Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
6e84f315 | 38 | #include <linux/sched/mm.h> |
3a4f8a0b | 39 | #include <linux/shmem_fs.h> |
4ffef5fe | 40 | #include <linux/hugetlb.h> |
d13d1443 | 41 | #include <linux/pagemap.h> |
d52aa412 | 42 | #include <linux/smp.h> |
8a9f3ccd | 43 | #include <linux/page-flags.h> |
66e1707b | 44 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
45 | #include <linux/bit_spinlock.h> |
46 | #include <linux/rcupdate.h> | |
e222432b | 47 | #include <linux/limits.h> |
b9e15baf | 48 | #include <linux/export.h> |
8c7c6e34 | 49 | #include <linux/mutex.h> |
bb4cc1a8 | 50 | #include <linux/rbtree.h> |
b6ac57d5 | 51 | #include <linux/slab.h> |
66e1707b | 52 | #include <linux/swap.h> |
02491447 | 53 | #include <linux/swapops.h> |
66e1707b | 54 | #include <linux/spinlock.h> |
2e72b634 | 55 | #include <linux/eventfd.h> |
79bd9814 | 56 | #include <linux/poll.h> |
2e72b634 | 57 | #include <linux/sort.h> |
66e1707b | 58 | #include <linux/fs.h> |
d2ceb9b7 | 59 | #include <linux/seq_file.h> |
70ddf637 | 60 | #include <linux/vmpressure.h> |
b69408e8 | 61 | #include <linux/mm_inline.h> |
5d1ea48b | 62 | #include <linux/swap_cgroup.h> |
cdec2e42 | 63 | #include <linux/cpu.h> |
158e0a2d | 64 | #include <linux/oom.h> |
0056f4e6 | 65 | #include <linux/lockdep.h> |
79bd9814 | 66 | #include <linux/file.h> |
b23afb93 | 67 | #include <linux/tracehook.h> |
08e552c6 | 68 | #include "internal.h" |
d1a4c0b3 | 69 | #include <net/sock.h> |
4bd2c1ee | 70 | #include <net/ip.h> |
f35c3a8e | 71 | #include "slab.h" |
8cdea7c0 | 72 | |
7c0f6ba6 | 73 | #include <linux/uaccess.h> |
8697d331 | 74 | |
cc8e970c KM |
75 | #include <trace/events/vmscan.h> |
76 | ||
073219e9 TH |
77 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
78 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 79 | |
7d828602 JW |
80 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
81 | ||
a181b0e8 | 82 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 83 | |
f7e1cb6e JW |
84 | /* Socket memory accounting disabled? */ |
85 | static bool cgroup_memory_nosocket; | |
86 | ||
04823c83 VD |
87 | /* Kernel memory accounting disabled? */ |
88 | static bool cgroup_memory_nokmem; | |
89 | ||
21afa38e | 90 | /* Whether the swap controller is active */ |
c255a458 | 91 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 92 | int do_swap_account __read_mostly; |
c077719b | 93 | #else |
a0db00fc | 94 | #define do_swap_account 0 |
c077719b KH |
95 | #endif |
96 | ||
7941d214 JW |
97 | /* Whether legacy memory+swap accounting is active */ |
98 | static bool do_memsw_account(void) | |
99 | { | |
100 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
101 | } | |
102 | ||
af7c4b0e JW |
103 | static const char * const mem_cgroup_stat_names[] = { |
104 | "cache", | |
105 | "rss", | |
b070e65c | 106 | "rss_huge", |
af7c4b0e | 107 | "mapped_file", |
c4843a75 | 108 | "dirty", |
3ea67d06 | 109 | "writeback", |
af7c4b0e JW |
110 | "swap", |
111 | }; | |
112 | ||
af7c4b0e JW |
113 | static const char * const mem_cgroup_events_names[] = { |
114 | "pgpgin", | |
115 | "pgpgout", | |
116 | "pgfault", | |
117 | "pgmajfault", | |
118 | }; | |
119 | ||
58cf188e SZ |
120 | static const char * const mem_cgroup_lru_names[] = { |
121 | "inactive_anon", | |
122 | "active_anon", | |
123 | "inactive_file", | |
124 | "active_file", | |
125 | "unevictable", | |
126 | }; | |
127 | ||
a0db00fc KS |
128 | #define THRESHOLDS_EVENTS_TARGET 128 |
129 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
130 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 131 | |
bb4cc1a8 AM |
132 | /* |
133 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
134 | * their hierarchy representation | |
135 | */ | |
136 | ||
ef8f2327 | 137 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 AM |
138 | struct rb_root rb_root; |
139 | spinlock_t lock; | |
140 | }; | |
141 | ||
bb4cc1a8 AM |
142 | struct mem_cgroup_tree { |
143 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
144 | }; | |
145 | ||
146 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
147 | ||
9490ff27 KH |
148 | /* for OOM */ |
149 | struct mem_cgroup_eventfd_list { | |
150 | struct list_head list; | |
151 | struct eventfd_ctx *eventfd; | |
152 | }; | |
2e72b634 | 153 | |
79bd9814 TH |
154 | /* |
155 | * cgroup_event represents events which userspace want to receive. | |
156 | */ | |
3bc942f3 | 157 | struct mem_cgroup_event { |
79bd9814 | 158 | /* |
59b6f873 | 159 | * memcg which the event belongs to. |
79bd9814 | 160 | */ |
59b6f873 | 161 | struct mem_cgroup *memcg; |
79bd9814 TH |
162 | /* |
163 | * eventfd to signal userspace about the event. | |
164 | */ | |
165 | struct eventfd_ctx *eventfd; | |
166 | /* | |
167 | * Each of these stored in a list by the cgroup. | |
168 | */ | |
169 | struct list_head list; | |
fba94807 TH |
170 | /* |
171 | * register_event() callback will be used to add new userspace | |
172 | * waiter for changes related to this event. Use eventfd_signal() | |
173 | * on eventfd to send notification to userspace. | |
174 | */ | |
59b6f873 | 175 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 176 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
177 | /* |
178 | * unregister_event() callback will be called when userspace closes | |
179 | * the eventfd or on cgroup removing. This callback must be set, | |
180 | * if you want provide notification functionality. | |
181 | */ | |
59b6f873 | 182 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 183 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
184 | /* |
185 | * All fields below needed to unregister event when | |
186 | * userspace closes eventfd. | |
187 | */ | |
188 | poll_table pt; | |
189 | wait_queue_head_t *wqh; | |
190 | wait_queue_t wait; | |
191 | struct work_struct remove; | |
192 | }; | |
193 | ||
c0ff4b85 R |
194 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
195 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 196 | |
7dc74be0 DN |
197 | /* Stuffs for move charges at task migration. */ |
198 | /* | |
1dfab5ab | 199 | * Types of charges to be moved. |
7dc74be0 | 200 | */ |
1dfab5ab JW |
201 | #define MOVE_ANON 0x1U |
202 | #define MOVE_FILE 0x2U | |
203 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 204 | |
4ffef5fe DN |
205 | /* "mc" and its members are protected by cgroup_mutex */ |
206 | static struct move_charge_struct { | |
b1dd693e | 207 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 208 | struct mm_struct *mm; |
4ffef5fe DN |
209 | struct mem_cgroup *from; |
210 | struct mem_cgroup *to; | |
1dfab5ab | 211 | unsigned long flags; |
4ffef5fe | 212 | unsigned long precharge; |
854ffa8d | 213 | unsigned long moved_charge; |
483c30b5 | 214 | unsigned long moved_swap; |
8033b97c DN |
215 | struct task_struct *moving_task; /* a task moving charges */ |
216 | wait_queue_head_t waitq; /* a waitq for other context */ | |
217 | } mc = { | |
2bd9bb20 | 218 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
219 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
220 | }; | |
4ffef5fe | 221 | |
4e416953 BS |
222 | /* |
223 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
224 | * limit reclaim to prevent infinite loops, if they ever occur. | |
225 | */ | |
a0db00fc | 226 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 227 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 228 | |
217bc319 KH |
229 | enum charge_type { |
230 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 231 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 232 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 233 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
234 | NR_CHARGE_TYPE, |
235 | }; | |
236 | ||
8c7c6e34 | 237 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
238 | enum res_type { |
239 | _MEM, | |
240 | _MEMSWAP, | |
241 | _OOM_TYPE, | |
510fc4e1 | 242 | _KMEM, |
d55f90bf | 243 | _TCP, |
86ae53e1 GC |
244 | }; |
245 | ||
a0db00fc KS |
246 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
247 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 248 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
249 | /* Used for OOM nofiier */ |
250 | #define OOM_CONTROL (0) | |
8c7c6e34 | 251 | |
70ddf637 AV |
252 | /* Some nice accessors for the vmpressure. */ |
253 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
254 | { | |
255 | if (!memcg) | |
256 | memcg = root_mem_cgroup; | |
257 | return &memcg->vmpressure; | |
258 | } | |
259 | ||
260 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
261 | { | |
262 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
263 | } | |
264 | ||
7ffc0edc MH |
265 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
266 | { | |
267 | return (memcg == root_mem_cgroup); | |
268 | } | |
269 | ||
127424c8 | 270 | #ifndef CONFIG_SLOB |
55007d84 | 271 | /* |
f7ce3190 | 272 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
273 | * The main reason for not using cgroup id for this: |
274 | * this works better in sparse environments, where we have a lot of memcgs, | |
275 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
276 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
277 | * 200 entry array for that. | |
55007d84 | 278 | * |
dbcf73e2 VD |
279 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
280 | * will double each time we have to increase it. | |
55007d84 | 281 | */ |
dbcf73e2 VD |
282 | static DEFINE_IDA(memcg_cache_ida); |
283 | int memcg_nr_cache_ids; | |
749c5415 | 284 | |
05257a1a VD |
285 | /* Protects memcg_nr_cache_ids */ |
286 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
287 | ||
288 | void memcg_get_cache_ids(void) | |
289 | { | |
290 | down_read(&memcg_cache_ids_sem); | |
291 | } | |
292 | ||
293 | void memcg_put_cache_ids(void) | |
294 | { | |
295 | up_read(&memcg_cache_ids_sem); | |
296 | } | |
297 | ||
55007d84 GC |
298 | /* |
299 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
300 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
301 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
302 | * tunable, but that is strictly not necessary. | |
303 | * | |
b8627835 | 304 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
305 | * this constant directly from cgroup, but it is understandable that this is |
306 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 307 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
308 | * increase ours as well if it increases. |
309 | */ | |
310 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 311 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 312 | |
d7f25f8a GC |
313 | /* |
314 | * A lot of the calls to the cache allocation functions are expected to be | |
315 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
316 | * conditional to this static branch, we'll have to allow modules that does | |
317 | * kmem_cache_alloc and the such to see this symbol as well | |
318 | */ | |
ef12947c | 319 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 320 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 321 | |
17cc4dfe TH |
322 | struct workqueue_struct *memcg_kmem_cache_wq; |
323 | ||
127424c8 | 324 | #endif /* !CONFIG_SLOB */ |
a8964b9b | 325 | |
ad7fa852 TH |
326 | /** |
327 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
328 | * @page: page of interest | |
329 | * | |
330 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
331 | * with @page is returned. The returned css remains associated with @page | |
332 | * until it is released. | |
333 | * | |
334 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
335 | * is returned. | |
ad7fa852 TH |
336 | */ |
337 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
338 | { | |
339 | struct mem_cgroup *memcg; | |
340 | ||
ad7fa852 TH |
341 | memcg = page->mem_cgroup; |
342 | ||
9e10a130 | 343 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
344 | memcg = root_mem_cgroup; |
345 | ||
ad7fa852 TH |
346 | return &memcg->css; |
347 | } | |
348 | ||
2fc04524 VD |
349 | /** |
350 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
351 | * @page: the page | |
352 | * | |
353 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
354 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
355 | * is safe to call this function without holding a reference to @page. | |
356 | * | |
357 | * Note, this function is inherently racy, because there is nothing to prevent | |
358 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
359 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
360 | * do not care (such as procfs interfaces). | |
361 | */ | |
362 | ino_t page_cgroup_ino(struct page *page) | |
363 | { | |
364 | struct mem_cgroup *memcg; | |
365 | unsigned long ino = 0; | |
366 | ||
367 | rcu_read_lock(); | |
368 | memcg = READ_ONCE(page->mem_cgroup); | |
369 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
370 | memcg = parent_mem_cgroup(memcg); | |
371 | if (memcg) | |
372 | ino = cgroup_ino(memcg->css.cgroup); | |
373 | rcu_read_unlock(); | |
374 | return ino; | |
375 | } | |
376 | ||
ef8f2327 MG |
377 | static struct mem_cgroup_per_node * |
378 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 379 | { |
97a6c37b | 380 | int nid = page_to_nid(page); |
f64c3f54 | 381 | |
ef8f2327 | 382 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
383 | } |
384 | ||
ef8f2327 MG |
385 | static struct mem_cgroup_tree_per_node * |
386 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 387 | { |
ef8f2327 | 388 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
389 | } |
390 | ||
ef8f2327 | 391 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
392 | soft_limit_tree_from_page(struct page *page) |
393 | { | |
394 | int nid = page_to_nid(page); | |
bb4cc1a8 | 395 | |
ef8f2327 | 396 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
397 | } |
398 | ||
ef8f2327 MG |
399 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
400 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 401 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
402 | { |
403 | struct rb_node **p = &mctz->rb_root.rb_node; | |
404 | struct rb_node *parent = NULL; | |
ef8f2327 | 405 | struct mem_cgroup_per_node *mz_node; |
bb4cc1a8 AM |
406 | |
407 | if (mz->on_tree) | |
408 | return; | |
409 | ||
410 | mz->usage_in_excess = new_usage_in_excess; | |
411 | if (!mz->usage_in_excess) | |
412 | return; | |
413 | while (*p) { | |
414 | parent = *p; | |
ef8f2327 | 415 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 AM |
416 | tree_node); |
417 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
418 | p = &(*p)->rb_left; | |
419 | /* | |
420 | * We can't avoid mem cgroups that are over their soft | |
421 | * limit by the same amount | |
422 | */ | |
423 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
424 | p = &(*p)->rb_right; | |
425 | } | |
426 | rb_link_node(&mz->tree_node, parent, p); | |
427 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
428 | mz->on_tree = true; | |
429 | } | |
430 | ||
ef8f2327 MG |
431 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
432 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
433 | { |
434 | if (!mz->on_tree) | |
435 | return; | |
436 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
437 | mz->on_tree = false; | |
438 | } | |
439 | ||
ef8f2327 MG |
440 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
441 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 442 | { |
0a31bc97 JW |
443 | unsigned long flags; |
444 | ||
445 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 446 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 447 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
448 | } |
449 | ||
3e32cb2e JW |
450 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
451 | { | |
452 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 453 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
454 | unsigned long excess = 0; |
455 | ||
456 | if (nr_pages > soft_limit) | |
457 | excess = nr_pages - soft_limit; | |
458 | ||
459 | return excess; | |
460 | } | |
bb4cc1a8 AM |
461 | |
462 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
463 | { | |
3e32cb2e | 464 | unsigned long excess; |
ef8f2327 MG |
465 | struct mem_cgroup_per_node *mz; |
466 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 467 | |
e231875b | 468 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
469 | if (!mctz) |
470 | return; | |
bb4cc1a8 AM |
471 | /* |
472 | * Necessary to update all ancestors when hierarchy is used. | |
473 | * because their event counter is not touched. | |
474 | */ | |
475 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 476 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 477 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
478 | /* |
479 | * We have to update the tree if mz is on RB-tree or | |
480 | * mem is over its softlimit. | |
481 | */ | |
482 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
483 | unsigned long flags; |
484 | ||
485 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
486 | /* if on-tree, remove it */ |
487 | if (mz->on_tree) | |
cf2c8127 | 488 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
489 | /* |
490 | * Insert again. mz->usage_in_excess will be updated. | |
491 | * If excess is 0, no tree ops. | |
492 | */ | |
cf2c8127 | 493 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 494 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
495 | } |
496 | } | |
497 | } | |
498 | ||
499 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
500 | { | |
ef8f2327 MG |
501 | struct mem_cgroup_tree_per_node *mctz; |
502 | struct mem_cgroup_per_node *mz; | |
503 | int nid; | |
bb4cc1a8 | 504 | |
e231875b | 505 | for_each_node(nid) { |
ef8f2327 MG |
506 | mz = mem_cgroup_nodeinfo(memcg, nid); |
507 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
508 | if (mctz) |
509 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
510 | } |
511 | } | |
512 | ||
ef8f2327 MG |
513 | static struct mem_cgroup_per_node * |
514 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
515 | { |
516 | struct rb_node *rightmost = NULL; | |
ef8f2327 | 517 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
518 | |
519 | retry: | |
520 | mz = NULL; | |
521 | rightmost = rb_last(&mctz->rb_root); | |
522 | if (!rightmost) | |
523 | goto done; /* Nothing to reclaim from */ | |
524 | ||
ef8f2327 | 525 | mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node); |
bb4cc1a8 AM |
526 | /* |
527 | * Remove the node now but someone else can add it back, | |
528 | * we will to add it back at the end of reclaim to its correct | |
529 | * position in the tree. | |
530 | */ | |
cf2c8127 | 531 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 532 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 533 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
534 | goto retry; |
535 | done: | |
536 | return mz; | |
537 | } | |
538 | ||
ef8f2327 MG |
539 | static struct mem_cgroup_per_node * |
540 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 541 | { |
ef8f2327 | 542 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 543 | |
0a31bc97 | 544 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 545 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 546 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
547 | return mz; |
548 | } | |
549 | ||
711d3d2c | 550 | /* |
484ebb3b GT |
551 | * Return page count for single (non recursive) @memcg. |
552 | * | |
711d3d2c KH |
553 | * Implementation Note: reading percpu statistics for memcg. |
554 | * | |
555 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
556 | * synchronization to implement "quick" read. There are trade-off between | |
557 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 558 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
559 | * |
560 | * But this _read() function is used for user interface now. The user accounts | |
561 | * memory usage by memory cgroup and he _always_ requires exact value because | |
562 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
563 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
564 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
565 | * | |
566 | * If there are kernel internal actions which can make use of some not-exact | |
567 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 568 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
569 | * implemented. |
570 | */ | |
484ebb3b GT |
571 | static unsigned long |
572 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 573 | { |
7a159cc9 | 574 | long val = 0; |
c62b1a3b | 575 | int cpu; |
c62b1a3b | 576 | |
484ebb3b | 577 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 578 | for_each_possible_cpu(cpu) |
c0ff4b85 | 579 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
580 | /* |
581 | * Summing races with updates, so val may be negative. Avoid exposing | |
582 | * transient negative values. | |
583 | */ | |
584 | if (val < 0) | |
585 | val = 0; | |
c62b1a3b KH |
586 | return val; |
587 | } | |
588 | ||
c0ff4b85 | 589 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
590 | enum mem_cgroup_events_index idx) |
591 | { | |
592 | unsigned long val = 0; | |
593 | int cpu; | |
594 | ||
733a572e | 595 | for_each_possible_cpu(cpu) |
c0ff4b85 | 596 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
597 | return val; |
598 | } | |
599 | ||
c0ff4b85 | 600 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 601 | struct page *page, |
f627c2f5 | 602 | bool compound, int nr_pages) |
d52aa412 | 603 | { |
b2402857 KH |
604 | /* |
605 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
606 | * counted as CACHE even if it's on ANON LRU. | |
607 | */ | |
0a31bc97 | 608 | if (PageAnon(page)) |
b2402857 | 609 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 610 | nr_pages); |
d52aa412 | 611 | else |
b2402857 | 612 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 613 | nr_pages); |
55e462b0 | 614 | |
f627c2f5 KS |
615 | if (compound) { |
616 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
b070e65c DR |
617 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
618 | nr_pages); | |
f627c2f5 | 619 | } |
b070e65c | 620 | |
e401f176 KH |
621 | /* pagein of a big page is an event. So, ignore page size */ |
622 | if (nr_pages > 0) | |
c0ff4b85 | 623 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 624 | else { |
c0ff4b85 | 625 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
626 | nr_pages = -nr_pages; /* for event */ |
627 | } | |
e401f176 | 628 | |
13114716 | 629 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
630 | } |
631 | ||
0a6b76dd VD |
632 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
633 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 634 | { |
b4536f0c | 635 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
e231875b | 636 | unsigned long nr = 0; |
ef8f2327 | 637 | enum lru_list lru; |
889976db | 638 | |
e231875b | 639 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 640 | |
ef8f2327 MG |
641 | for_each_lru(lru) { |
642 | if (!(BIT(lru) & lru_mask)) | |
643 | continue; | |
b4536f0c | 644 | nr += mem_cgroup_get_lru_size(lruvec, lru); |
e231875b JZ |
645 | } |
646 | return nr; | |
889976db | 647 | } |
bb2a0de9 | 648 | |
c0ff4b85 | 649 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 650 | unsigned int lru_mask) |
6d12e2d8 | 651 | { |
e231875b | 652 | unsigned long nr = 0; |
889976db | 653 | int nid; |
6d12e2d8 | 654 | |
31aaea4a | 655 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
656 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
657 | return nr; | |
d52aa412 KH |
658 | } |
659 | ||
f53d7ce3 JW |
660 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
661 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
662 | { |
663 | unsigned long val, next; | |
664 | ||
13114716 | 665 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 666 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 667 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
668 | if ((long)next - (long)val < 0) { |
669 | switch (target) { | |
670 | case MEM_CGROUP_TARGET_THRESH: | |
671 | next = val + THRESHOLDS_EVENTS_TARGET; | |
672 | break; | |
bb4cc1a8 AM |
673 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
674 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
675 | break; | |
f53d7ce3 JW |
676 | case MEM_CGROUP_TARGET_NUMAINFO: |
677 | next = val + NUMAINFO_EVENTS_TARGET; | |
678 | break; | |
679 | default: | |
680 | break; | |
681 | } | |
682 | __this_cpu_write(memcg->stat->targets[target], next); | |
683 | return true; | |
7a159cc9 | 684 | } |
f53d7ce3 | 685 | return false; |
d2265e6f KH |
686 | } |
687 | ||
688 | /* | |
689 | * Check events in order. | |
690 | * | |
691 | */ | |
c0ff4b85 | 692 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
693 | { |
694 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
695 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
696 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 697 | bool do_softlimit; |
82b3f2a7 | 698 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 699 | |
bb4cc1a8 AM |
700 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
701 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
702 | #if MAX_NUMNODES > 1 |
703 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
704 | MEM_CGROUP_TARGET_NUMAINFO); | |
705 | #endif | |
c0ff4b85 | 706 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
707 | if (unlikely(do_softlimit)) |
708 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 709 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 710 | if (unlikely(do_numainfo)) |
c0ff4b85 | 711 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 712 | #endif |
0a31bc97 | 713 | } |
d2265e6f KH |
714 | } |
715 | ||
cf475ad2 | 716 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 717 | { |
31a78f23 BS |
718 | /* |
719 | * mm_update_next_owner() may clear mm->owner to NULL | |
720 | * if it races with swapoff, page migration, etc. | |
721 | * So this can be called with p == NULL. | |
722 | */ | |
723 | if (unlikely(!p)) | |
724 | return NULL; | |
725 | ||
073219e9 | 726 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 727 | } |
33398cf2 | 728 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 729 | |
df381975 | 730 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 731 | { |
c0ff4b85 | 732 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 733 | |
54595fe2 KH |
734 | rcu_read_lock(); |
735 | do { | |
6f6acb00 MH |
736 | /* |
737 | * Page cache insertions can happen withou an | |
738 | * actual mm context, e.g. during disk probing | |
739 | * on boot, loopback IO, acct() writes etc. | |
740 | */ | |
741 | if (unlikely(!mm)) | |
df381975 | 742 | memcg = root_mem_cgroup; |
6f6acb00 MH |
743 | else { |
744 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
745 | if (unlikely(!memcg)) | |
746 | memcg = root_mem_cgroup; | |
747 | } | |
ec903c0c | 748 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 749 | rcu_read_unlock(); |
c0ff4b85 | 750 | return memcg; |
54595fe2 KH |
751 | } |
752 | ||
5660048c JW |
753 | /** |
754 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
755 | * @root: hierarchy root | |
756 | * @prev: previously returned memcg, NULL on first invocation | |
757 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
758 | * | |
759 | * Returns references to children of the hierarchy below @root, or | |
760 | * @root itself, or %NULL after a full round-trip. | |
761 | * | |
762 | * Caller must pass the return value in @prev on subsequent | |
763 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
764 | * to cancel a hierarchy walk before the round-trip is complete. | |
765 | * | |
766 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
767 | * divide up the memcgs in the hierarchy among all concurrent | |
768 | * reclaimers operating on the same zone and priority. | |
769 | */ | |
694fbc0f | 770 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 771 | struct mem_cgroup *prev, |
694fbc0f | 772 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 773 | { |
33398cf2 | 774 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 775 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 776 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 777 | struct mem_cgroup *pos = NULL; |
711d3d2c | 778 | |
694fbc0f AM |
779 | if (mem_cgroup_disabled()) |
780 | return NULL; | |
5660048c | 781 | |
9f3a0d09 JW |
782 | if (!root) |
783 | root = root_mem_cgroup; | |
7d74b06f | 784 | |
9f3a0d09 | 785 | if (prev && !reclaim) |
5ac8fb31 | 786 | pos = prev; |
14067bb3 | 787 | |
9f3a0d09 JW |
788 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
789 | if (prev) | |
5ac8fb31 | 790 | goto out; |
694fbc0f | 791 | return root; |
9f3a0d09 | 792 | } |
14067bb3 | 793 | |
542f85f9 | 794 | rcu_read_lock(); |
5f578161 | 795 | |
5ac8fb31 | 796 | if (reclaim) { |
ef8f2327 | 797 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 798 | |
ef8f2327 | 799 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
800 | iter = &mz->iter[reclaim->priority]; |
801 | ||
802 | if (prev && reclaim->generation != iter->generation) | |
803 | goto out_unlock; | |
804 | ||
6df38689 | 805 | while (1) { |
4db0c3c2 | 806 | pos = READ_ONCE(iter->position); |
6df38689 VD |
807 | if (!pos || css_tryget(&pos->css)) |
808 | break; | |
5ac8fb31 | 809 | /* |
6df38689 VD |
810 | * css reference reached zero, so iter->position will |
811 | * be cleared by ->css_released. However, we should not | |
812 | * rely on this happening soon, because ->css_released | |
813 | * is called from a work queue, and by busy-waiting we | |
814 | * might block it. So we clear iter->position right | |
815 | * away. | |
5ac8fb31 | 816 | */ |
6df38689 VD |
817 | (void)cmpxchg(&iter->position, pos, NULL); |
818 | } | |
5ac8fb31 JW |
819 | } |
820 | ||
821 | if (pos) | |
822 | css = &pos->css; | |
823 | ||
824 | for (;;) { | |
825 | css = css_next_descendant_pre(css, &root->css); | |
826 | if (!css) { | |
827 | /* | |
828 | * Reclaimers share the hierarchy walk, and a | |
829 | * new one might jump in right at the end of | |
830 | * the hierarchy - make sure they see at least | |
831 | * one group and restart from the beginning. | |
832 | */ | |
833 | if (!prev) | |
834 | continue; | |
835 | break; | |
527a5ec9 | 836 | } |
7d74b06f | 837 | |
5ac8fb31 JW |
838 | /* |
839 | * Verify the css and acquire a reference. The root | |
840 | * is provided by the caller, so we know it's alive | |
841 | * and kicking, and don't take an extra reference. | |
842 | */ | |
843 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 844 | |
5ac8fb31 JW |
845 | if (css == &root->css) |
846 | break; | |
14067bb3 | 847 | |
0b8f73e1 JW |
848 | if (css_tryget(css)) |
849 | break; | |
9f3a0d09 | 850 | |
5ac8fb31 | 851 | memcg = NULL; |
9f3a0d09 | 852 | } |
5ac8fb31 JW |
853 | |
854 | if (reclaim) { | |
5ac8fb31 | 855 | /* |
6df38689 VD |
856 | * The position could have already been updated by a competing |
857 | * thread, so check that the value hasn't changed since we read | |
858 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 859 | */ |
6df38689 VD |
860 | (void)cmpxchg(&iter->position, pos, memcg); |
861 | ||
5ac8fb31 JW |
862 | if (pos) |
863 | css_put(&pos->css); | |
864 | ||
865 | if (!memcg) | |
866 | iter->generation++; | |
867 | else if (!prev) | |
868 | reclaim->generation = iter->generation; | |
9f3a0d09 | 869 | } |
5ac8fb31 | 870 | |
542f85f9 MH |
871 | out_unlock: |
872 | rcu_read_unlock(); | |
5ac8fb31 | 873 | out: |
c40046f3 MH |
874 | if (prev && prev != root) |
875 | css_put(&prev->css); | |
876 | ||
9f3a0d09 | 877 | return memcg; |
14067bb3 | 878 | } |
7d74b06f | 879 | |
5660048c JW |
880 | /** |
881 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
882 | * @root: hierarchy root | |
883 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
884 | */ | |
885 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
886 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
887 | { |
888 | if (!root) | |
889 | root = root_mem_cgroup; | |
890 | if (prev && prev != root) | |
891 | css_put(&prev->css); | |
892 | } | |
7d74b06f | 893 | |
6df38689 VD |
894 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
895 | { | |
896 | struct mem_cgroup *memcg = dead_memcg; | |
897 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
898 | struct mem_cgroup_per_node *mz; |
899 | int nid; | |
6df38689 VD |
900 | int i; |
901 | ||
902 | while ((memcg = parent_mem_cgroup(memcg))) { | |
903 | for_each_node(nid) { | |
ef8f2327 MG |
904 | mz = mem_cgroup_nodeinfo(memcg, nid); |
905 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
906 | iter = &mz->iter[i]; | |
907 | cmpxchg(&iter->position, | |
908 | dead_memcg, NULL); | |
6df38689 VD |
909 | } |
910 | } | |
911 | } | |
912 | } | |
913 | ||
9f3a0d09 JW |
914 | /* |
915 | * Iteration constructs for visiting all cgroups (under a tree). If | |
916 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
917 | * be used for reference counting. | |
918 | */ | |
919 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 920 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 921 | iter != NULL; \ |
527a5ec9 | 922 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 923 | |
9f3a0d09 | 924 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 925 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 926 | iter != NULL; \ |
527a5ec9 | 927 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 928 | |
7c5f64f8 VD |
929 | /** |
930 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
931 | * @memcg: hierarchy root | |
932 | * @fn: function to call for each task | |
933 | * @arg: argument passed to @fn | |
934 | * | |
935 | * This function iterates over tasks attached to @memcg or to any of its | |
936 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
937 | * value, the function breaks the iteration loop and returns the value. | |
938 | * Otherwise, it will iterate over all tasks and return 0. | |
939 | * | |
940 | * This function must not be called for the root memory cgroup. | |
941 | */ | |
942 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
943 | int (*fn)(struct task_struct *, void *), void *arg) | |
944 | { | |
945 | struct mem_cgroup *iter; | |
946 | int ret = 0; | |
947 | ||
948 | BUG_ON(memcg == root_mem_cgroup); | |
949 | ||
950 | for_each_mem_cgroup_tree(iter, memcg) { | |
951 | struct css_task_iter it; | |
952 | struct task_struct *task; | |
953 | ||
954 | css_task_iter_start(&iter->css, &it); | |
955 | while (!ret && (task = css_task_iter_next(&it))) | |
956 | ret = fn(task, arg); | |
957 | css_task_iter_end(&it); | |
958 | if (ret) { | |
959 | mem_cgroup_iter_break(memcg, iter); | |
960 | break; | |
961 | } | |
962 | } | |
963 | return ret; | |
964 | } | |
965 | ||
925b7673 | 966 | /** |
dfe0e773 | 967 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 968 | * @page: the page |
fa9add64 | 969 | * @zone: zone of the page |
dfe0e773 JW |
970 | * |
971 | * This function is only safe when following the LRU page isolation | |
972 | * and putback protocol: the LRU lock must be held, and the page must | |
973 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 974 | */ |
599d0c95 | 975 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 976 | { |
ef8f2327 | 977 | struct mem_cgroup_per_node *mz; |
925b7673 | 978 | struct mem_cgroup *memcg; |
bea8c150 | 979 | struct lruvec *lruvec; |
6d12e2d8 | 980 | |
bea8c150 | 981 | if (mem_cgroup_disabled()) { |
599d0c95 | 982 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
983 | goto out; |
984 | } | |
925b7673 | 985 | |
1306a85a | 986 | memcg = page->mem_cgroup; |
7512102c | 987 | /* |
dfe0e773 | 988 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 989 | * possibly migrated - before they are charged. |
7512102c | 990 | */ |
29833315 JW |
991 | if (!memcg) |
992 | memcg = root_mem_cgroup; | |
7512102c | 993 | |
ef8f2327 | 994 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
995 | lruvec = &mz->lruvec; |
996 | out: | |
997 | /* | |
998 | * Since a node can be onlined after the mem_cgroup was created, | |
999 | * we have to be prepared to initialize lruvec->zone here; | |
1000 | * and if offlined then reonlined, we need to reinitialize it. | |
1001 | */ | |
599d0c95 MG |
1002 | if (unlikely(lruvec->pgdat != pgdat)) |
1003 | lruvec->pgdat = pgdat; | |
bea8c150 | 1004 | return lruvec; |
08e552c6 | 1005 | } |
b69408e8 | 1006 | |
925b7673 | 1007 | /** |
fa9add64 HD |
1008 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1009 | * @lruvec: mem_cgroup per zone lru vector | |
1010 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1011 | * @zid: zone id of the accounted pages |
fa9add64 | 1012 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1013 | * |
ca707239 HD |
1014 | * This function must be called under lru_lock, just before a page is added |
1015 | * to or just after a page is removed from an lru list (that ordering being | |
1016 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1017 | */ |
fa9add64 | 1018 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1019 | int zid, int nr_pages) |
3f58a829 | 1020 | { |
ef8f2327 | 1021 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1022 | unsigned long *lru_size; |
ca707239 | 1023 | long size; |
3f58a829 MK |
1024 | |
1025 | if (mem_cgroup_disabled()) | |
1026 | return; | |
1027 | ||
ef8f2327 | 1028 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1029 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1030 | |
1031 | if (nr_pages < 0) | |
1032 | *lru_size += nr_pages; | |
1033 | ||
1034 | size = *lru_size; | |
b4536f0c MH |
1035 | if (WARN_ONCE(size < 0, |
1036 | "%s(%p, %d, %d): lru_size %ld\n", | |
1037 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1038 | VM_BUG_ON(1); |
1039 | *lru_size = 0; | |
1040 | } | |
1041 | ||
1042 | if (nr_pages > 0) | |
1043 | *lru_size += nr_pages; | |
08e552c6 | 1044 | } |
544122e5 | 1045 | |
2314b42d | 1046 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1047 | { |
2314b42d | 1048 | struct mem_cgroup *task_memcg; |
158e0a2d | 1049 | struct task_struct *p; |
ffbdccf5 | 1050 | bool ret; |
4c4a2214 | 1051 | |
158e0a2d | 1052 | p = find_lock_task_mm(task); |
de077d22 | 1053 | if (p) { |
2314b42d | 1054 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1055 | task_unlock(p); |
1056 | } else { | |
1057 | /* | |
1058 | * All threads may have already detached their mm's, but the oom | |
1059 | * killer still needs to detect if they have already been oom | |
1060 | * killed to prevent needlessly killing additional tasks. | |
1061 | */ | |
ffbdccf5 | 1062 | rcu_read_lock(); |
2314b42d JW |
1063 | task_memcg = mem_cgroup_from_task(task); |
1064 | css_get(&task_memcg->css); | |
ffbdccf5 | 1065 | rcu_read_unlock(); |
de077d22 | 1066 | } |
2314b42d JW |
1067 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1068 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1069 | return ret; |
1070 | } | |
1071 | ||
19942822 | 1072 | /** |
9d11ea9f | 1073 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1074 | * @memcg: the memory cgroup |
19942822 | 1075 | * |
9d11ea9f | 1076 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1077 | * pages. |
19942822 | 1078 | */ |
c0ff4b85 | 1079 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1080 | { |
3e32cb2e JW |
1081 | unsigned long margin = 0; |
1082 | unsigned long count; | |
1083 | unsigned long limit; | |
9d11ea9f | 1084 | |
3e32cb2e | 1085 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1086 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1087 | if (count < limit) |
1088 | margin = limit - count; | |
1089 | ||
7941d214 | 1090 | if (do_memsw_account()) { |
3e32cb2e | 1091 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1092 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1093 | if (count <= limit) |
1094 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1095 | else |
1096 | margin = 0; | |
3e32cb2e JW |
1097 | } |
1098 | ||
1099 | return margin; | |
19942822 JW |
1100 | } |
1101 | ||
32047e2a | 1102 | /* |
bdcbb659 | 1103 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1104 | * |
bdcbb659 QH |
1105 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1106 | * moving cgroups. This is for waiting at high-memory pressure | |
1107 | * caused by "move". | |
32047e2a | 1108 | */ |
c0ff4b85 | 1109 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1110 | { |
2bd9bb20 KH |
1111 | struct mem_cgroup *from; |
1112 | struct mem_cgroup *to; | |
4b534334 | 1113 | bool ret = false; |
2bd9bb20 KH |
1114 | /* |
1115 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1116 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1117 | */ | |
1118 | spin_lock(&mc.lock); | |
1119 | from = mc.from; | |
1120 | to = mc.to; | |
1121 | if (!from) | |
1122 | goto unlock; | |
3e92041d | 1123 | |
2314b42d JW |
1124 | ret = mem_cgroup_is_descendant(from, memcg) || |
1125 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1126 | unlock: |
1127 | spin_unlock(&mc.lock); | |
4b534334 KH |
1128 | return ret; |
1129 | } | |
1130 | ||
c0ff4b85 | 1131 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1132 | { |
1133 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1134 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1135 | DEFINE_WAIT(wait); |
1136 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1137 | /* moving charge context might have finished. */ | |
1138 | if (mc.moving_task) | |
1139 | schedule(); | |
1140 | finish_wait(&mc.waitq, &wait); | |
1141 | return true; | |
1142 | } | |
1143 | } | |
1144 | return false; | |
1145 | } | |
1146 | ||
58cf188e | 1147 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1148 | /** |
58cf188e | 1149 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1150 | * @memcg: The memory cgroup that went over limit |
1151 | * @p: Task that is going to be killed | |
1152 | * | |
1153 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1154 | * enabled | |
1155 | */ | |
1156 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1157 | { | |
58cf188e SZ |
1158 | struct mem_cgroup *iter; |
1159 | unsigned int i; | |
e222432b | 1160 | |
e222432b BS |
1161 | rcu_read_lock(); |
1162 | ||
2415b9f5 BV |
1163 | if (p) { |
1164 | pr_info("Task in "); | |
1165 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1166 | pr_cont(" killed as a result of limit of "); | |
1167 | } else { | |
1168 | pr_info("Memory limit reached of cgroup "); | |
1169 | } | |
1170 | ||
e61734c5 | 1171 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1172 | pr_cont("\n"); |
e222432b | 1173 | |
e222432b BS |
1174 | rcu_read_unlock(); |
1175 | ||
3e32cb2e JW |
1176 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1177 | K((u64)page_counter_read(&memcg->memory)), | |
1178 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1179 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1180 | K((u64)page_counter_read(&memcg->memsw)), | |
1181 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1182 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1183 | K((u64)page_counter_read(&memcg->kmem)), | |
1184 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1185 | |
1186 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1187 | pr_info("Memory cgroup stats for "); |
1188 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1189 | pr_cont(":"); |
1190 | ||
1191 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
37e84351 | 1192 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
58cf188e | 1193 | continue; |
484ebb3b | 1194 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1195 | K(mem_cgroup_read_stat(iter, i))); |
1196 | } | |
1197 | ||
1198 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1199 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1200 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1201 | ||
1202 | pr_cont("\n"); | |
1203 | } | |
e222432b BS |
1204 | } |
1205 | ||
81d39c20 KH |
1206 | /* |
1207 | * This function returns the number of memcg under hierarchy tree. Returns | |
1208 | * 1(self count) if no children. | |
1209 | */ | |
c0ff4b85 | 1210 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1211 | { |
1212 | int num = 0; | |
7d74b06f KH |
1213 | struct mem_cgroup *iter; |
1214 | ||
c0ff4b85 | 1215 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1216 | num++; |
81d39c20 KH |
1217 | return num; |
1218 | } | |
1219 | ||
a63d83f4 DR |
1220 | /* |
1221 | * Return the memory (and swap, if configured) limit for a memcg. | |
1222 | */ | |
7c5f64f8 | 1223 | unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1224 | { |
3e32cb2e | 1225 | unsigned long limit; |
f3e8eb70 | 1226 | |
3e32cb2e | 1227 | limit = memcg->memory.limit; |
9a5a8f19 | 1228 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1229 | unsigned long memsw_limit; |
37e84351 | 1230 | unsigned long swap_limit; |
9a5a8f19 | 1231 | |
3e32cb2e | 1232 | memsw_limit = memcg->memsw.limit; |
37e84351 VD |
1233 | swap_limit = memcg->swap.limit; |
1234 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); | |
1235 | limit = min(limit + swap_limit, memsw_limit); | |
9a5a8f19 | 1236 | } |
9a5a8f19 | 1237 | return limit; |
a63d83f4 DR |
1238 | } |
1239 | ||
b6e6edcf | 1240 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1241 | int order) |
9cbb78bb | 1242 | { |
6e0fc46d DR |
1243 | struct oom_control oc = { |
1244 | .zonelist = NULL, | |
1245 | .nodemask = NULL, | |
2a966b77 | 1246 | .memcg = memcg, |
6e0fc46d DR |
1247 | .gfp_mask = gfp_mask, |
1248 | .order = order, | |
6e0fc46d | 1249 | }; |
7c5f64f8 | 1250 | bool ret; |
9cbb78bb | 1251 | |
dc56401f | 1252 | mutex_lock(&oom_lock); |
7c5f64f8 | 1253 | ret = out_of_memory(&oc); |
dc56401f | 1254 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1255 | return ret; |
9cbb78bb DR |
1256 | } |
1257 | ||
ae6e71d3 MC |
1258 | #if MAX_NUMNODES > 1 |
1259 | ||
4d0c066d KH |
1260 | /** |
1261 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1262 | * @memcg: the target memcg |
4d0c066d KH |
1263 | * @nid: the node ID to be checked. |
1264 | * @noswap : specify true here if the user wants flle only information. | |
1265 | * | |
1266 | * This function returns whether the specified memcg contains any | |
1267 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1268 | * pages in the node. | |
1269 | */ | |
c0ff4b85 | 1270 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1271 | int nid, bool noswap) |
1272 | { | |
c0ff4b85 | 1273 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1274 | return true; |
1275 | if (noswap || !total_swap_pages) | |
1276 | return false; | |
c0ff4b85 | 1277 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1278 | return true; |
1279 | return false; | |
1280 | ||
1281 | } | |
889976db YH |
1282 | |
1283 | /* | |
1284 | * Always updating the nodemask is not very good - even if we have an empty | |
1285 | * list or the wrong list here, we can start from some node and traverse all | |
1286 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1287 | * | |
1288 | */ | |
c0ff4b85 | 1289 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1290 | { |
1291 | int nid; | |
453a9bf3 KH |
1292 | /* |
1293 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1294 | * pagein/pageout changes since the last update. | |
1295 | */ | |
c0ff4b85 | 1296 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1297 | return; |
c0ff4b85 | 1298 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1299 | return; |
1300 | ||
889976db | 1301 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1302 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1303 | |
31aaea4a | 1304 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1305 | |
c0ff4b85 R |
1306 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1307 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1308 | } |
453a9bf3 | 1309 | |
c0ff4b85 R |
1310 | atomic_set(&memcg->numainfo_events, 0); |
1311 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1312 | } |
1313 | ||
1314 | /* | |
1315 | * Selecting a node where we start reclaim from. Because what we need is just | |
1316 | * reducing usage counter, start from anywhere is O,K. Considering | |
1317 | * memory reclaim from current node, there are pros. and cons. | |
1318 | * | |
1319 | * Freeing memory from current node means freeing memory from a node which | |
1320 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1321 | * hit limits, it will see a contention on a node. But freeing from remote | |
1322 | * node means more costs for memory reclaim because of memory latency. | |
1323 | * | |
1324 | * Now, we use round-robin. Better algorithm is welcomed. | |
1325 | */ | |
c0ff4b85 | 1326 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1327 | { |
1328 | int node; | |
1329 | ||
c0ff4b85 R |
1330 | mem_cgroup_may_update_nodemask(memcg); |
1331 | node = memcg->last_scanned_node; | |
889976db | 1332 | |
0edaf86c | 1333 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1334 | /* |
fda3d69b MH |
1335 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1336 | * last time it really checked all the LRUs due to rate limiting. | |
1337 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1338 | */ |
1339 | if (unlikely(node == MAX_NUMNODES)) | |
1340 | node = numa_node_id(); | |
1341 | ||
c0ff4b85 | 1342 | memcg->last_scanned_node = node; |
889976db YH |
1343 | return node; |
1344 | } | |
889976db | 1345 | #else |
c0ff4b85 | 1346 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1347 | { |
1348 | return 0; | |
1349 | } | |
1350 | #endif | |
1351 | ||
0608f43d | 1352 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1353 | pg_data_t *pgdat, |
0608f43d AM |
1354 | gfp_t gfp_mask, |
1355 | unsigned long *total_scanned) | |
1356 | { | |
1357 | struct mem_cgroup *victim = NULL; | |
1358 | int total = 0; | |
1359 | int loop = 0; | |
1360 | unsigned long excess; | |
1361 | unsigned long nr_scanned; | |
1362 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1363 | .pgdat = pgdat, |
0608f43d AM |
1364 | .priority = 0, |
1365 | }; | |
1366 | ||
3e32cb2e | 1367 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1368 | |
1369 | while (1) { | |
1370 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1371 | if (!victim) { | |
1372 | loop++; | |
1373 | if (loop >= 2) { | |
1374 | /* | |
1375 | * If we have not been able to reclaim | |
1376 | * anything, it might because there are | |
1377 | * no reclaimable pages under this hierarchy | |
1378 | */ | |
1379 | if (!total) | |
1380 | break; | |
1381 | /* | |
1382 | * We want to do more targeted reclaim. | |
1383 | * excess >> 2 is not to excessive so as to | |
1384 | * reclaim too much, nor too less that we keep | |
1385 | * coming back to reclaim from this cgroup | |
1386 | */ | |
1387 | if (total >= (excess >> 2) || | |
1388 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1389 | break; | |
1390 | } | |
1391 | continue; | |
1392 | } | |
a9dd0a83 | 1393 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1394 | pgdat, &nr_scanned); |
0608f43d | 1395 | *total_scanned += nr_scanned; |
3e32cb2e | 1396 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1397 | break; |
6d61ef40 | 1398 | } |
0608f43d AM |
1399 | mem_cgroup_iter_break(root_memcg, victim); |
1400 | return total; | |
6d61ef40 BS |
1401 | } |
1402 | ||
0056f4e6 JW |
1403 | #ifdef CONFIG_LOCKDEP |
1404 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1405 | .name = "memcg_oom_lock", | |
1406 | }; | |
1407 | #endif | |
1408 | ||
fb2a6fc5 JW |
1409 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1410 | ||
867578cb KH |
1411 | /* |
1412 | * Check OOM-Killer is already running under our hierarchy. | |
1413 | * If someone is running, return false. | |
1414 | */ | |
fb2a6fc5 | 1415 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1416 | { |
79dfdacc | 1417 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1418 | |
fb2a6fc5 JW |
1419 | spin_lock(&memcg_oom_lock); |
1420 | ||
9f3a0d09 | 1421 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1422 | if (iter->oom_lock) { |
79dfdacc MH |
1423 | /* |
1424 | * this subtree of our hierarchy is already locked | |
1425 | * so we cannot give a lock. | |
1426 | */ | |
79dfdacc | 1427 | failed = iter; |
9f3a0d09 JW |
1428 | mem_cgroup_iter_break(memcg, iter); |
1429 | break; | |
23751be0 JW |
1430 | } else |
1431 | iter->oom_lock = true; | |
7d74b06f | 1432 | } |
867578cb | 1433 | |
fb2a6fc5 JW |
1434 | if (failed) { |
1435 | /* | |
1436 | * OK, we failed to lock the whole subtree so we have | |
1437 | * to clean up what we set up to the failing subtree | |
1438 | */ | |
1439 | for_each_mem_cgroup_tree(iter, memcg) { | |
1440 | if (iter == failed) { | |
1441 | mem_cgroup_iter_break(memcg, iter); | |
1442 | break; | |
1443 | } | |
1444 | iter->oom_lock = false; | |
79dfdacc | 1445 | } |
0056f4e6 JW |
1446 | } else |
1447 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1448 | |
1449 | spin_unlock(&memcg_oom_lock); | |
1450 | ||
1451 | return !failed; | |
a636b327 | 1452 | } |
0b7f569e | 1453 | |
fb2a6fc5 | 1454 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1455 | { |
7d74b06f KH |
1456 | struct mem_cgroup *iter; |
1457 | ||
fb2a6fc5 | 1458 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1459 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1460 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1461 | iter->oom_lock = false; |
fb2a6fc5 | 1462 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1463 | } |
1464 | ||
c0ff4b85 | 1465 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1466 | { |
1467 | struct mem_cgroup *iter; | |
1468 | ||
c2b42d3c | 1469 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1470 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1471 | iter->under_oom++; |
1472 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1473 | } |
1474 | ||
c0ff4b85 | 1475 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1476 | { |
1477 | struct mem_cgroup *iter; | |
1478 | ||
867578cb KH |
1479 | /* |
1480 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1481 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1482 | */ |
c2b42d3c | 1483 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1484 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1485 | if (iter->under_oom > 0) |
1486 | iter->under_oom--; | |
1487 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1488 | } |
1489 | ||
867578cb KH |
1490 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1491 | ||
dc98df5a | 1492 | struct oom_wait_info { |
d79154bb | 1493 | struct mem_cgroup *memcg; |
dc98df5a KH |
1494 | wait_queue_t wait; |
1495 | }; | |
1496 | ||
1497 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1498 | unsigned mode, int sync, void *arg) | |
1499 | { | |
d79154bb HD |
1500 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1501 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1502 | struct oom_wait_info *oom_wait_info; |
1503 | ||
1504 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1505 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1506 | |
2314b42d JW |
1507 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1508 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1509 | return 0; |
dc98df5a KH |
1510 | return autoremove_wake_function(wait, mode, sync, arg); |
1511 | } | |
1512 | ||
c0ff4b85 | 1513 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1514 | { |
c2b42d3c TH |
1515 | /* |
1516 | * For the following lockless ->under_oom test, the only required | |
1517 | * guarantee is that it must see the state asserted by an OOM when | |
1518 | * this function is called as a result of userland actions | |
1519 | * triggered by the notification of the OOM. This is trivially | |
1520 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1521 | * triggering notification. | |
1522 | */ | |
1523 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1524 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1525 | } |
1526 | ||
3812c8c8 | 1527 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1528 | { |
d0db7afa | 1529 | if (!current->memcg_may_oom) |
3812c8c8 | 1530 | return; |
867578cb | 1531 | /* |
49426420 JW |
1532 | * We are in the middle of the charge context here, so we |
1533 | * don't want to block when potentially sitting on a callstack | |
1534 | * that holds all kinds of filesystem and mm locks. | |
1535 | * | |
1536 | * Also, the caller may handle a failed allocation gracefully | |
1537 | * (like optional page cache readahead) and so an OOM killer | |
1538 | * invocation might not even be necessary. | |
1539 | * | |
1540 | * That's why we don't do anything here except remember the | |
1541 | * OOM context and then deal with it at the end of the page | |
1542 | * fault when the stack is unwound, the locks are released, | |
1543 | * and when we know whether the fault was overall successful. | |
867578cb | 1544 | */ |
49426420 | 1545 | css_get(&memcg->css); |
626ebc41 TH |
1546 | current->memcg_in_oom = memcg; |
1547 | current->memcg_oom_gfp_mask = mask; | |
1548 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1549 | } |
1550 | ||
1551 | /** | |
1552 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1553 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1554 | * |
49426420 JW |
1555 | * This has to be called at the end of a page fault if the memcg OOM |
1556 | * handler was enabled. | |
3812c8c8 | 1557 | * |
49426420 | 1558 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1559 | * sleep on a waitqueue until the userspace task resolves the |
1560 | * situation. Sleeping directly in the charge context with all kinds | |
1561 | * of locks held is not a good idea, instead we remember an OOM state | |
1562 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1563 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1564 | * |
1565 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1566 | * completed, %false otherwise. |
3812c8c8 | 1567 | */ |
49426420 | 1568 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1569 | { |
626ebc41 | 1570 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1571 | struct oom_wait_info owait; |
49426420 | 1572 | bool locked; |
3812c8c8 JW |
1573 | |
1574 | /* OOM is global, do not handle */ | |
3812c8c8 | 1575 | if (!memcg) |
49426420 | 1576 | return false; |
3812c8c8 | 1577 | |
7c5f64f8 | 1578 | if (!handle) |
49426420 | 1579 | goto cleanup; |
3812c8c8 JW |
1580 | |
1581 | owait.memcg = memcg; | |
1582 | owait.wait.flags = 0; | |
1583 | owait.wait.func = memcg_oom_wake_function; | |
1584 | owait.wait.private = current; | |
1585 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1586 | |
3812c8c8 | 1587 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1588 | mem_cgroup_mark_under_oom(memcg); |
1589 | ||
1590 | locked = mem_cgroup_oom_trylock(memcg); | |
1591 | ||
1592 | if (locked) | |
1593 | mem_cgroup_oom_notify(memcg); | |
1594 | ||
1595 | if (locked && !memcg->oom_kill_disable) { | |
1596 | mem_cgroup_unmark_under_oom(memcg); | |
1597 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1598 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1599 | current->memcg_oom_order); | |
49426420 | 1600 | } else { |
3812c8c8 | 1601 | schedule(); |
49426420 JW |
1602 | mem_cgroup_unmark_under_oom(memcg); |
1603 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1604 | } | |
1605 | ||
1606 | if (locked) { | |
fb2a6fc5 JW |
1607 | mem_cgroup_oom_unlock(memcg); |
1608 | /* | |
1609 | * There is no guarantee that an OOM-lock contender | |
1610 | * sees the wakeups triggered by the OOM kill | |
1611 | * uncharges. Wake any sleepers explicitely. | |
1612 | */ | |
1613 | memcg_oom_recover(memcg); | |
1614 | } | |
49426420 | 1615 | cleanup: |
626ebc41 | 1616 | current->memcg_in_oom = NULL; |
3812c8c8 | 1617 | css_put(&memcg->css); |
867578cb | 1618 | return true; |
0b7f569e KH |
1619 | } |
1620 | ||
d7365e78 | 1621 | /** |
81f8c3a4 JW |
1622 | * lock_page_memcg - lock a page->mem_cgroup binding |
1623 | * @page: the page | |
32047e2a | 1624 | * |
81f8c3a4 JW |
1625 | * This function protects unlocked LRU pages from being moved to |
1626 | * another cgroup and stabilizes their page->mem_cgroup binding. | |
d69b042f | 1627 | */ |
62cccb8c | 1628 | void lock_page_memcg(struct page *page) |
89c06bd5 KH |
1629 | { |
1630 | struct mem_cgroup *memcg; | |
6de22619 | 1631 | unsigned long flags; |
89c06bd5 | 1632 | |
6de22619 JW |
1633 | /* |
1634 | * The RCU lock is held throughout the transaction. The fast | |
1635 | * path can get away without acquiring the memcg->move_lock | |
1636 | * because page moving starts with an RCU grace period. | |
6de22619 | 1637 | */ |
d7365e78 JW |
1638 | rcu_read_lock(); |
1639 | ||
1640 | if (mem_cgroup_disabled()) | |
62cccb8c | 1641 | return; |
89c06bd5 | 1642 | again: |
1306a85a | 1643 | memcg = page->mem_cgroup; |
29833315 | 1644 | if (unlikely(!memcg)) |
62cccb8c | 1645 | return; |
d7365e78 | 1646 | |
bdcbb659 | 1647 | if (atomic_read(&memcg->moving_account) <= 0) |
62cccb8c | 1648 | return; |
89c06bd5 | 1649 | |
6de22619 | 1650 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1651 | if (memcg != page->mem_cgroup) { |
6de22619 | 1652 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1653 | goto again; |
1654 | } | |
6de22619 JW |
1655 | |
1656 | /* | |
1657 | * When charge migration first begins, we can have locked and | |
1658 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1659 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1660 | */ |
1661 | memcg->move_lock_task = current; | |
1662 | memcg->move_lock_flags = flags; | |
d7365e78 | 1663 | |
62cccb8c | 1664 | return; |
89c06bd5 | 1665 | } |
81f8c3a4 | 1666 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1667 | |
d7365e78 | 1668 | /** |
81f8c3a4 | 1669 | * unlock_page_memcg - unlock a page->mem_cgroup binding |
62cccb8c | 1670 | * @page: the page |
d7365e78 | 1671 | */ |
62cccb8c | 1672 | void unlock_page_memcg(struct page *page) |
89c06bd5 | 1673 | { |
62cccb8c JW |
1674 | struct mem_cgroup *memcg = page->mem_cgroup; |
1675 | ||
6de22619 JW |
1676 | if (memcg && memcg->move_lock_task == current) { |
1677 | unsigned long flags = memcg->move_lock_flags; | |
1678 | ||
1679 | memcg->move_lock_task = NULL; | |
1680 | memcg->move_lock_flags = 0; | |
1681 | ||
1682 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1683 | } | |
89c06bd5 | 1684 | |
d7365e78 | 1685 | rcu_read_unlock(); |
89c06bd5 | 1686 | } |
81f8c3a4 | 1687 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1688 | |
cdec2e42 KH |
1689 | /* |
1690 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1691 | * TODO: maybe necessary to use big numbers in big irons. | |
1692 | */ | |
7ec99d62 | 1693 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1694 | struct memcg_stock_pcp { |
1695 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1696 | unsigned int nr_pages; |
cdec2e42 | 1697 | struct work_struct work; |
26fe6168 | 1698 | unsigned long flags; |
a0db00fc | 1699 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1700 | }; |
1701 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1702 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1703 | |
a0956d54 SS |
1704 | /** |
1705 | * consume_stock: Try to consume stocked charge on this cpu. | |
1706 | * @memcg: memcg to consume from. | |
1707 | * @nr_pages: how many pages to charge. | |
1708 | * | |
1709 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1710 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1711 | * service an allocation will refill the stock. | |
1712 | * | |
1713 | * returns true if successful, false otherwise. | |
cdec2e42 | 1714 | */ |
a0956d54 | 1715 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1716 | { |
1717 | struct memcg_stock_pcp *stock; | |
db2ba40c | 1718 | unsigned long flags; |
3e32cb2e | 1719 | bool ret = false; |
cdec2e42 | 1720 | |
a0956d54 | 1721 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1722 | return ret; |
a0956d54 | 1723 | |
db2ba40c JW |
1724 | local_irq_save(flags); |
1725 | ||
1726 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 1727 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1728 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1729 | ret = true; |
1730 | } | |
db2ba40c JW |
1731 | |
1732 | local_irq_restore(flags); | |
1733 | ||
cdec2e42 KH |
1734 | return ret; |
1735 | } | |
1736 | ||
1737 | /* | |
3e32cb2e | 1738 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1739 | */ |
1740 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1741 | { | |
1742 | struct mem_cgroup *old = stock->cached; | |
1743 | ||
11c9ea4e | 1744 | if (stock->nr_pages) { |
3e32cb2e | 1745 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1746 | if (do_memsw_account()) |
3e32cb2e | 1747 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1748 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1749 | stock->nr_pages = 0; |
cdec2e42 KH |
1750 | } |
1751 | stock->cached = NULL; | |
cdec2e42 KH |
1752 | } |
1753 | ||
cdec2e42 KH |
1754 | static void drain_local_stock(struct work_struct *dummy) |
1755 | { | |
db2ba40c JW |
1756 | struct memcg_stock_pcp *stock; |
1757 | unsigned long flags; | |
1758 | ||
1759 | local_irq_save(flags); | |
1760 | ||
1761 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 1762 | drain_stock(stock); |
26fe6168 | 1763 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
1764 | |
1765 | local_irq_restore(flags); | |
cdec2e42 KH |
1766 | } |
1767 | ||
1768 | /* | |
3e32cb2e | 1769 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1770 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1771 | */ |
c0ff4b85 | 1772 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 1773 | { |
db2ba40c JW |
1774 | struct memcg_stock_pcp *stock; |
1775 | unsigned long flags; | |
1776 | ||
1777 | local_irq_save(flags); | |
cdec2e42 | 1778 | |
db2ba40c | 1779 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 1780 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1781 | drain_stock(stock); |
c0ff4b85 | 1782 | stock->cached = memcg; |
cdec2e42 | 1783 | } |
11c9ea4e | 1784 | stock->nr_pages += nr_pages; |
db2ba40c JW |
1785 | |
1786 | local_irq_restore(flags); | |
cdec2e42 KH |
1787 | } |
1788 | ||
1789 | /* | |
c0ff4b85 | 1790 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1791 | * of the hierarchy under it. |
cdec2e42 | 1792 | */ |
6d3d6aa2 | 1793 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1794 | { |
26fe6168 | 1795 | int cpu, curcpu; |
d38144b7 | 1796 | |
6d3d6aa2 JW |
1797 | /* If someone's already draining, avoid adding running more workers. */ |
1798 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1799 | return; | |
cdec2e42 | 1800 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1801 | get_online_cpus(); |
5af12d0e | 1802 | curcpu = get_cpu(); |
cdec2e42 KH |
1803 | for_each_online_cpu(cpu) { |
1804 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1805 | struct mem_cgroup *memcg; |
26fe6168 | 1806 | |
c0ff4b85 R |
1807 | memcg = stock->cached; |
1808 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1809 | continue; |
2314b42d | 1810 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1811 | continue; |
d1a05b69 MH |
1812 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1813 | if (cpu == curcpu) | |
1814 | drain_local_stock(&stock->work); | |
1815 | else | |
1816 | schedule_work_on(cpu, &stock->work); | |
1817 | } | |
cdec2e42 | 1818 | } |
5af12d0e | 1819 | put_cpu(); |
f894ffa8 | 1820 | put_online_cpus(); |
9f50fad6 | 1821 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1822 | } |
1823 | ||
308167fc | 1824 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 1825 | { |
cdec2e42 KH |
1826 | struct memcg_stock_pcp *stock; |
1827 | ||
cdec2e42 KH |
1828 | stock = &per_cpu(memcg_stock, cpu); |
1829 | drain_stock(stock); | |
308167fc | 1830 | return 0; |
cdec2e42 KH |
1831 | } |
1832 | ||
f7e1cb6e JW |
1833 | static void reclaim_high(struct mem_cgroup *memcg, |
1834 | unsigned int nr_pages, | |
1835 | gfp_t gfp_mask) | |
1836 | { | |
1837 | do { | |
1838 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1839 | continue; | |
1840 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
1841 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
1842 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1843 | } | |
1844 | ||
1845 | static void high_work_func(struct work_struct *work) | |
1846 | { | |
1847 | struct mem_cgroup *memcg; | |
1848 | ||
1849 | memcg = container_of(work, struct mem_cgroup, high_work); | |
1850 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); | |
1851 | } | |
1852 | ||
b23afb93 TH |
1853 | /* |
1854 | * Scheduled by try_charge() to be executed from the userland return path | |
1855 | * and reclaims memory over the high limit. | |
1856 | */ | |
1857 | void mem_cgroup_handle_over_high(void) | |
1858 | { | |
1859 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1860 | struct mem_cgroup *memcg; |
b23afb93 TH |
1861 | |
1862 | if (likely(!nr_pages)) | |
1863 | return; | |
1864 | ||
f7e1cb6e JW |
1865 | memcg = get_mem_cgroup_from_mm(current->mm); |
1866 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1867 | css_put(&memcg->css); |
1868 | current->memcg_nr_pages_over_high = 0; | |
1869 | } | |
1870 | ||
00501b53 JW |
1871 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1872 | unsigned int nr_pages) | |
8a9f3ccd | 1873 | { |
7ec99d62 | 1874 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1875 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1876 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1877 | struct page_counter *counter; |
6539cc05 | 1878 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1879 | bool may_swap = true; |
1880 | bool drained = false; | |
a636b327 | 1881 | |
ce00a967 | 1882 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1883 | return 0; |
6539cc05 | 1884 | retry: |
b6b6cc72 | 1885 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1886 | return 0; |
8a9f3ccd | 1887 | |
7941d214 | 1888 | if (!do_memsw_account() || |
6071ca52 JW |
1889 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1890 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1891 | goto done_restock; |
7941d214 | 1892 | if (do_memsw_account()) |
3e32cb2e JW |
1893 | page_counter_uncharge(&memcg->memsw, batch); |
1894 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1895 | } else { |
3e32cb2e | 1896 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1897 | may_swap = false; |
3fbe7244 | 1898 | } |
7a81b88c | 1899 | |
6539cc05 JW |
1900 | if (batch > nr_pages) { |
1901 | batch = nr_pages; | |
1902 | goto retry; | |
1903 | } | |
6d61ef40 | 1904 | |
06b078fc JW |
1905 | /* |
1906 | * Unlike in global OOM situations, memcg is not in a physical | |
1907 | * memory shortage. Allow dying and OOM-killed tasks to | |
1908 | * bypass the last charges so that they can exit quickly and | |
1909 | * free their memory. | |
1910 | */ | |
1911 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
1912 | fatal_signal_pending(current) || | |
1913 | current->flags & PF_EXITING)) | |
10d53c74 | 1914 | goto force; |
06b078fc | 1915 | |
89a28483 JW |
1916 | /* |
1917 | * Prevent unbounded recursion when reclaim operations need to | |
1918 | * allocate memory. This might exceed the limits temporarily, | |
1919 | * but we prefer facilitating memory reclaim and getting back | |
1920 | * under the limit over triggering OOM kills in these cases. | |
1921 | */ | |
1922 | if (unlikely(current->flags & PF_MEMALLOC)) | |
1923 | goto force; | |
1924 | ||
06b078fc JW |
1925 | if (unlikely(task_in_memcg_oom(current))) |
1926 | goto nomem; | |
1927 | ||
d0164adc | 1928 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 1929 | goto nomem; |
4b534334 | 1930 | |
241994ed JW |
1931 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
1932 | ||
b70a2a21 JW |
1933 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
1934 | gfp_mask, may_swap); | |
6539cc05 | 1935 | |
61e02c74 | 1936 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 1937 | goto retry; |
28c34c29 | 1938 | |
b70a2a21 | 1939 | if (!drained) { |
6d3d6aa2 | 1940 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
1941 | drained = true; |
1942 | goto retry; | |
1943 | } | |
1944 | ||
28c34c29 JW |
1945 | if (gfp_mask & __GFP_NORETRY) |
1946 | goto nomem; | |
6539cc05 JW |
1947 | /* |
1948 | * Even though the limit is exceeded at this point, reclaim | |
1949 | * may have been able to free some pages. Retry the charge | |
1950 | * before killing the task. | |
1951 | * | |
1952 | * Only for regular pages, though: huge pages are rather | |
1953 | * unlikely to succeed so close to the limit, and we fall back | |
1954 | * to regular pages anyway in case of failure. | |
1955 | */ | |
61e02c74 | 1956 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
1957 | goto retry; |
1958 | /* | |
1959 | * At task move, charge accounts can be doubly counted. So, it's | |
1960 | * better to wait until the end of task_move if something is going on. | |
1961 | */ | |
1962 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
1963 | goto retry; | |
1964 | ||
9b130619 JW |
1965 | if (nr_retries--) |
1966 | goto retry; | |
1967 | ||
06b078fc | 1968 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 1969 | goto force; |
06b078fc | 1970 | |
6539cc05 | 1971 | if (fatal_signal_pending(current)) |
10d53c74 | 1972 | goto force; |
6539cc05 | 1973 | |
241994ed JW |
1974 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
1975 | ||
3608de07 JM |
1976 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
1977 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 1978 | nomem: |
6d1fdc48 | 1979 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 1980 | return -ENOMEM; |
10d53c74 TH |
1981 | force: |
1982 | /* | |
1983 | * The allocation either can't fail or will lead to more memory | |
1984 | * being freed very soon. Allow memory usage go over the limit | |
1985 | * temporarily by force charging it. | |
1986 | */ | |
1987 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 1988 | if (do_memsw_account()) |
10d53c74 TH |
1989 | page_counter_charge(&memcg->memsw, nr_pages); |
1990 | css_get_many(&memcg->css, nr_pages); | |
1991 | ||
1992 | return 0; | |
6539cc05 JW |
1993 | |
1994 | done_restock: | |
e8ea14cc | 1995 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
1996 | if (batch > nr_pages) |
1997 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 1998 | |
241994ed | 1999 | /* |
b23afb93 TH |
2000 | * If the hierarchy is above the normal consumption range, schedule |
2001 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2002 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2003 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2004 | * not recorded as it most likely matches current's and won't | |
2005 | * change in the meantime. As high limit is checked again before | |
2006 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2007 | */ |
2008 | do { | |
b23afb93 | 2009 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2010 | /* Don't bother a random interrupted task */ |
2011 | if (in_interrupt()) { | |
2012 | schedule_work(&memcg->high_work); | |
2013 | break; | |
2014 | } | |
9516a18a | 2015 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2016 | set_notify_resume(current); |
2017 | break; | |
2018 | } | |
241994ed | 2019 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2020 | |
2021 | return 0; | |
7a81b88c | 2022 | } |
8a9f3ccd | 2023 | |
00501b53 | 2024 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2025 | { |
ce00a967 JW |
2026 | if (mem_cgroup_is_root(memcg)) |
2027 | return; | |
2028 | ||
3e32cb2e | 2029 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2030 | if (do_memsw_account()) |
3e32cb2e | 2031 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2032 | |
e8ea14cc | 2033 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2034 | } |
2035 | ||
0a31bc97 JW |
2036 | static void lock_page_lru(struct page *page, int *isolated) |
2037 | { | |
2038 | struct zone *zone = page_zone(page); | |
2039 | ||
a52633d8 | 2040 | spin_lock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2041 | if (PageLRU(page)) { |
2042 | struct lruvec *lruvec; | |
2043 | ||
599d0c95 | 2044 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2045 | ClearPageLRU(page); |
2046 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2047 | *isolated = 1; | |
2048 | } else | |
2049 | *isolated = 0; | |
2050 | } | |
2051 | ||
2052 | static void unlock_page_lru(struct page *page, int isolated) | |
2053 | { | |
2054 | struct zone *zone = page_zone(page); | |
2055 | ||
2056 | if (isolated) { | |
2057 | struct lruvec *lruvec; | |
2058 | ||
599d0c95 | 2059 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2060 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2061 | SetPageLRU(page); | |
2062 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2063 | } | |
a52633d8 | 2064 | spin_unlock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2065 | } |
2066 | ||
00501b53 | 2067 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2068 | bool lrucare) |
7a81b88c | 2069 | { |
0a31bc97 | 2070 | int isolated; |
9ce70c02 | 2071 | |
1306a85a | 2072 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2073 | |
2074 | /* | |
2075 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2076 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2077 | */ | |
0a31bc97 JW |
2078 | if (lrucare) |
2079 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2080 | |
0a31bc97 JW |
2081 | /* |
2082 | * Nobody should be changing or seriously looking at | |
1306a85a | 2083 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2084 | * |
2085 | * - the page is uncharged | |
2086 | * | |
2087 | * - the page is off-LRU | |
2088 | * | |
2089 | * - an anonymous fault has exclusive page access, except for | |
2090 | * a locked page table | |
2091 | * | |
2092 | * - a page cache insertion, a swapin fault, or a migration | |
2093 | * have the page locked | |
2094 | */ | |
1306a85a | 2095 | page->mem_cgroup = memcg; |
9ce70c02 | 2096 | |
0a31bc97 JW |
2097 | if (lrucare) |
2098 | unlock_page_lru(page, isolated); | |
7a81b88c | 2099 | } |
66e1707b | 2100 | |
127424c8 | 2101 | #ifndef CONFIG_SLOB |
f3bb3043 | 2102 | static int memcg_alloc_cache_id(void) |
55007d84 | 2103 | { |
f3bb3043 VD |
2104 | int id, size; |
2105 | int err; | |
2106 | ||
dbcf73e2 | 2107 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2108 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2109 | if (id < 0) | |
2110 | return id; | |
55007d84 | 2111 | |
dbcf73e2 | 2112 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2113 | return id; |
2114 | ||
2115 | /* | |
2116 | * There's no space for the new id in memcg_caches arrays, | |
2117 | * so we have to grow them. | |
2118 | */ | |
05257a1a | 2119 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2120 | |
2121 | size = 2 * (id + 1); | |
55007d84 GC |
2122 | if (size < MEMCG_CACHES_MIN_SIZE) |
2123 | size = MEMCG_CACHES_MIN_SIZE; | |
2124 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2125 | size = MEMCG_CACHES_MAX_SIZE; | |
2126 | ||
f3bb3043 | 2127 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2128 | if (!err) |
2129 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2130 | if (!err) |
2131 | memcg_nr_cache_ids = size; | |
2132 | ||
2133 | up_write(&memcg_cache_ids_sem); | |
2134 | ||
f3bb3043 | 2135 | if (err) { |
dbcf73e2 | 2136 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2137 | return err; |
2138 | } | |
2139 | return id; | |
2140 | } | |
2141 | ||
2142 | static void memcg_free_cache_id(int id) | |
2143 | { | |
dbcf73e2 | 2144 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2145 | } |
2146 | ||
d5b3cf71 | 2147 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2148 | struct mem_cgroup *memcg; |
2149 | struct kmem_cache *cachep; | |
2150 | struct work_struct work; | |
2151 | }; | |
2152 | ||
d5b3cf71 | 2153 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2154 | { |
d5b3cf71 VD |
2155 | struct memcg_kmem_cache_create_work *cw = |
2156 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2157 | struct mem_cgroup *memcg = cw->memcg; |
2158 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2159 | |
d5b3cf71 | 2160 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2161 | |
5722d094 | 2162 | css_put(&memcg->css); |
d7f25f8a GC |
2163 | kfree(cw); |
2164 | } | |
2165 | ||
2166 | /* | |
2167 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2168 | */ |
d5b3cf71 VD |
2169 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2170 | struct kmem_cache *cachep) | |
d7f25f8a | 2171 | { |
d5b3cf71 | 2172 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2173 | |
776ed0f0 | 2174 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2175 | if (!cw) |
d7f25f8a | 2176 | return; |
8135be5a VD |
2177 | |
2178 | css_get(&memcg->css); | |
d7f25f8a GC |
2179 | |
2180 | cw->memcg = memcg; | |
2181 | cw->cachep = cachep; | |
d5b3cf71 | 2182 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2183 | |
17cc4dfe | 2184 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2185 | } |
2186 | ||
d5b3cf71 VD |
2187 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2188 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2189 | { |
2190 | /* | |
2191 | * We need to stop accounting when we kmalloc, because if the | |
2192 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2193 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2194 | * |
2195 | * However, it is better to enclose the whole function. Depending on | |
2196 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2197 | * trigger an allocation. This too, will make us recurse. Because at | |
2198 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2199 | * the safest choice is to do it like this, wrapping the whole function. | |
2200 | */ | |
6f185c29 | 2201 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2202 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2203 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2204 | } |
c67a8a68 | 2205 | |
45264778 VD |
2206 | static inline bool memcg_kmem_bypass(void) |
2207 | { | |
2208 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2209 | return true; | |
2210 | return false; | |
2211 | } | |
2212 | ||
2213 | /** | |
2214 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2215 | * @cachep: the original global kmem cache | |
2216 | * | |
d7f25f8a GC |
2217 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2218 | * We try to use the current memcg's version of the cache. | |
2219 | * | |
45264778 VD |
2220 | * If the cache does not exist yet, if we are the first user of it, we |
2221 | * create it asynchronously in a workqueue and let the current allocation | |
2222 | * go through with the original cache. | |
d7f25f8a | 2223 | * |
45264778 VD |
2224 | * This function takes a reference to the cache it returns to assure it |
2225 | * won't get destroyed while we are working with it. Once the caller is | |
2226 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2227 | * reference. | |
d7f25f8a | 2228 | */ |
45264778 | 2229 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2230 | { |
2231 | struct mem_cgroup *memcg; | |
959c8963 | 2232 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2233 | int kmemcg_id; |
d7f25f8a | 2234 | |
f7ce3190 | 2235 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2236 | |
45264778 | 2237 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2238 | return cachep; |
2239 | ||
9d100c5e | 2240 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2241 | return cachep; |
2242 | ||
8135be5a | 2243 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2244 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2245 | if (kmemcg_id < 0) |
ca0dde97 | 2246 | goto out; |
d7f25f8a | 2247 | |
2a4db7eb | 2248 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2249 | if (likely(memcg_cachep)) |
2250 | return memcg_cachep; | |
ca0dde97 LZ |
2251 | |
2252 | /* | |
2253 | * If we are in a safe context (can wait, and not in interrupt | |
2254 | * context), we could be be predictable and return right away. | |
2255 | * This would guarantee that the allocation being performed | |
2256 | * already belongs in the new cache. | |
2257 | * | |
2258 | * However, there are some clashes that can arrive from locking. | |
2259 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2260 | * memcg_create_kmem_cache, this means no further allocation |
2261 | * could happen with the slab_mutex held. So it's better to | |
2262 | * defer everything. | |
ca0dde97 | 2263 | */ |
d5b3cf71 | 2264 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2265 | out: |
8135be5a | 2266 | css_put(&memcg->css); |
ca0dde97 | 2267 | return cachep; |
d7f25f8a | 2268 | } |
d7f25f8a | 2269 | |
45264778 VD |
2270 | /** |
2271 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2272 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2273 | */ | |
2274 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2275 | { |
2276 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2277 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2278 | } |
2279 | ||
45264778 VD |
2280 | /** |
2281 | * memcg_kmem_charge: charge a kmem page | |
2282 | * @page: page to charge | |
2283 | * @gfp: reclaim mode | |
2284 | * @order: allocation order | |
2285 | * @memcg: memory cgroup to charge | |
2286 | * | |
2287 | * Returns 0 on success, an error code on failure. | |
2288 | */ | |
2289 | int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, | |
2290 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2291 | { |
f3ccb2c4 VD |
2292 | unsigned int nr_pages = 1 << order; |
2293 | struct page_counter *counter; | |
7ae1e1d0 GC |
2294 | int ret; |
2295 | ||
f3ccb2c4 | 2296 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2297 | if (ret) |
f3ccb2c4 | 2298 | return ret; |
52c29b04 JW |
2299 | |
2300 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2301 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2302 | cancel_charge(memcg, nr_pages); | |
2303 | return -ENOMEM; | |
7ae1e1d0 GC |
2304 | } |
2305 | ||
f3ccb2c4 | 2306 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2307 | |
f3ccb2c4 | 2308 | return 0; |
7ae1e1d0 GC |
2309 | } |
2310 | ||
45264778 VD |
2311 | /** |
2312 | * memcg_kmem_charge: charge a kmem page to the current memory cgroup | |
2313 | * @page: page to charge | |
2314 | * @gfp: reclaim mode | |
2315 | * @order: allocation order | |
2316 | * | |
2317 | * Returns 0 on success, an error code on failure. | |
2318 | */ | |
2319 | int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) | |
7ae1e1d0 | 2320 | { |
f3ccb2c4 | 2321 | struct mem_cgroup *memcg; |
fcff7d7e | 2322 | int ret = 0; |
7ae1e1d0 | 2323 | |
45264778 VD |
2324 | if (memcg_kmem_bypass()) |
2325 | return 0; | |
2326 | ||
f3ccb2c4 | 2327 | memcg = get_mem_cgroup_from_mm(current->mm); |
c4159a75 | 2328 | if (!mem_cgroup_is_root(memcg)) { |
45264778 | 2329 | ret = memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2330 | if (!ret) |
2331 | __SetPageKmemcg(page); | |
2332 | } | |
7ae1e1d0 | 2333 | css_put(&memcg->css); |
d05e83a6 | 2334 | return ret; |
7ae1e1d0 | 2335 | } |
45264778 VD |
2336 | /** |
2337 | * memcg_kmem_uncharge: uncharge a kmem page | |
2338 | * @page: page to uncharge | |
2339 | * @order: allocation order | |
2340 | */ | |
2341 | void memcg_kmem_uncharge(struct page *page, int order) | |
7ae1e1d0 | 2342 | { |
1306a85a | 2343 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2344 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2345 | |
7ae1e1d0 GC |
2346 | if (!memcg) |
2347 | return; | |
2348 | ||
309381fe | 2349 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2350 | |
52c29b04 JW |
2351 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2352 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2353 | ||
f3ccb2c4 | 2354 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2355 | if (do_memsw_account()) |
f3ccb2c4 | 2356 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2357 | |
1306a85a | 2358 | page->mem_cgroup = NULL; |
c4159a75 VD |
2359 | |
2360 | /* slab pages do not have PageKmemcg flag set */ | |
2361 | if (PageKmemcg(page)) | |
2362 | __ClearPageKmemcg(page); | |
2363 | ||
f3ccb2c4 | 2364 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2365 | } |
127424c8 | 2366 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2367 | |
ca3e0214 KH |
2368 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2369 | ||
ca3e0214 KH |
2370 | /* |
2371 | * Because tail pages are not marked as "used", set it. We're under | |
a52633d8 | 2372 | * zone_lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2373 | */ |
e94c8a9c | 2374 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2375 | { |
e94c8a9c | 2376 | int i; |
ca3e0214 | 2377 | |
3d37c4a9 KH |
2378 | if (mem_cgroup_disabled()) |
2379 | return; | |
b070e65c | 2380 | |
29833315 | 2381 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2382 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2383 | |
1306a85a | 2384 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2385 | HPAGE_PMD_NR); |
ca3e0214 | 2386 | } |
12d27107 | 2387 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2388 | |
c255a458 | 2389 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2390 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2391 | bool charge) | |
d13d1443 | 2392 | { |
0a31bc97 JW |
2393 | int val = (charge) ? 1 : -1; |
2394 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2395 | } |
02491447 DN |
2396 | |
2397 | /** | |
2398 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2399 | * @entry: swap entry to be moved | |
2400 | * @from: mem_cgroup which the entry is moved from | |
2401 | * @to: mem_cgroup which the entry is moved to | |
2402 | * | |
2403 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2404 | * as the mem_cgroup's id of @from. | |
2405 | * | |
2406 | * Returns 0 on success, -EINVAL on failure. | |
2407 | * | |
3e32cb2e | 2408 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2409 | * both res and memsw, and called css_get(). |
2410 | */ | |
2411 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2412 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2413 | { |
2414 | unsigned short old_id, new_id; | |
2415 | ||
34c00c31 LZ |
2416 | old_id = mem_cgroup_id(from); |
2417 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2418 | |
2419 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2420 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2421 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2422 | return 0; |
2423 | } | |
2424 | return -EINVAL; | |
2425 | } | |
2426 | #else | |
2427 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2428 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2429 | { |
2430 | return -EINVAL; | |
2431 | } | |
8c7c6e34 | 2432 | #endif |
d13d1443 | 2433 | |
3e32cb2e | 2434 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2435 | |
d38d2a75 | 2436 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2437 | unsigned long limit) |
628f4235 | 2438 | { |
3e32cb2e JW |
2439 | unsigned long curusage; |
2440 | unsigned long oldusage; | |
2441 | bool enlarge = false; | |
81d39c20 | 2442 | int retry_count; |
3e32cb2e | 2443 | int ret; |
81d39c20 KH |
2444 | |
2445 | /* | |
2446 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2447 | * is depends on callers. We set our retry-count to be function | |
2448 | * of # of children which we should visit in this loop. | |
2449 | */ | |
3e32cb2e JW |
2450 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2451 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2452 | |
3e32cb2e | 2453 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2454 | |
3e32cb2e | 2455 | do { |
628f4235 KH |
2456 | if (signal_pending(current)) { |
2457 | ret = -EINTR; | |
2458 | break; | |
2459 | } | |
3e32cb2e JW |
2460 | |
2461 | mutex_lock(&memcg_limit_mutex); | |
2462 | if (limit > memcg->memsw.limit) { | |
2463 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2464 | ret = -EINVAL; |
628f4235 KH |
2465 | break; |
2466 | } | |
3e32cb2e JW |
2467 | if (limit > memcg->memory.limit) |
2468 | enlarge = true; | |
2469 | ret = page_counter_limit(&memcg->memory, limit); | |
2470 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2471 | |
2472 | if (!ret) | |
2473 | break; | |
2474 | ||
b70a2a21 JW |
2475 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2476 | ||
3e32cb2e | 2477 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2478 | /* Usage is reduced ? */ |
f894ffa8 | 2479 | if (curusage >= oldusage) |
81d39c20 KH |
2480 | retry_count--; |
2481 | else | |
2482 | oldusage = curusage; | |
3e32cb2e JW |
2483 | } while (retry_count); |
2484 | ||
3c11ecf4 KH |
2485 | if (!ret && enlarge) |
2486 | memcg_oom_recover(memcg); | |
14797e23 | 2487 | |
8c7c6e34 KH |
2488 | return ret; |
2489 | } | |
2490 | ||
338c8431 | 2491 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2492 | unsigned long limit) |
8c7c6e34 | 2493 | { |
3e32cb2e JW |
2494 | unsigned long curusage; |
2495 | unsigned long oldusage; | |
2496 | bool enlarge = false; | |
81d39c20 | 2497 | int retry_count; |
3e32cb2e | 2498 | int ret; |
8c7c6e34 | 2499 | |
81d39c20 | 2500 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2501 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2502 | mem_cgroup_count_children(memcg); | |
2503 | ||
2504 | oldusage = page_counter_read(&memcg->memsw); | |
2505 | ||
2506 | do { | |
8c7c6e34 KH |
2507 | if (signal_pending(current)) { |
2508 | ret = -EINTR; | |
2509 | break; | |
2510 | } | |
3e32cb2e JW |
2511 | |
2512 | mutex_lock(&memcg_limit_mutex); | |
2513 | if (limit < memcg->memory.limit) { | |
2514 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2515 | ret = -EINVAL; |
8c7c6e34 KH |
2516 | break; |
2517 | } | |
3e32cb2e JW |
2518 | if (limit > memcg->memsw.limit) |
2519 | enlarge = true; | |
2520 | ret = page_counter_limit(&memcg->memsw, limit); | |
2521 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2522 | |
2523 | if (!ret) | |
2524 | break; | |
2525 | ||
b70a2a21 JW |
2526 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2527 | ||
3e32cb2e | 2528 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2529 | /* Usage is reduced ? */ |
8c7c6e34 | 2530 | if (curusage >= oldusage) |
628f4235 | 2531 | retry_count--; |
81d39c20 KH |
2532 | else |
2533 | oldusage = curusage; | |
3e32cb2e JW |
2534 | } while (retry_count); |
2535 | ||
3c11ecf4 KH |
2536 | if (!ret && enlarge) |
2537 | memcg_oom_recover(memcg); | |
3e32cb2e | 2538 | |
628f4235 KH |
2539 | return ret; |
2540 | } | |
2541 | ||
ef8f2327 | 2542 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2543 | gfp_t gfp_mask, |
2544 | unsigned long *total_scanned) | |
2545 | { | |
2546 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2547 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2548 | unsigned long reclaimed; |
2549 | int loop = 0; | |
ef8f2327 | 2550 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2551 | unsigned long excess; |
0608f43d AM |
2552 | unsigned long nr_scanned; |
2553 | ||
2554 | if (order > 0) | |
2555 | return 0; | |
2556 | ||
ef8f2327 | 2557 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2558 | |
2559 | /* | |
2560 | * Do not even bother to check the largest node if the root | |
2561 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2562 | * are acceptable as soft limit is best effort anyway. | |
2563 | */ | |
bfc7228b | 2564 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2565 | return 0; |
2566 | ||
0608f43d AM |
2567 | /* |
2568 | * This loop can run a while, specially if mem_cgroup's continuously | |
2569 | * keep exceeding their soft limit and putting the system under | |
2570 | * pressure | |
2571 | */ | |
2572 | do { | |
2573 | if (next_mz) | |
2574 | mz = next_mz; | |
2575 | else | |
2576 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2577 | if (!mz) | |
2578 | break; | |
2579 | ||
2580 | nr_scanned = 0; | |
ef8f2327 | 2581 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2582 | gfp_mask, &nr_scanned); |
2583 | nr_reclaimed += reclaimed; | |
2584 | *total_scanned += nr_scanned; | |
0a31bc97 | 2585 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2586 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2587 | |
2588 | /* | |
2589 | * If we failed to reclaim anything from this memory cgroup | |
2590 | * it is time to move on to the next cgroup | |
2591 | */ | |
2592 | next_mz = NULL; | |
bc2f2e7f VD |
2593 | if (!reclaimed) |
2594 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2595 | ||
3e32cb2e | 2596 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2597 | /* |
2598 | * One school of thought says that we should not add | |
2599 | * back the node to the tree if reclaim returns 0. | |
2600 | * But our reclaim could return 0, simply because due | |
2601 | * to priority we are exposing a smaller subset of | |
2602 | * memory to reclaim from. Consider this as a longer | |
2603 | * term TODO. | |
2604 | */ | |
2605 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2606 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2607 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2608 | css_put(&mz->memcg->css); |
2609 | loop++; | |
2610 | /* | |
2611 | * Could not reclaim anything and there are no more | |
2612 | * mem cgroups to try or we seem to be looping without | |
2613 | * reclaiming anything. | |
2614 | */ | |
2615 | if (!nr_reclaimed && | |
2616 | (next_mz == NULL || | |
2617 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2618 | break; | |
2619 | } while (!nr_reclaimed); | |
2620 | if (next_mz) | |
2621 | css_put(&next_mz->memcg->css); | |
2622 | return nr_reclaimed; | |
2623 | } | |
2624 | ||
ea280e7b TH |
2625 | /* |
2626 | * Test whether @memcg has children, dead or alive. Note that this | |
2627 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2628 | * returns %true if there are child csses according to the cgroup | |
2629 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2630 | */ | |
b5f99b53 GC |
2631 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2632 | { | |
ea280e7b TH |
2633 | bool ret; |
2634 | ||
ea280e7b TH |
2635 | rcu_read_lock(); |
2636 | ret = css_next_child(NULL, &memcg->css); | |
2637 | rcu_read_unlock(); | |
2638 | return ret; | |
b5f99b53 GC |
2639 | } |
2640 | ||
c26251f9 | 2641 | /* |
51038171 | 2642 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2643 | * |
2644 | * Caller is responsible for holding css reference for memcg. | |
2645 | */ | |
2646 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2647 | { | |
2648 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2649 | |
c1e862c1 KH |
2650 | /* we call try-to-free pages for make this cgroup empty */ |
2651 | lru_add_drain_all(); | |
f817ed48 | 2652 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2653 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2654 | int progress; |
c1e862c1 | 2655 | |
c26251f9 MH |
2656 | if (signal_pending(current)) |
2657 | return -EINTR; | |
2658 | ||
b70a2a21 JW |
2659 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2660 | GFP_KERNEL, true); | |
c1e862c1 | 2661 | if (!progress) { |
f817ed48 | 2662 | nr_retries--; |
c1e862c1 | 2663 | /* maybe some writeback is necessary */ |
8aa7e847 | 2664 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2665 | } |
f817ed48 KH |
2666 | |
2667 | } | |
ab5196c2 MH |
2668 | |
2669 | return 0; | |
cc847582 KH |
2670 | } |
2671 | ||
6770c64e TH |
2672 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2673 | char *buf, size_t nbytes, | |
2674 | loff_t off) | |
c1e862c1 | 2675 | { |
6770c64e | 2676 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2677 | |
d8423011 MH |
2678 | if (mem_cgroup_is_root(memcg)) |
2679 | return -EINVAL; | |
6770c64e | 2680 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2681 | } |
2682 | ||
182446d0 TH |
2683 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2684 | struct cftype *cft) | |
18f59ea7 | 2685 | { |
182446d0 | 2686 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2687 | } |
2688 | ||
182446d0 TH |
2689 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2690 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2691 | { |
2692 | int retval = 0; | |
182446d0 | 2693 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2694 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2695 | |
567fb435 | 2696 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2697 | return 0; |
567fb435 | 2698 | |
18f59ea7 | 2699 | /* |
af901ca1 | 2700 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2701 | * in the child subtrees. If it is unset, then the change can |
2702 | * occur, provided the current cgroup has no children. | |
2703 | * | |
2704 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2705 | * set if there are no children. | |
2706 | */ | |
c0ff4b85 | 2707 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2708 | (val == 1 || val == 0)) { |
ea280e7b | 2709 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2710 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2711 | else |
2712 | retval = -EBUSY; | |
2713 | } else | |
2714 | retval = -EINVAL; | |
567fb435 | 2715 | |
18f59ea7 BS |
2716 | return retval; |
2717 | } | |
2718 | ||
72b54e73 | 2719 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
ce00a967 JW |
2720 | { |
2721 | struct mem_cgroup *iter; | |
72b54e73 | 2722 | int i; |
ce00a967 | 2723 | |
72b54e73 | 2724 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
ce00a967 | 2725 | |
72b54e73 VD |
2726 | for_each_mem_cgroup_tree(iter, memcg) { |
2727 | for (i = 0; i < MEMCG_NR_STAT; i++) | |
2728 | stat[i] += mem_cgroup_read_stat(iter, i); | |
2729 | } | |
ce00a967 JW |
2730 | } |
2731 | ||
72b54e73 | 2732 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
587d9f72 JW |
2733 | { |
2734 | struct mem_cgroup *iter; | |
72b54e73 | 2735 | int i; |
587d9f72 | 2736 | |
72b54e73 | 2737 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
587d9f72 | 2738 | |
72b54e73 VD |
2739 | for_each_mem_cgroup_tree(iter, memcg) { |
2740 | for (i = 0; i < MEMCG_NR_EVENTS; i++) | |
2741 | events[i] += mem_cgroup_read_events(iter, i); | |
2742 | } | |
587d9f72 JW |
2743 | } |
2744 | ||
6f646156 | 2745 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2746 | { |
72b54e73 | 2747 | unsigned long val = 0; |
ce00a967 | 2748 | |
3e32cb2e | 2749 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2750 | struct mem_cgroup *iter; |
2751 | ||
2752 | for_each_mem_cgroup_tree(iter, memcg) { | |
2753 | val += mem_cgroup_read_stat(iter, | |
2754 | MEM_CGROUP_STAT_CACHE); | |
2755 | val += mem_cgroup_read_stat(iter, | |
2756 | MEM_CGROUP_STAT_RSS); | |
2757 | if (swap) | |
2758 | val += mem_cgroup_read_stat(iter, | |
2759 | MEM_CGROUP_STAT_SWAP); | |
2760 | } | |
3e32cb2e | 2761 | } else { |
ce00a967 | 2762 | if (!swap) |
3e32cb2e | 2763 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2764 | else |
3e32cb2e | 2765 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2766 | } |
c12176d3 | 2767 | return val; |
ce00a967 JW |
2768 | } |
2769 | ||
3e32cb2e JW |
2770 | enum { |
2771 | RES_USAGE, | |
2772 | RES_LIMIT, | |
2773 | RES_MAX_USAGE, | |
2774 | RES_FAILCNT, | |
2775 | RES_SOFT_LIMIT, | |
2776 | }; | |
ce00a967 | 2777 | |
791badbd | 2778 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2779 | struct cftype *cft) |
8cdea7c0 | 2780 | { |
182446d0 | 2781 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2782 | struct page_counter *counter; |
af36f906 | 2783 | |
3e32cb2e | 2784 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2785 | case _MEM: |
3e32cb2e JW |
2786 | counter = &memcg->memory; |
2787 | break; | |
8c7c6e34 | 2788 | case _MEMSWAP: |
3e32cb2e JW |
2789 | counter = &memcg->memsw; |
2790 | break; | |
510fc4e1 | 2791 | case _KMEM: |
3e32cb2e | 2792 | counter = &memcg->kmem; |
510fc4e1 | 2793 | break; |
d55f90bf | 2794 | case _TCP: |
0db15298 | 2795 | counter = &memcg->tcpmem; |
d55f90bf | 2796 | break; |
8c7c6e34 KH |
2797 | default: |
2798 | BUG(); | |
8c7c6e34 | 2799 | } |
3e32cb2e JW |
2800 | |
2801 | switch (MEMFILE_ATTR(cft->private)) { | |
2802 | case RES_USAGE: | |
2803 | if (counter == &memcg->memory) | |
c12176d3 | 2804 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2805 | if (counter == &memcg->memsw) |
c12176d3 | 2806 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2807 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2808 | case RES_LIMIT: | |
2809 | return (u64)counter->limit * PAGE_SIZE; | |
2810 | case RES_MAX_USAGE: | |
2811 | return (u64)counter->watermark * PAGE_SIZE; | |
2812 | case RES_FAILCNT: | |
2813 | return counter->failcnt; | |
2814 | case RES_SOFT_LIMIT: | |
2815 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2816 | default: | |
2817 | BUG(); | |
2818 | } | |
8cdea7c0 | 2819 | } |
510fc4e1 | 2820 | |
127424c8 | 2821 | #ifndef CONFIG_SLOB |
567e9ab2 | 2822 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2823 | { |
d6441637 VD |
2824 | int memcg_id; |
2825 | ||
b313aeee VD |
2826 | if (cgroup_memory_nokmem) |
2827 | return 0; | |
2828 | ||
2a4db7eb | 2829 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2830 | BUG_ON(memcg->kmem_state); |
d6441637 | 2831 | |
f3bb3043 | 2832 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
2833 | if (memcg_id < 0) |
2834 | return memcg_id; | |
d6441637 | 2835 | |
ef12947c | 2836 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2837 | /* |
567e9ab2 | 2838 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2839 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2840 | * guarantee no one starts accounting before all call sites are |
2841 | * patched. | |
2842 | */ | |
900a38f0 | 2843 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2844 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 2845 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
2846 | |
2847 | return 0; | |
d6441637 VD |
2848 | } |
2849 | ||
8e0a8912 JW |
2850 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2851 | { | |
2852 | struct cgroup_subsys_state *css; | |
2853 | struct mem_cgroup *parent, *child; | |
2854 | int kmemcg_id; | |
2855 | ||
2856 | if (memcg->kmem_state != KMEM_ONLINE) | |
2857 | return; | |
2858 | /* | |
2859 | * Clear the online state before clearing memcg_caches array | |
2860 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2861 | * guarantees that no cache will be created for this cgroup | |
2862 | * after we are done (see memcg_create_kmem_cache()). | |
2863 | */ | |
2864 | memcg->kmem_state = KMEM_ALLOCATED; | |
2865 | ||
2866 | memcg_deactivate_kmem_caches(memcg); | |
2867 | ||
2868 | kmemcg_id = memcg->kmemcg_id; | |
2869 | BUG_ON(kmemcg_id < 0); | |
2870 | ||
2871 | parent = parent_mem_cgroup(memcg); | |
2872 | if (!parent) | |
2873 | parent = root_mem_cgroup; | |
2874 | ||
2875 | /* | |
2876 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2877 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2878 | * to ones of the parent. After we have finished, all list_lrus | |
2879 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2880 | * ordering is imposed by list_lru_node->lock taken by | |
2881 | * memcg_drain_all_list_lrus(). | |
2882 | */ | |
3a06bb78 | 2883 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
2884 | css_for_each_descendant_pre(css, &memcg->css) { |
2885 | child = mem_cgroup_from_css(css); | |
2886 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2887 | child->kmemcg_id = parent->kmemcg_id; | |
2888 | if (!memcg->use_hierarchy) | |
2889 | break; | |
2890 | } | |
3a06bb78 TH |
2891 | rcu_read_unlock(); |
2892 | ||
8e0a8912 JW |
2893 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
2894 | ||
2895 | memcg_free_cache_id(kmemcg_id); | |
2896 | } | |
2897 | ||
2898 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2899 | { | |
0b8f73e1 JW |
2900 | /* css_alloc() failed, offlining didn't happen */ |
2901 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
2902 | memcg_offline_kmem(memcg); | |
2903 | ||
8e0a8912 JW |
2904 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
2905 | memcg_destroy_kmem_caches(memcg); | |
2906 | static_branch_dec(&memcg_kmem_enabled_key); | |
2907 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2908 | } | |
8e0a8912 | 2909 | } |
d6441637 | 2910 | #else |
0b8f73e1 | 2911 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
2912 | { |
2913 | return 0; | |
2914 | } | |
2915 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
2916 | { | |
2917 | } | |
2918 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2919 | { | |
2920 | } | |
2921 | #endif /* !CONFIG_SLOB */ | |
2922 | ||
d6441637 | 2923 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2924 | unsigned long limit) |
d6441637 | 2925 | { |
b313aeee | 2926 | int ret; |
127424c8 JW |
2927 | |
2928 | mutex_lock(&memcg_limit_mutex); | |
127424c8 | 2929 | ret = page_counter_limit(&memcg->kmem, limit); |
127424c8 JW |
2930 | mutex_unlock(&memcg_limit_mutex); |
2931 | return ret; | |
d6441637 | 2932 | } |
510fc4e1 | 2933 | |
d55f90bf VD |
2934 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
2935 | { | |
2936 | int ret; | |
2937 | ||
2938 | mutex_lock(&memcg_limit_mutex); | |
2939 | ||
0db15298 | 2940 | ret = page_counter_limit(&memcg->tcpmem, limit); |
d55f90bf VD |
2941 | if (ret) |
2942 | goto out; | |
2943 | ||
0db15298 | 2944 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
2945 | /* |
2946 | * The active flag needs to be written after the static_key | |
2947 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
2948 | * function is the last one to run. See mem_cgroup_sk_alloc() |
2949 | * for details, and note that we don't mark any socket as | |
2950 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
2951 | * |
2952 | * We need to do this, because static_keys will span multiple | |
2953 | * sites, but we can't control their order. If we mark a socket | |
2954 | * as accounted, but the accounting functions are not patched in | |
2955 | * yet, we'll lose accounting. | |
2956 | * | |
2d758073 | 2957 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
2958 | * because when this value change, the code to process it is not |
2959 | * patched in yet. | |
2960 | */ | |
2961 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 2962 | memcg->tcpmem_active = true; |
d55f90bf VD |
2963 | } |
2964 | out: | |
2965 | mutex_unlock(&memcg_limit_mutex); | |
2966 | return ret; | |
2967 | } | |
d55f90bf | 2968 | |
628f4235 KH |
2969 | /* |
2970 | * The user of this function is... | |
2971 | * RES_LIMIT. | |
2972 | */ | |
451af504 TH |
2973 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2974 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2975 | { |
451af504 | 2976 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2977 | unsigned long nr_pages; |
628f4235 KH |
2978 | int ret; |
2979 | ||
451af504 | 2980 | buf = strstrip(buf); |
650c5e56 | 2981 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
2982 | if (ret) |
2983 | return ret; | |
af36f906 | 2984 | |
3e32cb2e | 2985 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 2986 | case RES_LIMIT: |
4b3bde4c BS |
2987 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2988 | ret = -EINVAL; | |
2989 | break; | |
2990 | } | |
3e32cb2e JW |
2991 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
2992 | case _MEM: | |
2993 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 2994 | break; |
3e32cb2e JW |
2995 | case _MEMSWAP: |
2996 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 2997 | break; |
3e32cb2e JW |
2998 | case _KMEM: |
2999 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3000 | break; | |
d55f90bf VD |
3001 | case _TCP: |
3002 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
3003 | break; | |
3e32cb2e | 3004 | } |
296c81d8 | 3005 | break; |
3e32cb2e JW |
3006 | case RES_SOFT_LIMIT: |
3007 | memcg->soft_limit = nr_pages; | |
3008 | ret = 0; | |
628f4235 KH |
3009 | break; |
3010 | } | |
451af504 | 3011 | return ret ?: nbytes; |
8cdea7c0 BS |
3012 | } |
3013 | ||
6770c64e TH |
3014 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3015 | size_t nbytes, loff_t off) | |
c84872e1 | 3016 | { |
6770c64e | 3017 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3018 | struct page_counter *counter; |
c84872e1 | 3019 | |
3e32cb2e JW |
3020 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3021 | case _MEM: | |
3022 | counter = &memcg->memory; | |
3023 | break; | |
3024 | case _MEMSWAP: | |
3025 | counter = &memcg->memsw; | |
3026 | break; | |
3027 | case _KMEM: | |
3028 | counter = &memcg->kmem; | |
3029 | break; | |
d55f90bf | 3030 | case _TCP: |
0db15298 | 3031 | counter = &memcg->tcpmem; |
d55f90bf | 3032 | break; |
3e32cb2e JW |
3033 | default: |
3034 | BUG(); | |
3035 | } | |
af36f906 | 3036 | |
3e32cb2e | 3037 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3038 | case RES_MAX_USAGE: |
3e32cb2e | 3039 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3040 | break; |
3041 | case RES_FAILCNT: | |
3e32cb2e | 3042 | counter->failcnt = 0; |
29f2a4da | 3043 | break; |
3e32cb2e JW |
3044 | default: |
3045 | BUG(); | |
29f2a4da | 3046 | } |
f64c3f54 | 3047 | |
6770c64e | 3048 | return nbytes; |
c84872e1 PE |
3049 | } |
3050 | ||
182446d0 | 3051 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3052 | struct cftype *cft) |
3053 | { | |
182446d0 | 3054 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3055 | } |
3056 | ||
02491447 | 3057 | #ifdef CONFIG_MMU |
182446d0 | 3058 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3059 | struct cftype *cft, u64 val) |
3060 | { | |
182446d0 | 3061 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3062 | |
1dfab5ab | 3063 | if (val & ~MOVE_MASK) |
7dc74be0 | 3064 | return -EINVAL; |
ee5e8472 | 3065 | |
7dc74be0 | 3066 | /* |
ee5e8472 GC |
3067 | * No kind of locking is needed in here, because ->can_attach() will |
3068 | * check this value once in the beginning of the process, and then carry | |
3069 | * on with stale data. This means that changes to this value will only | |
3070 | * affect task migrations starting after the change. | |
7dc74be0 | 3071 | */ |
c0ff4b85 | 3072 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3073 | return 0; |
3074 | } | |
02491447 | 3075 | #else |
182446d0 | 3076 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3077 | struct cftype *cft, u64 val) |
3078 | { | |
3079 | return -ENOSYS; | |
3080 | } | |
3081 | #endif | |
7dc74be0 | 3082 | |
406eb0c9 | 3083 | #ifdef CONFIG_NUMA |
2da8ca82 | 3084 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3085 | { |
25485de6 GT |
3086 | struct numa_stat { |
3087 | const char *name; | |
3088 | unsigned int lru_mask; | |
3089 | }; | |
3090 | ||
3091 | static const struct numa_stat stats[] = { | |
3092 | { "total", LRU_ALL }, | |
3093 | { "file", LRU_ALL_FILE }, | |
3094 | { "anon", LRU_ALL_ANON }, | |
3095 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3096 | }; | |
3097 | const struct numa_stat *stat; | |
406eb0c9 | 3098 | int nid; |
25485de6 | 3099 | unsigned long nr; |
2da8ca82 | 3100 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3101 | |
25485de6 GT |
3102 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3103 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3104 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3105 | for_each_node_state(nid, N_MEMORY) { | |
3106 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3107 | stat->lru_mask); | |
3108 | seq_printf(m, " N%d=%lu", nid, nr); | |
3109 | } | |
3110 | seq_putc(m, '\n'); | |
406eb0c9 | 3111 | } |
406eb0c9 | 3112 | |
071aee13 YH |
3113 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3114 | struct mem_cgroup *iter; | |
3115 | ||
3116 | nr = 0; | |
3117 | for_each_mem_cgroup_tree(iter, memcg) | |
3118 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3119 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3120 | for_each_node_state(nid, N_MEMORY) { | |
3121 | nr = 0; | |
3122 | for_each_mem_cgroup_tree(iter, memcg) | |
3123 | nr += mem_cgroup_node_nr_lru_pages( | |
3124 | iter, nid, stat->lru_mask); | |
3125 | seq_printf(m, " N%d=%lu", nid, nr); | |
3126 | } | |
3127 | seq_putc(m, '\n'); | |
406eb0c9 | 3128 | } |
406eb0c9 | 3129 | |
406eb0c9 YH |
3130 | return 0; |
3131 | } | |
3132 | #endif /* CONFIG_NUMA */ | |
3133 | ||
2da8ca82 | 3134 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3135 | { |
2da8ca82 | 3136 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3137 | unsigned long memory, memsw; |
af7c4b0e JW |
3138 | struct mem_cgroup *mi; |
3139 | unsigned int i; | |
406eb0c9 | 3140 | |
0ca44b14 GT |
3141 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3142 | MEM_CGROUP_STAT_NSTATS); | |
3143 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3144 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3145 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3146 | ||
af7c4b0e | 3147 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
7941d214 | 3148 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3149 | continue; |
484ebb3b | 3150 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3151 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3152 | } |
7b854121 | 3153 | |
af7c4b0e JW |
3154 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3155 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3156 | mem_cgroup_read_events(memcg, i)); | |
3157 | ||
3158 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3159 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3160 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3161 | ||
14067bb3 | 3162 | /* Hierarchical information */ |
3e32cb2e JW |
3163 | memory = memsw = PAGE_COUNTER_MAX; |
3164 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3165 | memory = min(memory, mi->memory.limit); | |
3166 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3167 | } |
3e32cb2e JW |
3168 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3169 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3170 | if (do_memsw_account()) |
3e32cb2e JW |
3171 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3172 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3173 | |
af7c4b0e | 3174 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3175 | unsigned long long val = 0; |
af7c4b0e | 3176 | |
7941d214 | 3177 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3178 | continue; |
af7c4b0e JW |
3179 | for_each_mem_cgroup_tree(mi, memcg) |
3180 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3181 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3182 | } |
3183 | ||
3184 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3185 | unsigned long long val = 0; | |
3186 | ||
3187 | for_each_mem_cgroup_tree(mi, memcg) | |
3188 | val += mem_cgroup_read_events(mi, i); | |
3189 | seq_printf(m, "total_%s %llu\n", | |
3190 | mem_cgroup_events_names[i], val); | |
3191 | } | |
3192 | ||
3193 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3194 | unsigned long long val = 0; | |
3195 | ||
3196 | for_each_mem_cgroup_tree(mi, memcg) | |
3197 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3198 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3199 | } |
14067bb3 | 3200 | |
7f016ee8 | 3201 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3202 | { |
ef8f2327 MG |
3203 | pg_data_t *pgdat; |
3204 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3205 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3206 | unsigned long recent_rotated[2] = {0, 0}; |
3207 | unsigned long recent_scanned[2] = {0, 0}; | |
3208 | ||
ef8f2327 MG |
3209 | for_each_online_pgdat(pgdat) { |
3210 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3211 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3212 | |
ef8f2327 MG |
3213 | recent_rotated[0] += rstat->recent_rotated[0]; |
3214 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3215 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3216 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3217 | } | |
78ccf5b5 JW |
3218 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3219 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3220 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3221 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3222 | } |
3223 | #endif | |
3224 | ||
d2ceb9b7 KH |
3225 | return 0; |
3226 | } | |
3227 | ||
182446d0 TH |
3228 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3229 | struct cftype *cft) | |
a7885eb8 | 3230 | { |
182446d0 | 3231 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3232 | |
1f4c025b | 3233 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3234 | } |
3235 | ||
182446d0 TH |
3236 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3237 | struct cftype *cft, u64 val) | |
a7885eb8 | 3238 | { |
182446d0 | 3239 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3240 | |
3dae7fec | 3241 | if (val > 100) |
a7885eb8 KM |
3242 | return -EINVAL; |
3243 | ||
14208b0e | 3244 | if (css->parent) |
3dae7fec JW |
3245 | memcg->swappiness = val; |
3246 | else | |
3247 | vm_swappiness = val; | |
068b38c1 | 3248 | |
a7885eb8 KM |
3249 | return 0; |
3250 | } | |
3251 | ||
2e72b634 KS |
3252 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3253 | { | |
3254 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3255 | unsigned long usage; |
2e72b634 KS |
3256 | int i; |
3257 | ||
3258 | rcu_read_lock(); | |
3259 | if (!swap) | |
2c488db2 | 3260 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3261 | else |
2c488db2 | 3262 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3263 | |
3264 | if (!t) | |
3265 | goto unlock; | |
3266 | ||
ce00a967 | 3267 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3268 | |
3269 | /* | |
748dad36 | 3270 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3271 | * If it's not true, a threshold was crossed after last |
3272 | * call of __mem_cgroup_threshold(). | |
3273 | */ | |
5407a562 | 3274 | i = t->current_threshold; |
2e72b634 KS |
3275 | |
3276 | /* | |
3277 | * Iterate backward over array of thresholds starting from | |
3278 | * current_threshold and check if a threshold is crossed. | |
3279 | * If none of thresholds below usage is crossed, we read | |
3280 | * only one element of the array here. | |
3281 | */ | |
3282 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3283 | eventfd_signal(t->entries[i].eventfd, 1); | |
3284 | ||
3285 | /* i = current_threshold + 1 */ | |
3286 | i++; | |
3287 | ||
3288 | /* | |
3289 | * Iterate forward over array of thresholds starting from | |
3290 | * current_threshold+1 and check if a threshold is crossed. | |
3291 | * If none of thresholds above usage is crossed, we read | |
3292 | * only one element of the array here. | |
3293 | */ | |
3294 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3295 | eventfd_signal(t->entries[i].eventfd, 1); | |
3296 | ||
3297 | /* Update current_threshold */ | |
5407a562 | 3298 | t->current_threshold = i - 1; |
2e72b634 KS |
3299 | unlock: |
3300 | rcu_read_unlock(); | |
3301 | } | |
3302 | ||
3303 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3304 | { | |
ad4ca5f4 KS |
3305 | while (memcg) { |
3306 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3307 | if (do_memsw_account()) |
ad4ca5f4 KS |
3308 | __mem_cgroup_threshold(memcg, true); |
3309 | ||
3310 | memcg = parent_mem_cgroup(memcg); | |
3311 | } | |
2e72b634 KS |
3312 | } |
3313 | ||
3314 | static int compare_thresholds(const void *a, const void *b) | |
3315 | { | |
3316 | const struct mem_cgroup_threshold *_a = a; | |
3317 | const struct mem_cgroup_threshold *_b = b; | |
3318 | ||
2bff24a3 GT |
3319 | if (_a->threshold > _b->threshold) |
3320 | return 1; | |
3321 | ||
3322 | if (_a->threshold < _b->threshold) | |
3323 | return -1; | |
3324 | ||
3325 | return 0; | |
2e72b634 KS |
3326 | } |
3327 | ||
c0ff4b85 | 3328 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3329 | { |
3330 | struct mem_cgroup_eventfd_list *ev; | |
3331 | ||
2bcf2e92 MH |
3332 | spin_lock(&memcg_oom_lock); |
3333 | ||
c0ff4b85 | 3334 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3335 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3336 | |
3337 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3338 | return 0; |
3339 | } | |
3340 | ||
c0ff4b85 | 3341 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3342 | { |
7d74b06f KH |
3343 | struct mem_cgroup *iter; |
3344 | ||
c0ff4b85 | 3345 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3346 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3347 | } |
3348 | ||
59b6f873 | 3349 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3350 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3351 | { |
2c488db2 KS |
3352 | struct mem_cgroup_thresholds *thresholds; |
3353 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3354 | unsigned long threshold; |
3355 | unsigned long usage; | |
2c488db2 | 3356 | int i, size, ret; |
2e72b634 | 3357 | |
650c5e56 | 3358 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3359 | if (ret) |
3360 | return ret; | |
3361 | ||
3362 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3363 | |
05b84301 | 3364 | if (type == _MEM) { |
2c488db2 | 3365 | thresholds = &memcg->thresholds; |
ce00a967 | 3366 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3367 | } else if (type == _MEMSWAP) { |
2c488db2 | 3368 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3369 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3370 | } else |
2e72b634 KS |
3371 | BUG(); |
3372 | ||
2e72b634 | 3373 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3374 | if (thresholds->primary) |
2e72b634 KS |
3375 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3376 | ||
2c488db2 | 3377 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3378 | |
3379 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3380 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3381 | GFP_KERNEL); |
2c488db2 | 3382 | if (!new) { |
2e72b634 KS |
3383 | ret = -ENOMEM; |
3384 | goto unlock; | |
3385 | } | |
2c488db2 | 3386 | new->size = size; |
2e72b634 KS |
3387 | |
3388 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3389 | if (thresholds->primary) { |
3390 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3391 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3392 | } |
3393 | ||
2e72b634 | 3394 | /* Add new threshold */ |
2c488db2 KS |
3395 | new->entries[size - 1].eventfd = eventfd; |
3396 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3397 | |
3398 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3399 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3400 | compare_thresholds, NULL); |
3401 | ||
3402 | /* Find current threshold */ | |
2c488db2 | 3403 | new->current_threshold = -1; |
2e72b634 | 3404 | for (i = 0; i < size; i++) { |
748dad36 | 3405 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3406 | /* |
2c488db2 KS |
3407 | * new->current_threshold will not be used until |
3408 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3409 | * it here. |
3410 | */ | |
2c488db2 | 3411 | ++new->current_threshold; |
748dad36 SZ |
3412 | } else |
3413 | break; | |
2e72b634 KS |
3414 | } |
3415 | ||
2c488db2 KS |
3416 | /* Free old spare buffer and save old primary buffer as spare */ |
3417 | kfree(thresholds->spare); | |
3418 | thresholds->spare = thresholds->primary; | |
3419 | ||
3420 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3421 | |
907860ed | 3422 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3423 | synchronize_rcu(); |
3424 | ||
2e72b634 KS |
3425 | unlock: |
3426 | mutex_unlock(&memcg->thresholds_lock); | |
3427 | ||
3428 | return ret; | |
3429 | } | |
3430 | ||
59b6f873 | 3431 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3432 | struct eventfd_ctx *eventfd, const char *args) |
3433 | { | |
59b6f873 | 3434 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3435 | } |
3436 | ||
59b6f873 | 3437 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3438 | struct eventfd_ctx *eventfd, const char *args) |
3439 | { | |
59b6f873 | 3440 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3441 | } |
3442 | ||
59b6f873 | 3443 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3444 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3445 | { |
2c488db2 KS |
3446 | struct mem_cgroup_thresholds *thresholds; |
3447 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3448 | unsigned long usage; |
2c488db2 | 3449 | int i, j, size; |
2e72b634 KS |
3450 | |
3451 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3452 | |
3453 | if (type == _MEM) { | |
2c488db2 | 3454 | thresholds = &memcg->thresholds; |
ce00a967 | 3455 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3456 | } else if (type == _MEMSWAP) { |
2c488db2 | 3457 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3458 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3459 | } else |
2e72b634 KS |
3460 | BUG(); |
3461 | ||
371528ca AV |
3462 | if (!thresholds->primary) |
3463 | goto unlock; | |
3464 | ||
2e72b634 KS |
3465 | /* Check if a threshold crossed before removing */ |
3466 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3467 | ||
3468 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3469 | size = 0; |
3470 | for (i = 0; i < thresholds->primary->size; i++) { | |
3471 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3472 | size++; |
3473 | } | |
3474 | ||
2c488db2 | 3475 | new = thresholds->spare; |
907860ed | 3476 | |
2e72b634 KS |
3477 | /* Set thresholds array to NULL if we don't have thresholds */ |
3478 | if (!size) { | |
2c488db2 KS |
3479 | kfree(new); |
3480 | new = NULL; | |
907860ed | 3481 | goto swap_buffers; |
2e72b634 KS |
3482 | } |
3483 | ||
2c488db2 | 3484 | new->size = size; |
2e72b634 KS |
3485 | |
3486 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3487 | new->current_threshold = -1; |
3488 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3489 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3490 | continue; |
3491 | ||
2c488db2 | 3492 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3493 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3494 | /* |
2c488db2 | 3495 | * new->current_threshold will not be used |
2e72b634 KS |
3496 | * until rcu_assign_pointer(), so it's safe to increment |
3497 | * it here. | |
3498 | */ | |
2c488db2 | 3499 | ++new->current_threshold; |
2e72b634 KS |
3500 | } |
3501 | j++; | |
3502 | } | |
3503 | ||
907860ed | 3504 | swap_buffers: |
2c488db2 KS |
3505 | /* Swap primary and spare array */ |
3506 | thresholds->spare = thresholds->primary; | |
8c757763 | 3507 | |
2c488db2 | 3508 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3509 | |
907860ed | 3510 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3511 | synchronize_rcu(); |
6611d8d7 MC |
3512 | |
3513 | /* If all events are unregistered, free the spare array */ | |
3514 | if (!new) { | |
3515 | kfree(thresholds->spare); | |
3516 | thresholds->spare = NULL; | |
3517 | } | |
371528ca | 3518 | unlock: |
2e72b634 | 3519 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3520 | } |
c1e862c1 | 3521 | |
59b6f873 | 3522 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3523 | struct eventfd_ctx *eventfd) |
3524 | { | |
59b6f873 | 3525 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3526 | } |
3527 | ||
59b6f873 | 3528 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3529 | struct eventfd_ctx *eventfd) |
3530 | { | |
59b6f873 | 3531 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3532 | } |
3533 | ||
59b6f873 | 3534 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3535 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3536 | { |
9490ff27 | 3537 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3538 | |
9490ff27 KH |
3539 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3540 | if (!event) | |
3541 | return -ENOMEM; | |
3542 | ||
1af8efe9 | 3543 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3544 | |
3545 | event->eventfd = eventfd; | |
3546 | list_add(&event->list, &memcg->oom_notify); | |
3547 | ||
3548 | /* already in OOM ? */ | |
c2b42d3c | 3549 | if (memcg->under_oom) |
9490ff27 | 3550 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3551 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3552 | |
3553 | return 0; | |
3554 | } | |
3555 | ||
59b6f873 | 3556 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3557 | struct eventfd_ctx *eventfd) |
9490ff27 | 3558 | { |
9490ff27 | 3559 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3560 | |
1af8efe9 | 3561 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3562 | |
c0ff4b85 | 3563 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3564 | if (ev->eventfd == eventfd) { |
3565 | list_del(&ev->list); | |
3566 | kfree(ev); | |
3567 | } | |
3568 | } | |
3569 | ||
1af8efe9 | 3570 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3571 | } |
3572 | ||
2da8ca82 | 3573 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3574 | { |
2da8ca82 | 3575 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3576 | |
791badbd | 3577 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3578 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3579 | return 0; |
3580 | } | |
3581 | ||
182446d0 | 3582 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3583 | struct cftype *cft, u64 val) |
3584 | { | |
182446d0 | 3585 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3586 | |
3587 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3588 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3589 | return -EINVAL; |
3590 | ||
c0ff4b85 | 3591 | memcg->oom_kill_disable = val; |
4d845ebf | 3592 | if (!val) |
c0ff4b85 | 3593 | memcg_oom_recover(memcg); |
3dae7fec | 3594 | |
3c11ecf4 KH |
3595 | return 0; |
3596 | } | |
3597 | ||
52ebea74 TH |
3598 | #ifdef CONFIG_CGROUP_WRITEBACK |
3599 | ||
3600 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3601 | { | |
3602 | return &memcg->cgwb_list; | |
3603 | } | |
3604 | ||
841710aa TH |
3605 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3606 | { | |
3607 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3608 | } | |
3609 | ||
3610 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3611 | { | |
3612 | wb_domain_exit(&memcg->cgwb_domain); | |
3613 | } | |
3614 | ||
2529bb3a TH |
3615 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3616 | { | |
3617 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3618 | } | |
3619 | ||
841710aa TH |
3620 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3621 | { | |
3622 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3623 | ||
3624 | if (!memcg->css.parent) | |
3625 | return NULL; | |
3626 | ||
3627 | return &memcg->cgwb_domain; | |
3628 | } | |
3629 | ||
c2aa723a TH |
3630 | /** |
3631 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3632 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3633 | * @pfilepages: out parameter for number of file pages |
3634 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3635 | * @pdirty: out parameter for number of dirty pages |
3636 | * @pwriteback: out parameter for number of pages under writeback | |
3637 | * | |
c5edf9cd TH |
3638 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3639 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3640 | * is a bit more involved. | |
c2aa723a | 3641 | * |
c5edf9cd TH |
3642 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3643 | * headroom is calculated as the lowest headroom of itself and the | |
3644 | * ancestors. Note that this doesn't consider the actual amount of | |
3645 | * available memory in the system. The caller should further cap | |
3646 | * *@pheadroom accordingly. | |
c2aa723a | 3647 | */ |
c5edf9cd TH |
3648 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3649 | unsigned long *pheadroom, unsigned long *pdirty, | |
3650 | unsigned long *pwriteback) | |
c2aa723a TH |
3651 | { |
3652 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3653 | struct mem_cgroup *parent; | |
c2aa723a TH |
3654 | |
3655 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3656 | ||
3657 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3658 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3659 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3660 | (1 << LRU_ACTIVE_FILE)); | |
3661 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3662 | |
c2aa723a TH |
3663 | while ((parent = parent_mem_cgroup(memcg))) { |
3664 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3665 | unsigned long used = page_counter_read(&memcg->memory); | |
3666 | ||
c5edf9cd | 3667 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3668 | memcg = parent; |
3669 | } | |
c2aa723a TH |
3670 | } |
3671 | ||
841710aa TH |
3672 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3673 | ||
3674 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3675 | { | |
3676 | return 0; | |
3677 | } | |
3678 | ||
3679 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3680 | { | |
3681 | } | |
3682 | ||
2529bb3a TH |
3683 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3684 | { | |
3685 | } | |
3686 | ||
52ebea74 TH |
3687 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3688 | ||
3bc942f3 TH |
3689 | /* |
3690 | * DO NOT USE IN NEW FILES. | |
3691 | * | |
3692 | * "cgroup.event_control" implementation. | |
3693 | * | |
3694 | * This is way over-engineered. It tries to support fully configurable | |
3695 | * events for each user. Such level of flexibility is completely | |
3696 | * unnecessary especially in the light of the planned unified hierarchy. | |
3697 | * | |
3698 | * Please deprecate this and replace with something simpler if at all | |
3699 | * possible. | |
3700 | */ | |
3701 | ||
79bd9814 TH |
3702 | /* |
3703 | * Unregister event and free resources. | |
3704 | * | |
3705 | * Gets called from workqueue. | |
3706 | */ | |
3bc942f3 | 3707 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3708 | { |
3bc942f3 TH |
3709 | struct mem_cgroup_event *event = |
3710 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3711 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3712 | |
3713 | remove_wait_queue(event->wqh, &event->wait); | |
3714 | ||
59b6f873 | 3715 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3716 | |
3717 | /* Notify userspace the event is going away. */ | |
3718 | eventfd_signal(event->eventfd, 1); | |
3719 | ||
3720 | eventfd_ctx_put(event->eventfd); | |
3721 | kfree(event); | |
59b6f873 | 3722 | css_put(&memcg->css); |
79bd9814 TH |
3723 | } |
3724 | ||
3725 | /* | |
3726 | * Gets called on POLLHUP on eventfd when user closes it. | |
3727 | * | |
3728 | * Called with wqh->lock held and interrupts disabled. | |
3729 | */ | |
3bc942f3 TH |
3730 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3731 | int sync, void *key) | |
79bd9814 | 3732 | { |
3bc942f3 TH |
3733 | struct mem_cgroup_event *event = |
3734 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3735 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3736 | unsigned long flags = (unsigned long)key; |
3737 | ||
3738 | if (flags & POLLHUP) { | |
3739 | /* | |
3740 | * If the event has been detached at cgroup removal, we | |
3741 | * can simply return knowing the other side will cleanup | |
3742 | * for us. | |
3743 | * | |
3744 | * We can't race against event freeing since the other | |
3745 | * side will require wqh->lock via remove_wait_queue(), | |
3746 | * which we hold. | |
3747 | */ | |
fba94807 | 3748 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3749 | if (!list_empty(&event->list)) { |
3750 | list_del_init(&event->list); | |
3751 | /* | |
3752 | * We are in atomic context, but cgroup_event_remove() | |
3753 | * may sleep, so we have to call it in workqueue. | |
3754 | */ | |
3755 | schedule_work(&event->remove); | |
3756 | } | |
fba94807 | 3757 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3758 | } |
3759 | ||
3760 | return 0; | |
3761 | } | |
3762 | ||
3bc942f3 | 3763 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3764 | wait_queue_head_t *wqh, poll_table *pt) |
3765 | { | |
3bc942f3 TH |
3766 | struct mem_cgroup_event *event = |
3767 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3768 | |
3769 | event->wqh = wqh; | |
3770 | add_wait_queue(wqh, &event->wait); | |
3771 | } | |
3772 | ||
3773 | /* | |
3bc942f3 TH |
3774 | * DO NOT USE IN NEW FILES. |
3775 | * | |
79bd9814 TH |
3776 | * Parse input and register new cgroup event handler. |
3777 | * | |
3778 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3779 | * Interpretation of args is defined by control file implementation. | |
3780 | */ | |
451af504 TH |
3781 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3782 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3783 | { |
451af504 | 3784 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3785 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3786 | struct mem_cgroup_event *event; |
79bd9814 TH |
3787 | struct cgroup_subsys_state *cfile_css; |
3788 | unsigned int efd, cfd; | |
3789 | struct fd efile; | |
3790 | struct fd cfile; | |
fba94807 | 3791 | const char *name; |
79bd9814 TH |
3792 | char *endp; |
3793 | int ret; | |
3794 | ||
451af504 TH |
3795 | buf = strstrip(buf); |
3796 | ||
3797 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3798 | if (*endp != ' ') |
3799 | return -EINVAL; | |
451af504 | 3800 | buf = endp + 1; |
79bd9814 | 3801 | |
451af504 | 3802 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3803 | if ((*endp != ' ') && (*endp != '\0')) |
3804 | return -EINVAL; | |
451af504 | 3805 | buf = endp + 1; |
79bd9814 TH |
3806 | |
3807 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3808 | if (!event) | |
3809 | return -ENOMEM; | |
3810 | ||
59b6f873 | 3811 | event->memcg = memcg; |
79bd9814 | 3812 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3813 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3814 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3815 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3816 | |
3817 | efile = fdget(efd); | |
3818 | if (!efile.file) { | |
3819 | ret = -EBADF; | |
3820 | goto out_kfree; | |
3821 | } | |
3822 | ||
3823 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3824 | if (IS_ERR(event->eventfd)) { | |
3825 | ret = PTR_ERR(event->eventfd); | |
3826 | goto out_put_efile; | |
3827 | } | |
3828 | ||
3829 | cfile = fdget(cfd); | |
3830 | if (!cfile.file) { | |
3831 | ret = -EBADF; | |
3832 | goto out_put_eventfd; | |
3833 | } | |
3834 | ||
3835 | /* the process need read permission on control file */ | |
3836 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3837 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3838 | if (ret < 0) | |
3839 | goto out_put_cfile; | |
3840 | ||
fba94807 TH |
3841 | /* |
3842 | * Determine the event callbacks and set them in @event. This used | |
3843 | * to be done via struct cftype but cgroup core no longer knows | |
3844 | * about these events. The following is crude but the whole thing | |
3845 | * is for compatibility anyway. | |
3bc942f3 TH |
3846 | * |
3847 | * DO NOT ADD NEW FILES. | |
fba94807 | 3848 | */ |
b583043e | 3849 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3850 | |
3851 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3852 | event->register_event = mem_cgroup_usage_register_event; | |
3853 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3854 | } else if (!strcmp(name, "memory.oom_control")) { | |
3855 | event->register_event = mem_cgroup_oom_register_event; | |
3856 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3857 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3858 | event->register_event = vmpressure_register_event; | |
3859 | event->unregister_event = vmpressure_unregister_event; | |
3860 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3861 | event->register_event = memsw_cgroup_usage_register_event; |
3862 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3863 | } else { |
3864 | ret = -EINVAL; | |
3865 | goto out_put_cfile; | |
3866 | } | |
3867 | ||
79bd9814 | 3868 | /* |
b5557c4c TH |
3869 | * Verify @cfile should belong to @css. Also, remaining events are |
3870 | * automatically removed on cgroup destruction but the removal is | |
3871 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3872 | */ |
b583043e | 3873 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3874 | &memory_cgrp_subsys); |
79bd9814 | 3875 | ret = -EINVAL; |
5a17f543 | 3876 | if (IS_ERR(cfile_css)) |
79bd9814 | 3877 | goto out_put_cfile; |
5a17f543 TH |
3878 | if (cfile_css != css) { |
3879 | css_put(cfile_css); | |
79bd9814 | 3880 | goto out_put_cfile; |
5a17f543 | 3881 | } |
79bd9814 | 3882 | |
451af504 | 3883 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3884 | if (ret) |
3885 | goto out_put_css; | |
3886 | ||
3887 | efile.file->f_op->poll(efile.file, &event->pt); | |
3888 | ||
fba94807 TH |
3889 | spin_lock(&memcg->event_list_lock); |
3890 | list_add(&event->list, &memcg->event_list); | |
3891 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3892 | |
3893 | fdput(cfile); | |
3894 | fdput(efile); | |
3895 | ||
451af504 | 3896 | return nbytes; |
79bd9814 TH |
3897 | |
3898 | out_put_css: | |
b5557c4c | 3899 | css_put(css); |
79bd9814 TH |
3900 | out_put_cfile: |
3901 | fdput(cfile); | |
3902 | out_put_eventfd: | |
3903 | eventfd_ctx_put(event->eventfd); | |
3904 | out_put_efile: | |
3905 | fdput(efile); | |
3906 | out_kfree: | |
3907 | kfree(event); | |
3908 | ||
3909 | return ret; | |
3910 | } | |
3911 | ||
241994ed | 3912 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 3913 | { |
0eea1030 | 3914 | .name = "usage_in_bytes", |
8c7c6e34 | 3915 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 3916 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3917 | }, |
c84872e1 PE |
3918 | { |
3919 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3920 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 3921 | .write = mem_cgroup_reset, |
791badbd | 3922 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 3923 | }, |
8cdea7c0 | 3924 | { |
0eea1030 | 3925 | .name = "limit_in_bytes", |
8c7c6e34 | 3926 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 3927 | .write = mem_cgroup_write, |
791badbd | 3928 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3929 | }, |
296c81d8 BS |
3930 | { |
3931 | .name = "soft_limit_in_bytes", | |
3932 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 3933 | .write = mem_cgroup_write, |
791badbd | 3934 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 3935 | }, |
8cdea7c0 BS |
3936 | { |
3937 | .name = "failcnt", | |
8c7c6e34 | 3938 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 3939 | .write = mem_cgroup_reset, |
791badbd | 3940 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3941 | }, |
d2ceb9b7 KH |
3942 | { |
3943 | .name = "stat", | |
2da8ca82 | 3944 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 3945 | }, |
c1e862c1 KH |
3946 | { |
3947 | .name = "force_empty", | |
6770c64e | 3948 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 3949 | }, |
18f59ea7 BS |
3950 | { |
3951 | .name = "use_hierarchy", | |
3952 | .write_u64 = mem_cgroup_hierarchy_write, | |
3953 | .read_u64 = mem_cgroup_hierarchy_read, | |
3954 | }, | |
79bd9814 | 3955 | { |
3bc942f3 | 3956 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 3957 | .write = memcg_write_event_control, |
7dbdb199 | 3958 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 3959 | }, |
a7885eb8 KM |
3960 | { |
3961 | .name = "swappiness", | |
3962 | .read_u64 = mem_cgroup_swappiness_read, | |
3963 | .write_u64 = mem_cgroup_swappiness_write, | |
3964 | }, | |
7dc74be0 DN |
3965 | { |
3966 | .name = "move_charge_at_immigrate", | |
3967 | .read_u64 = mem_cgroup_move_charge_read, | |
3968 | .write_u64 = mem_cgroup_move_charge_write, | |
3969 | }, | |
9490ff27 KH |
3970 | { |
3971 | .name = "oom_control", | |
2da8ca82 | 3972 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 3973 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
3974 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
3975 | }, | |
70ddf637 AV |
3976 | { |
3977 | .name = "pressure_level", | |
70ddf637 | 3978 | }, |
406eb0c9 YH |
3979 | #ifdef CONFIG_NUMA |
3980 | { | |
3981 | .name = "numa_stat", | |
2da8ca82 | 3982 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
3983 | }, |
3984 | #endif | |
510fc4e1 GC |
3985 | { |
3986 | .name = "kmem.limit_in_bytes", | |
3987 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 3988 | .write = mem_cgroup_write, |
791badbd | 3989 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
3990 | }, |
3991 | { | |
3992 | .name = "kmem.usage_in_bytes", | |
3993 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 3994 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
3995 | }, |
3996 | { | |
3997 | .name = "kmem.failcnt", | |
3998 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 3999 | .write = mem_cgroup_reset, |
791badbd | 4000 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4001 | }, |
4002 | { | |
4003 | .name = "kmem.max_usage_in_bytes", | |
4004 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4005 | .write = mem_cgroup_reset, |
791badbd | 4006 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4007 | }, |
749c5415 GC |
4008 | #ifdef CONFIG_SLABINFO |
4009 | { | |
4010 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4011 | .seq_start = memcg_slab_start, |
4012 | .seq_next = memcg_slab_next, | |
4013 | .seq_stop = memcg_slab_stop, | |
b047501c | 4014 | .seq_show = memcg_slab_show, |
749c5415 GC |
4015 | }, |
4016 | #endif | |
d55f90bf VD |
4017 | { |
4018 | .name = "kmem.tcp.limit_in_bytes", | |
4019 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4020 | .write = mem_cgroup_write, | |
4021 | .read_u64 = mem_cgroup_read_u64, | |
4022 | }, | |
4023 | { | |
4024 | .name = "kmem.tcp.usage_in_bytes", | |
4025 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4026 | .read_u64 = mem_cgroup_read_u64, | |
4027 | }, | |
4028 | { | |
4029 | .name = "kmem.tcp.failcnt", | |
4030 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4031 | .write = mem_cgroup_reset, | |
4032 | .read_u64 = mem_cgroup_read_u64, | |
4033 | }, | |
4034 | { | |
4035 | .name = "kmem.tcp.max_usage_in_bytes", | |
4036 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4037 | .write = mem_cgroup_reset, | |
4038 | .read_u64 = mem_cgroup_read_u64, | |
4039 | }, | |
6bc10349 | 4040 | { }, /* terminate */ |
af36f906 | 4041 | }; |
8c7c6e34 | 4042 | |
73f576c0 JW |
4043 | /* |
4044 | * Private memory cgroup IDR | |
4045 | * | |
4046 | * Swap-out records and page cache shadow entries need to store memcg | |
4047 | * references in constrained space, so we maintain an ID space that is | |
4048 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4049 | * memory-controlled cgroups to 64k. | |
4050 | * | |
4051 | * However, there usually are many references to the oflline CSS after | |
4052 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4053 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4054 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4055 | * relatively small ID space and prevent the creation of new cgroups | |
4056 | * even when there are much fewer than 64k cgroups - possibly none. | |
4057 | * | |
4058 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4059 | * be freed and recycled when it's no longer needed, which is usually | |
4060 | * when the CSS is offlined. | |
4061 | * | |
4062 | * The only exception to that are records of swapped out tmpfs/shmem | |
4063 | * pages that need to be attributed to live ancestors on swapin. But | |
4064 | * those references are manageable from userspace. | |
4065 | */ | |
4066 | ||
4067 | static DEFINE_IDR(mem_cgroup_idr); | |
4068 | ||
615d66c3 | 4069 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4070 | { |
58fa2a55 | 4071 | VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0); |
615d66c3 | 4072 | atomic_add(n, &memcg->id.ref); |
73f576c0 JW |
4073 | } |
4074 | ||
615d66c3 | 4075 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4076 | { |
58fa2a55 | 4077 | VM_BUG_ON(atomic_read(&memcg->id.ref) < n); |
615d66c3 | 4078 | if (atomic_sub_and_test(n, &memcg->id.ref)) { |
73f576c0 JW |
4079 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
4080 | memcg->id.id = 0; | |
4081 | ||
4082 | /* Memcg ID pins CSS */ | |
4083 | css_put(&memcg->css); | |
4084 | } | |
4085 | } | |
4086 | ||
615d66c3 VD |
4087 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4088 | { | |
4089 | mem_cgroup_id_get_many(memcg, 1); | |
4090 | } | |
4091 | ||
4092 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4093 | { | |
4094 | mem_cgroup_id_put_many(memcg, 1); | |
4095 | } | |
4096 | ||
73f576c0 JW |
4097 | /** |
4098 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4099 | * @id: the memcg id to look up | |
4100 | * | |
4101 | * Caller must hold rcu_read_lock(). | |
4102 | */ | |
4103 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4104 | { | |
4105 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4106 | return idr_find(&mem_cgroup_idr, id); | |
4107 | } | |
4108 | ||
ef8f2327 | 4109 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4110 | { |
4111 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4112 | int tmp = node; |
1ecaab2b KH |
4113 | /* |
4114 | * This routine is called against possible nodes. | |
4115 | * But it's BUG to call kmalloc() against offline node. | |
4116 | * | |
4117 | * TODO: this routine can waste much memory for nodes which will | |
4118 | * never be onlined. It's better to use memory hotplug callback | |
4119 | * function. | |
4120 | */ | |
41e3355d KH |
4121 | if (!node_state(node, N_NORMAL_MEMORY)) |
4122 | tmp = -1; | |
17295c88 | 4123 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4124 | if (!pn) |
4125 | return 1; | |
1ecaab2b | 4126 | |
ef8f2327 MG |
4127 | lruvec_init(&pn->lruvec); |
4128 | pn->usage_in_excess = 0; | |
4129 | pn->on_tree = false; | |
4130 | pn->memcg = memcg; | |
4131 | ||
54f72fe0 | 4132 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4133 | return 0; |
4134 | } | |
4135 | ||
ef8f2327 | 4136 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4137 | { |
54f72fe0 | 4138 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4139 | } |
4140 | ||
0b8f73e1 | 4141 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4142 | { |
c8b2a36f | 4143 | int node; |
59927fb9 | 4144 | |
0b8f73e1 | 4145 | memcg_wb_domain_exit(memcg); |
c8b2a36f | 4146 | for_each_node(node) |
ef8f2327 | 4147 | free_mem_cgroup_per_node_info(memcg, node); |
c8b2a36f | 4148 | free_percpu(memcg->stat); |
8ff69e2c | 4149 | kfree(memcg); |
59927fb9 | 4150 | } |
3afe36b1 | 4151 | |
0b8f73e1 | 4152 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4153 | { |
d142e3e6 | 4154 | struct mem_cgroup *memcg; |
0b8f73e1 | 4155 | size_t size; |
6d12e2d8 | 4156 | int node; |
8cdea7c0 | 4157 | |
0b8f73e1 JW |
4158 | size = sizeof(struct mem_cgroup); |
4159 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4160 | ||
4161 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4162 | if (!memcg) |
0b8f73e1 JW |
4163 | return NULL; |
4164 | ||
73f576c0 JW |
4165 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4166 | 1, MEM_CGROUP_ID_MAX, | |
4167 | GFP_KERNEL); | |
4168 | if (memcg->id.id < 0) | |
4169 | goto fail; | |
4170 | ||
0b8f73e1 JW |
4171 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4172 | if (!memcg->stat) | |
4173 | goto fail; | |
78fb7466 | 4174 | |
3ed28fa1 | 4175 | for_each_node(node) |
ef8f2327 | 4176 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4177 | goto fail; |
f64c3f54 | 4178 | |
0b8f73e1 JW |
4179 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4180 | goto fail; | |
28dbc4b6 | 4181 | |
f7e1cb6e | 4182 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4183 | memcg->last_scanned_node = MAX_NUMNODES; |
4184 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4185 | mutex_init(&memcg->thresholds_lock); |
4186 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4187 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4188 | INIT_LIST_HEAD(&memcg->event_list); |
4189 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4190 | memcg->socket_pressure = jiffies; |
127424c8 | 4191 | #ifndef CONFIG_SLOB |
900a38f0 | 4192 | memcg->kmemcg_id = -1; |
900a38f0 | 4193 | #endif |
52ebea74 TH |
4194 | #ifdef CONFIG_CGROUP_WRITEBACK |
4195 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4196 | #endif | |
73f576c0 | 4197 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4198 | return memcg; |
4199 | fail: | |
73f576c0 JW |
4200 | if (memcg->id.id > 0) |
4201 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
0b8f73e1 JW |
4202 | mem_cgroup_free(memcg); |
4203 | return NULL; | |
d142e3e6 GC |
4204 | } |
4205 | ||
0b8f73e1 JW |
4206 | static struct cgroup_subsys_state * __ref |
4207 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4208 | { |
0b8f73e1 JW |
4209 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4210 | struct mem_cgroup *memcg; | |
4211 | long error = -ENOMEM; | |
d142e3e6 | 4212 | |
0b8f73e1 JW |
4213 | memcg = mem_cgroup_alloc(); |
4214 | if (!memcg) | |
4215 | return ERR_PTR(error); | |
d142e3e6 | 4216 | |
0b8f73e1 JW |
4217 | memcg->high = PAGE_COUNTER_MAX; |
4218 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4219 | if (parent) { | |
4220 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4221 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4222 | } | |
4223 | if (parent && parent->use_hierarchy) { | |
4224 | memcg->use_hierarchy = true; | |
3e32cb2e | 4225 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4226 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4227 | page_counter_init(&memcg->memsw, &parent->memsw); |
4228 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4229 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4230 | } else { |
3e32cb2e | 4231 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4232 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4233 | page_counter_init(&memcg->memsw, NULL); |
4234 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4235 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4236 | /* |
4237 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4238 | * much sense so let cgroup subsystem know about this | |
4239 | * unfortunate state in our controller. | |
4240 | */ | |
d142e3e6 | 4241 | if (parent != root_mem_cgroup) |
073219e9 | 4242 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4243 | } |
d6441637 | 4244 | |
0b8f73e1 JW |
4245 | /* The following stuff does not apply to the root */ |
4246 | if (!parent) { | |
4247 | root_mem_cgroup = memcg; | |
4248 | return &memcg->css; | |
4249 | } | |
4250 | ||
b313aeee | 4251 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4252 | if (error) |
4253 | goto fail; | |
127424c8 | 4254 | |
f7e1cb6e | 4255 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4256 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4257 | |
0b8f73e1 JW |
4258 | return &memcg->css; |
4259 | fail: | |
4260 | mem_cgroup_free(memcg); | |
ea3a9645 | 4261 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4262 | } |
4263 | ||
73f576c0 | 4264 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4265 | { |
58fa2a55 VD |
4266 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4267 | ||
73f576c0 | 4268 | /* Online state pins memcg ID, memcg ID pins CSS */ |
58fa2a55 | 4269 | atomic_set(&memcg->id.ref, 1); |
73f576c0 | 4270 | css_get(css); |
2f7dd7a4 | 4271 | return 0; |
8cdea7c0 BS |
4272 | } |
4273 | ||
eb95419b | 4274 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4275 | { |
eb95419b | 4276 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4277 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4278 | |
4279 | /* | |
4280 | * Unregister events and notify userspace. | |
4281 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4282 | * directory to avoid race between userspace and kernelspace. | |
4283 | */ | |
fba94807 TH |
4284 | spin_lock(&memcg->event_list_lock); |
4285 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4286 | list_del_init(&event->list); |
4287 | schedule_work(&event->remove); | |
4288 | } | |
fba94807 | 4289 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4290 | |
567e9ab2 | 4291 | memcg_offline_kmem(memcg); |
52ebea74 | 4292 | wb_memcg_offline(memcg); |
73f576c0 JW |
4293 | |
4294 | mem_cgroup_id_put(memcg); | |
df878fb0 KH |
4295 | } |
4296 | ||
6df38689 VD |
4297 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4298 | { | |
4299 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4300 | ||
4301 | invalidate_reclaim_iterators(memcg); | |
4302 | } | |
4303 | ||
eb95419b | 4304 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4305 | { |
eb95419b | 4306 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4307 | |
f7e1cb6e | 4308 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4309 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4310 | |
0db15298 | 4311 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4312 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4313 | |
0b8f73e1 JW |
4314 | vmpressure_cleanup(&memcg->vmpressure); |
4315 | cancel_work_sync(&memcg->high_work); | |
4316 | mem_cgroup_remove_from_trees(memcg); | |
d886f4e4 | 4317 | memcg_free_kmem(memcg); |
0b8f73e1 | 4318 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4319 | } |
4320 | ||
1ced953b TH |
4321 | /** |
4322 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4323 | * @css: the target css | |
4324 | * | |
4325 | * Reset the states of the mem_cgroup associated with @css. This is | |
4326 | * invoked when the userland requests disabling on the default hierarchy | |
4327 | * but the memcg is pinned through dependency. The memcg should stop | |
4328 | * applying policies and should revert to the vanilla state as it may be | |
4329 | * made visible again. | |
4330 | * | |
4331 | * The current implementation only resets the essential configurations. | |
4332 | * This needs to be expanded to cover all the visible parts. | |
4333 | */ | |
4334 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4335 | { | |
4336 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4337 | ||
d334c9bc VD |
4338 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
4339 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); | |
4340 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); | |
4341 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); | |
4342 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
241994ed JW |
4343 | memcg->low = 0; |
4344 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4345 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4346 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4347 | } |
4348 | ||
02491447 | 4349 | #ifdef CONFIG_MMU |
7dc74be0 | 4350 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4351 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4352 | { |
05b84301 | 4353 | int ret; |
9476db97 | 4354 | |
d0164adc MG |
4355 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4356 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4357 | if (!ret) { |
854ffa8d | 4358 | mc.precharge += count; |
854ffa8d DN |
4359 | return ret; |
4360 | } | |
9476db97 | 4361 | |
3674534b | 4362 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4363 | while (count--) { |
3674534b | 4364 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4365 | if (ret) |
38c5d72f | 4366 | return ret; |
854ffa8d | 4367 | mc.precharge++; |
9476db97 | 4368 | cond_resched(); |
854ffa8d | 4369 | } |
9476db97 | 4370 | return 0; |
4ffef5fe DN |
4371 | } |
4372 | ||
4ffef5fe DN |
4373 | union mc_target { |
4374 | struct page *page; | |
02491447 | 4375 | swp_entry_t ent; |
4ffef5fe DN |
4376 | }; |
4377 | ||
4ffef5fe | 4378 | enum mc_target_type { |
8d32ff84 | 4379 | MC_TARGET_NONE = 0, |
4ffef5fe | 4380 | MC_TARGET_PAGE, |
02491447 | 4381 | MC_TARGET_SWAP, |
4ffef5fe DN |
4382 | }; |
4383 | ||
90254a65 DN |
4384 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4385 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4386 | { |
90254a65 | 4387 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4388 | |
90254a65 DN |
4389 | if (!page || !page_mapped(page)) |
4390 | return NULL; | |
4391 | if (PageAnon(page)) { | |
1dfab5ab | 4392 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4393 | return NULL; |
1dfab5ab JW |
4394 | } else { |
4395 | if (!(mc.flags & MOVE_FILE)) | |
4396 | return NULL; | |
4397 | } | |
90254a65 DN |
4398 | if (!get_page_unless_zero(page)) |
4399 | return NULL; | |
4400 | ||
4401 | return page; | |
4402 | } | |
4403 | ||
4b91355e | 4404 | #ifdef CONFIG_SWAP |
90254a65 | 4405 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4406 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4407 | { |
90254a65 DN |
4408 | struct page *page = NULL; |
4409 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4410 | ||
1dfab5ab | 4411 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4412 | return NULL; |
4b91355e KH |
4413 | /* |
4414 | * Because lookup_swap_cache() updates some statistics counter, | |
4415 | * we call find_get_page() with swapper_space directly. | |
4416 | */ | |
f6ab1f7f | 4417 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4418 | if (do_memsw_account()) |
90254a65 DN |
4419 | entry->val = ent.val; |
4420 | ||
4421 | return page; | |
4422 | } | |
4b91355e KH |
4423 | #else |
4424 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4425 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4426 | { |
4427 | return NULL; | |
4428 | } | |
4429 | #endif | |
90254a65 | 4430 | |
87946a72 DN |
4431 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4432 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4433 | { | |
4434 | struct page *page = NULL; | |
87946a72 DN |
4435 | struct address_space *mapping; |
4436 | pgoff_t pgoff; | |
4437 | ||
4438 | if (!vma->vm_file) /* anonymous vma */ | |
4439 | return NULL; | |
1dfab5ab | 4440 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4441 | return NULL; |
4442 | ||
87946a72 | 4443 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4444 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4445 | |
4446 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4447 | #ifdef CONFIG_SWAP |
4448 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4449 | if (shmem_mapping(mapping)) { |
4450 | page = find_get_entry(mapping, pgoff); | |
4451 | if (radix_tree_exceptional_entry(page)) { | |
4452 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4453 | if (do_memsw_account()) |
139b6a6f | 4454 | *entry = swp; |
f6ab1f7f HY |
4455 | page = find_get_page(swap_address_space(swp), |
4456 | swp_offset(swp)); | |
139b6a6f JW |
4457 | } |
4458 | } else | |
4459 | page = find_get_page(mapping, pgoff); | |
4460 | #else | |
4461 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4462 | #endif |
87946a72 DN |
4463 | return page; |
4464 | } | |
4465 | ||
b1b0deab CG |
4466 | /** |
4467 | * mem_cgroup_move_account - move account of the page | |
4468 | * @page: the page | |
25843c2b | 4469 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4470 | * @from: mem_cgroup which the page is moved from. |
4471 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4472 | * | |
3ac808fd | 4473 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4474 | * |
4475 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4476 | * from old cgroup. | |
4477 | */ | |
4478 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4479 | bool compound, |
b1b0deab CG |
4480 | struct mem_cgroup *from, |
4481 | struct mem_cgroup *to) | |
4482 | { | |
4483 | unsigned long flags; | |
f627c2f5 | 4484 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4485 | int ret; |
c4843a75 | 4486 | bool anon; |
b1b0deab CG |
4487 | |
4488 | VM_BUG_ON(from == to); | |
4489 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4490 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4491 | |
4492 | /* | |
6a93ca8f | 4493 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4494 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4495 | */ |
f627c2f5 | 4496 | ret = -EBUSY; |
b1b0deab CG |
4497 | if (!trylock_page(page)) |
4498 | goto out; | |
4499 | ||
4500 | ret = -EINVAL; | |
4501 | if (page->mem_cgroup != from) | |
4502 | goto out_unlock; | |
4503 | ||
c4843a75 GT |
4504 | anon = PageAnon(page); |
4505 | ||
b1b0deab CG |
4506 | spin_lock_irqsave(&from->move_lock, flags); |
4507 | ||
c4843a75 | 4508 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4509 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4510 | nr_pages); | |
4511 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4512 | nr_pages); | |
4513 | } | |
4514 | ||
c4843a75 GT |
4515 | /* |
4516 | * move_lock grabbed above and caller set from->moving_account, so | |
4517 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4518 | * So mapping should be stable for dirty pages. | |
4519 | */ | |
4520 | if (!anon && PageDirty(page)) { | |
4521 | struct address_space *mapping = page_mapping(page); | |
4522 | ||
4523 | if (mapping_cap_account_dirty(mapping)) { | |
4524 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4525 | nr_pages); | |
4526 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4527 | nr_pages); | |
4528 | } | |
4529 | } | |
4530 | ||
b1b0deab CG |
4531 | if (PageWriteback(page)) { |
4532 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4533 | nr_pages); | |
4534 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4535 | nr_pages); | |
4536 | } | |
4537 | ||
4538 | /* | |
4539 | * It is safe to change page->mem_cgroup here because the page | |
4540 | * is referenced, charged, and isolated - we can't race with | |
4541 | * uncharging, charging, migration, or LRU putback. | |
4542 | */ | |
4543 | ||
4544 | /* caller should have done css_get */ | |
4545 | page->mem_cgroup = to; | |
4546 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4547 | ||
4548 | ret = 0; | |
4549 | ||
4550 | local_irq_disable(); | |
f627c2f5 | 4551 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4552 | memcg_check_events(to, page); |
f627c2f5 | 4553 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4554 | memcg_check_events(from, page); |
4555 | local_irq_enable(); | |
4556 | out_unlock: | |
4557 | unlock_page(page); | |
4558 | out: | |
4559 | return ret; | |
4560 | } | |
4561 | ||
7cf7806c LR |
4562 | /** |
4563 | * get_mctgt_type - get target type of moving charge | |
4564 | * @vma: the vma the pte to be checked belongs | |
4565 | * @addr: the address corresponding to the pte to be checked | |
4566 | * @ptent: the pte to be checked | |
4567 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4568 | * | |
4569 | * Returns | |
4570 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4571 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4572 | * move charge. if @target is not NULL, the page is stored in target->page | |
4573 | * with extra refcnt got(Callers should handle it). | |
4574 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4575 | * target for charge migration. if @target is not NULL, the entry is stored | |
4576 | * in target->ent. | |
4577 | * | |
4578 | * Called with pte lock held. | |
4579 | */ | |
4580 | ||
8d32ff84 | 4581 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4582 | unsigned long addr, pte_t ptent, union mc_target *target) |
4583 | { | |
4584 | struct page *page = NULL; | |
8d32ff84 | 4585 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4586 | swp_entry_t ent = { .val = 0 }; |
4587 | ||
4588 | if (pte_present(ptent)) | |
4589 | page = mc_handle_present_pte(vma, addr, ptent); | |
4590 | else if (is_swap_pte(ptent)) | |
48406ef8 | 4591 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 4592 | else if (pte_none(ptent)) |
87946a72 | 4593 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4594 | |
4595 | if (!page && !ent.val) | |
8d32ff84 | 4596 | return ret; |
02491447 | 4597 | if (page) { |
02491447 | 4598 | /* |
0a31bc97 | 4599 | * Do only loose check w/o serialization. |
1306a85a | 4600 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4601 | * not under LRU exclusion. |
02491447 | 4602 | */ |
1306a85a | 4603 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4604 | ret = MC_TARGET_PAGE; |
4605 | if (target) | |
4606 | target->page = page; | |
4607 | } | |
4608 | if (!ret || !target) | |
4609 | put_page(page); | |
4610 | } | |
90254a65 DN |
4611 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4612 | if (ent.val && !ret && | |
34c00c31 | 4613 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4614 | ret = MC_TARGET_SWAP; |
4615 | if (target) | |
4616 | target->ent = ent; | |
4ffef5fe | 4617 | } |
4ffef5fe DN |
4618 | return ret; |
4619 | } | |
4620 | ||
12724850 NH |
4621 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4622 | /* | |
4623 | * We don't consider swapping or file mapped pages because THP does not | |
4624 | * support them for now. | |
4625 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4626 | */ | |
4627 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4628 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4629 | { | |
4630 | struct page *page = NULL; | |
12724850 NH |
4631 | enum mc_target_type ret = MC_TARGET_NONE; |
4632 | ||
4633 | page = pmd_page(pmd); | |
309381fe | 4634 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4635 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4636 | return ret; |
1306a85a | 4637 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4638 | ret = MC_TARGET_PAGE; |
4639 | if (target) { | |
4640 | get_page(page); | |
4641 | target->page = page; | |
4642 | } | |
4643 | } | |
4644 | return ret; | |
4645 | } | |
4646 | #else | |
4647 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4648 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4649 | { | |
4650 | return MC_TARGET_NONE; | |
4651 | } | |
4652 | #endif | |
4653 | ||
4ffef5fe DN |
4654 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4655 | unsigned long addr, unsigned long end, | |
4656 | struct mm_walk *walk) | |
4657 | { | |
26bcd64a | 4658 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4659 | pte_t *pte; |
4660 | spinlock_t *ptl; | |
4661 | ||
b6ec57f4 KS |
4662 | ptl = pmd_trans_huge_lock(pmd, vma); |
4663 | if (ptl) { | |
12724850 NH |
4664 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4665 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4666 | spin_unlock(ptl); |
1a5a9906 | 4667 | return 0; |
12724850 | 4668 | } |
03319327 | 4669 | |
45f83cef AA |
4670 | if (pmd_trans_unstable(pmd)) |
4671 | return 0; | |
4ffef5fe DN |
4672 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4673 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4674 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4675 | mc.precharge++; /* increment precharge temporarily */ |
4676 | pte_unmap_unlock(pte - 1, ptl); | |
4677 | cond_resched(); | |
4678 | ||
7dc74be0 DN |
4679 | return 0; |
4680 | } | |
4681 | ||
4ffef5fe DN |
4682 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4683 | { | |
4684 | unsigned long precharge; | |
4ffef5fe | 4685 | |
26bcd64a NH |
4686 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4687 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4688 | .mm = mm, | |
4689 | }; | |
dfe076b0 | 4690 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
4691 | walk_page_range(0, mm->highest_vm_end, |
4692 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 4693 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4694 | |
4695 | precharge = mc.precharge; | |
4696 | mc.precharge = 0; | |
4697 | ||
4698 | return precharge; | |
4699 | } | |
4700 | ||
4ffef5fe DN |
4701 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4702 | { | |
dfe076b0 DN |
4703 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4704 | ||
4705 | VM_BUG_ON(mc.moving_task); | |
4706 | mc.moving_task = current; | |
4707 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4708 | } |
4709 | ||
dfe076b0 DN |
4710 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4711 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4712 | { |
2bd9bb20 KH |
4713 | struct mem_cgroup *from = mc.from; |
4714 | struct mem_cgroup *to = mc.to; | |
4715 | ||
4ffef5fe | 4716 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4717 | if (mc.precharge) { |
00501b53 | 4718 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4719 | mc.precharge = 0; |
4720 | } | |
4721 | /* | |
4722 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4723 | * we must uncharge here. | |
4724 | */ | |
4725 | if (mc.moved_charge) { | |
00501b53 | 4726 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4727 | mc.moved_charge = 0; |
4ffef5fe | 4728 | } |
483c30b5 DN |
4729 | /* we must fixup refcnts and charges */ |
4730 | if (mc.moved_swap) { | |
483c30b5 | 4731 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4732 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4733 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4734 | |
615d66c3 VD |
4735 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
4736 | ||
05b84301 | 4737 | /* |
3e32cb2e JW |
4738 | * we charged both to->memory and to->memsw, so we |
4739 | * should uncharge to->memory. | |
05b84301 | 4740 | */ |
ce00a967 | 4741 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4742 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4743 | ||
615d66c3 VD |
4744 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
4745 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 4746 | |
483c30b5 DN |
4747 | mc.moved_swap = 0; |
4748 | } | |
dfe076b0 DN |
4749 | memcg_oom_recover(from); |
4750 | memcg_oom_recover(to); | |
4751 | wake_up_all(&mc.waitq); | |
4752 | } | |
4753 | ||
4754 | static void mem_cgroup_clear_mc(void) | |
4755 | { | |
264a0ae1 TH |
4756 | struct mm_struct *mm = mc.mm; |
4757 | ||
dfe076b0 DN |
4758 | /* |
4759 | * we must clear moving_task before waking up waiters at the end of | |
4760 | * task migration. | |
4761 | */ | |
4762 | mc.moving_task = NULL; | |
4763 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4764 | spin_lock(&mc.lock); |
4ffef5fe DN |
4765 | mc.from = NULL; |
4766 | mc.to = NULL; | |
264a0ae1 | 4767 | mc.mm = NULL; |
2bd9bb20 | 4768 | spin_unlock(&mc.lock); |
264a0ae1 TH |
4769 | |
4770 | mmput(mm); | |
4ffef5fe DN |
4771 | } |
4772 | ||
1f7dd3e5 | 4773 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4774 | { |
1f7dd3e5 | 4775 | struct cgroup_subsys_state *css; |
eed67d75 | 4776 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4777 | struct mem_cgroup *from; |
4530eddb | 4778 | struct task_struct *leader, *p; |
9f2115f9 | 4779 | struct mm_struct *mm; |
1dfab5ab | 4780 | unsigned long move_flags; |
9f2115f9 | 4781 | int ret = 0; |
7dc74be0 | 4782 | |
1f7dd3e5 TH |
4783 | /* charge immigration isn't supported on the default hierarchy */ |
4784 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4785 | return 0; |
4786 | ||
4530eddb TH |
4787 | /* |
4788 | * Multi-process migrations only happen on the default hierarchy | |
4789 | * where charge immigration is not used. Perform charge | |
4790 | * immigration if @tset contains a leader and whine if there are | |
4791 | * multiple. | |
4792 | */ | |
4793 | p = NULL; | |
1f7dd3e5 | 4794 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4795 | WARN_ON_ONCE(p); |
4796 | p = leader; | |
1f7dd3e5 | 4797 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4798 | } |
4799 | if (!p) | |
4800 | return 0; | |
4801 | ||
1f7dd3e5 TH |
4802 | /* |
4803 | * We are now commited to this value whatever it is. Changes in this | |
4804 | * tunable will only affect upcoming migrations, not the current one. | |
4805 | * So we need to save it, and keep it going. | |
4806 | */ | |
4807 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4808 | if (!move_flags) | |
4809 | return 0; | |
4810 | ||
9f2115f9 TH |
4811 | from = mem_cgroup_from_task(p); |
4812 | ||
4813 | VM_BUG_ON(from == memcg); | |
4814 | ||
4815 | mm = get_task_mm(p); | |
4816 | if (!mm) | |
4817 | return 0; | |
4818 | /* We move charges only when we move a owner of the mm */ | |
4819 | if (mm->owner == p) { | |
4820 | VM_BUG_ON(mc.from); | |
4821 | VM_BUG_ON(mc.to); | |
4822 | VM_BUG_ON(mc.precharge); | |
4823 | VM_BUG_ON(mc.moved_charge); | |
4824 | VM_BUG_ON(mc.moved_swap); | |
4825 | ||
4826 | spin_lock(&mc.lock); | |
264a0ae1 | 4827 | mc.mm = mm; |
9f2115f9 TH |
4828 | mc.from = from; |
4829 | mc.to = memcg; | |
4830 | mc.flags = move_flags; | |
4831 | spin_unlock(&mc.lock); | |
4832 | /* We set mc.moving_task later */ | |
4833 | ||
4834 | ret = mem_cgroup_precharge_mc(mm); | |
4835 | if (ret) | |
4836 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
4837 | } else { |
4838 | mmput(mm); | |
7dc74be0 DN |
4839 | } |
4840 | return ret; | |
4841 | } | |
4842 | ||
1f7dd3e5 | 4843 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4844 | { |
4e2f245d JW |
4845 | if (mc.to) |
4846 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4847 | } |
4848 | ||
4ffef5fe DN |
4849 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4850 | unsigned long addr, unsigned long end, | |
4851 | struct mm_walk *walk) | |
7dc74be0 | 4852 | { |
4ffef5fe | 4853 | int ret = 0; |
26bcd64a | 4854 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4855 | pte_t *pte; |
4856 | spinlock_t *ptl; | |
12724850 NH |
4857 | enum mc_target_type target_type; |
4858 | union mc_target target; | |
4859 | struct page *page; | |
4ffef5fe | 4860 | |
b6ec57f4 KS |
4861 | ptl = pmd_trans_huge_lock(pmd, vma); |
4862 | if (ptl) { | |
62ade86a | 4863 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4864 | spin_unlock(ptl); |
12724850 NH |
4865 | return 0; |
4866 | } | |
4867 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4868 | if (target_type == MC_TARGET_PAGE) { | |
4869 | page = target.page; | |
4870 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 4871 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 4872 | mc.from, mc.to)) { |
12724850 NH |
4873 | mc.precharge -= HPAGE_PMD_NR; |
4874 | mc.moved_charge += HPAGE_PMD_NR; | |
4875 | } | |
4876 | putback_lru_page(page); | |
4877 | } | |
4878 | put_page(page); | |
4879 | } | |
bf929152 | 4880 | spin_unlock(ptl); |
1a5a9906 | 4881 | return 0; |
12724850 NH |
4882 | } |
4883 | ||
45f83cef AA |
4884 | if (pmd_trans_unstable(pmd)) |
4885 | return 0; | |
4ffef5fe DN |
4886 | retry: |
4887 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4888 | for (; addr != end; addr += PAGE_SIZE) { | |
4889 | pte_t ptent = *(pte++); | |
02491447 | 4890 | swp_entry_t ent; |
4ffef5fe DN |
4891 | |
4892 | if (!mc.precharge) | |
4893 | break; | |
4894 | ||
8d32ff84 | 4895 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4896 | case MC_TARGET_PAGE: |
4897 | page = target.page; | |
53f9263b KS |
4898 | /* |
4899 | * We can have a part of the split pmd here. Moving it | |
4900 | * can be done but it would be too convoluted so simply | |
4901 | * ignore such a partial THP and keep it in original | |
4902 | * memcg. There should be somebody mapping the head. | |
4903 | */ | |
4904 | if (PageTransCompound(page)) | |
4905 | goto put; | |
4ffef5fe DN |
4906 | if (isolate_lru_page(page)) |
4907 | goto put; | |
f627c2f5 KS |
4908 | if (!mem_cgroup_move_account(page, false, |
4909 | mc.from, mc.to)) { | |
4ffef5fe | 4910 | mc.precharge--; |
854ffa8d DN |
4911 | /* we uncharge from mc.from later. */ |
4912 | mc.moved_charge++; | |
4ffef5fe DN |
4913 | } |
4914 | putback_lru_page(page); | |
8d32ff84 | 4915 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4916 | put_page(page); |
4917 | break; | |
02491447 DN |
4918 | case MC_TARGET_SWAP: |
4919 | ent = target.ent; | |
e91cbb42 | 4920 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4921 | mc.precharge--; |
483c30b5 DN |
4922 | /* we fixup refcnts and charges later. */ |
4923 | mc.moved_swap++; | |
4924 | } | |
02491447 | 4925 | break; |
4ffef5fe DN |
4926 | default: |
4927 | break; | |
4928 | } | |
4929 | } | |
4930 | pte_unmap_unlock(pte - 1, ptl); | |
4931 | cond_resched(); | |
4932 | ||
4933 | if (addr != end) { | |
4934 | /* | |
4935 | * We have consumed all precharges we got in can_attach(). | |
4936 | * We try charge one by one, but don't do any additional | |
4937 | * charges to mc.to if we have failed in charge once in attach() | |
4938 | * phase. | |
4939 | */ | |
854ffa8d | 4940 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4941 | if (!ret) |
4942 | goto retry; | |
4943 | } | |
4944 | ||
4945 | return ret; | |
4946 | } | |
4947 | ||
264a0ae1 | 4948 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 4949 | { |
26bcd64a NH |
4950 | struct mm_walk mem_cgroup_move_charge_walk = { |
4951 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 4952 | .mm = mc.mm, |
26bcd64a | 4953 | }; |
4ffef5fe DN |
4954 | |
4955 | lru_add_drain_all(); | |
312722cb | 4956 | /* |
81f8c3a4 JW |
4957 | * Signal lock_page_memcg() to take the memcg's move_lock |
4958 | * while we're moving its pages to another memcg. Then wait | |
4959 | * for already started RCU-only updates to finish. | |
312722cb JW |
4960 | */ |
4961 | atomic_inc(&mc.from->moving_account); | |
4962 | synchronize_rcu(); | |
dfe076b0 | 4963 | retry: |
264a0ae1 | 4964 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
4965 | /* |
4966 | * Someone who are holding the mmap_sem might be waiting in | |
4967 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4968 | * and retry. Because we cancel precharges, we might not be able | |
4969 | * to move enough charges, but moving charge is a best-effort | |
4970 | * feature anyway, so it wouldn't be a big problem. | |
4971 | */ | |
4972 | __mem_cgroup_clear_mc(); | |
4973 | cond_resched(); | |
4974 | goto retry; | |
4975 | } | |
26bcd64a NH |
4976 | /* |
4977 | * When we have consumed all precharges and failed in doing | |
4978 | * additional charge, the page walk just aborts. | |
4979 | */ | |
0247f3f4 JM |
4980 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
4981 | ||
264a0ae1 | 4982 | up_read(&mc.mm->mmap_sem); |
312722cb | 4983 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
4984 | } |
4985 | ||
264a0ae1 | 4986 | static void mem_cgroup_move_task(void) |
67e465a7 | 4987 | { |
264a0ae1 TH |
4988 | if (mc.to) { |
4989 | mem_cgroup_move_charge(); | |
a433658c | 4990 | mem_cgroup_clear_mc(); |
264a0ae1 | 4991 | } |
67e465a7 | 4992 | } |
5cfb80a7 | 4993 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 4994 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4995 | { |
4996 | return 0; | |
4997 | } | |
1f7dd3e5 | 4998 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4999 | { |
5000 | } | |
264a0ae1 | 5001 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5002 | { |
5003 | } | |
5004 | #endif | |
67e465a7 | 5005 | |
f00baae7 TH |
5006 | /* |
5007 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5008 | * to verify whether we're attached to the default hierarchy on each mount |
5009 | * attempt. | |
f00baae7 | 5010 | */ |
eb95419b | 5011 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5012 | { |
5013 | /* | |
aa6ec29b | 5014 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5015 | * guarantees that @root doesn't have any children, so turning it |
5016 | * on for the root memcg is enough. | |
5017 | */ | |
9e10a130 | 5018 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5019 | root_mem_cgroup->use_hierarchy = true; |
5020 | else | |
5021 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5022 | } |
5023 | ||
241994ed JW |
5024 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5025 | struct cftype *cft) | |
5026 | { | |
f5fc3c5d JW |
5027 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5028 | ||
5029 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5030 | } |
5031 | ||
5032 | static int memory_low_show(struct seq_file *m, void *v) | |
5033 | { | |
5034 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5035 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5036 | |
5037 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5038 | seq_puts(m, "max\n"); |
241994ed JW |
5039 | else |
5040 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5041 | ||
5042 | return 0; | |
5043 | } | |
5044 | ||
5045 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5046 | char *buf, size_t nbytes, loff_t off) | |
5047 | { | |
5048 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5049 | unsigned long low; | |
5050 | int err; | |
5051 | ||
5052 | buf = strstrip(buf); | |
d2973697 | 5053 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5054 | if (err) |
5055 | return err; | |
5056 | ||
5057 | memcg->low = low; | |
5058 | ||
5059 | return nbytes; | |
5060 | } | |
5061 | ||
5062 | static int memory_high_show(struct seq_file *m, void *v) | |
5063 | { | |
5064 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5065 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5066 | |
5067 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5068 | seq_puts(m, "max\n"); |
241994ed JW |
5069 | else |
5070 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5071 | ||
5072 | return 0; | |
5073 | } | |
5074 | ||
5075 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5076 | char *buf, size_t nbytes, loff_t off) | |
5077 | { | |
5078 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5079 | unsigned long nr_pages; |
241994ed JW |
5080 | unsigned long high; |
5081 | int err; | |
5082 | ||
5083 | buf = strstrip(buf); | |
d2973697 | 5084 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5085 | if (err) |
5086 | return err; | |
5087 | ||
5088 | memcg->high = high; | |
5089 | ||
588083bb JW |
5090 | nr_pages = page_counter_read(&memcg->memory); |
5091 | if (nr_pages > high) | |
5092 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5093 | GFP_KERNEL, true); | |
5094 | ||
2529bb3a | 5095 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5096 | return nbytes; |
5097 | } | |
5098 | ||
5099 | static int memory_max_show(struct seq_file *m, void *v) | |
5100 | { | |
5101 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5102 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5103 | |
5104 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5105 | seq_puts(m, "max\n"); |
241994ed JW |
5106 | else |
5107 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5108 | ||
5109 | return 0; | |
5110 | } | |
5111 | ||
5112 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5113 | char *buf, size_t nbytes, loff_t off) | |
5114 | { | |
5115 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5116 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5117 | bool drained = false; | |
241994ed JW |
5118 | unsigned long max; |
5119 | int err; | |
5120 | ||
5121 | buf = strstrip(buf); | |
d2973697 | 5122 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5123 | if (err) |
5124 | return err; | |
5125 | ||
b6e6edcf JW |
5126 | xchg(&memcg->memory.limit, max); |
5127 | ||
5128 | for (;;) { | |
5129 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5130 | ||
5131 | if (nr_pages <= max) | |
5132 | break; | |
5133 | ||
5134 | if (signal_pending(current)) { | |
5135 | err = -EINTR; | |
5136 | break; | |
5137 | } | |
5138 | ||
5139 | if (!drained) { | |
5140 | drain_all_stock(memcg); | |
5141 | drained = true; | |
5142 | continue; | |
5143 | } | |
5144 | ||
5145 | if (nr_reclaims) { | |
5146 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5147 | GFP_KERNEL, true)) | |
5148 | nr_reclaims--; | |
5149 | continue; | |
5150 | } | |
5151 | ||
5152 | mem_cgroup_events(memcg, MEMCG_OOM, 1); | |
5153 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) | |
5154 | break; | |
5155 | } | |
241994ed | 5156 | |
2529bb3a | 5157 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5158 | return nbytes; |
5159 | } | |
5160 | ||
5161 | static int memory_events_show(struct seq_file *m, void *v) | |
5162 | { | |
5163 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5164 | ||
5165 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5166 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5167 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5168 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5169 | ||
5170 | return 0; | |
5171 | } | |
5172 | ||
587d9f72 JW |
5173 | static int memory_stat_show(struct seq_file *m, void *v) |
5174 | { | |
5175 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
72b54e73 VD |
5176 | unsigned long stat[MEMCG_NR_STAT]; |
5177 | unsigned long events[MEMCG_NR_EVENTS]; | |
587d9f72 JW |
5178 | int i; |
5179 | ||
5180 | /* | |
5181 | * Provide statistics on the state of the memory subsystem as | |
5182 | * well as cumulative event counters that show past behavior. | |
5183 | * | |
5184 | * This list is ordered following a combination of these gradients: | |
5185 | * 1) generic big picture -> specifics and details | |
5186 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5187 | * | |
5188 | * Current memory state: | |
5189 | */ | |
5190 | ||
72b54e73 VD |
5191 | tree_stat(memcg, stat); |
5192 | tree_events(memcg, events); | |
5193 | ||
587d9f72 | 5194 | seq_printf(m, "anon %llu\n", |
72b54e73 | 5195 | (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE); |
587d9f72 | 5196 | seq_printf(m, "file %llu\n", |
72b54e73 | 5197 | (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE); |
12580e4b | 5198 | seq_printf(m, "kernel_stack %llu\n", |
efdc9490 | 5199 | (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024); |
27ee57c9 VD |
5200 | seq_printf(m, "slab %llu\n", |
5201 | (u64)(stat[MEMCG_SLAB_RECLAIMABLE] + | |
5202 | stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5203 | seq_printf(m, "sock %llu\n", |
72b54e73 | 5204 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 JW |
5205 | |
5206 | seq_printf(m, "file_mapped %llu\n", | |
72b54e73 | 5207 | (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5208 | seq_printf(m, "file_dirty %llu\n", |
72b54e73 | 5209 | (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE); |
587d9f72 | 5210 | seq_printf(m, "file_writeback %llu\n", |
72b54e73 | 5211 | (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE); |
587d9f72 JW |
5212 | |
5213 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5214 | struct mem_cgroup *mi; | |
5215 | unsigned long val = 0; | |
5216 | ||
5217 | for_each_mem_cgroup_tree(mi, memcg) | |
5218 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
5219 | seq_printf(m, "%s %llu\n", | |
5220 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); | |
5221 | } | |
5222 | ||
27ee57c9 VD |
5223 | seq_printf(m, "slab_reclaimable %llu\n", |
5224 | (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE); | |
5225 | seq_printf(m, "slab_unreclaimable %llu\n", | |
5226 | (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE); | |
5227 | ||
587d9f72 JW |
5228 | /* Accumulated memory events */ |
5229 | ||
5230 | seq_printf(m, "pgfault %lu\n", | |
72b54e73 | 5231 | events[MEM_CGROUP_EVENTS_PGFAULT]); |
587d9f72 | 5232 | seq_printf(m, "pgmajfault %lu\n", |
72b54e73 | 5233 | events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
587d9f72 JW |
5234 | |
5235 | return 0; | |
5236 | } | |
5237 | ||
241994ed JW |
5238 | static struct cftype memory_files[] = { |
5239 | { | |
5240 | .name = "current", | |
f5fc3c5d | 5241 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5242 | .read_u64 = memory_current_read, |
5243 | }, | |
5244 | { | |
5245 | .name = "low", | |
5246 | .flags = CFTYPE_NOT_ON_ROOT, | |
5247 | .seq_show = memory_low_show, | |
5248 | .write = memory_low_write, | |
5249 | }, | |
5250 | { | |
5251 | .name = "high", | |
5252 | .flags = CFTYPE_NOT_ON_ROOT, | |
5253 | .seq_show = memory_high_show, | |
5254 | .write = memory_high_write, | |
5255 | }, | |
5256 | { | |
5257 | .name = "max", | |
5258 | .flags = CFTYPE_NOT_ON_ROOT, | |
5259 | .seq_show = memory_max_show, | |
5260 | .write = memory_max_write, | |
5261 | }, | |
5262 | { | |
5263 | .name = "events", | |
5264 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5265 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5266 | .seq_show = memory_events_show, |
5267 | }, | |
587d9f72 JW |
5268 | { |
5269 | .name = "stat", | |
5270 | .flags = CFTYPE_NOT_ON_ROOT, | |
5271 | .seq_show = memory_stat_show, | |
5272 | }, | |
241994ed JW |
5273 | { } /* terminate */ |
5274 | }; | |
5275 | ||
073219e9 | 5276 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5277 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5278 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5279 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5280 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5281 | .css_free = mem_cgroup_css_free, |
1ced953b | 5282 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5283 | .can_attach = mem_cgroup_can_attach, |
5284 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5285 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5286 | .bind = mem_cgroup_bind, |
241994ed JW |
5287 | .dfl_cftypes = memory_files, |
5288 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5289 | .early_init = 0, |
8cdea7c0 | 5290 | }; |
c077719b | 5291 | |
241994ed JW |
5292 | /** |
5293 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5294 | * @root: the highest ancestor to consider | |
5295 | * @memcg: the memory cgroup to check | |
5296 | * | |
5297 | * Returns %true if memory consumption of @memcg, and that of all | |
5298 | * configurable ancestors up to @root, is below the normal range. | |
5299 | */ | |
5300 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5301 | { | |
5302 | if (mem_cgroup_disabled()) | |
5303 | return false; | |
5304 | ||
5305 | /* | |
5306 | * The toplevel group doesn't have a configurable range, so | |
5307 | * it's never low when looked at directly, and it is not | |
5308 | * considered an ancestor when assessing the hierarchy. | |
5309 | */ | |
5310 | ||
5311 | if (memcg == root_mem_cgroup) | |
5312 | return false; | |
5313 | ||
4e54dede | 5314 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5315 | return false; |
5316 | ||
5317 | while (memcg != root) { | |
5318 | memcg = parent_mem_cgroup(memcg); | |
5319 | ||
5320 | if (memcg == root_mem_cgroup) | |
5321 | break; | |
5322 | ||
4e54dede | 5323 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5324 | return false; |
5325 | } | |
5326 | return true; | |
5327 | } | |
5328 | ||
00501b53 JW |
5329 | /** |
5330 | * mem_cgroup_try_charge - try charging a page | |
5331 | * @page: page to charge | |
5332 | * @mm: mm context of the victim | |
5333 | * @gfp_mask: reclaim mode | |
5334 | * @memcgp: charged memcg return | |
25843c2b | 5335 | * @compound: charge the page as compound or small page |
00501b53 JW |
5336 | * |
5337 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5338 | * pages according to @gfp_mask if necessary. | |
5339 | * | |
5340 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5341 | * Otherwise, an error code is returned. | |
5342 | * | |
5343 | * After page->mapping has been set up, the caller must finalize the | |
5344 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5345 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5346 | */ | |
5347 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5348 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5349 | bool compound) | |
00501b53 JW |
5350 | { |
5351 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5352 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5353 | int ret = 0; |
5354 | ||
5355 | if (mem_cgroup_disabled()) | |
5356 | goto out; | |
5357 | ||
5358 | if (PageSwapCache(page)) { | |
00501b53 JW |
5359 | /* |
5360 | * Every swap fault against a single page tries to charge the | |
5361 | * page, bail as early as possible. shmem_unuse() encounters | |
5362 | * already charged pages, too. The USED bit is protected by | |
5363 | * the page lock, which serializes swap cache removal, which | |
5364 | * in turn serializes uncharging. | |
5365 | */ | |
e993d905 | 5366 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5367 | if (page->mem_cgroup) |
00501b53 | 5368 | goto out; |
e993d905 | 5369 | |
37e84351 | 5370 | if (do_swap_account) { |
e993d905 VD |
5371 | swp_entry_t ent = { .val = page_private(page), }; |
5372 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5373 | ||
5374 | rcu_read_lock(); | |
5375 | memcg = mem_cgroup_from_id(id); | |
5376 | if (memcg && !css_tryget_online(&memcg->css)) | |
5377 | memcg = NULL; | |
5378 | rcu_read_unlock(); | |
5379 | } | |
00501b53 JW |
5380 | } |
5381 | ||
00501b53 JW |
5382 | if (!memcg) |
5383 | memcg = get_mem_cgroup_from_mm(mm); | |
5384 | ||
5385 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5386 | ||
5387 | css_put(&memcg->css); | |
00501b53 JW |
5388 | out: |
5389 | *memcgp = memcg; | |
5390 | return ret; | |
5391 | } | |
5392 | ||
5393 | /** | |
5394 | * mem_cgroup_commit_charge - commit a page charge | |
5395 | * @page: page to charge | |
5396 | * @memcg: memcg to charge the page to | |
5397 | * @lrucare: page might be on LRU already | |
25843c2b | 5398 | * @compound: charge the page as compound or small page |
00501b53 JW |
5399 | * |
5400 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5401 | * after page->mapping has been set up. This must happen atomically | |
5402 | * as part of the page instantiation, i.e. under the page table lock | |
5403 | * for anonymous pages, under the page lock for page and swap cache. | |
5404 | * | |
5405 | * In addition, the page must not be on the LRU during the commit, to | |
5406 | * prevent racing with task migration. If it might be, use @lrucare. | |
5407 | * | |
5408 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5409 | */ | |
5410 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5411 | bool lrucare, bool compound) |
00501b53 | 5412 | { |
f627c2f5 | 5413 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5414 | |
5415 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5416 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5417 | ||
5418 | if (mem_cgroup_disabled()) | |
5419 | return; | |
5420 | /* | |
5421 | * Swap faults will attempt to charge the same page multiple | |
5422 | * times. But reuse_swap_page() might have removed the page | |
5423 | * from swapcache already, so we can't check PageSwapCache(). | |
5424 | */ | |
5425 | if (!memcg) | |
5426 | return; | |
5427 | ||
6abb5a86 JW |
5428 | commit_charge(page, memcg, lrucare); |
5429 | ||
6abb5a86 | 5430 | local_irq_disable(); |
f627c2f5 | 5431 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5432 | memcg_check_events(memcg, page); |
5433 | local_irq_enable(); | |
00501b53 | 5434 | |
7941d214 | 5435 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5436 | swp_entry_t entry = { .val = page_private(page) }; |
5437 | /* | |
5438 | * The swap entry might not get freed for a long time, | |
5439 | * let's not wait for it. The page already received a | |
5440 | * memory+swap charge, drop the swap entry duplicate. | |
5441 | */ | |
5442 | mem_cgroup_uncharge_swap(entry); | |
5443 | } | |
5444 | } | |
5445 | ||
5446 | /** | |
5447 | * mem_cgroup_cancel_charge - cancel a page charge | |
5448 | * @page: page to charge | |
5449 | * @memcg: memcg to charge the page to | |
25843c2b | 5450 | * @compound: charge the page as compound or small page |
00501b53 JW |
5451 | * |
5452 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5453 | */ | |
f627c2f5 KS |
5454 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5455 | bool compound) | |
00501b53 | 5456 | { |
f627c2f5 | 5457 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5458 | |
5459 | if (mem_cgroup_disabled()) | |
5460 | return; | |
5461 | /* | |
5462 | * Swap faults will attempt to charge the same page multiple | |
5463 | * times. But reuse_swap_page() might have removed the page | |
5464 | * from swapcache already, so we can't check PageSwapCache(). | |
5465 | */ | |
5466 | if (!memcg) | |
5467 | return; | |
5468 | ||
00501b53 JW |
5469 | cancel_charge(memcg, nr_pages); |
5470 | } | |
5471 | ||
747db954 | 5472 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 | 5473 | unsigned long nr_anon, unsigned long nr_file, |
5e8d35f8 VD |
5474 | unsigned long nr_huge, unsigned long nr_kmem, |
5475 | struct page *dummy_page) | |
747db954 | 5476 | { |
5e8d35f8 | 5477 | unsigned long nr_pages = nr_anon + nr_file + nr_kmem; |
747db954 JW |
5478 | unsigned long flags; |
5479 | ||
ce00a967 | 5480 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 | 5481 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 5482 | if (do_memsw_account()) |
18eca2e6 | 5483 | page_counter_uncharge(&memcg->memsw, nr_pages); |
5e8d35f8 VD |
5484 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem) |
5485 | page_counter_uncharge(&memcg->kmem, nr_kmem); | |
ce00a967 JW |
5486 | memcg_oom_recover(memcg); |
5487 | } | |
747db954 JW |
5488 | |
5489 | local_irq_save(flags); | |
5490 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5491 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5492 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5493 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5494 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5495 | memcg_check_events(memcg, dummy_page); |
5496 | local_irq_restore(flags); | |
e8ea14cc JW |
5497 | |
5498 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5499 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5500 | } |
5501 | ||
5502 | static void uncharge_list(struct list_head *page_list) | |
5503 | { | |
5504 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5505 | unsigned long nr_anon = 0; |
5506 | unsigned long nr_file = 0; | |
5507 | unsigned long nr_huge = 0; | |
5e8d35f8 | 5508 | unsigned long nr_kmem = 0; |
747db954 | 5509 | unsigned long pgpgout = 0; |
747db954 JW |
5510 | struct list_head *next; |
5511 | struct page *page; | |
5512 | ||
8b592656 JW |
5513 | /* |
5514 | * Note that the list can be a single page->lru; hence the | |
5515 | * do-while loop instead of a simple list_for_each_entry(). | |
5516 | */ | |
747db954 JW |
5517 | next = page_list->next; |
5518 | do { | |
747db954 JW |
5519 | page = list_entry(next, struct page, lru); |
5520 | next = page->lru.next; | |
5521 | ||
5522 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5523 | VM_BUG_ON_PAGE(page_count(page), page); | |
5524 | ||
1306a85a | 5525 | if (!page->mem_cgroup) |
747db954 JW |
5526 | continue; |
5527 | ||
5528 | /* | |
5529 | * Nobody should be changing or seriously looking at | |
1306a85a | 5530 | * page->mem_cgroup at this point, we have fully |
29833315 | 5531 | * exclusive access to the page. |
747db954 JW |
5532 | */ |
5533 | ||
1306a85a | 5534 | if (memcg != page->mem_cgroup) { |
747db954 | 5535 | if (memcg) { |
18eca2e6 | 5536 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5e8d35f8 VD |
5537 | nr_huge, nr_kmem, page); |
5538 | pgpgout = nr_anon = nr_file = | |
5539 | nr_huge = nr_kmem = 0; | |
747db954 | 5540 | } |
1306a85a | 5541 | memcg = page->mem_cgroup; |
747db954 JW |
5542 | } |
5543 | ||
5e8d35f8 VD |
5544 | if (!PageKmemcg(page)) { |
5545 | unsigned int nr_pages = 1; | |
747db954 | 5546 | |
5e8d35f8 VD |
5547 | if (PageTransHuge(page)) { |
5548 | nr_pages <<= compound_order(page); | |
5e8d35f8 VD |
5549 | nr_huge += nr_pages; |
5550 | } | |
5551 | if (PageAnon(page)) | |
5552 | nr_anon += nr_pages; | |
5553 | else | |
5554 | nr_file += nr_pages; | |
5555 | pgpgout++; | |
c4159a75 | 5556 | } else { |
5e8d35f8 | 5557 | nr_kmem += 1 << compound_order(page); |
c4159a75 VD |
5558 | __ClearPageKmemcg(page); |
5559 | } | |
747db954 | 5560 | |
1306a85a | 5561 | page->mem_cgroup = NULL; |
747db954 JW |
5562 | } while (next != page_list); |
5563 | ||
5564 | if (memcg) | |
18eca2e6 | 5565 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5e8d35f8 | 5566 | nr_huge, nr_kmem, page); |
747db954 JW |
5567 | } |
5568 | ||
0a31bc97 JW |
5569 | /** |
5570 | * mem_cgroup_uncharge - uncharge a page | |
5571 | * @page: page to uncharge | |
5572 | * | |
5573 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5574 | * mem_cgroup_commit_charge(). | |
5575 | */ | |
5576 | void mem_cgroup_uncharge(struct page *page) | |
5577 | { | |
0a31bc97 JW |
5578 | if (mem_cgroup_disabled()) |
5579 | return; | |
5580 | ||
747db954 | 5581 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5582 | if (!page->mem_cgroup) |
0a31bc97 JW |
5583 | return; |
5584 | ||
747db954 JW |
5585 | INIT_LIST_HEAD(&page->lru); |
5586 | uncharge_list(&page->lru); | |
5587 | } | |
0a31bc97 | 5588 | |
747db954 JW |
5589 | /** |
5590 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5591 | * @page_list: list of pages to uncharge | |
5592 | * | |
5593 | * Uncharge a list of pages previously charged with | |
5594 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5595 | */ | |
5596 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5597 | { | |
5598 | if (mem_cgroup_disabled()) | |
5599 | return; | |
0a31bc97 | 5600 | |
747db954 JW |
5601 | if (!list_empty(page_list)) |
5602 | uncharge_list(page_list); | |
0a31bc97 JW |
5603 | } |
5604 | ||
5605 | /** | |
6a93ca8f JW |
5606 | * mem_cgroup_migrate - charge a page's replacement |
5607 | * @oldpage: currently circulating page | |
5608 | * @newpage: replacement page | |
0a31bc97 | 5609 | * |
6a93ca8f JW |
5610 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
5611 | * be uncharged upon free. | |
0a31bc97 JW |
5612 | * |
5613 | * Both pages must be locked, @newpage->mapping must be set up. | |
5614 | */ | |
6a93ca8f | 5615 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5616 | { |
29833315 | 5617 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
5618 | unsigned int nr_pages; |
5619 | bool compound; | |
d93c4130 | 5620 | unsigned long flags; |
0a31bc97 JW |
5621 | |
5622 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5623 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5624 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5625 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5626 | newpage); | |
0a31bc97 JW |
5627 | |
5628 | if (mem_cgroup_disabled()) | |
5629 | return; | |
5630 | ||
5631 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5632 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5633 | return; |
5634 | ||
45637bab | 5635 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5636 | memcg = oldpage->mem_cgroup; |
29833315 | 5637 | if (!memcg) |
0a31bc97 JW |
5638 | return; |
5639 | ||
44b7a8d3 JW |
5640 | /* Force-charge the new page. The old one will be freed soon */ |
5641 | compound = PageTransHuge(newpage); | |
5642 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
5643 | ||
5644 | page_counter_charge(&memcg->memory, nr_pages); | |
5645 | if (do_memsw_account()) | |
5646 | page_counter_charge(&memcg->memsw, nr_pages); | |
5647 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 5648 | |
9cf7666a | 5649 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 5650 | |
d93c4130 | 5651 | local_irq_save(flags); |
44b7a8d3 JW |
5652 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
5653 | memcg_check_events(memcg, newpage); | |
d93c4130 | 5654 | local_irq_restore(flags); |
0a31bc97 JW |
5655 | } |
5656 | ||
ef12947c | 5657 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5658 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5659 | ||
2d758073 | 5660 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
5661 | { |
5662 | struct mem_cgroup *memcg; | |
5663 | ||
2d758073 JW |
5664 | if (!mem_cgroup_sockets_enabled) |
5665 | return; | |
5666 | ||
5667 | /* | |
5668 | * Socket cloning can throw us here with sk_memcg already | |
11092087 JW |
5669 | * filled. It won't however, necessarily happen from |
5670 | * process context. So the test for root memcg given | |
5671 | * the current task's memcg won't help us in this case. | |
5672 | * | |
5673 | * Respecting the original socket's memcg is a better | |
5674 | * decision in this case. | |
5675 | */ | |
5676 | if (sk->sk_memcg) { | |
5677 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
5678 | css_get(&sk->sk_memcg->css); | |
5679 | return; | |
5680 | } | |
5681 | ||
5682 | rcu_read_lock(); | |
5683 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5684 | if (memcg == root_mem_cgroup) |
5685 | goto out; | |
0db15298 | 5686 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 5687 | goto out; |
f7e1cb6e | 5688 | if (css_tryget_online(&memcg->css)) |
11092087 | 5689 | sk->sk_memcg = memcg; |
f7e1cb6e | 5690 | out: |
11092087 JW |
5691 | rcu_read_unlock(); |
5692 | } | |
11092087 | 5693 | |
2d758073 | 5694 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 5695 | { |
2d758073 JW |
5696 | if (sk->sk_memcg) |
5697 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
5698 | } |
5699 | ||
5700 | /** | |
5701 | * mem_cgroup_charge_skmem - charge socket memory | |
5702 | * @memcg: memcg to charge | |
5703 | * @nr_pages: number of pages to charge | |
5704 | * | |
5705 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5706 | * @memcg's configured limit, %false if the charge had to be forced. | |
5707 | */ | |
5708 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5709 | { | |
f7e1cb6e | 5710 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5711 | |
f7e1cb6e | 5712 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5713 | struct page_counter *fail; |
f7e1cb6e | 5714 | |
0db15298 JW |
5715 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
5716 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
5717 | return true; |
5718 | } | |
0db15298 JW |
5719 | page_counter_charge(&memcg->tcpmem, nr_pages); |
5720 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 5721 | return false; |
11092087 | 5722 | } |
d886f4e4 | 5723 | |
f7e1cb6e JW |
5724 | /* Don't block in the packet receive path */ |
5725 | if (in_softirq()) | |
5726 | gfp_mask = GFP_NOWAIT; | |
5727 | ||
b2807f07 JW |
5728 | this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5729 | ||
f7e1cb6e JW |
5730 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
5731 | return true; | |
5732 | ||
5733 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5734 | return false; |
5735 | } | |
5736 | ||
5737 | /** | |
5738 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5739 | * @memcg - memcg to uncharge | |
5740 | * @nr_pages - number of pages to uncharge | |
5741 | */ | |
5742 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5743 | { | |
f7e1cb6e | 5744 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5745 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
5746 | return; |
5747 | } | |
d886f4e4 | 5748 | |
b2807f07 JW |
5749 | this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5750 | ||
f7e1cb6e JW |
5751 | page_counter_uncharge(&memcg->memory, nr_pages); |
5752 | css_put_many(&memcg->css, nr_pages); | |
11092087 JW |
5753 | } |
5754 | ||
f7e1cb6e JW |
5755 | static int __init cgroup_memory(char *s) |
5756 | { | |
5757 | char *token; | |
5758 | ||
5759 | while ((token = strsep(&s, ",")) != NULL) { | |
5760 | if (!*token) | |
5761 | continue; | |
5762 | if (!strcmp(token, "nosocket")) | |
5763 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5764 | if (!strcmp(token, "nokmem")) |
5765 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5766 | } |
5767 | return 0; | |
5768 | } | |
5769 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5770 | |
2d11085e | 5771 | /* |
1081312f MH |
5772 | * subsys_initcall() for memory controller. |
5773 | * | |
308167fc SAS |
5774 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
5775 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
5776 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
5777 | * should be initialized from here. | |
2d11085e MH |
5778 | */ |
5779 | static int __init mem_cgroup_init(void) | |
5780 | { | |
95a045f6 JW |
5781 | int cpu, node; |
5782 | ||
13583c3d VD |
5783 | #ifndef CONFIG_SLOB |
5784 | /* | |
5785 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
5786 | * so use a workqueue with limited concurrency to avoid stalling |
5787 | * all worker threads in case lots of cgroups are created and | |
5788 | * destroyed simultaneously. | |
13583c3d | 5789 | */ |
17cc4dfe TH |
5790 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
5791 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
5792 | #endif |
5793 | ||
308167fc SAS |
5794 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
5795 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
5796 | |
5797 | for_each_possible_cpu(cpu) | |
5798 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5799 | drain_local_stock); | |
5800 | ||
5801 | for_each_node(node) { | |
5802 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
5803 | |
5804 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5805 | node_online(node) ? node : NUMA_NO_NODE); | |
5806 | ||
ef8f2327 MG |
5807 | rtpn->rb_root = RB_ROOT; |
5808 | spin_lock_init(&rtpn->lock); | |
95a045f6 JW |
5809 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
5810 | } | |
5811 | ||
2d11085e MH |
5812 | return 0; |
5813 | } | |
5814 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5815 | |
5816 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
5817 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
5818 | { | |
5819 | while (!atomic_inc_not_zero(&memcg->id.ref)) { | |
5820 | /* | |
5821 | * The root cgroup cannot be destroyed, so it's refcount must | |
5822 | * always be >= 1. | |
5823 | */ | |
5824 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
5825 | VM_BUG_ON(1); | |
5826 | break; | |
5827 | } | |
5828 | memcg = parent_mem_cgroup(memcg); | |
5829 | if (!memcg) | |
5830 | memcg = root_mem_cgroup; | |
5831 | } | |
5832 | return memcg; | |
5833 | } | |
5834 | ||
21afa38e JW |
5835 | /** |
5836 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5837 | * @page: page whose memsw charge to transfer | |
5838 | * @entry: swap entry to move the charge to | |
5839 | * | |
5840 | * Transfer the memsw charge of @page to @entry. | |
5841 | */ | |
5842 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5843 | { | |
1f47b61f | 5844 | struct mem_cgroup *memcg, *swap_memcg; |
21afa38e JW |
5845 | unsigned short oldid; |
5846 | ||
5847 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5848 | VM_BUG_ON_PAGE(page_count(page), page); | |
5849 | ||
7941d214 | 5850 | if (!do_memsw_account()) |
21afa38e JW |
5851 | return; |
5852 | ||
5853 | memcg = page->mem_cgroup; | |
5854 | ||
5855 | /* Readahead page, never charged */ | |
5856 | if (!memcg) | |
5857 | return; | |
5858 | ||
1f47b61f VD |
5859 | /* |
5860 | * In case the memcg owning these pages has been offlined and doesn't | |
5861 | * have an ID allocated to it anymore, charge the closest online | |
5862 | * ancestor for the swap instead and transfer the memory+swap charge. | |
5863 | */ | |
5864 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
5865 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg)); | |
21afa38e | 5866 | VM_BUG_ON_PAGE(oldid, page); |
1f47b61f | 5867 | mem_cgroup_swap_statistics(swap_memcg, true); |
21afa38e JW |
5868 | |
5869 | page->mem_cgroup = NULL; | |
5870 | ||
5871 | if (!mem_cgroup_is_root(memcg)) | |
5872 | page_counter_uncharge(&memcg->memory, 1); | |
5873 | ||
1f47b61f VD |
5874 | if (memcg != swap_memcg) { |
5875 | if (!mem_cgroup_is_root(swap_memcg)) | |
5876 | page_counter_charge(&swap_memcg->memsw, 1); | |
5877 | page_counter_uncharge(&memcg->memsw, 1); | |
5878 | } | |
5879 | ||
ce9ce665 SAS |
5880 | /* |
5881 | * Interrupts should be disabled here because the caller holds the | |
5882 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5883 | * important here to have the interrupts disabled because it is the | |
5884 | * only synchronisation we have for udpating the per-CPU variables. | |
5885 | */ | |
5886 | VM_BUG_ON(!irqs_disabled()); | |
f627c2f5 | 5887 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
21afa38e | 5888 | memcg_check_events(memcg, page); |
73f576c0 JW |
5889 | |
5890 | if (!mem_cgroup_is_root(memcg)) | |
5891 | css_put(&memcg->css); | |
21afa38e JW |
5892 | } |
5893 | ||
37e84351 VD |
5894 | /* |
5895 | * mem_cgroup_try_charge_swap - try charging a swap entry | |
5896 | * @page: page being added to swap | |
5897 | * @entry: swap entry to charge | |
5898 | * | |
5899 | * Try to charge @entry to the memcg that @page belongs to. | |
5900 | * | |
5901 | * Returns 0 on success, -ENOMEM on failure. | |
5902 | */ | |
5903 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
5904 | { | |
5905 | struct mem_cgroup *memcg; | |
5906 | struct page_counter *counter; | |
5907 | unsigned short oldid; | |
5908 | ||
5909 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
5910 | return 0; | |
5911 | ||
5912 | memcg = page->mem_cgroup; | |
5913 | ||
5914 | /* Readahead page, never charged */ | |
5915 | if (!memcg) | |
5916 | return 0; | |
5917 | ||
1f47b61f VD |
5918 | memcg = mem_cgroup_id_get_online(memcg); |
5919 | ||
37e84351 | 5920 | if (!mem_cgroup_is_root(memcg) && |
1f47b61f VD |
5921 | !page_counter_try_charge(&memcg->swap, 1, &counter)) { |
5922 | mem_cgroup_id_put(memcg); | |
37e84351 | 5923 | return -ENOMEM; |
1f47b61f | 5924 | } |
37e84351 VD |
5925 | |
5926 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5927 | VM_BUG_ON_PAGE(oldid, page); | |
5928 | mem_cgroup_swap_statistics(memcg, true); | |
5929 | ||
37e84351 VD |
5930 | return 0; |
5931 | } | |
5932 | ||
21afa38e JW |
5933 | /** |
5934 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5935 | * @entry: swap entry to uncharge | |
5936 | * | |
37e84351 | 5937 | * Drop the swap charge associated with @entry. |
21afa38e JW |
5938 | */ |
5939 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5940 | { | |
5941 | struct mem_cgroup *memcg; | |
5942 | unsigned short id; | |
5943 | ||
37e84351 | 5944 | if (!do_swap_account) |
21afa38e JW |
5945 | return; |
5946 | ||
5947 | id = swap_cgroup_record(entry, 0); | |
5948 | rcu_read_lock(); | |
adbe427b | 5949 | memcg = mem_cgroup_from_id(id); |
21afa38e | 5950 | if (memcg) { |
37e84351 VD |
5951 | if (!mem_cgroup_is_root(memcg)) { |
5952 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5953 | page_counter_uncharge(&memcg->swap, 1); | |
5954 | else | |
5955 | page_counter_uncharge(&memcg->memsw, 1); | |
5956 | } | |
21afa38e | 5957 | mem_cgroup_swap_statistics(memcg, false); |
73f576c0 | 5958 | mem_cgroup_id_put(memcg); |
21afa38e JW |
5959 | } |
5960 | rcu_read_unlock(); | |
5961 | } | |
5962 | ||
d8b38438 VD |
5963 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
5964 | { | |
5965 | long nr_swap_pages = get_nr_swap_pages(); | |
5966 | ||
5967 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5968 | return nr_swap_pages; | |
5969 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5970 | nr_swap_pages = min_t(long, nr_swap_pages, | |
5971 | READ_ONCE(memcg->swap.limit) - | |
5972 | page_counter_read(&memcg->swap)); | |
5973 | return nr_swap_pages; | |
5974 | } | |
5975 | ||
5ccc5aba VD |
5976 | bool mem_cgroup_swap_full(struct page *page) |
5977 | { | |
5978 | struct mem_cgroup *memcg; | |
5979 | ||
5980 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
5981 | ||
5982 | if (vm_swap_full()) | |
5983 | return true; | |
5984 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5985 | return false; | |
5986 | ||
5987 | memcg = page->mem_cgroup; | |
5988 | if (!memcg) | |
5989 | return false; | |
5990 | ||
5991 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5992 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) | |
5993 | return true; | |
5994 | ||
5995 | return false; | |
5996 | } | |
5997 | ||
21afa38e JW |
5998 | /* for remember boot option*/ |
5999 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6000 | static int really_do_swap_account __initdata = 1; | |
6001 | #else | |
6002 | static int really_do_swap_account __initdata; | |
6003 | #endif | |
6004 | ||
6005 | static int __init enable_swap_account(char *s) | |
6006 | { | |
6007 | if (!strcmp(s, "1")) | |
6008 | really_do_swap_account = 1; | |
6009 | else if (!strcmp(s, "0")) | |
6010 | really_do_swap_account = 0; | |
6011 | return 1; | |
6012 | } | |
6013 | __setup("swapaccount=", enable_swap_account); | |
6014 | ||
37e84351 VD |
6015 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6016 | struct cftype *cft) | |
6017 | { | |
6018 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6019 | ||
6020 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6021 | } | |
6022 | ||
6023 | static int swap_max_show(struct seq_file *m, void *v) | |
6024 | { | |
6025 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
6026 | unsigned long max = READ_ONCE(memcg->swap.limit); | |
6027 | ||
6028 | if (max == PAGE_COUNTER_MAX) | |
6029 | seq_puts(m, "max\n"); | |
6030 | else | |
6031 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
6032 | ||
6033 | return 0; | |
6034 | } | |
6035 | ||
6036 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6037 | char *buf, size_t nbytes, loff_t off) | |
6038 | { | |
6039 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6040 | unsigned long max; | |
6041 | int err; | |
6042 | ||
6043 | buf = strstrip(buf); | |
6044 | err = page_counter_memparse(buf, "max", &max); | |
6045 | if (err) | |
6046 | return err; | |
6047 | ||
6048 | mutex_lock(&memcg_limit_mutex); | |
6049 | err = page_counter_limit(&memcg->swap, max); | |
6050 | mutex_unlock(&memcg_limit_mutex); | |
6051 | if (err) | |
6052 | return err; | |
6053 | ||
6054 | return nbytes; | |
6055 | } | |
6056 | ||
6057 | static struct cftype swap_files[] = { | |
6058 | { | |
6059 | .name = "swap.current", | |
6060 | .flags = CFTYPE_NOT_ON_ROOT, | |
6061 | .read_u64 = swap_current_read, | |
6062 | }, | |
6063 | { | |
6064 | .name = "swap.max", | |
6065 | .flags = CFTYPE_NOT_ON_ROOT, | |
6066 | .seq_show = swap_max_show, | |
6067 | .write = swap_max_write, | |
6068 | }, | |
6069 | { } /* terminate */ | |
6070 | }; | |
6071 | ||
21afa38e JW |
6072 | static struct cftype memsw_cgroup_files[] = { |
6073 | { | |
6074 | .name = "memsw.usage_in_bytes", | |
6075 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6076 | .read_u64 = mem_cgroup_read_u64, | |
6077 | }, | |
6078 | { | |
6079 | .name = "memsw.max_usage_in_bytes", | |
6080 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6081 | .write = mem_cgroup_reset, | |
6082 | .read_u64 = mem_cgroup_read_u64, | |
6083 | }, | |
6084 | { | |
6085 | .name = "memsw.limit_in_bytes", | |
6086 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6087 | .write = mem_cgroup_write, | |
6088 | .read_u64 = mem_cgroup_read_u64, | |
6089 | }, | |
6090 | { | |
6091 | .name = "memsw.failcnt", | |
6092 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6093 | .write = mem_cgroup_reset, | |
6094 | .read_u64 = mem_cgroup_read_u64, | |
6095 | }, | |
6096 | { }, /* terminate */ | |
6097 | }; | |
6098 | ||
6099 | static int __init mem_cgroup_swap_init(void) | |
6100 | { | |
6101 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6102 | do_swap_account = 1; | |
37e84351 VD |
6103 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6104 | swap_files)); | |
21afa38e JW |
6105 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6106 | memsw_cgroup_files)); | |
6107 | } | |
6108 | return 0; | |
6109 | } | |
6110 | subsys_initcall(mem_cgroup_swap_init); | |
6111 | ||
6112 | #endif /* CONFIG_MEMCG_SWAP */ |