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 | * | |
8cdea7c0 BS |
17 | * This program is free software; you can redistribute it and/or modify |
18 | * it under the terms of the GNU General Public License as published by | |
19 | * the Free Software Foundation; either version 2 of the License, or | |
20 | * (at your option) any later version. | |
21 | * | |
22 | * This program is distributed in the hope that it will be useful, | |
23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
25 | * GNU General Public License for more details. | |
26 | */ | |
27 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
29 | #include <linux/memcontrol.h> |
30 | #include <linux/cgroup.h> | |
78fb7466 | 31 | #include <linux/mm.h> |
4ffef5fe | 32 | #include <linux/hugetlb.h> |
d13d1443 | 33 | #include <linux/pagemap.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
bb4cc1a8 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 | 47 | #include <linux/eventfd.h> |
79bd9814 | 48 | #include <linux/poll.h> |
2e72b634 | 49 | #include <linux/sort.h> |
66e1707b | 50 | #include <linux/fs.h> |
d2ceb9b7 | 51 | #include <linux/seq_file.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
5d1ea48b | 54 | #include <linux/swap_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
08e552c6 | 59 | #include "internal.h" |
d1a4c0b3 | 60 | #include <net/sock.h> |
4bd2c1ee | 61 | #include <net/ip.h> |
d1a4c0b3 | 62 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 63 | #include "slab.h" |
8cdea7c0 | 64 | |
8697d331 BS |
65 | #include <asm/uaccess.h> |
66 | ||
cc8e970c KM |
67 | #include <trace/events/vmscan.h> |
68 | ||
073219e9 TH |
69 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
70 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 71 | |
a181b0e8 | 72 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 73 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 74 | |
21afa38e | 75 | /* Whether the swap controller is active */ |
c255a458 | 76 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 77 | int do_swap_account __read_mostly; |
c077719b | 78 | #else |
a0db00fc | 79 | #define do_swap_account 0 |
c077719b KH |
80 | #endif |
81 | ||
af7c4b0e JW |
82 | static const char * const mem_cgroup_stat_names[] = { |
83 | "cache", | |
84 | "rss", | |
b070e65c | 85 | "rss_huge", |
af7c4b0e | 86 | "mapped_file", |
3ea67d06 | 87 | "writeback", |
af7c4b0e JW |
88 | "swap", |
89 | }; | |
90 | ||
af7c4b0e JW |
91 | static const char * const mem_cgroup_events_names[] = { |
92 | "pgpgin", | |
93 | "pgpgout", | |
94 | "pgfault", | |
95 | "pgmajfault", | |
96 | }; | |
97 | ||
58cf188e SZ |
98 | static const char * const mem_cgroup_lru_names[] = { |
99 | "inactive_anon", | |
100 | "active_anon", | |
101 | "inactive_file", | |
102 | "active_file", | |
103 | "unevictable", | |
104 | }; | |
105 | ||
7a159cc9 JW |
106 | /* |
107 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
108 | * it will be incremated by the number of pages. This counter is used for | |
109 | * for trigger some periodic events. This is straightforward and better | |
110 | * than using jiffies etc. to handle periodic memcg event. | |
111 | */ | |
112 | enum mem_cgroup_events_target { | |
113 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 114 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 115 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
116 | MEM_CGROUP_NTARGETS, |
117 | }; | |
a0db00fc KS |
118 | #define THRESHOLDS_EVENTS_TARGET 128 |
119 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
120 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 121 | |
d52aa412 | 122 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 123 | long count[MEM_CGROUP_STAT_NSTATS]; |
241994ed | 124 | unsigned long events[MEMCG_NR_EVENTS]; |
13114716 | 125 | unsigned long nr_page_events; |
7a159cc9 | 126 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
127 | }; |
128 | ||
5ac8fb31 JW |
129 | struct reclaim_iter { |
130 | struct mem_cgroup *position; | |
527a5ec9 JW |
131 | /* scan generation, increased every round-trip */ |
132 | unsigned int generation; | |
133 | }; | |
134 | ||
6d12e2d8 KH |
135 | /* |
136 | * per-zone information in memory controller. | |
137 | */ | |
6d12e2d8 | 138 | struct mem_cgroup_per_zone { |
6290df54 | 139 | struct lruvec lruvec; |
1eb49272 | 140 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 141 | |
5ac8fb31 | 142 | struct reclaim_iter iter[DEF_PRIORITY + 1]; |
527a5ec9 | 143 | |
bb4cc1a8 | 144 | struct rb_node tree_node; /* RB tree node */ |
3e32cb2e | 145 | unsigned long usage_in_excess;/* Set to the value by which */ |
bb4cc1a8 AM |
146 | /* the soft limit is exceeded*/ |
147 | bool on_tree; | |
d79154bb | 148 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 149 | /* use container_of */ |
6d12e2d8 | 150 | }; |
6d12e2d8 KH |
151 | |
152 | struct mem_cgroup_per_node { | |
153 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
154 | }; | |
155 | ||
bb4cc1a8 AM |
156 | /* |
157 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
158 | * their hierarchy representation | |
159 | */ | |
160 | ||
161 | struct mem_cgroup_tree_per_zone { | |
162 | struct rb_root rb_root; | |
163 | spinlock_t lock; | |
164 | }; | |
165 | ||
166 | struct mem_cgroup_tree_per_node { | |
167 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
168 | }; | |
169 | ||
170 | struct mem_cgroup_tree { | |
171 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
172 | }; | |
173 | ||
174 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
175 | ||
2e72b634 KS |
176 | struct mem_cgroup_threshold { |
177 | struct eventfd_ctx *eventfd; | |
3e32cb2e | 178 | unsigned long threshold; |
2e72b634 KS |
179 | }; |
180 | ||
9490ff27 | 181 | /* For threshold */ |
2e72b634 | 182 | struct mem_cgroup_threshold_ary { |
748dad36 | 183 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 184 | int current_threshold; |
2e72b634 KS |
185 | /* Size of entries[] */ |
186 | unsigned int size; | |
187 | /* Array of thresholds */ | |
188 | struct mem_cgroup_threshold entries[0]; | |
189 | }; | |
2c488db2 KS |
190 | |
191 | struct mem_cgroup_thresholds { | |
192 | /* Primary thresholds array */ | |
193 | struct mem_cgroup_threshold_ary *primary; | |
194 | /* | |
195 | * Spare threshold array. | |
196 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
197 | * It must be able to store at least primary->size - 1 entries. | |
198 | */ | |
199 | struct mem_cgroup_threshold_ary *spare; | |
200 | }; | |
201 | ||
9490ff27 KH |
202 | /* for OOM */ |
203 | struct mem_cgroup_eventfd_list { | |
204 | struct list_head list; | |
205 | struct eventfd_ctx *eventfd; | |
206 | }; | |
2e72b634 | 207 | |
79bd9814 TH |
208 | /* |
209 | * cgroup_event represents events which userspace want to receive. | |
210 | */ | |
3bc942f3 | 211 | struct mem_cgroup_event { |
79bd9814 | 212 | /* |
59b6f873 | 213 | * memcg which the event belongs to. |
79bd9814 | 214 | */ |
59b6f873 | 215 | struct mem_cgroup *memcg; |
79bd9814 TH |
216 | /* |
217 | * eventfd to signal userspace about the event. | |
218 | */ | |
219 | struct eventfd_ctx *eventfd; | |
220 | /* | |
221 | * Each of these stored in a list by the cgroup. | |
222 | */ | |
223 | struct list_head list; | |
fba94807 TH |
224 | /* |
225 | * register_event() callback will be used to add new userspace | |
226 | * waiter for changes related to this event. Use eventfd_signal() | |
227 | * on eventfd to send notification to userspace. | |
228 | */ | |
59b6f873 | 229 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 230 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
231 | /* |
232 | * unregister_event() callback will be called when userspace closes | |
233 | * the eventfd or on cgroup removing. This callback must be set, | |
234 | * if you want provide notification functionality. | |
235 | */ | |
59b6f873 | 236 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 237 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
238 | /* |
239 | * All fields below needed to unregister event when | |
240 | * userspace closes eventfd. | |
241 | */ | |
242 | poll_table pt; | |
243 | wait_queue_head_t *wqh; | |
244 | wait_queue_t wait; | |
245 | struct work_struct remove; | |
246 | }; | |
247 | ||
c0ff4b85 R |
248 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
249 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 250 | |
8cdea7c0 BS |
251 | /* |
252 | * The memory controller data structure. The memory controller controls both | |
253 | * page cache and RSS per cgroup. We would eventually like to provide | |
254 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
255 | * to help the administrator determine what knobs to tune. | |
256 | * | |
257 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
258 | * we hit the water mark. May be even add a low water mark, such that |
259 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
260 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
261 | */ |
262 | struct mem_cgroup { | |
263 | struct cgroup_subsys_state css; | |
3e32cb2e JW |
264 | |
265 | /* Accounted resources */ | |
266 | struct page_counter memory; | |
267 | struct page_counter memsw; | |
268 | struct page_counter kmem; | |
269 | ||
241994ed JW |
270 | /* Normal memory consumption range */ |
271 | unsigned long low; | |
272 | unsigned long high; | |
273 | ||
3e32cb2e | 274 | unsigned long soft_limit; |
59927fb9 | 275 | |
70ddf637 AV |
276 | /* vmpressure notifications */ |
277 | struct vmpressure vmpressure; | |
278 | ||
2f7dd7a4 JW |
279 | /* css_online() has been completed */ |
280 | int initialized; | |
281 | ||
18f59ea7 BS |
282 | /* |
283 | * Should the accounting and control be hierarchical, per subtree? | |
284 | */ | |
285 | bool use_hierarchy; | |
79dfdacc MH |
286 | |
287 | bool oom_lock; | |
288 | atomic_t under_oom; | |
3812c8c8 | 289 | atomic_t oom_wakeups; |
79dfdacc | 290 | |
1f4c025b | 291 | int swappiness; |
3c11ecf4 KH |
292 | /* OOM-Killer disable */ |
293 | int oom_kill_disable; | |
a7885eb8 | 294 | |
2e72b634 KS |
295 | /* protect arrays of thresholds */ |
296 | struct mutex thresholds_lock; | |
297 | ||
298 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 299 | struct mem_cgroup_thresholds thresholds; |
907860ed | 300 | |
2e72b634 | 301 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 302 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 303 | |
9490ff27 KH |
304 | /* For oom notifier event fd */ |
305 | struct list_head oom_notify; | |
185efc0f | 306 | |
7dc74be0 DN |
307 | /* |
308 | * Should we move charges of a task when a task is moved into this | |
309 | * mem_cgroup ? And what type of charges should we move ? | |
310 | */ | |
f894ffa8 | 311 | unsigned long move_charge_at_immigrate; |
619d094b KH |
312 | /* |
313 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
314 | */ | |
6de22619 | 315 | atomic_t moving_account; |
312734c0 | 316 | /* taken only while moving_account > 0 */ |
6de22619 JW |
317 | spinlock_t move_lock; |
318 | struct task_struct *move_lock_task; | |
319 | unsigned long move_lock_flags; | |
d52aa412 | 320 | /* |
c62b1a3b | 321 | * percpu counter. |
d52aa412 | 322 | */ |
3a7951b4 | 323 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
324 | /* |
325 | * used when a cpu is offlined or other synchronizations | |
326 | * See mem_cgroup_read_stat(). | |
327 | */ | |
328 | struct mem_cgroup_stat_cpu nocpu_base; | |
329 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 330 | |
4bd2c1ee | 331 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 332 | struct cg_proto tcp_mem; |
d1a4c0b3 | 333 | #endif |
2633d7a0 | 334 | #if defined(CONFIG_MEMCG_KMEM) |
f7ce3190 | 335 | /* Index in the kmem_cache->memcg_params.memcg_caches array */ |
2633d7a0 | 336 | int kmemcg_id; |
2a4db7eb | 337 | bool kmem_acct_active; |
2633d7a0 | 338 | #endif |
45cf7ebd GC |
339 | |
340 | int last_scanned_node; | |
341 | #if MAX_NUMNODES > 1 | |
342 | nodemask_t scan_nodes; | |
343 | atomic_t numainfo_events; | |
344 | atomic_t numainfo_updating; | |
345 | #endif | |
70ddf637 | 346 | |
fba94807 TH |
347 | /* List of events which userspace want to receive */ |
348 | struct list_head event_list; | |
349 | spinlock_t event_list_lock; | |
350 | ||
54f72fe0 JW |
351 | struct mem_cgroup_per_node *nodeinfo[0]; |
352 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
353 | }; |
354 | ||
510fc4e1 | 355 | #ifdef CONFIG_MEMCG_KMEM |
cb731d6c | 356 | bool memcg_kmem_is_active(struct mem_cgroup *memcg) |
7de37682 | 357 | { |
2a4db7eb | 358 | return memcg->kmem_acct_active; |
7de37682 | 359 | } |
510fc4e1 GC |
360 | #endif |
361 | ||
7dc74be0 DN |
362 | /* Stuffs for move charges at task migration. */ |
363 | /* | |
1dfab5ab | 364 | * Types of charges to be moved. |
7dc74be0 | 365 | */ |
1dfab5ab JW |
366 | #define MOVE_ANON 0x1U |
367 | #define MOVE_FILE 0x2U | |
368 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 369 | |
4ffef5fe DN |
370 | /* "mc" and its members are protected by cgroup_mutex */ |
371 | static struct move_charge_struct { | |
b1dd693e | 372 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
373 | struct mem_cgroup *from; |
374 | struct mem_cgroup *to; | |
1dfab5ab | 375 | unsigned long flags; |
4ffef5fe | 376 | unsigned long precharge; |
854ffa8d | 377 | unsigned long moved_charge; |
483c30b5 | 378 | unsigned long moved_swap; |
8033b97c DN |
379 | struct task_struct *moving_task; /* a task moving charges */ |
380 | wait_queue_head_t waitq; /* a waitq for other context */ | |
381 | } mc = { | |
2bd9bb20 | 382 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
383 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
384 | }; | |
4ffef5fe | 385 | |
4e416953 BS |
386 | /* |
387 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
388 | * limit reclaim to prevent infinite loops, if they ever occur. | |
389 | */ | |
a0db00fc | 390 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 391 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 392 | |
217bc319 KH |
393 | enum charge_type { |
394 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 395 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 396 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 397 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
398 | NR_CHARGE_TYPE, |
399 | }; | |
400 | ||
8c7c6e34 | 401 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
402 | enum res_type { |
403 | _MEM, | |
404 | _MEMSWAP, | |
405 | _OOM_TYPE, | |
510fc4e1 | 406 | _KMEM, |
86ae53e1 GC |
407 | }; |
408 | ||
a0db00fc KS |
409 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
410 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 411 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
412 | /* Used for OOM nofiier */ |
413 | #define OOM_CONTROL (0) | |
8c7c6e34 | 414 | |
0999821b GC |
415 | /* |
416 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
417 | * As a consequence, any change that needs to protect against new child cgroups | |
418 | * appearing has to hold it as well. | |
419 | */ | |
420 | static DEFINE_MUTEX(memcg_create_mutex); | |
421 | ||
b2145145 WL |
422 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
423 | { | |
a7c6d554 | 424 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
425 | } |
426 | ||
70ddf637 AV |
427 | /* Some nice accessors for the vmpressure. */ |
428 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
429 | { | |
430 | if (!memcg) | |
431 | memcg = root_mem_cgroup; | |
432 | return &memcg->vmpressure; | |
433 | } | |
434 | ||
435 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
436 | { | |
437 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
438 | } | |
439 | ||
7ffc0edc MH |
440 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
441 | { | |
442 | return (memcg == root_mem_cgroup); | |
443 | } | |
444 | ||
4219b2da LZ |
445 | /* |
446 | * We restrict the id in the range of [1, 65535], so it can fit into | |
447 | * an unsigned short. | |
448 | */ | |
449 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
450 | ||
34c00c31 LZ |
451 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
452 | { | |
15a4c835 | 453 | return memcg->css.id; |
34c00c31 LZ |
454 | } |
455 | ||
456 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
457 | { | |
458 | struct cgroup_subsys_state *css; | |
459 | ||
7d699ddb | 460 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
461 | return mem_cgroup_from_css(css); |
462 | } | |
463 | ||
e1aab161 | 464 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 465 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 466 | |
e1aab161 GC |
467 | void sock_update_memcg(struct sock *sk) |
468 | { | |
376be5ff | 469 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 470 | struct mem_cgroup *memcg; |
3f134619 | 471 | struct cg_proto *cg_proto; |
e1aab161 GC |
472 | |
473 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
474 | ||
f3f511e1 GC |
475 | /* Socket cloning can throw us here with sk_cgrp already |
476 | * filled. It won't however, necessarily happen from | |
477 | * process context. So the test for root memcg given | |
478 | * the current task's memcg won't help us in this case. | |
479 | * | |
480 | * Respecting the original socket's memcg is a better | |
481 | * decision in this case. | |
482 | */ | |
483 | if (sk->sk_cgrp) { | |
484 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 485 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
486 | return; |
487 | } | |
488 | ||
e1aab161 GC |
489 | rcu_read_lock(); |
490 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 491 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 492 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
493 | memcg_proto_active(cg_proto) && |
494 | css_tryget_online(&memcg->css)) { | |
3f134619 | 495 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
496 | } |
497 | rcu_read_unlock(); | |
498 | } | |
499 | } | |
500 | EXPORT_SYMBOL(sock_update_memcg); | |
501 | ||
502 | void sock_release_memcg(struct sock *sk) | |
503 | { | |
376be5ff | 504 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
505 | struct mem_cgroup *memcg; |
506 | WARN_ON(!sk->sk_cgrp->memcg); | |
507 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 508 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
509 | } |
510 | } | |
d1a4c0b3 GC |
511 | |
512 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
513 | { | |
514 | if (!memcg || mem_cgroup_is_root(memcg)) | |
515 | return NULL; | |
516 | ||
2e685cad | 517 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
518 | } |
519 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 520 | |
3f134619 GC |
521 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
522 | { | |
2e685cad | 523 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
524 | return; |
525 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
526 | } | |
527 | #else | |
528 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
529 | { | |
530 | } | |
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 | |
f3bb3043 VD |
585 | static void memcg_free_cache_id(int id); |
586 | ||
a8964b9b GC |
587 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
588 | { | |
2a4db7eb | 589 | if (memcg->kmemcg_id >= 0) { |
a8964b9b | 590 | static_key_slow_dec(&memcg_kmem_enabled_key); |
f3bb3043 | 591 | memcg_free_cache_id(memcg->kmemcg_id); |
55007d84 | 592 | } |
bea207c8 GC |
593 | /* |
594 | * This check can't live in kmem destruction function, | |
595 | * since the charges will outlive the cgroup | |
596 | */ | |
3e32cb2e | 597 | WARN_ON(page_counter_read(&memcg->kmem)); |
a8964b9b GC |
598 | } |
599 | #else | |
600 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
601 | { | |
602 | } | |
603 | #endif /* CONFIG_MEMCG_KMEM */ | |
604 | ||
605 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
606 | { | |
607 | disarm_sock_keys(memcg); | |
608 | disarm_kmem_keys(memcg); | |
609 | } | |
610 | ||
f64c3f54 | 611 | static struct mem_cgroup_per_zone * |
e231875b | 612 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 613 | { |
e231875b JZ |
614 | int nid = zone_to_nid(zone); |
615 | int zid = zone_idx(zone); | |
616 | ||
54f72fe0 | 617 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
618 | } |
619 | ||
c0ff4b85 | 620 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 621 | { |
c0ff4b85 | 622 | return &memcg->css; |
d324236b WF |
623 | } |
624 | ||
f64c3f54 | 625 | static struct mem_cgroup_per_zone * |
e231875b | 626 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 627 | { |
97a6c37b JW |
628 | int nid = page_to_nid(page); |
629 | int zid = page_zonenum(page); | |
f64c3f54 | 630 | |
e231875b | 631 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
632 | } |
633 | ||
bb4cc1a8 AM |
634 | static struct mem_cgroup_tree_per_zone * |
635 | soft_limit_tree_node_zone(int nid, int zid) | |
636 | { | |
637 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
638 | } | |
639 | ||
640 | static struct mem_cgroup_tree_per_zone * | |
641 | soft_limit_tree_from_page(struct page *page) | |
642 | { | |
643 | int nid = page_to_nid(page); | |
644 | int zid = page_zonenum(page); | |
645 | ||
646 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
647 | } | |
648 | ||
cf2c8127 JW |
649 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
650 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 651 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
652 | { |
653 | struct rb_node **p = &mctz->rb_root.rb_node; | |
654 | struct rb_node *parent = NULL; | |
655 | struct mem_cgroup_per_zone *mz_node; | |
656 | ||
657 | if (mz->on_tree) | |
658 | return; | |
659 | ||
660 | mz->usage_in_excess = new_usage_in_excess; | |
661 | if (!mz->usage_in_excess) | |
662 | return; | |
663 | while (*p) { | |
664 | parent = *p; | |
665 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
666 | tree_node); | |
667 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
668 | p = &(*p)->rb_left; | |
669 | /* | |
670 | * We can't avoid mem cgroups that are over their soft | |
671 | * limit by the same amount | |
672 | */ | |
673 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
674 | p = &(*p)->rb_right; | |
675 | } | |
676 | rb_link_node(&mz->tree_node, parent, p); | |
677 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
678 | mz->on_tree = true; | |
679 | } | |
680 | ||
cf2c8127 JW |
681 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
682 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
683 | { |
684 | if (!mz->on_tree) | |
685 | return; | |
686 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
687 | mz->on_tree = false; | |
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 | 692 | { |
0a31bc97 JW |
693 | unsigned long flags; |
694 | ||
695 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 696 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 697 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
698 | } |
699 | ||
3e32cb2e JW |
700 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
701 | { | |
702 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
703 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); | |
704 | unsigned long excess = 0; | |
705 | ||
706 | if (nr_pages > soft_limit) | |
707 | excess = nr_pages - soft_limit; | |
708 | ||
709 | return excess; | |
710 | } | |
bb4cc1a8 AM |
711 | |
712 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
713 | { | |
3e32cb2e | 714 | unsigned long excess; |
bb4cc1a8 AM |
715 | struct mem_cgroup_per_zone *mz; |
716 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 717 | |
e231875b | 718 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
719 | /* |
720 | * Necessary to update all ancestors when hierarchy is used. | |
721 | * because their event counter is not touched. | |
722 | */ | |
723 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 724 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 725 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
726 | /* |
727 | * We have to update the tree if mz is on RB-tree or | |
728 | * mem is over its softlimit. | |
729 | */ | |
730 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
731 | unsigned long flags; |
732 | ||
733 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
734 | /* if on-tree, remove it */ |
735 | if (mz->on_tree) | |
cf2c8127 | 736 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
737 | /* |
738 | * Insert again. mz->usage_in_excess will be updated. | |
739 | * If excess is 0, no tree ops. | |
740 | */ | |
cf2c8127 | 741 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 742 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
743 | } |
744 | } | |
745 | } | |
746 | ||
747 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
748 | { | |
bb4cc1a8 | 749 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
750 | struct mem_cgroup_per_zone *mz; |
751 | int nid, zid; | |
bb4cc1a8 | 752 | |
e231875b JZ |
753 | for_each_node(nid) { |
754 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
755 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
756 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 757 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
758 | } |
759 | } | |
760 | } | |
761 | ||
762 | static struct mem_cgroup_per_zone * | |
763 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
764 | { | |
765 | struct rb_node *rightmost = NULL; | |
766 | struct mem_cgroup_per_zone *mz; | |
767 | ||
768 | retry: | |
769 | mz = NULL; | |
770 | rightmost = rb_last(&mctz->rb_root); | |
771 | if (!rightmost) | |
772 | goto done; /* Nothing to reclaim from */ | |
773 | ||
774 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
775 | /* | |
776 | * Remove the node now but someone else can add it back, | |
777 | * we will to add it back at the end of reclaim to its correct | |
778 | * position in the tree. | |
779 | */ | |
cf2c8127 | 780 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 781 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 782 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
783 | goto retry; |
784 | done: | |
785 | return mz; | |
786 | } | |
787 | ||
788 | static struct mem_cgroup_per_zone * | |
789 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
790 | { | |
791 | struct mem_cgroup_per_zone *mz; | |
792 | ||
0a31bc97 | 793 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 794 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 795 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
796 | return mz; |
797 | } | |
798 | ||
711d3d2c KH |
799 | /* |
800 | * Implementation Note: reading percpu statistics for memcg. | |
801 | * | |
802 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
803 | * synchronization to implement "quick" read. There are trade-off between | |
804 | * reading cost and precision of value. Then, we may have a chance to implement | |
805 | * a periodic synchronizion of counter in memcg's counter. | |
806 | * | |
807 | * But this _read() function is used for user interface now. The user accounts | |
808 | * memory usage by memory cgroup and he _always_ requires exact value because | |
809 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
810 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
811 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
812 | * | |
813 | * If there are kernel internal actions which can make use of some not-exact | |
814 | * value, and reading all cpu value can be performance bottleneck in some | |
815 | * common workload, threashold and synchonization as vmstat[] should be | |
816 | * implemented. | |
817 | */ | |
c0ff4b85 | 818 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 819 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 820 | { |
7a159cc9 | 821 | long val = 0; |
c62b1a3b | 822 | int cpu; |
c62b1a3b | 823 | |
711d3d2c KH |
824 | get_online_cpus(); |
825 | for_each_online_cpu(cpu) | |
c0ff4b85 | 826 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 827 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
828 | spin_lock(&memcg->pcp_counter_lock); |
829 | val += memcg->nocpu_base.count[idx]; | |
830 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
831 | #endif |
832 | put_online_cpus(); | |
c62b1a3b KH |
833 | return val; |
834 | } | |
835 | ||
c0ff4b85 | 836 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
837 | enum mem_cgroup_events_index idx) |
838 | { | |
839 | unsigned long val = 0; | |
840 | int cpu; | |
841 | ||
9c567512 | 842 | get_online_cpus(); |
e9f8974f | 843 | for_each_online_cpu(cpu) |
c0ff4b85 | 844 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 845 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
846 | spin_lock(&memcg->pcp_counter_lock); |
847 | val += memcg->nocpu_base.events[idx]; | |
848 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 849 | #endif |
9c567512 | 850 | put_online_cpus(); |
e9f8974f JW |
851 | return val; |
852 | } | |
853 | ||
c0ff4b85 | 854 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 855 | struct page *page, |
0a31bc97 | 856 | int nr_pages) |
d52aa412 | 857 | { |
b2402857 KH |
858 | /* |
859 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
860 | * counted as CACHE even if it's on ANON LRU. | |
861 | */ | |
0a31bc97 | 862 | if (PageAnon(page)) |
b2402857 | 863 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 864 | nr_pages); |
d52aa412 | 865 | else |
b2402857 | 866 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 867 | nr_pages); |
55e462b0 | 868 | |
b070e65c DR |
869 | if (PageTransHuge(page)) |
870 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
871 | nr_pages); | |
872 | ||
e401f176 KH |
873 | /* pagein of a big page is an event. So, ignore page size */ |
874 | if (nr_pages > 0) | |
c0ff4b85 | 875 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 876 | else { |
c0ff4b85 | 877 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
878 | nr_pages = -nr_pages; /* for event */ |
879 | } | |
e401f176 | 880 | |
13114716 | 881 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
882 | } |
883 | ||
e231875b | 884 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
885 | { |
886 | struct mem_cgroup_per_zone *mz; | |
887 | ||
888 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
889 | return mz->lru_size[lru]; | |
890 | } | |
891 | ||
e231875b JZ |
892 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
893 | int nid, | |
894 | unsigned int lru_mask) | |
bb2a0de9 | 895 | { |
e231875b | 896 | unsigned long nr = 0; |
889976db YH |
897 | int zid; |
898 | ||
e231875b | 899 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 900 | |
e231875b JZ |
901 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
902 | struct mem_cgroup_per_zone *mz; | |
903 | enum lru_list lru; | |
904 | ||
905 | for_each_lru(lru) { | |
906 | if (!(BIT(lru) & lru_mask)) | |
907 | continue; | |
908 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
909 | nr += mz->lru_size[lru]; | |
910 | } | |
911 | } | |
912 | return nr; | |
889976db | 913 | } |
bb2a0de9 | 914 | |
c0ff4b85 | 915 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 916 | unsigned int lru_mask) |
6d12e2d8 | 917 | { |
e231875b | 918 | unsigned long nr = 0; |
889976db | 919 | int nid; |
6d12e2d8 | 920 | |
31aaea4a | 921 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
922 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
923 | return nr; | |
d52aa412 KH |
924 | } |
925 | ||
f53d7ce3 JW |
926 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
927 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
928 | { |
929 | unsigned long val, next; | |
930 | ||
13114716 | 931 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 932 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 933 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
934 | if ((long)next - (long)val < 0) { |
935 | switch (target) { | |
936 | case MEM_CGROUP_TARGET_THRESH: | |
937 | next = val + THRESHOLDS_EVENTS_TARGET; | |
938 | break; | |
bb4cc1a8 AM |
939 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
940 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
941 | break; | |
f53d7ce3 JW |
942 | case MEM_CGROUP_TARGET_NUMAINFO: |
943 | next = val + NUMAINFO_EVENTS_TARGET; | |
944 | break; | |
945 | default: | |
946 | break; | |
947 | } | |
948 | __this_cpu_write(memcg->stat->targets[target], next); | |
949 | return true; | |
7a159cc9 | 950 | } |
f53d7ce3 | 951 | return false; |
d2265e6f KH |
952 | } |
953 | ||
954 | /* | |
955 | * Check events in order. | |
956 | * | |
957 | */ | |
c0ff4b85 | 958 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
959 | { |
960 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
961 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
962 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 963 | bool do_softlimit; |
82b3f2a7 | 964 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 965 | |
bb4cc1a8 AM |
966 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
967 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
968 | #if MAX_NUMNODES > 1 |
969 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
970 | MEM_CGROUP_TARGET_NUMAINFO); | |
971 | #endif | |
c0ff4b85 | 972 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
973 | if (unlikely(do_softlimit)) |
974 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 975 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 976 | if (unlikely(do_numainfo)) |
c0ff4b85 | 977 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 978 | #endif |
0a31bc97 | 979 | } |
d2265e6f KH |
980 | } |
981 | ||
cf475ad2 | 982 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 983 | { |
31a78f23 BS |
984 | /* |
985 | * mm_update_next_owner() may clear mm->owner to NULL | |
986 | * if it races with swapoff, page migration, etc. | |
987 | * So this can be called with p == NULL. | |
988 | */ | |
989 | if (unlikely(!p)) | |
990 | return NULL; | |
991 | ||
073219e9 | 992 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
993 | } |
994 | ||
df381975 | 995 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 996 | { |
c0ff4b85 | 997 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 998 | |
54595fe2 KH |
999 | rcu_read_lock(); |
1000 | do { | |
6f6acb00 MH |
1001 | /* |
1002 | * Page cache insertions can happen withou an | |
1003 | * actual mm context, e.g. during disk probing | |
1004 | * on boot, loopback IO, acct() writes etc. | |
1005 | */ | |
1006 | if (unlikely(!mm)) | |
df381975 | 1007 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1008 | else { |
1009 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1010 | if (unlikely(!memcg)) | |
1011 | memcg = root_mem_cgroup; | |
1012 | } | |
ec903c0c | 1013 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1014 | rcu_read_unlock(); |
c0ff4b85 | 1015 | return memcg; |
54595fe2 KH |
1016 | } |
1017 | ||
5660048c JW |
1018 | /** |
1019 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1020 | * @root: hierarchy root | |
1021 | * @prev: previously returned memcg, NULL on first invocation | |
1022 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1023 | * | |
1024 | * Returns references to children of the hierarchy below @root, or | |
1025 | * @root itself, or %NULL after a full round-trip. | |
1026 | * | |
1027 | * Caller must pass the return value in @prev on subsequent | |
1028 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1029 | * to cancel a hierarchy walk before the round-trip is complete. | |
1030 | * | |
1031 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1032 | * divide up the memcgs in the hierarchy among all concurrent | |
1033 | * reclaimers operating on the same zone and priority. | |
1034 | */ | |
694fbc0f | 1035 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1036 | struct mem_cgroup *prev, |
694fbc0f | 1037 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1038 | { |
5ac8fb31 JW |
1039 | struct reclaim_iter *uninitialized_var(iter); |
1040 | struct cgroup_subsys_state *css = NULL; | |
9f3a0d09 | 1041 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1042 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1043 | |
694fbc0f AM |
1044 | if (mem_cgroup_disabled()) |
1045 | return NULL; | |
5660048c | 1046 | |
9f3a0d09 JW |
1047 | if (!root) |
1048 | root = root_mem_cgroup; | |
7d74b06f | 1049 | |
9f3a0d09 | 1050 | if (prev && !reclaim) |
5ac8fb31 | 1051 | pos = prev; |
14067bb3 | 1052 | |
9f3a0d09 JW |
1053 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1054 | if (prev) | |
5ac8fb31 | 1055 | goto out; |
694fbc0f | 1056 | return root; |
9f3a0d09 | 1057 | } |
14067bb3 | 1058 | |
542f85f9 | 1059 | rcu_read_lock(); |
5f578161 | 1060 | |
5ac8fb31 JW |
1061 | if (reclaim) { |
1062 | struct mem_cgroup_per_zone *mz; | |
1063 | ||
1064 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
1065 | iter = &mz->iter[reclaim->priority]; | |
1066 | ||
1067 | if (prev && reclaim->generation != iter->generation) | |
1068 | goto out_unlock; | |
1069 | ||
1070 | do { | |
1071 | pos = ACCESS_ONCE(iter->position); | |
1072 | /* | |
1073 | * A racing update may change the position and | |
1074 | * put the last reference, hence css_tryget(), | |
1075 | * or retry to see the updated position. | |
1076 | */ | |
1077 | } while (pos && !css_tryget(&pos->css)); | |
1078 | } | |
1079 | ||
1080 | if (pos) | |
1081 | css = &pos->css; | |
1082 | ||
1083 | for (;;) { | |
1084 | css = css_next_descendant_pre(css, &root->css); | |
1085 | if (!css) { | |
1086 | /* | |
1087 | * Reclaimers share the hierarchy walk, and a | |
1088 | * new one might jump in right at the end of | |
1089 | * the hierarchy - make sure they see at least | |
1090 | * one group and restart from the beginning. | |
1091 | */ | |
1092 | if (!prev) | |
1093 | continue; | |
1094 | break; | |
527a5ec9 | 1095 | } |
7d74b06f | 1096 | |
5ac8fb31 JW |
1097 | /* |
1098 | * Verify the css and acquire a reference. The root | |
1099 | * is provided by the caller, so we know it's alive | |
1100 | * and kicking, and don't take an extra reference. | |
1101 | */ | |
1102 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1103 | |
5ac8fb31 JW |
1104 | if (css == &root->css) |
1105 | break; | |
14067bb3 | 1106 | |
b2052564 | 1107 | if (css_tryget(css)) { |
5ac8fb31 JW |
1108 | /* |
1109 | * Make sure the memcg is initialized: | |
1110 | * mem_cgroup_css_online() orders the the | |
1111 | * initialization against setting the flag. | |
1112 | */ | |
1113 | if (smp_load_acquire(&memcg->initialized)) | |
1114 | break; | |
542f85f9 | 1115 | |
5ac8fb31 | 1116 | css_put(css); |
527a5ec9 | 1117 | } |
9f3a0d09 | 1118 | |
5ac8fb31 | 1119 | memcg = NULL; |
9f3a0d09 | 1120 | } |
5ac8fb31 JW |
1121 | |
1122 | if (reclaim) { | |
1123 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
1124 | if (memcg) | |
1125 | css_get(&memcg->css); | |
1126 | if (pos) | |
1127 | css_put(&pos->css); | |
1128 | } | |
1129 | ||
1130 | /* | |
1131 | * pairs with css_tryget when dereferencing iter->position | |
1132 | * above. | |
1133 | */ | |
1134 | if (pos) | |
1135 | css_put(&pos->css); | |
1136 | ||
1137 | if (!memcg) | |
1138 | iter->generation++; | |
1139 | else if (!prev) | |
1140 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1141 | } |
5ac8fb31 | 1142 | |
542f85f9 MH |
1143 | out_unlock: |
1144 | rcu_read_unlock(); | |
5ac8fb31 | 1145 | out: |
c40046f3 MH |
1146 | if (prev && prev != root) |
1147 | css_put(&prev->css); | |
1148 | ||
9f3a0d09 | 1149 | return memcg; |
14067bb3 | 1150 | } |
7d74b06f | 1151 | |
5660048c JW |
1152 | /** |
1153 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1154 | * @root: hierarchy root | |
1155 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1156 | */ | |
1157 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1158 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1159 | { |
1160 | if (!root) | |
1161 | root = root_mem_cgroup; | |
1162 | if (prev && prev != root) | |
1163 | css_put(&prev->css); | |
1164 | } | |
7d74b06f | 1165 | |
9f3a0d09 JW |
1166 | /* |
1167 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1168 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1169 | * be used for reference counting. | |
1170 | */ | |
1171 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1172 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1173 | iter != NULL; \ |
527a5ec9 | 1174 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1175 | |
9f3a0d09 | 1176 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1177 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1178 | iter != NULL; \ |
527a5ec9 | 1179 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1180 | |
68ae564b | 1181 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1182 | { |
c0ff4b85 | 1183 | struct mem_cgroup *memcg; |
456f998e | 1184 | |
456f998e | 1185 | rcu_read_lock(); |
c0ff4b85 R |
1186 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1187 | if (unlikely(!memcg)) | |
456f998e YH |
1188 | goto out; |
1189 | ||
1190 | switch (idx) { | |
456f998e | 1191 | case PGFAULT: |
0e574a93 JW |
1192 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1193 | break; | |
1194 | case PGMAJFAULT: | |
1195 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1196 | break; |
1197 | default: | |
1198 | BUG(); | |
1199 | } | |
1200 | out: | |
1201 | rcu_read_unlock(); | |
1202 | } | |
68ae564b | 1203 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1204 | |
925b7673 JW |
1205 | /** |
1206 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1207 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1208 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1209 | * |
1210 | * Returns the lru list vector holding pages for the given @zone and | |
1211 | * @mem. This can be the global zone lruvec, if the memory controller | |
1212 | * is disabled. | |
1213 | */ | |
1214 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1215 | struct mem_cgroup *memcg) | |
1216 | { | |
1217 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1218 | struct lruvec *lruvec; |
925b7673 | 1219 | |
bea8c150 HD |
1220 | if (mem_cgroup_disabled()) { |
1221 | lruvec = &zone->lruvec; | |
1222 | goto out; | |
1223 | } | |
925b7673 | 1224 | |
e231875b | 1225 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1226 | lruvec = &mz->lruvec; |
1227 | out: | |
1228 | /* | |
1229 | * Since a node can be onlined after the mem_cgroup was created, | |
1230 | * we have to be prepared to initialize lruvec->zone here; | |
1231 | * and if offlined then reonlined, we need to reinitialize it. | |
1232 | */ | |
1233 | if (unlikely(lruvec->zone != zone)) | |
1234 | lruvec->zone = zone; | |
1235 | return lruvec; | |
925b7673 JW |
1236 | } |
1237 | ||
925b7673 | 1238 | /** |
dfe0e773 | 1239 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1240 | * @page: the page |
fa9add64 | 1241 | * @zone: zone of the page |
dfe0e773 JW |
1242 | * |
1243 | * This function is only safe when following the LRU page isolation | |
1244 | * and putback protocol: the LRU lock must be held, and the page must | |
1245 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1246 | */ |
fa9add64 | 1247 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1248 | { |
08e552c6 | 1249 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1250 | struct mem_cgroup *memcg; |
bea8c150 | 1251 | struct lruvec *lruvec; |
6d12e2d8 | 1252 | |
bea8c150 HD |
1253 | if (mem_cgroup_disabled()) { |
1254 | lruvec = &zone->lruvec; | |
1255 | goto out; | |
1256 | } | |
925b7673 | 1257 | |
1306a85a | 1258 | memcg = page->mem_cgroup; |
7512102c | 1259 | /* |
dfe0e773 | 1260 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1261 | * possibly migrated - before they are charged. |
7512102c | 1262 | */ |
29833315 JW |
1263 | if (!memcg) |
1264 | memcg = root_mem_cgroup; | |
7512102c | 1265 | |
e231875b | 1266 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1267 | lruvec = &mz->lruvec; |
1268 | out: | |
1269 | /* | |
1270 | * Since a node can be onlined after the mem_cgroup was created, | |
1271 | * we have to be prepared to initialize lruvec->zone here; | |
1272 | * and if offlined then reonlined, we need to reinitialize it. | |
1273 | */ | |
1274 | if (unlikely(lruvec->zone != zone)) | |
1275 | lruvec->zone = zone; | |
1276 | return lruvec; | |
08e552c6 | 1277 | } |
b69408e8 | 1278 | |
925b7673 | 1279 | /** |
fa9add64 HD |
1280 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1281 | * @lruvec: mem_cgroup per zone lru vector | |
1282 | * @lru: index of lru list the page is sitting on | |
1283 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1284 | * |
fa9add64 HD |
1285 | * This function must be called when a page is added to or removed from an |
1286 | * lru list. | |
3f58a829 | 1287 | */ |
fa9add64 HD |
1288 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1289 | int nr_pages) | |
3f58a829 MK |
1290 | { |
1291 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1292 | unsigned long *lru_size; |
3f58a829 MK |
1293 | |
1294 | if (mem_cgroup_disabled()) | |
1295 | return; | |
1296 | ||
fa9add64 HD |
1297 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1298 | lru_size = mz->lru_size + lru; | |
1299 | *lru_size += nr_pages; | |
1300 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1301 | } |
544122e5 | 1302 | |
2314b42d | 1303 | bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root) |
3e92041d | 1304 | { |
2314b42d | 1305 | if (root == memcg) |
91c63734 | 1306 | return true; |
2314b42d | 1307 | if (!root->use_hierarchy) |
91c63734 | 1308 | return false; |
2314b42d | 1309 | return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); |
c3ac9a8a JW |
1310 | } |
1311 | ||
2314b42d | 1312 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1313 | { |
2314b42d | 1314 | struct mem_cgroup *task_memcg; |
158e0a2d | 1315 | struct task_struct *p; |
ffbdccf5 | 1316 | bool ret; |
4c4a2214 | 1317 | |
158e0a2d | 1318 | p = find_lock_task_mm(task); |
de077d22 | 1319 | if (p) { |
2314b42d | 1320 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1321 | task_unlock(p); |
1322 | } else { | |
1323 | /* | |
1324 | * All threads may have already detached their mm's, but the oom | |
1325 | * killer still needs to detect if they have already been oom | |
1326 | * killed to prevent needlessly killing additional tasks. | |
1327 | */ | |
ffbdccf5 | 1328 | rcu_read_lock(); |
2314b42d JW |
1329 | task_memcg = mem_cgroup_from_task(task); |
1330 | css_get(&task_memcg->css); | |
ffbdccf5 | 1331 | rcu_read_unlock(); |
de077d22 | 1332 | } |
2314b42d JW |
1333 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1334 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1335 | return ret; |
1336 | } | |
1337 | ||
c56d5c7d | 1338 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1339 | { |
9b272977 | 1340 | unsigned long inactive_ratio; |
14797e23 | 1341 | unsigned long inactive; |
9b272977 | 1342 | unsigned long active; |
c772be93 | 1343 | unsigned long gb; |
14797e23 | 1344 | |
4d7dcca2 HD |
1345 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1346 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1347 | |
c772be93 KM |
1348 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1349 | if (gb) | |
1350 | inactive_ratio = int_sqrt(10 * gb); | |
1351 | else | |
1352 | inactive_ratio = 1; | |
1353 | ||
9b272977 | 1354 | return inactive * inactive_ratio < active; |
14797e23 KM |
1355 | } |
1356 | ||
90cbc250 VD |
1357 | bool mem_cgroup_lruvec_online(struct lruvec *lruvec) |
1358 | { | |
1359 | struct mem_cgroup_per_zone *mz; | |
1360 | struct mem_cgroup *memcg; | |
1361 | ||
1362 | if (mem_cgroup_disabled()) | |
1363 | return true; | |
1364 | ||
1365 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
1366 | memcg = mz->memcg; | |
1367 | ||
1368 | return !!(memcg->css.flags & CSS_ONLINE); | |
1369 | } | |
1370 | ||
3e32cb2e | 1371 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1372 | container_of(counter, struct mem_cgroup, member) |
1373 | ||
19942822 | 1374 | /** |
9d11ea9f | 1375 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1376 | * @memcg: the memory cgroup |
19942822 | 1377 | * |
9d11ea9f | 1378 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1379 | * pages. |
19942822 | 1380 | */ |
c0ff4b85 | 1381 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1382 | { |
3e32cb2e JW |
1383 | unsigned long margin = 0; |
1384 | unsigned long count; | |
1385 | unsigned long limit; | |
9d11ea9f | 1386 | |
3e32cb2e JW |
1387 | count = page_counter_read(&memcg->memory); |
1388 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1389 | if (count < limit) | |
1390 | margin = limit - count; | |
1391 | ||
1392 | if (do_swap_account) { | |
1393 | count = page_counter_read(&memcg->memsw); | |
1394 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1395 | if (count <= limit) | |
1396 | margin = min(margin, limit - count); | |
1397 | } | |
1398 | ||
1399 | return margin; | |
19942822 JW |
1400 | } |
1401 | ||
1f4c025b | 1402 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1403 | { |
a7885eb8 | 1404 | /* root ? */ |
14208b0e | 1405 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1406 | return vm_swappiness; |
1407 | ||
bf1ff263 | 1408 | return memcg->swappiness; |
a7885eb8 KM |
1409 | } |
1410 | ||
32047e2a | 1411 | /* |
bdcbb659 | 1412 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1413 | * |
bdcbb659 QH |
1414 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1415 | * moving cgroups. This is for waiting at high-memory pressure | |
1416 | * caused by "move". | |
32047e2a | 1417 | */ |
c0ff4b85 | 1418 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1419 | { |
2bd9bb20 KH |
1420 | struct mem_cgroup *from; |
1421 | struct mem_cgroup *to; | |
4b534334 | 1422 | bool ret = false; |
2bd9bb20 KH |
1423 | /* |
1424 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1425 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1426 | */ | |
1427 | spin_lock(&mc.lock); | |
1428 | from = mc.from; | |
1429 | to = mc.to; | |
1430 | if (!from) | |
1431 | goto unlock; | |
3e92041d | 1432 | |
2314b42d JW |
1433 | ret = mem_cgroup_is_descendant(from, memcg) || |
1434 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1435 | unlock: |
1436 | spin_unlock(&mc.lock); | |
4b534334 KH |
1437 | return ret; |
1438 | } | |
1439 | ||
c0ff4b85 | 1440 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1441 | { |
1442 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1443 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1444 | DEFINE_WAIT(wait); |
1445 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1446 | /* moving charge context might have finished. */ | |
1447 | if (mc.moving_task) | |
1448 | schedule(); | |
1449 | finish_wait(&mc.waitq, &wait); | |
1450 | return true; | |
1451 | } | |
1452 | } | |
1453 | return false; | |
1454 | } | |
1455 | ||
58cf188e | 1456 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1457 | /** |
58cf188e | 1458 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1459 | * @memcg: The memory cgroup that went over limit |
1460 | * @p: Task that is going to be killed | |
1461 | * | |
1462 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1463 | * enabled | |
1464 | */ | |
1465 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1466 | { | |
e61734c5 | 1467 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1468 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1469 | struct mem_cgroup *iter; |
1470 | unsigned int i; | |
e222432b | 1471 | |
58cf188e | 1472 | if (!p) |
e222432b BS |
1473 | return; |
1474 | ||
08088cb9 | 1475 | mutex_lock(&oom_info_lock); |
e222432b BS |
1476 | rcu_read_lock(); |
1477 | ||
e61734c5 TH |
1478 | pr_info("Task in "); |
1479 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
0346dadb | 1480 | pr_cont(" killed as a result of limit of "); |
e61734c5 | 1481 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1482 | pr_cont("\n"); |
e222432b | 1483 | |
e222432b BS |
1484 | rcu_read_unlock(); |
1485 | ||
3e32cb2e JW |
1486 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1487 | K((u64)page_counter_read(&memcg->memory)), | |
1488 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1489 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1490 | K((u64)page_counter_read(&memcg->memsw)), | |
1491 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1492 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1493 | K((u64)page_counter_read(&memcg->kmem)), | |
1494 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1495 | |
1496 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1497 | pr_info("Memory cgroup stats for "); |
1498 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1499 | pr_cont(":"); |
1500 | ||
1501 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1502 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1503 | continue; | |
1504 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1505 | K(mem_cgroup_read_stat(iter, i))); | |
1506 | } | |
1507 | ||
1508 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1509 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1510 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1511 | ||
1512 | pr_cont("\n"); | |
1513 | } | |
08088cb9 | 1514 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1515 | } |
1516 | ||
81d39c20 KH |
1517 | /* |
1518 | * This function returns the number of memcg under hierarchy tree. Returns | |
1519 | * 1(self count) if no children. | |
1520 | */ | |
c0ff4b85 | 1521 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1522 | { |
1523 | int num = 0; | |
7d74b06f KH |
1524 | struct mem_cgroup *iter; |
1525 | ||
c0ff4b85 | 1526 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1527 | num++; |
81d39c20 KH |
1528 | return num; |
1529 | } | |
1530 | ||
a63d83f4 DR |
1531 | /* |
1532 | * Return the memory (and swap, if configured) limit for a memcg. | |
1533 | */ | |
3e32cb2e | 1534 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1535 | { |
3e32cb2e | 1536 | unsigned long limit; |
f3e8eb70 | 1537 | |
3e32cb2e | 1538 | limit = memcg->memory.limit; |
9a5a8f19 | 1539 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1540 | unsigned long memsw_limit; |
9a5a8f19 | 1541 | |
3e32cb2e JW |
1542 | memsw_limit = memcg->memsw.limit; |
1543 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1544 | } |
9a5a8f19 | 1545 | return limit; |
a63d83f4 DR |
1546 | } |
1547 | ||
19965460 DR |
1548 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1549 | int order) | |
9cbb78bb DR |
1550 | { |
1551 | struct mem_cgroup *iter; | |
1552 | unsigned long chosen_points = 0; | |
1553 | unsigned long totalpages; | |
1554 | unsigned int points = 0; | |
1555 | struct task_struct *chosen = NULL; | |
1556 | ||
876aafbf | 1557 | /* |
465adcf1 DR |
1558 | * If current has a pending SIGKILL or is exiting, then automatically |
1559 | * select it. The goal is to allow it to allocate so that it may | |
1560 | * quickly exit and free its memory. | |
876aafbf | 1561 | */ |
d003f371 | 1562 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
49550b60 | 1563 | mark_tsk_oom_victim(current); |
876aafbf DR |
1564 | return; |
1565 | } | |
1566 | ||
1567 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1568 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1569 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1570 | struct css_task_iter it; |
9cbb78bb DR |
1571 | struct task_struct *task; |
1572 | ||
72ec7029 TH |
1573 | css_task_iter_start(&iter->css, &it); |
1574 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1575 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1576 | false)) { | |
1577 | case OOM_SCAN_SELECT: | |
1578 | if (chosen) | |
1579 | put_task_struct(chosen); | |
1580 | chosen = task; | |
1581 | chosen_points = ULONG_MAX; | |
1582 | get_task_struct(chosen); | |
1583 | /* fall through */ | |
1584 | case OOM_SCAN_CONTINUE: | |
1585 | continue; | |
1586 | case OOM_SCAN_ABORT: | |
72ec7029 | 1587 | css_task_iter_end(&it); |
9cbb78bb DR |
1588 | mem_cgroup_iter_break(memcg, iter); |
1589 | if (chosen) | |
1590 | put_task_struct(chosen); | |
1591 | return; | |
1592 | case OOM_SCAN_OK: | |
1593 | break; | |
1594 | }; | |
1595 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1596 | if (!points || points < chosen_points) |
1597 | continue; | |
1598 | /* Prefer thread group leaders for display purposes */ | |
1599 | if (points == chosen_points && | |
1600 | thread_group_leader(chosen)) | |
1601 | continue; | |
1602 | ||
1603 | if (chosen) | |
1604 | put_task_struct(chosen); | |
1605 | chosen = task; | |
1606 | chosen_points = points; | |
1607 | get_task_struct(chosen); | |
9cbb78bb | 1608 | } |
72ec7029 | 1609 | css_task_iter_end(&it); |
9cbb78bb DR |
1610 | } |
1611 | ||
1612 | if (!chosen) | |
1613 | return; | |
1614 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1615 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1616 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1617 | } |
1618 | ||
ae6e71d3 MC |
1619 | #if MAX_NUMNODES > 1 |
1620 | ||
4d0c066d KH |
1621 | /** |
1622 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1623 | * @memcg: the target memcg |
4d0c066d KH |
1624 | * @nid: the node ID to be checked. |
1625 | * @noswap : specify true here if the user wants flle only information. | |
1626 | * | |
1627 | * This function returns whether the specified memcg contains any | |
1628 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1629 | * pages in the node. | |
1630 | */ | |
c0ff4b85 | 1631 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1632 | int nid, bool noswap) |
1633 | { | |
c0ff4b85 | 1634 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1635 | return true; |
1636 | if (noswap || !total_swap_pages) | |
1637 | return false; | |
c0ff4b85 | 1638 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1639 | return true; |
1640 | return false; | |
1641 | ||
1642 | } | |
889976db YH |
1643 | |
1644 | /* | |
1645 | * Always updating the nodemask is not very good - even if we have an empty | |
1646 | * list or the wrong list here, we can start from some node and traverse all | |
1647 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1648 | * | |
1649 | */ | |
c0ff4b85 | 1650 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1651 | { |
1652 | int nid; | |
453a9bf3 KH |
1653 | /* |
1654 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1655 | * pagein/pageout changes since the last update. | |
1656 | */ | |
c0ff4b85 | 1657 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1658 | return; |
c0ff4b85 | 1659 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1660 | return; |
1661 | ||
889976db | 1662 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1663 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1664 | |
31aaea4a | 1665 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1666 | |
c0ff4b85 R |
1667 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1668 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1669 | } |
453a9bf3 | 1670 | |
c0ff4b85 R |
1671 | atomic_set(&memcg->numainfo_events, 0); |
1672 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1673 | } |
1674 | ||
1675 | /* | |
1676 | * Selecting a node where we start reclaim from. Because what we need is just | |
1677 | * reducing usage counter, start from anywhere is O,K. Considering | |
1678 | * memory reclaim from current node, there are pros. and cons. | |
1679 | * | |
1680 | * Freeing memory from current node means freeing memory from a node which | |
1681 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1682 | * hit limits, it will see a contention on a node. But freeing from remote | |
1683 | * node means more costs for memory reclaim because of memory latency. | |
1684 | * | |
1685 | * Now, we use round-robin. Better algorithm is welcomed. | |
1686 | */ | |
c0ff4b85 | 1687 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1688 | { |
1689 | int node; | |
1690 | ||
c0ff4b85 R |
1691 | mem_cgroup_may_update_nodemask(memcg); |
1692 | node = memcg->last_scanned_node; | |
889976db | 1693 | |
c0ff4b85 | 1694 | node = next_node(node, memcg->scan_nodes); |
889976db | 1695 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1696 | node = first_node(memcg->scan_nodes); |
889976db YH |
1697 | /* |
1698 | * We call this when we hit limit, not when pages are added to LRU. | |
1699 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1700 | * memcg is too small and all pages are not on LRU. In that case, | |
1701 | * we use curret node. | |
1702 | */ | |
1703 | if (unlikely(node == MAX_NUMNODES)) | |
1704 | node = numa_node_id(); | |
1705 | ||
c0ff4b85 | 1706 | memcg->last_scanned_node = node; |
889976db YH |
1707 | return node; |
1708 | } | |
889976db | 1709 | #else |
c0ff4b85 | 1710 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1711 | { |
1712 | return 0; | |
1713 | } | |
1714 | #endif | |
1715 | ||
0608f43d AM |
1716 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1717 | struct zone *zone, | |
1718 | gfp_t gfp_mask, | |
1719 | unsigned long *total_scanned) | |
1720 | { | |
1721 | struct mem_cgroup *victim = NULL; | |
1722 | int total = 0; | |
1723 | int loop = 0; | |
1724 | unsigned long excess; | |
1725 | unsigned long nr_scanned; | |
1726 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1727 | .zone = zone, | |
1728 | .priority = 0, | |
1729 | }; | |
1730 | ||
3e32cb2e | 1731 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1732 | |
1733 | while (1) { | |
1734 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1735 | if (!victim) { | |
1736 | loop++; | |
1737 | if (loop >= 2) { | |
1738 | /* | |
1739 | * If we have not been able to reclaim | |
1740 | * anything, it might because there are | |
1741 | * no reclaimable pages under this hierarchy | |
1742 | */ | |
1743 | if (!total) | |
1744 | break; | |
1745 | /* | |
1746 | * We want to do more targeted reclaim. | |
1747 | * excess >> 2 is not to excessive so as to | |
1748 | * reclaim too much, nor too less that we keep | |
1749 | * coming back to reclaim from this cgroup | |
1750 | */ | |
1751 | if (total >= (excess >> 2) || | |
1752 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1753 | break; | |
1754 | } | |
1755 | continue; | |
1756 | } | |
0608f43d AM |
1757 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1758 | zone, &nr_scanned); | |
1759 | *total_scanned += nr_scanned; | |
3e32cb2e | 1760 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1761 | break; |
6d61ef40 | 1762 | } |
0608f43d AM |
1763 | mem_cgroup_iter_break(root_memcg, victim); |
1764 | return total; | |
6d61ef40 BS |
1765 | } |
1766 | ||
0056f4e6 JW |
1767 | #ifdef CONFIG_LOCKDEP |
1768 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1769 | .name = "memcg_oom_lock", | |
1770 | }; | |
1771 | #endif | |
1772 | ||
fb2a6fc5 JW |
1773 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1774 | ||
867578cb KH |
1775 | /* |
1776 | * Check OOM-Killer is already running under our hierarchy. | |
1777 | * If someone is running, return false. | |
1778 | */ | |
fb2a6fc5 | 1779 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1780 | { |
79dfdacc | 1781 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1782 | |
fb2a6fc5 JW |
1783 | spin_lock(&memcg_oom_lock); |
1784 | ||
9f3a0d09 | 1785 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1786 | if (iter->oom_lock) { |
79dfdacc MH |
1787 | /* |
1788 | * this subtree of our hierarchy is already locked | |
1789 | * so we cannot give a lock. | |
1790 | */ | |
79dfdacc | 1791 | failed = iter; |
9f3a0d09 JW |
1792 | mem_cgroup_iter_break(memcg, iter); |
1793 | break; | |
23751be0 JW |
1794 | } else |
1795 | iter->oom_lock = true; | |
7d74b06f | 1796 | } |
867578cb | 1797 | |
fb2a6fc5 JW |
1798 | if (failed) { |
1799 | /* | |
1800 | * OK, we failed to lock the whole subtree so we have | |
1801 | * to clean up what we set up to the failing subtree | |
1802 | */ | |
1803 | for_each_mem_cgroup_tree(iter, memcg) { | |
1804 | if (iter == failed) { | |
1805 | mem_cgroup_iter_break(memcg, iter); | |
1806 | break; | |
1807 | } | |
1808 | iter->oom_lock = false; | |
79dfdacc | 1809 | } |
0056f4e6 JW |
1810 | } else |
1811 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1812 | |
1813 | spin_unlock(&memcg_oom_lock); | |
1814 | ||
1815 | return !failed; | |
a636b327 | 1816 | } |
0b7f569e | 1817 | |
fb2a6fc5 | 1818 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1819 | { |
7d74b06f KH |
1820 | struct mem_cgroup *iter; |
1821 | ||
fb2a6fc5 | 1822 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1823 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1824 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1825 | iter->oom_lock = false; |
fb2a6fc5 | 1826 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1827 | } |
1828 | ||
c0ff4b85 | 1829 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1830 | { |
1831 | struct mem_cgroup *iter; | |
1832 | ||
c0ff4b85 | 1833 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1834 | atomic_inc(&iter->under_oom); |
1835 | } | |
1836 | ||
c0ff4b85 | 1837 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1838 | { |
1839 | struct mem_cgroup *iter; | |
1840 | ||
867578cb KH |
1841 | /* |
1842 | * When a new child is created while the hierarchy is under oom, | |
1843 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1844 | * atomic_add_unless() here. | |
1845 | */ | |
c0ff4b85 | 1846 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1847 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1848 | } |
1849 | ||
867578cb KH |
1850 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1851 | ||
dc98df5a | 1852 | struct oom_wait_info { |
d79154bb | 1853 | struct mem_cgroup *memcg; |
dc98df5a KH |
1854 | wait_queue_t wait; |
1855 | }; | |
1856 | ||
1857 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1858 | unsigned mode, int sync, void *arg) | |
1859 | { | |
d79154bb HD |
1860 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1861 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1862 | struct oom_wait_info *oom_wait_info; |
1863 | ||
1864 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1865 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1866 | |
2314b42d JW |
1867 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1868 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1869 | return 0; |
dc98df5a KH |
1870 | return autoremove_wake_function(wait, mode, sync, arg); |
1871 | } | |
1872 | ||
c0ff4b85 | 1873 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1874 | { |
3812c8c8 | 1875 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
1876 | /* for filtering, pass "memcg" as argument. */ |
1877 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1878 | } |
1879 | ||
c0ff4b85 | 1880 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1881 | { |
c0ff4b85 R |
1882 | if (memcg && atomic_read(&memcg->under_oom)) |
1883 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1884 | } |
1885 | ||
3812c8c8 | 1886 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1887 | { |
3812c8c8 JW |
1888 | if (!current->memcg_oom.may_oom) |
1889 | return; | |
867578cb | 1890 | /* |
49426420 JW |
1891 | * We are in the middle of the charge context here, so we |
1892 | * don't want to block when potentially sitting on a callstack | |
1893 | * that holds all kinds of filesystem and mm locks. | |
1894 | * | |
1895 | * Also, the caller may handle a failed allocation gracefully | |
1896 | * (like optional page cache readahead) and so an OOM killer | |
1897 | * invocation might not even be necessary. | |
1898 | * | |
1899 | * That's why we don't do anything here except remember the | |
1900 | * OOM context and then deal with it at the end of the page | |
1901 | * fault when the stack is unwound, the locks are released, | |
1902 | * and when we know whether the fault was overall successful. | |
867578cb | 1903 | */ |
49426420 JW |
1904 | css_get(&memcg->css); |
1905 | current->memcg_oom.memcg = memcg; | |
1906 | current->memcg_oom.gfp_mask = mask; | |
1907 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1908 | } |
1909 | ||
1910 | /** | |
1911 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1912 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1913 | * |
49426420 JW |
1914 | * This has to be called at the end of a page fault if the memcg OOM |
1915 | * handler was enabled. | |
3812c8c8 | 1916 | * |
49426420 | 1917 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1918 | * sleep on a waitqueue until the userspace task resolves the |
1919 | * situation. Sleeping directly in the charge context with all kinds | |
1920 | * of locks held is not a good idea, instead we remember an OOM state | |
1921 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1922 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1923 | * |
1924 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1925 | * completed, %false otherwise. |
3812c8c8 | 1926 | */ |
49426420 | 1927 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1928 | { |
49426420 | 1929 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1930 | struct oom_wait_info owait; |
49426420 | 1931 | bool locked; |
3812c8c8 JW |
1932 | |
1933 | /* OOM is global, do not handle */ | |
3812c8c8 | 1934 | if (!memcg) |
49426420 | 1935 | return false; |
3812c8c8 | 1936 | |
c32b3cbe | 1937 | if (!handle || oom_killer_disabled) |
49426420 | 1938 | goto cleanup; |
3812c8c8 JW |
1939 | |
1940 | owait.memcg = memcg; | |
1941 | owait.wait.flags = 0; | |
1942 | owait.wait.func = memcg_oom_wake_function; | |
1943 | owait.wait.private = current; | |
1944 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1945 | |
3812c8c8 | 1946 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1947 | mem_cgroup_mark_under_oom(memcg); |
1948 | ||
1949 | locked = mem_cgroup_oom_trylock(memcg); | |
1950 | ||
1951 | if (locked) | |
1952 | mem_cgroup_oom_notify(memcg); | |
1953 | ||
1954 | if (locked && !memcg->oom_kill_disable) { | |
1955 | mem_cgroup_unmark_under_oom(memcg); | |
1956 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1957 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1958 | current->memcg_oom.order); | |
1959 | } else { | |
3812c8c8 | 1960 | schedule(); |
49426420 JW |
1961 | mem_cgroup_unmark_under_oom(memcg); |
1962 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1963 | } | |
1964 | ||
1965 | if (locked) { | |
fb2a6fc5 JW |
1966 | mem_cgroup_oom_unlock(memcg); |
1967 | /* | |
1968 | * There is no guarantee that an OOM-lock contender | |
1969 | * sees the wakeups triggered by the OOM kill | |
1970 | * uncharges. Wake any sleepers explicitely. | |
1971 | */ | |
1972 | memcg_oom_recover(memcg); | |
1973 | } | |
49426420 JW |
1974 | cleanup: |
1975 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 1976 | css_put(&memcg->css); |
867578cb | 1977 | return true; |
0b7f569e KH |
1978 | } |
1979 | ||
d7365e78 JW |
1980 | /** |
1981 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1982 | * @page: page that is going to change accounted state | |
32047e2a | 1983 | * |
d7365e78 JW |
1984 | * This function must mark the beginning of an accounted page state |
1985 | * change to prevent double accounting when the page is concurrently | |
1986 | * being moved to another memcg: | |
32047e2a | 1987 | * |
6de22619 | 1988 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1989 | * if (TestClearPageState(page)) |
1990 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1991 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1992 | */ |
6de22619 | 1993 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1994 | { |
1995 | struct mem_cgroup *memcg; | |
6de22619 | 1996 | unsigned long flags; |
89c06bd5 | 1997 | |
6de22619 JW |
1998 | /* |
1999 | * The RCU lock is held throughout the transaction. The fast | |
2000 | * path can get away without acquiring the memcg->move_lock | |
2001 | * because page moving starts with an RCU grace period. | |
2002 | * | |
2003 | * The RCU lock also protects the memcg from being freed when | |
2004 | * the page state that is going to change is the only thing | |
2005 | * preventing the page from being uncharged. | |
2006 | * E.g. end-writeback clearing PageWriteback(), which allows | |
2007 | * migration to go ahead and uncharge the page before the | |
2008 | * account transaction might be complete. | |
2009 | */ | |
d7365e78 JW |
2010 | rcu_read_lock(); |
2011 | ||
2012 | if (mem_cgroup_disabled()) | |
2013 | return NULL; | |
89c06bd5 | 2014 | again: |
1306a85a | 2015 | memcg = page->mem_cgroup; |
29833315 | 2016 | if (unlikely(!memcg)) |
d7365e78 JW |
2017 | return NULL; |
2018 | ||
bdcbb659 | 2019 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2020 | return memcg; |
89c06bd5 | 2021 | |
6de22619 | 2022 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 2023 | if (memcg != page->mem_cgroup) { |
6de22619 | 2024 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2025 | goto again; |
2026 | } | |
6de22619 JW |
2027 | |
2028 | /* | |
2029 | * When charge migration first begins, we can have locked and | |
2030 | * unlocked page stat updates happening concurrently. Track | |
2031 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
2032 | */ | |
2033 | memcg->move_lock_task = current; | |
2034 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
2035 | |
2036 | return memcg; | |
89c06bd5 KH |
2037 | } |
2038 | ||
d7365e78 JW |
2039 | /** |
2040 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2041 | * @memcg: the memcg that was accounted against | |
d7365e78 | 2042 | */ |
6de22619 | 2043 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 2044 | { |
6de22619 JW |
2045 | if (memcg && memcg->move_lock_task == current) { |
2046 | unsigned long flags = memcg->move_lock_flags; | |
2047 | ||
2048 | memcg->move_lock_task = NULL; | |
2049 | memcg->move_lock_flags = 0; | |
2050 | ||
2051 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2052 | } | |
89c06bd5 | 2053 | |
d7365e78 | 2054 | rcu_read_unlock(); |
89c06bd5 KH |
2055 | } |
2056 | ||
d7365e78 JW |
2057 | /** |
2058 | * mem_cgroup_update_page_stat - update page state statistics | |
2059 | * @memcg: memcg to account against | |
2060 | * @idx: page state item to account | |
2061 | * @val: number of pages (positive or negative) | |
2062 | * | |
2063 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2064 | */ | |
2065 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2066 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2067 | { |
658b72c5 | 2068 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2069 | |
d7365e78 JW |
2070 | if (memcg) |
2071 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2072 | } |
26174efd | 2073 | |
cdec2e42 KH |
2074 | /* |
2075 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2076 | * TODO: maybe necessary to use big numbers in big irons. | |
2077 | */ | |
7ec99d62 | 2078 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2079 | struct memcg_stock_pcp { |
2080 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2081 | unsigned int nr_pages; |
cdec2e42 | 2082 | struct work_struct work; |
26fe6168 | 2083 | unsigned long flags; |
a0db00fc | 2084 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2085 | }; |
2086 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2087 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2088 | |
a0956d54 SS |
2089 | /** |
2090 | * consume_stock: Try to consume stocked charge on this cpu. | |
2091 | * @memcg: memcg to consume from. | |
2092 | * @nr_pages: how many pages to charge. | |
2093 | * | |
2094 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2095 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2096 | * service an allocation will refill the stock. | |
2097 | * | |
2098 | * returns true if successful, false otherwise. | |
cdec2e42 | 2099 | */ |
a0956d54 | 2100 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2101 | { |
2102 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2103 | bool ret = false; |
cdec2e42 | 2104 | |
a0956d54 | 2105 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2106 | return ret; |
a0956d54 | 2107 | |
cdec2e42 | 2108 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2109 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2110 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2111 | ret = true; |
2112 | } | |
cdec2e42 KH |
2113 | put_cpu_var(memcg_stock); |
2114 | return ret; | |
2115 | } | |
2116 | ||
2117 | /* | |
3e32cb2e | 2118 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2119 | */ |
2120 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2121 | { | |
2122 | struct mem_cgroup *old = stock->cached; | |
2123 | ||
11c9ea4e | 2124 | if (stock->nr_pages) { |
3e32cb2e | 2125 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2126 | if (do_swap_account) |
3e32cb2e | 2127 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2128 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2129 | stock->nr_pages = 0; |
cdec2e42 KH |
2130 | } |
2131 | stock->cached = NULL; | |
cdec2e42 KH |
2132 | } |
2133 | ||
2134 | /* | |
2135 | * This must be called under preempt disabled or must be called by | |
2136 | * a thread which is pinned to local cpu. | |
2137 | */ | |
2138 | static void drain_local_stock(struct work_struct *dummy) | |
2139 | { | |
7c8e0181 | 2140 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2141 | drain_stock(stock); |
26fe6168 | 2142 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2143 | } |
2144 | ||
2145 | /* | |
3e32cb2e | 2146 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2147 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2148 | */ |
c0ff4b85 | 2149 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2150 | { |
2151 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2152 | ||
c0ff4b85 | 2153 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2154 | drain_stock(stock); |
c0ff4b85 | 2155 | stock->cached = memcg; |
cdec2e42 | 2156 | } |
11c9ea4e | 2157 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2158 | put_cpu_var(memcg_stock); |
2159 | } | |
2160 | ||
2161 | /* | |
c0ff4b85 | 2162 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2163 | * of the hierarchy under it. |
cdec2e42 | 2164 | */ |
6d3d6aa2 | 2165 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2166 | { |
26fe6168 | 2167 | int cpu, curcpu; |
d38144b7 | 2168 | |
6d3d6aa2 JW |
2169 | /* If someone's already draining, avoid adding running more workers. */ |
2170 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2171 | return; | |
cdec2e42 | 2172 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2173 | get_online_cpus(); |
5af12d0e | 2174 | curcpu = get_cpu(); |
cdec2e42 KH |
2175 | for_each_online_cpu(cpu) { |
2176 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2177 | struct mem_cgroup *memcg; |
26fe6168 | 2178 | |
c0ff4b85 R |
2179 | memcg = stock->cached; |
2180 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2181 | continue; |
2314b42d | 2182 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 2183 | continue; |
d1a05b69 MH |
2184 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2185 | if (cpu == curcpu) | |
2186 | drain_local_stock(&stock->work); | |
2187 | else | |
2188 | schedule_work_on(cpu, &stock->work); | |
2189 | } | |
cdec2e42 | 2190 | } |
5af12d0e | 2191 | put_cpu(); |
f894ffa8 | 2192 | put_online_cpus(); |
9f50fad6 | 2193 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2194 | } |
2195 | ||
711d3d2c KH |
2196 | /* |
2197 | * This function drains percpu counter value from DEAD cpu and | |
2198 | * move it to local cpu. Note that this function can be preempted. | |
2199 | */ | |
c0ff4b85 | 2200 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2201 | { |
2202 | int i; | |
2203 | ||
c0ff4b85 | 2204 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2205 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2206 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2207 | |
c0ff4b85 R |
2208 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2209 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2210 | } |
e9f8974f | 2211 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2212 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2213 | |
c0ff4b85 R |
2214 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2215 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2216 | } |
c0ff4b85 | 2217 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2218 | } |
2219 | ||
0db0628d | 2220 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2221 | unsigned long action, |
2222 | void *hcpu) | |
2223 | { | |
2224 | int cpu = (unsigned long)hcpu; | |
2225 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2226 | struct mem_cgroup *iter; |
cdec2e42 | 2227 | |
619d094b | 2228 | if (action == CPU_ONLINE) |
1489ebad | 2229 | return NOTIFY_OK; |
1489ebad | 2230 | |
d833049b | 2231 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2232 | return NOTIFY_OK; |
711d3d2c | 2233 | |
9f3a0d09 | 2234 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2235 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2236 | ||
cdec2e42 KH |
2237 | stock = &per_cpu(memcg_stock, cpu); |
2238 | drain_stock(stock); | |
2239 | return NOTIFY_OK; | |
2240 | } | |
2241 | ||
00501b53 JW |
2242 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2243 | unsigned int nr_pages) | |
8a9f3ccd | 2244 | { |
7ec99d62 | 2245 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2246 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2247 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2248 | struct page_counter *counter; |
6539cc05 | 2249 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2250 | bool may_swap = true; |
2251 | bool drained = false; | |
05b84301 | 2252 | int ret = 0; |
a636b327 | 2253 | |
ce00a967 JW |
2254 | if (mem_cgroup_is_root(memcg)) |
2255 | goto done; | |
6539cc05 | 2256 | retry: |
b6b6cc72 MH |
2257 | if (consume_stock(memcg, nr_pages)) |
2258 | goto done; | |
8a9f3ccd | 2259 | |
3fbe7244 | 2260 | if (!do_swap_account || |
3e32cb2e JW |
2261 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2262 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2263 | goto done_restock; |
3fbe7244 | 2264 | if (do_swap_account) |
3e32cb2e JW |
2265 | page_counter_uncharge(&memcg->memsw, batch); |
2266 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2267 | } else { |
3e32cb2e | 2268 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2269 | may_swap = false; |
3fbe7244 | 2270 | } |
7a81b88c | 2271 | |
6539cc05 JW |
2272 | if (batch > nr_pages) { |
2273 | batch = nr_pages; | |
2274 | goto retry; | |
2275 | } | |
6d61ef40 | 2276 | |
06b078fc JW |
2277 | /* |
2278 | * Unlike in global OOM situations, memcg is not in a physical | |
2279 | * memory shortage. Allow dying and OOM-killed tasks to | |
2280 | * bypass the last charges so that they can exit quickly and | |
2281 | * free their memory. | |
2282 | */ | |
2283 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2284 | fatal_signal_pending(current) || | |
2285 | current->flags & PF_EXITING)) | |
2286 | goto bypass; | |
2287 | ||
2288 | if (unlikely(task_in_memcg_oom(current))) | |
2289 | goto nomem; | |
2290 | ||
6539cc05 JW |
2291 | if (!(gfp_mask & __GFP_WAIT)) |
2292 | goto nomem; | |
4b534334 | 2293 | |
241994ed JW |
2294 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2295 | ||
b70a2a21 JW |
2296 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2297 | gfp_mask, may_swap); | |
6539cc05 | 2298 | |
61e02c74 | 2299 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2300 | goto retry; |
28c34c29 | 2301 | |
b70a2a21 | 2302 | if (!drained) { |
6d3d6aa2 | 2303 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2304 | drained = true; |
2305 | goto retry; | |
2306 | } | |
2307 | ||
28c34c29 JW |
2308 | if (gfp_mask & __GFP_NORETRY) |
2309 | goto nomem; | |
6539cc05 JW |
2310 | /* |
2311 | * Even though the limit is exceeded at this point, reclaim | |
2312 | * may have been able to free some pages. Retry the charge | |
2313 | * before killing the task. | |
2314 | * | |
2315 | * Only for regular pages, though: huge pages are rather | |
2316 | * unlikely to succeed so close to the limit, and we fall back | |
2317 | * to regular pages anyway in case of failure. | |
2318 | */ | |
61e02c74 | 2319 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2320 | goto retry; |
2321 | /* | |
2322 | * At task move, charge accounts can be doubly counted. So, it's | |
2323 | * better to wait until the end of task_move if something is going on. | |
2324 | */ | |
2325 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2326 | goto retry; | |
2327 | ||
9b130619 JW |
2328 | if (nr_retries--) |
2329 | goto retry; | |
2330 | ||
06b078fc JW |
2331 | if (gfp_mask & __GFP_NOFAIL) |
2332 | goto bypass; | |
2333 | ||
6539cc05 JW |
2334 | if (fatal_signal_pending(current)) |
2335 | goto bypass; | |
2336 | ||
241994ed JW |
2337 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2338 | ||
61e02c74 | 2339 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2340 | nomem: |
6d1fdc48 | 2341 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2342 | return -ENOMEM; |
867578cb | 2343 | bypass: |
ce00a967 | 2344 | return -EINTR; |
6539cc05 JW |
2345 | |
2346 | done_restock: | |
e8ea14cc | 2347 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2348 | if (batch > nr_pages) |
2349 | refill_stock(memcg, batch - nr_pages); | |
241994ed JW |
2350 | /* |
2351 | * If the hierarchy is above the normal consumption range, | |
2352 | * make the charging task trim their excess contribution. | |
2353 | */ | |
2354 | do { | |
2355 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2356 | continue; | |
2357 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
2358 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
2359 | } while ((memcg = parent_mem_cgroup(memcg))); | |
6539cc05 | 2360 | done: |
05b84301 | 2361 | return ret; |
7a81b88c | 2362 | } |
8a9f3ccd | 2363 | |
00501b53 | 2364 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2365 | { |
ce00a967 JW |
2366 | if (mem_cgroup_is_root(memcg)) |
2367 | return; | |
2368 | ||
3e32cb2e | 2369 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2370 | if (do_swap_account) |
3e32cb2e | 2371 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2372 | |
e8ea14cc | 2373 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2374 | } |
2375 | ||
a3b2d692 KH |
2376 | /* |
2377 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2378 | * rcu_read_lock(). The caller is responsible for calling |
2379 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2380 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2381 | */ |
2382 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2383 | { | |
a3b2d692 KH |
2384 | /* ID 0 is unused ID */ |
2385 | if (!id) | |
2386 | return NULL; | |
34c00c31 | 2387 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2388 | } |
2389 | ||
0a31bc97 JW |
2390 | /* |
2391 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2392 | * @page: the page | |
2393 | * | |
2394 | * Look up, get a css reference, and return the memcg that owns @page. | |
2395 | * | |
2396 | * The page must be locked to prevent racing with swap-in and page | |
2397 | * cache charges. If coming from an unlocked page table, the caller | |
2398 | * must ensure the page is on the LRU or this can race with charging. | |
2399 | */ | |
e42d9d5d | 2400 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2401 | { |
29833315 | 2402 | struct mem_cgroup *memcg; |
a3b2d692 | 2403 | unsigned short id; |
b5a84319 KH |
2404 | swp_entry_t ent; |
2405 | ||
309381fe | 2406 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2407 | |
1306a85a | 2408 | memcg = page->mem_cgroup; |
29833315 JW |
2409 | if (memcg) { |
2410 | if (!css_tryget_online(&memcg->css)) | |
c0ff4b85 | 2411 | memcg = NULL; |
e42d9d5d | 2412 | } else if (PageSwapCache(page)) { |
3c776e64 | 2413 | ent.val = page_private(page); |
9fb4b7cc | 2414 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2415 | rcu_read_lock(); |
c0ff4b85 | 2416 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2417 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2418 | memcg = NULL; |
a3b2d692 | 2419 | rcu_read_unlock(); |
3c776e64 | 2420 | } |
c0ff4b85 | 2421 | return memcg; |
b5a84319 KH |
2422 | } |
2423 | ||
0a31bc97 JW |
2424 | static void lock_page_lru(struct page *page, int *isolated) |
2425 | { | |
2426 | struct zone *zone = page_zone(page); | |
2427 | ||
2428 | spin_lock_irq(&zone->lru_lock); | |
2429 | if (PageLRU(page)) { | |
2430 | struct lruvec *lruvec; | |
2431 | ||
2432 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2433 | ClearPageLRU(page); | |
2434 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2435 | *isolated = 1; | |
2436 | } else | |
2437 | *isolated = 0; | |
2438 | } | |
2439 | ||
2440 | static void unlock_page_lru(struct page *page, int isolated) | |
2441 | { | |
2442 | struct zone *zone = page_zone(page); | |
2443 | ||
2444 | if (isolated) { | |
2445 | struct lruvec *lruvec; | |
2446 | ||
2447 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2448 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2449 | SetPageLRU(page); | |
2450 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2451 | } | |
2452 | spin_unlock_irq(&zone->lru_lock); | |
2453 | } | |
2454 | ||
00501b53 | 2455 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2456 | bool lrucare) |
7a81b88c | 2457 | { |
0a31bc97 | 2458 | int isolated; |
9ce70c02 | 2459 | |
1306a85a | 2460 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2461 | |
2462 | /* | |
2463 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2464 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2465 | */ | |
0a31bc97 JW |
2466 | if (lrucare) |
2467 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2468 | |
0a31bc97 JW |
2469 | /* |
2470 | * Nobody should be changing or seriously looking at | |
1306a85a | 2471 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2472 | * |
2473 | * - the page is uncharged | |
2474 | * | |
2475 | * - the page is off-LRU | |
2476 | * | |
2477 | * - an anonymous fault has exclusive page access, except for | |
2478 | * a locked page table | |
2479 | * | |
2480 | * - a page cache insertion, a swapin fault, or a migration | |
2481 | * have the page locked | |
2482 | */ | |
1306a85a | 2483 | page->mem_cgroup = memcg; |
9ce70c02 | 2484 | |
0a31bc97 JW |
2485 | if (lrucare) |
2486 | unlock_page_lru(page, isolated); | |
7a81b88c | 2487 | } |
66e1707b | 2488 | |
7ae1e1d0 | 2489 | #ifdef CONFIG_MEMCG_KMEM |
dbf22eb6 VD |
2490 | int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2491 | unsigned long nr_pages) | |
7ae1e1d0 | 2492 | { |
3e32cb2e | 2493 | struct page_counter *counter; |
7ae1e1d0 | 2494 | int ret = 0; |
7ae1e1d0 | 2495 | |
3e32cb2e JW |
2496 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2497 | if (ret < 0) | |
7ae1e1d0 GC |
2498 | return ret; |
2499 | ||
3e32cb2e | 2500 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2501 | if (ret == -EINTR) { |
2502 | /* | |
00501b53 JW |
2503 | * try_charge() chose to bypass to root due to OOM kill or |
2504 | * fatal signal. Since our only options are to either fail | |
2505 | * the allocation or charge it to this cgroup, do it as a | |
2506 | * temporary condition. But we can't fail. From a kmem/slab | |
2507 | * perspective, the cache has already been selected, by | |
2508 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2509 | * our minds. |
2510 | * | |
2511 | * This condition will only trigger if the task entered | |
00501b53 JW |
2512 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2513 | * during try_charge() above. Tasks that were already dying | |
2514 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2515 | * directed to the root cgroup in memcontrol.h |
2516 | */ | |
3e32cb2e | 2517 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2518 | if (do_swap_account) |
3e32cb2e | 2519 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2520 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2521 | ret = 0; |
2522 | } else if (ret) | |
3e32cb2e | 2523 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2524 | |
2525 | return ret; | |
2526 | } | |
2527 | ||
dbf22eb6 | 2528 | void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages) |
7ae1e1d0 | 2529 | { |
3e32cb2e | 2530 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2531 | if (do_swap_account) |
3e32cb2e | 2532 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2533 | |
64f21993 | 2534 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7de37682 | 2535 | |
e8ea14cc | 2536 | css_put_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2537 | } |
2538 | ||
2633d7a0 GC |
2539 | /* |
2540 | * helper for acessing a memcg's index. It will be used as an index in the | |
2541 | * child cache array in kmem_cache, and also to derive its name. This function | |
2542 | * will return -1 when this is not a kmem-limited memcg. | |
2543 | */ | |
2544 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2545 | { | |
2546 | return memcg ? memcg->kmemcg_id : -1; | |
2547 | } | |
2548 | ||
f3bb3043 | 2549 | static int memcg_alloc_cache_id(void) |
55007d84 | 2550 | { |
f3bb3043 VD |
2551 | int id, size; |
2552 | int err; | |
2553 | ||
dbcf73e2 | 2554 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2555 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2556 | if (id < 0) | |
2557 | return id; | |
55007d84 | 2558 | |
dbcf73e2 | 2559 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2560 | return id; |
2561 | ||
2562 | /* | |
2563 | * There's no space for the new id in memcg_caches arrays, | |
2564 | * so we have to grow them. | |
2565 | */ | |
05257a1a | 2566 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2567 | |
2568 | size = 2 * (id + 1); | |
55007d84 GC |
2569 | if (size < MEMCG_CACHES_MIN_SIZE) |
2570 | size = MEMCG_CACHES_MIN_SIZE; | |
2571 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2572 | size = MEMCG_CACHES_MAX_SIZE; | |
2573 | ||
f3bb3043 | 2574 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2575 | if (!err) |
2576 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2577 | if (!err) |
2578 | memcg_nr_cache_ids = size; | |
2579 | ||
2580 | up_write(&memcg_cache_ids_sem); | |
2581 | ||
f3bb3043 | 2582 | if (err) { |
dbcf73e2 | 2583 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2584 | return err; |
2585 | } | |
2586 | return id; | |
2587 | } | |
2588 | ||
2589 | static void memcg_free_cache_id(int id) | |
2590 | { | |
dbcf73e2 | 2591 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2592 | } |
2593 | ||
d5b3cf71 | 2594 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2595 | struct mem_cgroup *memcg; |
2596 | struct kmem_cache *cachep; | |
2597 | struct work_struct work; | |
2598 | }; | |
2599 | ||
d5b3cf71 | 2600 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2601 | { |
d5b3cf71 VD |
2602 | struct memcg_kmem_cache_create_work *cw = |
2603 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2604 | struct mem_cgroup *memcg = cw->memcg; |
2605 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2606 | |
d5b3cf71 | 2607 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2608 | |
5722d094 | 2609 | css_put(&memcg->css); |
d7f25f8a GC |
2610 | kfree(cw); |
2611 | } | |
2612 | ||
2613 | /* | |
2614 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2615 | */ |
d5b3cf71 VD |
2616 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2617 | struct kmem_cache *cachep) | |
d7f25f8a | 2618 | { |
d5b3cf71 | 2619 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2620 | |
776ed0f0 | 2621 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2622 | if (!cw) |
d7f25f8a | 2623 | return; |
8135be5a VD |
2624 | |
2625 | css_get(&memcg->css); | |
d7f25f8a GC |
2626 | |
2627 | cw->memcg = memcg; | |
2628 | cw->cachep = cachep; | |
d5b3cf71 | 2629 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2630 | |
d7f25f8a GC |
2631 | schedule_work(&cw->work); |
2632 | } | |
2633 | ||
d5b3cf71 VD |
2634 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2635 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2636 | { |
2637 | /* | |
2638 | * We need to stop accounting when we kmalloc, because if the | |
2639 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2640 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2641 | * |
2642 | * However, it is better to enclose the whole function. Depending on | |
2643 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2644 | * trigger an allocation. This too, will make us recurse. Because at | |
2645 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2646 | * the safest choice is to do it like this, wrapping the whole function. | |
2647 | */ | |
6f185c29 | 2648 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2649 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2650 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2651 | } |
c67a8a68 | 2652 | |
d7f25f8a GC |
2653 | /* |
2654 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2655 | * We try to use the current memcg's version of the cache. | |
2656 | * | |
2657 | * If the cache does not exist yet, if we are the first user of it, | |
2658 | * we either create it immediately, if possible, or create it asynchronously | |
2659 | * in a workqueue. | |
2660 | * In the latter case, we will let the current allocation go through with | |
2661 | * the original cache. | |
2662 | * | |
2663 | * Can't be called in interrupt context or from kernel threads. | |
2664 | * This function needs to be called with rcu_read_lock() held. | |
2665 | */ | |
056b7cce | 2666 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2667 | { |
2668 | struct mem_cgroup *memcg; | |
959c8963 | 2669 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2670 | int kmemcg_id; |
d7f25f8a | 2671 | |
f7ce3190 | 2672 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2673 | |
9d100c5e | 2674 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2675 | return cachep; |
2676 | ||
8135be5a | 2677 | memcg = get_mem_cgroup_from_mm(current->mm); |
2a4db7eb VD |
2678 | kmemcg_id = ACCESS_ONCE(memcg->kmemcg_id); |
2679 | if (kmemcg_id < 0) | |
ca0dde97 | 2680 | goto out; |
d7f25f8a | 2681 | |
2a4db7eb | 2682 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2683 | if (likely(memcg_cachep)) |
2684 | return memcg_cachep; | |
ca0dde97 LZ |
2685 | |
2686 | /* | |
2687 | * If we are in a safe context (can wait, and not in interrupt | |
2688 | * context), we could be be predictable and return right away. | |
2689 | * This would guarantee that the allocation being performed | |
2690 | * already belongs in the new cache. | |
2691 | * | |
2692 | * However, there are some clashes that can arrive from locking. | |
2693 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2694 | * memcg_create_kmem_cache, this means no further allocation |
2695 | * could happen with the slab_mutex held. So it's better to | |
2696 | * defer everything. | |
ca0dde97 | 2697 | */ |
d5b3cf71 | 2698 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2699 | out: |
8135be5a | 2700 | css_put(&memcg->css); |
ca0dde97 | 2701 | return cachep; |
d7f25f8a | 2702 | } |
d7f25f8a | 2703 | |
8135be5a VD |
2704 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2705 | { | |
2706 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2707 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2708 | } |
2709 | ||
7ae1e1d0 GC |
2710 | /* |
2711 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2712 | * possible for the given order. But the page is not allocated yet, so we'll | |
2713 | * need a further commit step to do the final arrangements. | |
2714 | * | |
2715 | * It is possible for the task to switch cgroups in this mean time, so at | |
2716 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2717 | * the handle argument to return to the caller which cgroup we should commit | |
2718 | * against. We could also return the memcg directly and avoid the pointer | |
2719 | * passing, but a boolean return value gives better semantics considering | |
2720 | * the compiled-out case as well. | |
2721 | * | |
2722 | * Returning true means the allocation is possible. | |
2723 | */ | |
2724 | bool | |
2725 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2726 | { | |
2727 | struct mem_cgroup *memcg; | |
2728 | int ret; | |
2729 | ||
2730 | *_memcg = NULL; | |
6d42c232 | 2731 | |
df381975 | 2732 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 2733 | |
cf2b8fbf | 2734 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
2735 | css_put(&memcg->css); |
2736 | return true; | |
2737 | } | |
2738 | ||
3e32cb2e | 2739 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
2740 | if (!ret) |
2741 | *_memcg = memcg; | |
7ae1e1d0 GC |
2742 | |
2743 | css_put(&memcg->css); | |
2744 | return (ret == 0); | |
2745 | } | |
2746 | ||
2747 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
2748 | int order) | |
2749 | { | |
7ae1e1d0 GC |
2750 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
2751 | ||
2752 | /* The page allocation failed. Revert */ | |
2753 | if (!page) { | |
3e32cb2e | 2754 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
2755 | return; |
2756 | } | |
1306a85a | 2757 | page->mem_cgroup = memcg; |
7ae1e1d0 GC |
2758 | } |
2759 | ||
2760 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
2761 | { | |
1306a85a | 2762 | struct mem_cgroup *memcg = page->mem_cgroup; |
7ae1e1d0 | 2763 | |
7ae1e1d0 GC |
2764 | if (!memcg) |
2765 | return; | |
2766 | ||
309381fe | 2767 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2768 | |
3e32cb2e | 2769 | memcg_uncharge_kmem(memcg, 1 << order); |
1306a85a | 2770 | page->mem_cgroup = NULL; |
7ae1e1d0 | 2771 | } |
60d3fd32 VD |
2772 | |
2773 | struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr) | |
2774 | { | |
2775 | struct mem_cgroup *memcg = NULL; | |
2776 | struct kmem_cache *cachep; | |
2777 | struct page *page; | |
2778 | ||
2779 | page = virt_to_head_page(ptr); | |
2780 | if (PageSlab(page)) { | |
2781 | cachep = page->slab_cache; | |
2782 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2783 | memcg = cachep->memcg_params.memcg; |
60d3fd32 VD |
2784 | } else |
2785 | /* page allocated by alloc_kmem_pages */ | |
2786 | memcg = page->mem_cgroup; | |
2787 | ||
2788 | return memcg; | |
2789 | } | |
7ae1e1d0 GC |
2790 | #endif /* CONFIG_MEMCG_KMEM */ |
2791 | ||
ca3e0214 KH |
2792 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2793 | ||
ca3e0214 KH |
2794 | /* |
2795 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2796 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2797 | * charge/uncharge will be never happen and move_account() is done under | |
2798 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2799 | */ |
e94c8a9c | 2800 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2801 | { |
e94c8a9c | 2802 | int i; |
ca3e0214 | 2803 | |
3d37c4a9 KH |
2804 | if (mem_cgroup_disabled()) |
2805 | return; | |
b070e65c | 2806 | |
29833315 | 2807 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2808 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2809 | |
1306a85a | 2810 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2811 | HPAGE_PMD_NR); |
ca3e0214 | 2812 | } |
12d27107 | 2813 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2814 | |
f817ed48 | 2815 | /** |
de3638d9 | 2816 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2817 | * @page: the page |
7ec99d62 | 2818 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2819 | * @from: mem_cgroup which the page is moved from. |
2820 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2821 | * | |
2822 | * The caller must confirm following. | |
08e552c6 | 2823 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2824 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2825 | * |
2f3479b1 KH |
2826 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2827 | * from old cgroup. | |
f817ed48 | 2828 | */ |
7ec99d62 JW |
2829 | static int mem_cgroup_move_account(struct page *page, |
2830 | unsigned int nr_pages, | |
7ec99d62 | 2831 | struct mem_cgroup *from, |
2f3479b1 | 2832 | struct mem_cgroup *to) |
f817ed48 | 2833 | { |
de3638d9 JW |
2834 | unsigned long flags; |
2835 | int ret; | |
987eba66 | 2836 | |
f817ed48 | 2837 | VM_BUG_ON(from == to); |
309381fe | 2838 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
2839 | /* |
2840 | * The page is isolated from LRU. So, collapse function | |
2841 | * will not handle this page. But page splitting can happen. | |
2842 | * Do this check under compound_page_lock(). The caller should | |
2843 | * hold it. | |
2844 | */ | |
2845 | ret = -EBUSY; | |
7ec99d62 | 2846 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2847 | goto out; |
2848 | ||
0a31bc97 | 2849 | /* |
1306a85a | 2850 | * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup |
0a31bc97 JW |
2851 | * of its source page while we change it: page migration takes |
2852 | * both pages off the LRU, but page cache replacement doesn't. | |
2853 | */ | |
2854 | if (!trylock_page(page)) | |
2855 | goto out; | |
de3638d9 JW |
2856 | |
2857 | ret = -EINVAL; | |
1306a85a | 2858 | if (page->mem_cgroup != from) |
0a31bc97 | 2859 | goto out_unlock; |
de3638d9 | 2860 | |
354a4783 | 2861 | spin_lock_irqsave(&from->move_lock, flags); |
f817ed48 | 2862 | |
0a31bc97 | 2863 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
2864 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
2865 | nr_pages); | |
2866 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
2867 | nr_pages); | |
2868 | } | |
3ea67d06 | 2869 | |
59d1d256 JW |
2870 | if (PageWriteback(page)) { |
2871 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
2872 | nr_pages); | |
2873 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
2874 | nr_pages); | |
2875 | } | |
3ea67d06 | 2876 | |
0a31bc97 | 2877 | /* |
1306a85a | 2878 | * It is safe to change page->mem_cgroup here because the page |
0a31bc97 JW |
2879 | * is referenced, charged, and isolated - we can't race with |
2880 | * uncharging, charging, migration, or LRU putback. | |
2881 | */ | |
d69b042f | 2882 | |
854ffa8d | 2883 | /* caller should have done css_get */ |
1306a85a | 2884 | page->mem_cgroup = to; |
354a4783 JW |
2885 | spin_unlock_irqrestore(&from->move_lock, flags); |
2886 | ||
de3638d9 | 2887 | ret = 0; |
0a31bc97 JW |
2888 | |
2889 | local_irq_disable(); | |
2890 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 2891 | memcg_check_events(to, page); |
0a31bc97 | 2892 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 2893 | memcg_check_events(from, page); |
0a31bc97 JW |
2894 | local_irq_enable(); |
2895 | out_unlock: | |
2896 | unlock_page(page); | |
de3638d9 | 2897 | out: |
f817ed48 KH |
2898 | return ret; |
2899 | } | |
2900 | ||
c255a458 | 2901 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2902 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2903 | bool charge) | |
d13d1443 | 2904 | { |
0a31bc97 JW |
2905 | int val = (charge) ? 1 : -1; |
2906 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2907 | } |
02491447 DN |
2908 | |
2909 | /** | |
2910 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2911 | * @entry: swap entry to be moved | |
2912 | * @from: mem_cgroup which the entry is moved from | |
2913 | * @to: mem_cgroup which the entry is moved to | |
2914 | * | |
2915 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2916 | * as the mem_cgroup's id of @from. | |
2917 | * | |
2918 | * Returns 0 on success, -EINVAL on failure. | |
2919 | * | |
3e32cb2e | 2920 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2921 | * both res and memsw, and called css_get(). |
2922 | */ | |
2923 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2924 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2925 | { |
2926 | unsigned short old_id, new_id; | |
2927 | ||
34c00c31 LZ |
2928 | old_id = mem_cgroup_id(from); |
2929 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2930 | |
2931 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2932 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2933 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2934 | return 0; |
2935 | } | |
2936 | return -EINVAL; | |
2937 | } | |
2938 | #else | |
2939 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2940 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2941 | { |
2942 | return -EINVAL; | |
2943 | } | |
8c7c6e34 | 2944 | #endif |
d13d1443 | 2945 | |
3e32cb2e | 2946 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2947 | |
d38d2a75 | 2948 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2949 | unsigned long limit) |
628f4235 | 2950 | { |
3e32cb2e JW |
2951 | unsigned long curusage; |
2952 | unsigned long oldusage; | |
2953 | bool enlarge = false; | |
81d39c20 | 2954 | int retry_count; |
3e32cb2e | 2955 | int ret; |
81d39c20 KH |
2956 | |
2957 | /* | |
2958 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2959 | * is depends on callers. We set our retry-count to be function | |
2960 | * of # of children which we should visit in this loop. | |
2961 | */ | |
3e32cb2e JW |
2962 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2963 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2964 | |
3e32cb2e | 2965 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2966 | |
3e32cb2e | 2967 | do { |
628f4235 KH |
2968 | if (signal_pending(current)) { |
2969 | ret = -EINTR; | |
2970 | break; | |
2971 | } | |
3e32cb2e JW |
2972 | |
2973 | mutex_lock(&memcg_limit_mutex); | |
2974 | if (limit > memcg->memsw.limit) { | |
2975 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2976 | ret = -EINVAL; |
628f4235 KH |
2977 | break; |
2978 | } | |
3e32cb2e JW |
2979 | if (limit > memcg->memory.limit) |
2980 | enlarge = true; | |
2981 | ret = page_counter_limit(&memcg->memory, limit); | |
2982 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2983 | |
2984 | if (!ret) | |
2985 | break; | |
2986 | ||
b70a2a21 JW |
2987 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2988 | ||
3e32cb2e | 2989 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2990 | /* Usage is reduced ? */ |
f894ffa8 | 2991 | if (curusage >= oldusage) |
81d39c20 KH |
2992 | retry_count--; |
2993 | else | |
2994 | oldusage = curusage; | |
3e32cb2e JW |
2995 | } while (retry_count); |
2996 | ||
3c11ecf4 KH |
2997 | if (!ret && enlarge) |
2998 | memcg_oom_recover(memcg); | |
14797e23 | 2999 | |
8c7c6e34 KH |
3000 | return ret; |
3001 | } | |
3002 | ||
338c8431 | 3003 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3004 | unsigned long limit) |
8c7c6e34 | 3005 | { |
3e32cb2e JW |
3006 | unsigned long curusage; |
3007 | unsigned long oldusage; | |
3008 | bool enlarge = false; | |
81d39c20 | 3009 | int retry_count; |
3e32cb2e | 3010 | int ret; |
8c7c6e34 | 3011 | |
81d39c20 | 3012 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
3013 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3014 | mem_cgroup_count_children(memcg); | |
3015 | ||
3016 | oldusage = page_counter_read(&memcg->memsw); | |
3017 | ||
3018 | do { | |
8c7c6e34 KH |
3019 | if (signal_pending(current)) { |
3020 | ret = -EINTR; | |
3021 | break; | |
3022 | } | |
3e32cb2e JW |
3023 | |
3024 | mutex_lock(&memcg_limit_mutex); | |
3025 | if (limit < memcg->memory.limit) { | |
3026 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3027 | ret = -EINVAL; |
8c7c6e34 KH |
3028 | break; |
3029 | } | |
3e32cb2e JW |
3030 | if (limit > memcg->memsw.limit) |
3031 | enlarge = true; | |
3032 | ret = page_counter_limit(&memcg->memsw, limit); | |
3033 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3034 | |
3035 | if (!ret) | |
3036 | break; | |
3037 | ||
b70a2a21 JW |
3038 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3039 | ||
3e32cb2e | 3040 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3041 | /* Usage is reduced ? */ |
8c7c6e34 | 3042 | if (curusage >= oldusage) |
628f4235 | 3043 | retry_count--; |
81d39c20 KH |
3044 | else |
3045 | oldusage = curusage; | |
3e32cb2e JW |
3046 | } while (retry_count); |
3047 | ||
3c11ecf4 KH |
3048 | if (!ret && enlarge) |
3049 | memcg_oom_recover(memcg); | |
3e32cb2e | 3050 | |
628f4235 KH |
3051 | return ret; |
3052 | } | |
3053 | ||
0608f43d AM |
3054 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3055 | gfp_t gfp_mask, | |
3056 | unsigned long *total_scanned) | |
3057 | { | |
3058 | unsigned long nr_reclaimed = 0; | |
3059 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3060 | unsigned long reclaimed; | |
3061 | int loop = 0; | |
3062 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3063 | unsigned long excess; |
0608f43d AM |
3064 | unsigned long nr_scanned; |
3065 | ||
3066 | if (order > 0) | |
3067 | return 0; | |
3068 | ||
3069 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3070 | /* | |
3071 | * This loop can run a while, specially if mem_cgroup's continuously | |
3072 | * keep exceeding their soft limit and putting the system under | |
3073 | * pressure | |
3074 | */ | |
3075 | do { | |
3076 | if (next_mz) | |
3077 | mz = next_mz; | |
3078 | else | |
3079 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3080 | if (!mz) | |
3081 | break; | |
3082 | ||
3083 | nr_scanned = 0; | |
3084 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3085 | gfp_mask, &nr_scanned); | |
3086 | nr_reclaimed += reclaimed; | |
3087 | *total_scanned += nr_scanned; | |
0a31bc97 | 3088 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3089 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3090 | |
3091 | /* | |
3092 | * If we failed to reclaim anything from this memory cgroup | |
3093 | * it is time to move on to the next cgroup | |
3094 | */ | |
3095 | next_mz = NULL; | |
bc2f2e7f VD |
3096 | if (!reclaimed) |
3097 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3098 | ||
3e32cb2e | 3099 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3100 | /* |
3101 | * One school of thought says that we should not add | |
3102 | * back the node to the tree if reclaim returns 0. | |
3103 | * But our reclaim could return 0, simply because due | |
3104 | * to priority we are exposing a smaller subset of | |
3105 | * memory to reclaim from. Consider this as a longer | |
3106 | * term TODO. | |
3107 | */ | |
3108 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3109 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3110 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3111 | css_put(&mz->memcg->css); |
3112 | loop++; | |
3113 | /* | |
3114 | * Could not reclaim anything and there are no more | |
3115 | * mem cgroups to try or we seem to be looping without | |
3116 | * reclaiming anything. | |
3117 | */ | |
3118 | if (!nr_reclaimed && | |
3119 | (next_mz == NULL || | |
3120 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3121 | break; | |
3122 | } while (!nr_reclaimed); | |
3123 | if (next_mz) | |
3124 | css_put(&next_mz->memcg->css); | |
3125 | return nr_reclaimed; | |
3126 | } | |
3127 | ||
ea280e7b TH |
3128 | /* |
3129 | * Test whether @memcg has children, dead or alive. Note that this | |
3130 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3131 | * returns %true if there are child csses according to the cgroup | |
3132 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3133 | */ | |
b5f99b53 GC |
3134 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3135 | { | |
ea280e7b TH |
3136 | bool ret; |
3137 | ||
696ac172 | 3138 | /* |
ea280e7b TH |
3139 | * The lock does not prevent addition or deletion of children, but |
3140 | * it prevents a new child from being initialized based on this | |
3141 | * parent in css_online(), so it's enough to decide whether | |
3142 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3143 | */ |
ea280e7b TH |
3144 | lockdep_assert_held(&memcg_create_mutex); |
3145 | ||
3146 | rcu_read_lock(); | |
3147 | ret = css_next_child(NULL, &memcg->css); | |
3148 | rcu_read_unlock(); | |
3149 | return ret; | |
b5f99b53 GC |
3150 | } |
3151 | ||
c26251f9 MH |
3152 | /* |
3153 | * Reclaims as many pages from the given memcg as possible and moves | |
3154 | * the rest to the parent. | |
3155 | * | |
3156 | * Caller is responsible for holding css reference for memcg. | |
3157 | */ | |
3158 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3159 | { | |
3160 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3161 | |
c1e862c1 KH |
3162 | /* we call try-to-free pages for make this cgroup empty */ |
3163 | lru_add_drain_all(); | |
f817ed48 | 3164 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3165 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3166 | int progress; |
c1e862c1 | 3167 | |
c26251f9 MH |
3168 | if (signal_pending(current)) |
3169 | return -EINTR; | |
3170 | ||
b70a2a21 JW |
3171 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3172 | GFP_KERNEL, true); | |
c1e862c1 | 3173 | if (!progress) { |
f817ed48 | 3174 | nr_retries--; |
c1e862c1 | 3175 | /* maybe some writeback is necessary */ |
8aa7e847 | 3176 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3177 | } |
f817ed48 KH |
3178 | |
3179 | } | |
ab5196c2 MH |
3180 | |
3181 | return 0; | |
cc847582 KH |
3182 | } |
3183 | ||
6770c64e TH |
3184 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3185 | char *buf, size_t nbytes, | |
3186 | loff_t off) | |
c1e862c1 | 3187 | { |
6770c64e | 3188 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3189 | |
d8423011 MH |
3190 | if (mem_cgroup_is_root(memcg)) |
3191 | return -EINVAL; | |
6770c64e | 3192 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3193 | } |
3194 | ||
182446d0 TH |
3195 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3196 | struct cftype *cft) | |
18f59ea7 | 3197 | { |
182446d0 | 3198 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3199 | } |
3200 | ||
182446d0 TH |
3201 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3202 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3203 | { |
3204 | int retval = 0; | |
182446d0 | 3205 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3206 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3207 | |
0999821b | 3208 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3209 | |
3210 | if (memcg->use_hierarchy == val) | |
3211 | goto out; | |
3212 | ||
18f59ea7 | 3213 | /* |
af901ca1 | 3214 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3215 | * in the child subtrees. If it is unset, then the change can |
3216 | * occur, provided the current cgroup has no children. | |
3217 | * | |
3218 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3219 | * set if there are no children. | |
3220 | */ | |
c0ff4b85 | 3221 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3222 | (val == 1 || val == 0)) { |
ea280e7b | 3223 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3224 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3225 | else |
3226 | retval = -EBUSY; | |
3227 | } else | |
3228 | retval = -EINVAL; | |
567fb435 GC |
3229 | |
3230 | out: | |
0999821b | 3231 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3232 | |
3233 | return retval; | |
3234 | } | |
3235 | ||
3e32cb2e JW |
3236 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3237 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3238 | { |
3239 | struct mem_cgroup *iter; | |
3240 | long val = 0; | |
3241 | ||
3242 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3243 | for_each_mem_cgroup_tree(iter, memcg) | |
3244 | val += mem_cgroup_read_stat(iter, idx); | |
3245 | ||
3246 | if (val < 0) /* race ? */ | |
3247 | val = 0; | |
3248 | return val; | |
3249 | } | |
3250 | ||
3251 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3252 | { | |
3253 | u64 val; | |
3254 | ||
3e32cb2e JW |
3255 | if (mem_cgroup_is_root(memcg)) { |
3256 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3257 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3258 | if (swap) | |
3259 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3260 | } else { | |
ce00a967 | 3261 | if (!swap) |
3e32cb2e | 3262 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3263 | else |
3e32cb2e | 3264 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3265 | } |
ce00a967 JW |
3266 | return val << PAGE_SHIFT; |
3267 | } | |
3268 | ||
3e32cb2e JW |
3269 | enum { |
3270 | RES_USAGE, | |
3271 | RES_LIMIT, | |
3272 | RES_MAX_USAGE, | |
3273 | RES_FAILCNT, | |
3274 | RES_SOFT_LIMIT, | |
3275 | }; | |
ce00a967 | 3276 | |
791badbd | 3277 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3278 | struct cftype *cft) |
8cdea7c0 | 3279 | { |
182446d0 | 3280 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3281 | struct page_counter *counter; |
af36f906 | 3282 | |
3e32cb2e | 3283 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3284 | case _MEM: |
3e32cb2e JW |
3285 | counter = &memcg->memory; |
3286 | break; | |
8c7c6e34 | 3287 | case _MEMSWAP: |
3e32cb2e JW |
3288 | counter = &memcg->memsw; |
3289 | break; | |
510fc4e1 | 3290 | case _KMEM: |
3e32cb2e | 3291 | counter = &memcg->kmem; |
510fc4e1 | 3292 | break; |
8c7c6e34 KH |
3293 | default: |
3294 | BUG(); | |
8c7c6e34 | 3295 | } |
3e32cb2e JW |
3296 | |
3297 | switch (MEMFILE_ATTR(cft->private)) { | |
3298 | case RES_USAGE: | |
3299 | if (counter == &memcg->memory) | |
3300 | return mem_cgroup_usage(memcg, false); | |
3301 | if (counter == &memcg->memsw) | |
3302 | return mem_cgroup_usage(memcg, true); | |
3303 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3304 | case RES_LIMIT: | |
3305 | return (u64)counter->limit * PAGE_SIZE; | |
3306 | case RES_MAX_USAGE: | |
3307 | return (u64)counter->watermark * PAGE_SIZE; | |
3308 | case RES_FAILCNT: | |
3309 | return counter->failcnt; | |
3310 | case RES_SOFT_LIMIT: | |
3311 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3312 | default: | |
3313 | BUG(); | |
3314 | } | |
8cdea7c0 | 3315 | } |
510fc4e1 | 3316 | |
510fc4e1 | 3317 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3318 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3319 | unsigned long nr_pages) | |
d6441637 VD |
3320 | { |
3321 | int err = 0; | |
3322 | int memcg_id; | |
3323 | ||
2a4db7eb VD |
3324 | BUG_ON(memcg->kmemcg_id >= 0); |
3325 | BUG_ON(memcg->kmem_acct_active); | |
d6441637 | 3326 | |
510fc4e1 GC |
3327 | /* |
3328 | * For simplicity, we won't allow this to be disabled. It also can't | |
3329 | * be changed if the cgroup has children already, or if tasks had | |
3330 | * already joined. | |
3331 | * | |
3332 | * If tasks join before we set the limit, a person looking at | |
3333 | * kmem.usage_in_bytes will have no way to determine when it took | |
3334 | * place, which makes the value quite meaningless. | |
3335 | * | |
3336 | * After it first became limited, changes in the value of the limit are | |
3337 | * of course permitted. | |
510fc4e1 | 3338 | */ |
0999821b | 3339 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3340 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3341 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3342 | err = -EBUSY; |
3343 | mutex_unlock(&memcg_create_mutex); | |
3344 | if (err) | |
3345 | goto out; | |
510fc4e1 | 3346 | |
f3bb3043 | 3347 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3348 | if (memcg_id < 0) { |
3349 | err = memcg_id; | |
3350 | goto out; | |
3351 | } | |
3352 | ||
d6441637 | 3353 | /* |
900a38f0 VD |
3354 | * We couldn't have accounted to this cgroup, because it hasn't got |
3355 | * activated yet, so this should succeed. | |
d6441637 | 3356 | */ |
3e32cb2e | 3357 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3358 | VM_BUG_ON(err); |
3359 | ||
3360 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3361 | /* | |
900a38f0 VD |
3362 | * A memory cgroup is considered kmem-active as soon as it gets |
3363 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
3364 | * guarantee no one starts accounting before all call sites are |
3365 | * patched. | |
3366 | */ | |
900a38f0 | 3367 | memcg->kmemcg_id = memcg_id; |
2a4db7eb | 3368 | memcg->kmem_acct_active = true; |
510fc4e1 | 3369 | out: |
d6441637 | 3370 | return err; |
d6441637 VD |
3371 | } |
3372 | ||
d6441637 | 3373 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3374 | unsigned long limit) |
d6441637 VD |
3375 | { |
3376 | int ret; | |
3377 | ||
3e32cb2e | 3378 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3379 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3380 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3381 | else |
3e32cb2e JW |
3382 | ret = page_counter_limit(&memcg->kmem, limit); |
3383 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3384 | return ret; |
3385 | } | |
3386 | ||
55007d84 | 3387 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3388 | { |
55007d84 | 3389 | int ret = 0; |
510fc4e1 | 3390 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3391 | |
d6441637 VD |
3392 | if (!parent) |
3393 | return 0; | |
55007d84 | 3394 | |
8c0145b6 | 3395 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3396 | /* |
d6441637 VD |
3397 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3398 | * after this point, because it has at least one child already. | |
55007d84 | 3399 | */ |
d6441637 | 3400 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3401 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3402 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3403 | return ret; |
510fc4e1 | 3404 | } |
d6441637 VD |
3405 | #else |
3406 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3407 | unsigned long limit) |
d6441637 VD |
3408 | { |
3409 | return -EINVAL; | |
3410 | } | |
6d043990 | 3411 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3412 | |
628f4235 KH |
3413 | /* |
3414 | * The user of this function is... | |
3415 | * RES_LIMIT. | |
3416 | */ | |
451af504 TH |
3417 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3418 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3419 | { |
451af504 | 3420 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3421 | unsigned long nr_pages; |
628f4235 KH |
3422 | int ret; |
3423 | ||
451af504 | 3424 | buf = strstrip(buf); |
650c5e56 | 3425 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3426 | if (ret) |
3427 | return ret; | |
af36f906 | 3428 | |
3e32cb2e | 3429 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3430 | case RES_LIMIT: |
4b3bde4c BS |
3431 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3432 | ret = -EINVAL; | |
3433 | break; | |
3434 | } | |
3e32cb2e JW |
3435 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3436 | case _MEM: | |
3437 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3438 | break; |
3e32cb2e JW |
3439 | case _MEMSWAP: |
3440 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3441 | break; |
3e32cb2e JW |
3442 | case _KMEM: |
3443 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3444 | break; | |
3445 | } | |
296c81d8 | 3446 | break; |
3e32cb2e JW |
3447 | case RES_SOFT_LIMIT: |
3448 | memcg->soft_limit = nr_pages; | |
3449 | ret = 0; | |
628f4235 KH |
3450 | break; |
3451 | } | |
451af504 | 3452 | return ret ?: nbytes; |
8cdea7c0 BS |
3453 | } |
3454 | ||
6770c64e TH |
3455 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3456 | size_t nbytes, loff_t off) | |
c84872e1 | 3457 | { |
6770c64e | 3458 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3459 | struct page_counter *counter; |
c84872e1 | 3460 | |
3e32cb2e JW |
3461 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3462 | case _MEM: | |
3463 | counter = &memcg->memory; | |
3464 | break; | |
3465 | case _MEMSWAP: | |
3466 | counter = &memcg->memsw; | |
3467 | break; | |
3468 | case _KMEM: | |
3469 | counter = &memcg->kmem; | |
3470 | break; | |
3471 | default: | |
3472 | BUG(); | |
3473 | } | |
af36f906 | 3474 | |
3e32cb2e | 3475 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3476 | case RES_MAX_USAGE: |
3e32cb2e | 3477 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3478 | break; |
3479 | case RES_FAILCNT: | |
3e32cb2e | 3480 | counter->failcnt = 0; |
29f2a4da | 3481 | break; |
3e32cb2e JW |
3482 | default: |
3483 | BUG(); | |
29f2a4da | 3484 | } |
f64c3f54 | 3485 | |
6770c64e | 3486 | return nbytes; |
c84872e1 PE |
3487 | } |
3488 | ||
182446d0 | 3489 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3490 | struct cftype *cft) |
3491 | { | |
182446d0 | 3492 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3493 | } |
3494 | ||
02491447 | 3495 | #ifdef CONFIG_MMU |
182446d0 | 3496 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3497 | struct cftype *cft, u64 val) |
3498 | { | |
182446d0 | 3499 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3500 | |
1dfab5ab | 3501 | if (val & ~MOVE_MASK) |
7dc74be0 | 3502 | return -EINVAL; |
ee5e8472 | 3503 | |
7dc74be0 | 3504 | /* |
ee5e8472 GC |
3505 | * No kind of locking is needed in here, because ->can_attach() will |
3506 | * check this value once in the beginning of the process, and then carry | |
3507 | * on with stale data. This means that changes to this value will only | |
3508 | * affect task migrations starting after the change. | |
7dc74be0 | 3509 | */ |
c0ff4b85 | 3510 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3511 | return 0; |
3512 | } | |
02491447 | 3513 | #else |
182446d0 | 3514 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3515 | struct cftype *cft, u64 val) |
3516 | { | |
3517 | return -ENOSYS; | |
3518 | } | |
3519 | #endif | |
7dc74be0 | 3520 | |
406eb0c9 | 3521 | #ifdef CONFIG_NUMA |
2da8ca82 | 3522 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3523 | { |
25485de6 GT |
3524 | struct numa_stat { |
3525 | const char *name; | |
3526 | unsigned int lru_mask; | |
3527 | }; | |
3528 | ||
3529 | static const struct numa_stat stats[] = { | |
3530 | { "total", LRU_ALL }, | |
3531 | { "file", LRU_ALL_FILE }, | |
3532 | { "anon", LRU_ALL_ANON }, | |
3533 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3534 | }; | |
3535 | const struct numa_stat *stat; | |
406eb0c9 | 3536 | int nid; |
25485de6 | 3537 | unsigned long nr; |
2da8ca82 | 3538 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3539 | |
25485de6 GT |
3540 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3541 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3542 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3543 | for_each_node_state(nid, N_MEMORY) { | |
3544 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3545 | stat->lru_mask); | |
3546 | seq_printf(m, " N%d=%lu", nid, nr); | |
3547 | } | |
3548 | seq_putc(m, '\n'); | |
406eb0c9 | 3549 | } |
406eb0c9 | 3550 | |
071aee13 YH |
3551 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3552 | struct mem_cgroup *iter; | |
3553 | ||
3554 | nr = 0; | |
3555 | for_each_mem_cgroup_tree(iter, memcg) | |
3556 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3557 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3558 | for_each_node_state(nid, N_MEMORY) { | |
3559 | nr = 0; | |
3560 | for_each_mem_cgroup_tree(iter, memcg) | |
3561 | nr += mem_cgroup_node_nr_lru_pages( | |
3562 | iter, nid, stat->lru_mask); | |
3563 | seq_printf(m, " N%d=%lu", nid, nr); | |
3564 | } | |
3565 | seq_putc(m, '\n'); | |
406eb0c9 | 3566 | } |
406eb0c9 | 3567 | |
406eb0c9 YH |
3568 | return 0; |
3569 | } | |
3570 | #endif /* CONFIG_NUMA */ | |
3571 | ||
2da8ca82 | 3572 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3573 | { |
2da8ca82 | 3574 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3575 | unsigned long memory, memsw; |
af7c4b0e JW |
3576 | struct mem_cgroup *mi; |
3577 | unsigned int i; | |
406eb0c9 | 3578 | |
0ca44b14 GT |
3579 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3580 | MEM_CGROUP_STAT_NSTATS); | |
3581 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3582 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3583 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3584 | ||
af7c4b0e | 3585 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3586 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3587 | continue; |
af7c4b0e JW |
3588 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3589 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3590 | } |
7b854121 | 3591 | |
af7c4b0e JW |
3592 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3593 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3594 | mem_cgroup_read_events(memcg, i)); | |
3595 | ||
3596 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3597 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3598 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3599 | ||
14067bb3 | 3600 | /* Hierarchical information */ |
3e32cb2e JW |
3601 | memory = memsw = PAGE_COUNTER_MAX; |
3602 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3603 | memory = min(memory, mi->memory.limit); | |
3604 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3605 | } |
3e32cb2e JW |
3606 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3607 | (u64)memory * PAGE_SIZE); | |
3608 | if (do_swap_account) | |
3609 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3610 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3611 | |
af7c4b0e JW |
3612 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3613 | long long val = 0; | |
3614 | ||
bff6bb83 | 3615 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3616 | continue; |
af7c4b0e JW |
3617 | for_each_mem_cgroup_tree(mi, memcg) |
3618 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3619 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3620 | } | |
3621 | ||
3622 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3623 | unsigned long long val = 0; | |
3624 | ||
3625 | for_each_mem_cgroup_tree(mi, memcg) | |
3626 | val += mem_cgroup_read_events(mi, i); | |
3627 | seq_printf(m, "total_%s %llu\n", | |
3628 | mem_cgroup_events_names[i], val); | |
3629 | } | |
3630 | ||
3631 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3632 | unsigned long long val = 0; | |
3633 | ||
3634 | for_each_mem_cgroup_tree(mi, memcg) | |
3635 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3636 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3637 | } |
14067bb3 | 3638 | |
7f016ee8 | 3639 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3640 | { |
3641 | int nid, zid; | |
3642 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3643 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3644 | unsigned long recent_rotated[2] = {0, 0}; |
3645 | unsigned long recent_scanned[2] = {0, 0}; | |
3646 | ||
3647 | for_each_online_node(nid) | |
3648 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3649 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3650 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3651 | |
89abfab1 HD |
3652 | recent_rotated[0] += rstat->recent_rotated[0]; |
3653 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3654 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3655 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3656 | } |
78ccf5b5 JW |
3657 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3658 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3659 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3660 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3661 | } |
3662 | #endif | |
3663 | ||
d2ceb9b7 KH |
3664 | return 0; |
3665 | } | |
3666 | ||
182446d0 TH |
3667 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3668 | struct cftype *cft) | |
a7885eb8 | 3669 | { |
182446d0 | 3670 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3671 | |
1f4c025b | 3672 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3673 | } |
3674 | ||
182446d0 TH |
3675 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3676 | struct cftype *cft, u64 val) | |
a7885eb8 | 3677 | { |
182446d0 | 3678 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3679 | |
3dae7fec | 3680 | if (val > 100) |
a7885eb8 KM |
3681 | return -EINVAL; |
3682 | ||
14208b0e | 3683 | if (css->parent) |
3dae7fec JW |
3684 | memcg->swappiness = val; |
3685 | else | |
3686 | vm_swappiness = val; | |
068b38c1 | 3687 | |
a7885eb8 KM |
3688 | return 0; |
3689 | } | |
3690 | ||
2e72b634 KS |
3691 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3692 | { | |
3693 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3694 | unsigned long usage; |
2e72b634 KS |
3695 | int i; |
3696 | ||
3697 | rcu_read_lock(); | |
3698 | if (!swap) | |
2c488db2 | 3699 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3700 | else |
2c488db2 | 3701 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3702 | |
3703 | if (!t) | |
3704 | goto unlock; | |
3705 | ||
ce00a967 | 3706 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3707 | |
3708 | /* | |
748dad36 | 3709 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3710 | * If it's not true, a threshold was crossed after last |
3711 | * call of __mem_cgroup_threshold(). | |
3712 | */ | |
5407a562 | 3713 | i = t->current_threshold; |
2e72b634 KS |
3714 | |
3715 | /* | |
3716 | * Iterate backward over array of thresholds starting from | |
3717 | * current_threshold and check if a threshold is crossed. | |
3718 | * If none of thresholds below usage is crossed, we read | |
3719 | * only one element of the array here. | |
3720 | */ | |
3721 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3722 | eventfd_signal(t->entries[i].eventfd, 1); | |
3723 | ||
3724 | /* i = current_threshold + 1 */ | |
3725 | i++; | |
3726 | ||
3727 | /* | |
3728 | * Iterate forward over array of thresholds starting from | |
3729 | * current_threshold+1 and check if a threshold is crossed. | |
3730 | * If none of thresholds above usage is crossed, we read | |
3731 | * only one element of the array here. | |
3732 | */ | |
3733 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3734 | eventfd_signal(t->entries[i].eventfd, 1); | |
3735 | ||
3736 | /* Update current_threshold */ | |
5407a562 | 3737 | t->current_threshold = i - 1; |
2e72b634 KS |
3738 | unlock: |
3739 | rcu_read_unlock(); | |
3740 | } | |
3741 | ||
3742 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3743 | { | |
ad4ca5f4 KS |
3744 | while (memcg) { |
3745 | __mem_cgroup_threshold(memcg, false); | |
3746 | if (do_swap_account) | |
3747 | __mem_cgroup_threshold(memcg, true); | |
3748 | ||
3749 | memcg = parent_mem_cgroup(memcg); | |
3750 | } | |
2e72b634 KS |
3751 | } |
3752 | ||
3753 | static int compare_thresholds(const void *a, const void *b) | |
3754 | { | |
3755 | const struct mem_cgroup_threshold *_a = a; | |
3756 | const struct mem_cgroup_threshold *_b = b; | |
3757 | ||
2bff24a3 GT |
3758 | if (_a->threshold > _b->threshold) |
3759 | return 1; | |
3760 | ||
3761 | if (_a->threshold < _b->threshold) | |
3762 | return -1; | |
3763 | ||
3764 | return 0; | |
2e72b634 KS |
3765 | } |
3766 | ||
c0ff4b85 | 3767 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3768 | { |
3769 | struct mem_cgroup_eventfd_list *ev; | |
3770 | ||
2bcf2e92 MH |
3771 | spin_lock(&memcg_oom_lock); |
3772 | ||
c0ff4b85 | 3773 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3774 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3775 | |
3776 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3777 | return 0; |
3778 | } | |
3779 | ||
c0ff4b85 | 3780 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3781 | { |
7d74b06f KH |
3782 | struct mem_cgroup *iter; |
3783 | ||
c0ff4b85 | 3784 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3785 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3786 | } |
3787 | ||
59b6f873 | 3788 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3789 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3790 | { |
2c488db2 KS |
3791 | struct mem_cgroup_thresholds *thresholds; |
3792 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3793 | unsigned long threshold; |
3794 | unsigned long usage; | |
2c488db2 | 3795 | int i, size, ret; |
2e72b634 | 3796 | |
650c5e56 | 3797 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3798 | if (ret) |
3799 | return ret; | |
3800 | ||
3801 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3802 | |
05b84301 | 3803 | if (type == _MEM) { |
2c488db2 | 3804 | thresholds = &memcg->thresholds; |
ce00a967 | 3805 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3806 | } else if (type == _MEMSWAP) { |
2c488db2 | 3807 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3808 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3809 | } else |
2e72b634 KS |
3810 | BUG(); |
3811 | ||
2e72b634 | 3812 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3813 | if (thresholds->primary) |
2e72b634 KS |
3814 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3815 | ||
2c488db2 | 3816 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3817 | |
3818 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3819 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3820 | GFP_KERNEL); |
2c488db2 | 3821 | if (!new) { |
2e72b634 KS |
3822 | ret = -ENOMEM; |
3823 | goto unlock; | |
3824 | } | |
2c488db2 | 3825 | new->size = size; |
2e72b634 KS |
3826 | |
3827 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3828 | if (thresholds->primary) { |
3829 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3830 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3831 | } |
3832 | ||
2e72b634 | 3833 | /* Add new threshold */ |
2c488db2 KS |
3834 | new->entries[size - 1].eventfd = eventfd; |
3835 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3836 | |
3837 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3838 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3839 | compare_thresholds, NULL); |
3840 | ||
3841 | /* Find current threshold */ | |
2c488db2 | 3842 | new->current_threshold = -1; |
2e72b634 | 3843 | for (i = 0; i < size; i++) { |
748dad36 | 3844 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3845 | /* |
2c488db2 KS |
3846 | * new->current_threshold will not be used until |
3847 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3848 | * it here. |
3849 | */ | |
2c488db2 | 3850 | ++new->current_threshold; |
748dad36 SZ |
3851 | } else |
3852 | break; | |
2e72b634 KS |
3853 | } |
3854 | ||
2c488db2 KS |
3855 | /* Free old spare buffer and save old primary buffer as spare */ |
3856 | kfree(thresholds->spare); | |
3857 | thresholds->spare = thresholds->primary; | |
3858 | ||
3859 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3860 | |
907860ed | 3861 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3862 | synchronize_rcu(); |
3863 | ||
2e72b634 KS |
3864 | unlock: |
3865 | mutex_unlock(&memcg->thresholds_lock); | |
3866 | ||
3867 | return ret; | |
3868 | } | |
3869 | ||
59b6f873 | 3870 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3871 | struct eventfd_ctx *eventfd, const char *args) |
3872 | { | |
59b6f873 | 3873 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3874 | } |
3875 | ||
59b6f873 | 3876 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3877 | struct eventfd_ctx *eventfd, const char *args) |
3878 | { | |
59b6f873 | 3879 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3880 | } |
3881 | ||
59b6f873 | 3882 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3883 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3884 | { |
2c488db2 KS |
3885 | struct mem_cgroup_thresholds *thresholds; |
3886 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3887 | unsigned long usage; |
2c488db2 | 3888 | int i, j, size; |
2e72b634 KS |
3889 | |
3890 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3891 | |
3892 | if (type == _MEM) { | |
2c488db2 | 3893 | thresholds = &memcg->thresholds; |
ce00a967 | 3894 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3895 | } else if (type == _MEMSWAP) { |
2c488db2 | 3896 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3897 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3898 | } else |
2e72b634 KS |
3899 | BUG(); |
3900 | ||
371528ca AV |
3901 | if (!thresholds->primary) |
3902 | goto unlock; | |
3903 | ||
2e72b634 KS |
3904 | /* Check if a threshold crossed before removing */ |
3905 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3906 | ||
3907 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3908 | size = 0; |
3909 | for (i = 0; i < thresholds->primary->size; i++) { | |
3910 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3911 | size++; |
3912 | } | |
3913 | ||
2c488db2 | 3914 | new = thresholds->spare; |
907860ed | 3915 | |
2e72b634 KS |
3916 | /* Set thresholds array to NULL if we don't have thresholds */ |
3917 | if (!size) { | |
2c488db2 KS |
3918 | kfree(new); |
3919 | new = NULL; | |
907860ed | 3920 | goto swap_buffers; |
2e72b634 KS |
3921 | } |
3922 | ||
2c488db2 | 3923 | new->size = size; |
2e72b634 KS |
3924 | |
3925 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3926 | new->current_threshold = -1; |
3927 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3928 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3929 | continue; |
3930 | ||
2c488db2 | 3931 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3932 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3933 | /* |
2c488db2 | 3934 | * new->current_threshold will not be used |
2e72b634 KS |
3935 | * until rcu_assign_pointer(), so it's safe to increment |
3936 | * it here. | |
3937 | */ | |
2c488db2 | 3938 | ++new->current_threshold; |
2e72b634 KS |
3939 | } |
3940 | j++; | |
3941 | } | |
3942 | ||
907860ed | 3943 | swap_buffers: |
2c488db2 KS |
3944 | /* Swap primary and spare array */ |
3945 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
3946 | /* If all events are unregistered, free the spare array */ |
3947 | if (!new) { | |
3948 | kfree(thresholds->spare); | |
3949 | thresholds->spare = NULL; | |
3950 | } | |
3951 | ||
2c488db2 | 3952 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3953 | |
907860ed | 3954 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3955 | synchronize_rcu(); |
371528ca | 3956 | unlock: |
2e72b634 | 3957 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3958 | } |
c1e862c1 | 3959 | |
59b6f873 | 3960 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3961 | struct eventfd_ctx *eventfd) |
3962 | { | |
59b6f873 | 3963 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3964 | } |
3965 | ||
59b6f873 | 3966 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3967 | struct eventfd_ctx *eventfd) |
3968 | { | |
59b6f873 | 3969 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3970 | } |
3971 | ||
59b6f873 | 3972 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3973 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3974 | { |
9490ff27 | 3975 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3976 | |
9490ff27 KH |
3977 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3978 | if (!event) | |
3979 | return -ENOMEM; | |
3980 | ||
1af8efe9 | 3981 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3982 | |
3983 | event->eventfd = eventfd; | |
3984 | list_add(&event->list, &memcg->oom_notify); | |
3985 | ||
3986 | /* already in OOM ? */ | |
79dfdacc | 3987 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 3988 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3989 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3990 | |
3991 | return 0; | |
3992 | } | |
3993 | ||
59b6f873 | 3994 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3995 | struct eventfd_ctx *eventfd) |
9490ff27 | 3996 | { |
9490ff27 | 3997 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3998 | |
1af8efe9 | 3999 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4000 | |
c0ff4b85 | 4001 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4002 | if (ev->eventfd == eventfd) { |
4003 | list_del(&ev->list); | |
4004 | kfree(ev); | |
4005 | } | |
4006 | } | |
4007 | ||
1af8efe9 | 4008 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4009 | } |
4010 | ||
2da8ca82 | 4011 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4012 | { |
2da8ca82 | 4013 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4014 | |
791badbd TH |
4015 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4016 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4017 | return 0; |
4018 | } | |
4019 | ||
182446d0 | 4020 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4021 | struct cftype *cft, u64 val) |
4022 | { | |
182446d0 | 4023 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4024 | |
4025 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4026 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4027 | return -EINVAL; |
4028 | ||
c0ff4b85 | 4029 | memcg->oom_kill_disable = val; |
4d845ebf | 4030 | if (!val) |
c0ff4b85 | 4031 | memcg_oom_recover(memcg); |
3dae7fec | 4032 | |
3c11ecf4 KH |
4033 | return 0; |
4034 | } | |
4035 | ||
c255a458 | 4036 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4037 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4038 | { |
55007d84 GC |
4039 | int ret; |
4040 | ||
55007d84 GC |
4041 | ret = memcg_propagate_kmem(memcg); |
4042 | if (ret) | |
4043 | return ret; | |
2633d7a0 | 4044 | |
1d62e436 | 4045 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4046 | } |
e5671dfa | 4047 | |
2a4db7eb VD |
4048 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
4049 | { | |
4050 | if (!memcg->kmem_acct_active) | |
4051 | return; | |
4052 | ||
4053 | /* | |
4054 | * Clear the 'active' flag before clearing memcg_caches arrays entries. | |
4055 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it | |
4056 | * guarantees no cache will be created for this cgroup after we are | |
4057 | * done (see memcg_create_kmem_cache()). | |
4058 | */ | |
4059 | memcg->kmem_acct_active = false; | |
4060 | ||
4061 | memcg_deactivate_kmem_caches(memcg); | |
4062 | } | |
4063 | ||
10d5ebf4 | 4064 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4065 | { |
d5b3cf71 | 4066 | memcg_destroy_kmem_caches(memcg); |
1d62e436 | 4067 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 4068 | } |
e5671dfa | 4069 | #else |
cbe128e3 | 4070 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4071 | { |
4072 | return 0; | |
4073 | } | |
d1a4c0b3 | 4074 | |
2a4db7eb VD |
4075 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
4076 | { | |
4077 | } | |
4078 | ||
10d5ebf4 LZ |
4079 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4080 | { | |
4081 | } | |
e5671dfa GC |
4082 | #endif |
4083 | ||
3bc942f3 TH |
4084 | /* |
4085 | * DO NOT USE IN NEW FILES. | |
4086 | * | |
4087 | * "cgroup.event_control" implementation. | |
4088 | * | |
4089 | * This is way over-engineered. It tries to support fully configurable | |
4090 | * events for each user. Such level of flexibility is completely | |
4091 | * unnecessary especially in the light of the planned unified hierarchy. | |
4092 | * | |
4093 | * Please deprecate this and replace with something simpler if at all | |
4094 | * possible. | |
4095 | */ | |
4096 | ||
79bd9814 TH |
4097 | /* |
4098 | * Unregister event and free resources. | |
4099 | * | |
4100 | * Gets called from workqueue. | |
4101 | */ | |
3bc942f3 | 4102 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4103 | { |
3bc942f3 TH |
4104 | struct mem_cgroup_event *event = |
4105 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4106 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4107 | |
4108 | remove_wait_queue(event->wqh, &event->wait); | |
4109 | ||
59b6f873 | 4110 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4111 | |
4112 | /* Notify userspace the event is going away. */ | |
4113 | eventfd_signal(event->eventfd, 1); | |
4114 | ||
4115 | eventfd_ctx_put(event->eventfd); | |
4116 | kfree(event); | |
59b6f873 | 4117 | css_put(&memcg->css); |
79bd9814 TH |
4118 | } |
4119 | ||
4120 | /* | |
4121 | * Gets called on POLLHUP on eventfd when user closes it. | |
4122 | * | |
4123 | * Called with wqh->lock held and interrupts disabled. | |
4124 | */ | |
3bc942f3 TH |
4125 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4126 | int sync, void *key) | |
79bd9814 | 4127 | { |
3bc942f3 TH |
4128 | struct mem_cgroup_event *event = |
4129 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4130 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4131 | unsigned long flags = (unsigned long)key; |
4132 | ||
4133 | if (flags & POLLHUP) { | |
4134 | /* | |
4135 | * If the event has been detached at cgroup removal, we | |
4136 | * can simply return knowing the other side will cleanup | |
4137 | * for us. | |
4138 | * | |
4139 | * We can't race against event freeing since the other | |
4140 | * side will require wqh->lock via remove_wait_queue(), | |
4141 | * which we hold. | |
4142 | */ | |
fba94807 | 4143 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4144 | if (!list_empty(&event->list)) { |
4145 | list_del_init(&event->list); | |
4146 | /* | |
4147 | * We are in atomic context, but cgroup_event_remove() | |
4148 | * may sleep, so we have to call it in workqueue. | |
4149 | */ | |
4150 | schedule_work(&event->remove); | |
4151 | } | |
fba94807 | 4152 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4153 | } |
4154 | ||
4155 | return 0; | |
4156 | } | |
4157 | ||
3bc942f3 | 4158 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4159 | wait_queue_head_t *wqh, poll_table *pt) |
4160 | { | |
3bc942f3 TH |
4161 | struct mem_cgroup_event *event = |
4162 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4163 | |
4164 | event->wqh = wqh; | |
4165 | add_wait_queue(wqh, &event->wait); | |
4166 | } | |
4167 | ||
4168 | /* | |
3bc942f3 TH |
4169 | * DO NOT USE IN NEW FILES. |
4170 | * | |
79bd9814 TH |
4171 | * Parse input and register new cgroup event handler. |
4172 | * | |
4173 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4174 | * Interpretation of args is defined by control file implementation. | |
4175 | */ | |
451af504 TH |
4176 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4177 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4178 | { |
451af504 | 4179 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4180 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4181 | struct mem_cgroup_event *event; |
79bd9814 TH |
4182 | struct cgroup_subsys_state *cfile_css; |
4183 | unsigned int efd, cfd; | |
4184 | struct fd efile; | |
4185 | struct fd cfile; | |
fba94807 | 4186 | const char *name; |
79bd9814 TH |
4187 | char *endp; |
4188 | int ret; | |
4189 | ||
451af504 TH |
4190 | buf = strstrip(buf); |
4191 | ||
4192 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4193 | if (*endp != ' ') |
4194 | return -EINVAL; | |
451af504 | 4195 | buf = endp + 1; |
79bd9814 | 4196 | |
451af504 | 4197 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4198 | if ((*endp != ' ') && (*endp != '\0')) |
4199 | return -EINVAL; | |
451af504 | 4200 | buf = endp + 1; |
79bd9814 TH |
4201 | |
4202 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4203 | if (!event) | |
4204 | return -ENOMEM; | |
4205 | ||
59b6f873 | 4206 | event->memcg = memcg; |
79bd9814 | 4207 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4208 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4209 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4210 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4211 | |
4212 | efile = fdget(efd); | |
4213 | if (!efile.file) { | |
4214 | ret = -EBADF; | |
4215 | goto out_kfree; | |
4216 | } | |
4217 | ||
4218 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4219 | if (IS_ERR(event->eventfd)) { | |
4220 | ret = PTR_ERR(event->eventfd); | |
4221 | goto out_put_efile; | |
4222 | } | |
4223 | ||
4224 | cfile = fdget(cfd); | |
4225 | if (!cfile.file) { | |
4226 | ret = -EBADF; | |
4227 | goto out_put_eventfd; | |
4228 | } | |
4229 | ||
4230 | /* the process need read permission on control file */ | |
4231 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4232 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4233 | if (ret < 0) | |
4234 | goto out_put_cfile; | |
4235 | ||
fba94807 TH |
4236 | /* |
4237 | * Determine the event callbacks and set them in @event. This used | |
4238 | * to be done via struct cftype but cgroup core no longer knows | |
4239 | * about these events. The following is crude but the whole thing | |
4240 | * is for compatibility anyway. | |
3bc942f3 TH |
4241 | * |
4242 | * DO NOT ADD NEW FILES. | |
fba94807 | 4243 | */ |
b583043e | 4244 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4245 | |
4246 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4247 | event->register_event = mem_cgroup_usage_register_event; | |
4248 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4249 | } else if (!strcmp(name, "memory.oom_control")) { | |
4250 | event->register_event = mem_cgroup_oom_register_event; | |
4251 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4252 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4253 | event->register_event = vmpressure_register_event; | |
4254 | event->unregister_event = vmpressure_unregister_event; | |
4255 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4256 | event->register_event = memsw_cgroup_usage_register_event; |
4257 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4258 | } else { |
4259 | ret = -EINVAL; | |
4260 | goto out_put_cfile; | |
4261 | } | |
4262 | ||
79bd9814 | 4263 | /* |
b5557c4c TH |
4264 | * Verify @cfile should belong to @css. Also, remaining events are |
4265 | * automatically removed on cgroup destruction but the removal is | |
4266 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4267 | */ |
b583043e | 4268 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4269 | &memory_cgrp_subsys); |
79bd9814 | 4270 | ret = -EINVAL; |
5a17f543 | 4271 | if (IS_ERR(cfile_css)) |
79bd9814 | 4272 | goto out_put_cfile; |
5a17f543 TH |
4273 | if (cfile_css != css) { |
4274 | css_put(cfile_css); | |
79bd9814 | 4275 | goto out_put_cfile; |
5a17f543 | 4276 | } |
79bd9814 | 4277 | |
451af504 | 4278 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4279 | if (ret) |
4280 | goto out_put_css; | |
4281 | ||
4282 | efile.file->f_op->poll(efile.file, &event->pt); | |
4283 | ||
fba94807 TH |
4284 | spin_lock(&memcg->event_list_lock); |
4285 | list_add(&event->list, &memcg->event_list); | |
4286 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4287 | |
4288 | fdput(cfile); | |
4289 | fdput(efile); | |
4290 | ||
451af504 | 4291 | return nbytes; |
79bd9814 TH |
4292 | |
4293 | out_put_css: | |
b5557c4c | 4294 | css_put(css); |
79bd9814 TH |
4295 | out_put_cfile: |
4296 | fdput(cfile); | |
4297 | out_put_eventfd: | |
4298 | eventfd_ctx_put(event->eventfd); | |
4299 | out_put_efile: | |
4300 | fdput(efile); | |
4301 | out_kfree: | |
4302 | kfree(event); | |
4303 | ||
4304 | return ret; | |
4305 | } | |
4306 | ||
241994ed | 4307 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4308 | { |
0eea1030 | 4309 | .name = "usage_in_bytes", |
8c7c6e34 | 4310 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4311 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4312 | }, |
c84872e1 PE |
4313 | { |
4314 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4315 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4316 | .write = mem_cgroup_reset, |
791badbd | 4317 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4318 | }, |
8cdea7c0 | 4319 | { |
0eea1030 | 4320 | .name = "limit_in_bytes", |
8c7c6e34 | 4321 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4322 | .write = mem_cgroup_write, |
791badbd | 4323 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4324 | }, |
296c81d8 BS |
4325 | { |
4326 | .name = "soft_limit_in_bytes", | |
4327 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4328 | .write = mem_cgroup_write, |
791badbd | 4329 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4330 | }, |
8cdea7c0 BS |
4331 | { |
4332 | .name = "failcnt", | |
8c7c6e34 | 4333 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4334 | .write = mem_cgroup_reset, |
791badbd | 4335 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4336 | }, |
d2ceb9b7 KH |
4337 | { |
4338 | .name = "stat", | |
2da8ca82 | 4339 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4340 | }, |
c1e862c1 KH |
4341 | { |
4342 | .name = "force_empty", | |
6770c64e | 4343 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4344 | }, |
18f59ea7 BS |
4345 | { |
4346 | .name = "use_hierarchy", | |
4347 | .write_u64 = mem_cgroup_hierarchy_write, | |
4348 | .read_u64 = mem_cgroup_hierarchy_read, | |
4349 | }, | |
79bd9814 | 4350 | { |
3bc942f3 | 4351 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4352 | .write = memcg_write_event_control, |
79bd9814 TH |
4353 | .flags = CFTYPE_NO_PREFIX, |
4354 | .mode = S_IWUGO, | |
4355 | }, | |
a7885eb8 KM |
4356 | { |
4357 | .name = "swappiness", | |
4358 | .read_u64 = mem_cgroup_swappiness_read, | |
4359 | .write_u64 = mem_cgroup_swappiness_write, | |
4360 | }, | |
7dc74be0 DN |
4361 | { |
4362 | .name = "move_charge_at_immigrate", | |
4363 | .read_u64 = mem_cgroup_move_charge_read, | |
4364 | .write_u64 = mem_cgroup_move_charge_write, | |
4365 | }, | |
9490ff27 KH |
4366 | { |
4367 | .name = "oom_control", | |
2da8ca82 | 4368 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4369 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4370 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4371 | }, | |
70ddf637 AV |
4372 | { |
4373 | .name = "pressure_level", | |
70ddf637 | 4374 | }, |
406eb0c9 YH |
4375 | #ifdef CONFIG_NUMA |
4376 | { | |
4377 | .name = "numa_stat", | |
2da8ca82 | 4378 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4379 | }, |
4380 | #endif | |
510fc4e1 GC |
4381 | #ifdef CONFIG_MEMCG_KMEM |
4382 | { | |
4383 | .name = "kmem.limit_in_bytes", | |
4384 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4385 | .write = mem_cgroup_write, |
791badbd | 4386 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4387 | }, |
4388 | { | |
4389 | .name = "kmem.usage_in_bytes", | |
4390 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4391 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4392 | }, |
4393 | { | |
4394 | .name = "kmem.failcnt", | |
4395 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4396 | .write = mem_cgroup_reset, |
791badbd | 4397 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4398 | }, |
4399 | { | |
4400 | .name = "kmem.max_usage_in_bytes", | |
4401 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4402 | .write = mem_cgroup_reset, |
791badbd | 4403 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4404 | }, |
749c5415 GC |
4405 | #ifdef CONFIG_SLABINFO |
4406 | { | |
4407 | .name = "kmem.slabinfo", | |
b047501c VD |
4408 | .seq_start = slab_start, |
4409 | .seq_next = slab_next, | |
4410 | .seq_stop = slab_stop, | |
4411 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4412 | }, |
4413 | #endif | |
8c7c6e34 | 4414 | #endif |
6bc10349 | 4415 | { }, /* terminate */ |
af36f906 | 4416 | }; |
8c7c6e34 | 4417 | |
c0ff4b85 | 4418 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4419 | { |
4420 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4421 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4422 | int zone, tmp = node; |
1ecaab2b KH |
4423 | /* |
4424 | * This routine is called against possible nodes. | |
4425 | * But it's BUG to call kmalloc() against offline node. | |
4426 | * | |
4427 | * TODO: this routine can waste much memory for nodes which will | |
4428 | * never be onlined. It's better to use memory hotplug callback | |
4429 | * function. | |
4430 | */ | |
41e3355d KH |
4431 | if (!node_state(node, N_NORMAL_MEMORY)) |
4432 | tmp = -1; | |
17295c88 | 4433 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4434 | if (!pn) |
4435 | return 1; | |
1ecaab2b | 4436 | |
1ecaab2b KH |
4437 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4438 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4439 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4440 | mz->usage_in_excess = 0; |
4441 | mz->on_tree = false; | |
d79154bb | 4442 | mz->memcg = memcg; |
1ecaab2b | 4443 | } |
54f72fe0 | 4444 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4445 | return 0; |
4446 | } | |
4447 | ||
c0ff4b85 | 4448 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4449 | { |
54f72fe0 | 4450 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4451 | } |
4452 | ||
33327948 KH |
4453 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4454 | { | |
d79154bb | 4455 | struct mem_cgroup *memcg; |
8ff69e2c | 4456 | size_t size; |
33327948 | 4457 | |
8ff69e2c VD |
4458 | size = sizeof(struct mem_cgroup); |
4459 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4460 | |
8ff69e2c | 4461 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4462 | if (!memcg) |
e7bbcdf3 DC |
4463 | return NULL; |
4464 | ||
d79154bb HD |
4465 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4466 | if (!memcg->stat) | |
d2e61b8d | 4467 | goto out_free; |
d79154bb HD |
4468 | spin_lock_init(&memcg->pcp_counter_lock); |
4469 | return memcg; | |
d2e61b8d DC |
4470 | |
4471 | out_free: | |
8ff69e2c | 4472 | kfree(memcg); |
d2e61b8d | 4473 | return NULL; |
33327948 KH |
4474 | } |
4475 | ||
59927fb9 | 4476 | /* |
c8b2a36f GC |
4477 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4478 | * (scanning all at force_empty is too costly...) | |
4479 | * | |
4480 | * Instead of clearing all references at force_empty, we remember | |
4481 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4482 | * it goes down to 0. | |
4483 | * | |
4484 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4485 | */ |
c8b2a36f GC |
4486 | |
4487 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4488 | { |
c8b2a36f | 4489 | int node; |
59927fb9 | 4490 | |
bb4cc1a8 | 4491 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4492 | |
4493 | for_each_node(node) | |
4494 | free_mem_cgroup_per_zone_info(memcg, node); | |
4495 | ||
4496 | free_percpu(memcg->stat); | |
4497 | ||
a8964b9b | 4498 | disarm_static_keys(memcg); |
8ff69e2c | 4499 | kfree(memcg); |
59927fb9 | 4500 | } |
3afe36b1 | 4501 | |
7bcc1bb1 DN |
4502 | /* |
4503 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4504 | */ | |
e1aab161 | 4505 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4506 | { |
3e32cb2e | 4507 | if (!memcg->memory.parent) |
7bcc1bb1 | 4508 | return NULL; |
3e32cb2e | 4509 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4510 | } |
e1aab161 | 4511 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4512 | |
0eb253e2 | 4513 | static struct cgroup_subsys_state * __ref |
eb95419b | 4514 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4515 | { |
d142e3e6 | 4516 | struct mem_cgroup *memcg; |
04046e1a | 4517 | long error = -ENOMEM; |
6d12e2d8 | 4518 | int node; |
8cdea7c0 | 4519 | |
c0ff4b85 R |
4520 | memcg = mem_cgroup_alloc(); |
4521 | if (!memcg) | |
04046e1a | 4522 | return ERR_PTR(error); |
78fb7466 | 4523 | |
3ed28fa1 | 4524 | for_each_node(node) |
c0ff4b85 | 4525 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4526 | goto free_out; |
f64c3f54 | 4527 | |
c077719b | 4528 | /* root ? */ |
eb95419b | 4529 | if (parent_css == NULL) { |
a41c58a6 | 4530 | root_mem_cgroup = memcg; |
3e32cb2e | 4531 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4532 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4533 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4534 | page_counter_init(&memcg->memsw, NULL); |
4535 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4536 | } |
28dbc4b6 | 4537 | |
d142e3e6 GC |
4538 | memcg->last_scanned_node = MAX_NUMNODES; |
4539 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4540 | memcg->move_charge_at_immigrate = 0; |
4541 | mutex_init(&memcg->thresholds_lock); | |
4542 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4543 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4544 | INIT_LIST_HEAD(&memcg->event_list); |
4545 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4546 | #ifdef CONFIG_MEMCG_KMEM |
4547 | memcg->kmemcg_id = -1; | |
900a38f0 | 4548 | #endif |
d142e3e6 GC |
4549 | |
4550 | return &memcg->css; | |
4551 | ||
4552 | free_out: | |
4553 | __mem_cgroup_free(memcg); | |
4554 | return ERR_PTR(error); | |
4555 | } | |
4556 | ||
4557 | static int | |
eb95419b | 4558 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4559 | { |
eb95419b | 4560 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4561 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4562 | int ret; |
d142e3e6 | 4563 | |
15a4c835 | 4564 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4565 | return -ENOSPC; |
4566 | ||
63876986 | 4567 | if (!parent) |
d142e3e6 GC |
4568 | return 0; |
4569 | ||
0999821b | 4570 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4571 | |
4572 | memcg->use_hierarchy = parent->use_hierarchy; | |
4573 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4574 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4575 | ||
4576 | if (parent->use_hierarchy) { | |
3e32cb2e | 4577 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4578 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4579 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4580 | page_counter_init(&memcg->memsw, &parent->memsw); |
4581 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4582 | |
7bcc1bb1 | 4583 | /* |
8d76a979 LZ |
4584 | * No need to take a reference to the parent because cgroup |
4585 | * core guarantees its existence. | |
7bcc1bb1 | 4586 | */ |
18f59ea7 | 4587 | } else { |
3e32cb2e | 4588 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4589 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4590 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4591 | page_counter_init(&memcg->memsw, NULL); |
4592 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4593 | /* |
4594 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4595 | * much sense so let cgroup subsystem know about this | |
4596 | * unfortunate state in our controller. | |
4597 | */ | |
d142e3e6 | 4598 | if (parent != root_mem_cgroup) |
073219e9 | 4599 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4600 | } |
0999821b | 4601 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4602 | |
2f7dd7a4 JW |
4603 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4604 | if (ret) | |
4605 | return ret; | |
4606 | ||
4607 | /* | |
4608 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4609 | * orders reading memcg->initialized against its callers | |
4610 | * reading the memcg members. | |
4611 | */ | |
4612 | smp_store_release(&memcg->initialized, 1); | |
4613 | ||
4614 | return 0; | |
8cdea7c0 BS |
4615 | } |
4616 | ||
eb95419b | 4617 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4618 | { |
eb95419b | 4619 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4620 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4621 | |
4622 | /* | |
4623 | * Unregister events and notify userspace. | |
4624 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4625 | * directory to avoid race between userspace and kernelspace. | |
4626 | */ | |
fba94807 TH |
4627 | spin_lock(&memcg->event_list_lock); |
4628 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4629 | list_del_init(&event->list); |
4630 | schedule_work(&event->remove); | |
4631 | } | |
fba94807 | 4632 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4633 | |
33cb876e | 4634 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb VD |
4635 | |
4636 | memcg_deactivate_kmem(memcg); | |
df878fb0 KH |
4637 | } |
4638 | ||
eb95419b | 4639 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4640 | { |
eb95419b | 4641 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4642 | |
10d5ebf4 | 4643 | memcg_destroy_kmem(memcg); |
465939a1 | 4644 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4645 | } |
4646 | ||
1ced953b TH |
4647 | /** |
4648 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4649 | * @css: the target css | |
4650 | * | |
4651 | * Reset the states of the mem_cgroup associated with @css. This is | |
4652 | * invoked when the userland requests disabling on the default hierarchy | |
4653 | * but the memcg is pinned through dependency. The memcg should stop | |
4654 | * applying policies and should revert to the vanilla state as it may be | |
4655 | * made visible again. | |
4656 | * | |
4657 | * The current implementation only resets the essential configurations. | |
4658 | * This needs to be expanded to cover all the visible parts. | |
4659 | */ | |
4660 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4661 | { | |
4662 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4663 | ||
3e32cb2e JW |
4664 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4665 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4666 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4667 | memcg->low = 0; |
4668 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4669 | memcg->soft_limit = PAGE_COUNTER_MAX; |
1ced953b TH |
4670 | } |
4671 | ||
02491447 | 4672 | #ifdef CONFIG_MMU |
7dc74be0 | 4673 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4674 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4675 | { |
05b84301 | 4676 | int ret; |
9476db97 JW |
4677 | |
4678 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 4679 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 4680 | if (!ret) { |
854ffa8d | 4681 | mc.precharge += count; |
854ffa8d DN |
4682 | return ret; |
4683 | } | |
692e7c45 | 4684 | if (ret == -EINTR) { |
00501b53 | 4685 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
4686 | return ret; |
4687 | } | |
9476db97 JW |
4688 | |
4689 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4690 | while (count--) { |
00501b53 | 4691 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
4692 | /* |
4693 | * In case of failure, any residual charges against | |
4694 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
4695 | * later on. However, cancel any charges that are |
4696 | * bypassed to root right away or they'll be lost. | |
9476db97 | 4697 | */ |
692e7c45 | 4698 | if (ret == -EINTR) |
00501b53 | 4699 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 4700 | if (ret) |
38c5d72f | 4701 | return ret; |
854ffa8d | 4702 | mc.precharge++; |
9476db97 | 4703 | cond_resched(); |
854ffa8d | 4704 | } |
9476db97 | 4705 | return 0; |
4ffef5fe DN |
4706 | } |
4707 | ||
4708 | /** | |
8d32ff84 | 4709 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4710 | * @vma: the vma the pte to be checked belongs |
4711 | * @addr: the address corresponding to the pte to be checked | |
4712 | * @ptent: the pte to be checked | |
02491447 | 4713 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4714 | * |
4715 | * Returns | |
4716 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4717 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4718 | * move charge. if @target is not NULL, the page is stored in target->page | |
4719 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4720 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4721 | * target for charge migration. if @target is not NULL, the entry is stored | |
4722 | * in target->ent. | |
4ffef5fe DN |
4723 | * |
4724 | * Called with pte lock held. | |
4725 | */ | |
4ffef5fe DN |
4726 | union mc_target { |
4727 | struct page *page; | |
02491447 | 4728 | swp_entry_t ent; |
4ffef5fe DN |
4729 | }; |
4730 | ||
4ffef5fe | 4731 | enum mc_target_type { |
8d32ff84 | 4732 | MC_TARGET_NONE = 0, |
4ffef5fe | 4733 | MC_TARGET_PAGE, |
02491447 | 4734 | MC_TARGET_SWAP, |
4ffef5fe DN |
4735 | }; |
4736 | ||
90254a65 DN |
4737 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4738 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4739 | { |
90254a65 | 4740 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4741 | |
90254a65 DN |
4742 | if (!page || !page_mapped(page)) |
4743 | return NULL; | |
4744 | if (PageAnon(page)) { | |
1dfab5ab | 4745 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4746 | return NULL; |
1dfab5ab JW |
4747 | } else { |
4748 | if (!(mc.flags & MOVE_FILE)) | |
4749 | return NULL; | |
4750 | } | |
90254a65 DN |
4751 | if (!get_page_unless_zero(page)) |
4752 | return NULL; | |
4753 | ||
4754 | return page; | |
4755 | } | |
4756 | ||
4b91355e | 4757 | #ifdef CONFIG_SWAP |
90254a65 DN |
4758 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4759 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4760 | { | |
90254a65 DN |
4761 | struct page *page = NULL; |
4762 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4763 | ||
1dfab5ab | 4764 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4765 | return NULL; |
4b91355e KH |
4766 | /* |
4767 | * Because lookup_swap_cache() updates some statistics counter, | |
4768 | * we call find_get_page() with swapper_space directly. | |
4769 | */ | |
33806f06 | 4770 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4771 | if (do_swap_account) |
4772 | entry->val = ent.val; | |
4773 | ||
4774 | return page; | |
4775 | } | |
4b91355e KH |
4776 | #else |
4777 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4778 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4779 | { | |
4780 | return NULL; | |
4781 | } | |
4782 | #endif | |
90254a65 | 4783 | |
87946a72 DN |
4784 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4785 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4786 | { | |
4787 | struct page *page = NULL; | |
87946a72 DN |
4788 | struct address_space *mapping; |
4789 | pgoff_t pgoff; | |
4790 | ||
4791 | if (!vma->vm_file) /* anonymous vma */ | |
4792 | return NULL; | |
1dfab5ab | 4793 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4794 | return NULL; |
4795 | ||
87946a72 | 4796 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4797 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4798 | |
4799 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4800 | #ifdef CONFIG_SWAP |
4801 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4802 | if (shmem_mapping(mapping)) { |
4803 | page = find_get_entry(mapping, pgoff); | |
4804 | if (radix_tree_exceptional_entry(page)) { | |
4805 | swp_entry_t swp = radix_to_swp_entry(page); | |
4806 | if (do_swap_account) | |
4807 | *entry = swp; | |
4808 | page = find_get_page(swap_address_space(swp), swp.val); | |
4809 | } | |
4810 | } else | |
4811 | page = find_get_page(mapping, pgoff); | |
4812 | #else | |
4813 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4814 | #endif |
87946a72 DN |
4815 | return page; |
4816 | } | |
4817 | ||
8d32ff84 | 4818 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4819 | unsigned long addr, pte_t ptent, union mc_target *target) |
4820 | { | |
4821 | struct page *page = NULL; | |
8d32ff84 | 4822 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4823 | swp_entry_t ent = { .val = 0 }; |
4824 | ||
4825 | if (pte_present(ptent)) | |
4826 | page = mc_handle_present_pte(vma, addr, ptent); | |
4827 | else if (is_swap_pte(ptent)) | |
4828 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4829 | else if (pte_none(ptent)) |
87946a72 | 4830 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4831 | |
4832 | if (!page && !ent.val) | |
8d32ff84 | 4833 | return ret; |
02491447 | 4834 | if (page) { |
02491447 | 4835 | /* |
0a31bc97 | 4836 | * Do only loose check w/o serialization. |
1306a85a | 4837 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4838 | * not under LRU exclusion. |
02491447 | 4839 | */ |
1306a85a | 4840 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4841 | ret = MC_TARGET_PAGE; |
4842 | if (target) | |
4843 | target->page = page; | |
4844 | } | |
4845 | if (!ret || !target) | |
4846 | put_page(page); | |
4847 | } | |
90254a65 DN |
4848 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4849 | if (ent.val && !ret && | |
34c00c31 | 4850 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4851 | ret = MC_TARGET_SWAP; |
4852 | if (target) | |
4853 | target->ent = ent; | |
4ffef5fe | 4854 | } |
4ffef5fe DN |
4855 | return ret; |
4856 | } | |
4857 | ||
12724850 NH |
4858 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4859 | /* | |
4860 | * We don't consider swapping or file mapped pages because THP does not | |
4861 | * support them for now. | |
4862 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4863 | */ | |
4864 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4865 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4866 | { | |
4867 | struct page *page = NULL; | |
12724850 NH |
4868 | enum mc_target_type ret = MC_TARGET_NONE; |
4869 | ||
4870 | page = pmd_page(pmd); | |
309381fe | 4871 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4872 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4873 | return ret; |
1306a85a | 4874 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4875 | ret = MC_TARGET_PAGE; |
4876 | if (target) { | |
4877 | get_page(page); | |
4878 | target->page = page; | |
4879 | } | |
4880 | } | |
4881 | return ret; | |
4882 | } | |
4883 | #else | |
4884 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4885 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4886 | { | |
4887 | return MC_TARGET_NONE; | |
4888 | } | |
4889 | #endif | |
4890 | ||
4ffef5fe DN |
4891 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4892 | unsigned long addr, unsigned long end, | |
4893 | struct mm_walk *walk) | |
4894 | { | |
26bcd64a | 4895 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4896 | pte_t *pte; |
4897 | spinlock_t *ptl; | |
4898 | ||
bf929152 | 4899 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4900 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4901 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4902 | spin_unlock(ptl); |
1a5a9906 | 4903 | return 0; |
12724850 | 4904 | } |
03319327 | 4905 | |
45f83cef AA |
4906 | if (pmd_trans_unstable(pmd)) |
4907 | return 0; | |
4ffef5fe DN |
4908 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4909 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4910 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4911 | mc.precharge++; /* increment precharge temporarily */ |
4912 | pte_unmap_unlock(pte - 1, ptl); | |
4913 | cond_resched(); | |
4914 | ||
7dc74be0 DN |
4915 | return 0; |
4916 | } | |
4917 | ||
4ffef5fe DN |
4918 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4919 | { | |
4920 | unsigned long precharge; | |
4ffef5fe | 4921 | |
26bcd64a NH |
4922 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4923 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4924 | .mm = mm, | |
4925 | }; | |
dfe076b0 | 4926 | down_read(&mm->mmap_sem); |
26bcd64a | 4927 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4928 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4929 | |
4930 | precharge = mc.precharge; | |
4931 | mc.precharge = 0; | |
4932 | ||
4933 | return precharge; | |
4934 | } | |
4935 | ||
4ffef5fe DN |
4936 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4937 | { | |
dfe076b0 DN |
4938 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4939 | ||
4940 | VM_BUG_ON(mc.moving_task); | |
4941 | mc.moving_task = current; | |
4942 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4943 | } |
4944 | ||
dfe076b0 DN |
4945 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4946 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4947 | { |
2bd9bb20 KH |
4948 | struct mem_cgroup *from = mc.from; |
4949 | struct mem_cgroup *to = mc.to; | |
4950 | ||
4ffef5fe | 4951 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4952 | if (mc.precharge) { |
00501b53 | 4953 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4954 | mc.precharge = 0; |
4955 | } | |
4956 | /* | |
4957 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4958 | * we must uncharge here. | |
4959 | */ | |
4960 | if (mc.moved_charge) { | |
00501b53 | 4961 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4962 | mc.moved_charge = 0; |
4ffef5fe | 4963 | } |
483c30b5 DN |
4964 | /* we must fixup refcnts and charges */ |
4965 | if (mc.moved_swap) { | |
483c30b5 | 4966 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4967 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4968 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4969 | |
05b84301 | 4970 | /* |
3e32cb2e JW |
4971 | * we charged both to->memory and to->memsw, so we |
4972 | * should uncharge to->memory. | |
05b84301 | 4973 | */ |
ce00a967 | 4974 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4975 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4976 | ||
e8ea14cc | 4977 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4978 | |
4050377b | 4979 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4980 | mc.moved_swap = 0; |
4981 | } | |
dfe076b0 DN |
4982 | memcg_oom_recover(from); |
4983 | memcg_oom_recover(to); | |
4984 | wake_up_all(&mc.waitq); | |
4985 | } | |
4986 | ||
4987 | static void mem_cgroup_clear_mc(void) | |
4988 | { | |
dfe076b0 DN |
4989 | /* |
4990 | * we must clear moving_task before waking up waiters at the end of | |
4991 | * task migration. | |
4992 | */ | |
4993 | mc.moving_task = NULL; | |
4994 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4995 | spin_lock(&mc.lock); |
4ffef5fe DN |
4996 | mc.from = NULL; |
4997 | mc.to = NULL; | |
2bd9bb20 | 4998 | spin_unlock(&mc.lock); |
4ffef5fe DN |
4999 | } |
5000 | ||
eb95419b | 5001 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5002 | struct cgroup_taskset *tset) |
7dc74be0 | 5003 | { |
2f7ee569 | 5004 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5005 | int ret = 0; |
eb95419b | 5006 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
1dfab5ab | 5007 | unsigned long move_flags; |
7dc74be0 | 5008 | |
ee5e8472 GC |
5009 | /* |
5010 | * We are now commited to this value whatever it is. Changes in this | |
5011 | * tunable will only affect upcoming migrations, not the current one. | |
5012 | * So we need to save it, and keep it going. | |
5013 | */ | |
1dfab5ab JW |
5014 | move_flags = ACCESS_ONCE(memcg->move_charge_at_immigrate); |
5015 | if (move_flags) { | |
7dc74be0 DN |
5016 | struct mm_struct *mm; |
5017 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5018 | ||
c0ff4b85 | 5019 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5020 | |
5021 | mm = get_task_mm(p); | |
5022 | if (!mm) | |
5023 | return 0; | |
7dc74be0 | 5024 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5025 | if (mm->owner == p) { |
5026 | VM_BUG_ON(mc.from); | |
5027 | VM_BUG_ON(mc.to); | |
5028 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5029 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5030 | VM_BUG_ON(mc.moved_swap); |
247b1447 | 5031 | |
2bd9bb20 | 5032 | spin_lock(&mc.lock); |
4ffef5fe | 5033 | mc.from = from; |
c0ff4b85 | 5034 | mc.to = memcg; |
1dfab5ab | 5035 | mc.flags = move_flags; |
2bd9bb20 | 5036 | spin_unlock(&mc.lock); |
dfe076b0 | 5037 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5038 | |
5039 | ret = mem_cgroup_precharge_mc(mm); | |
5040 | if (ret) | |
5041 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5042 | } |
5043 | mmput(mm); | |
7dc74be0 DN |
5044 | } |
5045 | return ret; | |
5046 | } | |
5047 | ||
eb95419b | 5048 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5049 | struct cgroup_taskset *tset) |
7dc74be0 | 5050 | { |
4e2f245d JW |
5051 | if (mc.to) |
5052 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5053 | } |
5054 | ||
4ffef5fe DN |
5055 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5056 | unsigned long addr, unsigned long end, | |
5057 | struct mm_walk *walk) | |
7dc74be0 | 5058 | { |
4ffef5fe | 5059 | int ret = 0; |
26bcd64a | 5060 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5061 | pte_t *pte; |
5062 | spinlock_t *ptl; | |
12724850 NH |
5063 | enum mc_target_type target_type; |
5064 | union mc_target target; | |
5065 | struct page *page; | |
4ffef5fe | 5066 | |
12724850 NH |
5067 | /* |
5068 | * We don't take compound_lock() here but no race with splitting thp | |
5069 | * happens because: | |
5070 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5071 | * under splitting, which means there's no concurrent thp split, | |
5072 | * - if another thread runs into split_huge_page() just after we | |
5073 | * entered this if-block, the thread must wait for page table lock | |
5074 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5075 | * part of thp split is not executed yet. | |
5076 | */ | |
bf929152 | 5077 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5078 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5079 | spin_unlock(ptl); |
12724850 NH |
5080 | return 0; |
5081 | } | |
5082 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5083 | if (target_type == MC_TARGET_PAGE) { | |
5084 | page = target.page; | |
5085 | if (!isolate_lru_page(page)) { | |
12724850 | 5086 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 5087 | mc.from, mc.to)) { |
12724850 NH |
5088 | mc.precharge -= HPAGE_PMD_NR; |
5089 | mc.moved_charge += HPAGE_PMD_NR; | |
5090 | } | |
5091 | putback_lru_page(page); | |
5092 | } | |
5093 | put_page(page); | |
5094 | } | |
bf929152 | 5095 | spin_unlock(ptl); |
1a5a9906 | 5096 | return 0; |
12724850 NH |
5097 | } |
5098 | ||
45f83cef AA |
5099 | if (pmd_trans_unstable(pmd)) |
5100 | return 0; | |
4ffef5fe DN |
5101 | retry: |
5102 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5103 | for (; addr != end; addr += PAGE_SIZE) { | |
5104 | pte_t ptent = *(pte++); | |
02491447 | 5105 | swp_entry_t ent; |
4ffef5fe DN |
5106 | |
5107 | if (!mc.precharge) | |
5108 | break; | |
5109 | ||
8d32ff84 | 5110 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5111 | case MC_TARGET_PAGE: |
5112 | page = target.page; | |
5113 | if (isolate_lru_page(page)) | |
5114 | goto put; | |
1306a85a | 5115 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 5116 | mc.precharge--; |
854ffa8d DN |
5117 | /* we uncharge from mc.from later. */ |
5118 | mc.moved_charge++; | |
4ffef5fe DN |
5119 | } |
5120 | putback_lru_page(page); | |
8d32ff84 | 5121 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5122 | put_page(page); |
5123 | break; | |
02491447 DN |
5124 | case MC_TARGET_SWAP: |
5125 | ent = target.ent; | |
e91cbb42 | 5126 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5127 | mc.precharge--; |
483c30b5 DN |
5128 | /* we fixup refcnts and charges later. */ |
5129 | mc.moved_swap++; | |
5130 | } | |
02491447 | 5131 | break; |
4ffef5fe DN |
5132 | default: |
5133 | break; | |
5134 | } | |
5135 | } | |
5136 | pte_unmap_unlock(pte - 1, ptl); | |
5137 | cond_resched(); | |
5138 | ||
5139 | if (addr != end) { | |
5140 | /* | |
5141 | * We have consumed all precharges we got in can_attach(). | |
5142 | * We try charge one by one, but don't do any additional | |
5143 | * charges to mc.to if we have failed in charge once in attach() | |
5144 | * phase. | |
5145 | */ | |
854ffa8d | 5146 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5147 | if (!ret) |
5148 | goto retry; | |
5149 | } | |
5150 | ||
5151 | return ret; | |
5152 | } | |
5153 | ||
5154 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5155 | { | |
26bcd64a NH |
5156 | struct mm_walk mem_cgroup_move_charge_walk = { |
5157 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5158 | .mm = mm, | |
5159 | }; | |
4ffef5fe DN |
5160 | |
5161 | lru_add_drain_all(); | |
312722cb JW |
5162 | /* |
5163 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5164 | * move_lock while we're moving its pages to another memcg. | |
5165 | * Then wait for already started RCU-only updates to finish. | |
5166 | */ | |
5167 | atomic_inc(&mc.from->moving_account); | |
5168 | synchronize_rcu(); | |
dfe076b0 DN |
5169 | retry: |
5170 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5171 | /* | |
5172 | * Someone who are holding the mmap_sem might be waiting in | |
5173 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5174 | * and retry. Because we cancel precharges, we might not be able | |
5175 | * to move enough charges, but moving charge is a best-effort | |
5176 | * feature anyway, so it wouldn't be a big problem. | |
5177 | */ | |
5178 | __mem_cgroup_clear_mc(); | |
5179 | cond_resched(); | |
5180 | goto retry; | |
5181 | } | |
26bcd64a NH |
5182 | /* |
5183 | * When we have consumed all precharges and failed in doing | |
5184 | * additional charge, the page walk just aborts. | |
5185 | */ | |
5186 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 5187 | up_read(&mm->mmap_sem); |
312722cb | 5188 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5189 | } |
5190 | ||
eb95419b | 5191 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5192 | struct cgroup_taskset *tset) |
67e465a7 | 5193 | { |
2f7ee569 | 5194 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5195 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5196 | |
dfe076b0 | 5197 | if (mm) { |
a433658c KM |
5198 | if (mc.to) |
5199 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5200 | mmput(mm); |
5201 | } | |
a433658c KM |
5202 | if (mc.to) |
5203 | mem_cgroup_clear_mc(); | |
67e465a7 | 5204 | } |
5cfb80a7 | 5205 | #else /* !CONFIG_MMU */ |
eb95419b | 5206 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5207 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5208 | { |
5209 | return 0; | |
5210 | } | |
eb95419b | 5211 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5212 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5213 | { |
5214 | } | |
eb95419b | 5215 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5216 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5217 | { |
5218 | } | |
5219 | #endif | |
67e465a7 | 5220 | |
f00baae7 TH |
5221 | /* |
5222 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5223 | * to verify whether we're attached to the default hierarchy on each mount |
5224 | * attempt. | |
f00baae7 | 5225 | */ |
eb95419b | 5226 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5227 | { |
5228 | /* | |
aa6ec29b | 5229 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5230 | * guarantees that @root doesn't have any children, so turning it |
5231 | * on for the root memcg is enough. | |
5232 | */ | |
aa6ec29b | 5233 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 5234 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
5235 | } |
5236 | ||
241994ed JW |
5237 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5238 | struct cftype *cft) | |
5239 | { | |
5240 | return mem_cgroup_usage(mem_cgroup_from_css(css), false); | |
5241 | } | |
5242 | ||
5243 | static int memory_low_show(struct seq_file *m, void *v) | |
5244 | { | |
5245 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5246 | unsigned long low = ACCESS_ONCE(memcg->low); | |
5247 | ||
5248 | if (low == PAGE_COUNTER_MAX) | |
5249 | seq_puts(m, "infinity\n"); | |
5250 | else | |
5251 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5252 | ||
5253 | return 0; | |
5254 | } | |
5255 | ||
5256 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5257 | char *buf, size_t nbytes, loff_t off) | |
5258 | { | |
5259 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5260 | unsigned long low; | |
5261 | int err; | |
5262 | ||
5263 | buf = strstrip(buf); | |
5264 | err = page_counter_memparse(buf, "infinity", &low); | |
5265 | if (err) | |
5266 | return err; | |
5267 | ||
5268 | memcg->low = low; | |
5269 | ||
5270 | return nbytes; | |
5271 | } | |
5272 | ||
5273 | static int memory_high_show(struct seq_file *m, void *v) | |
5274 | { | |
5275 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5276 | unsigned long high = ACCESS_ONCE(memcg->high); | |
5277 | ||
5278 | if (high == PAGE_COUNTER_MAX) | |
5279 | seq_puts(m, "infinity\n"); | |
5280 | else | |
5281 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5282 | ||
5283 | return 0; | |
5284 | } | |
5285 | ||
5286 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5287 | char *buf, size_t nbytes, loff_t off) | |
5288 | { | |
5289 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5290 | unsigned long high; | |
5291 | int err; | |
5292 | ||
5293 | buf = strstrip(buf); | |
5294 | err = page_counter_memparse(buf, "infinity", &high); | |
5295 | if (err) | |
5296 | return err; | |
5297 | ||
5298 | memcg->high = high; | |
5299 | ||
5300 | return nbytes; | |
5301 | } | |
5302 | ||
5303 | static int memory_max_show(struct seq_file *m, void *v) | |
5304 | { | |
5305 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5306 | unsigned long max = ACCESS_ONCE(memcg->memory.limit); | |
5307 | ||
5308 | if (max == PAGE_COUNTER_MAX) | |
5309 | seq_puts(m, "infinity\n"); | |
5310 | else | |
5311 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5312 | ||
5313 | return 0; | |
5314 | } | |
5315 | ||
5316 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5317 | char *buf, size_t nbytes, loff_t off) | |
5318 | { | |
5319 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5320 | unsigned long max; | |
5321 | int err; | |
5322 | ||
5323 | buf = strstrip(buf); | |
5324 | err = page_counter_memparse(buf, "infinity", &max); | |
5325 | if (err) | |
5326 | return err; | |
5327 | ||
5328 | err = mem_cgroup_resize_limit(memcg, max); | |
5329 | if (err) | |
5330 | return err; | |
5331 | ||
5332 | return nbytes; | |
5333 | } | |
5334 | ||
5335 | static int memory_events_show(struct seq_file *m, void *v) | |
5336 | { | |
5337 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5338 | ||
5339 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5340 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5341 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5342 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5343 | ||
5344 | return 0; | |
5345 | } | |
5346 | ||
5347 | static struct cftype memory_files[] = { | |
5348 | { | |
5349 | .name = "current", | |
5350 | .read_u64 = memory_current_read, | |
5351 | }, | |
5352 | { | |
5353 | .name = "low", | |
5354 | .flags = CFTYPE_NOT_ON_ROOT, | |
5355 | .seq_show = memory_low_show, | |
5356 | .write = memory_low_write, | |
5357 | }, | |
5358 | { | |
5359 | .name = "high", | |
5360 | .flags = CFTYPE_NOT_ON_ROOT, | |
5361 | .seq_show = memory_high_show, | |
5362 | .write = memory_high_write, | |
5363 | }, | |
5364 | { | |
5365 | .name = "max", | |
5366 | .flags = CFTYPE_NOT_ON_ROOT, | |
5367 | .seq_show = memory_max_show, | |
5368 | .write = memory_max_write, | |
5369 | }, | |
5370 | { | |
5371 | .name = "events", | |
5372 | .flags = CFTYPE_NOT_ON_ROOT, | |
5373 | .seq_show = memory_events_show, | |
5374 | }, | |
5375 | { } /* terminate */ | |
5376 | }; | |
5377 | ||
073219e9 | 5378 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5379 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5380 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5381 | .css_offline = mem_cgroup_css_offline, |
5382 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5383 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5384 | .can_attach = mem_cgroup_can_attach, |
5385 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5386 | .attach = mem_cgroup_move_task, |
f00baae7 | 5387 | .bind = mem_cgroup_bind, |
241994ed JW |
5388 | .dfl_cftypes = memory_files, |
5389 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5390 | .early_init = 0, |
8cdea7c0 | 5391 | }; |
c077719b | 5392 | |
241994ed JW |
5393 | /** |
5394 | * mem_cgroup_events - count memory events against a cgroup | |
5395 | * @memcg: the memory cgroup | |
5396 | * @idx: the event index | |
5397 | * @nr: the number of events to account for | |
5398 | */ | |
5399 | void mem_cgroup_events(struct mem_cgroup *memcg, | |
5400 | enum mem_cgroup_events_index idx, | |
5401 | unsigned int nr) | |
5402 | { | |
5403 | this_cpu_add(memcg->stat->events[idx], nr); | |
5404 | } | |
5405 | ||
5406 | /** | |
5407 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5408 | * @root: the highest ancestor to consider | |
5409 | * @memcg: the memory cgroup to check | |
5410 | * | |
5411 | * Returns %true if memory consumption of @memcg, and that of all | |
5412 | * configurable ancestors up to @root, is below the normal range. | |
5413 | */ | |
5414 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5415 | { | |
5416 | if (mem_cgroup_disabled()) | |
5417 | return false; | |
5418 | ||
5419 | /* | |
5420 | * The toplevel group doesn't have a configurable range, so | |
5421 | * it's never low when looked at directly, and it is not | |
5422 | * considered an ancestor when assessing the hierarchy. | |
5423 | */ | |
5424 | ||
5425 | if (memcg == root_mem_cgroup) | |
5426 | return false; | |
5427 | ||
5428 | if (page_counter_read(&memcg->memory) > memcg->low) | |
5429 | return false; | |
5430 | ||
5431 | while (memcg != root) { | |
5432 | memcg = parent_mem_cgroup(memcg); | |
5433 | ||
5434 | if (memcg == root_mem_cgroup) | |
5435 | break; | |
5436 | ||
5437 | if (page_counter_read(&memcg->memory) > memcg->low) | |
5438 | return false; | |
5439 | } | |
5440 | return true; | |
5441 | } | |
5442 | ||
00501b53 JW |
5443 | /** |
5444 | * mem_cgroup_try_charge - try charging a page | |
5445 | * @page: page to charge | |
5446 | * @mm: mm context of the victim | |
5447 | * @gfp_mask: reclaim mode | |
5448 | * @memcgp: charged memcg return | |
5449 | * | |
5450 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5451 | * pages according to @gfp_mask if necessary. | |
5452 | * | |
5453 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5454 | * Otherwise, an error code is returned. | |
5455 | * | |
5456 | * After page->mapping has been set up, the caller must finalize the | |
5457 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5458 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5459 | */ | |
5460 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5461 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5462 | { | |
5463 | struct mem_cgroup *memcg = NULL; | |
5464 | unsigned int nr_pages = 1; | |
5465 | int ret = 0; | |
5466 | ||
5467 | if (mem_cgroup_disabled()) | |
5468 | goto out; | |
5469 | ||
5470 | if (PageSwapCache(page)) { | |
00501b53 JW |
5471 | /* |
5472 | * Every swap fault against a single page tries to charge the | |
5473 | * page, bail as early as possible. shmem_unuse() encounters | |
5474 | * already charged pages, too. The USED bit is protected by | |
5475 | * the page lock, which serializes swap cache removal, which | |
5476 | * in turn serializes uncharging. | |
5477 | */ | |
1306a85a | 5478 | if (page->mem_cgroup) |
00501b53 JW |
5479 | goto out; |
5480 | } | |
5481 | ||
5482 | if (PageTransHuge(page)) { | |
5483 | nr_pages <<= compound_order(page); | |
5484 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5485 | } | |
5486 | ||
5487 | if (do_swap_account && PageSwapCache(page)) | |
5488 | memcg = try_get_mem_cgroup_from_page(page); | |
5489 | if (!memcg) | |
5490 | memcg = get_mem_cgroup_from_mm(mm); | |
5491 | ||
5492 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5493 | ||
5494 | css_put(&memcg->css); | |
5495 | ||
5496 | if (ret == -EINTR) { | |
5497 | memcg = root_mem_cgroup; | |
5498 | ret = 0; | |
5499 | } | |
5500 | out: | |
5501 | *memcgp = memcg; | |
5502 | return ret; | |
5503 | } | |
5504 | ||
5505 | /** | |
5506 | * mem_cgroup_commit_charge - commit a page charge | |
5507 | * @page: page to charge | |
5508 | * @memcg: memcg to charge the page to | |
5509 | * @lrucare: page might be on LRU already | |
5510 | * | |
5511 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5512 | * after page->mapping has been set up. This must happen atomically | |
5513 | * as part of the page instantiation, i.e. under the page table lock | |
5514 | * for anonymous pages, under the page lock for page and swap cache. | |
5515 | * | |
5516 | * In addition, the page must not be on the LRU during the commit, to | |
5517 | * prevent racing with task migration. If it might be, use @lrucare. | |
5518 | * | |
5519 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5520 | */ | |
5521 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5522 | bool lrucare) | |
5523 | { | |
5524 | unsigned int nr_pages = 1; | |
5525 | ||
5526 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5527 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5528 | ||
5529 | if (mem_cgroup_disabled()) | |
5530 | return; | |
5531 | /* | |
5532 | * Swap faults will attempt to charge the same page multiple | |
5533 | * times. But reuse_swap_page() might have removed the page | |
5534 | * from swapcache already, so we can't check PageSwapCache(). | |
5535 | */ | |
5536 | if (!memcg) | |
5537 | return; | |
5538 | ||
6abb5a86 JW |
5539 | commit_charge(page, memcg, lrucare); |
5540 | ||
00501b53 JW |
5541 | if (PageTransHuge(page)) { |
5542 | nr_pages <<= compound_order(page); | |
5543 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5544 | } | |
5545 | ||
6abb5a86 JW |
5546 | local_irq_disable(); |
5547 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5548 | memcg_check_events(memcg, page); | |
5549 | local_irq_enable(); | |
00501b53 JW |
5550 | |
5551 | if (do_swap_account && PageSwapCache(page)) { | |
5552 | swp_entry_t entry = { .val = page_private(page) }; | |
5553 | /* | |
5554 | * The swap entry might not get freed for a long time, | |
5555 | * let's not wait for it. The page already received a | |
5556 | * memory+swap charge, drop the swap entry duplicate. | |
5557 | */ | |
5558 | mem_cgroup_uncharge_swap(entry); | |
5559 | } | |
5560 | } | |
5561 | ||
5562 | /** | |
5563 | * mem_cgroup_cancel_charge - cancel a page charge | |
5564 | * @page: page to charge | |
5565 | * @memcg: memcg to charge the page to | |
5566 | * | |
5567 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5568 | */ | |
5569 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5570 | { | |
5571 | unsigned int nr_pages = 1; | |
5572 | ||
5573 | if (mem_cgroup_disabled()) | |
5574 | return; | |
5575 | /* | |
5576 | * Swap faults will attempt to charge the same page multiple | |
5577 | * times. But reuse_swap_page() might have removed the page | |
5578 | * from swapcache already, so we can't check PageSwapCache(). | |
5579 | */ | |
5580 | if (!memcg) | |
5581 | return; | |
5582 | ||
5583 | if (PageTransHuge(page)) { | |
5584 | nr_pages <<= compound_order(page); | |
5585 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5586 | } | |
5587 | ||
5588 | cancel_charge(memcg, nr_pages); | |
5589 | } | |
5590 | ||
747db954 | 5591 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5592 | unsigned long nr_anon, unsigned long nr_file, |
5593 | unsigned long nr_huge, struct page *dummy_page) | |
5594 | { | |
18eca2e6 | 5595 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5596 | unsigned long flags; |
5597 | ||
ce00a967 | 5598 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5599 | page_counter_uncharge(&memcg->memory, nr_pages); |
5600 | if (do_swap_account) | |
5601 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5602 | memcg_oom_recover(memcg); |
5603 | } | |
747db954 JW |
5604 | |
5605 | local_irq_save(flags); | |
5606 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5607 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5608 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5609 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5610 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5611 | memcg_check_events(memcg, dummy_page); |
5612 | local_irq_restore(flags); | |
e8ea14cc JW |
5613 | |
5614 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5615 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5616 | } |
5617 | ||
5618 | static void uncharge_list(struct list_head *page_list) | |
5619 | { | |
5620 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5621 | unsigned long nr_anon = 0; |
5622 | unsigned long nr_file = 0; | |
5623 | unsigned long nr_huge = 0; | |
5624 | unsigned long pgpgout = 0; | |
747db954 JW |
5625 | struct list_head *next; |
5626 | struct page *page; | |
5627 | ||
5628 | next = page_list->next; | |
5629 | do { | |
5630 | unsigned int nr_pages = 1; | |
747db954 JW |
5631 | |
5632 | page = list_entry(next, struct page, lru); | |
5633 | next = page->lru.next; | |
5634 | ||
5635 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5636 | VM_BUG_ON_PAGE(page_count(page), page); | |
5637 | ||
1306a85a | 5638 | if (!page->mem_cgroup) |
747db954 JW |
5639 | continue; |
5640 | ||
5641 | /* | |
5642 | * Nobody should be changing or seriously looking at | |
1306a85a | 5643 | * page->mem_cgroup at this point, we have fully |
29833315 | 5644 | * exclusive access to the page. |
747db954 JW |
5645 | */ |
5646 | ||
1306a85a | 5647 | if (memcg != page->mem_cgroup) { |
747db954 | 5648 | if (memcg) { |
18eca2e6 JW |
5649 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5650 | nr_huge, page); | |
5651 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5652 | } |
1306a85a | 5653 | memcg = page->mem_cgroup; |
747db954 JW |
5654 | } |
5655 | ||
5656 | if (PageTransHuge(page)) { | |
5657 | nr_pages <<= compound_order(page); | |
5658 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5659 | nr_huge += nr_pages; | |
5660 | } | |
5661 | ||
5662 | if (PageAnon(page)) | |
5663 | nr_anon += nr_pages; | |
5664 | else | |
5665 | nr_file += nr_pages; | |
5666 | ||
1306a85a | 5667 | page->mem_cgroup = NULL; |
747db954 JW |
5668 | |
5669 | pgpgout++; | |
5670 | } while (next != page_list); | |
5671 | ||
5672 | if (memcg) | |
18eca2e6 JW |
5673 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5674 | nr_huge, page); | |
747db954 JW |
5675 | } |
5676 | ||
0a31bc97 JW |
5677 | /** |
5678 | * mem_cgroup_uncharge - uncharge a page | |
5679 | * @page: page to uncharge | |
5680 | * | |
5681 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5682 | * mem_cgroup_commit_charge(). | |
5683 | */ | |
5684 | void mem_cgroup_uncharge(struct page *page) | |
5685 | { | |
0a31bc97 JW |
5686 | if (mem_cgroup_disabled()) |
5687 | return; | |
5688 | ||
747db954 | 5689 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5690 | if (!page->mem_cgroup) |
0a31bc97 JW |
5691 | return; |
5692 | ||
747db954 JW |
5693 | INIT_LIST_HEAD(&page->lru); |
5694 | uncharge_list(&page->lru); | |
5695 | } | |
0a31bc97 | 5696 | |
747db954 JW |
5697 | /** |
5698 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5699 | * @page_list: list of pages to uncharge | |
5700 | * | |
5701 | * Uncharge a list of pages previously charged with | |
5702 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5703 | */ | |
5704 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5705 | { | |
5706 | if (mem_cgroup_disabled()) | |
5707 | return; | |
0a31bc97 | 5708 | |
747db954 JW |
5709 | if (!list_empty(page_list)) |
5710 | uncharge_list(page_list); | |
0a31bc97 JW |
5711 | } |
5712 | ||
5713 | /** | |
5714 | * mem_cgroup_migrate - migrate a charge to another page | |
5715 | * @oldpage: currently charged page | |
5716 | * @newpage: page to transfer the charge to | |
f5e03a49 | 5717 | * @lrucare: either or both pages might be on the LRU already |
0a31bc97 JW |
5718 | * |
5719 | * Migrate the charge from @oldpage to @newpage. | |
5720 | * | |
5721 | * Both pages must be locked, @newpage->mapping must be set up. | |
5722 | */ | |
5723 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
5724 | bool lrucare) | |
5725 | { | |
29833315 | 5726 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5727 | int isolated; |
5728 | ||
5729 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5730 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
5731 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
5732 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
5733 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
5734 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5735 | newpage); | |
0a31bc97 JW |
5736 | |
5737 | if (mem_cgroup_disabled()) | |
5738 | return; | |
5739 | ||
5740 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5741 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5742 | return; |
5743 | ||
7d5e3245 JW |
5744 | /* |
5745 | * Swapcache readahead pages can get migrated before being | |
5746 | * charged, and migration from compaction can happen to an | |
5747 | * uncharged page when the PFN walker finds a page that | |
5748 | * reclaim just put back on the LRU but has not released yet. | |
5749 | */ | |
1306a85a | 5750 | memcg = oldpage->mem_cgroup; |
29833315 | 5751 | if (!memcg) |
0a31bc97 JW |
5752 | return; |
5753 | ||
0a31bc97 JW |
5754 | if (lrucare) |
5755 | lock_page_lru(oldpage, &isolated); | |
5756 | ||
1306a85a | 5757 | oldpage->mem_cgroup = NULL; |
0a31bc97 JW |
5758 | |
5759 | if (lrucare) | |
5760 | unlock_page_lru(oldpage, isolated); | |
5761 | ||
29833315 | 5762 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
5763 | } |
5764 | ||
2d11085e | 5765 | /* |
1081312f MH |
5766 | * subsys_initcall() for memory controller. |
5767 | * | |
5768 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5769 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5770 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5771 | * be initialized from here. | |
2d11085e MH |
5772 | */ |
5773 | static int __init mem_cgroup_init(void) | |
5774 | { | |
95a045f6 JW |
5775 | int cpu, node; |
5776 | ||
2d11085e | 5777 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5778 | |
5779 | for_each_possible_cpu(cpu) | |
5780 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5781 | drain_local_stock); | |
5782 | ||
5783 | for_each_node(node) { | |
5784 | struct mem_cgroup_tree_per_node *rtpn; | |
5785 | int zone; | |
5786 | ||
5787 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5788 | node_online(node) ? node : NUMA_NO_NODE); | |
5789 | ||
5790 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5791 | struct mem_cgroup_tree_per_zone *rtpz; | |
5792 | ||
5793 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5794 | rtpz->rb_root = RB_ROOT; | |
5795 | spin_lock_init(&rtpz->lock); | |
5796 | } | |
5797 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5798 | } | |
5799 | ||
2d11085e MH |
5800 | return 0; |
5801 | } | |
5802 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5803 | |
5804 | #ifdef CONFIG_MEMCG_SWAP | |
5805 | /** | |
5806 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5807 | * @page: page whose memsw charge to transfer | |
5808 | * @entry: swap entry to move the charge to | |
5809 | * | |
5810 | * Transfer the memsw charge of @page to @entry. | |
5811 | */ | |
5812 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5813 | { | |
5814 | struct mem_cgroup *memcg; | |
5815 | unsigned short oldid; | |
5816 | ||
5817 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5818 | VM_BUG_ON_PAGE(page_count(page), page); | |
5819 | ||
5820 | if (!do_swap_account) | |
5821 | return; | |
5822 | ||
5823 | memcg = page->mem_cgroup; | |
5824 | ||
5825 | /* Readahead page, never charged */ | |
5826 | if (!memcg) | |
5827 | return; | |
5828 | ||
5829 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5830 | VM_BUG_ON_PAGE(oldid, page); | |
5831 | mem_cgroup_swap_statistics(memcg, true); | |
5832 | ||
5833 | page->mem_cgroup = NULL; | |
5834 | ||
5835 | if (!mem_cgroup_is_root(memcg)) | |
5836 | page_counter_uncharge(&memcg->memory, 1); | |
5837 | ||
5838 | /* XXX: caller holds IRQ-safe mapping->tree_lock */ | |
5839 | VM_BUG_ON(!irqs_disabled()); | |
5840 | ||
5841 | mem_cgroup_charge_statistics(memcg, page, -1); | |
5842 | memcg_check_events(memcg, page); | |
5843 | } | |
5844 | ||
5845 | /** | |
5846 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5847 | * @entry: swap entry to uncharge | |
5848 | * | |
5849 | * Drop the memsw charge associated with @entry. | |
5850 | */ | |
5851 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5852 | { | |
5853 | struct mem_cgroup *memcg; | |
5854 | unsigned short id; | |
5855 | ||
5856 | if (!do_swap_account) | |
5857 | return; | |
5858 | ||
5859 | id = swap_cgroup_record(entry, 0); | |
5860 | rcu_read_lock(); | |
5861 | memcg = mem_cgroup_lookup(id); | |
5862 | if (memcg) { | |
5863 | if (!mem_cgroup_is_root(memcg)) | |
5864 | page_counter_uncharge(&memcg->memsw, 1); | |
5865 | mem_cgroup_swap_statistics(memcg, false); | |
5866 | css_put(&memcg->css); | |
5867 | } | |
5868 | rcu_read_unlock(); | |
5869 | } | |
5870 | ||
5871 | /* for remember boot option*/ | |
5872 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5873 | static int really_do_swap_account __initdata = 1; | |
5874 | #else | |
5875 | static int really_do_swap_account __initdata; | |
5876 | #endif | |
5877 | ||
5878 | static int __init enable_swap_account(char *s) | |
5879 | { | |
5880 | if (!strcmp(s, "1")) | |
5881 | really_do_swap_account = 1; | |
5882 | else if (!strcmp(s, "0")) | |
5883 | really_do_swap_account = 0; | |
5884 | return 1; | |
5885 | } | |
5886 | __setup("swapaccount=", enable_swap_account); | |
5887 | ||
5888 | static struct cftype memsw_cgroup_files[] = { | |
5889 | { | |
5890 | .name = "memsw.usage_in_bytes", | |
5891 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5892 | .read_u64 = mem_cgroup_read_u64, | |
5893 | }, | |
5894 | { | |
5895 | .name = "memsw.max_usage_in_bytes", | |
5896 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5897 | .write = mem_cgroup_reset, | |
5898 | .read_u64 = mem_cgroup_read_u64, | |
5899 | }, | |
5900 | { | |
5901 | .name = "memsw.limit_in_bytes", | |
5902 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5903 | .write = mem_cgroup_write, | |
5904 | .read_u64 = mem_cgroup_read_u64, | |
5905 | }, | |
5906 | { | |
5907 | .name = "memsw.failcnt", | |
5908 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5909 | .write = mem_cgroup_reset, | |
5910 | .read_u64 = mem_cgroup_read_u64, | |
5911 | }, | |
5912 | { }, /* terminate */ | |
5913 | }; | |
5914 | ||
5915 | static int __init mem_cgroup_swap_init(void) | |
5916 | { | |
5917 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5918 | do_swap_account = 1; | |
5919 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
5920 | memsw_cgroup_files)); | |
5921 | } | |
5922 | return 0; | |
5923 | } | |
5924 | subsys_initcall(mem_cgroup_swap_init); | |
5925 | ||
5926 | #endif /* CONFIG_MEMCG_SWAP */ |