_bytes",
4300                 .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT),
4301                 .write = mem_cgroup_write,
4302                 .read_u64 = mem_cgroup_read_u64,
4303         },
4304         {
4305                 .name = "kmem.usage_in_bytes",
4306                 .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE),
4307                 .read_u64 = mem_cgroup_read_u64,
4308         },
4309         {
4310                 .name = "kmem.failcnt",
4311                 .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT),
4312                 .write = mem_cgroup_reset,
4313                 .read_u64 = mem_cgroup_read_u64,
4314         },
4315         {
4316                 .name = "kmem.max_usage_in_bytes",
4317                 .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE),
4318                 .write = mem_cgroup_reset,
4319                 .read_u64 = mem_cgroup_read_u64,
4320         },
4321 #ifdef CONFIG_SLABINFO
4322         {
4323                 .name = "kmem.slabinfo",
4324                 .seq_start = slab_start,
4325                 .seq_next = slab_next,
4326                 .seq_stop = slab_stop,
4327                 .seq_show = memcg_slab_show,
4328         },
4329 #endif
4330 #endif
4331         { },    /* terminate */
4332 };
4333 
4334 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4335 {
4336         struct mem_cgroup_per_node *pn;
4337         struct mem_cgroup_per_zone *mz;
4338         int zone, tmp = node;
4339         /*
4340          * This routine is called against possible nodes.
4341          * But it's BUG to call kmalloc() against offline node.
4342          *
4343          * TODO: this routine can waste much memory for nodes which will
4344          *       never be onlined. It's better to use memory hotplug callback
4345          *       function.
4346          */
4347         if (!node_state(node, N_NORMAL_MEMORY))
4348                 tmp = -1;
4349         pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
4350         if (!pn)
4351                 return 1;
4352 
4353         for (zone = 0; zone < MAX_NR_ZONES; zone++) {
4354                 mz = &pn->zoneinfo[zone];
4355                 lruvec_init(&mz->lruvec);
4356                 mz->usage_in_excess = 0;
4357                 mz->on_tree = false;
4358                 mz->memcg = memcg;
4359         }
4360         memcg->nodeinfo[node] = pn;
4361         return 0;
4362 }
4363 
4364 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node)
4365 {
4366         kfree(memcg->nodeinfo[node]);
4367 }
4368 
4369 static struct mem_cgroup *mem_cgroup_alloc(void)
4370 {
4371         struct mem_cgroup *memcg;
4372         size_t size;
4373 
4374         size = sizeof(struct mem_cgroup);
4375         size += nr_node_ids * sizeof(struct mem_cgroup_per_node *);
4376 
4377         memcg = kzalloc(size, GFP_KERNEL);
4378         if (!memcg)
4379                 return NULL;
4380 
4381         memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu);
4382         if (!memcg->stat)
4383                 goto out_free;
4384         spin_lock_init(&memcg->pcp_counter_lock);
4385         return memcg;
4386 
4387 out_free:
4388         kfree(memcg);
4389         return NULL;
4390 }
4391 
4392 /*
4393  * At destroying mem_cgroup, references from swap_cgroup can remain.
4394  * (scanning all at force_empty is too costly...)
4395  *
4396  * Instead of clearing all references at force_empty, we remember
4397  * the number of reference from swap_cgroup and free mem_cgroup when
4398  * it goes down to 0.
4399  *
4400  * Removal of cgroup itself succeeds regardless of refs from swap.
4401  */
4402 
4403 static void __mem_cgroup_free(struct mem_cgroup *memcg)
4404 {
4405         int node;
4406 
4407         mem_cgroup_remove_from_trees(memcg);
4408 
4409         for_each_node(node)
4410                 free_mem_cgroup_per_zone_info(memcg, node);
4411 
4412         free_percpu(memcg->stat);
4413         kfree(memcg);
4414 }
4415 
4416 /*
4417  * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4418  */
4419 struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg)
4420 {
4421         if (!memcg->memory.parent)
4422                 return NULL;
4423         return mem_cgroup_from_counter(memcg->memory.parent, memory);
4424 }
4425 EXPORT_SYMBOL(parent_mem_cgroup);
4426 
4427 static struct cgroup_subsys_state * __ref
4428 mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
4429 {
4430         struct mem_cgroup *memcg;
4431         long error = -ENOMEM;
4432         int node;
4433 
4434         memcg = mem_cgroup_alloc();
4435         if (!memcg)
4436                 return ERR_PTR(error);
4437 
4438         for_each_node(node)
4439                 if (alloc_mem_cgroup_per_zone_info(memcg, node))
4440                         goto free_out;
4441 
4442         /* root ? */
4443         if (parent_css == NULL) {
4444                 root_mem_cgroup = memcg;
4445                 page_counter_init(&memcg->memory, NULL);
4446                 memcg->high = PAGE_COUNTER_MAX;
4447                 memcg->soft_limit = PAGE_COUNTER_MAX;
4448                 page_counter_init(&memcg->memsw, NULL);
4449                 page_counter_init(&memcg->kmem, NULL);
4450         }
4451 
4452         memcg->last_scanned_node = MAX_NUMNODES;
4453         INIT_LIST_HEAD(&memcg->oom_notify);
4454         memcg->move_charge_at_immigrate = 0;
4455         mutex_init(&memcg->thresholds_lock);
4456         spin_lock_init(&memcg->move_lock);
4457         vmpressure_init(&memcg->vmpressure);
4458         INIT_LIST_HEAD(&memcg->event_list);
4459         spin_lock_init(&memcg->event_list_lock);
4460 #ifdef CONFIG_MEMCG_KMEM
4461         memcg->kmemcg_id = -1;
4462 #endif
4463 
4464         return &memcg->css;
4465 
4466 free_out:
4467         __mem_cgroup_free(memcg);
4468         return ERR_PTR(error);
4469 }
4470 
4471 static int
4472 mem_cgroup_css_online(struct cgroup_subsys_state *css)
4473 {
4474         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4475         struct mem_cgroup *parent = mem_cgroup_from_css(css->parent);
4476         int ret;
4477 
4478         if (css->id > MEM_CGROUP_ID_MAX)
4479                 return -ENOSPC;
4480 
4481         if (!parent)
4482                 return 0;
4483 
4484         mutex_lock(&memcg_create_mutex);
4485 
4486         memcg->use_hierarchy = parent->use_hierarchy;
4487         memcg->oom_kill_disable = parent->oom_kill_disable;
4488         memcg->swappiness = mem_cgroup_swappiness(parent);
4489 
4490         if (parent->use_hierarchy) {
4491                 page_counter_init(&memcg->memory, &parent->memory);
4492                 memcg->high = PAGE_COUNTER_MAX;
4493                 memcg->soft_limit = PAGE_COUNTER_MAX;
4494                 page_counter_init(&memcg->memsw, &parent->memsw);
4495                 page_counter_init(&memcg->kmem, &parent->kmem);
4496 
4497                 /*
4498                  * No need to take a reference to the parent because cgroup
4499                  * core guarantees its existence.
4500                  */
4501         } else {
4502                 page_counter_init(&memcg->memory, NULL);
4503                 memcg->high = PAGE_COUNTER_MAX;
4504                 memcg->soft_limit = PAGE_COUNTER_MAX;
4505                 page_counter_init(&memcg->memsw, NULL);
4506                 page_counter_init(&memcg->kmem, NULL);
4507                 /*
4508                  * Deeper hierachy with use_hierarchy == false doesn't make
4509                  * much sense so let cgroup subsystem know about this
4510                  * unfortunate state in our controller.
4511                  */
4512                 if (parent != root_mem_cgroup)
4513                         memory_cgrp_subsys.broken_hierarchy = true;
4514         }
4515         mutex_unlock(&memcg_create_mutex);
4516 
4517         ret = memcg_init_kmem(memcg, &memory_cgrp_subsys);
4518         if (ret)
4519                 return ret;
4520 
4521         /*
4522          * Make sure the memcg is initialized: mem_cgroup_iter()
4523          * orders reading memcg->initialized against its callers
4524          * reading the memcg members.
4525          */
4526         smp_store_release(&memcg->initialized, 1);
4527 
4528         return 0;
4529 }
4530 
4531 static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
4532 {
4533         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4534         struct mem_cgroup_event *event, *tmp;
4535 
4536         /*
4537          * Unregister events and notify userspace.
4538          * Notify userspace about cgroup removing only after rmdir of cgroup
4539          * directory to avoid race between userspace and kernelspace.
4540          */
4541         spin_lock(&memcg->event_list_lock);
4542         list_for_each_entry_safe(event, tmp, &memcg->event_list, list) {
4543                 list_del_init(&event->list);
4544                 schedule_work(&event->remove);
4545         }
4546         spin_unlock(&memcg->event_list_lock);
4547 
4548         vmpressure_cleanup(&memcg->vmpressure);
4549 
4550         memcg_deactivate_kmem(memcg);
4551 }
4552 
4553 static void mem_cgroup_css_free(struct cgroup_subsys_state *css)
4554 {
4555         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4556 
4557         memcg_destroy_kmem(memcg);
4558         __mem_cgroup_free(memcg);
4559 }
4560 
4561 /**
4562  * mem_cgroup_css_reset - reset the states of a mem_cgroup
4563  * @css: the target css
4564  *
4565  * Reset the states of the mem_cgroup associated with @css.  This is
4566  * invoked when the userland requests disabling on the default hierarchy
4567  * but the memcg is pinned through dependency.  The memcg should stop
4568  * applying policies and should revert to the vanilla state as it may be
4569  * made visible again.
4570  *
4571  * The current implementation only resets the essential configurations.
4572  * This needs to be expanded to cover all the visible parts.
4573  */
4574 static void mem_cgroup_css_reset(struct cgroup_subsys_state *css)
4575 {
4576         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
4577 
4578         mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX);
4579         mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX);
4580         memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX);
4581         memcg->low = 0;
4582         memcg->high = PAGE_COUNTER_MAX;
4583         memcg->soft_limit = PAGE_COUNTER_MAX;
4584 }
4585 
4586 #ifdef CONFIG_MMU
4587 /* Handlers for move charge at task migration. */
4588 static int mem_cgroup_do_precharge(unsigned long count)
4589 {
4590         int ret;
4591 
4592         /* Try a single bulk charge without reclaim first */
4593         ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count);
4594         if (!ret) {
4595                 mc.precharge += count;
4596                 return ret;
4597         }
4598         if (ret == -EINTR) {
4599                 cancel_charge(root_mem_cgroup, count);
4600                 return ret;
4601         }
4602 
4603         /* Try charges one by one with reclaim */
4604         while (count--) {
4605                 ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1);
4606                 /*
4607                  * In case of failure, any residual charges against
4608                  * mc.to will be dropped by mem_cgroup_clear_mc()
4609                  * later on.  However, cancel any charges that are
4610                  * bypassed to root right away or they'll be lost.
4611                  */
4612                 if (ret == -EINTR)
4613                         cancel_charge(root_mem_cgroup, 1);
4614                 if (ret)
4615                         return ret;
4616                 mc.precharge++;
4617                 cond_resched();
4618         }
4619         return 0;
4620 }
4621 
4622 /**
4623  * get_mctgt_type - get target type of moving charge
4624  * @vma: the vma the pte to be checked belongs
4625  * @addr: the address corresponding to the pte to be checked
4626  * @ptent: the pte to be checked
4627  * @target: the pointer the target page or swap ent will be stored(can be NULL)
4628  *
4629  * Returns
4630  *   0(MC_TARGET_NONE): if the pte is not a target for move charge.
4631  *   1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4632  *     move charge. if @target is not NULL, the page is stored in target->page
4633  *     with extra refcnt got(Callers should handle it).
4634  *   2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4635  *     target for charge migration. if @target is not NULL, the entry is stored
4636  *     in target->ent.
4637  *
4638  * Called with pte lock held.
4639  */
4640 union mc_target {
4641         struct page     *page;
4642         swp_entry_t     ent;
4643 };
4644 
4645 enum mc_target_type {
4646         MC_TARGET_NONE = 0,
4647         MC_TARGET_PAGE,
4648         MC_TARGET_SWAP,
4649 };
4650 
4651 static struct page *mc_handle_present_pte(struct vm_area_struct *vma,
4652                                                 unsigned long addr, pte_t ptent)
4653 {
4654         struct page *page = vm_normal_page(vma, addr, ptent);
4655 
4656         if (!page || !page_mapped(page))
4657                 return NULL;
4658         if (PageAnon(page)) {
4659                 if (!(mc.flags & MOVE_ANON))
4660                         return NULL;
4661         } else {
4662                 if (!(mc.flags & MOVE_FILE))
4663                         return NULL;
4664         }
4665         if (!get_page_unless_zero(page))
4666                 return NULL;
4667 
4668         return page;
4669 }
4670 
4671 #ifdef CONFIG_SWAP
4672 static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4673                         unsigned long addr, pte_t ptent, swp_entry_t *entry)
4674 {
4675         struct page *page = NULL;
4676         swp_entry_t ent = pte_to_swp_entry(ptent);
4677 
4678         if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent))
4679                 return NULL;
4680         /*
4681          * Because lookup_swap_cache() updates some statistics counter,
4682          * we call find_get_page() with swapper_space directly.
4683          */
4684         page = find_get_page(swap_address_space(ent), ent.val);
4685         if (do_swap_account)
4686                 entry->val = ent.val;
4687 
4688         return page;
4689 }
4690 #else
4691 static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
4692                         unsigned long addr, pte_t ptent, swp_entry_t *entry)
4693 {
4694         return NULL;
4695 }
4696 #endif
4697 
4698 static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
4699                         unsigned long addr, pte_t ptent, swp_entry_t *entry)
4700 {
4701         struct page *page = NULL;
4702         struct address_space *mapping;
4703         pgoff_t pgoff;
4704 
4705         if (!vma->vm_file) /* anonymous vma */
4706                 return NULL;
4707         if (!(mc.flags & MOVE_FILE))
4708                 return NULL;
4709 
4710         mapping = vma->vm_file->f_mapping;
4711         pgoff = linear_page_index(vma, addr);
4712 
4713         /* page is moved even if it's not RSS of this task(page-faulted). */
4714 #ifdef CONFIG_SWAP
4715         /* shmem/tmpfs may report page out on swap: account for that too. */
4716         if (shmem_mapping(mapping)) {
4717                 page = find_get_entry(mapping, pgoff);
4718                 if (radix_tree_exceptional_entry(page)) {
4719                         swp_entry_t swp = radix_to_swp_entry(page);
4720                         if (do_swap_account)
4721                                 *entry = swp;
4722                         page = find_get_page(swap_address_space(swp), swp.val);
4723                 }
4724         } else
4725                 page = find_get_page(mapping, pgoff);
4726 #else
4727         page = find_get_page(mapping, pgoff);
4728 #endif
4729         return page;
4730 }
4731 
4732 /**
4733  * mem_cgroup_move_account - move account of the page
4734  * @page: the page
4735  * @nr_pages: number of regular pages (>1 for huge pages)
4736  * @from: mem_cgroup which the page is moved from.
4737  * @to: mem_cgroup which the page is moved to. @from != @to.
4738  *
4739  * The caller must confirm following.
4740  * - page is not on LRU (isolate_page() is useful.)
4741  * - compound_lock is held when nr_pages > 1
4742  *
4743  * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4744  * from old cgroup.
4745  */
4746 static int mem_cgroup_move_account(struct page *page,
4747                                    unsigned int nr_pages,
4748                                    struct mem_cgroup *from,
4749                                    struct mem_cgroup *to)
4750 {
4751         unsigned long flags;
4752         int ret;
4753 
4754         VM_BUG_ON(from == to);
4755         VM_BUG_ON_PAGE(PageLRU(page), page);
4756         /*
4757          * The page is isolated from LRU. So, collapse function
4758          * will not handle this page. But page splitting can happen.
4759          * Do this check under compound_page_lock(). The caller should
4760          * hold it.
4761          */
4762         ret = -EBUSY;
4763         if (nr_pages > 1 && !PageTransHuge(page))
4764                 goto out;
4765 
4766         /*
4767          * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup
4768          * of its source page while we change it: page migration takes
4769          * both pages off the LRU, but page cache replacement doesn't.
4770          */
4771         if (!trylock_page(page))
4772                 goto out;
4773 
4774         ret = -EINVAL;
4775         if (page->mem_cgroup != from)
4776                 goto out_unlock;
4777 
4778         spin_lock_irqsave(&from->move_lock, flags);
4779 
4780         if (!PageAnon(page) && page_mapped(page)) {
4781                 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4782                                nr_pages);
4783                 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
4784                                nr_pages);
4785         }
4786 
4787         if (PageWriteback(page)) {
4788                 __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4789                                nr_pages);
4790                 __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
4791                                nr_pages);
4792         }
4793 
4794         /*
4795          * It is safe to change page->mem_cgroup here because the page
4796          * is referenced, charged, and isolated - we can't race with
4797          * uncharging, charging, migration, or LRU putback.
4798          */
4799 
4800         /* caller should have done css_get */
4801         page->mem_cgroup = to;
4802         spin_unlock_irqrestore(&from->move_lock, flags);
4803 
4804         ret = 0;
4805 
4806         local_irq_disable();
4807         mem_cgroup_charge_statistics(to, page, nr_pages);
4808         memcg_check_events(to, page);
4809         mem_cgroup_charge_statistics(from, page, -nr_pages);
4810         memcg_check_events(from, page);
4811         local_irq_enable();
4812 out_unlock:
4813         unlock_page(page);
4814 out:
4815         return ret;
4816 }
4817 
4818 static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
4819                 unsigned long addr, pte_t ptent, union mc_target *target)
4820 {
4821         struct page *page = NULL;
4822         enum mc_target_type ret = MC_TARGET_NONE;
4823         swp_entry_t ent = { .val = 0 };
4824 
4825         if (pte_present(ptent))
4826                 page = mc_handle_present_pte(vma, addr, ptent);
4827         else if (is_swap_pte(ptent))
4828                 page = mc_handle_swap_pte(vma, addr, ptent, &ent);
4829         else if (pte_none(ptent))
4830                 page = mc_handle_file_pte(vma, addr, ptent, &ent);
4831 
4832         if (!page && !ent.val)
4833                 return ret;
4834         if (page) {
4835                 /*
4836                  * Do only loose check w/o serialization.
4837                  * mem_cgroup_move_account() checks the page is valid or
4838                  * not under LRU exclusion.
4839                  */
4840                 if (page->mem_cgroup == mc.from) {
4841                         ret = MC_TARGET_PAGE;
4842                         if (target)
4843                                 target->page = page;
4844                 }
4845                 if (!ret || !target)
4846                         put_page(page);
4847         }
4848         /* There is a swap entry and a page doesn't exist or isn't charged */
4849         if (ent.val && !ret &&
4850             mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) {
4851                 ret = MC_TARGET_SWAP;
4852                 if (target)
4853                         target->ent = ent;
4854         }
4855         return ret;
4856 }
4857 
4858 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4859 /*
4860  * We don't consider swapping or file mapped pages because THP does not
4861  * support them for now.
4862  * Caller should make sure that pmd_trans_huge(pmd) is true.
4863  */
4864 static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4865                 unsigned long addr, pmd_t pmd, union mc_target *target)
4866 {
4867         struct page *page = NULL;
4868         enum mc_target_type ret = MC_TARGET_NONE;
4869 
4870         page = pmd_page(pmd);
4871         VM_BUG_ON_PAGE(!page || !PageHead(page), page);
4872         if (!(mc.flags & MOVE_ANON))
4873                 return ret;
4874         if (page->mem_cgroup == mc.from) {
4875                 ret = MC_TARGET_PAGE;
4876                 if (target) {
4877                         get_page(page);
4878                         target->page = page;
4879                 }
4880         }
4881         return ret;
4882 }
4883 #else
4884 static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma,
4885                 unsigned long addr, pmd_t pmd, union mc_target *target)
4886 {
4887         return MC_TARGET_NONE;
4888 }
4889 #endif
4890 
4891 static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd,
4892                                         unsigned long addr, unsigned long end,
4893                                         struct mm_walk *walk)
4894 {
4895         struct vm_area_struct *vma = walk->vma;
4896         pte_t *pte;
4897         spinlock_t *ptl;
4898 
4899         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
4900                 if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE)
4901                         mc.precharge += HPAGE_PMD_NR;
4902                 spin_unlock(ptl);
4903                 return 0;
4904         }
4905 
4906         if (pmd_trans_unstable(pmd))
4907                 return 0;
4908         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
4909         for (; addr != end; pte++, addr += PAGE_SIZE)
4910                 if (get_mctgt_type(vma, addr, *pte, NULL))
4911                         mc.precharge++; /* increment precharge temporarily */
4912         pte_unmap_unlock(pte - 1, ptl);
4913         cond_resched();
4914 
4915         return 0;
4916 }
4917 
4918 static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm)
4919 {
4920         unsigned long precharge;
4921 
4922         struct mm_walk mem_cgroup_count_precharge_walk = {
4923                 .pmd_entry = mem_cgroup_count_precharge_pte_range,
4924                 .mm = mm,
4925         };
4926         down_read(&mm->mmap_sem);
4927         walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk);
4928         up_read(&mm->mmap_sem);
4929 
4930         precharge = mc.precharge;
4931         mc.precharge = 0;
4932 
4933         return precharge;
4934 }
4935 
4936 static int mem_cgroup_precharge_mc(struct mm_struct *mm)
4937 {
4938         unsigned long precharge = mem_cgroup_count_precharge(mm);
4939 
4940         VM_BUG_ON(mc.moving_task);
4941         mc.moving_task = current;
4942         return mem_cgroup_do_precharge(precharge);
4943 }
4944 
4945 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4946 static void __mem_cgroup_clear_mc(void)
4947 {
4948         struct mem_cgroup *from = mc.from;
4949         struct mem_cgroup *to = mc.to;
4950 
4951         /* we must uncharge all the leftover precharges from mc.to */
4952         if (mc.precharge) {
4953                 cancel_charge(mc.to, mc.precharge);
4954                 mc.precharge = 0;
4955         }
4956         /*
4957          * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4958          * we must uncharge here.
4959          */
4960         if (mc.moved_charge) {
4961                 cancel_charge(mc.from, mc.moved_charge);
4962                 mc.moved_charge = 0;
4963         }
4964         /* we must fixup refcnts and charges */
4965         if (mc.moved_swap) {
4966                 /* uncharge swap account from the old cgroup */
4967                 if (!mem_cgroup_is_root(mc.from))
4968                         page_counter_uncharge(&mc.from->memsw, mc.moved_swap);
4969 
4970                 /*
4971                  * we charged both to->memory and to->memsw, so we
4972                  * should uncharge to->memory.
4973                  */
4974                 if (!mem_cgroup_is_root(mc.to))
4975                         page_counter_uncharge(&mc.to->memory, mc.moved_swap);
4976 
4977                 css_put_many(&mc.from->css, mc.moved_swap);
4978 
4979                 /* we've already done css_get(mc.to) */
4980                 mc.moved_swap = 0;
4981         }
4982         memcg_oom_recover(from);
4983         memcg_oom_recover(to);
4984         wake_up_all(&mc.waitq);
4985 }
4986 
4987 static void mem_cgroup_clear_mc(void)
4988 {
4989         /*
4990          * we must clear moving_task before waking up waiters at the end of
4991          * task migration.
4992          */
4993         mc.moving_task = NULL;
4994         __mem_cgroup_clear_mc();
4995         spin_lock(&mc.lock);
4996         mc.from = NULL;
4997         mc.to = NULL;
4998         spin_unlock(&mc.lock);
4999 }
5000 
5001 static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
5002                                  struct cgroup_taskset *tset)
5003 {
5004         struct task_struct *p = cgroup_taskset_first(tset);
5005         int ret = 0;
5006         struct mem_cgroup *memcg = mem_cgroup_from_css(css);
5007         unsigned long move_flags;
5008 
5009         /*
5010          * We are now commited to this value whatever it is. Changes in this
5011          * tunable will only affect upcoming migrations, not the current one.
5012          * So we need to save it, and keep it going.
5013          */
5014         move_flags = READ_ONCE(memcg->move_charge_at_immigrate);
5015         if (move_flags) {
5016                 struct mm_struct *mm;
5017                 struct mem_cgroup *from = mem_cgroup_from_task(p);
5018 
5019                 VM_BUG_ON(from == memcg);
5020 
5021                 mm = get_task_mm(p);
5022                 if (!mm)
5023                         return 0;
5024                 /* We move charges only when we move a owner of the mm */
5025                 if (mm->owner == p) {
5026                         VM_BUG_ON(mc.from);
5027                         VM_BUG_ON(mc.to);
5028                         VM_BUG_ON(mc.precharge);
5029                         VM_BUG_ON(mc.moved_charge);
5030                         VM_BUG_ON(mc.moved_swap);
5031 
5032                         spin_lock(&mc.lock);
5033                         mc.from = from;
5034                         mc.to = memcg;
5035                         mc.flags = move_flags;
5036                         spin_unlock(&mc.lock);
5037                         /* We set mc.moving_task later */
5038 
5039                         ret = mem_cgroup_precharge_mc(mm);
5040                         if (ret)
5041                                 mem_cgroup_clear_mc();
5042                 }
5043                 mmput(mm);
5044         }
5045         return ret;
5046 }
5047 
5048 static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
5049                                      struct cgroup_taskset *tset)
5050 {
5051         if (mc.to)
5052                 mem_cgroup_clear_mc();
5053 }
5054 
5055 static int mem_cgroup_move_charge_pte_range(pmd_t *pmd,
5056                                 unsigned long addr, unsigned long end,
5057                                 struct mm_walk *walk)
5058 {
5059         int ret = 0;
5060         struct vm_area_struct *vma = walk->vma;
5061         pte_t *pte;
5062         spinlock_t *ptl;
5063         enum mc_target_type target_type;
5064         union mc_target target;
5065         struct page *page;
5066 
5067         /*
5068          * We don't take compound_lock() here but no race with splitting thp
5069          * happens because:
5070          *  - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5071          *    under splitting, which means there's no concurrent thp split,
5072          *  - if another thread runs into split_huge_page() just after we
5073          *    entered this if-block, the thread must wait for page table lock
5074          *    to be unlocked in __split_huge_page_splitting(), where the main
5075          *    part of thp split is not executed yet.
5076          */
5077         if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
5078                 if (mc.precharge < HPAGE_PMD_NR) {
5079                         spin_unlock(ptl);
5080                         return 0;
5081                 }
5082                 target_type = get_mctgt_type_thp(vma, addr, *pmd, &target);
5083                 if (target_type == MC_TARGET_PAGE) {
5084                         page = target.page;
5085                         if (!isolate_lru_page(page)) {
5086                                 if (!mem_cgroup_move_account(page, HPAGE_PMD_NR,
5087                                                              mc.from, mc.to)) {
5088                                         mc.precharge -= HPAGE_PMD_NR;
5089                                         mc.moved_charge += HPAGE_PMD_NR;
5090                                 }
5091                                 putback_lru_page(page);
5092                         }
5093                         put_page(page);
5094                 }
5095                 spin_unlock(ptl);
5096                 return 0;
5097         }
5098 
5099         if (pmd_trans_unstable(pmd))
5100                 return 0;
5101 retry:
5102         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
5103         for (; addr != end; addr += PAGE_SIZE) {
5104                 pte_t ptent = *(pte++);
5105                 swp_entry_t ent;
5106 
5107                 if (!mc.precharge)
5108                         break;
5109 
5110                 switch (get_mctgt_type(vma, addr, ptent, &target)) {
5111                 case MC_TARGET_PAGE:
5112                         page = target.page;
5113                         if (isolate_lru_page(page))
5114                                 goto put;
5115                         if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) {
5116                                 mc.precharge--;
5117                                 /* we uncharge from mc.from later. */
5118                                 mc.moved_charge++;
5119                         }
5120                         putback_lru_page(page);
5121 put:                    /* get_mctgt_type() gets the page */
5122                         put_page(page);
5123                         break;
5124                 case MC_TARGET_SWAP:
5125                         ent = target.ent;
5126                         if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) {
5127                                 mc.precharge--;
5128                                 /* we fixup refcnts and charges later. */
5129                                 mc.moved_swap++;
5130                         }
5131                         break;
5132                 default:
5133                         break;
5134                 }
5135         }
5136         pte_unmap_unlock(pte - 1, ptl);
5137         cond_resched();
5138 
5139         if (addr != end) {
5140                 /*
5141                  * We have consumed all precharges we got in can_attach().
5142                  * We try charge one by one, but don't do any additional
5143                  * charges to mc.to if we have failed in charge once in attach()
5144                  * phase.
5145                  */
5146                 ret = mem_cgroup_do_precharge(1);
5147                 if (!ret)
5148                         goto retry;
5149         }
5150 
5151         return ret;
5152 }
5153 
5154 static void mem_cgroup_move_charge(struct mm_struct *mm)
5155 {
5156         struct mm_walk mem_cgroup_move_charge_walk = {
5157                 .pmd_entry = mem_cgroup_move_charge_pte_range,
5158                 .mm = mm,
5159         };
5160 
5161         lru_add_drain_all();
5162         /*
5163          * Signal mem_cgroup_begin_page_stat() to take the memcg's
5164          * move_lock while we're moving its pages to another memcg.
5165          * Then wait for already started RCU-only updates to finish.
5166          */
5167         atomic_inc(&mc.from->moving_account);
5168         synchronize_rcu();
5169 retry:
5170         if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
5171                 /*
5172                  * Someone who are holding the mmap_sem might be waiting in
5173                  * waitq. So we cancel all extra charges, wake up all waiters,
5174                  * and retry. Because we cancel precharges, we might not be able
5175                  * to move enough charges, but moving charge is a best-effort
5176                  * feature anyway, so it wouldn't be a big problem.
5177                  */
5178                 __mem_cgroup_clear_mc();
5179                 cond_resched();
5180                 goto retry;
5181         }
5182         /*
5183          * When we have consumed all precharges and failed in doing
5184          * additional charge, the page walk just aborts.
5185          */
5186         walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk);
5187         up_read(&mm->mmap_sem);
5188         atomic_dec(&mc.from->moving_account);
5189 }
5190 
5191 static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
5192                                  struct cgroup_taskset *tset)
5193 {
5194         struct task_struct *p = cgroup_taskset_first(tset);
5195         struct mm_struct *mm = get_task_mm(p);
5196 
5197         if (mm) {
5198                 if (mc.to)
5199                         mem_cgroup_move_charge(mm);
5200                 mmput(mm);
5201         }
5202         if (mc.to)
5203                 mem_cgroup_clear_mc();
5204 }
5205 #else   /* !CONFIG_MMU */
5206 static int mem_cgroup_can_attach(struct cgroup_subsys_state *css,
5207                                  struct cgroup_taskset *tset)
5208 {
5209         return 0;
5210 }
5211 static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css,
5212                                      struct cgroup_taskset *tset)
5213 {
5214 }
5215 static void mem_cgroup_move_task(struct cgroup_subsys_state *css,
5216                                  struct cgroup_taskset *tset)
5217 {
5218 }
5219 #endif
5220 
5221 /*
5222  * Cgroup retains root cgroups across [un]mount cycles making it necessary
5223  * to verify whether we're attached to the default hierarchy on each mount
5224  * attempt.
5225  */
5226 static void mem_cgroup_bind(struct cgroup_subsys_state *root_css)
5227 {
5228         /*
5229          * use_hierarchy is forced on the default hierarchy.  cgroup core
5230          * guarantees that @root doesn't have any children, so turning it
5231          * on for the root memcg is enough.
5232          */
5233         if (cgroup_on_dfl(root_css->cgroup))
5234                 root_mem_cgroup->use_hierarchy = true;
5235         else
5236                 root_mem_cgroup->use_hierarchy = false;
5237 }
5238 
5239 static u64 memory_current_read(struct cgroup_subsys_state *css,
5240                                struct cftype *cft)
5241 {
5242         return mem_cgroup_usage(mem_cgroup_from_css(css), false);
5243 }
5244 
5245 static int memory_low_show(struct seq_file *m, void *v)
5246 {
5247         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5248         unsigned long low = READ_ONCE(memcg->low);
5249 
5250         if (low == PAGE_COUNTER_MAX)
5251                 seq_puts(m, "max\n");
5252         else
5253                 seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE);
5254 
5255         return 0;
5256 }
5257 
5258 static ssize_t memory_low_write(struct kernfs_open_file *of,
5259                                 char *buf, size_t nbytes, loff_t off)
5260 {
5261         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5262         unsigned long low;
5263         int err;
5264 
5265         buf = strstrip(buf);
5266         err = page_counter_memparse(buf, "max", &low);
5267         if (err)
5268                 return err;
5269 
5270         memcg->low = low;
5271 
5272         return nbytes;
5273 }
5274 
5275 static int memory_high_show(struct seq_file *m, void *v)
5276 {
5277         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5278         unsigned long high = READ_ONCE(memcg->high);
5279 
5280         if (high == PAGE_COUNTER_MAX)
5281                 seq_puts(m, "max\n");
5282         else
5283                 seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE);
5284 
5285         return 0;
5286 }
5287 
5288 static ssize_t memory_high_write(struct kernfs_open_file *of,
5289                                  char *buf, size_t nbytes, loff_t off)
5290 {
5291         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5292         unsigned long nr_pages;
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         nr_pages = page_counter_read(&memcg->memory);
5304         if (nr_pages > high)
5305                 try_to_free_mem_cgroup_pages(memcg, nr_pages - high,
5306                                              GFP_KERNEL, true);
5307 
5308         return nbytes;
5309 }
5310 
5311 static int memory_max_show(struct seq_file *m, void *v)
5312 {
5313         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5314         unsigned long max = READ_ONCE(memcg->memory.limit);
5315 
5316         if (max == PAGE_COUNTER_MAX)
5317                 seq_puts(m, "max\n");
5318         else
5319                 seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE);
5320 
5321         return 0;
5322 }
5323 
5324 static ssize_t memory_max_write(struct kernfs_open_file *of,
5325                                 char *buf, size_t nbytes, loff_t off)
5326 {
5327         struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
5328         unsigned long max;
5329         int err;
5330 
5331         buf = strstrip(buf);
5332         err = page_counter_memparse(buf, "max", &max);
5333         if (err)
5334                 return err;
5335 
5336         err = mem_cgroup_resize_limit(memcg, max);
5337         if (err)
5338                 return err;
5339 
5340         return nbytes;
5341 }
5342 
5343 static int memory_events_show(struct seq_file *m, void *v)
5344 {
5345         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
5346 
5347         seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW));
5348         seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH));
5349         seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX));
5350         seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM));
5351 
5352         return 0;
5353 }
5354 
5355 static struct cftype memory_files[] = {
5356         {
5357                 .name = "current",
5358                 .read_u64 = memory_current_read,
5359         },
5360         {
5361                 .name = "low",
5362                 .flags = CFTYPE_NOT_ON_ROOT,
5363                 .seq_show = memory_low_show,
5364                 .write = memory_low_write,
5365         },
5366         {
5367                 .name = "high",
5368                 .flags = CFTYPE_NOT_ON_ROOT,
5369                 .seq_show = memory_high_show,
5370                 .write = memory_high_write,
5371         },
5372         {
5373                 .name = "max",
5374                 .flags = CFTYPE_NOT_ON_ROOT,
5375                 .seq_show = memory_max_show,
5376                 .write = memory_max_write,
5377         },
5378         {
5379                 .name = "events",
5380                 .flags = CFTYPE_NOT_ON_ROOT,
5381                 .seq_show = memory_events_show,
5382         },
5383         { }     /* terminate */
5384 };
5385 
5386 struct cgroup_subsys memory_cgrp_subsys = {
5387         .css_alloc = mem_cgroup_css_alloc,
5388         .css_online = mem_cgroup_css_online,
5389         .css_offline = mem_cgroup_css_offline,
5390         .css_free = mem_cgroup_css_free,
5391         .css_reset = mem_cgroup_css_reset,
5392         .can_attach = mem_cgroup_can_attach,
5393         .cancel_attach = mem_cgroup_cancel_attach,
5394         .attach = mem_cgroup_move_task,
5395         .bind = mem_cgroup_bind,
5396         .dfl_cftypes = memory_files,
5397         .legacy_cftypes = mem_cgroup_legacy_files,
5398         .early_init = 0,
5399 };
5400 
5401 /**
5402  * mem_cgroup_events - count memory events against a cgroup
5403  * @memcg: the memory cgroup
5404  * @idx: the event index
5405  * @nr: the number of events to account for
5406  */
5407 void mem_cgroup_events(struct mem_cgroup *memcg,
5408                        enum mem_cgroup_events_index idx,
5409                        unsigned int nr)
5410 {
5411         this_cpu_add(memcg->stat->events[idx], nr);
5412 }
5413 
5414 /**
5415  * mem_cgroup_low - check if memory consumption is below the normal range
5416  * @root: the highest ancestor to consider
5417  * @memcg: the memory cgroup to check
5418  *
5419  * Returns %true if memory consumption of @memcg, and that of all
5420  * configurable ancestors up to @root, is below the normal range.
5421  */
5422 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg)
5423 {
5424         if (mem_cgroup_disabled())
5425                 return false;
5426 
5427         /*
5428          * The toplevel group doesn't have a configurable range, so
5429          * it's never low when looked at directly, and it is not
5430          * considered an ancestor when assessing the hierarchy.
5431          */
5432 
5433         if (memcg == root_mem_cgroup)
5434                 return false;
5435 
5436         if (page_counter_read(&memcg->memory) >= memcg->low)
5437                 return false;
5438 
5439         while (memcg != root) {
5440                 memcg = parent_mem_cgroup(memcg);
5441 
5442                 if (memcg == root_mem_cgroup)
5443                         break;
5444 
5445                 if (page_counter_read(&memcg->memory) >= memcg->low)
5446                         return false;
5447         }
5448         return true;
5449 }
5450 
5451 /**
5452  * mem_cgroup_try_charge - try charging a page
5453  * @page: page to charge
5454  * @mm: mm context of the victim
5455  * @gfp_mask: reclaim mode
5456  * @memcgp: charged memcg return
5457  *
5458  * Try to charge @page to the memcg that @mm belongs to, reclaiming
5459  * pages according to @gfp_mask if necessary.
5460  *
5461  * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5462  * Otherwise, an error code is returned.
5463  *
5464  * After page->mapping has been set up, the caller must finalize the
5465  * charge with mem_cgroup_commit_charge().  Or abort the transaction
5466  * with mem_cgroup_cancel_charge() in case page instantiation fails.
5467  */
5468 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
5469                           gfp_t gfp_mask, struct mem_cgroup **memcgp)
5470 {
5471         struct mem_cgroup *memcg = NULL;
5472         unsigned int nr_pages = 1;
5473         int ret = 0;
5474 
5475         if (mem_cgroup_disabled())
5476                 goto out;
5477 
5478         if (PageSwapCache(page)) {
5479                 /*
5480                  * Every swap fault against a single page tries to charge the
5481                  * page, bail as early as possible.  shmem_unuse() encounters
5482                  * already charged pages, too.  The USED bit is protected by
5483                  * the page lock, which serializes swap cache removal, which
5484                  * in turn serializes uncharging.
5485                  */
5486                 if (page->mem_cgroup)
5487                         goto out;
5488         }
5489 
5490         if (PageTransHuge(page)) {
5491                 nr_pages <<= compound_order(page);
5492                 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5493         }
5494 
5495         if (do_swap_account && PageSwapCache(page))
5496                 memcg = try_get_mem_cgroup_from_page(page);
5497         if (!memcg)
5498                 memcg = get_mem_cgroup_from_mm(mm);
5499 
5500         ret = try_charge(memcg, gfp_mask, nr_pages);
5501 
5502         css_put(&memcg->css);
5503 
5504         if (ret == -EINTR) {
5505                 memcg = root_mem_cgroup;
5506                 ret = 0;
5507         }
5508 out:
5509         *memcgp = memcg;
5510         return ret;
5511 }
5512 
5513 /**
5514  * mem_cgroup_commit_charge - commit a page charge
5515  * @page: page to charge
5516  * @memcg: memcg to charge the page to
5517  * @lrucare: page might be on LRU already
5518  *
5519  * Finalize a charge transaction started by mem_cgroup_try_charge(),
5520  * after page->mapping has been set up.  This must happen atomically
5521  * as part of the page instantiation, i.e. under the page table lock
5522  * for anonymous pages, under the page lock for page and swap cache.
5523  *
5524  * In addition, the page must not be on the LRU during the commit, to
5525  * prevent racing with task migration.  If it might be, use @lrucare.
5526  *
5527  * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5528  */
5529 void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg,
5530                               bool lrucare)
5531 {
5532         unsigned int nr_pages = 1;
5533 
5534         VM_BUG_ON_PAGE(!page->mapping, page);
5535         VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page);
5536 
5537         if (mem_cgroup_disabled())
5538                 return;
5539         /*
5540          * Swap faults will attempt to charge the same page multiple
5541          * times.  But reuse_swap_page() might have removed the page
5542          * from swapcache already, so we can't check PageSwapCache().
5543          */
5544         if (!memcg)
5545                 return;
5546 
5547         commit_charge(page, memcg, lrucare);
5548 
5549         if (PageTransHuge(page)) {
5550                 nr_pages <<= compound_order(page);
5551                 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5552         }
5553 
5554         local_irq_disable();
5555         mem_cgroup_charge_statistics(memcg, page, nr_pages);
5556         memcg_check_events(memcg, page);
5557         local_irq_enable();
5558 
5559         if (do_swap_account && PageSwapCache(page)) {
5560                 swp_entry_t entry = { .val = page_private(page) };
5561                 /*
5562                  * The swap entry might not get freed for a long time,
5563                  * let's not wait for it.  The page already received a
5564                  * memory+swap charge, drop the swap entry duplicate.
5565                  */
5566                 mem_cgroup_uncharge_swap(entry);
5567         }
5568 }
5569 
5570 /**
5571  * mem_cgroup_cancel_charge - cancel a page charge
5572  * @page: page to charge
5573  * @memcg: memcg to charge the page to
5574  *
5575  * Cancel a charge transaction started by mem_cgroup_try_charge().
5576  */
5577 void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg)
5578 {
5579         unsigned int nr_pages = 1;
5580 
5581         if (mem_cgroup_disabled())
5582                 return;
5583         /*
5584          * Swap faults will attempt to charge the same page multiple
5585          * times.  But reuse_swap_page() might have removed the page
5586          * from swapcache already, so we can't check PageSwapCache().
5587          */
5588         if (!memcg)
5589                 return;
5590 
5591         if (PageTransHuge(page)) {
5592                 nr_pages <<= compound_order(page);
5593                 VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5594         }
5595 
5596         cancel_charge(memcg, nr_pages);
5597 }
5598 
5599 static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout,
5600                            unsigned long nr_anon, unsigned long nr_file,
5601                            unsigned long nr_huge, struct page *dummy_page)
5602 {
5603         unsigned long nr_pages = nr_anon + nr_file;
5604         unsigned long flags;
5605 
5606         if (!mem_cgroup_is_root(memcg)) {
5607                 page_counter_uncharge(&memcg->memory, nr_pages);
5608                 if (do_swap_account)
5609                         page_counter_uncharge(&memcg->memsw, nr_pages);
5610                 memcg_oom_recover(memcg);
5611         }
5612 
5613         local_irq_save(flags);
5614         __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon);
5615         __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file);
5616         __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge);
5617         __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout);
5618         __this_cpu_add(memcg->stat->nr_page_events, nr_pages);
5619         memcg_check_events(memcg, dummy_page);
5620         local_irq_restore(flags);
5621 
5622         if (!mem_cgroup_is_root(memcg))
5623                 css_put_many(&memcg->css, nr_pages);
5624 }
5625 
5626 static void uncharge_list(struct list_head *page_list)
5627 {
5628         struct mem_cgroup *memcg = NULL;
5629         unsigned long nr_anon = 0;
5630         unsigned long nr_file = 0;
5631         unsigned long nr_huge = 0;
5632         unsigned long pgpgout = 0;
5633         struct list_head *next;
5634         struct page *page;
5635 
5636         next = page_list->next;
5637         do {
5638                 unsigned int nr_pages = 1;
5639 
5640                 page = list_entry(next, struct page, lru);
5641                 next = page->lru.next;
5642 
5643                 VM_BUG_ON_PAGE(PageLRU(page), page);
5644                 VM_BUG_ON_PAGE(!PageHWPoison(page) && page_count(page), page);
5645 
5646                 if (!page->mem_cgroup)
5647                         continue;
5648 
5649                 /*
5650                  * Nobody should be changing or seriously looking at
5651                  * page->mem_cgroup at this point, we have fully
5652                  * exclusive access to the page.
5653                  */
5654 
5655                 if (memcg != page->mem_cgroup) {
5656                         if (memcg) {
5657                                 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5658                                                nr_huge, page);
5659                                 pgpgout = nr_anon = nr_file = nr_huge = 0;
5660                         }
5661                         memcg = page->mem_cgroup;
5662                 }
5663 
5664                 if (PageTransHuge(page)) {
5665                         nr_pages <<= compound_order(page);
5666                         VM_BUG_ON_PAGE(!PageTransHuge(page), page);
5667                         nr_huge += nr_pages;
5668                 }
5669 
5670                 if (PageAnon(page))
5671                         nr_anon += nr_pages;
5672                 else
5673                         nr_file += nr_pages;
5674 
5675                 page->mem_cgroup = NULL;
5676 
5677                 pgpgout++;
5678         } while (next != page_list);
5679 
5680         if (memcg)
5681                 uncharge_batch(memcg, pgpgout, nr_anon, nr_file,
5682                                nr_huge, page);
5683 }
5684 
5685 /**
5686  * mem_cgroup_uncharge - uncharge a page
5687  * @page: page to uncharge
5688  *
5689  * Uncharge a page previously charged with mem_cgroup_try_charge() and
5690  * mem_cgroup_commit_charge().
5691  */
5692 void mem_cgroup_uncharge(struct page *page)
5693 {
5694         if (mem_cgroup_disabled())
5695                 return;
5696 
5697         /* Don't touch page->lru of any random page, pre-check: */
5698         if (!page->mem_cgroup)
5699                 return;
5700 
5701         INIT_LIST_HEAD(&page->lru);
5702         uncharge_list(&page->lru);
5703 }
5704 
5705 /**
5706  * mem_cgroup_uncharge_list - uncharge a list of page
5707  * @page_list: list of pages to uncharge
5708  *
5709  * Uncharge a list of pages previously charged with
5710  * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5711  */
5712 void mem_cgroup_uncharge_list(struct list_head *page_list)
5713 {
5714         if (mem_cgroup_disabled())
5715                 return;
5716 
5717         if (!list_empty(page_list))
5718                 uncharge_list(page_list);
5719 }
5720 
5721 /**
5722  * mem_cgroup_migrate - migrate a charge to another page
5723  * @oldpage: currently charged page
5724  * @newpage: page to transfer the charge to
5725  * @lrucare: either or both pages might be on the LRU already
5726  *
5727  * Migrate the charge from @oldpage to @newpage.
5728  *
5729  * Both pages must be locked, @newpage->mapping must be set up.
5730  */
5731 void mem_cgroup_migrate(struct page *oldpage, struct page *newpage,
5732                         bool lrucare)
5733 {
5734         struct mem_cgroup *memcg;
5735         int isolated;
5736 
5737         VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage);
5738         VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
5739         VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage);
5740         VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage);
5741         VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage);
5742         VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage),
5743                        newpage);
5744 
5745         if (mem_cgroup_disabled())
5746                 return;
5747 
5748         /* Page cache replacement: new page already charged? */
5749         if (newpage->mem_cgroup)
5750                 return;
5751 
5752         /*
5753          * Swapcache readahead pages can get migrated before being
5754          * charged, and migration from compaction can happen to an
5755          * uncharged page when the PFN walker finds a page that
5756          * reclaim just put back on the LRU but has not released yet.
5757          */
5758         memcg = oldpage->mem_cgroup;
5759         if (!memcg)
5760                 return;
5761 
5762         if (lrucare)
5763                 lock_page_lru(oldpage, &isolated);
5764 
5765         oldpage->mem_cgroup = NULL;
5766 
5767         if (lrucare)
5768                 unlock_page_lru(oldpage, isolated);
5769 
5770         commit_charge(newpage, memcg, lrucare);
5771 }
5772 
5773 /*
5774  * subsys_initcall() for memory controller.
5775  *
5776  * Some parts like hotcpu_notifier() have to be initialized from this context
5777  * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5778  * everything that doesn't depend on a specific mem_cgroup structure should
5779  * be initialized from here.
5780  */
5781 static int __init mem_cgroup_init(void)
5782 {
5783         int cpu, node;
5784 
5785         hotcpu_notifier(memcg_cpu_hotplug_callback, 0);
5786 
5787         for_each_possible_cpu(cpu)
5788                 INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work,
5789                           drain_local_stock);
5790 
5791         for_each_node(node) {
5792                 struct mem_cgroup_tree_per_node *rtpn;
5793                 int zone;
5794 
5795                 rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
5796                                     node_online(node) ? node : NUMA_NO_NODE);
5797 
5798                 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
5799                         struct mem_cgroup_tree_per_zone *rtpz;
5800 
5801                         rtpz = &rtpn->rb_tree_per_zone[zone];
5802                         rtpz->rb_root = RB_ROOT;
5803                         spin_lock_init(&rtpz->lock);
5804                 }
5805                 soft_limit_tree.rb_tree_per_node[node] = rtpn;
5806         }
5807 
5808         return 0;
5809 }
5810 subsys_initcall(mem_cgroup_init);
5811 
5812 #ifdef CONFIG_MEMCG_SWAP
5813 /**
5814  * mem_cgroup_swapout - transfer a memsw charge to swap
5815  * @page: page whose memsw charge to transfer
5816  * @entry: swap entry to move the charge to
5817  *
5818  * Transfer the memsw charge of @page to @entry.
5819  */
5820 void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
5821 {
5822         struct mem_cgroup *memcg;
5823         unsigned short oldid;
5824 
5825         VM_BUG_ON_PAGE(PageLRU(page), page);
5826         VM_BUG_ON_PAGE(page_count(page), page);
5827 
5828         if (!do_swap_account)
5829                 return;
5830 
5831         memcg = page->mem_cgroup;
5832 
5833         /* Readahead page, never charged */
5834         if (!memcg)
5835                 return;
5836 
5837         oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg));
5838         VM_BUG_ON_PAGE(oldid, page);
5839         mem_cgroup_swap_statistics(memcg, true);
5840 
5841         page->mem_cgroup = NULL;
5842 
5843         if (!mem_cgroup_is_root(memcg))
5844                 page_counter_uncharge(&memcg->memory, 1);
5845 
5846         /* Caller disabled preemption with mapping->tree_lock */
5847         mem_cgroup_charge_statistics(memcg, page, -1);
5848         memcg_check_events(memcg, page);
5849 }
5850 
5851 /**
5852  * mem_cgroup_uncharge_swap - uncharge a swap entry
5853  * @entry: swap entry to uncharge
5854  *
5855  * Drop the memsw charge associated with @entry.
5856  */
5857 void mem_cgroup_uncharge_swap(swp_entry_t entry)
5858 {
5859         struct mem_cgroup *memcg;
5860         unsigned short id;
5861 
5862         if (!do_swap_account)
5863                 return;
5864 
5865         id = swap_cgroup_record(entry, 0);
5866         rcu_read_lock();
5867         memcg = mem_cgroup_from_id(id);
5868         if (memcg) {
5869                 if (!mem_cgroup_is_root(memcg))
5870                         page_counter_uncharge(&memcg->memsw, 1);
5871                 mem_cgroup_swap_statistics(memcg, false);
5872                 css_put(&memcg->css);
5873         }
5874         rcu_read_unlock();
5875 }
5876 
5877 /* for remember boot option*/
5878 #ifdef CONFIG_MEMCG_SWAP_ENABLED
5879 static int really_do_swap_account __initdata = 1;
5880 #else
5881 static int really_do_swap_account __initdata;
5882 #endif
5883 
5884 static int __init enable_swap_account(char *s)
5885 {
5886         if (!strcmp(s, "1"))
5887                 really_do_swap_account = 1;
5888         else if (!strcmp(s, ""))
5889                 really_do_swap_account = 0;
5890         return 1;
5891 }
5892 __setup("swapaccount=", enable_swap_account);
5893 
5894 static struct cftype memsw_cgroup_files[] = {
5895         {
5896                 .name = "memsw.usage_in_bytes",
5897                 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE),
5898                 .read_u64 = mem_cgroup_read_u64,
5899         },
5900         {
5901                 .name = "memsw.max_usage_in_bytes",
5902                 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE),
5903                 .write = mem_cgroup_reset,
5904                 .read_u64 = mem_cgroup_read_u64,
5905         },
5906         {
5907                 .name = "memsw.limit_in_bytes",
5908                 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT),
5909                 .write = mem_cgroup_write,
5910                 .read_u64 = mem_cgroup_read_u64,
5911         },
5912         {
5913                 .name = "memsw.failcnt",
5914                 .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT),
5915                 .write = mem_cgroup_reset,
5916                 .read_u64 = mem_cgroup_read_u64,
5917         },
5918         { },    /* terminate */
5919 };
5920 
5921 static int __init mem_cgroup_swap_init(void)
5922 {
5923         if (!mem_cgroup_disabled() && really_do_swap_account) {
5924                 do_swap_account = 1;
5925                 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys,
5926                                                   memsw_cgroup_files));
5927         }
5928         return 0;
5929 }
5930 subsys_initcall(mem_cgroup_swap_init);
5931 
5932 #endif /* CONFIG_MEMCG_SWAP */
5933 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp