Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
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 | * | |
8cdea7c0 BS |
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 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
4ffef5fe | 38 | #include <linux/hugetlb.h> |
d13d1443 | 39 | #include <linux/pagemap.h> |
d52aa412 | 40 | #include <linux/smp.h> |
8a9f3ccd | 41 | #include <linux/page-flags.h> |
66e1707b | 42 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
43 | #include <linux/bit_spinlock.h> |
44 | #include <linux/rcupdate.h> | |
e222432b | 45 | #include <linux/limits.h> |
b9e15baf | 46 | #include <linux/export.h> |
8c7c6e34 | 47 | #include <linux/mutex.h> |
bb4cc1a8 | 48 | #include <linux/rbtree.h> |
b6ac57d5 | 49 | #include <linux/slab.h> |
66e1707b | 50 | #include <linux/swap.h> |
02491447 | 51 | #include <linux/swapops.h> |
66e1707b | 52 | #include <linux/spinlock.h> |
2e72b634 | 53 | #include <linux/eventfd.h> |
79bd9814 | 54 | #include <linux/poll.h> |
2e72b634 | 55 | #include <linux/sort.h> |
66e1707b | 56 | #include <linux/fs.h> |
d2ceb9b7 | 57 | #include <linux/seq_file.h> |
70ddf637 | 58 | #include <linux/vmpressure.h> |
b69408e8 | 59 | #include <linux/mm_inline.h> |
5d1ea48b | 60 | #include <linux/swap_cgroup.h> |
cdec2e42 | 61 | #include <linux/cpu.h> |
158e0a2d | 62 | #include <linux/oom.h> |
0056f4e6 | 63 | #include <linux/lockdep.h> |
79bd9814 | 64 | #include <linux/file.h> |
08e552c6 | 65 | #include "internal.h" |
d1a4c0b3 | 66 | #include <net/sock.h> |
4bd2c1ee | 67 | #include <net/ip.h> |
d1a4c0b3 | 68 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 69 | #include "slab.h" |
8cdea7c0 | 70 | |
8697d331 BS |
71 | #include <asm/uaccess.h> |
72 | ||
cc8e970c KM |
73 | #include <trace/events/vmscan.h> |
74 | ||
073219e9 TH |
75 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
76 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 77 | |
a181b0e8 | 78 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 79 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
56161634 | 80 | struct cgroup_subsys_state *mem_cgroup_root_css __read_mostly; |
8cdea7c0 | 81 | |
21afa38e | 82 | /* Whether the swap controller is active */ |
c255a458 | 83 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 84 | int do_swap_account __read_mostly; |
c077719b | 85 | #else |
a0db00fc | 86 | #define do_swap_account 0 |
c077719b KH |
87 | #endif |
88 | ||
af7c4b0e JW |
89 | static const char * const mem_cgroup_stat_names[] = { |
90 | "cache", | |
91 | "rss", | |
b070e65c | 92 | "rss_huge", |
af7c4b0e | 93 | "mapped_file", |
c4843a75 | 94 | "dirty", |
3ea67d06 | 95 | "writeback", |
af7c4b0e JW |
96 | "swap", |
97 | }; | |
98 | ||
af7c4b0e JW |
99 | static const char * const mem_cgroup_events_names[] = { |
100 | "pgpgin", | |
101 | "pgpgout", | |
102 | "pgfault", | |
103 | "pgmajfault", | |
104 | }; | |
105 | ||
58cf188e SZ |
106 | static const char * const mem_cgroup_lru_names[] = { |
107 | "inactive_anon", | |
108 | "active_anon", | |
109 | "inactive_file", | |
110 | "active_file", | |
111 | "unevictable", | |
112 | }; | |
113 | ||
7a159cc9 JW |
114 | /* |
115 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
116 | * it will be incremated by the number of pages. This counter is used for | |
117 | * for trigger some periodic events. This is straightforward and better | |
118 | * than using jiffies etc. to handle periodic memcg event. | |
119 | */ | |
120 | enum mem_cgroup_events_target { | |
121 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 122 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 123 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
124 | MEM_CGROUP_NTARGETS, |
125 | }; | |
a0db00fc KS |
126 | #define THRESHOLDS_EVENTS_TARGET 128 |
127 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
128 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 129 | |
d52aa412 | 130 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 131 | long count[MEM_CGROUP_STAT_NSTATS]; |
241994ed | 132 | unsigned long events[MEMCG_NR_EVENTS]; |
13114716 | 133 | unsigned long nr_page_events; |
7a159cc9 | 134 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
135 | }; |
136 | ||
5ac8fb31 JW |
137 | struct reclaim_iter { |
138 | struct mem_cgroup *position; | |
527a5ec9 JW |
139 | /* scan generation, increased every round-trip */ |
140 | unsigned int generation; | |
141 | }; | |
142 | ||
6d12e2d8 KH |
143 | /* |
144 | * per-zone information in memory controller. | |
145 | */ | |
6d12e2d8 | 146 | struct mem_cgroup_per_zone { |
6290df54 | 147 | struct lruvec lruvec; |
1eb49272 | 148 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 149 | |
5ac8fb31 | 150 | struct reclaim_iter iter[DEF_PRIORITY + 1]; |
527a5ec9 | 151 | |
bb4cc1a8 | 152 | struct rb_node tree_node; /* RB tree node */ |
3e32cb2e | 153 | unsigned long usage_in_excess;/* Set to the value by which */ |
bb4cc1a8 AM |
154 | /* the soft limit is exceeded*/ |
155 | bool on_tree; | |
d79154bb | 156 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 157 | /* use container_of */ |
6d12e2d8 | 158 | }; |
6d12e2d8 KH |
159 | |
160 | struct mem_cgroup_per_node { | |
161 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
162 | }; | |
163 | ||
bb4cc1a8 AM |
164 | /* |
165 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
166 | * their hierarchy representation | |
167 | */ | |
168 | ||
169 | struct mem_cgroup_tree_per_zone { | |
170 | struct rb_root rb_root; | |
171 | spinlock_t lock; | |
172 | }; | |
173 | ||
174 | struct mem_cgroup_tree_per_node { | |
175 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
176 | }; | |
177 | ||
178 | struct mem_cgroup_tree { | |
179 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
180 | }; | |
181 | ||
182 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
183 | ||
2e72b634 KS |
184 | struct mem_cgroup_threshold { |
185 | struct eventfd_ctx *eventfd; | |
3e32cb2e | 186 | unsigned long threshold; |
2e72b634 KS |
187 | }; |
188 | ||
9490ff27 | 189 | /* For threshold */ |
2e72b634 | 190 | struct mem_cgroup_threshold_ary { |
748dad36 | 191 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 192 | int current_threshold; |
2e72b634 KS |
193 | /* Size of entries[] */ |
194 | unsigned int size; | |
195 | /* Array of thresholds */ | |
196 | struct mem_cgroup_threshold entries[0]; | |
197 | }; | |
2c488db2 KS |
198 | |
199 | struct mem_cgroup_thresholds { | |
200 | /* Primary thresholds array */ | |
201 | struct mem_cgroup_threshold_ary *primary; | |
202 | /* | |
203 | * Spare threshold array. | |
204 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
205 | * It must be able to store at least primary->size - 1 entries. | |
206 | */ | |
207 | struct mem_cgroup_threshold_ary *spare; | |
208 | }; | |
209 | ||
9490ff27 KH |
210 | /* for OOM */ |
211 | struct mem_cgroup_eventfd_list { | |
212 | struct list_head list; | |
213 | struct eventfd_ctx *eventfd; | |
214 | }; | |
2e72b634 | 215 | |
79bd9814 TH |
216 | /* |
217 | * cgroup_event represents events which userspace want to receive. | |
218 | */ | |
3bc942f3 | 219 | struct mem_cgroup_event { |
79bd9814 | 220 | /* |
59b6f873 | 221 | * memcg which the event belongs to. |
79bd9814 | 222 | */ |
59b6f873 | 223 | struct mem_cgroup *memcg; |
79bd9814 TH |
224 | /* |
225 | * eventfd to signal userspace about the event. | |
226 | */ | |
227 | struct eventfd_ctx *eventfd; | |
228 | /* | |
229 | * Each of these stored in a list by the cgroup. | |
230 | */ | |
231 | struct list_head list; | |
fba94807 TH |
232 | /* |
233 | * register_event() callback will be used to add new userspace | |
234 | * waiter for changes related to this event. Use eventfd_signal() | |
235 | * on eventfd to send notification to userspace. | |
236 | */ | |
59b6f873 | 237 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 238 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
239 | /* |
240 | * unregister_event() callback will be called when userspace closes | |
241 | * the eventfd or on cgroup removing. This callback must be set, | |
242 | * if you want provide notification functionality. | |
243 | */ | |
59b6f873 | 244 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 245 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
246 | /* |
247 | * All fields below needed to unregister event when | |
248 | * userspace closes eventfd. | |
249 | */ | |
250 | poll_table pt; | |
251 | wait_queue_head_t *wqh; | |
252 | wait_queue_t wait; | |
253 | struct work_struct remove; | |
254 | }; | |
255 | ||
c0ff4b85 R |
256 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
257 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 258 | |
8cdea7c0 BS |
259 | /* |
260 | * The memory controller data structure. The memory controller controls both | |
261 | * page cache and RSS per cgroup. We would eventually like to provide | |
262 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
263 | * to help the administrator determine what knobs to tune. | |
8cdea7c0 BS |
264 | */ |
265 | struct mem_cgroup { | |
266 | struct cgroup_subsys_state css; | |
3e32cb2e JW |
267 | |
268 | /* Accounted resources */ | |
269 | struct page_counter memory; | |
270 | struct page_counter memsw; | |
271 | struct page_counter kmem; | |
272 | ||
241994ed JW |
273 | /* Normal memory consumption range */ |
274 | unsigned long low; | |
275 | unsigned long high; | |
276 | ||
3e32cb2e | 277 | unsigned long soft_limit; |
59927fb9 | 278 | |
70ddf637 AV |
279 | /* vmpressure notifications */ |
280 | struct vmpressure vmpressure; | |
281 | ||
2f7dd7a4 JW |
282 | /* css_online() has been completed */ |
283 | int initialized; | |
284 | ||
18f59ea7 BS |
285 | /* |
286 | * Should the accounting and control be hierarchical, per subtree? | |
287 | */ | |
288 | bool use_hierarchy; | |
79dfdacc | 289 | |
c2b42d3c | 290 | /* protected by memcg_oom_lock */ |
79dfdacc | 291 | bool oom_lock; |
c2b42d3c | 292 | int under_oom; |
79dfdacc | 293 | |
1f4c025b | 294 | int swappiness; |
3c11ecf4 KH |
295 | /* OOM-Killer disable */ |
296 | int oom_kill_disable; | |
a7885eb8 | 297 | |
2e72b634 KS |
298 | /* protect arrays of thresholds */ |
299 | struct mutex thresholds_lock; | |
300 | ||
301 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 302 | struct mem_cgroup_thresholds thresholds; |
907860ed | 303 | |
2e72b634 | 304 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 305 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 306 | |
9490ff27 KH |
307 | /* For oom notifier event fd */ |
308 | struct list_head oom_notify; | |
185efc0f | 309 | |
7dc74be0 DN |
310 | /* |
311 | * Should we move charges of a task when a task is moved into this | |
312 | * mem_cgroup ? And what type of charges should we move ? | |
313 | */ | |
f894ffa8 | 314 | unsigned long move_charge_at_immigrate; |
619d094b KH |
315 | /* |
316 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
317 | */ | |
6de22619 | 318 | atomic_t moving_account; |
312734c0 | 319 | /* taken only while moving_account > 0 */ |
6de22619 JW |
320 | spinlock_t move_lock; |
321 | struct task_struct *move_lock_task; | |
322 | unsigned long move_lock_flags; | |
d52aa412 | 323 | /* |
c62b1a3b | 324 | * percpu counter. |
d52aa412 | 325 | */ |
3a7951b4 | 326 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c | 327 | spinlock_t pcp_counter_lock; |
d1a4c0b3 | 328 | |
4bd2c1ee | 329 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 330 | struct cg_proto tcp_mem; |
d1a4c0b3 | 331 | #endif |
2633d7a0 | 332 | #if defined(CONFIG_MEMCG_KMEM) |
f7ce3190 | 333 | /* Index in the kmem_cache->memcg_params.memcg_caches array */ |
2633d7a0 | 334 | int kmemcg_id; |
2788cf0c | 335 | bool kmem_acct_activated; |
2a4db7eb | 336 | bool kmem_acct_active; |
2633d7a0 | 337 | #endif |
45cf7ebd GC |
338 | |
339 | int last_scanned_node; | |
340 | #if MAX_NUMNODES > 1 | |
341 | nodemask_t scan_nodes; | |
342 | atomic_t numainfo_events; | |
343 | atomic_t numainfo_updating; | |
344 | #endif | |
70ddf637 | 345 | |
52ebea74 TH |
346 | #ifdef CONFIG_CGROUP_WRITEBACK |
347 | struct list_head cgwb_list; | |
841710aa | 348 | struct wb_domain cgwb_domain; |
52ebea74 TH |
349 | #endif |
350 | ||
fba94807 TH |
351 | /* List of events which userspace want to receive */ |
352 | struct list_head event_list; | |
353 | spinlock_t event_list_lock; | |
354 | ||
54f72fe0 JW |
355 | struct mem_cgroup_per_node *nodeinfo[0]; |
356 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
357 | }; |
358 | ||
510fc4e1 | 359 | #ifdef CONFIG_MEMCG_KMEM |
cb731d6c | 360 | bool memcg_kmem_is_active(struct mem_cgroup *memcg) |
7de37682 | 361 | { |
2a4db7eb | 362 | return memcg->kmem_acct_active; |
7de37682 | 363 | } |
510fc4e1 GC |
364 | #endif |
365 | ||
7dc74be0 DN |
366 | /* Stuffs for move charges at task migration. */ |
367 | /* | |
1dfab5ab | 368 | * Types of charges to be moved. |
7dc74be0 | 369 | */ |
1dfab5ab JW |
370 | #define MOVE_ANON 0x1U |
371 | #define MOVE_FILE 0x2U | |
372 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 373 | |
4ffef5fe DN |
374 | /* "mc" and its members are protected by cgroup_mutex */ |
375 | static struct move_charge_struct { | |
b1dd693e | 376 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
377 | struct mem_cgroup *from; |
378 | struct mem_cgroup *to; | |
1dfab5ab | 379 | unsigned long flags; |
4ffef5fe | 380 | unsigned long precharge; |
854ffa8d | 381 | unsigned long moved_charge; |
483c30b5 | 382 | unsigned long moved_swap; |
8033b97c DN |
383 | struct task_struct *moving_task; /* a task moving charges */ |
384 | wait_queue_head_t waitq; /* a waitq for other context */ | |
385 | } mc = { | |
2bd9bb20 | 386 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
387 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
388 | }; | |
4ffef5fe | 389 | |
4e416953 BS |
390 | /* |
391 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
392 | * limit reclaim to prevent infinite loops, if they ever occur. | |
393 | */ | |
a0db00fc | 394 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 395 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 396 | |
217bc319 KH |
397 | enum charge_type { |
398 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 399 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 400 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 401 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
402 | NR_CHARGE_TYPE, |
403 | }; | |
404 | ||
8c7c6e34 | 405 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
406 | enum res_type { |
407 | _MEM, | |
408 | _MEMSWAP, | |
409 | _OOM_TYPE, | |
510fc4e1 | 410 | _KMEM, |
86ae53e1 GC |
411 | }; |
412 | ||
a0db00fc KS |
413 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
414 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 415 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
416 | /* Used for OOM nofiier */ |
417 | #define OOM_CONTROL (0) | |
8c7c6e34 | 418 | |
0999821b GC |
419 | /* |
420 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
421 | * As a consequence, any change that needs to protect against new child cgroups | |
422 | * appearing has to hold it as well. | |
423 | */ | |
424 | static DEFINE_MUTEX(memcg_create_mutex); | |
425 | ||
b2145145 WL |
426 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
427 | { | |
a7c6d554 | 428 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
429 | } |
430 | ||
70ddf637 AV |
431 | /* Some nice accessors for the vmpressure. */ |
432 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
433 | { | |
434 | if (!memcg) | |
435 | memcg = root_mem_cgroup; | |
436 | return &memcg->vmpressure; | |
437 | } | |
438 | ||
439 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
440 | { | |
441 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
442 | } | |
443 | ||
7ffc0edc MH |
444 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
445 | { | |
446 | return (memcg == root_mem_cgroup); | |
447 | } | |
448 | ||
4219b2da LZ |
449 | /* |
450 | * We restrict the id in the range of [1, 65535], so it can fit into | |
451 | * an unsigned short. | |
452 | */ | |
453 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
454 | ||
34c00c31 LZ |
455 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
456 | { | |
15a4c835 | 457 | return memcg->css.id; |
34c00c31 LZ |
458 | } |
459 | ||
adbe427b VD |
460 | /* |
461 | * A helper function to get mem_cgroup from ID. must be called under | |
462 | * rcu_read_lock(). The caller is responsible for calling | |
463 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
464 | * refcnt from swap can be called against removed memcg.) | |
465 | */ | |
34c00c31 LZ |
466 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
467 | { | |
468 | struct cgroup_subsys_state *css; | |
469 | ||
7d699ddb | 470 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
471 | return mem_cgroup_from_css(css); |
472 | } | |
473 | ||
e1aab161 | 474 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 475 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 476 | |
e1aab161 GC |
477 | void sock_update_memcg(struct sock *sk) |
478 | { | |
376be5ff | 479 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 480 | struct mem_cgroup *memcg; |
3f134619 | 481 | struct cg_proto *cg_proto; |
e1aab161 GC |
482 | |
483 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
484 | ||
f3f511e1 GC |
485 | /* Socket cloning can throw us here with sk_cgrp already |
486 | * filled. It won't however, necessarily happen from | |
487 | * process context. So the test for root memcg given | |
488 | * the current task's memcg won't help us in this case. | |
489 | * | |
490 | * Respecting the original socket's memcg is a better | |
491 | * decision in this case. | |
492 | */ | |
493 | if (sk->sk_cgrp) { | |
494 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 495 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
496 | return; |
497 | } | |
498 | ||
e1aab161 GC |
499 | rcu_read_lock(); |
500 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 501 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 502 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
503 | memcg_proto_active(cg_proto) && |
504 | css_tryget_online(&memcg->css)) { | |
3f134619 | 505 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
506 | } |
507 | rcu_read_unlock(); | |
508 | } | |
509 | } | |
510 | EXPORT_SYMBOL(sock_update_memcg); | |
511 | ||
512 | void sock_release_memcg(struct sock *sk) | |
513 | { | |
376be5ff | 514 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
515 | struct mem_cgroup *memcg; |
516 | WARN_ON(!sk->sk_cgrp->memcg); | |
517 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 518 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
519 | } |
520 | } | |
d1a4c0b3 GC |
521 | |
522 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
523 | { | |
524 | if (!memcg || mem_cgroup_is_root(memcg)) | |
525 | return NULL; | |
526 | ||
2e685cad | 527 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
528 | } |
529 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 530 | |
3f134619 GC |
531 | #endif |
532 | ||
a8964b9b | 533 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 534 | /* |
f7ce3190 | 535 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
536 | * The main reason for not using cgroup id for this: |
537 | * this works better in sparse environments, where we have a lot of memcgs, | |
538 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
539 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
540 | * 200 entry array for that. | |
55007d84 | 541 | * |
dbcf73e2 VD |
542 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
543 | * will double each time we have to increase it. | |
55007d84 | 544 | */ |
dbcf73e2 VD |
545 | static DEFINE_IDA(memcg_cache_ida); |
546 | int memcg_nr_cache_ids; | |
749c5415 | 547 | |
05257a1a VD |
548 | /* Protects memcg_nr_cache_ids */ |
549 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
550 | ||
551 | void memcg_get_cache_ids(void) | |
552 | { | |
553 | down_read(&memcg_cache_ids_sem); | |
554 | } | |
555 | ||
556 | void memcg_put_cache_ids(void) | |
557 | { | |
558 | up_read(&memcg_cache_ids_sem); | |
559 | } | |
560 | ||
55007d84 GC |
561 | /* |
562 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
563 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
564 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
565 | * tunable, but that is strictly not necessary. | |
566 | * | |
b8627835 | 567 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
568 | * this constant directly from cgroup, but it is understandable that this is |
569 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 570 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
571 | * increase ours as well if it increases. |
572 | */ | |
573 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 574 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 575 | |
d7f25f8a GC |
576 | /* |
577 | * A lot of the calls to the cache allocation functions are expected to be | |
578 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
579 | * conditional to this static branch, we'll have to allow modules that does | |
580 | * kmem_cache_alloc and the such to see this symbol as well | |
581 | */ | |
a8964b9b | 582 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 583 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 584 | |
a8964b9b GC |
585 | #endif /* CONFIG_MEMCG_KMEM */ |
586 | ||
f64c3f54 | 587 | static struct mem_cgroup_per_zone * |
e231875b | 588 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 589 | { |
e231875b JZ |
590 | int nid = zone_to_nid(zone); |
591 | int zid = zone_idx(zone); | |
592 | ||
54f72fe0 | 593 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
594 | } |
595 | ||
c0ff4b85 | 596 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 597 | { |
c0ff4b85 | 598 | return &memcg->css; |
d324236b WF |
599 | } |
600 | ||
ad7fa852 TH |
601 | /** |
602 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
603 | * @page: page of interest | |
604 | * | |
605 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
606 | * with @page is returned. The returned css remains associated with @page | |
607 | * until it is released. | |
608 | * | |
609 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
610 | * is returned. | |
611 | * | |
612 | * XXX: The above description of behavior on the default hierarchy isn't | |
613 | * strictly true yet as replace_page_cache_page() can modify the | |
614 | * association before @page is released even on the default hierarchy; | |
615 | * however, the current and planned usages don't mix the the two functions | |
616 | * and replace_page_cache_page() will soon be updated to make the invariant | |
617 | * actually true. | |
618 | */ | |
619 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
620 | { | |
621 | struct mem_cgroup *memcg; | |
622 | ||
623 | rcu_read_lock(); | |
624 | ||
625 | memcg = page->mem_cgroup; | |
626 | ||
627 | if (!memcg || !cgroup_on_dfl(memcg->css.cgroup)) | |
628 | memcg = root_mem_cgroup; | |
629 | ||
630 | rcu_read_unlock(); | |
631 | return &memcg->css; | |
632 | } | |
633 | ||
f64c3f54 | 634 | static struct mem_cgroup_per_zone * |
e231875b | 635 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 636 | { |
97a6c37b JW |
637 | int nid = page_to_nid(page); |
638 | int zid = page_zonenum(page); | |
f64c3f54 | 639 | |
e231875b | 640 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
641 | } |
642 | ||
bb4cc1a8 AM |
643 | static struct mem_cgroup_tree_per_zone * |
644 | soft_limit_tree_node_zone(int nid, int zid) | |
645 | { | |
646 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
647 | } | |
648 | ||
649 | static struct mem_cgroup_tree_per_zone * | |
650 | soft_limit_tree_from_page(struct page *page) | |
651 | { | |
652 | int nid = page_to_nid(page); | |
653 | int zid = page_zonenum(page); | |
654 | ||
655 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
656 | } | |
657 | ||
cf2c8127 JW |
658 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
659 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 660 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
661 | { |
662 | struct rb_node **p = &mctz->rb_root.rb_node; | |
663 | struct rb_node *parent = NULL; | |
664 | struct mem_cgroup_per_zone *mz_node; | |
665 | ||
666 | if (mz->on_tree) | |
667 | return; | |
668 | ||
669 | mz->usage_in_excess = new_usage_in_excess; | |
670 | if (!mz->usage_in_excess) | |
671 | return; | |
672 | while (*p) { | |
673 | parent = *p; | |
674 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
675 | tree_node); | |
676 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
677 | p = &(*p)->rb_left; | |
678 | /* | |
679 | * We can't avoid mem cgroups that are over their soft | |
680 | * limit by the same amount | |
681 | */ | |
682 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
683 | p = &(*p)->rb_right; | |
684 | } | |
685 | rb_link_node(&mz->tree_node, parent, p); | |
686 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
687 | mz->on_tree = true; | |
688 | } | |
689 | ||
cf2c8127 JW |
690 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
691 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
692 | { |
693 | if (!mz->on_tree) | |
694 | return; | |
695 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
696 | mz->on_tree = false; | |
697 | } | |
698 | ||
cf2c8127 JW |
699 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
700 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 701 | { |
0a31bc97 JW |
702 | unsigned long flags; |
703 | ||
704 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 705 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 706 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
707 | } |
708 | ||
3e32cb2e JW |
709 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
710 | { | |
711 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 712 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
713 | unsigned long excess = 0; |
714 | ||
715 | if (nr_pages > soft_limit) | |
716 | excess = nr_pages - soft_limit; | |
717 | ||
718 | return excess; | |
719 | } | |
bb4cc1a8 AM |
720 | |
721 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
722 | { | |
3e32cb2e | 723 | unsigned long excess; |
bb4cc1a8 AM |
724 | struct mem_cgroup_per_zone *mz; |
725 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 726 | |
e231875b | 727 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
728 | /* |
729 | * Necessary to update all ancestors when hierarchy is used. | |
730 | * because their event counter is not touched. | |
731 | */ | |
732 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 733 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 734 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
735 | /* |
736 | * We have to update the tree if mz is on RB-tree or | |
737 | * mem is over its softlimit. | |
738 | */ | |
739 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
740 | unsigned long flags; |
741 | ||
742 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
743 | /* if on-tree, remove it */ |
744 | if (mz->on_tree) | |
cf2c8127 | 745 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
746 | /* |
747 | * Insert again. mz->usage_in_excess will be updated. | |
748 | * If excess is 0, no tree ops. | |
749 | */ | |
cf2c8127 | 750 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 751 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
752 | } |
753 | } | |
754 | } | |
755 | ||
756 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
757 | { | |
bb4cc1a8 | 758 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
759 | struct mem_cgroup_per_zone *mz; |
760 | int nid, zid; | |
bb4cc1a8 | 761 | |
e231875b JZ |
762 | for_each_node(nid) { |
763 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
764 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
765 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 766 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
767 | } |
768 | } | |
769 | } | |
770 | ||
771 | static struct mem_cgroup_per_zone * | |
772 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
773 | { | |
774 | struct rb_node *rightmost = NULL; | |
775 | struct mem_cgroup_per_zone *mz; | |
776 | ||
777 | retry: | |
778 | mz = NULL; | |
779 | rightmost = rb_last(&mctz->rb_root); | |
780 | if (!rightmost) | |
781 | goto done; /* Nothing to reclaim from */ | |
782 | ||
783 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
784 | /* | |
785 | * Remove the node now but someone else can add it back, | |
786 | * we will to add it back at the end of reclaim to its correct | |
787 | * position in the tree. | |
788 | */ | |
cf2c8127 | 789 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 790 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 791 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
792 | goto retry; |
793 | done: | |
794 | return mz; | |
795 | } | |
796 | ||
797 | static struct mem_cgroup_per_zone * | |
798 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
799 | { | |
800 | struct mem_cgroup_per_zone *mz; | |
801 | ||
0a31bc97 | 802 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 803 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 804 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
805 | return mz; |
806 | } | |
807 | ||
711d3d2c KH |
808 | /* |
809 | * Implementation Note: reading percpu statistics for memcg. | |
810 | * | |
811 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
812 | * synchronization to implement "quick" read. There are trade-off between | |
813 | * reading cost and precision of value. Then, we may have a chance to implement | |
814 | * a periodic synchronizion of counter in memcg's counter. | |
815 | * | |
816 | * But this _read() function is used for user interface now. The user accounts | |
817 | * memory usage by memory cgroup and he _always_ requires exact value because | |
818 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
819 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
820 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
821 | * | |
822 | * If there are kernel internal actions which can make use of some not-exact | |
823 | * value, and reading all cpu value can be performance bottleneck in some | |
824 | * common workload, threashold and synchonization as vmstat[] should be | |
825 | * implemented. | |
826 | */ | |
c0ff4b85 | 827 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 828 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 829 | { |
7a159cc9 | 830 | long val = 0; |
c62b1a3b | 831 | int cpu; |
c62b1a3b | 832 | |
733a572e | 833 | for_each_possible_cpu(cpu) |
c0ff4b85 | 834 | val += per_cpu(memcg->stat->count[idx], cpu); |
c62b1a3b KH |
835 | return val; |
836 | } | |
837 | ||
c0ff4b85 | 838 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
839 | enum mem_cgroup_events_index idx) |
840 | { | |
841 | unsigned long val = 0; | |
842 | int cpu; | |
843 | ||
733a572e | 844 | for_each_possible_cpu(cpu) |
c0ff4b85 | 845 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
846 | return val; |
847 | } | |
848 | ||
c0ff4b85 | 849 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 850 | struct page *page, |
0a31bc97 | 851 | int nr_pages) |
d52aa412 | 852 | { |
b2402857 KH |
853 | /* |
854 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
855 | * counted as CACHE even if it's on ANON LRU. | |
856 | */ | |
0a31bc97 | 857 | if (PageAnon(page)) |
b2402857 | 858 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 859 | nr_pages); |
d52aa412 | 860 | else |
b2402857 | 861 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 862 | nr_pages); |
55e462b0 | 863 | |
b070e65c DR |
864 | if (PageTransHuge(page)) |
865 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
866 | nr_pages); | |
867 | ||
e401f176 KH |
868 | /* pagein of a big page is an event. So, ignore page size */ |
869 | if (nr_pages > 0) | |
c0ff4b85 | 870 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 871 | else { |
c0ff4b85 | 872 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
873 | nr_pages = -nr_pages; /* for event */ |
874 | } | |
e401f176 | 875 | |
13114716 | 876 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
877 | } |
878 | ||
e231875b | 879 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
880 | { |
881 | struct mem_cgroup_per_zone *mz; | |
882 | ||
883 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
884 | return mz->lru_size[lru]; | |
885 | } | |
886 | ||
e231875b JZ |
887 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
888 | int nid, | |
889 | unsigned int lru_mask) | |
bb2a0de9 | 890 | { |
e231875b | 891 | unsigned long nr = 0; |
889976db YH |
892 | int zid; |
893 | ||
e231875b | 894 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 895 | |
e231875b JZ |
896 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
897 | struct mem_cgroup_per_zone *mz; | |
898 | enum lru_list lru; | |
899 | ||
900 | for_each_lru(lru) { | |
901 | if (!(BIT(lru) & lru_mask)) | |
902 | continue; | |
903 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
904 | nr += mz->lru_size[lru]; | |
905 | } | |
906 | } | |
907 | return nr; | |
889976db | 908 | } |
bb2a0de9 | 909 | |
c0ff4b85 | 910 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 911 | unsigned int lru_mask) |
6d12e2d8 | 912 | { |
e231875b | 913 | unsigned long nr = 0; |
889976db | 914 | int nid; |
6d12e2d8 | 915 | |
31aaea4a | 916 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
917 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
918 | return nr; | |
d52aa412 KH |
919 | } |
920 | ||
f53d7ce3 JW |
921 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
922 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
923 | { |
924 | unsigned long val, next; | |
925 | ||
13114716 | 926 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 927 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 928 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
929 | if ((long)next - (long)val < 0) { |
930 | switch (target) { | |
931 | case MEM_CGROUP_TARGET_THRESH: | |
932 | next = val + THRESHOLDS_EVENTS_TARGET; | |
933 | break; | |
bb4cc1a8 AM |
934 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
935 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
936 | break; | |
f53d7ce3 JW |
937 | case MEM_CGROUP_TARGET_NUMAINFO: |
938 | next = val + NUMAINFO_EVENTS_TARGET; | |
939 | break; | |
940 | default: | |
941 | break; | |
942 | } | |
943 | __this_cpu_write(memcg->stat->targets[target], next); | |
944 | return true; | |
7a159cc9 | 945 | } |
f53d7ce3 | 946 | return false; |
d2265e6f KH |
947 | } |
948 | ||
949 | /* | |
950 | * Check events in order. | |
951 | * | |
952 | */ | |
c0ff4b85 | 953 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
954 | { |
955 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
956 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
957 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 958 | bool do_softlimit; |
82b3f2a7 | 959 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 960 | |
bb4cc1a8 AM |
961 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
962 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
963 | #if MAX_NUMNODES > 1 |
964 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
965 | MEM_CGROUP_TARGET_NUMAINFO); | |
966 | #endif | |
c0ff4b85 | 967 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
968 | if (unlikely(do_softlimit)) |
969 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 970 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 971 | if (unlikely(do_numainfo)) |
c0ff4b85 | 972 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 973 | #endif |
0a31bc97 | 974 | } |
d2265e6f KH |
975 | } |
976 | ||
cf475ad2 | 977 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 978 | { |
31a78f23 BS |
979 | /* |
980 | * mm_update_next_owner() may clear mm->owner to NULL | |
981 | * if it races with swapoff, page migration, etc. | |
982 | * So this can be called with p == NULL. | |
983 | */ | |
984 | if (unlikely(!p)) | |
985 | return NULL; | |
986 | ||
073219e9 | 987 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
988 | } |
989 | ||
df381975 | 990 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 991 | { |
c0ff4b85 | 992 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 993 | |
54595fe2 KH |
994 | rcu_read_lock(); |
995 | do { | |
6f6acb00 MH |
996 | /* |
997 | * Page cache insertions can happen withou an | |
998 | * actual mm context, e.g. during disk probing | |
999 | * on boot, loopback IO, acct() writes etc. | |
1000 | */ | |
1001 | if (unlikely(!mm)) | |
df381975 | 1002 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1003 | else { |
1004 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1005 | if (unlikely(!memcg)) | |
1006 | memcg = root_mem_cgroup; | |
1007 | } | |
ec903c0c | 1008 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1009 | rcu_read_unlock(); |
c0ff4b85 | 1010 | return memcg; |
54595fe2 KH |
1011 | } |
1012 | ||
5660048c JW |
1013 | /** |
1014 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1015 | * @root: hierarchy root | |
1016 | * @prev: previously returned memcg, NULL on first invocation | |
1017 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1018 | * | |
1019 | * Returns references to children of the hierarchy below @root, or | |
1020 | * @root itself, or %NULL after a full round-trip. | |
1021 | * | |
1022 | * Caller must pass the return value in @prev on subsequent | |
1023 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1024 | * to cancel a hierarchy walk before the round-trip is complete. | |
1025 | * | |
1026 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1027 | * divide up the memcgs in the hierarchy among all concurrent | |
1028 | * reclaimers operating on the same zone and priority. | |
1029 | */ | |
694fbc0f | 1030 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1031 | struct mem_cgroup *prev, |
694fbc0f | 1032 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1033 | { |
5ac8fb31 JW |
1034 | struct reclaim_iter *uninitialized_var(iter); |
1035 | struct cgroup_subsys_state *css = NULL; | |
9f3a0d09 | 1036 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1037 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1038 | |
694fbc0f AM |
1039 | if (mem_cgroup_disabled()) |
1040 | return NULL; | |
5660048c | 1041 | |
9f3a0d09 JW |
1042 | if (!root) |
1043 | root = root_mem_cgroup; | |
7d74b06f | 1044 | |
9f3a0d09 | 1045 | if (prev && !reclaim) |
5ac8fb31 | 1046 | pos = prev; |
14067bb3 | 1047 | |
9f3a0d09 JW |
1048 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1049 | if (prev) | |
5ac8fb31 | 1050 | goto out; |
694fbc0f | 1051 | return root; |
9f3a0d09 | 1052 | } |
14067bb3 | 1053 | |
542f85f9 | 1054 | rcu_read_lock(); |
5f578161 | 1055 | |
5ac8fb31 JW |
1056 | if (reclaim) { |
1057 | struct mem_cgroup_per_zone *mz; | |
1058 | ||
1059 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
1060 | iter = &mz->iter[reclaim->priority]; | |
1061 | ||
1062 | if (prev && reclaim->generation != iter->generation) | |
1063 | goto out_unlock; | |
1064 | ||
1065 | do { | |
4db0c3c2 | 1066 | pos = READ_ONCE(iter->position); |
5ac8fb31 JW |
1067 | /* |
1068 | * A racing update may change the position and | |
1069 | * put the last reference, hence css_tryget(), | |
1070 | * or retry to see the updated position. | |
1071 | */ | |
1072 | } while (pos && !css_tryget(&pos->css)); | |
1073 | } | |
1074 | ||
1075 | if (pos) | |
1076 | css = &pos->css; | |
1077 | ||
1078 | for (;;) { | |
1079 | css = css_next_descendant_pre(css, &root->css); | |
1080 | if (!css) { | |
1081 | /* | |
1082 | * Reclaimers share the hierarchy walk, and a | |
1083 | * new one might jump in right at the end of | |
1084 | * the hierarchy - make sure they see at least | |
1085 | * one group and restart from the beginning. | |
1086 | */ | |
1087 | if (!prev) | |
1088 | continue; | |
1089 | break; | |
527a5ec9 | 1090 | } |
7d74b06f | 1091 | |
5ac8fb31 JW |
1092 | /* |
1093 | * Verify the css and acquire a reference. The root | |
1094 | * is provided by the caller, so we know it's alive | |
1095 | * and kicking, and don't take an extra reference. | |
1096 | */ | |
1097 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1098 | |
5ac8fb31 JW |
1099 | if (css == &root->css) |
1100 | break; | |
14067bb3 | 1101 | |
b2052564 | 1102 | if (css_tryget(css)) { |
5ac8fb31 JW |
1103 | /* |
1104 | * Make sure the memcg is initialized: | |
1105 | * mem_cgroup_css_online() orders the the | |
1106 | * initialization against setting the flag. | |
1107 | */ | |
1108 | if (smp_load_acquire(&memcg->initialized)) | |
1109 | break; | |
542f85f9 | 1110 | |
5ac8fb31 | 1111 | css_put(css); |
527a5ec9 | 1112 | } |
9f3a0d09 | 1113 | |
5ac8fb31 | 1114 | memcg = NULL; |
9f3a0d09 | 1115 | } |
5ac8fb31 JW |
1116 | |
1117 | if (reclaim) { | |
1118 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
1119 | if (memcg) | |
1120 | css_get(&memcg->css); | |
1121 | if (pos) | |
1122 | css_put(&pos->css); | |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * pairs with css_tryget when dereferencing iter->position | |
1127 | * above. | |
1128 | */ | |
1129 | if (pos) | |
1130 | css_put(&pos->css); | |
1131 | ||
1132 | if (!memcg) | |
1133 | iter->generation++; | |
1134 | else if (!prev) | |
1135 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1136 | } |
5ac8fb31 | 1137 | |
542f85f9 MH |
1138 | out_unlock: |
1139 | rcu_read_unlock(); | |
5ac8fb31 | 1140 | out: |
c40046f3 MH |
1141 | if (prev && prev != root) |
1142 | css_put(&prev->css); | |
1143 | ||
9f3a0d09 | 1144 | return memcg; |
14067bb3 | 1145 | } |
7d74b06f | 1146 | |
5660048c JW |
1147 | /** |
1148 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1149 | * @root: hierarchy root | |
1150 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1151 | */ | |
1152 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1153 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1154 | { |
1155 | if (!root) | |
1156 | root = root_mem_cgroup; | |
1157 | if (prev && prev != root) | |
1158 | css_put(&prev->css); | |
1159 | } | |
7d74b06f | 1160 | |
9f3a0d09 JW |
1161 | /* |
1162 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1163 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1164 | * be used for reference counting. | |
1165 | */ | |
1166 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1167 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1168 | iter != NULL; \ |
527a5ec9 | 1169 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1170 | |
9f3a0d09 | 1171 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1172 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1173 | iter != NULL; \ |
527a5ec9 | 1174 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1175 | |
68ae564b | 1176 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1177 | { |
c0ff4b85 | 1178 | struct mem_cgroup *memcg; |
456f998e | 1179 | |
456f998e | 1180 | rcu_read_lock(); |
c0ff4b85 R |
1181 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1182 | if (unlikely(!memcg)) | |
456f998e YH |
1183 | goto out; |
1184 | ||
1185 | switch (idx) { | |
456f998e | 1186 | case PGFAULT: |
0e574a93 JW |
1187 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1188 | break; | |
1189 | case PGMAJFAULT: | |
1190 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1191 | break; |
1192 | default: | |
1193 | BUG(); | |
1194 | } | |
1195 | out: | |
1196 | rcu_read_unlock(); | |
1197 | } | |
68ae564b | 1198 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1199 | |
925b7673 JW |
1200 | /** |
1201 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1202 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1203 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1204 | * |
1205 | * Returns the lru list vector holding pages for the given @zone and | |
1206 | * @mem. This can be the global zone lruvec, if the memory controller | |
1207 | * is disabled. | |
1208 | */ | |
1209 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1210 | struct mem_cgroup *memcg) | |
1211 | { | |
1212 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1213 | struct lruvec *lruvec; |
925b7673 | 1214 | |
bea8c150 HD |
1215 | if (mem_cgroup_disabled()) { |
1216 | lruvec = &zone->lruvec; | |
1217 | goto out; | |
1218 | } | |
925b7673 | 1219 | |
e231875b | 1220 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1221 | lruvec = &mz->lruvec; |
1222 | out: | |
1223 | /* | |
1224 | * Since a node can be onlined after the mem_cgroup was created, | |
1225 | * we have to be prepared to initialize lruvec->zone here; | |
1226 | * and if offlined then reonlined, we need to reinitialize it. | |
1227 | */ | |
1228 | if (unlikely(lruvec->zone != zone)) | |
1229 | lruvec->zone = zone; | |
1230 | return lruvec; | |
925b7673 JW |
1231 | } |
1232 | ||
925b7673 | 1233 | /** |
dfe0e773 | 1234 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1235 | * @page: the page |
fa9add64 | 1236 | * @zone: zone of the page |
dfe0e773 JW |
1237 | * |
1238 | * This function is only safe when following the LRU page isolation | |
1239 | * and putback protocol: the LRU lock must be held, and the page must | |
1240 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1241 | */ |
fa9add64 | 1242 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1243 | { |
08e552c6 | 1244 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1245 | struct mem_cgroup *memcg; |
bea8c150 | 1246 | struct lruvec *lruvec; |
6d12e2d8 | 1247 | |
bea8c150 HD |
1248 | if (mem_cgroup_disabled()) { |
1249 | lruvec = &zone->lruvec; | |
1250 | goto out; | |
1251 | } | |
925b7673 | 1252 | |
1306a85a | 1253 | memcg = page->mem_cgroup; |
7512102c | 1254 | /* |
dfe0e773 | 1255 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1256 | * possibly migrated - before they are charged. |
7512102c | 1257 | */ |
29833315 JW |
1258 | if (!memcg) |
1259 | memcg = root_mem_cgroup; | |
7512102c | 1260 | |
e231875b | 1261 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1262 | lruvec = &mz->lruvec; |
1263 | out: | |
1264 | /* | |
1265 | * Since a node can be onlined after the mem_cgroup was created, | |
1266 | * we have to be prepared to initialize lruvec->zone here; | |
1267 | * and if offlined then reonlined, we need to reinitialize it. | |
1268 | */ | |
1269 | if (unlikely(lruvec->zone != zone)) | |
1270 | lruvec->zone = zone; | |
1271 | return lruvec; | |
08e552c6 | 1272 | } |
b69408e8 | 1273 | |
925b7673 | 1274 | /** |
fa9add64 HD |
1275 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1276 | * @lruvec: mem_cgroup per zone lru vector | |
1277 | * @lru: index of lru list the page is sitting on | |
1278 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1279 | * |
fa9add64 HD |
1280 | * This function must be called when a page is added to or removed from an |
1281 | * lru list. | |
3f58a829 | 1282 | */ |
fa9add64 HD |
1283 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1284 | int nr_pages) | |
3f58a829 MK |
1285 | { |
1286 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1287 | unsigned long *lru_size; |
3f58a829 MK |
1288 | |
1289 | if (mem_cgroup_disabled()) | |
1290 | return; | |
1291 | ||
fa9add64 HD |
1292 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1293 | lru_size = mz->lru_size + lru; | |
1294 | *lru_size += nr_pages; | |
1295 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1296 | } |
544122e5 | 1297 | |
2314b42d | 1298 | bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root) |
3e92041d | 1299 | { |
2314b42d | 1300 | if (root == memcg) |
91c63734 | 1301 | return true; |
2314b42d | 1302 | if (!root->use_hierarchy) |
91c63734 | 1303 | return false; |
2314b42d | 1304 | return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); |
c3ac9a8a JW |
1305 | } |
1306 | ||
2314b42d | 1307 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1308 | { |
2314b42d | 1309 | struct mem_cgroup *task_memcg; |
158e0a2d | 1310 | struct task_struct *p; |
ffbdccf5 | 1311 | bool ret; |
4c4a2214 | 1312 | |
158e0a2d | 1313 | p = find_lock_task_mm(task); |
de077d22 | 1314 | if (p) { |
2314b42d | 1315 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1316 | task_unlock(p); |
1317 | } else { | |
1318 | /* | |
1319 | * All threads may have already detached their mm's, but the oom | |
1320 | * killer still needs to detect if they have already been oom | |
1321 | * killed to prevent needlessly killing additional tasks. | |
1322 | */ | |
ffbdccf5 | 1323 | rcu_read_lock(); |
2314b42d JW |
1324 | task_memcg = mem_cgroup_from_task(task); |
1325 | css_get(&task_memcg->css); | |
ffbdccf5 | 1326 | rcu_read_unlock(); |
de077d22 | 1327 | } |
2314b42d JW |
1328 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1329 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1330 | return ret; |
1331 | } | |
1332 | ||
c56d5c7d | 1333 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1334 | { |
9b272977 | 1335 | unsigned long inactive_ratio; |
14797e23 | 1336 | unsigned long inactive; |
9b272977 | 1337 | unsigned long active; |
c772be93 | 1338 | unsigned long gb; |
14797e23 | 1339 | |
4d7dcca2 HD |
1340 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1341 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1342 | |
c772be93 KM |
1343 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1344 | if (gb) | |
1345 | inactive_ratio = int_sqrt(10 * gb); | |
1346 | else | |
1347 | inactive_ratio = 1; | |
1348 | ||
9b272977 | 1349 | return inactive * inactive_ratio < active; |
14797e23 KM |
1350 | } |
1351 | ||
90cbc250 VD |
1352 | bool mem_cgroup_lruvec_online(struct lruvec *lruvec) |
1353 | { | |
1354 | struct mem_cgroup_per_zone *mz; | |
1355 | struct mem_cgroup *memcg; | |
1356 | ||
1357 | if (mem_cgroup_disabled()) | |
1358 | return true; | |
1359 | ||
1360 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
1361 | memcg = mz->memcg; | |
1362 | ||
1363 | return !!(memcg->css.flags & CSS_ONLINE); | |
1364 | } | |
1365 | ||
3e32cb2e | 1366 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1367 | container_of(counter, struct mem_cgroup, member) |
1368 | ||
19942822 | 1369 | /** |
9d11ea9f | 1370 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1371 | * @memcg: the memory cgroup |
19942822 | 1372 | * |
9d11ea9f | 1373 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1374 | * pages. |
19942822 | 1375 | */ |
c0ff4b85 | 1376 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1377 | { |
3e32cb2e JW |
1378 | unsigned long margin = 0; |
1379 | unsigned long count; | |
1380 | unsigned long limit; | |
9d11ea9f | 1381 | |
3e32cb2e | 1382 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1383 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1384 | if (count < limit) |
1385 | margin = limit - count; | |
1386 | ||
1387 | if (do_swap_account) { | |
1388 | count = page_counter_read(&memcg->memsw); | |
4db0c3c2 | 1389 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1390 | if (count <= limit) |
1391 | margin = min(margin, limit - count); | |
1392 | } | |
1393 | ||
1394 | return margin; | |
19942822 JW |
1395 | } |
1396 | ||
1f4c025b | 1397 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1398 | { |
a7885eb8 | 1399 | /* root ? */ |
14208b0e | 1400 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1401 | return vm_swappiness; |
1402 | ||
bf1ff263 | 1403 | return memcg->swappiness; |
a7885eb8 KM |
1404 | } |
1405 | ||
32047e2a | 1406 | /* |
bdcbb659 | 1407 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1408 | * |
bdcbb659 QH |
1409 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1410 | * moving cgroups. This is for waiting at high-memory pressure | |
1411 | * caused by "move". | |
32047e2a | 1412 | */ |
c0ff4b85 | 1413 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1414 | { |
2bd9bb20 KH |
1415 | struct mem_cgroup *from; |
1416 | struct mem_cgroup *to; | |
4b534334 | 1417 | bool ret = false; |
2bd9bb20 KH |
1418 | /* |
1419 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1420 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1421 | */ | |
1422 | spin_lock(&mc.lock); | |
1423 | from = mc.from; | |
1424 | to = mc.to; | |
1425 | if (!from) | |
1426 | goto unlock; | |
3e92041d | 1427 | |
2314b42d JW |
1428 | ret = mem_cgroup_is_descendant(from, memcg) || |
1429 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1430 | unlock: |
1431 | spin_unlock(&mc.lock); | |
4b534334 KH |
1432 | return ret; |
1433 | } | |
1434 | ||
c0ff4b85 | 1435 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1436 | { |
1437 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1438 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1439 | DEFINE_WAIT(wait); |
1440 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1441 | /* moving charge context might have finished. */ | |
1442 | if (mc.moving_task) | |
1443 | schedule(); | |
1444 | finish_wait(&mc.waitq, &wait); | |
1445 | return true; | |
1446 | } | |
1447 | } | |
1448 | return false; | |
1449 | } | |
1450 | ||
58cf188e | 1451 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1452 | /** |
58cf188e | 1453 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1454 | * @memcg: The memory cgroup that went over limit |
1455 | * @p: Task that is going to be killed | |
1456 | * | |
1457 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1458 | * enabled | |
1459 | */ | |
1460 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1461 | { | |
e61734c5 | 1462 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1463 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1464 | struct mem_cgroup *iter; |
1465 | unsigned int i; | |
e222432b | 1466 | |
08088cb9 | 1467 | mutex_lock(&oom_info_lock); |
e222432b BS |
1468 | rcu_read_lock(); |
1469 | ||
2415b9f5 BV |
1470 | if (p) { |
1471 | pr_info("Task in "); | |
1472 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1473 | pr_cont(" killed as a result of limit of "); | |
1474 | } else { | |
1475 | pr_info("Memory limit reached of cgroup "); | |
1476 | } | |
1477 | ||
e61734c5 | 1478 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1479 | pr_cont("\n"); |
e222432b | 1480 | |
e222432b BS |
1481 | rcu_read_unlock(); |
1482 | ||
3e32cb2e JW |
1483 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1484 | K((u64)page_counter_read(&memcg->memory)), | |
1485 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1486 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1487 | K((u64)page_counter_read(&memcg->memsw)), | |
1488 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1489 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1490 | K((u64)page_counter_read(&memcg->kmem)), | |
1491 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1492 | |
1493 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1494 | pr_info("Memory cgroup stats for "); |
1495 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1496 | pr_cont(":"); |
1497 | ||
1498 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1499 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1500 | continue; | |
1501 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1502 | K(mem_cgroup_read_stat(iter, i))); | |
1503 | } | |
1504 | ||
1505 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1506 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1507 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1508 | ||
1509 | pr_cont("\n"); | |
1510 | } | |
08088cb9 | 1511 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1512 | } |
1513 | ||
81d39c20 KH |
1514 | /* |
1515 | * This function returns the number of memcg under hierarchy tree. Returns | |
1516 | * 1(self count) if no children. | |
1517 | */ | |
c0ff4b85 | 1518 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1519 | { |
1520 | int num = 0; | |
7d74b06f KH |
1521 | struct mem_cgroup *iter; |
1522 | ||
c0ff4b85 | 1523 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1524 | num++; |
81d39c20 KH |
1525 | return num; |
1526 | } | |
1527 | ||
a63d83f4 DR |
1528 | /* |
1529 | * Return the memory (and swap, if configured) limit for a memcg. | |
1530 | */ | |
3e32cb2e | 1531 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1532 | { |
3e32cb2e | 1533 | unsigned long limit; |
f3e8eb70 | 1534 | |
3e32cb2e | 1535 | limit = memcg->memory.limit; |
9a5a8f19 | 1536 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1537 | unsigned long memsw_limit; |
9a5a8f19 | 1538 | |
3e32cb2e JW |
1539 | memsw_limit = memcg->memsw.limit; |
1540 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1541 | } |
9a5a8f19 | 1542 | return limit; |
a63d83f4 DR |
1543 | } |
1544 | ||
19965460 DR |
1545 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1546 | int order) | |
9cbb78bb | 1547 | { |
6e0fc46d DR |
1548 | struct oom_control oc = { |
1549 | .zonelist = NULL, | |
1550 | .nodemask = NULL, | |
1551 | .gfp_mask = gfp_mask, | |
1552 | .order = order, | |
6e0fc46d | 1553 | }; |
9cbb78bb DR |
1554 | struct mem_cgroup *iter; |
1555 | unsigned long chosen_points = 0; | |
1556 | unsigned long totalpages; | |
1557 | unsigned int points = 0; | |
1558 | struct task_struct *chosen = NULL; | |
1559 | ||
dc56401f JW |
1560 | mutex_lock(&oom_lock); |
1561 | ||
876aafbf | 1562 | /* |
465adcf1 DR |
1563 | * If current has a pending SIGKILL or is exiting, then automatically |
1564 | * select it. The goal is to allow it to allocate so that it may | |
1565 | * quickly exit and free its memory. | |
876aafbf | 1566 | */ |
d003f371 | 1567 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1568 | mark_oom_victim(current); |
dc56401f | 1569 | goto unlock; |
876aafbf DR |
1570 | } |
1571 | ||
6e0fc46d | 1572 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1573 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1574 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1575 | struct css_task_iter it; |
9cbb78bb DR |
1576 | struct task_struct *task; |
1577 | ||
72ec7029 TH |
1578 | css_task_iter_start(&iter->css, &it); |
1579 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1580 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1581 | case OOM_SCAN_SELECT: |
1582 | if (chosen) | |
1583 | put_task_struct(chosen); | |
1584 | chosen = task; | |
1585 | chosen_points = ULONG_MAX; | |
1586 | get_task_struct(chosen); | |
1587 | /* fall through */ | |
1588 | case OOM_SCAN_CONTINUE: | |
1589 | continue; | |
1590 | case OOM_SCAN_ABORT: | |
72ec7029 | 1591 | css_task_iter_end(&it); |
9cbb78bb DR |
1592 | mem_cgroup_iter_break(memcg, iter); |
1593 | if (chosen) | |
1594 | put_task_struct(chosen); | |
dc56401f | 1595 | goto unlock; |
9cbb78bb DR |
1596 | case OOM_SCAN_OK: |
1597 | break; | |
1598 | }; | |
1599 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1600 | if (!points || points < chosen_points) |
1601 | continue; | |
1602 | /* Prefer thread group leaders for display purposes */ | |
1603 | if (points == chosen_points && | |
1604 | thread_group_leader(chosen)) | |
1605 | continue; | |
1606 | ||
1607 | if (chosen) | |
1608 | put_task_struct(chosen); | |
1609 | chosen = task; | |
1610 | chosen_points = points; | |
1611 | get_task_struct(chosen); | |
9cbb78bb | 1612 | } |
72ec7029 | 1613 | css_task_iter_end(&it); |
9cbb78bb DR |
1614 | } |
1615 | ||
dc56401f JW |
1616 | if (chosen) { |
1617 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1618 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1619 | "Memory cgroup out of memory"); | |
dc56401f JW |
1620 | } |
1621 | unlock: | |
1622 | mutex_unlock(&oom_lock); | |
9cbb78bb DR |
1623 | } |
1624 | ||
ae6e71d3 MC |
1625 | #if MAX_NUMNODES > 1 |
1626 | ||
4d0c066d KH |
1627 | /** |
1628 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1629 | * @memcg: the target memcg |
4d0c066d KH |
1630 | * @nid: the node ID to be checked. |
1631 | * @noswap : specify true here if the user wants flle only information. | |
1632 | * | |
1633 | * This function returns whether the specified memcg contains any | |
1634 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1635 | * pages in the node. | |
1636 | */ | |
c0ff4b85 | 1637 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1638 | int nid, bool noswap) |
1639 | { | |
c0ff4b85 | 1640 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1641 | return true; |
1642 | if (noswap || !total_swap_pages) | |
1643 | return false; | |
c0ff4b85 | 1644 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1645 | return true; |
1646 | return false; | |
1647 | ||
1648 | } | |
889976db YH |
1649 | |
1650 | /* | |
1651 | * Always updating the nodemask is not very good - even if we have an empty | |
1652 | * list or the wrong list here, we can start from some node and traverse all | |
1653 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1654 | * | |
1655 | */ | |
c0ff4b85 | 1656 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1657 | { |
1658 | int nid; | |
453a9bf3 KH |
1659 | /* |
1660 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1661 | * pagein/pageout changes since the last update. | |
1662 | */ | |
c0ff4b85 | 1663 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1664 | return; |
c0ff4b85 | 1665 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1666 | return; |
1667 | ||
889976db | 1668 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1669 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1670 | |
31aaea4a | 1671 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1672 | |
c0ff4b85 R |
1673 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1674 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1675 | } |
453a9bf3 | 1676 | |
c0ff4b85 R |
1677 | atomic_set(&memcg->numainfo_events, 0); |
1678 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1679 | } |
1680 | ||
1681 | /* | |
1682 | * Selecting a node where we start reclaim from. Because what we need is just | |
1683 | * reducing usage counter, start from anywhere is O,K. Considering | |
1684 | * memory reclaim from current node, there are pros. and cons. | |
1685 | * | |
1686 | * Freeing memory from current node means freeing memory from a node which | |
1687 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1688 | * hit limits, it will see a contention on a node. But freeing from remote | |
1689 | * node means more costs for memory reclaim because of memory latency. | |
1690 | * | |
1691 | * Now, we use round-robin. Better algorithm is welcomed. | |
1692 | */ | |
c0ff4b85 | 1693 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1694 | { |
1695 | int node; | |
1696 | ||
c0ff4b85 R |
1697 | mem_cgroup_may_update_nodemask(memcg); |
1698 | node = memcg->last_scanned_node; | |
889976db | 1699 | |
c0ff4b85 | 1700 | node = next_node(node, memcg->scan_nodes); |
889976db | 1701 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1702 | node = first_node(memcg->scan_nodes); |
889976db YH |
1703 | /* |
1704 | * We call this when we hit limit, not when pages are added to LRU. | |
1705 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1706 | * memcg is too small and all pages are not on LRU. In that case, | |
1707 | * we use curret node. | |
1708 | */ | |
1709 | if (unlikely(node == MAX_NUMNODES)) | |
1710 | node = numa_node_id(); | |
1711 | ||
c0ff4b85 | 1712 | memcg->last_scanned_node = node; |
889976db YH |
1713 | return node; |
1714 | } | |
889976db | 1715 | #else |
c0ff4b85 | 1716 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1717 | { |
1718 | return 0; | |
1719 | } | |
1720 | #endif | |
1721 | ||
0608f43d AM |
1722 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1723 | struct zone *zone, | |
1724 | gfp_t gfp_mask, | |
1725 | unsigned long *total_scanned) | |
1726 | { | |
1727 | struct mem_cgroup *victim = NULL; | |
1728 | int total = 0; | |
1729 | int loop = 0; | |
1730 | unsigned long excess; | |
1731 | unsigned long nr_scanned; | |
1732 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1733 | .zone = zone, | |
1734 | .priority = 0, | |
1735 | }; | |
1736 | ||
3e32cb2e | 1737 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1738 | |
1739 | while (1) { | |
1740 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1741 | if (!victim) { | |
1742 | loop++; | |
1743 | if (loop >= 2) { | |
1744 | /* | |
1745 | * If we have not been able to reclaim | |
1746 | * anything, it might because there are | |
1747 | * no reclaimable pages under this hierarchy | |
1748 | */ | |
1749 | if (!total) | |
1750 | break; | |
1751 | /* | |
1752 | * We want to do more targeted reclaim. | |
1753 | * excess >> 2 is not to excessive so as to | |
1754 | * reclaim too much, nor too less that we keep | |
1755 | * coming back to reclaim from this cgroup | |
1756 | */ | |
1757 | if (total >= (excess >> 2) || | |
1758 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1759 | break; | |
1760 | } | |
1761 | continue; | |
1762 | } | |
0608f43d AM |
1763 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1764 | zone, &nr_scanned); | |
1765 | *total_scanned += nr_scanned; | |
3e32cb2e | 1766 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1767 | break; |
6d61ef40 | 1768 | } |
0608f43d AM |
1769 | mem_cgroup_iter_break(root_memcg, victim); |
1770 | return total; | |
6d61ef40 BS |
1771 | } |
1772 | ||
0056f4e6 JW |
1773 | #ifdef CONFIG_LOCKDEP |
1774 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1775 | .name = "memcg_oom_lock", | |
1776 | }; | |
1777 | #endif | |
1778 | ||
fb2a6fc5 JW |
1779 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1780 | ||
867578cb KH |
1781 | /* |
1782 | * Check OOM-Killer is already running under our hierarchy. | |
1783 | * If someone is running, return false. | |
1784 | */ | |
fb2a6fc5 | 1785 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1786 | { |
79dfdacc | 1787 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1788 | |
fb2a6fc5 JW |
1789 | spin_lock(&memcg_oom_lock); |
1790 | ||
9f3a0d09 | 1791 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1792 | if (iter->oom_lock) { |
79dfdacc MH |
1793 | /* |
1794 | * this subtree of our hierarchy is already locked | |
1795 | * so we cannot give a lock. | |
1796 | */ | |
79dfdacc | 1797 | failed = iter; |
9f3a0d09 JW |
1798 | mem_cgroup_iter_break(memcg, iter); |
1799 | break; | |
23751be0 JW |
1800 | } else |
1801 | iter->oom_lock = true; | |
7d74b06f | 1802 | } |
867578cb | 1803 | |
fb2a6fc5 JW |
1804 | if (failed) { |
1805 | /* | |
1806 | * OK, we failed to lock the whole subtree so we have | |
1807 | * to clean up what we set up to the failing subtree | |
1808 | */ | |
1809 | for_each_mem_cgroup_tree(iter, memcg) { | |
1810 | if (iter == failed) { | |
1811 | mem_cgroup_iter_break(memcg, iter); | |
1812 | break; | |
1813 | } | |
1814 | iter->oom_lock = false; | |
79dfdacc | 1815 | } |
0056f4e6 JW |
1816 | } else |
1817 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1818 | |
1819 | spin_unlock(&memcg_oom_lock); | |
1820 | ||
1821 | return !failed; | |
a636b327 | 1822 | } |
0b7f569e | 1823 | |
fb2a6fc5 | 1824 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1825 | { |
7d74b06f KH |
1826 | struct mem_cgroup *iter; |
1827 | ||
fb2a6fc5 | 1828 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1829 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1830 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1831 | iter->oom_lock = false; |
fb2a6fc5 | 1832 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1833 | } |
1834 | ||
c0ff4b85 | 1835 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1836 | { |
1837 | struct mem_cgroup *iter; | |
1838 | ||
c2b42d3c | 1839 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1840 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1841 | iter->under_oom++; |
1842 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1843 | } |
1844 | ||
c0ff4b85 | 1845 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1846 | { |
1847 | struct mem_cgroup *iter; | |
1848 | ||
867578cb KH |
1849 | /* |
1850 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1851 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1852 | */ |
c2b42d3c | 1853 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1854 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1855 | if (iter->under_oom > 0) |
1856 | iter->under_oom--; | |
1857 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1858 | } |
1859 | ||
867578cb KH |
1860 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1861 | ||
dc98df5a | 1862 | struct oom_wait_info { |
d79154bb | 1863 | struct mem_cgroup *memcg; |
dc98df5a KH |
1864 | wait_queue_t wait; |
1865 | }; | |
1866 | ||
1867 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1868 | unsigned mode, int sync, void *arg) | |
1869 | { | |
d79154bb HD |
1870 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1871 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1872 | struct oom_wait_info *oom_wait_info; |
1873 | ||
1874 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1875 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1876 | |
2314b42d JW |
1877 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1878 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1879 | return 0; |
dc98df5a KH |
1880 | return autoremove_wake_function(wait, mode, sync, arg); |
1881 | } | |
1882 | ||
c0ff4b85 | 1883 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1884 | { |
c2b42d3c TH |
1885 | /* |
1886 | * For the following lockless ->under_oom test, the only required | |
1887 | * guarantee is that it must see the state asserted by an OOM when | |
1888 | * this function is called as a result of userland actions | |
1889 | * triggered by the notification of the OOM. This is trivially | |
1890 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1891 | * triggering notification. | |
1892 | */ | |
1893 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1894 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1895 | } |
1896 | ||
3812c8c8 | 1897 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1898 | { |
3812c8c8 JW |
1899 | if (!current->memcg_oom.may_oom) |
1900 | return; | |
867578cb | 1901 | /* |
49426420 JW |
1902 | * We are in the middle of the charge context here, so we |
1903 | * don't want to block when potentially sitting on a callstack | |
1904 | * that holds all kinds of filesystem and mm locks. | |
1905 | * | |
1906 | * Also, the caller may handle a failed allocation gracefully | |
1907 | * (like optional page cache readahead) and so an OOM killer | |
1908 | * invocation might not even be necessary. | |
1909 | * | |
1910 | * That's why we don't do anything here except remember the | |
1911 | * OOM context and then deal with it at the end of the page | |
1912 | * fault when the stack is unwound, the locks are released, | |
1913 | * and when we know whether the fault was overall successful. | |
867578cb | 1914 | */ |
49426420 JW |
1915 | css_get(&memcg->css); |
1916 | current->memcg_oom.memcg = memcg; | |
1917 | current->memcg_oom.gfp_mask = mask; | |
1918 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1919 | } |
1920 | ||
1921 | /** | |
1922 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1923 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1924 | * |
49426420 JW |
1925 | * This has to be called at the end of a page fault if the memcg OOM |
1926 | * handler was enabled. | |
3812c8c8 | 1927 | * |
49426420 | 1928 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1929 | * sleep on a waitqueue until the userspace task resolves the |
1930 | * situation. Sleeping directly in the charge context with all kinds | |
1931 | * of locks held is not a good idea, instead we remember an OOM state | |
1932 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1933 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1934 | * |
1935 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1936 | * completed, %false otherwise. |
3812c8c8 | 1937 | */ |
49426420 | 1938 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1939 | { |
49426420 | 1940 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1941 | struct oom_wait_info owait; |
49426420 | 1942 | bool locked; |
3812c8c8 JW |
1943 | |
1944 | /* OOM is global, do not handle */ | |
3812c8c8 | 1945 | if (!memcg) |
49426420 | 1946 | return false; |
3812c8c8 | 1947 | |
c32b3cbe | 1948 | if (!handle || oom_killer_disabled) |
49426420 | 1949 | goto cleanup; |
3812c8c8 JW |
1950 | |
1951 | owait.memcg = memcg; | |
1952 | owait.wait.flags = 0; | |
1953 | owait.wait.func = memcg_oom_wake_function; | |
1954 | owait.wait.private = current; | |
1955 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1956 | |
3812c8c8 | 1957 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1958 | mem_cgroup_mark_under_oom(memcg); |
1959 | ||
1960 | locked = mem_cgroup_oom_trylock(memcg); | |
1961 | ||
1962 | if (locked) | |
1963 | mem_cgroup_oom_notify(memcg); | |
1964 | ||
1965 | if (locked && !memcg->oom_kill_disable) { | |
1966 | mem_cgroup_unmark_under_oom(memcg); | |
1967 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1968 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1969 | current->memcg_oom.order); | |
1970 | } else { | |
3812c8c8 | 1971 | schedule(); |
49426420 JW |
1972 | mem_cgroup_unmark_under_oom(memcg); |
1973 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1974 | } | |
1975 | ||
1976 | if (locked) { | |
fb2a6fc5 JW |
1977 | mem_cgroup_oom_unlock(memcg); |
1978 | /* | |
1979 | * There is no guarantee that an OOM-lock contender | |
1980 | * sees the wakeups triggered by the OOM kill | |
1981 | * uncharges. Wake any sleepers explicitely. | |
1982 | */ | |
1983 | memcg_oom_recover(memcg); | |
1984 | } | |
49426420 JW |
1985 | cleanup: |
1986 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 1987 | css_put(&memcg->css); |
867578cb | 1988 | return true; |
0b7f569e KH |
1989 | } |
1990 | ||
d7365e78 JW |
1991 | /** |
1992 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1993 | * @page: page that is going to change accounted state | |
32047e2a | 1994 | * |
d7365e78 JW |
1995 | * This function must mark the beginning of an accounted page state |
1996 | * change to prevent double accounting when the page is concurrently | |
1997 | * being moved to another memcg: | |
32047e2a | 1998 | * |
6de22619 | 1999 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
2000 | * if (TestClearPageState(page)) |
2001 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 2002 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 2003 | */ |
6de22619 | 2004 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
2005 | { |
2006 | struct mem_cgroup *memcg; | |
6de22619 | 2007 | unsigned long flags; |
89c06bd5 | 2008 | |
6de22619 JW |
2009 | /* |
2010 | * The RCU lock is held throughout the transaction. The fast | |
2011 | * path can get away without acquiring the memcg->move_lock | |
2012 | * because page moving starts with an RCU grace period. | |
2013 | * | |
2014 | * The RCU lock also protects the memcg from being freed when | |
2015 | * the page state that is going to change is the only thing | |
2016 | * preventing the page from being uncharged. | |
2017 | * E.g. end-writeback clearing PageWriteback(), which allows | |
2018 | * migration to go ahead and uncharge the page before the | |
2019 | * account transaction might be complete. | |
2020 | */ | |
d7365e78 JW |
2021 | rcu_read_lock(); |
2022 | ||
2023 | if (mem_cgroup_disabled()) | |
2024 | return NULL; | |
89c06bd5 | 2025 | again: |
1306a85a | 2026 | memcg = page->mem_cgroup; |
29833315 | 2027 | if (unlikely(!memcg)) |
d7365e78 JW |
2028 | return NULL; |
2029 | ||
bdcbb659 | 2030 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2031 | return memcg; |
89c06bd5 | 2032 | |
6de22619 | 2033 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 2034 | if (memcg != page->mem_cgroup) { |
6de22619 | 2035 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2036 | goto again; |
2037 | } | |
6de22619 JW |
2038 | |
2039 | /* | |
2040 | * When charge migration first begins, we can have locked and | |
2041 | * unlocked page stat updates happening concurrently. Track | |
2042 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
2043 | */ | |
2044 | memcg->move_lock_task = current; | |
2045 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
2046 | |
2047 | return memcg; | |
89c06bd5 | 2048 | } |
c4843a75 | 2049 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
89c06bd5 | 2050 | |
d7365e78 JW |
2051 | /** |
2052 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2053 | * @memcg: the memcg that was accounted against | |
d7365e78 | 2054 | */ |
6de22619 | 2055 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 2056 | { |
6de22619 JW |
2057 | if (memcg && memcg->move_lock_task == current) { |
2058 | unsigned long flags = memcg->move_lock_flags; | |
2059 | ||
2060 | memcg->move_lock_task = NULL; | |
2061 | memcg->move_lock_flags = 0; | |
2062 | ||
2063 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2064 | } | |
89c06bd5 | 2065 | |
d7365e78 | 2066 | rcu_read_unlock(); |
89c06bd5 | 2067 | } |
c4843a75 | 2068 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
89c06bd5 | 2069 | |
d7365e78 JW |
2070 | /** |
2071 | * mem_cgroup_update_page_stat - update page state statistics | |
2072 | * @memcg: memcg to account against | |
2073 | * @idx: page state item to account | |
2074 | * @val: number of pages (positive or negative) | |
2075 | * | |
2076 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2077 | */ | |
2078 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2079 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2080 | { |
658b72c5 | 2081 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2082 | |
d7365e78 JW |
2083 | if (memcg) |
2084 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2085 | } |
26174efd | 2086 | |
cdec2e42 KH |
2087 | /* |
2088 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2089 | * TODO: maybe necessary to use big numbers in big irons. | |
2090 | */ | |
7ec99d62 | 2091 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2092 | struct memcg_stock_pcp { |
2093 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2094 | unsigned int nr_pages; |
cdec2e42 | 2095 | struct work_struct work; |
26fe6168 | 2096 | unsigned long flags; |
a0db00fc | 2097 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2098 | }; |
2099 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2100 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2101 | |
a0956d54 SS |
2102 | /** |
2103 | * consume_stock: Try to consume stocked charge on this cpu. | |
2104 | * @memcg: memcg to consume from. | |
2105 | * @nr_pages: how many pages to charge. | |
2106 | * | |
2107 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2108 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2109 | * service an allocation will refill the stock. | |
2110 | * | |
2111 | * returns true if successful, false otherwise. | |
cdec2e42 | 2112 | */ |
a0956d54 | 2113 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2114 | { |
2115 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2116 | bool ret = false; |
cdec2e42 | 2117 | |
a0956d54 | 2118 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2119 | return ret; |
a0956d54 | 2120 | |
cdec2e42 | 2121 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2122 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2123 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2124 | ret = true; |
2125 | } | |
cdec2e42 KH |
2126 | put_cpu_var(memcg_stock); |
2127 | return ret; | |
2128 | } | |
2129 | ||
2130 | /* | |
3e32cb2e | 2131 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2132 | */ |
2133 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2134 | { | |
2135 | struct mem_cgroup *old = stock->cached; | |
2136 | ||
11c9ea4e | 2137 | if (stock->nr_pages) { |
3e32cb2e | 2138 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2139 | if (do_swap_account) |
3e32cb2e | 2140 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2141 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2142 | stock->nr_pages = 0; |
cdec2e42 KH |
2143 | } |
2144 | stock->cached = NULL; | |
cdec2e42 KH |
2145 | } |
2146 | ||
2147 | /* | |
2148 | * This must be called under preempt disabled or must be called by | |
2149 | * a thread which is pinned to local cpu. | |
2150 | */ | |
2151 | static void drain_local_stock(struct work_struct *dummy) | |
2152 | { | |
7c8e0181 | 2153 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2154 | drain_stock(stock); |
26fe6168 | 2155 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2156 | } |
2157 | ||
2158 | /* | |
3e32cb2e | 2159 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2160 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2161 | */ |
c0ff4b85 | 2162 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2163 | { |
2164 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2165 | ||
c0ff4b85 | 2166 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2167 | drain_stock(stock); |
c0ff4b85 | 2168 | stock->cached = memcg; |
cdec2e42 | 2169 | } |
11c9ea4e | 2170 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2171 | put_cpu_var(memcg_stock); |
2172 | } | |
2173 | ||
2174 | /* | |
c0ff4b85 | 2175 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2176 | * of the hierarchy under it. |
cdec2e42 | 2177 | */ |
6d3d6aa2 | 2178 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2179 | { |
26fe6168 | 2180 | int cpu, curcpu; |
d38144b7 | 2181 | |
6d3d6aa2 JW |
2182 | /* If someone's already draining, avoid adding running more workers. */ |
2183 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2184 | return; | |
cdec2e42 | 2185 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2186 | get_online_cpus(); |
5af12d0e | 2187 | curcpu = get_cpu(); |
cdec2e42 KH |
2188 | for_each_online_cpu(cpu) { |
2189 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2190 | struct mem_cgroup *memcg; |
26fe6168 | 2191 | |
c0ff4b85 R |
2192 | memcg = stock->cached; |
2193 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2194 | continue; |
2314b42d | 2195 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 2196 | continue; |
d1a05b69 MH |
2197 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2198 | if (cpu == curcpu) | |
2199 | drain_local_stock(&stock->work); | |
2200 | else | |
2201 | schedule_work_on(cpu, &stock->work); | |
2202 | } | |
cdec2e42 | 2203 | } |
5af12d0e | 2204 | put_cpu(); |
f894ffa8 | 2205 | put_online_cpus(); |
9f50fad6 | 2206 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2207 | } |
2208 | ||
0db0628d | 2209 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2210 | unsigned long action, |
2211 | void *hcpu) | |
2212 | { | |
2213 | int cpu = (unsigned long)hcpu; | |
2214 | struct memcg_stock_pcp *stock; | |
2215 | ||
619d094b | 2216 | if (action == CPU_ONLINE) |
1489ebad | 2217 | return NOTIFY_OK; |
1489ebad | 2218 | |
d833049b | 2219 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2220 | return NOTIFY_OK; |
711d3d2c | 2221 | |
cdec2e42 KH |
2222 | stock = &per_cpu(memcg_stock, cpu); |
2223 | drain_stock(stock); | |
2224 | return NOTIFY_OK; | |
2225 | } | |
2226 | ||
00501b53 JW |
2227 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2228 | unsigned int nr_pages) | |
8a9f3ccd | 2229 | { |
7ec99d62 | 2230 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2231 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2232 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2233 | struct page_counter *counter; |
6539cc05 | 2234 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2235 | bool may_swap = true; |
2236 | bool drained = false; | |
05b84301 | 2237 | int ret = 0; |
a636b327 | 2238 | |
ce00a967 JW |
2239 | if (mem_cgroup_is_root(memcg)) |
2240 | goto done; | |
6539cc05 | 2241 | retry: |
b6b6cc72 MH |
2242 | if (consume_stock(memcg, nr_pages)) |
2243 | goto done; | |
8a9f3ccd | 2244 | |
3fbe7244 | 2245 | if (!do_swap_account || |
3e32cb2e JW |
2246 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2247 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2248 | goto done_restock; |
3fbe7244 | 2249 | if (do_swap_account) |
3e32cb2e JW |
2250 | page_counter_uncharge(&memcg->memsw, batch); |
2251 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2252 | } else { |
3e32cb2e | 2253 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2254 | may_swap = false; |
3fbe7244 | 2255 | } |
7a81b88c | 2256 | |
6539cc05 JW |
2257 | if (batch > nr_pages) { |
2258 | batch = nr_pages; | |
2259 | goto retry; | |
2260 | } | |
6d61ef40 | 2261 | |
06b078fc JW |
2262 | /* |
2263 | * Unlike in global OOM situations, memcg is not in a physical | |
2264 | * memory shortage. Allow dying and OOM-killed tasks to | |
2265 | * bypass the last charges so that they can exit quickly and | |
2266 | * free their memory. | |
2267 | */ | |
2268 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2269 | fatal_signal_pending(current) || | |
2270 | current->flags & PF_EXITING)) | |
2271 | goto bypass; | |
2272 | ||
2273 | if (unlikely(task_in_memcg_oom(current))) | |
2274 | goto nomem; | |
2275 | ||
6539cc05 JW |
2276 | if (!(gfp_mask & __GFP_WAIT)) |
2277 | goto nomem; | |
4b534334 | 2278 | |
241994ed JW |
2279 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2280 | ||
b70a2a21 JW |
2281 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2282 | gfp_mask, may_swap); | |
6539cc05 | 2283 | |
61e02c74 | 2284 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2285 | goto retry; |
28c34c29 | 2286 | |
b70a2a21 | 2287 | if (!drained) { |
6d3d6aa2 | 2288 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2289 | drained = true; |
2290 | goto retry; | |
2291 | } | |
2292 | ||
28c34c29 JW |
2293 | if (gfp_mask & __GFP_NORETRY) |
2294 | goto nomem; | |
6539cc05 JW |
2295 | /* |
2296 | * Even though the limit is exceeded at this point, reclaim | |
2297 | * may have been able to free some pages. Retry the charge | |
2298 | * before killing the task. | |
2299 | * | |
2300 | * Only for regular pages, though: huge pages are rather | |
2301 | * unlikely to succeed so close to the limit, and we fall back | |
2302 | * to regular pages anyway in case of failure. | |
2303 | */ | |
61e02c74 | 2304 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2305 | goto retry; |
2306 | /* | |
2307 | * At task move, charge accounts can be doubly counted. So, it's | |
2308 | * better to wait until the end of task_move if something is going on. | |
2309 | */ | |
2310 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2311 | goto retry; | |
2312 | ||
9b130619 JW |
2313 | if (nr_retries--) |
2314 | goto retry; | |
2315 | ||
06b078fc JW |
2316 | if (gfp_mask & __GFP_NOFAIL) |
2317 | goto bypass; | |
2318 | ||
6539cc05 JW |
2319 | if (fatal_signal_pending(current)) |
2320 | goto bypass; | |
2321 | ||
241994ed JW |
2322 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2323 | ||
61e02c74 | 2324 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2325 | nomem: |
6d1fdc48 | 2326 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2327 | return -ENOMEM; |
867578cb | 2328 | bypass: |
ce00a967 | 2329 | return -EINTR; |
6539cc05 JW |
2330 | |
2331 | done_restock: | |
e8ea14cc | 2332 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2333 | if (batch > nr_pages) |
2334 | refill_stock(memcg, batch - nr_pages); | |
7d638093 VD |
2335 | if (!(gfp_mask & __GFP_WAIT)) |
2336 | goto done; | |
241994ed JW |
2337 | /* |
2338 | * If the hierarchy is above the normal consumption range, | |
2339 | * make the charging task trim their excess contribution. | |
2340 | */ | |
2341 | do { | |
2342 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2343 | continue; | |
2344 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
2345 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
2346 | } while ((memcg = parent_mem_cgroup(memcg))); | |
6539cc05 | 2347 | done: |
05b84301 | 2348 | return ret; |
7a81b88c | 2349 | } |
8a9f3ccd | 2350 | |
00501b53 | 2351 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2352 | { |
ce00a967 JW |
2353 | if (mem_cgroup_is_root(memcg)) |
2354 | return; | |
2355 | ||
3e32cb2e | 2356 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2357 | if (do_swap_account) |
3e32cb2e | 2358 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2359 | |
e8ea14cc | 2360 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2361 | } |
2362 | ||
0a31bc97 JW |
2363 | /* |
2364 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2365 | * @page: the page | |
2366 | * | |
2367 | * Look up, get a css reference, and return the memcg that owns @page. | |
2368 | * | |
2369 | * The page must be locked to prevent racing with swap-in and page | |
2370 | * cache charges. If coming from an unlocked page table, the caller | |
2371 | * must ensure the page is on the LRU or this can race with charging. | |
2372 | */ | |
e42d9d5d | 2373 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2374 | { |
29833315 | 2375 | struct mem_cgroup *memcg; |
a3b2d692 | 2376 | unsigned short id; |
b5a84319 KH |
2377 | swp_entry_t ent; |
2378 | ||
309381fe | 2379 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2380 | |
1306a85a | 2381 | memcg = page->mem_cgroup; |
29833315 JW |
2382 | if (memcg) { |
2383 | if (!css_tryget_online(&memcg->css)) | |
c0ff4b85 | 2384 | memcg = NULL; |
e42d9d5d | 2385 | } else if (PageSwapCache(page)) { |
3c776e64 | 2386 | ent.val = page_private(page); |
9fb4b7cc | 2387 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2388 | rcu_read_lock(); |
adbe427b | 2389 | memcg = mem_cgroup_from_id(id); |
ec903c0c | 2390 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2391 | memcg = NULL; |
a3b2d692 | 2392 | rcu_read_unlock(); |
3c776e64 | 2393 | } |
c0ff4b85 | 2394 | return memcg; |
b5a84319 KH |
2395 | } |
2396 | ||
0a31bc97 JW |
2397 | static void lock_page_lru(struct page *page, int *isolated) |
2398 | { | |
2399 | struct zone *zone = page_zone(page); | |
2400 | ||
2401 | spin_lock_irq(&zone->lru_lock); | |
2402 | if (PageLRU(page)) { | |
2403 | struct lruvec *lruvec; | |
2404 | ||
2405 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2406 | ClearPageLRU(page); | |
2407 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2408 | *isolated = 1; | |
2409 | } else | |
2410 | *isolated = 0; | |
2411 | } | |
2412 | ||
2413 | static void unlock_page_lru(struct page *page, int isolated) | |
2414 | { | |
2415 | struct zone *zone = page_zone(page); | |
2416 | ||
2417 | if (isolated) { | |
2418 | struct lruvec *lruvec; | |
2419 | ||
2420 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2421 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2422 | SetPageLRU(page); | |
2423 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2424 | } | |
2425 | spin_unlock_irq(&zone->lru_lock); | |
2426 | } | |
2427 | ||
00501b53 | 2428 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2429 | bool lrucare) |
7a81b88c | 2430 | { |
0a31bc97 | 2431 | int isolated; |
9ce70c02 | 2432 | |
1306a85a | 2433 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2434 | |
2435 | /* | |
2436 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2437 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2438 | */ | |
0a31bc97 JW |
2439 | if (lrucare) |
2440 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2441 | |
0a31bc97 JW |
2442 | /* |
2443 | * Nobody should be changing or seriously looking at | |
1306a85a | 2444 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2445 | * |
2446 | * - the page is uncharged | |
2447 | * | |
2448 | * - the page is off-LRU | |
2449 | * | |
2450 | * - an anonymous fault has exclusive page access, except for | |
2451 | * a locked page table | |
2452 | * | |
2453 | * - a page cache insertion, a swapin fault, or a migration | |
2454 | * have the page locked | |
2455 | */ | |
1306a85a | 2456 | page->mem_cgroup = memcg; |
9ce70c02 | 2457 | |
0a31bc97 JW |
2458 | if (lrucare) |
2459 | unlock_page_lru(page, isolated); | |
7a81b88c | 2460 | } |
66e1707b | 2461 | |
7ae1e1d0 | 2462 | #ifdef CONFIG_MEMCG_KMEM |
dbf22eb6 VD |
2463 | int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2464 | unsigned long nr_pages) | |
7ae1e1d0 | 2465 | { |
3e32cb2e | 2466 | struct page_counter *counter; |
7ae1e1d0 | 2467 | int ret = 0; |
7ae1e1d0 | 2468 | |
3e32cb2e JW |
2469 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2470 | if (ret < 0) | |
7ae1e1d0 GC |
2471 | return ret; |
2472 | ||
3e32cb2e | 2473 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2474 | if (ret == -EINTR) { |
2475 | /* | |
00501b53 JW |
2476 | * try_charge() chose to bypass to root due to OOM kill or |
2477 | * fatal signal. Since our only options are to either fail | |
2478 | * the allocation or charge it to this cgroup, do it as a | |
2479 | * temporary condition. But we can't fail. From a kmem/slab | |
2480 | * perspective, the cache has already been selected, by | |
2481 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2482 | * our minds. |
2483 | * | |
2484 | * This condition will only trigger if the task entered | |
00501b53 JW |
2485 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2486 | * during try_charge() above. Tasks that were already dying | |
2487 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2488 | * directed to the root cgroup in memcontrol.h |
2489 | */ | |
3e32cb2e | 2490 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2491 | if (do_swap_account) |
3e32cb2e | 2492 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2493 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2494 | ret = 0; |
2495 | } else if (ret) | |
3e32cb2e | 2496 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2497 | |
2498 | return ret; | |
2499 | } | |
2500 | ||
dbf22eb6 | 2501 | void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages) |
7ae1e1d0 | 2502 | { |
3e32cb2e | 2503 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2504 | if (do_swap_account) |
3e32cb2e | 2505 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2506 | |
64f21993 | 2507 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7de37682 | 2508 | |
e8ea14cc | 2509 | css_put_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2510 | } |
2511 | ||
2633d7a0 GC |
2512 | /* |
2513 | * helper for acessing a memcg's index. It will be used as an index in the | |
2514 | * child cache array in kmem_cache, and also to derive its name. This function | |
2515 | * will return -1 when this is not a kmem-limited memcg. | |
2516 | */ | |
2517 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2518 | { | |
2519 | return memcg ? memcg->kmemcg_id : -1; | |
2520 | } | |
2521 | ||
f3bb3043 | 2522 | static int memcg_alloc_cache_id(void) |
55007d84 | 2523 | { |
f3bb3043 VD |
2524 | int id, size; |
2525 | int err; | |
2526 | ||
dbcf73e2 | 2527 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2528 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2529 | if (id < 0) | |
2530 | return id; | |
55007d84 | 2531 | |
dbcf73e2 | 2532 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2533 | return id; |
2534 | ||
2535 | /* | |
2536 | * There's no space for the new id in memcg_caches arrays, | |
2537 | * so we have to grow them. | |
2538 | */ | |
05257a1a | 2539 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2540 | |
2541 | size = 2 * (id + 1); | |
55007d84 GC |
2542 | if (size < MEMCG_CACHES_MIN_SIZE) |
2543 | size = MEMCG_CACHES_MIN_SIZE; | |
2544 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2545 | size = MEMCG_CACHES_MAX_SIZE; | |
2546 | ||
f3bb3043 | 2547 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2548 | if (!err) |
2549 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2550 | if (!err) |
2551 | memcg_nr_cache_ids = size; | |
2552 | ||
2553 | up_write(&memcg_cache_ids_sem); | |
2554 | ||
f3bb3043 | 2555 | if (err) { |
dbcf73e2 | 2556 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2557 | return err; |
2558 | } | |
2559 | return id; | |
2560 | } | |
2561 | ||
2562 | static void memcg_free_cache_id(int id) | |
2563 | { | |
dbcf73e2 | 2564 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2565 | } |
2566 | ||
d5b3cf71 | 2567 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2568 | struct mem_cgroup *memcg; |
2569 | struct kmem_cache *cachep; | |
2570 | struct work_struct work; | |
2571 | }; | |
2572 | ||
d5b3cf71 | 2573 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2574 | { |
d5b3cf71 VD |
2575 | struct memcg_kmem_cache_create_work *cw = |
2576 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2577 | struct mem_cgroup *memcg = cw->memcg; |
2578 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2579 | |
d5b3cf71 | 2580 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2581 | |
5722d094 | 2582 | css_put(&memcg->css); |
d7f25f8a GC |
2583 | kfree(cw); |
2584 | } | |
2585 | ||
2586 | /* | |
2587 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2588 | */ |
d5b3cf71 VD |
2589 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2590 | struct kmem_cache *cachep) | |
d7f25f8a | 2591 | { |
d5b3cf71 | 2592 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2593 | |
776ed0f0 | 2594 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2595 | if (!cw) |
d7f25f8a | 2596 | return; |
8135be5a VD |
2597 | |
2598 | css_get(&memcg->css); | |
d7f25f8a GC |
2599 | |
2600 | cw->memcg = memcg; | |
2601 | cw->cachep = cachep; | |
d5b3cf71 | 2602 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2603 | |
d7f25f8a GC |
2604 | schedule_work(&cw->work); |
2605 | } | |
2606 | ||
d5b3cf71 VD |
2607 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2608 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2609 | { |
2610 | /* | |
2611 | * We need to stop accounting when we kmalloc, because if the | |
2612 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2613 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2614 | * |
2615 | * However, it is better to enclose the whole function. Depending on | |
2616 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2617 | * trigger an allocation. This too, will make us recurse. Because at | |
2618 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2619 | * the safest choice is to do it like this, wrapping the whole function. | |
2620 | */ | |
6f185c29 | 2621 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2622 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2623 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2624 | } |
c67a8a68 | 2625 | |
d7f25f8a GC |
2626 | /* |
2627 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2628 | * We try to use the current memcg's version of the cache. | |
2629 | * | |
2630 | * If the cache does not exist yet, if we are the first user of it, | |
2631 | * we either create it immediately, if possible, or create it asynchronously | |
2632 | * in a workqueue. | |
2633 | * In the latter case, we will let the current allocation go through with | |
2634 | * the original cache. | |
2635 | * | |
2636 | * Can't be called in interrupt context or from kernel threads. | |
2637 | * This function needs to be called with rcu_read_lock() held. | |
2638 | */ | |
056b7cce | 2639 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2640 | { |
2641 | struct mem_cgroup *memcg; | |
959c8963 | 2642 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2643 | int kmemcg_id; |
d7f25f8a | 2644 | |
f7ce3190 | 2645 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2646 | |
9d100c5e | 2647 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2648 | return cachep; |
2649 | ||
8135be5a | 2650 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2651 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2652 | if (kmemcg_id < 0) |
ca0dde97 | 2653 | goto out; |
d7f25f8a | 2654 | |
2a4db7eb | 2655 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2656 | if (likely(memcg_cachep)) |
2657 | return memcg_cachep; | |
ca0dde97 LZ |
2658 | |
2659 | /* | |
2660 | * If we are in a safe context (can wait, and not in interrupt | |
2661 | * context), we could be be predictable and return right away. | |
2662 | * This would guarantee that the allocation being performed | |
2663 | * already belongs in the new cache. | |
2664 | * | |
2665 | * However, there are some clashes that can arrive from locking. | |
2666 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2667 | * memcg_create_kmem_cache, this means no further allocation |
2668 | * could happen with the slab_mutex held. So it's better to | |
2669 | * defer everything. | |
ca0dde97 | 2670 | */ |
d5b3cf71 | 2671 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2672 | out: |
8135be5a | 2673 | css_put(&memcg->css); |
ca0dde97 | 2674 | return cachep; |
d7f25f8a | 2675 | } |
d7f25f8a | 2676 | |
8135be5a VD |
2677 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2678 | { | |
2679 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2680 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2681 | } |
2682 | ||
7ae1e1d0 GC |
2683 | /* |
2684 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2685 | * possible for the given order. But the page is not allocated yet, so we'll | |
2686 | * need a further commit step to do the final arrangements. | |
2687 | * | |
2688 | * It is possible for the task to switch cgroups in this mean time, so at | |
2689 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2690 | * the handle argument to return to the caller which cgroup we should commit | |
2691 | * against. We could also return the memcg directly and avoid the pointer | |
2692 | * passing, but a boolean return value gives better semantics considering | |
2693 | * the compiled-out case as well. | |
2694 | * | |
2695 | * Returning true means the allocation is possible. | |
2696 | */ | |
2697 | bool | |
2698 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2699 | { | |
2700 | struct mem_cgroup *memcg; | |
2701 | int ret; | |
2702 | ||
2703 | *_memcg = NULL; | |
6d42c232 | 2704 | |
df381975 | 2705 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 2706 | |
cf2b8fbf | 2707 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
2708 | css_put(&memcg->css); |
2709 | return true; | |
2710 | } | |
2711 | ||
3e32cb2e | 2712 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
2713 | if (!ret) |
2714 | *_memcg = memcg; | |
7ae1e1d0 GC |
2715 | |
2716 | css_put(&memcg->css); | |
2717 | return (ret == 0); | |
2718 | } | |
2719 | ||
2720 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
2721 | int order) | |
2722 | { | |
7ae1e1d0 GC |
2723 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
2724 | ||
2725 | /* The page allocation failed. Revert */ | |
2726 | if (!page) { | |
3e32cb2e | 2727 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
2728 | return; |
2729 | } | |
1306a85a | 2730 | page->mem_cgroup = memcg; |
7ae1e1d0 GC |
2731 | } |
2732 | ||
2733 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
2734 | { | |
1306a85a | 2735 | struct mem_cgroup *memcg = page->mem_cgroup; |
7ae1e1d0 | 2736 | |
7ae1e1d0 GC |
2737 | if (!memcg) |
2738 | return; | |
2739 | ||
309381fe | 2740 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2741 | |
3e32cb2e | 2742 | memcg_uncharge_kmem(memcg, 1 << order); |
1306a85a | 2743 | page->mem_cgroup = NULL; |
7ae1e1d0 | 2744 | } |
60d3fd32 VD |
2745 | |
2746 | struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr) | |
2747 | { | |
2748 | struct mem_cgroup *memcg = NULL; | |
2749 | struct kmem_cache *cachep; | |
2750 | struct page *page; | |
2751 | ||
2752 | page = virt_to_head_page(ptr); | |
2753 | if (PageSlab(page)) { | |
2754 | cachep = page->slab_cache; | |
2755 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2756 | memcg = cachep->memcg_params.memcg; |
60d3fd32 VD |
2757 | } else |
2758 | /* page allocated by alloc_kmem_pages */ | |
2759 | memcg = page->mem_cgroup; | |
2760 | ||
2761 | return memcg; | |
2762 | } | |
7ae1e1d0 GC |
2763 | #endif /* CONFIG_MEMCG_KMEM */ |
2764 | ||
ca3e0214 KH |
2765 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2766 | ||
ca3e0214 KH |
2767 | /* |
2768 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2769 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2770 | * charge/uncharge will be never happen and move_account() is done under | |
2771 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2772 | */ |
e94c8a9c | 2773 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2774 | { |
e94c8a9c | 2775 | int i; |
ca3e0214 | 2776 | |
3d37c4a9 KH |
2777 | if (mem_cgroup_disabled()) |
2778 | return; | |
b070e65c | 2779 | |
29833315 | 2780 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2781 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2782 | |
1306a85a | 2783 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2784 | HPAGE_PMD_NR); |
ca3e0214 | 2785 | } |
12d27107 | 2786 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2787 | |
c255a458 | 2788 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2789 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2790 | bool charge) | |
d13d1443 | 2791 | { |
0a31bc97 JW |
2792 | int val = (charge) ? 1 : -1; |
2793 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2794 | } |
02491447 DN |
2795 | |
2796 | /** | |
2797 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2798 | * @entry: swap entry to be moved | |
2799 | * @from: mem_cgroup which the entry is moved from | |
2800 | * @to: mem_cgroup which the entry is moved to | |
2801 | * | |
2802 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2803 | * as the mem_cgroup's id of @from. | |
2804 | * | |
2805 | * Returns 0 on success, -EINVAL on failure. | |
2806 | * | |
3e32cb2e | 2807 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2808 | * both res and memsw, and called css_get(). |
2809 | */ | |
2810 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2811 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2812 | { |
2813 | unsigned short old_id, new_id; | |
2814 | ||
34c00c31 LZ |
2815 | old_id = mem_cgroup_id(from); |
2816 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2817 | |
2818 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2819 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2820 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2821 | return 0; |
2822 | } | |
2823 | return -EINVAL; | |
2824 | } | |
2825 | #else | |
2826 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2827 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2828 | { |
2829 | return -EINVAL; | |
2830 | } | |
8c7c6e34 | 2831 | #endif |
d13d1443 | 2832 | |
3e32cb2e | 2833 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2834 | |
d38d2a75 | 2835 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2836 | unsigned long limit) |
628f4235 | 2837 | { |
3e32cb2e JW |
2838 | unsigned long curusage; |
2839 | unsigned long oldusage; | |
2840 | bool enlarge = false; | |
81d39c20 | 2841 | int retry_count; |
3e32cb2e | 2842 | int ret; |
81d39c20 KH |
2843 | |
2844 | /* | |
2845 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2846 | * is depends on callers. We set our retry-count to be function | |
2847 | * of # of children which we should visit in this loop. | |
2848 | */ | |
3e32cb2e JW |
2849 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2850 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2851 | |
3e32cb2e | 2852 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2853 | |
3e32cb2e | 2854 | do { |
628f4235 KH |
2855 | if (signal_pending(current)) { |
2856 | ret = -EINTR; | |
2857 | break; | |
2858 | } | |
3e32cb2e JW |
2859 | |
2860 | mutex_lock(&memcg_limit_mutex); | |
2861 | if (limit > memcg->memsw.limit) { | |
2862 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2863 | ret = -EINVAL; |
628f4235 KH |
2864 | break; |
2865 | } | |
3e32cb2e JW |
2866 | if (limit > memcg->memory.limit) |
2867 | enlarge = true; | |
2868 | ret = page_counter_limit(&memcg->memory, limit); | |
2869 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2870 | |
2871 | if (!ret) | |
2872 | break; | |
2873 | ||
b70a2a21 JW |
2874 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2875 | ||
3e32cb2e | 2876 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2877 | /* Usage is reduced ? */ |
f894ffa8 | 2878 | if (curusage >= oldusage) |
81d39c20 KH |
2879 | retry_count--; |
2880 | else | |
2881 | oldusage = curusage; | |
3e32cb2e JW |
2882 | } while (retry_count); |
2883 | ||
3c11ecf4 KH |
2884 | if (!ret && enlarge) |
2885 | memcg_oom_recover(memcg); | |
14797e23 | 2886 | |
8c7c6e34 KH |
2887 | return ret; |
2888 | } | |
2889 | ||
338c8431 | 2890 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2891 | unsigned long limit) |
8c7c6e34 | 2892 | { |
3e32cb2e JW |
2893 | unsigned long curusage; |
2894 | unsigned long oldusage; | |
2895 | bool enlarge = false; | |
81d39c20 | 2896 | int retry_count; |
3e32cb2e | 2897 | int ret; |
8c7c6e34 | 2898 | |
81d39c20 | 2899 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2900 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2901 | mem_cgroup_count_children(memcg); | |
2902 | ||
2903 | oldusage = page_counter_read(&memcg->memsw); | |
2904 | ||
2905 | do { | |
8c7c6e34 KH |
2906 | if (signal_pending(current)) { |
2907 | ret = -EINTR; | |
2908 | break; | |
2909 | } | |
3e32cb2e JW |
2910 | |
2911 | mutex_lock(&memcg_limit_mutex); | |
2912 | if (limit < memcg->memory.limit) { | |
2913 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2914 | ret = -EINVAL; |
8c7c6e34 KH |
2915 | break; |
2916 | } | |
3e32cb2e JW |
2917 | if (limit > memcg->memsw.limit) |
2918 | enlarge = true; | |
2919 | ret = page_counter_limit(&memcg->memsw, limit); | |
2920 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2921 | |
2922 | if (!ret) | |
2923 | break; | |
2924 | ||
b70a2a21 JW |
2925 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2926 | ||
3e32cb2e | 2927 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2928 | /* Usage is reduced ? */ |
8c7c6e34 | 2929 | if (curusage >= oldusage) |
628f4235 | 2930 | retry_count--; |
81d39c20 KH |
2931 | else |
2932 | oldusage = curusage; | |
3e32cb2e JW |
2933 | } while (retry_count); |
2934 | ||
3c11ecf4 KH |
2935 | if (!ret && enlarge) |
2936 | memcg_oom_recover(memcg); | |
3e32cb2e | 2937 | |
628f4235 KH |
2938 | return ret; |
2939 | } | |
2940 | ||
0608f43d AM |
2941 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2942 | gfp_t gfp_mask, | |
2943 | unsigned long *total_scanned) | |
2944 | { | |
2945 | unsigned long nr_reclaimed = 0; | |
2946 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2947 | unsigned long reclaimed; | |
2948 | int loop = 0; | |
2949 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2950 | unsigned long excess; |
0608f43d AM |
2951 | unsigned long nr_scanned; |
2952 | ||
2953 | if (order > 0) | |
2954 | return 0; | |
2955 | ||
2956 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2957 | /* | |
2958 | * This loop can run a while, specially if mem_cgroup's continuously | |
2959 | * keep exceeding their soft limit and putting the system under | |
2960 | * pressure | |
2961 | */ | |
2962 | do { | |
2963 | if (next_mz) | |
2964 | mz = next_mz; | |
2965 | else | |
2966 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2967 | if (!mz) | |
2968 | break; | |
2969 | ||
2970 | nr_scanned = 0; | |
2971 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2972 | gfp_mask, &nr_scanned); | |
2973 | nr_reclaimed += reclaimed; | |
2974 | *total_scanned += nr_scanned; | |
0a31bc97 | 2975 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2976 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2977 | |
2978 | /* | |
2979 | * If we failed to reclaim anything from this memory cgroup | |
2980 | * it is time to move on to the next cgroup | |
2981 | */ | |
2982 | next_mz = NULL; | |
bc2f2e7f VD |
2983 | if (!reclaimed) |
2984 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2985 | ||
3e32cb2e | 2986 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2987 | /* |
2988 | * One school of thought says that we should not add | |
2989 | * back the node to the tree if reclaim returns 0. | |
2990 | * But our reclaim could return 0, simply because due | |
2991 | * to priority we are exposing a smaller subset of | |
2992 | * memory to reclaim from. Consider this as a longer | |
2993 | * term TODO. | |
2994 | */ | |
2995 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2996 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2997 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2998 | css_put(&mz->memcg->css); |
2999 | loop++; | |
3000 | /* | |
3001 | * Could not reclaim anything and there are no more | |
3002 | * mem cgroups to try or we seem to be looping without | |
3003 | * reclaiming anything. | |
3004 | */ | |
3005 | if (!nr_reclaimed && | |
3006 | (next_mz == NULL || | |
3007 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3008 | break; | |
3009 | } while (!nr_reclaimed); | |
3010 | if (next_mz) | |
3011 | css_put(&next_mz->memcg->css); | |
3012 | return nr_reclaimed; | |
3013 | } | |
3014 | ||
ea280e7b TH |
3015 | /* |
3016 | * Test whether @memcg has children, dead or alive. Note that this | |
3017 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3018 | * returns %true if there are child csses according to the cgroup | |
3019 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3020 | */ | |
b5f99b53 GC |
3021 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3022 | { | |
ea280e7b TH |
3023 | bool ret; |
3024 | ||
696ac172 | 3025 | /* |
ea280e7b TH |
3026 | * The lock does not prevent addition or deletion of children, but |
3027 | * it prevents a new child from being initialized based on this | |
3028 | * parent in css_online(), so it's enough to decide whether | |
3029 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3030 | */ |
ea280e7b TH |
3031 | lockdep_assert_held(&memcg_create_mutex); |
3032 | ||
3033 | rcu_read_lock(); | |
3034 | ret = css_next_child(NULL, &memcg->css); | |
3035 | rcu_read_unlock(); | |
3036 | return ret; | |
b5f99b53 GC |
3037 | } |
3038 | ||
c26251f9 MH |
3039 | /* |
3040 | * Reclaims as many pages from the given memcg as possible and moves | |
3041 | * the rest to the parent. | |
3042 | * | |
3043 | * Caller is responsible for holding css reference for memcg. | |
3044 | */ | |
3045 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3046 | { | |
3047 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3048 | |
c1e862c1 KH |
3049 | /* we call try-to-free pages for make this cgroup empty */ |
3050 | lru_add_drain_all(); | |
f817ed48 | 3051 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3052 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3053 | int progress; |
c1e862c1 | 3054 | |
c26251f9 MH |
3055 | if (signal_pending(current)) |
3056 | return -EINTR; | |
3057 | ||
b70a2a21 JW |
3058 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3059 | GFP_KERNEL, true); | |
c1e862c1 | 3060 | if (!progress) { |
f817ed48 | 3061 | nr_retries--; |
c1e862c1 | 3062 | /* maybe some writeback is necessary */ |
8aa7e847 | 3063 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3064 | } |
f817ed48 KH |
3065 | |
3066 | } | |
ab5196c2 MH |
3067 | |
3068 | return 0; | |
cc847582 KH |
3069 | } |
3070 | ||
6770c64e TH |
3071 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3072 | char *buf, size_t nbytes, | |
3073 | loff_t off) | |
c1e862c1 | 3074 | { |
6770c64e | 3075 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3076 | |
d8423011 MH |
3077 | if (mem_cgroup_is_root(memcg)) |
3078 | return -EINVAL; | |
6770c64e | 3079 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3080 | } |
3081 | ||
182446d0 TH |
3082 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3083 | struct cftype *cft) | |
18f59ea7 | 3084 | { |
182446d0 | 3085 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3086 | } |
3087 | ||
182446d0 TH |
3088 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3089 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3090 | { |
3091 | int retval = 0; | |
182446d0 | 3092 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3093 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3094 | |
0999821b | 3095 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3096 | |
3097 | if (memcg->use_hierarchy == val) | |
3098 | goto out; | |
3099 | ||
18f59ea7 | 3100 | /* |
af901ca1 | 3101 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3102 | * in the child subtrees. If it is unset, then the change can |
3103 | * occur, provided the current cgroup has no children. | |
3104 | * | |
3105 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3106 | * set if there are no children. | |
3107 | */ | |
c0ff4b85 | 3108 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3109 | (val == 1 || val == 0)) { |
ea280e7b | 3110 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3111 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3112 | else |
3113 | retval = -EBUSY; | |
3114 | } else | |
3115 | retval = -EINVAL; | |
567fb435 GC |
3116 | |
3117 | out: | |
0999821b | 3118 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3119 | |
3120 | return retval; | |
3121 | } | |
3122 | ||
3e32cb2e JW |
3123 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3124 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3125 | { |
3126 | struct mem_cgroup *iter; | |
3127 | long val = 0; | |
3128 | ||
3129 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3130 | for_each_mem_cgroup_tree(iter, memcg) | |
3131 | val += mem_cgroup_read_stat(iter, idx); | |
3132 | ||
3133 | if (val < 0) /* race ? */ | |
3134 | val = 0; | |
3135 | return val; | |
3136 | } | |
3137 | ||
3138 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3139 | { | |
3140 | u64 val; | |
3141 | ||
3e32cb2e JW |
3142 | if (mem_cgroup_is_root(memcg)) { |
3143 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3144 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3145 | if (swap) | |
3146 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3147 | } else { | |
ce00a967 | 3148 | if (!swap) |
3e32cb2e | 3149 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3150 | else |
3e32cb2e | 3151 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3152 | } |
ce00a967 JW |
3153 | return val << PAGE_SHIFT; |
3154 | } | |
3155 | ||
3e32cb2e JW |
3156 | enum { |
3157 | RES_USAGE, | |
3158 | RES_LIMIT, | |
3159 | RES_MAX_USAGE, | |
3160 | RES_FAILCNT, | |
3161 | RES_SOFT_LIMIT, | |
3162 | }; | |
ce00a967 | 3163 | |
791badbd | 3164 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3165 | struct cftype *cft) |
8cdea7c0 | 3166 | { |
182446d0 | 3167 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3168 | struct page_counter *counter; |
af36f906 | 3169 | |
3e32cb2e | 3170 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3171 | case _MEM: |
3e32cb2e JW |
3172 | counter = &memcg->memory; |
3173 | break; | |
8c7c6e34 | 3174 | case _MEMSWAP: |
3e32cb2e JW |
3175 | counter = &memcg->memsw; |
3176 | break; | |
510fc4e1 | 3177 | case _KMEM: |
3e32cb2e | 3178 | counter = &memcg->kmem; |
510fc4e1 | 3179 | break; |
8c7c6e34 KH |
3180 | default: |
3181 | BUG(); | |
8c7c6e34 | 3182 | } |
3e32cb2e JW |
3183 | |
3184 | switch (MEMFILE_ATTR(cft->private)) { | |
3185 | case RES_USAGE: | |
3186 | if (counter == &memcg->memory) | |
3187 | return mem_cgroup_usage(memcg, false); | |
3188 | if (counter == &memcg->memsw) | |
3189 | return mem_cgroup_usage(memcg, true); | |
3190 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3191 | case RES_LIMIT: | |
3192 | return (u64)counter->limit * PAGE_SIZE; | |
3193 | case RES_MAX_USAGE: | |
3194 | return (u64)counter->watermark * PAGE_SIZE; | |
3195 | case RES_FAILCNT: | |
3196 | return counter->failcnt; | |
3197 | case RES_SOFT_LIMIT: | |
3198 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3199 | default: | |
3200 | BUG(); | |
3201 | } | |
8cdea7c0 | 3202 | } |
510fc4e1 | 3203 | |
510fc4e1 | 3204 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3205 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3206 | unsigned long nr_pages) | |
d6441637 VD |
3207 | { |
3208 | int err = 0; | |
3209 | int memcg_id; | |
3210 | ||
2a4db7eb | 3211 | BUG_ON(memcg->kmemcg_id >= 0); |
2788cf0c | 3212 | BUG_ON(memcg->kmem_acct_activated); |
2a4db7eb | 3213 | BUG_ON(memcg->kmem_acct_active); |
d6441637 | 3214 | |
510fc4e1 GC |
3215 | /* |
3216 | * For simplicity, we won't allow this to be disabled. It also can't | |
3217 | * be changed if the cgroup has children already, or if tasks had | |
3218 | * already joined. | |
3219 | * | |
3220 | * If tasks join before we set the limit, a person looking at | |
3221 | * kmem.usage_in_bytes will have no way to determine when it took | |
3222 | * place, which makes the value quite meaningless. | |
3223 | * | |
3224 | * After it first became limited, changes in the value of the limit are | |
3225 | * of course permitted. | |
510fc4e1 | 3226 | */ |
0999821b | 3227 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3228 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3229 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3230 | err = -EBUSY; |
3231 | mutex_unlock(&memcg_create_mutex); | |
3232 | if (err) | |
3233 | goto out; | |
510fc4e1 | 3234 | |
f3bb3043 | 3235 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3236 | if (memcg_id < 0) { |
3237 | err = memcg_id; | |
3238 | goto out; | |
3239 | } | |
3240 | ||
d6441637 | 3241 | /* |
900a38f0 VD |
3242 | * We couldn't have accounted to this cgroup, because it hasn't got |
3243 | * activated yet, so this should succeed. | |
d6441637 | 3244 | */ |
3e32cb2e | 3245 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3246 | VM_BUG_ON(err); |
3247 | ||
3248 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3249 | /* | |
900a38f0 VD |
3250 | * A memory cgroup is considered kmem-active as soon as it gets |
3251 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
3252 | * guarantee no one starts accounting before all call sites are |
3253 | * patched. | |
3254 | */ | |
900a38f0 | 3255 | memcg->kmemcg_id = memcg_id; |
2788cf0c | 3256 | memcg->kmem_acct_activated = true; |
2a4db7eb | 3257 | memcg->kmem_acct_active = true; |
510fc4e1 | 3258 | out: |
d6441637 | 3259 | return err; |
d6441637 VD |
3260 | } |
3261 | ||
d6441637 | 3262 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3263 | unsigned long limit) |
d6441637 VD |
3264 | { |
3265 | int ret; | |
3266 | ||
3e32cb2e | 3267 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3268 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3269 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3270 | else |
3e32cb2e JW |
3271 | ret = page_counter_limit(&memcg->kmem, limit); |
3272 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3273 | return ret; |
3274 | } | |
3275 | ||
55007d84 | 3276 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3277 | { |
55007d84 | 3278 | int ret = 0; |
510fc4e1 | 3279 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3280 | |
d6441637 VD |
3281 | if (!parent) |
3282 | return 0; | |
55007d84 | 3283 | |
8c0145b6 | 3284 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3285 | /* |
d6441637 VD |
3286 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3287 | * after this point, because it has at least one child already. | |
55007d84 | 3288 | */ |
d6441637 | 3289 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3290 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3291 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3292 | return ret; |
510fc4e1 | 3293 | } |
d6441637 VD |
3294 | #else |
3295 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3296 | unsigned long limit) |
d6441637 VD |
3297 | { |
3298 | return -EINVAL; | |
3299 | } | |
6d043990 | 3300 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3301 | |
628f4235 KH |
3302 | /* |
3303 | * The user of this function is... | |
3304 | * RES_LIMIT. | |
3305 | */ | |
451af504 TH |
3306 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3307 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3308 | { |
451af504 | 3309 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3310 | unsigned long nr_pages; |
628f4235 KH |
3311 | int ret; |
3312 | ||
451af504 | 3313 | buf = strstrip(buf); |
650c5e56 | 3314 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3315 | if (ret) |
3316 | return ret; | |
af36f906 | 3317 | |
3e32cb2e | 3318 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3319 | case RES_LIMIT: |
4b3bde4c BS |
3320 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3321 | ret = -EINVAL; | |
3322 | break; | |
3323 | } | |
3e32cb2e JW |
3324 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3325 | case _MEM: | |
3326 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3327 | break; |
3e32cb2e JW |
3328 | case _MEMSWAP: |
3329 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3330 | break; |
3e32cb2e JW |
3331 | case _KMEM: |
3332 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3333 | break; | |
3334 | } | |
296c81d8 | 3335 | break; |
3e32cb2e JW |
3336 | case RES_SOFT_LIMIT: |
3337 | memcg->soft_limit = nr_pages; | |
3338 | ret = 0; | |
628f4235 KH |
3339 | break; |
3340 | } | |
451af504 | 3341 | return ret ?: nbytes; |
8cdea7c0 BS |
3342 | } |
3343 | ||
6770c64e TH |
3344 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3345 | size_t nbytes, loff_t off) | |
c84872e1 | 3346 | { |
6770c64e | 3347 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3348 | struct page_counter *counter; |
c84872e1 | 3349 | |
3e32cb2e JW |
3350 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3351 | case _MEM: | |
3352 | counter = &memcg->memory; | |
3353 | break; | |
3354 | case _MEMSWAP: | |
3355 | counter = &memcg->memsw; | |
3356 | break; | |
3357 | case _KMEM: | |
3358 | counter = &memcg->kmem; | |
3359 | break; | |
3360 | default: | |
3361 | BUG(); | |
3362 | } | |
af36f906 | 3363 | |
3e32cb2e | 3364 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3365 | case RES_MAX_USAGE: |
3e32cb2e | 3366 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3367 | break; |
3368 | case RES_FAILCNT: | |
3e32cb2e | 3369 | counter->failcnt = 0; |
29f2a4da | 3370 | break; |
3e32cb2e JW |
3371 | default: |
3372 | BUG(); | |
29f2a4da | 3373 | } |
f64c3f54 | 3374 | |
6770c64e | 3375 | return nbytes; |
c84872e1 PE |
3376 | } |
3377 | ||
182446d0 | 3378 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3379 | struct cftype *cft) |
3380 | { | |
182446d0 | 3381 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3382 | } |
3383 | ||
02491447 | 3384 | #ifdef CONFIG_MMU |
182446d0 | 3385 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3386 | struct cftype *cft, u64 val) |
3387 | { | |
182446d0 | 3388 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3389 | |
1dfab5ab | 3390 | if (val & ~MOVE_MASK) |
7dc74be0 | 3391 | return -EINVAL; |
ee5e8472 | 3392 | |
7dc74be0 | 3393 | /* |
ee5e8472 GC |
3394 | * No kind of locking is needed in here, because ->can_attach() will |
3395 | * check this value once in the beginning of the process, and then carry | |
3396 | * on with stale data. This means that changes to this value will only | |
3397 | * affect task migrations starting after the change. | |
7dc74be0 | 3398 | */ |
c0ff4b85 | 3399 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3400 | return 0; |
3401 | } | |
02491447 | 3402 | #else |
182446d0 | 3403 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3404 | struct cftype *cft, u64 val) |
3405 | { | |
3406 | return -ENOSYS; | |
3407 | } | |
3408 | #endif | |
7dc74be0 | 3409 | |
406eb0c9 | 3410 | #ifdef CONFIG_NUMA |
2da8ca82 | 3411 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3412 | { |
25485de6 GT |
3413 | struct numa_stat { |
3414 | const char *name; | |
3415 | unsigned int lru_mask; | |
3416 | }; | |
3417 | ||
3418 | static const struct numa_stat stats[] = { | |
3419 | { "total", LRU_ALL }, | |
3420 | { "file", LRU_ALL_FILE }, | |
3421 | { "anon", LRU_ALL_ANON }, | |
3422 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3423 | }; | |
3424 | const struct numa_stat *stat; | |
406eb0c9 | 3425 | int nid; |
25485de6 | 3426 | unsigned long nr; |
2da8ca82 | 3427 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3428 | |
25485de6 GT |
3429 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3430 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3431 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3432 | for_each_node_state(nid, N_MEMORY) { | |
3433 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3434 | stat->lru_mask); | |
3435 | seq_printf(m, " N%d=%lu", nid, nr); | |
3436 | } | |
3437 | seq_putc(m, '\n'); | |
406eb0c9 | 3438 | } |
406eb0c9 | 3439 | |
071aee13 YH |
3440 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3441 | struct mem_cgroup *iter; | |
3442 | ||
3443 | nr = 0; | |
3444 | for_each_mem_cgroup_tree(iter, memcg) | |
3445 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3446 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3447 | for_each_node_state(nid, N_MEMORY) { | |
3448 | nr = 0; | |
3449 | for_each_mem_cgroup_tree(iter, memcg) | |
3450 | nr += mem_cgroup_node_nr_lru_pages( | |
3451 | iter, nid, stat->lru_mask); | |
3452 | seq_printf(m, " N%d=%lu", nid, nr); | |
3453 | } | |
3454 | seq_putc(m, '\n'); | |
406eb0c9 | 3455 | } |
406eb0c9 | 3456 | |
406eb0c9 YH |
3457 | return 0; |
3458 | } | |
3459 | #endif /* CONFIG_NUMA */ | |
3460 | ||
2da8ca82 | 3461 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3462 | { |
2da8ca82 | 3463 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3464 | unsigned long memory, memsw; |
af7c4b0e JW |
3465 | struct mem_cgroup *mi; |
3466 | unsigned int i; | |
406eb0c9 | 3467 | |
0ca44b14 GT |
3468 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3469 | MEM_CGROUP_STAT_NSTATS); | |
3470 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3471 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3472 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3473 | ||
af7c4b0e | 3474 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3475 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3476 | continue; |
af7c4b0e JW |
3477 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3478 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3479 | } |
7b854121 | 3480 | |
af7c4b0e JW |
3481 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3482 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3483 | mem_cgroup_read_events(memcg, i)); | |
3484 | ||
3485 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3486 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3487 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3488 | ||
14067bb3 | 3489 | /* Hierarchical information */ |
3e32cb2e JW |
3490 | memory = memsw = PAGE_COUNTER_MAX; |
3491 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3492 | memory = min(memory, mi->memory.limit); | |
3493 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3494 | } |
3e32cb2e JW |
3495 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3496 | (u64)memory * PAGE_SIZE); | |
3497 | if (do_swap_account) | |
3498 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3499 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3500 | |
af7c4b0e JW |
3501 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3502 | long long val = 0; | |
3503 | ||
bff6bb83 | 3504 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3505 | continue; |
af7c4b0e JW |
3506 | for_each_mem_cgroup_tree(mi, memcg) |
3507 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3508 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3509 | } | |
3510 | ||
3511 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3512 | unsigned long long val = 0; | |
3513 | ||
3514 | for_each_mem_cgroup_tree(mi, memcg) | |
3515 | val += mem_cgroup_read_events(mi, i); | |
3516 | seq_printf(m, "total_%s %llu\n", | |
3517 | mem_cgroup_events_names[i], val); | |
3518 | } | |
3519 | ||
3520 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3521 | unsigned long long val = 0; | |
3522 | ||
3523 | for_each_mem_cgroup_tree(mi, memcg) | |
3524 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3525 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3526 | } |
14067bb3 | 3527 | |
7f016ee8 | 3528 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3529 | { |
3530 | int nid, zid; | |
3531 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3532 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3533 | unsigned long recent_rotated[2] = {0, 0}; |
3534 | unsigned long recent_scanned[2] = {0, 0}; | |
3535 | ||
3536 | for_each_online_node(nid) | |
3537 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3538 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3539 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3540 | |
89abfab1 HD |
3541 | recent_rotated[0] += rstat->recent_rotated[0]; |
3542 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3543 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3544 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3545 | } |
78ccf5b5 JW |
3546 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3547 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3548 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3549 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3550 | } |
3551 | #endif | |
3552 | ||
d2ceb9b7 KH |
3553 | return 0; |
3554 | } | |
3555 | ||
182446d0 TH |
3556 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3557 | struct cftype *cft) | |
a7885eb8 | 3558 | { |
182446d0 | 3559 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3560 | |
1f4c025b | 3561 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3562 | } |
3563 | ||
182446d0 TH |
3564 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3565 | struct cftype *cft, u64 val) | |
a7885eb8 | 3566 | { |
182446d0 | 3567 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3568 | |
3dae7fec | 3569 | if (val > 100) |
a7885eb8 KM |
3570 | return -EINVAL; |
3571 | ||
14208b0e | 3572 | if (css->parent) |
3dae7fec JW |
3573 | memcg->swappiness = val; |
3574 | else | |
3575 | vm_swappiness = val; | |
068b38c1 | 3576 | |
a7885eb8 KM |
3577 | return 0; |
3578 | } | |
3579 | ||
2e72b634 KS |
3580 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3581 | { | |
3582 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3583 | unsigned long usage; |
2e72b634 KS |
3584 | int i; |
3585 | ||
3586 | rcu_read_lock(); | |
3587 | if (!swap) | |
2c488db2 | 3588 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3589 | else |
2c488db2 | 3590 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3591 | |
3592 | if (!t) | |
3593 | goto unlock; | |
3594 | ||
ce00a967 | 3595 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3596 | |
3597 | /* | |
748dad36 | 3598 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3599 | * If it's not true, a threshold was crossed after last |
3600 | * call of __mem_cgroup_threshold(). | |
3601 | */ | |
5407a562 | 3602 | i = t->current_threshold; |
2e72b634 KS |
3603 | |
3604 | /* | |
3605 | * Iterate backward over array of thresholds starting from | |
3606 | * current_threshold and check if a threshold is crossed. | |
3607 | * If none of thresholds below usage is crossed, we read | |
3608 | * only one element of the array here. | |
3609 | */ | |
3610 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3611 | eventfd_signal(t->entries[i].eventfd, 1); | |
3612 | ||
3613 | /* i = current_threshold + 1 */ | |
3614 | i++; | |
3615 | ||
3616 | /* | |
3617 | * Iterate forward over array of thresholds starting from | |
3618 | * current_threshold+1 and check if a threshold is crossed. | |
3619 | * If none of thresholds above usage is crossed, we read | |
3620 | * only one element of the array here. | |
3621 | */ | |
3622 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3623 | eventfd_signal(t->entries[i].eventfd, 1); | |
3624 | ||
3625 | /* Update current_threshold */ | |
5407a562 | 3626 | t->current_threshold = i - 1; |
2e72b634 KS |
3627 | unlock: |
3628 | rcu_read_unlock(); | |
3629 | } | |
3630 | ||
3631 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3632 | { | |
ad4ca5f4 KS |
3633 | while (memcg) { |
3634 | __mem_cgroup_threshold(memcg, false); | |
3635 | if (do_swap_account) | |
3636 | __mem_cgroup_threshold(memcg, true); | |
3637 | ||
3638 | memcg = parent_mem_cgroup(memcg); | |
3639 | } | |
2e72b634 KS |
3640 | } |
3641 | ||
3642 | static int compare_thresholds(const void *a, const void *b) | |
3643 | { | |
3644 | const struct mem_cgroup_threshold *_a = a; | |
3645 | const struct mem_cgroup_threshold *_b = b; | |
3646 | ||
2bff24a3 GT |
3647 | if (_a->threshold > _b->threshold) |
3648 | return 1; | |
3649 | ||
3650 | if (_a->threshold < _b->threshold) | |
3651 | return -1; | |
3652 | ||
3653 | return 0; | |
2e72b634 KS |
3654 | } |
3655 | ||
c0ff4b85 | 3656 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3657 | { |
3658 | struct mem_cgroup_eventfd_list *ev; | |
3659 | ||
2bcf2e92 MH |
3660 | spin_lock(&memcg_oom_lock); |
3661 | ||
c0ff4b85 | 3662 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3663 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3664 | |
3665 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3666 | return 0; |
3667 | } | |
3668 | ||
c0ff4b85 | 3669 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3670 | { |
7d74b06f KH |
3671 | struct mem_cgroup *iter; |
3672 | ||
c0ff4b85 | 3673 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3674 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3675 | } |
3676 | ||
59b6f873 | 3677 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3678 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3679 | { |
2c488db2 KS |
3680 | struct mem_cgroup_thresholds *thresholds; |
3681 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3682 | unsigned long threshold; |
3683 | unsigned long usage; | |
2c488db2 | 3684 | int i, size, ret; |
2e72b634 | 3685 | |
650c5e56 | 3686 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3687 | if (ret) |
3688 | return ret; | |
3689 | ||
3690 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3691 | |
05b84301 | 3692 | if (type == _MEM) { |
2c488db2 | 3693 | thresholds = &memcg->thresholds; |
ce00a967 | 3694 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3695 | } else if (type == _MEMSWAP) { |
2c488db2 | 3696 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3697 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3698 | } else |
2e72b634 KS |
3699 | BUG(); |
3700 | ||
2e72b634 | 3701 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3702 | if (thresholds->primary) |
2e72b634 KS |
3703 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3704 | ||
2c488db2 | 3705 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3706 | |
3707 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3708 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3709 | GFP_KERNEL); |
2c488db2 | 3710 | if (!new) { |
2e72b634 KS |
3711 | ret = -ENOMEM; |
3712 | goto unlock; | |
3713 | } | |
2c488db2 | 3714 | new->size = size; |
2e72b634 KS |
3715 | |
3716 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3717 | if (thresholds->primary) { |
3718 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3719 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3720 | } |
3721 | ||
2e72b634 | 3722 | /* Add new threshold */ |
2c488db2 KS |
3723 | new->entries[size - 1].eventfd = eventfd; |
3724 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3725 | |
3726 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3727 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3728 | compare_thresholds, NULL); |
3729 | ||
3730 | /* Find current threshold */ | |
2c488db2 | 3731 | new->current_threshold = -1; |
2e72b634 | 3732 | for (i = 0; i < size; i++) { |
748dad36 | 3733 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3734 | /* |
2c488db2 KS |
3735 | * new->current_threshold will not be used until |
3736 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3737 | * it here. |
3738 | */ | |
2c488db2 | 3739 | ++new->current_threshold; |
748dad36 SZ |
3740 | } else |
3741 | break; | |
2e72b634 KS |
3742 | } |
3743 | ||
2c488db2 KS |
3744 | /* Free old spare buffer and save old primary buffer as spare */ |
3745 | kfree(thresholds->spare); | |
3746 | thresholds->spare = thresholds->primary; | |
3747 | ||
3748 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3749 | |
907860ed | 3750 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3751 | synchronize_rcu(); |
3752 | ||
2e72b634 KS |
3753 | unlock: |
3754 | mutex_unlock(&memcg->thresholds_lock); | |
3755 | ||
3756 | return ret; | |
3757 | } | |
3758 | ||
59b6f873 | 3759 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3760 | struct eventfd_ctx *eventfd, const char *args) |
3761 | { | |
59b6f873 | 3762 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3763 | } |
3764 | ||
59b6f873 | 3765 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3766 | struct eventfd_ctx *eventfd, const char *args) |
3767 | { | |
59b6f873 | 3768 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3769 | } |
3770 | ||
59b6f873 | 3771 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3772 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3773 | { |
2c488db2 KS |
3774 | struct mem_cgroup_thresholds *thresholds; |
3775 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3776 | unsigned long usage; |
2c488db2 | 3777 | int i, j, size; |
2e72b634 KS |
3778 | |
3779 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3780 | |
3781 | if (type == _MEM) { | |
2c488db2 | 3782 | thresholds = &memcg->thresholds; |
ce00a967 | 3783 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3784 | } else if (type == _MEMSWAP) { |
2c488db2 | 3785 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3786 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3787 | } else |
2e72b634 KS |
3788 | BUG(); |
3789 | ||
371528ca AV |
3790 | if (!thresholds->primary) |
3791 | goto unlock; | |
3792 | ||
2e72b634 KS |
3793 | /* Check if a threshold crossed before removing */ |
3794 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3795 | ||
3796 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3797 | size = 0; |
3798 | for (i = 0; i < thresholds->primary->size; i++) { | |
3799 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3800 | size++; |
3801 | } | |
3802 | ||
2c488db2 | 3803 | new = thresholds->spare; |
907860ed | 3804 | |
2e72b634 KS |
3805 | /* Set thresholds array to NULL if we don't have thresholds */ |
3806 | if (!size) { | |
2c488db2 KS |
3807 | kfree(new); |
3808 | new = NULL; | |
907860ed | 3809 | goto swap_buffers; |
2e72b634 KS |
3810 | } |
3811 | ||
2c488db2 | 3812 | new->size = size; |
2e72b634 KS |
3813 | |
3814 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3815 | new->current_threshold = -1; |
3816 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3817 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3818 | continue; |
3819 | ||
2c488db2 | 3820 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3821 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3822 | /* |
2c488db2 | 3823 | * new->current_threshold will not be used |
2e72b634 KS |
3824 | * until rcu_assign_pointer(), so it's safe to increment |
3825 | * it here. | |
3826 | */ | |
2c488db2 | 3827 | ++new->current_threshold; |
2e72b634 KS |
3828 | } |
3829 | j++; | |
3830 | } | |
3831 | ||
907860ed | 3832 | swap_buffers: |
2c488db2 KS |
3833 | /* Swap primary and spare array */ |
3834 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
3835 | /* If all events are unregistered, free the spare array */ |
3836 | if (!new) { | |
3837 | kfree(thresholds->spare); | |
3838 | thresholds->spare = NULL; | |
3839 | } | |
3840 | ||
2c488db2 | 3841 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3842 | |
907860ed | 3843 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3844 | synchronize_rcu(); |
371528ca | 3845 | unlock: |
2e72b634 | 3846 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3847 | } |
c1e862c1 | 3848 | |
59b6f873 | 3849 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3850 | struct eventfd_ctx *eventfd) |
3851 | { | |
59b6f873 | 3852 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3853 | } |
3854 | ||
59b6f873 | 3855 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3856 | struct eventfd_ctx *eventfd) |
3857 | { | |
59b6f873 | 3858 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3859 | } |
3860 | ||
59b6f873 | 3861 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3862 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3863 | { |
9490ff27 | 3864 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3865 | |
9490ff27 KH |
3866 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3867 | if (!event) | |
3868 | return -ENOMEM; | |
3869 | ||
1af8efe9 | 3870 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3871 | |
3872 | event->eventfd = eventfd; | |
3873 | list_add(&event->list, &memcg->oom_notify); | |
3874 | ||
3875 | /* already in OOM ? */ | |
c2b42d3c | 3876 | if (memcg->under_oom) |
9490ff27 | 3877 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3878 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3879 | |
3880 | return 0; | |
3881 | } | |
3882 | ||
59b6f873 | 3883 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3884 | struct eventfd_ctx *eventfd) |
9490ff27 | 3885 | { |
9490ff27 | 3886 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3887 | |
1af8efe9 | 3888 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3889 | |
c0ff4b85 | 3890 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3891 | if (ev->eventfd == eventfd) { |
3892 | list_del(&ev->list); | |
3893 | kfree(ev); | |
3894 | } | |
3895 | } | |
3896 | ||
1af8efe9 | 3897 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3898 | } |
3899 | ||
2da8ca82 | 3900 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3901 | { |
2da8ca82 | 3902 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3903 | |
791badbd | 3904 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3905 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3906 | return 0; |
3907 | } | |
3908 | ||
182446d0 | 3909 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3910 | struct cftype *cft, u64 val) |
3911 | { | |
182446d0 | 3912 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3913 | |
3914 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3915 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3916 | return -EINVAL; |
3917 | ||
c0ff4b85 | 3918 | memcg->oom_kill_disable = val; |
4d845ebf | 3919 | if (!val) |
c0ff4b85 | 3920 | memcg_oom_recover(memcg); |
3dae7fec | 3921 | |
3c11ecf4 KH |
3922 | return 0; |
3923 | } | |
3924 | ||
c255a458 | 3925 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 3926 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 3927 | { |
55007d84 GC |
3928 | int ret; |
3929 | ||
55007d84 GC |
3930 | ret = memcg_propagate_kmem(memcg); |
3931 | if (ret) | |
3932 | return ret; | |
2633d7a0 | 3933 | |
1d62e436 | 3934 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 3935 | } |
e5671dfa | 3936 | |
2a4db7eb VD |
3937 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3938 | { | |
2788cf0c VD |
3939 | struct cgroup_subsys_state *css; |
3940 | struct mem_cgroup *parent, *child; | |
3941 | int kmemcg_id; | |
3942 | ||
2a4db7eb VD |
3943 | if (!memcg->kmem_acct_active) |
3944 | return; | |
3945 | ||
3946 | /* | |
3947 | * Clear the 'active' flag before clearing memcg_caches arrays entries. | |
3948 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it | |
3949 | * guarantees no cache will be created for this cgroup after we are | |
3950 | * done (see memcg_create_kmem_cache()). | |
3951 | */ | |
3952 | memcg->kmem_acct_active = false; | |
3953 | ||
3954 | memcg_deactivate_kmem_caches(memcg); | |
2788cf0c VD |
3955 | |
3956 | kmemcg_id = memcg->kmemcg_id; | |
3957 | BUG_ON(kmemcg_id < 0); | |
3958 | ||
3959 | parent = parent_mem_cgroup(memcg); | |
3960 | if (!parent) | |
3961 | parent = root_mem_cgroup; | |
3962 | ||
3963 | /* | |
3964 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3965 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3966 | * to ones of the parent. After we have finished, all list_lrus | |
3967 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3968 | * ordering is imposed by list_lru_node->lock taken by | |
3969 | * memcg_drain_all_list_lrus(). | |
3970 | */ | |
3971 | css_for_each_descendant_pre(css, &memcg->css) { | |
3972 | child = mem_cgroup_from_css(css); | |
3973 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3974 | child->kmemcg_id = parent->kmemcg_id; | |
3975 | if (!memcg->use_hierarchy) | |
3976 | break; | |
3977 | } | |
3978 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
3979 | ||
3980 | memcg_free_cache_id(kmemcg_id); | |
2a4db7eb VD |
3981 | } |
3982 | ||
10d5ebf4 | 3983 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 3984 | { |
f48b80a5 VD |
3985 | if (memcg->kmem_acct_activated) { |
3986 | memcg_destroy_kmem_caches(memcg); | |
3987 | static_key_slow_dec(&memcg_kmem_enabled_key); | |
3988 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3989 | } | |
1d62e436 | 3990 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 3991 | } |
e5671dfa | 3992 | #else |
cbe128e3 | 3993 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
3994 | { |
3995 | return 0; | |
3996 | } | |
d1a4c0b3 | 3997 | |
2a4db7eb VD |
3998 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3999 | { | |
4000 | } | |
4001 | ||
10d5ebf4 LZ |
4002 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4003 | { | |
4004 | } | |
e5671dfa GC |
4005 | #endif |
4006 | ||
52ebea74 TH |
4007 | #ifdef CONFIG_CGROUP_WRITEBACK |
4008 | ||
4009 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
4010 | { | |
4011 | return &memcg->cgwb_list; | |
4012 | } | |
4013 | ||
841710aa TH |
4014 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4015 | { | |
4016 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4017 | } | |
4018 | ||
4019 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4020 | { | |
4021 | wb_domain_exit(&memcg->cgwb_domain); | |
4022 | } | |
4023 | ||
2529bb3a TH |
4024 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4025 | { | |
4026 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4027 | } | |
4028 | ||
841710aa TH |
4029 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4030 | { | |
4031 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4032 | ||
4033 | if (!memcg->css.parent) | |
4034 | return NULL; | |
4035 | ||
4036 | return &memcg->cgwb_domain; | |
4037 | } | |
4038 | ||
c2aa723a TH |
4039 | /** |
4040 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4041 | * @wb: bdi_writeback in question | |
4042 | * @pavail: out parameter for number of available pages | |
4043 | * @pdirty: out parameter for number of dirty pages | |
4044 | * @pwriteback: out parameter for number of pages under writeback | |
4045 | * | |
4046 | * Determine the numbers of available, dirty, and writeback pages in @wb's | |
4047 | * memcg. Dirty and writeback are self-explanatory. Available is a bit | |
4048 | * more involved. | |
4049 | * | |
4050 | * A memcg's headroom is "min(max, high) - used". The available memory is | |
4051 | * calculated as the lowest headroom of itself and the ancestors plus the | |
4052 | * number of pages already being used for file pages. Note that this | |
4053 | * doesn't consider the actual amount of available memory in the system. | |
4054 | * The caller should further cap *@pavail accordingly. | |
4055 | */ | |
4056 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pavail, | |
4057 | unsigned long *pdirty, unsigned long *pwriteback) | |
4058 | { | |
4059 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4060 | struct mem_cgroup *parent; | |
4061 | unsigned long head_room = PAGE_COUNTER_MAX; | |
4062 | unsigned long file_pages; | |
4063 | ||
4064 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
4065 | ||
4066 | /* this should eventually include NR_UNSTABLE_NFS */ | |
4067 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
4068 | ||
4069 | file_pages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | | |
4070 | (1 << LRU_ACTIVE_FILE)); | |
4071 | while ((parent = parent_mem_cgroup(memcg))) { | |
4072 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
4073 | unsigned long used = page_counter_read(&memcg->memory); | |
4074 | ||
4075 | head_room = min(head_room, ceiling - min(ceiling, used)); | |
4076 | memcg = parent; | |
4077 | } | |
4078 | ||
4079 | *pavail = file_pages + head_room; | |
4080 | } | |
4081 | ||
841710aa TH |
4082 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4083 | ||
4084 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4085 | { | |
4086 | return 0; | |
4087 | } | |
4088 | ||
4089 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4090 | { | |
4091 | } | |
4092 | ||
2529bb3a TH |
4093 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4094 | { | |
4095 | } | |
4096 | ||
52ebea74 TH |
4097 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4098 | ||
3bc942f3 TH |
4099 | /* |
4100 | * DO NOT USE IN NEW FILES. | |
4101 | * | |
4102 | * "cgroup.event_control" implementation. | |
4103 | * | |
4104 | * This is way over-engineered. It tries to support fully configurable | |
4105 | * events for each user. Such level of flexibility is completely | |
4106 | * unnecessary especially in the light of the planned unified hierarchy. | |
4107 | * | |
4108 | * Please deprecate this and replace with something simpler if at all | |
4109 | * possible. | |
4110 | */ | |
4111 | ||
79bd9814 TH |
4112 | /* |
4113 | * Unregister event and free resources. | |
4114 | * | |
4115 | * Gets called from workqueue. | |
4116 | */ | |
3bc942f3 | 4117 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4118 | { |
3bc942f3 TH |
4119 | struct mem_cgroup_event *event = |
4120 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4121 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4122 | |
4123 | remove_wait_queue(event->wqh, &event->wait); | |
4124 | ||
59b6f873 | 4125 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4126 | |
4127 | /* Notify userspace the event is going away. */ | |
4128 | eventfd_signal(event->eventfd, 1); | |
4129 | ||
4130 | eventfd_ctx_put(event->eventfd); | |
4131 | kfree(event); | |
59b6f873 | 4132 | css_put(&memcg->css); |
79bd9814 TH |
4133 | } |
4134 | ||
4135 | /* | |
4136 | * Gets called on POLLHUP on eventfd when user closes it. | |
4137 | * | |
4138 | * Called with wqh->lock held and interrupts disabled. | |
4139 | */ | |
3bc942f3 TH |
4140 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4141 | int sync, void *key) | |
79bd9814 | 4142 | { |
3bc942f3 TH |
4143 | struct mem_cgroup_event *event = |
4144 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4145 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4146 | unsigned long flags = (unsigned long)key; |
4147 | ||
4148 | if (flags & POLLHUP) { | |
4149 | /* | |
4150 | * If the event has been detached at cgroup removal, we | |
4151 | * can simply return knowing the other side will cleanup | |
4152 | * for us. | |
4153 | * | |
4154 | * We can't race against event freeing since the other | |
4155 | * side will require wqh->lock via remove_wait_queue(), | |
4156 | * which we hold. | |
4157 | */ | |
fba94807 | 4158 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4159 | if (!list_empty(&event->list)) { |
4160 | list_del_init(&event->list); | |
4161 | /* | |
4162 | * We are in atomic context, but cgroup_event_remove() | |
4163 | * may sleep, so we have to call it in workqueue. | |
4164 | */ | |
4165 | schedule_work(&event->remove); | |
4166 | } | |
fba94807 | 4167 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4168 | } |
4169 | ||
4170 | return 0; | |
4171 | } | |
4172 | ||
3bc942f3 | 4173 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4174 | wait_queue_head_t *wqh, poll_table *pt) |
4175 | { | |
3bc942f3 TH |
4176 | struct mem_cgroup_event *event = |
4177 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4178 | |
4179 | event->wqh = wqh; | |
4180 | add_wait_queue(wqh, &event->wait); | |
4181 | } | |
4182 | ||
4183 | /* | |
3bc942f3 TH |
4184 | * DO NOT USE IN NEW FILES. |
4185 | * | |
79bd9814 TH |
4186 | * Parse input and register new cgroup event handler. |
4187 | * | |
4188 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4189 | * Interpretation of args is defined by control file implementation. | |
4190 | */ | |
451af504 TH |
4191 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4192 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4193 | { |
451af504 | 4194 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4195 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4196 | struct mem_cgroup_event *event; |
79bd9814 TH |
4197 | struct cgroup_subsys_state *cfile_css; |
4198 | unsigned int efd, cfd; | |
4199 | struct fd efile; | |
4200 | struct fd cfile; | |
fba94807 | 4201 | const char *name; |
79bd9814 TH |
4202 | char *endp; |
4203 | int ret; | |
4204 | ||
451af504 TH |
4205 | buf = strstrip(buf); |
4206 | ||
4207 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4208 | if (*endp != ' ') |
4209 | return -EINVAL; | |
451af504 | 4210 | buf = endp + 1; |
79bd9814 | 4211 | |
451af504 | 4212 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4213 | if ((*endp != ' ') && (*endp != '\0')) |
4214 | return -EINVAL; | |
451af504 | 4215 | buf = endp + 1; |
79bd9814 TH |
4216 | |
4217 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4218 | if (!event) | |
4219 | return -ENOMEM; | |
4220 | ||
59b6f873 | 4221 | event->memcg = memcg; |
79bd9814 | 4222 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4223 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4224 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4225 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4226 | |
4227 | efile = fdget(efd); | |
4228 | if (!efile.file) { | |
4229 | ret = -EBADF; | |
4230 | goto out_kfree; | |
4231 | } | |
4232 | ||
4233 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4234 | if (IS_ERR(event->eventfd)) { | |
4235 | ret = PTR_ERR(event->eventfd); | |
4236 | goto out_put_efile; | |
4237 | } | |
4238 | ||
4239 | cfile = fdget(cfd); | |
4240 | if (!cfile.file) { | |
4241 | ret = -EBADF; | |
4242 | goto out_put_eventfd; | |
4243 | } | |
4244 | ||
4245 | /* the process need read permission on control file */ | |
4246 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4247 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4248 | if (ret < 0) | |
4249 | goto out_put_cfile; | |
4250 | ||
fba94807 TH |
4251 | /* |
4252 | * Determine the event callbacks and set them in @event. This used | |
4253 | * to be done via struct cftype but cgroup core no longer knows | |
4254 | * about these events. The following is crude but the whole thing | |
4255 | * is for compatibility anyway. | |
3bc942f3 TH |
4256 | * |
4257 | * DO NOT ADD NEW FILES. | |
fba94807 | 4258 | */ |
b583043e | 4259 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4260 | |
4261 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4262 | event->register_event = mem_cgroup_usage_register_event; | |
4263 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4264 | } else if (!strcmp(name, "memory.oom_control")) { | |
4265 | event->register_event = mem_cgroup_oom_register_event; | |
4266 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4267 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4268 | event->register_event = vmpressure_register_event; | |
4269 | event->unregister_event = vmpressure_unregister_event; | |
4270 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4271 | event->register_event = memsw_cgroup_usage_register_event; |
4272 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4273 | } else { |
4274 | ret = -EINVAL; | |
4275 | goto out_put_cfile; | |
4276 | } | |
4277 | ||
79bd9814 | 4278 | /* |
b5557c4c TH |
4279 | * Verify @cfile should belong to @css. Also, remaining events are |
4280 | * automatically removed on cgroup destruction but the removal is | |
4281 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4282 | */ |
b583043e | 4283 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4284 | &memory_cgrp_subsys); |
79bd9814 | 4285 | ret = -EINVAL; |
5a17f543 | 4286 | if (IS_ERR(cfile_css)) |
79bd9814 | 4287 | goto out_put_cfile; |
5a17f543 TH |
4288 | if (cfile_css != css) { |
4289 | css_put(cfile_css); | |
79bd9814 | 4290 | goto out_put_cfile; |
5a17f543 | 4291 | } |
79bd9814 | 4292 | |
451af504 | 4293 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4294 | if (ret) |
4295 | goto out_put_css; | |
4296 | ||
4297 | efile.file->f_op->poll(efile.file, &event->pt); | |
4298 | ||
fba94807 TH |
4299 | spin_lock(&memcg->event_list_lock); |
4300 | list_add(&event->list, &memcg->event_list); | |
4301 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4302 | |
4303 | fdput(cfile); | |
4304 | fdput(efile); | |
4305 | ||
451af504 | 4306 | return nbytes; |
79bd9814 TH |
4307 | |
4308 | out_put_css: | |
b5557c4c | 4309 | css_put(css); |
79bd9814 TH |
4310 | out_put_cfile: |
4311 | fdput(cfile); | |
4312 | out_put_eventfd: | |
4313 | eventfd_ctx_put(event->eventfd); | |
4314 | out_put_efile: | |
4315 | fdput(efile); | |
4316 | out_kfree: | |
4317 | kfree(event); | |
4318 | ||
4319 | return ret; | |
4320 | } | |
4321 | ||
241994ed | 4322 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4323 | { |
0eea1030 | 4324 | .name = "usage_in_bytes", |
8c7c6e34 | 4325 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4326 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4327 | }, |
c84872e1 PE |
4328 | { |
4329 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4330 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4331 | .write = mem_cgroup_reset, |
791badbd | 4332 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4333 | }, |
8cdea7c0 | 4334 | { |
0eea1030 | 4335 | .name = "limit_in_bytes", |
8c7c6e34 | 4336 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4337 | .write = mem_cgroup_write, |
791badbd | 4338 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4339 | }, |
296c81d8 BS |
4340 | { |
4341 | .name = "soft_limit_in_bytes", | |
4342 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4343 | .write = mem_cgroup_write, |
791badbd | 4344 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4345 | }, |
8cdea7c0 BS |
4346 | { |
4347 | .name = "failcnt", | |
8c7c6e34 | 4348 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4349 | .write = mem_cgroup_reset, |
791badbd | 4350 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4351 | }, |
d2ceb9b7 KH |
4352 | { |
4353 | .name = "stat", | |
2da8ca82 | 4354 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4355 | }, |
c1e862c1 KH |
4356 | { |
4357 | .name = "force_empty", | |
6770c64e | 4358 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4359 | }, |
18f59ea7 BS |
4360 | { |
4361 | .name = "use_hierarchy", | |
4362 | .write_u64 = mem_cgroup_hierarchy_write, | |
4363 | .read_u64 = mem_cgroup_hierarchy_read, | |
4364 | }, | |
79bd9814 | 4365 | { |
3bc942f3 | 4366 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4367 | .write = memcg_write_event_control, |
79bd9814 TH |
4368 | .flags = CFTYPE_NO_PREFIX, |
4369 | .mode = S_IWUGO, | |
4370 | }, | |
a7885eb8 KM |
4371 | { |
4372 | .name = "swappiness", | |
4373 | .read_u64 = mem_cgroup_swappiness_read, | |
4374 | .write_u64 = mem_cgroup_swappiness_write, | |
4375 | }, | |
7dc74be0 DN |
4376 | { |
4377 | .name = "move_charge_at_immigrate", | |
4378 | .read_u64 = mem_cgroup_move_charge_read, | |
4379 | .write_u64 = mem_cgroup_move_charge_write, | |
4380 | }, | |
9490ff27 KH |
4381 | { |
4382 | .name = "oom_control", | |
2da8ca82 | 4383 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4384 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4385 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4386 | }, | |
70ddf637 AV |
4387 | { |
4388 | .name = "pressure_level", | |
70ddf637 | 4389 | }, |
406eb0c9 YH |
4390 | #ifdef CONFIG_NUMA |
4391 | { | |
4392 | .name = "numa_stat", | |
2da8ca82 | 4393 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4394 | }, |
4395 | #endif | |
510fc4e1 GC |
4396 | #ifdef CONFIG_MEMCG_KMEM |
4397 | { | |
4398 | .name = "kmem.limit_in_bytes", | |
4399 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4400 | .write = mem_cgroup_write, |
791badbd | 4401 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4402 | }, |
4403 | { | |
4404 | .name = "kmem.usage_in_bytes", | |
4405 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4406 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4407 | }, |
4408 | { | |
4409 | .name = "kmem.failcnt", | |
4410 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4411 | .write = mem_cgroup_reset, |
791badbd | 4412 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4413 | }, |
4414 | { | |
4415 | .name = "kmem.max_usage_in_bytes", | |
4416 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4417 | .write = mem_cgroup_reset, |
791badbd | 4418 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4419 | }, |
749c5415 GC |
4420 | #ifdef CONFIG_SLABINFO |
4421 | { | |
4422 | .name = "kmem.slabinfo", | |
b047501c VD |
4423 | .seq_start = slab_start, |
4424 | .seq_next = slab_next, | |
4425 | .seq_stop = slab_stop, | |
4426 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4427 | }, |
4428 | #endif | |
8c7c6e34 | 4429 | #endif |
6bc10349 | 4430 | { }, /* terminate */ |
af36f906 | 4431 | }; |
8c7c6e34 | 4432 | |
c0ff4b85 | 4433 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4434 | { |
4435 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4436 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4437 | int zone, tmp = node; |
1ecaab2b KH |
4438 | /* |
4439 | * This routine is called against possible nodes. | |
4440 | * But it's BUG to call kmalloc() against offline node. | |
4441 | * | |
4442 | * TODO: this routine can waste much memory for nodes which will | |
4443 | * never be onlined. It's better to use memory hotplug callback | |
4444 | * function. | |
4445 | */ | |
41e3355d KH |
4446 | if (!node_state(node, N_NORMAL_MEMORY)) |
4447 | tmp = -1; | |
17295c88 | 4448 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4449 | if (!pn) |
4450 | return 1; | |
1ecaab2b | 4451 | |
1ecaab2b KH |
4452 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4453 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4454 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4455 | mz->usage_in_excess = 0; |
4456 | mz->on_tree = false; | |
d79154bb | 4457 | mz->memcg = memcg; |
1ecaab2b | 4458 | } |
54f72fe0 | 4459 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4460 | return 0; |
4461 | } | |
4462 | ||
c0ff4b85 | 4463 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4464 | { |
54f72fe0 | 4465 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4466 | } |
4467 | ||
33327948 KH |
4468 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4469 | { | |
d79154bb | 4470 | struct mem_cgroup *memcg; |
8ff69e2c | 4471 | size_t size; |
33327948 | 4472 | |
8ff69e2c VD |
4473 | size = sizeof(struct mem_cgroup); |
4474 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4475 | |
8ff69e2c | 4476 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4477 | if (!memcg) |
e7bbcdf3 DC |
4478 | return NULL; |
4479 | ||
d79154bb HD |
4480 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4481 | if (!memcg->stat) | |
d2e61b8d | 4482 | goto out_free; |
841710aa TH |
4483 | |
4484 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) | |
4485 | goto out_free_stat; | |
4486 | ||
d79154bb HD |
4487 | spin_lock_init(&memcg->pcp_counter_lock); |
4488 | return memcg; | |
d2e61b8d | 4489 | |
841710aa TH |
4490 | out_free_stat: |
4491 | free_percpu(memcg->stat); | |
d2e61b8d | 4492 | out_free: |
8ff69e2c | 4493 | kfree(memcg); |
d2e61b8d | 4494 | return NULL; |
33327948 KH |
4495 | } |
4496 | ||
59927fb9 | 4497 | /* |
c8b2a36f GC |
4498 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4499 | * (scanning all at force_empty is too costly...) | |
4500 | * | |
4501 | * Instead of clearing all references at force_empty, we remember | |
4502 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4503 | * it goes down to 0. | |
4504 | * | |
4505 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4506 | */ |
c8b2a36f GC |
4507 | |
4508 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4509 | { |
c8b2a36f | 4510 | int node; |
59927fb9 | 4511 | |
bb4cc1a8 | 4512 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4513 | |
4514 | for_each_node(node) | |
4515 | free_mem_cgroup_per_zone_info(memcg, node); | |
4516 | ||
4517 | free_percpu(memcg->stat); | |
841710aa | 4518 | memcg_wb_domain_exit(memcg); |
8ff69e2c | 4519 | kfree(memcg); |
59927fb9 | 4520 | } |
3afe36b1 | 4521 | |
7bcc1bb1 DN |
4522 | /* |
4523 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4524 | */ | |
e1aab161 | 4525 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4526 | { |
3e32cb2e | 4527 | if (!memcg->memory.parent) |
7bcc1bb1 | 4528 | return NULL; |
3e32cb2e | 4529 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4530 | } |
e1aab161 | 4531 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4532 | |
0eb253e2 | 4533 | static struct cgroup_subsys_state * __ref |
eb95419b | 4534 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4535 | { |
d142e3e6 | 4536 | struct mem_cgroup *memcg; |
04046e1a | 4537 | long error = -ENOMEM; |
6d12e2d8 | 4538 | int node; |
8cdea7c0 | 4539 | |
c0ff4b85 R |
4540 | memcg = mem_cgroup_alloc(); |
4541 | if (!memcg) | |
04046e1a | 4542 | return ERR_PTR(error); |
78fb7466 | 4543 | |
3ed28fa1 | 4544 | for_each_node(node) |
c0ff4b85 | 4545 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4546 | goto free_out; |
f64c3f54 | 4547 | |
c077719b | 4548 | /* root ? */ |
eb95419b | 4549 | if (parent_css == NULL) { |
a41c58a6 | 4550 | root_mem_cgroup = memcg; |
56161634 | 4551 | mem_cgroup_root_css = &memcg->css; |
3e32cb2e | 4552 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4553 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4554 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4555 | page_counter_init(&memcg->memsw, NULL); |
4556 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4557 | } |
28dbc4b6 | 4558 | |
d142e3e6 GC |
4559 | memcg->last_scanned_node = MAX_NUMNODES; |
4560 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4561 | memcg->move_charge_at_immigrate = 0; |
4562 | mutex_init(&memcg->thresholds_lock); | |
4563 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4564 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4565 | INIT_LIST_HEAD(&memcg->event_list); |
4566 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4567 | #ifdef CONFIG_MEMCG_KMEM |
4568 | memcg->kmemcg_id = -1; | |
900a38f0 | 4569 | #endif |
52ebea74 TH |
4570 | #ifdef CONFIG_CGROUP_WRITEBACK |
4571 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4572 | #endif | |
d142e3e6 GC |
4573 | return &memcg->css; |
4574 | ||
4575 | free_out: | |
4576 | __mem_cgroup_free(memcg); | |
4577 | return ERR_PTR(error); | |
4578 | } | |
4579 | ||
4580 | static int | |
eb95419b | 4581 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4582 | { |
eb95419b | 4583 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4584 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4585 | int ret; |
d142e3e6 | 4586 | |
15a4c835 | 4587 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4588 | return -ENOSPC; |
4589 | ||
63876986 | 4590 | if (!parent) |
d142e3e6 GC |
4591 | return 0; |
4592 | ||
0999821b | 4593 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4594 | |
4595 | memcg->use_hierarchy = parent->use_hierarchy; | |
4596 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4597 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4598 | ||
4599 | if (parent->use_hierarchy) { | |
3e32cb2e | 4600 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4601 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4602 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4603 | page_counter_init(&memcg->memsw, &parent->memsw); |
4604 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4605 | |
7bcc1bb1 | 4606 | /* |
8d76a979 LZ |
4607 | * No need to take a reference to the parent because cgroup |
4608 | * core guarantees its existence. | |
7bcc1bb1 | 4609 | */ |
18f59ea7 | 4610 | } else { |
3e32cb2e | 4611 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4612 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4613 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4614 | page_counter_init(&memcg->memsw, NULL); |
4615 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4616 | /* |
4617 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4618 | * much sense so let cgroup subsystem know about this | |
4619 | * unfortunate state in our controller. | |
4620 | */ | |
d142e3e6 | 4621 | if (parent != root_mem_cgroup) |
073219e9 | 4622 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4623 | } |
0999821b | 4624 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4625 | |
2f7dd7a4 JW |
4626 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4627 | if (ret) | |
4628 | return ret; | |
4629 | ||
4630 | /* | |
4631 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4632 | * orders reading memcg->initialized against its callers | |
4633 | * reading the memcg members. | |
4634 | */ | |
4635 | smp_store_release(&memcg->initialized, 1); | |
4636 | ||
4637 | return 0; | |
8cdea7c0 BS |
4638 | } |
4639 | ||
eb95419b | 4640 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4641 | { |
eb95419b | 4642 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4643 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4644 | |
4645 | /* | |
4646 | * Unregister events and notify userspace. | |
4647 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4648 | * directory to avoid race between userspace and kernelspace. | |
4649 | */ | |
fba94807 TH |
4650 | spin_lock(&memcg->event_list_lock); |
4651 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4652 | list_del_init(&event->list); |
4653 | schedule_work(&event->remove); | |
4654 | } | |
fba94807 | 4655 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4656 | |
33cb876e | 4657 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb VD |
4658 | |
4659 | memcg_deactivate_kmem(memcg); | |
52ebea74 TH |
4660 | |
4661 | wb_memcg_offline(memcg); | |
df878fb0 KH |
4662 | } |
4663 | ||
eb95419b | 4664 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4665 | { |
eb95419b | 4666 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4667 | |
10d5ebf4 | 4668 | memcg_destroy_kmem(memcg); |
465939a1 | 4669 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4670 | } |
4671 | ||
1ced953b TH |
4672 | /** |
4673 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4674 | * @css: the target css | |
4675 | * | |
4676 | * Reset the states of the mem_cgroup associated with @css. This is | |
4677 | * invoked when the userland requests disabling on the default hierarchy | |
4678 | * but the memcg is pinned through dependency. The memcg should stop | |
4679 | * applying policies and should revert to the vanilla state as it may be | |
4680 | * made visible again. | |
4681 | * | |
4682 | * The current implementation only resets the essential configurations. | |
4683 | * This needs to be expanded to cover all the visible parts. | |
4684 | */ | |
4685 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4686 | { | |
4687 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4688 | ||
3e32cb2e JW |
4689 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4690 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4691 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4692 | memcg->low = 0; |
4693 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4694 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4695 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4696 | } |
4697 | ||
02491447 | 4698 | #ifdef CONFIG_MMU |
7dc74be0 | 4699 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4700 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4701 | { |
05b84301 | 4702 | int ret; |
9476db97 JW |
4703 | |
4704 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 4705 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 4706 | if (!ret) { |
854ffa8d | 4707 | mc.precharge += count; |
854ffa8d DN |
4708 | return ret; |
4709 | } | |
692e7c45 | 4710 | if (ret == -EINTR) { |
00501b53 | 4711 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
4712 | return ret; |
4713 | } | |
9476db97 JW |
4714 | |
4715 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4716 | while (count--) { |
00501b53 | 4717 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
4718 | /* |
4719 | * In case of failure, any residual charges against | |
4720 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
4721 | * later on. However, cancel any charges that are |
4722 | * bypassed to root right away or they'll be lost. | |
9476db97 | 4723 | */ |
692e7c45 | 4724 | if (ret == -EINTR) |
00501b53 | 4725 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 4726 | if (ret) |
38c5d72f | 4727 | return ret; |
854ffa8d | 4728 | mc.precharge++; |
9476db97 | 4729 | cond_resched(); |
854ffa8d | 4730 | } |
9476db97 | 4731 | return 0; |
4ffef5fe DN |
4732 | } |
4733 | ||
4734 | /** | |
8d32ff84 | 4735 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4736 | * @vma: the vma the pte to be checked belongs |
4737 | * @addr: the address corresponding to the pte to be checked | |
4738 | * @ptent: the pte to be checked | |
02491447 | 4739 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4740 | * |
4741 | * Returns | |
4742 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4743 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4744 | * move charge. if @target is not NULL, the page is stored in target->page | |
4745 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4746 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4747 | * target for charge migration. if @target is not NULL, the entry is stored | |
4748 | * in target->ent. | |
4ffef5fe DN |
4749 | * |
4750 | * Called with pte lock held. | |
4751 | */ | |
4ffef5fe DN |
4752 | union mc_target { |
4753 | struct page *page; | |
02491447 | 4754 | swp_entry_t ent; |
4ffef5fe DN |
4755 | }; |
4756 | ||
4ffef5fe | 4757 | enum mc_target_type { |
8d32ff84 | 4758 | MC_TARGET_NONE = 0, |
4ffef5fe | 4759 | MC_TARGET_PAGE, |
02491447 | 4760 | MC_TARGET_SWAP, |
4ffef5fe DN |
4761 | }; |
4762 | ||
90254a65 DN |
4763 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4764 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4765 | { |
90254a65 | 4766 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4767 | |
90254a65 DN |
4768 | if (!page || !page_mapped(page)) |
4769 | return NULL; | |
4770 | if (PageAnon(page)) { | |
1dfab5ab | 4771 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4772 | return NULL; |
1dfab5ab JW |
4773 | } else { |
4774 | if (!(mc.flags & MOVE_FILE)) | |
4775 | return NULL; | |
4776 | } | |
90254a65 DN |
4777 | if (!get_page_unless_zero(page)) |
4778 | return NULL; | |
4779 | ||
4780 | return page; | |
4781 | } | |
4782 | ||
4b91355e | 4783 | #ifdef CONFIG_SWAP |
90254a65 DN |
4784 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4785 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4786 | { | |
90254a65 DN |
4787 | struct page *page = NULL; |
4788 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4789 | ||
1dfab5ab | 4790 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4791 | return NULL; |
4b91355e KH |
4792 | /* |
4793 | * Because lookup_swap_cache() updates some statistics counter, | |
4794 | * we call find_get_page() with swapper_space directly. | |
4795 | */ | |
33806f06 | 4796 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4797 | if (do_swap_account) |
4798 | entry->val = ent.val; | |
4799 | ||
4800 | return page; | |
4801 | } | |
4b91355e KH |
4802 | #else |
4803 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4804 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4805 | { | |
4806 | return NULL; | |
4807 | } | |
4808 | #endif | |
90254a65 | 4809 | |
87946a72 DN |
4810 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4811 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4812 | { | |
4813 | struct page *page = NULL; | |
87946a72 DN |
4814 | struct address_space *mapping; |
4815 | pgoff_t pgoff; | |
4816 | ||
4817 | if (!vma->vm_file) /* anonymous vma */ | |
4818 | return NULL; | |
1dfab5ab | 4819 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4820 | return NULL; |
4821 | ||
87946a72 | 4822 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4823 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4824 | |
4825 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4826 | #ifdef CONFIG_SWAP |
4827 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4828 | if (shmem_mapping(mapping)) { |
4829 | page = find_get_entry(mapping, pgoff); | |
4830 | if (radix_tree_exceptional_entry(page)) { | |
4831 | swp_entry_t swp = radix_to_swp_entry(page); | |
4832 | if (do_swap_account) | |
4833 | *entry = swp; | |
4834 | page = find_get_page(swap_address_space(swp), swp.val); | |
4835 | } | |
4836 | } else | |
4837 | page = find_get_page(mapping, pgoff); | |
4838 | #else | |
4839 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4840 | #endif |
87946a72 DN |
4841 | return page; |
4842 | } | |
4843 | ||
b1b0deab CG |
4844 | /** |
4845 | * mem_cgroup_move_account - move account of the page | |
4846 | * @page: the page | |
4847 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4848 | * @from: mem_cgroup which the page is moved from. | |
4849 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4850 | * | |
4851 | * The caller must confirm following. | |
4852 | * - page is not on LRU (isolate_page() is useful.) | |
4853 | * - compound_lock is held when nr_pages > 1 | |
4854 | * | |
4855 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4856 | * from old cgroup. | |
4857 | */ | |
4858 | static int mem_cgroup_move_account(struct page *page, | |
4859 | unsigned int nr_pages, | |
4860 | struct mem_cgroup *from, | |
4861 | struct mem_cgroup *to) | |
4862 | { | |
4863 | unsigned long flags; | |
4864 | int ret; | |
c4843a75 | 4865 | bool anon; |
b1b0deab CG |
4866 | |
4867 | VM_BUG_ON(from == to); | |
4868 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
4869 | /* | |
4870 | * The page is isolated from LRU. So, collapse function | |
4871 | * will not handle this page. But page splitting can happen. | |
4872 | * Do this check under compound_page_lock(). The caller should | |
4873 | * hold it. | |
4874 | */ | |
4875 | ret = -EBUSY; | |
4876 | if (nr_pages > 1 && !PageTransHuge(page)) | |
4877 | goto out; | |
4878 | ||
4879 | /* | |
4880 | * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup | |
4881 | * of its source page while we change it: page migration takes | |
4882 | * both pages off the LRU, but page cache replacement doesn't. | |
4883 | */ | |
4884 | if (!trylock_page(page)) | |
4885 | goto out; | |
4886 | ||
4887 | ret = -EINVAL; | |
4888 | if (page->mem_cgroup != from) | |
4889 | goto out_unlock; | |
4890 | ||
c4843a75 GT |
4891 | anon = PageAnon(page); |
4892 | ||
b1b0deab CG |
4893 | spin_lock_irqsave(&from->move_lock, flags); |
4894 | ||
c4843a75 | 4895 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4896 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4897 | nr_pages); | |
4898 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4899 | nr_pages); | |
4900 | } | |
4901 | ||
c4843a75 GT |
4902 | /* |
4903 | * move_lock grabbed above and caller set from->moving_account, so | |
4904 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4905 | * So mapping should be stable for dirty pages. | |
4906 | */ | |
4907 | if (!anon && PageDirty(page)) { | |
4908 | struct address_space *mapping = page_mapping(page); | |
4909 | ||
4910 | if (mapping_cap_account_dirty(mapping)) { | |
4911 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4912 | nr_pages); | |
4913 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4914 | nr_pages); | |
4915 | } | |
4916 | } | |
4917 | ||
b1b0deab CG |
4918 | if (PageWriteback(page)) { |
4919 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4920 | nr_pages); | |
4921 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4922 | nr_pages); | |
4923 | } | |
4924 | ||
4925 | /* | |
4926 | * It is safe to change page->mem_cgroup here because the page | |
4927 | * is referenced, charged, and isolated - we can't race with | |
4928 | * uncharging, charging, migration, or LRU putback. | |
4929 | */ | |
4930 | ||
4931 | /* caller should have done css_get */ | |
4932 | page->mem_cgroup = to; | |
4933 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4934 | ||
4935 | ret = 0; | |
4936 | ||
4937 | local_irq_disable(); | |
4938 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
4939 | memcg_check_events(to, page); | |
4940 | mem_cgroup_charge_statistics(from, page, -nr_pages); | |
4941 | memcg_check_events(from, page); | |
4942 | local_irq_enable(); | |
4943 | out_unlock: | |
4944 | unlock_page(page); | |
4945 | out: | |
4946 | return ret; | |
4947 | } | |
4948 | ||
8d32ff84 | 4949 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4950 | unsigned long addr, pte_t ptent, union mc_target *target) |
4951 | { | |
4952 | struct page *page = NULL; | |
8d32ff84 | 4953 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4954 | swp_entry_t ent = { .val = 0 }; |
4955 | ||
4956 | if (pte_present(ptent)) | |
4957 | page = mc_handle_present_pte(vma, addr, ptent); | |
4958 | else if (is_swap_pte(ptent)) | |
4959 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4960 | else if (pte_none(ptent)) |
87946a72 | 4961 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4962 | |
4963 | if (!page && !ent.val) | |
8d32ff84 | 4964 | return ret; |
02491447 | 4965 | if (page) { |
02491447 | 4966 | /* |
0a31bc97 | 4967 | * Do only loose check w/o serialization. |
1306a85a | 4968 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4969 | * not under LRU exclusion. |
02491447 | 4970 | */ |
1306a85a | 4971 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4972 | ret = MC_TARGET_PAGE; |
4973 | if (target) | |
4974 | target->page = page; | |
4975 | } | |
4976 | if (!ret || !target) | |
4977 | put_page(page); | |
4978 | } | |
90254a65 DN |
4979 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4980 | if (ent.val && !ret && | |
34c00c31 | 4981 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4982 | ret = MC_TARGET_SWAP; |
4983 | if (target) | |
4984 | target->ent = ent; | |
4ffef5fe | 4985 | } |
4ffef5fe DN |
4986 | return ret; |
4987 | } | |
4988 | ||
12724850 NH |
4989 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4990 | /* | |
4991 | * We don't consider swapping or file mapped pages because THP does not | |
4992 | * support them for now. | |
4993 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4994 | */ | |
4995 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4996 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4997 | { | |
4998 | struct page *page = NULL; | |
12724850 NH |
4999 | enum mc_target_type ret = MC_TARGET_NONE; |
5000 | ||
5001 | page = pmd_page(pmd); | |
309381fe | 5002 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5003 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5004 | return ret; |
1306a85a | 5005 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5006 | ret = MC_TARGET_PAGE; |
5007 | if (target) { | |
5008 | get_page(page); | |
5009 | target->page = page; | |
5010 | } | |
5011 | } | |
5012 | return ret; | |
5013 | } | |
5014 | #else | |
5015 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5016 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5017 | { | |
5018 | return MC_TARGET_NONE; | |
5019 | } | |
5020 | #endif | |
5021 | ||
4ffef5fe DN |
5022 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5023 | unsigned long addr, unsigned long end, | |
5024 | struct mm_walk *walk) | |
5025 | { | |
26bcd64a | 5026 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5027 | pte_t *pte; |
5028 | spinlock_t *ptl; | |
5029 | ||
bf929152 | 5030 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5031 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5032 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5033 | spin_unlock(ptl); |
1a5a9906 | 5034 | return 0; |
12724850 | 5035 | } |
03319327 | 5036 | |
45f83cef AA |
5037 | if (pmd_trans_unstable(pmd)) |
5038 | return 0; | |
4ffef5fe DN |
5039 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5040 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5041 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5042 | mc.precharge++; /* increment precharge temporarily */ |
5043 | pte_unmap_unlock(pte - 1, ptl); | |
5044 | cond_resched(); | |
5045 | ||
7dc74be0 DN |
5046 | return 0; |
5047 | } | |
5048 | ||
4ffef5fe DN |
5049 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5050 | { | |
5051 | unsigned long precharge; | |
4ffef5fe | 5052 | |
26bcd64a NH |
5053 | struct mm_walk mem_cgroup_count_precharge_walk = { |
5054 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5055 | .mm = mm, | |
5056 | }; | |
dfe076b0 | 5057 | down_read(&mm->mmap_sem); |
26bcd64a | 5058 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 5059 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5060 | |
5061 | precharge = mc.precharge; | |
5062 | mc.precharge = 0; | |
5063 | ||
5064 | return precharge; | |
5065 | } | |
5066 | ||
4ffef5fe DN |
5067 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5068 | { | |
dfe076b0 DN |
5069 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5070 | ||
5071 | VM_BUG_ON(mc.moving_task); | |
5072 | mc.moving_task = current; | |
5073 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5074 | } |
5075 | ||
dfe076b0 DN |
5076 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5077 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5078 | { |
2bd9bb20 KH |
5079 | struct mem_cgroup *from = mc.from; |
5080 | struct mem_cgroup *to = mc.to; | |
5081 | ||
4ffef5fe | 5082 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5083 | if (mc.precharge) { |
00501b53 | 5084 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5085 | mc.precharge = 0; |
5086 | } | |
5087 | /* | |
5088 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5089 | * we must uncharge here. | |
5090 | */ | |
5091 | if (mc.moved_charge) { | |
00501b53 | 5092 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5093 | mc.moved_charge = 0; |
4ffef5fe | 5094 | } |
483c30b5 DN |
5095 | /* we must fixup refcnts and charges */ |
5096 | if (mc.moved_swap) { | |
483c30b5 | 5097 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5098 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5099 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5100 | |
05b84301 | 5101 | /* |
3e32cb2e JW |
5102 | * we charged both to->memory and to->memsw, so we |
5103 | * should uncharge to->memory. | |
05b84301 | 5104 | */ |
ce00a967 | 5105 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5106 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5107 | ||
e8ea14cc | 5108 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 5109 | |
4050377b | 5110 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5111 | mc.moved_swap = 0; |
5112 | } | |
dfe076b0 DN |
5113 | memcg_oom_recover(from); |
5114 | memcg_oom_recover(to); | |
5115 | wake_up_all(&mc.waitq); | |
5116 | } | |
5117 | ||
5118 | static void mem_cgroup_clear_mc(void) | |
5119 | { | |
dfe076b0 DN |
5120 | /* |
5121 | * we must clear moving_task before waking up waiters at the end of | |
5122 | * task migration. | |
5123 | */ | |
5124 | mc.moving_task = NULL; | |
5125 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5126 | spin_lock(&mc.lock); |
4ffef5fe DN |
5127 | mc.from = NULL; |
5128 | mc.to = NULL; | |
2bd9bb20 | 5129 | spin_unlock(&mc.lock); |
4ffef5fe DN |
5130 | } |
5131 | ||
eb95419b | 5132 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5133 | struct cgroup_taskset *tset) |
7dc74be0 | 5134 | { |
2f7ee569 | 5135 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5136 | int ret = 0; |
eb95419b | 5137 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
1dfab5ab | 5138 | unsigned long move_flags; |
7dc74be0 | 5139 | |
ee5e8472 GC |
5140 | /* |
5141 | * We are now commited to this value whatever it is. Changes in this | |
5142 | * tunable will only affect upcoming migrations, not the current one. | |
5143 | * So we need to save it, and keep it going. | |
5144 | */ | |
4db0c3c2 | 5145 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); |
1dfab5ab | 5146 | if (move_flags) { |
7dc74be0 DN |
5147 | struct mm_struct *mm; |
5148 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5149 | ||
c0ff4b85 | 5150 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5151 | |
5152 | mm = get_task_mm(p); | |
5153 | if (!mm) | |
5154 | return 0; | |
7dc74be0 | 5155 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5156 | if (mm->owner == p) { |
5157 | VM_BUG_ON(mc.from); | |
5158 | VM_BUG_ON(mc.to); | |
5159 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5160 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5161 | VM_BUG_ON(mc.moved_swap); |
247b1447 | 5162 | |
2bd9bb20 | 5163 | spin_lock(&mc.lock); |
4ffef5fe | 5164 | mc.from = from; |
c0ff4b85 | 5165 | mc.to = memcg; |
1dfab5ab | 5166 | mc.flags = move_flags; |
2bd9bb20 | 5167 | spin_unlock(&mc.lock); |
dfe076b0 | 5168 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5169 | |
5170 | ret = mem_cgroup_precharge_mc(mm); | |
5171 | if (ret) | |
5172 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5173 | } |
5174 | mmput(mm); | |
7dc74be0 DN |
5175 | } |
5176 | return ret; | |
5177 | } | |
5178 | ||
eb95419b | 5179 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5180 | struct cgroup_taskset *tset) |
7dc74be0 | 5181 | { |
4e2f245d JW |
5182 | if (mc.to) |
5183 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5184 | } |
5185 | ||
4ffef5fe DN |
5186 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5187 | unsigned long addr, unsigned long end, | |
5188 | struct mm_walk *walk) | |
7dc74be0 | 5189 | { |
4ffef5fe | 5190 | int ret = 0; |
26bcd64a | 5191 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5192 | pte_t *pte; |
5193 | spinlock_t *ptl; | |
12724850 NH |
5194 | enum mc_target_type target_type; |
5195 | union mc_target target; | |
5196 | struct page *page; | |
4ffef5fe | 5197 | |
12724850 NH |
5198 | /* |
5199 | * We don't take compound_lock() here but no race with splitting thp | |
5200 | * happens because: | |
5201 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5202 | * under splitting, which means there's no concurrent thp split, | |
5203 | * - if another thread runs into split_huge_page() just after we | |
5204 | * entered this if-block, the thread must wait for page table lock | |
5205 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5206 | * part of thp split is not executed yet. | |
5207 | */ | |
bf929152 | 5208 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5209 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5210 | spin_unlock(ptl); |
12724850 NH |
5211 | return 0; |
5212 | } | |
5213 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5214 | if (target_type == MC_TARGET_PAGE) { | |
5215 | page = target.page; | |
5216 | if (!isolate_lru_page(page)) { | |
12724850 | 5217 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 5218 | mc.from, mc.to)) { |
12724850 NH |
5219 | mc.precharge -= HPAGE_PMD_NR; |
5220 | mc.moved_charge += HPAGE_PMD_NR; | |
5221 | } | |
5222 | putback_lru_page(page); | |
5223 | } | |
5224 | put_page(page); | |
5225 | } | |
bf929152 | 5226 | spin_unlock(ptl); |
1a5a9906 | 5227 | return 0; |
12724850 NH |
5228 | } |
5229 | ||
45f83cef AA |
5230 | if (pmd_trans_unstable(pmd)) |
5231 | return 0; | |
4ffef5fe DN |
5232 | retry: |
5233 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5234 | for (; addr != end; addr += PAGE_SIZE) { | |
5235 | pte_t ptent = *(pte++); | |
02491447 | 5236 | swp_entry_t ent; |
4ffef5fe DN |
5237 | |
5238 | if (!mc.precharge) | |
5239 | break; | |
5240 | ||
8d32ff84 | 5241 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5242 | case MC_TARGET_PAGE: |
5243 | page = target.page; | |
5244 | if (isolate_lru_page(page)) | |
5245 | goto put; | |
1306a85a | 5246 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 5247 | mc.precharge--; |
854ffa8d DN |
5248 | /* we uncharge from mc.from later. */ |
5249 | mc.moved_charge++; | |
4ffef5fe DN |
5250 | } |
5251 | putback_lru_page(page); | |
8d32ff84 | 5252 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5253 | put_page(page); |
5254 | break; | |
02491447 DN |
5255 | case MC_TARGET_SWAP: |
5256 | ent = target.ent; | |
e91cbb42 | 5257 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5258 | mc.precharge--; |
483c30b5 DN |
5259 | /* we fixup refcnts and charges later. */ |
5260 | mc.moved_swap++; | |
5261 | } | |
02491447 | 5262 | break; |
4ffef5fe DN |
5263 | default: |
5264 | break; | |
5265 | } | |
5266 | } | |
5267 | pte_unmap_unlock(pte - 1, ptl); | |
5268 | cond_resched(); | |
5269 | ||
5270 | if (addr != end) { | |
5271 | /* | |
5272 | * We have consumed all precharges we got in can_attach(). | |
5273 | * We try charge one by one, but don't do any additional | |
5274 | * charges to mc.to if we have failed in charge once in attach() | |
5275 | * phase. | |
5276 | */ | |
854ffa8d | 5277 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5278 | if (!ret) |
5279 | goto retry; | |
5280 | } | |
5281 | ||
5282 | return ret; | |
5283 | } | |
5284 | ||
5285 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5286 | { | |
26bcd64a NH |
5287 | struct mm_walk mem_cgroup_move_charge_walk = { |
5288 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5289 | .mm = mm, | |
5290 | }; | |
4ffef5fe DN |
5291 | |
5292 | lru_add_drain_all(); | |
312722cb JW |
5293 | /* |
5294 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5295 | * move_lock while we're moving its pages to another memcg. | |
5296 | * Then wait for already started RCU-only updates to finish. | |
5297 | */ | |
5298 | atomic_inc(&mc.from->moving_account); | |
5299 | synchronize_rcu(); | |
dfe076b0 DN |
5300 | retry: |
5301 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5302 | /* | |
5303 | * Someone who are holding the mmap_sem might be waiting in | |
5304 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5305 | * and retry. Because we cancel precharges, we might not be able | |
5306 | * to move enough charges, but moving charge is a best-effort | |
5307 | * feature anyway, so it wouldn't be a big problem. | |
5308 | */ | |
5309 | __mem_cgroup_clear_mc(); | |
5310 | cond_resched(); | |
5311 | goto retry; | |
5312 | } | |
26bcd64a NH |
5313 | /* |
5314 | * When we have consumed all precharges and failed in doing | |
5315 | * additional charge, the page walk just aborts. | |
5316 | */ | |
5317 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 5318 | up_read(&mm->mmap_sem); |
312722cb | 5319 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5320 | } |
5321 | ||
eb95419b | 5322 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5323 | struct cgroup_taskset *tset) |
67e465a7 | 5324 | { |
2f7ee569 | 5325 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5326 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5327 | |
dfe076b0 | 5328 | if (mm) { |
a433658c KM |
5329 | if (mc.to) |
5330 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5331 | mmput(mm); |
5332 | } | |
a433658c KM |
5333 | if (mc.to) |
5334 | mem_cgroup_clear_mc(); | |
67e465a7 | 5335 | } |
5cfb80a7 | 5336 | #else /* !CONFIG_MMU */ |
eb95419b | 5337 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5338 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5339 | { |
5340 | return 0; | |
5341 | } | |
eb95419b | 5342 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5343 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5344 | { |
5345 | } | |
eb95419b | 5346 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5347 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5348 | { |
5349 | } | |
5350 | #endif | |
67e465a7 | 5351 | |
f00baae7 TH |
5352 | /* |
5353 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5354 | * to verify whether we're attached to the default hierarchy on each mount |
5355 | * attempt. | |
f00baae7 | 5356 | */ |
eb95419b | 5357 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5358 | { |
5359 | /* | |
aa6ec29b | 5360 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5361 | * guarantees that @root doesn't have any children, so turning it |
5362 | * on for the root memcg is enough. | |
5363 | */ | |
aa6ec29b | 5364 | if (cgroup_on_dfl(root_css->cgroup)) |
7feee590 VD |
5365 | root_mem_cgroup->use_hierarchy = true; |
5366 | else | |
5367 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5368 | } |
5369 | ||
241994ed JW |
5370 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5371 | struct cftype *cft) | |
5372 | { | |
5373 | return mem_cgroup_usage(mem_cgroup_from_css(css), false); | |
5374 | } | |
5375 | ||
5376 | static int memory_low_show(struct seq_file *m, void *v) | |
5377 | { | |
5378 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5379 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5380 | |
5381 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5382 | seq_puts(m, "max\n"); |
241994ed JW |
5383 | else |
5384 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5385 | ||
5386 | return 0; | |
5387 | } | |
5388 | ||
5389 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5390 | char *buf, size_t nbytes, loff_t off) | |
5391 | { | |
5392 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5393 | unsigned long low; | |
5394 | int err; | |
5395 | ||
5396 | buf = strstrip(buf); | |
d2973697 | 5397 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5398 | if (err) |
5399 | return err; | |
5400 | ||
5401 | memcg->low = low; | |
5402 | ||
5403 | return nbytes; | |
5404 | } | |
5405 | ||
5406 | static int memory_high_show(struct seq_file *m, void *v) | |
5407 | { | |
5408 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5409 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5410 | |
5411 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5412 | seq_puts(m, "max\n"); |
241994ed JW |
5413 | else |
5414 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5415 | ||
5416 | return 0; | |
5417 | } | |
5418 | ||
5419 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5420 | char *buf, size_t nbytes, loff_t off) | |
5421 | { | |
5422 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5423 | unsigned long high; | |
5424 | int err; | |
5425 | ||
5426 | buf = strstrip(buf); | |
d2973697 | 5427 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5428 | if (err) |
5429 | return err; | |
5430 | ||
5431 | memcg->high = high; | |
5432 | ||
2529bb3a | 5433 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5434 | return nbytes; |
5435 | } | |
5436 | ||
5437 | static int memory_max_show(struct seq_file *m, void *v) | |
5438 | { | |
5439 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5440 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5441 | |
5442 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5443 | seq_puts(m, "max\n"); |
241994ed JW |
5444 | else |
5445 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5446 | ||
5447 | return 0; | |
5448 | } | |
5449 | ||
5450 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5451 | char *buf, size_t nbytes, loff_t off) | |
5452 | { | |
5453 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5454 | unsigned long max; | |
5455 | int err; | |
5456 | ||
5457 | buf = strstrip(buf); | |
d2973697 | 5458 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5459 | if (err) |
5460 | return err; | |
5461 | ||
5462 | err = mem_cgroup_resize_limit(memcg, max); | |
5463 | if (err) | |
5464 | return err; | |
5465 | ||
2529bb3a | 5466 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5467 | return nbytes; |
5468 | } | |
5469 | ||
5470 | static int memory_events_show(struct seq_file *m, void *v) | |
5471 | { | |
5472 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5473 | ||
5474 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5475 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5476 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5477 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5478 | ||
5479 | return 0; | |
5480 | } | |
5481 | ||
5482 | static struct cftype memory_files[] = { | |
5483 | { | |
5484 | .name = "current", | |
5485 | .read_u64 = memory_current_read, | |
5486 | }, | |
5487 | { | |
5488 | .name = "low", | |
5489 | .flags = CFTYPE_NOT_ON_ROOT, | |
5490 | .seq_show = memory_low_show, | |
5491 | .write = memory_low_write, | |
5492 | }, | |
5493 | { | |
5494 | .name = "high", | |
5495 | .flags = CFTYPE_NOT_ON_ROOT, | |
5496 | .seq_show = memory_high_show, | |
5497 | .write = memory_high_write, | |
5498 | }, | |
5499 | { | |
5500 | .name = "max", | |
5501 | .flags = CFTYPE_NOT_ON_ROOT, | |
5502 | .seq_show = memory_max_show, | |
5503 | .write = memory_max_write, | |
5504 | }, | |
5505 | { | |
5506 | .name = "events", | |
5507 | .flags = CFTYPE_NOT_ON_ROOT, | |
5508 | .seq_show = memory_events_show, | |
5509 | }, | |
5510 | { } /* terminate */ | |
5511 | }; | |
5512 | ||
073219e9 | 5513 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5514 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5515 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5516 | .css_offline = mem_cgroup_css_offline, |
5517 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5518 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5519 | .can_attach = mem_cgroup_can_attach, |
5520 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5521 | .attach = mem_cgroup_move_task, |
f00baae7 | 5522 | .bind = mem_cgroup_bind, |
241994ed JW |
5523 | .dfl_cftypes = memory_files, |
5524 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5525 | .early_init = 0, |
8cdea7c0 | 5526 | }; |
c077719b | 5527 | |
241994ed JW |
5528 | /** |
5529 | * mem_cgroup_events - count memory events against a cgroup | |
5530 | * @memcg: the memory cgroup | |
5531 | * @idx: the event index | |
5532 | * @nr: the number of events to account for | |
5533 | */ | |
5534 | void mem_cgroup_events(struct mem_cgroup *memcg, | |
5535 | enum mem_cgroup_events_index idx, | |
5536 | unsigned int nr) | |
5537 | { | |
5538 | this_cpu_add(memcg->stat->events[idx], nr); | |
5539 | } | |
5540 | ||
5541 | /** | |
5542 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5543 | * @root: the highest ancestor to consider | |
5544 | * @memcg: the memory cgroup to check | |
5545 | * | |
5546 | * Returns %true if memory consumption of @memcg, and that of all | |
5547 | * configurable ancestors up to @root, is below the normal range. | |
5548 | */ | |
5549 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5550 | { | |
5551 | if (mem_cgroup_disabled()) | |
5552 | return false; | |
5553 | ||
5554 | /* | |
5555 | * The toplevel group doesn't have a configurable range, so | |
5556 | * it's never low when looked at directly, and it is not | |
5557 | * considered an ancestor when assessing the hierarchy. | |
5558 | */ | |
5559 | ||
5560 | if (memcg == root_mem_cgroup) | |
5561 | return false; | |
5562 | ||
4e54dede | 5563 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5564 | return false; |
5565 | ||
5566 | while (memcg != root) { | |
5567 | memcg = parent_mem_cgroup(memcg); | |
5568 | ||
5569 | if (memcg == root_mem_cgroup) | |
5570 | break; | |
5571 | ||
4e54dede | 5572 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5573 | return false; |
5574 | } | |
5575 | return true; | |
5576 | } | |
5577 | ||
00501b53 JW |
5578 | /** |
5579 | * mem_cgroup_try_charge - try charging a page | |
5580 | * @page: page to charge | |
5581 | * @mm: mm context of the victim | |
5582 | * @gfp_mask: reclaim mode | |
5583 | * @memcgp: charged memcg return | |
5584 | * | |
5585 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5586 | * pages according to @gfp_mask if necessary. | |
5587 | * | |
5588 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5589 | * Otherwise, an error code is returned. | |
5590 | * | |
5591 | * After page->mapping has been set up, the caller must finalize the | |
5592 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5593 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5594 | */ | |
5595 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5596 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5597 | { | |
5598 | struct mem_cgroup *memcg = NULL; | |
5599 | unsigned int nr_pages = 1; | |
5600 | int ret = 0; | |
5601 | ||
5602 | if (mem_cgroup_disabled()) | |
5603 | goto out; | |
5604 | ||
5605 | if (PageSwapCache(page)) { | |
00501b53 JW |
5606 | /* |
5607 | * Every swap fault against a single page tries to charge the | |
5608 | * page, bail as early as possible. shmem_unuse() encounters | |
5609 | * already charged pages, too. The USED bit is protected by | |
5610 | * the page lock, which serializes swap cache removal, which | |
5611 | * in turn serializes uncharging. | |
5612 | */ | |
1306a85a | 5613 | if (page->mem_cgroup) |
00501b53 JW |
5614 | goto out; |
5615 | } | |
5616 | ||
5617 | if (PageTransHuge(page)) { | |
5618 | nr_pages <<= compound_order(page); | |
5619 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5620 | } | |
5621 | ||
5622 | if (do_swap_account && PageSwapCache(page)) | |
5623 | memcg = try_get_mem_cgroup_from_page(page); | |
5624 | if (!memcg) | |
5625 | memcg = get_mem_cgroup_from_mm(mm); | |
5626 | ||
5627 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5628 | ||
5629 | css_put(&memcg->css); | |
5630 | ||
5631 | if (ret == -EINTR) { | |
5632 | memcg = root_mem_cgroup; | |
5633 | ret = 0; | |
5634 | } | |
5635 | out: | |
5636 | *memcgp = memcg; | |
5637 | return ret; | |
5638 | } | |
5639 | ||
5640 | /** | |
5641 | * mem_cgroup_commit_charge - commit a page charge | |
5642 | * @page: page to charge | |
5643 | * @memcg: memcg to charge the page to | |
5644 | * @lrucare: page might be on LRU already | |
5645 | * | |
5646 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5647 | * after page->mapping has been set up. This must happen atomically | |
5648 | * as part of the page instantiation, i.e. under the page table lock | |
5649 | * for anonymous pages, under the page lock for page and swap cache. | |
5650 | * | |
5651 | * In addition, the page must not be on the LRU during the commit, to | |
5652 | * prevent racing with task migration. If it might be, use @lrucare. | |
5653 | * | |
5654 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5655 | */ | |
5656 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5657 | bool lrucare) | |
5658 | { | |
5659 | unsigned int nr_pages = 1; | |
5660 | ||
5661 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5662 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5663 | ||
5664 | if (mem_cgroup_disabled()) | |
5665 | return; | |
5666 | /* | |
5667 | * Swap faults will attempt to charge the same page multiple | |
5668 | * times. But reuse_swap_page() might have removed the page | |
5669 | * from swapcache already, so we can't check PageSwapCache(). | |
5670 | */ | |
5671 | if (!memcg) | |
5672 | return; | |
5673 | ||
6abb5a86 JW |
5674 | commit_charge(page, memcg, lrucare); |
5675 | ||
00501b53 JW |
5676 | if (PageTransHuge(page)) { |
5677 | nr_pages <<= compound_order(page); | |
5678 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5679 | } | |
5680 | ||
6abb5a86 JW |
5681 | local_irq_disable(); |
5682 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5683 | memcg_check_events(memcg, page); | |
5684 | local_irq_enable(); | |
00501b53 JW |
5685 | |
5686 | if (do_swap_account && PageSwapCache(page)) { | |
5687 | swp_entry_t entry = { .val = page_private(page) }; | |
5688 | /* | |
5689 | * The swap entry might not get freed for a long time, | |
5690 | * let's not wait for it. The page already received a | |
5691 | * memory+swap charge, drop the swap entry duplicate. | |
5692 | */ | |
5693 | mem_cgroup_uncharge_swap(entry); | |
5694 | } | |
5695 | } | |
5696 | ||
5697 | /** | |
5698 | * mem_cgroup_cancel_charge - cancel a page charge | |
5699 | * @page: page to charge | |
5700 | * @memcg: memcg to charge the page to | |
5701 | * | |
5702 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5703 | */ | |
5704 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5705 | { | |
5706 | unsigned int nr_pages = 1; | |
5707 | ||
5708 | if (mem_cgroup_disabled()) | |
5709 | return; | |
5710 | /* | |
5711 | * Swap faults will attempt to charge the same page multiple | |
5712 | * times. But reuse_swap_page() might have removed the page | |
5713 | * from swapcache already, so we can't check PageSwapCache(). | |
5714 | */ | |
5715 | if (!memcg) | |
5716 | return; | |
5717 | ||
5718 | if (PageTransHuge(page)) { | |
5719 | nr_pages <<= compound_order(page); | |
5720 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5721 | } | |
5722 | ||
5723 | cancel_charge(memcg, nr_pages); | |
5724 | } | |
5725 | ||
747db954 | 5726 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5727 | unsigned long nr_anon, unsigned long nr_file, |
5728 | unsigned long nr_huge, struct page *dummy_page) | |
5729 | { | |
18eca2e6 | 5730 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5731 | unsigned long flags; |
5732 | ||
ce00a967 | 5733 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5734 | page_counter_uncharge(&memcg->memory, nr_pages); |
5735 | if (do_swap_account) | |
5736 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5737 | memcg_oom_recover(memcg); |
5738 | } | |
747db954 JW |
5739 | |
5740 | local_irq_save(flags); | |
5741 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5742 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5743 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5744 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5745 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5746 | memcg_check_events(memcg, dummy_page); |
5747 | local_irq_restore(flags); | |
e8ea14cc JW |
5748 | |
5749 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5750 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5751 | } |
5752 | ||
5753 | static void uncharge_list(struct list_head *page_list) | |
5754 | { | |
5755 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5756 | unsigned long nr_anon = 0; |
5757 | unsigned long nr_file = 0; | |
5758 | unsigned long nr_huge = 0; | |
5759 | unsigned long pgpgout = 0; | |
747db954 JW |
5760 | struct list_head *next; |
5761 | struct page *page; | |
5762 | ||
5763 | next = page_list->next; | |
5764 | do { | |
5765 | unsigned int nr_pages = 1; | |
747db954 JW |
5766 | |
5767 | page = list_entry(next, struct page, lru); | |
5768 | next = page->lru.next; | |
5769 | ||
5770 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5771 | VM_BUG_ON_PAGE(page_count(page), page); | |
5772 | ||
1306a85a | 5773 | if (!page->mem_cgroup) |
747db954 JW |
5774 | continue; |
5775 | ||
5776 | /* | |
5777 | * Nobody should be changing or seriously looking at | |
1306a85a | 5778 | * page->mem_cgroup at this point, we have fully |
29833315 | 5779 | * exclusive access to the page. |
747db954 JW |
5780 | */ |
5781 | ||
1306a85a | 5782 | if (memcg != page->mem_cgroup) { |
747db954 | 5783 | if (memcg) { |
18eca2e6 JW |
5784 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5785 | nr_huge, page); | |
5786 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5787 | } |
1306a85a | 5788 | memcg = page->mem_cgroup; |
747db954 JW |
5789 | } |
5790 | ||
5791 | if (PageTransHuge(page)) { | |
5792 | nr_pages <<= compound_order(page); | |
5793 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5794 | nr_huge += nr_pages; | |
5795 | } | |
5796 | ||
5797 | if (PageAnon(page)) | |
5798 | nr_anon += nr_pages; | |
5799 | else | |
5800 | nr_file += nr_pages; | |
5801 | ||
1306a85a | 5802 | page->mem_cgroup = NULL; |
747db954 JW |
5803 | |
5804 | pgpgout++; | |
5805 | } while (next != page_list); | |
5806 | ||
5807 | if (memcg) | |
18eca2e6 JW |
5808 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5809 | nr_huge, page); | |
747db954 JW |
5810 | } |
5811 | ||
0a31bc97 JW |
5812 | /** |
5813 | * mem_cgroup_uncharge - uncharge a page | |
5814 | * @page: page to uncharge | |
5815 | * | |
5816 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5817 | * mem_cgroup_commit_charge(). | |
5818 | */ | |
5819 | void mem_cgroup_uncharge(struct page *page) | |
5820 | { | |
0a31bc97 JW |
5821 | if (mem_cgroup_disabled()) |
5822 | return; | |
5823 | ||
747db954 | 5824 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5825 | if (!page->mem_cgroup) |
0a31bc97 JW |
5826 | return; |
5827 | ||
747db954 JW |
5828 | INIT_LIST_HEAD(&page->lru); |
5829 | uncharge_list(&page->lru); | |
5830 | } | |
0a31bc97 | 5831 | |
747db954 JW |
5832 | /** |
5833 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5834 | * @page_list: list of pages to uncharge | |
5835 | * | |
5836 | * Uncharge a list of pages previously charged with | |
5837 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5838 | */ | |
5839 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5840 | { | |
5841 | if (mem_cgroup_disabled()) | |
5842 | return; | |
0a31bc97 | 5843 | |
747db954 JW |
5844 | if (!list_empty(page_list)) |
5845 | uncharge_list(page_list); | |
0a31bc97 JW |
5846 | } |
5847 | ||
5848 | /** | |
5849 | * mem_cgroup_migrate - migrate a charge to another page | |
5850 | * @oldpage: currently charged page | |
5851 | * @newpage: page to transfer the charge to | |
f5e03a49 | 5852 | * @lrucare: either or both pages might be on the LRU already |
0a31bc97 JW |
5853 | * |
5854 | * Migrate the charge from @oldpage to @newpage. | |
5855 | * | |
5856 | * Both pages must be locked, @newpage->mapping must be set up. | |
5857 | */ | |
5858 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
5859 | bool lrucare) | |
5860 | { | |
29833315 | 5861 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5862 | int isolated; |
5863 | ||
5864 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5865 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
5866 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
5867 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
5868 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
5869 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5870 | newpage); | |
0a31bc97 JW |
5871 | |
5872 | if (mem_cgroup_disabled()) | |
5873 | return; | |
5874 | ||
5875 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5876 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5877 | return; |
5878 | ||
7d5e3245 JW |
5879 | /* |
5880 | * Swapcache readahead pages can get migrated before being | |
5881 | * charged, and migration from compaction can happen to an | |
5882 | * uncharged page when the PFN walker finds a page that | |
5883 | * reclaim just put back on the LRU but has not released yet. | |
5884 | */ | |
1306a85a | 5885 | memcg = oldpage->mem_cgroup; |
29833315 | 5886 | if (!memcg) |
0a31bc97 JW |
5887 | return; |
5888 | ||
0a31bc97 JW |
5889 | if (lrucare) |
5890 | lock_page_lru(oldpage, &isolated); | |
5891 | ||
1306a85a | 5892 | oldpage->mem_cgroup = NULL; |
0a31bc97 JW |
5893 | |
5894 | if (lrucare) | |
5895 | unlock_page_lru(oldpage, isolated); | |
5896 | ||
29833315 | 5897 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
5898 | } |
5899 | ||
2d11085e | 5900 | /* |
1081312f MH |
5901 | * subsys_initcall() for memory controller. |
5902 | * | |
5903 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5904 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5905 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5906 | * be initialized from here. | |
2d11085e MH |
5907 | */ |
5908 | static int __init mem_cgroup_init(void) | |
5909 | { | |
95a045f6 JW |
5910 | int cpu, node; |
5911 | ||
2d11085e | 5912 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5913 | |
5914 | for_each_possible_cpu(cpu) | |
5915 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5916 | drain_local_stock); | |
5917 | ||
5918 | for_each_node(node) { | |
5919 | struct mem_cgroup_tree_per_node *rtpn; | |
5920 | int zone; | |
5921 | ||
5922 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5923 | node_online(node) ? node : NUMA_NO_NODE); | |
5924 | ||
5925 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5926 | struct mem_cgroup_tree_per_zone *rtpz; | |
5927 | ||
5928 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5929 | rtpz->rb_root = RB_ROOT; | |
5930 | spin_lock_init(&rtpz->lock); | |
5931 | } | |
5932 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5933 | } | |
5934 | ||
2d11085e MH |
5935 | return 0; |
5936 | } | |
5937 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5938 | |
5939 | #ifdef CONFIG_MEMCG_SWAP | |
5940 | /** | |
5941 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5942 | * @page: page whose memsw charge to transfer | |
5943 | * @entry: swap entry to move the charge to | |
5944 | * | |
5945 | * Transfer the memsw charge of @page to @entry. | |
5946 | */ | |
5947 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5948 | { | |
5949 | struct mem_cgroup *memcg; | |
5950 | unsigned short oldid; | |
5951 | ||
5952 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5953 | VM_BUG_ON_PAGE(page_count(page), page); | |
5954 | ||
5955 | if (!do_swap_account) | |
5956 | return; | |
5957 | ||
5958 | memcg = page->mem_cgroup; | |
5959 | ||
5960 | /* Readahead page, never charged */ | |
5961 | if (!memcg) | |
5962 | return; | |
5963 | ||
5964 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5965 | VM_BUG_ON_PAGE(oldid, page); | |
5966 | mem_cgroup_swap_statistics(memcg, true); | |
5967 | ||
5968 | page->mem_cgroup = NULL; | |
5969 | ||
5970 | if (!mem_cgroup_is_root(memcg)) | |
5971 | page_counter_uncharge(&memcg->memory, 1); | |
5972 | ||
ce9ce665 SAS |
5973 | /* |
5974 | * Interrupts should be disabled here because the caller holds the | |
5975 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5976 | * important here to have the interrupts disabled because it is the | |
5977 | * only synchronisation we have for udpating the per-CPU variables. | |
5978 | */ | |
5979 | VM_BUG_ON(!irqs_disabled()); | |
21afa38e JW |
5980 | mem_cgroup_charge_statistics(memcg, page, -1); |
5981 | memcg_check_events(memcg, page); | |
5982 | } | |
5983 | ||
5984 | /** | |
5985 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5986 | * @entry: swap entry to uncharge | |
5987 | * | |
5988 | * Drop the memsw charge associated with @entry. | |
5989 | */ | |
5990 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5991 | { | |
5992 | struct mem_cgroup *memcg; | |
5993 | unsigned short id; | |
5994 | ||
5995 | if (!do_swap_account) | |
5996 | return; | |
5997 | ||
5998 | id = swap_cgroup_record(entry, 0); | |
5999 | rcu_read_lock(); | |
adbe427b | 6000 | memcg = mem_cgroup_from_id(id); |
21afa38e JW |
6001 | if (memcg) { |
6002 | if (!mem_cgroup_is_root(memcg)) | |
6003 | page_counter_uncharge(&memcg->memsw, 1); | |
6004 | mem_cgroup_swap_statistics(memcg, false); | |
6005 | css_put(&memcg->css); | |
6006 | } | |
6007 | rcu_read_unlock(); | |
6008 | } | |
6009 | ||
6010 | /* for remember boot option*/ | |
6011 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6012 | static int really_do_swap_account __initdata = 1; | |
6013 | #else | |
6014 | static int really_do_swap_account __initdata; | |
6015 | #endif | |
6016 | ||
6017 | static int __init enable_swap_account(char *s) | |
6018 | { | |
6019 | if (!strcmp(s, "1")) | |
6020 | really_do_swap_account = 1; | |
6021 | else if (!strcmp(s, "0")) | |
6022 | really_do_swap_account = 0; | |
6023 | return 1; | |
6024 | } | |
6025 | __setup("swapaccount=", enable_swap_account); | |
6026 | ||
6027 | static struct cftype memsw_cgroup_files[] = { | |
6028 | { | |
6029 | .name = "memsw.usage_in_bytes", | |
6030 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6031 | .read_u64 = mem_cgroup_read_u64, | |
6032 | }, | |
6033 | { | |
6034 | .name = "memsw.max_usage_in_bytes", | |
6035 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6036 | .write = mem_cgroup_reset, | |
6037 | .read_u64 = mem_cgroup_read_u64, | |
6038 | }, | |
6039 | { | |
6040 | .name = "memsw.limit_in_bytes", | |
6041 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6042 | .write = mem_cgroup_write, | |
6043 | .read_u64 = mem_cgroup_read_u64, | |
6044 | }, | |
6045 | { | |
6046 | .name = "memsw.failcnt", | |
6047 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6048 | .write = mem_cgroup_reset, | |
6049 | .read_u64 = mem_cgroup_read_u64, | |
6050 | }, | |
6051 | { }, /* terminate */ | |
6052 | }; | |
6053 | ||
6054 | static int __init mem_cgroup_swap_init(void) | |
6055 | { | |
6056 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6057 | do_swap_account = 1; | |
6058 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
6059 | memsw_cgroup_files)); | |
6060 | } | |
6061 | return 0; | |
6062 | } | |
6063 | subsys_initcall(mem_cgroup_swap_init); | |
6064 | ||
6065 | #endif /* CONFIG_MEMCG_SWAP */ |