]>
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", |
9a4caf1e | 107 | "shmem", |
af7c4b0e | 108 | "mapped_file", |
c4843a75 | 109 | "dirty", |
3ea67d06 | 110 | "writeback", |
af7c4b0e JW |
111 | "swap", |
112 | }; | |
113 | ||
af7c4b0e JW |
114 | static const char * const mem_cgroup_events_names[] = { |
115 | "pgpgin", | |
116 | "pgpgout", | |
117 | "pgfault", | |
118 | "pgmajfault", | |
119 | }; | |
120 | ||
58cf188e SZ |
121 | static const char * const mem_cgroup_lru_names[] = { |
122 | "inactive_anon", | |
123 | "active_anon", | |
124 | "inactive_file", | |
125 | "active_file", | |
126 | "unevictable", | |
127 | }; | |
128 | ||
a0db00fc KS |
129 | #define THRESHOLDS_EVENTS_TARGET 128 |
130 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
131 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 132 | |
bb4cc1a8 AM |
133 | /* |
134 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
135 | * their hierarchy representation | |
136 | */ | |
137 | ||
ef8f2327 | 138 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 AM |
139 | struct rb_root rb_root; |
140 | spinlock_t lock; | |
141 | }; | |
142 | ||
bb4cc1a8 AM |
143 | struct mem_cgroup_tree { |
144 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
145 | }; | |
146 | ||
147 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
148 | ||
9490ff27 KH |
149 | /* for OOM */ |
150 | struct mem_cgroup_eventfd_list { | |
151 | struct list_head list; | |
152 | struct eventfd_ctx *eventfd; | |
153 | }; | |
2e72b634 | 154 | |
79bd9814 TH |
155 | /* |
156 | * cgroup_event represents events which userspace want to receive. | |
157 | */ | |
3bc942f3 | 158 | struct mem_cgroup_event { |
79bd9814 | 159 | /* |
59b6f873 | 160 | * memcg which the event belongs to. |
79bd9814 | 161 | */ |
59b6f873 | 162 | struct mem_cgroup *memcg; |
79bd9814 TH |
163 | /* |
164 | * eventfd to signal userspace about the event. | |
165 | */ | |
166 | struct eventfd_ctx *eventfd; | |
167 | /* | |
168 | * Each of these stored in a list by the cgroup. | |
169 | */ | |
170 | struct list_head list; | |
fba94807 TH |
171 | /* |
172 | * register_event() callback will be used to add new userspace | |
173 | * waiter for changes related to this event. Use eventfd_signal() | |
174 | * on eventfd to send notification to userspace. | |
175 | */ | |
59b6f873 | 176 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 177 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
178 | /* |
179 | * unregister_event() callback will be called when userspace closes | |
180 | * the eventfd or on cgroup removing. This callback must be set, | |
181 | * if you want provide notification functionality. | |
182 | */ | |
59b6f873 | 183 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 184 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
185 | /* |
186 | * All fields below needed to unregister event when | |
187 | * userspace closes eventfd. | |
188 | */ | |
189 | poll_table pt; | |
190 | wait_queue_head_t *wqh; | |
191 | wait_queue_t wait; | |
192 | struct work_struct remove; | |
193 | }; | |
194 | ||
c0ff4b85 R |
195 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
196 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 197 | |
7dc74be0 DN |
198 | /* Stuffs for move charges at task migration. */ |
199 | /* | |
1dfab5ab | 200 | * Types of charges to be moved. |
7dc74be0 | 201 | */ |
1dfab5ab JW |
202 | #define MOVE_ANON 0x1U |
203 | #define MOVE_FILE 0x2U | |
204 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 205 | |
4ffef5fe DN |
206 | /* "mc" and its members are protected by cgroup_mutex */ |
207 | static struct move_charge_struct { | |
b1dd693e | 208 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 209 | struct mm_struct *mm; |
4ffef5fe DN |
210 | struct mem_cgroup *from; |
211 | struct mem_cgroup *to; | |
1dfab5ab | 212 | unsigned long flags; |
4ffef5fe | 213 | unsigned long precharge; |
854ffa8d | 214 | unsigned long moved_charge; |
483c30b5 | 215 | unsigned long moved_swap; |
8033b97c DN |
216 | struct task_struct *moving_task; /* a task moving charges */ |
217 | wait_queue_head_t waitq; /* a waitq for other context */ | |
218 | } mc = { | |
2bd9bb20 | 219 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
220 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
221 | }; | |
4ffef5fe | 222 | |
4e416953 BS |
223 | /* |
224 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
225 | * limit reclaim to prevent infinite loops, if they ever occur. | |
226 | */ | |
a0db00fc | 227 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 228 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 229 | |
217bc319 KH |
230 | enum charge_type { |
231 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 232 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 233 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 234 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
235 | NR_CHARGE_TYPE, |
236 | }; | |
237 | ||
8c7c6e34 | 238 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
239 | enum res_type { |
240 | _MEM, | |
241 | _MEMSWAP, | |
242 | _OOM_TYPE, | |
510fc4e1 | 243 | _KMEM, |
d55f90bf | 244 | _TCP, |
86ae53e1 GC |
245 | }; |
246 | ||
a0db00fc KS |
247 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
248 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 249 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
250 | /* Used for OOM nofiier */ |
251 | #define OOM_CONTROL (0) | |
8c7c6e34 | 252 | |
70ddf637 AV |
253 | /* Some nice accessors for the vmpressure. */ |
254 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
255 | { | |
256 | if (!memcg) | |
257 | memcg = root_mem_cgroup; | |
258 | return &memcg->vmpressure; | |
259 | } | |
260 | ||
261 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
262 | { | |
263 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
264 | } | |
265 | ||
7ffc0edc MH |
266 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
267 | { | |
268 | return (memcg == root_mem_cgroup); | |
269 | } | |
270 | ||
127424c8 | 271 | #ifndef CONFIG_SLOB |
55007d84 | 272 | /* |
f7ce3190 | 273 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
274 | * The main reason for not using cgroup id for this: |
275 | * this works better in sparse environments, where we have a lot of memcgs, | |
276 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
277 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
278 | * 200 entry array for that. | |
55007d84 | 279 | * |
dbcf73e2 VD |
280 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
281 | * will double each time we have to increase it. | |
55007d84 | 282 | */ |
dbcf73e2 VD |
283 | static DEFINE_IDA(memcg_cache_ida); |
284 | int memcg_nr_cache_ids; | |
749c5415 | 285 | |
05257a1a VD |
286 | /* Protects memcg_nr_cache_ids */ |
287 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
288 | ||
289 | void memcg_get_cache_ids(void) | |
290 | { | |
291 | down_read(&memcg_cache_ids_sem); | |
292 | } | |
293 | ||
294 | void memcg_put_cache_ids(void) | |
295 | { | |
296 | up_read(&memcg_cache_ids_sem); | |
297 | } | |
298 | ||
55007d84 GC |
299 | /* |
300 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
301 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
302 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
303 | * tunable, but that is strictly not necessary. | |
304 | * | |
b8627835 | 305 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
306 | * this constant directly from cgroup, but it is understandable that this is |
307 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 308 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
309 | * increase ours as well if it increases. |
310 | */ | |
311 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 312 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 313 | |
d7f25f8a GC |
314 | /* |
315 | * A lot of the calls to the cache allocation functions are expected to be | |
316 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
317 | * conditional to this static branch, we'll have to allow modules that does | |
318 | * kmem_cache_alloc and the such to see this symbol as well | |
319 | */ | |
ef12947c | 320 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 321 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 322 | |
17cc4dfe TH |
323 | struct workqueue_struct *memcg_kmem_cache_wq; |
324 | ||
127424c8 | 325 | #endif /* !CONFIG_SLOB */ |
a8964b9b | 326 | |
ad7fa852 TH |
327 | /** |
328 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
329 | * @page: page of interest | |
330 | * | |
331 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
332 | * with @page is returned. The returned css remains associated with @page | |
333 | * until it is released. | |
334 | * | |
335 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
336 | * is returned. | |
ad7fa852 TH |
337 | */ |
338 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
339 | { | |
340 | struct mem_cgroup *memcg; | |
341 | ||
ad7fa852 TH |
342 | memcg = page->mem_cgroup; |
343 | ||
9e10a130 | 344 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
345 | memcg = root_mem_cgroup; |
346 | ||
ad7fa852 TH |
347 | return &memcg->css; |
348 | } | |
349 | ||
2fc04524 VD |
350 | /** |
351 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
352 | * @page: the page | |
353 | * | |
354 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
355 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
356 | * is safe to call this function without holding a reference to @page. | |
357 | * | |
358 | * Note, this function is inherently racy, because there is nothing to prevent | |
359 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
360 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
361 | * do not care (such as procfs interfaces). | |
362 | */ | |
363 | ino_t page_cgroup_ino(struct page *page) | |
364 | { | |
365 | struct mem_cgroup *memcg; | |
366 | unsigned long ino = 0; | |
367 | ||
368 | rcu_read_lock(); | |
369 | memcg = READ_ONCE(page->mem_cgroup); | |
370 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
371 | memcg = parent_mem_cgroup(memcg); | |
372 | if (memcg) | |
373 | ino = cgroup_ino(memcg->css.cgroup); | |
374 | rcu_read_unlock(); | |
375 | return ino; | |
376 | } | |
377 | ||
ef8f2327 MG |
378 | static struct mem_cgroup_per_node * |
379 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 380 | { |
97a6c37b | 381 | int nid = page_to_nid(page); |
f64c3f54 | 382 | |
ef8f2327 | 383 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
384 | } |
385 | ||
ef8f2327 MG |
386 | static struct mem_cgroup_tree_per_node * |
387 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 388 | { |
ef8f2327 | 389 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
390 | } |
391 | ||
ef8f2327 | 392 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
393 | soft_limit_tree_from_page(struct page *page) |
394 | { | |
395 | int nid = page_to_nid(page); | |
bb4cc1a8 | 396 | |
ef8f2327 | 397 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
398 | } |
399 | ||
ef8f2327 MG |
400 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
401 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 402 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
403 | { |
404 | struct rb_node **p = &mctz->rb_root.rb_node; | |
405 | struct rb_node *parent = NULL; | |
ef8f2327 | 406 | struct mem_cgroup_per_node *mz_node; |
bb4cc1a8 AM |
407 | |
408 | if (mz->on_tree) | |
409 | return; | |
410 | ||
411 | mz->usage_in_excess = new_usage_in_excess; | |
412 | if (!mz->usage_in_excess) | |
413 | return; | |
414 | while (*p) { | |
415 | parent = *p; | |
ef8f2327 | 416 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 AM |
417 | tree_node); |
418 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
419 | p = &(*p)->rb_left; | |
420 | /* | |
421 | * We can't avoid mem cgroups that are over their soft | |
422 | * limit by the same amount | |
423 | */ | |
424 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
425 | p = &(*p)->rb_right; | |
426 | } | |
427 | rb_link_node(&mz->tree_node, parent, p); | |
428 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
429 | mz->on_tree = true; | |
430 | } | |
431 | ||
ef8f2327 MG |
432 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
433 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
434 | { |
435 | if (!mz->on_tree) | |
436 | return; | |
437 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
438 | mz->on_tree = false; | |
439 | } | |
440 | ||
ef8f2327 MG |
441 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
442 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 443 | { |
0a31bc97 JW |
444 | unsigned long flags; |
445 | ||
446 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 447 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 448 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
449 | } |
450 | ||
3e32cb2e JW |
451 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
452 | { | |
453 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 454 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
455 | unsigned long excess = 0; |
456 | ||
457 | if (nr_pages > soft_limit) | |
458 | excess = nr_pages - soft_limit; | |
459 | ||
460 | return excess; | |
461 | } | |
bb4cc1a8 AM |
462 | |
463 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
464 | { | |
3e32cb2e | 465 | unsigned long excess; |
ef8f2327 MG |
466 | struct mem_cgroup_per_node *mz; |
467 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 468 | |
e231875b | 469 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
470 | if (!mctz) |
471 | return; | |
bb4cc1a8 AM |
472 | /* |
473 | * Necessary to update all ancestors when hierarchy is used. | |
474 | * because their event counter is not touched. | |
475 | */ | |
476 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 477 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 478 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
479 | /* |
480 | * We have to update the tree if mz is on RB-tree or | |
481 | * mem is over its softlimit. | |
482 | */ | |
483 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
484 | unsigned long flags; |
485 | ||
486 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
487 | /* if on-tree, remove it */ |
488 | if (mz->on_tree) | |
cf2c8127 | 489 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
490 | /* |
491 | * Insert again. mz->usage_in_excess will be updated. | |
492 | * If excess is 0, no tree ops. | |
493 | */ | |
cf2c8127 | 494 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 495 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
496 | } |
497 | } | |
498 | } | |
499 | ||
500 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
501 | { | |
ef8f2327 MG |
502 | struct mem_cgroup_tree_per_node *mctz; |
503 | struct mem_cgroup_per_node *mz; | |
504 | int nid; | |
bb4cc1a8 | 505 | |
e231875b | 506 | for_each_node(nid) { |
ef8f2327 MG |
507 | mz = mem_cgroup_nodeinfo(memcg, nid); |
508 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
509 | if (mctz) |
510 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
511 | } |
512 | } | |
513 | ||
ef8f2327 MG |
514 | static struct mem_cgroup_per_node * |
515 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
516 | { |
517 | struct rb_node *rightmost = NULL; | |
ef8f2327 | 518 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
519 | |
520 | retry: | |
521 | mz = NULL; | |
522 | rightmost = rb_last(&mctz->rb_root); | |
523 | if (!rightmost) | |
524 | goto done; /* Nothing to reclaim from */ | |
525 | ||
ef8f2327 | 526 | mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node); |
bb4cc1a8 AM |
527 | /* |
528 | * Remove the node now but someone else can add it back, | |
529 | * we will to add it back at the end of reclaim to its correct | |
530 | * position in the tree. | |
531 | */ | |
cf2c8127 | 532 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 533 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 534 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
535 | goto retry; |
536 | done: | |
537 | return mz; | |
538 | } | |
539 | ||
ef8f2327 MG |
540 | static struct mem_cgroup_per_node * |
541 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 542 | { |
ef8f2327 | 543 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 544 | |
0a31bc97 | 545 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 546 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 547 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
548 | return mz; |
549 | } | |
550 | ||
711d3d2c | 551 | /* |
484ebb3b GT |
552 | * Return page count for single (non recursive) @memcg. |
553 | * | |
711d3d2c KH |
554 | * Implementation Note: reading percpu statistics for memcg. |
555 | * | |
556 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
557 | * synchronization to implement "quick" read. There are trade-off between | |
558 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 559 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
560 | * |
561 | * But this _read() function is used for user interface now. The user accounts | |
562 | * memory usage by memory cgroup and he _always_ requires exact value because | |
563 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
564 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
565 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
566 | * | |
567 | * If there are kernel internal actions which can make use of some not-exact | |
568 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 569 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
570 | * implemented. |
571 | */ | |
484ebb3b GT |
572 | static unsigned long |
573 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 574 | { |
7a159cc9 | 575 | long val = 0; |
c62b1a3b | 576 | int cpu; |
c62b1a3b | 577 | |
484ebb3b | 578 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 579 | for_each_possible_cpu(cpu) |
c0ff4b85 | 580 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
581 | /* |
582 | * Summing races with updates, so val may be negative. Avoid exposing | |
583 | * transient negative values. | |
584 | */ | |
585 | if (val < 0) | |
586 | val = 0; | |
c62b1a3b KH |
587 | return val; |
588 | } | |
589 | ||
c0ff4b85 | 590 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
591 | enum mem_cgroup_events_index idx) |
592 | { | |
593 | unsigned long val = 0; | |
594 | int cpu; | |
595 | ||
733a572e | 596 | for_each_possible_cpu(cpu) |
c0ff4b85 | 597 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
598 | return val; |
599 | } | |
600 | ||
c0ff4b85 | 601 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 602 | struct page *page, |
f627c2f5 | 603 | bool compound, int nr_pages) |
d52aa412 | 604 | { |
b2402857 KH |
605 | /* |
606 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
607 | * counted as CACHE even if it's on ANON LRU. | |
608 | */ | |
0a31bc97 | 609 | if (PageAnon(page)) |
b2402857 | 610 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 611 | nr_pages); |
9a4caf1e | 612 | else { |
b2402857 | 613 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 614 | nr_pages); |
9a4caf1e JW |
615 | if (PageSwapBacked(page)) |
616 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SHMEM], | |
617 | nr_pages); | |
618 | } | |
55e462b0 | 619 | |
f627c2f5 KS |
620 | if (compound) { |
621 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
b070e65c DR |
622 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
623 | nr_pages); | |
f627c2f5 | 624 | } |
b070e65c | 625 | |
e401f176 KH |
626 | /* pagein of a big page is an event. So, ignore page size */ |
627 | if (nr_pages > 0) | |
c0ff4b85 | 628 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 629 | else { |
c0ff4b85 | 630 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
631 | nr_pages = -nr_pages; /* for event */ |
632 | } | |
e401f176 | 633 | |
13114716 | 634 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
635 | } |
636 | ||
0a6b76dd VD |
637 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
638 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 639 | { |
b4536f0c | 640 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
e231875b | 641 | unsigned long nr = 0; |
ef8f2327 | 642 | enum lru_list lru; |
889976db | 643 | |
e231875b | 644 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 645 | |
ef8f2327 MG |
646 | for_each_lru(lru) { |
647 | if (!(BIT(lru) & lru_mask)) | |
648 | continue; | |
b4536f0c | 649 | nr += mem_cgroup_get_lru_size(lruvec, lru); |
e231875b JZ |
650 | } |
651 | return nr; | |
889976db | 652 | } |
bb2a0de9 | 653 | |
c0ff4b85 | 654 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 655 | unsigned int lru_mask) |
6d12e2d8 | 656 | { |
e231875b | 657 | unsigned long nr = 0; |
889976db | 658 | int nid; |
6d12e2d8 | 659 | |
31aaea4a | 660 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
661 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
662 | return nr; | |
d52aa412 KH |
663 | } |
664 | ||
f53d7ce3 JW |
665 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
666 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
667 | { |
668 | unsigned long val, next; | |
669 | ||
13114716 | 670 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 671 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 672 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
673 | if ((long)next - (long)val < 0) { |
674 | switch (target) { | |
675 | case MEM_CGROUP_TARGET_THRESH: | |
676 | next = val + THRESHOLDS_EVENTS_TARGET; | |
677 | break; | |
bb4cc1a8 AM |
678 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
679 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
680 | break; | |
f53d7ce3 JW |
681 | case MEM_CGROUP_TARGET_NUMAINFO: |
682 | next = val + NUMAINFO_EVENTS_TARGET; | |
683 | break; | |
684 | default: | |
685 | break; | |
686 | } | |
687 | __this_cpu_write(memcg->stat->targets[target], next); | |
688 | return true; | |
7a159cc9 | 689 | } |
f53d7ce3 | 690 | return false; |
d2265e6f KH |
691 | } |
692 | ||
693 | /* | |
694 | * Check events in order. | |
695 | * | |
696 | */ | |
c0ff4b85 | 697 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
698 | { |
699 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
700 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
701 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 702 | bool do_softlimit; |
82b3f2a7 | 703 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 704 | |
bb4cc1a8 AM |
705 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
706 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
707 | #if MAX_NUMNODES > 1 |
708 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
709 | MEM_CGROUP_TARGET_NUMAINFO); | |
710 | #endif | |
c0ff4b85 | 711 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
712 | if (unlikely(do_softlimit)) |
713 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 714 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 715 | if (unlikely(do_numainfo)) |
c0ff4b85 | 716 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 717 | #endif |
0a31bc97 | 718 | } |
d2265e6f KH |
719 | } |
720 | ||
cf475ad2 | 721 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 722 | { |
31a78f23 BS |
723 | /* |
724 | * mm_update_next_owner() may clear mm->owner to NULL | |
725 | * if it races with swapoff, page migration, etc. | |
726 | * So this can be called with p == NULL. | |
727 | */ | |
728 | if (unlikely(!p)) | |
729 | return NULL; | |
730 | ||
073219e9 | 731 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 732 | } |
33398cf2 | 733 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 734 | |
df381975 | 735 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 736 | { |
c0ff4b85 | 737 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 738 | |
54595fe2 KH |
739 | rcu_read_lock(); |
740 | do { | |
6f6acb00 MH |
741 | /* |
742 | * Page cache insertions can happen withou an | |
743 | * actual mm context, e.g. during disk probing | |
744 | * on boot, loopback IO, acct() writes etc. | |
745 | */ | |
746 | if (unlikely(!mm)) | |
df381975 | 747 | memcg = root_mem_cgroup; |
6f6acb00 MH |
748 | else { |
749 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
750 | if (unlikely(!memcg)) | |
751 | memcg = root_mem_cgroup; | |
752 | } | |
ec903c0c | 753 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 754 | rcu_read_unlock(); |
c0ff4b85 | 755 | return memcg; |
54595fe2 KH |
756 | } |
757 | ||
5660048c JW |
758 | /** |
759 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
760 | * @root: hierarchy root | |
761 | * @prev: previously returned memcg, NULL on first invocation | |
762 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
763 | * | |
764 | * Returns references to children of the hierarchy below @root, or | |
765 | * @root itself, or %NULL after a full round-trip. | |
766 | * | |
767 | * Caller must pass the return value in @prev on subsequent | |
768 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
769 | * to cancel a hierarchy walk before the round-trip is complete. | |
770 | * | |
771 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
772 | * divide up the memcgs in the hierarchy among all concurrent | |
773 | * reclaimers operating on the same zone and priority. | |
774 | */ | |
694fbc0f | 775 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 776 | struct mem_cgroup *prev, |
694fbc0f | 777 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 778 | { |
33398cf2 | 779 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 780 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 781 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 782 | struct mem_cgroup *pos = NULL; |
711d3d2c | 783 | |
694fbc0f AM |
784 | if (mem_cgroup_disabled()) |
785 | return NULL; | |
5660048c | 786 | |
9f3a0d09 JW |
787 | if (!root) |
788 | root = root_mem_cgroup; | |
7d74b06f | 789 | |
9f3a0d09 | 790 | if (prev && !reclaim) |
5ac8fb31 | 791 | pos = prev; |
14067bb3 | 792 | |
9f3a0d09 JW |
793 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
794 | if (prev) | |
5ac8fb31 | 795 | goto out; |
694fbc0f | 796 | return root; |
9f3a0d09 | 797 | } |
14067bb3 | 798 | |
542f85f9 | 799 | rcu_read_lock(); |
5f578161 | 800 | |
5ac8fb31 | 801 | if (reclaim) { |
ef8f2327 | 802 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 803 | |
ef8f2327 | 804 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
805 | iter = &mz->iter[reclaim->priority]; |
806 | ||
807 | if (prev && reclaim->generation != iter->generation) | |
808 | goto out_unlock; | |
809 | ||
6df38689 | 810 | while (1) { |
4db0c3c2 | 811 | pos = READ_ONCE(iter->position); |
6df38689 VD |
812 | if (!pos || css_tryget(&pos->css)) |
813 | break; | |
5ac8fb31 | 814 | /* |
6df38689 VD |
815 | * css reference reached zero, so iter->position will |
816 | * be cleared by ->css_released. However, we should not | |
817 | * rely on this happening soon, because ->css_released | |
818 | * is called from a work queue, and by busy-waiting we | |
819 | * might block it. So we clear iter->position right | |
820 | * away. | |
5ac8fb31 | 821 | */ |
6df38689 VD |
822 | (void)cmpxchg(&iter->position, pos, NULL); |
823 | } | |
5ac8fb31 JW |
824 | } |
825 | ||
826 | if (pos) | |
827 | css = &pos->css; | |
828 | ||
829 | for (;;) { | |
830 | css = css_next_descendant_pre(css, &root->css); | |
831 | if (!css) { | |
832 | /* | |
833 | * Reclaimers share the hierarchy walk, and a | |
834 | * new one might jump in right at the end of | |
835 | * the hierarchy - make sure they see at least | |
836 | * one group and restart from the beginning. | |
837 | */ | |
838 | if (!prev) | |
839 | continue; | |
840 | break; | |
527a5ec9 | 841 | } |
7d74b06f | 842 | |
5ac8fb31 JW |
843 | /* |
844 | * Verify the css and acquire a reference. The root | |
845 | * is provided by the caller, so we know it's alive | |
846 | * and kicking, and don't take an extra reference. | |
847 | */ | |
848 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 849 | |
5ac8fb31 JW |
850 | if (css == &root->css) |
851 | break; | |
14067bb3 | 852 | |
0b8f73e1 JW |
853 | if (css_tryget(css)) |
854 | break; | |
9f3a0d09 | 855 | |
5ac8fb31 | 856 | memcg = NULL; |
9f3a0d09 | 857 | } |
5ac8fb31 JW |
858 | |
859 | if (reclaim) { | |
5ac8fb31 | 860 | /* |
6df38689 VD |
861 | * The position could have already been updated by a competing |
862 | * thread, so check that the value hasn't changed since we read | |
863 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 864 | */ |
6df38689 VD |
865 | (void)cmpxchg(&iter->position, pos, memcg); |
866 | ||
5ac8fb31 JW |
867 | if (pos) |
868 | css_put(&pos->css); | |
869 | ||
870 | if (!memcg) | |
871 | iter->generation++; | |
872 | else if (!prev) | |
873 | reclaim->generation = iter->generation; | |
9f3a0d09 | 874 | } |
5ac8fb31 | 875 | |
542f85f9 MH |
876 | out_unlock: |
877 | rcu_read_unlock(); | |
5ac8fb31 | 878 | out: |
c40046f3 MH |
879 | if (prev && prev != root) |
880 | css_put(&prev->css); | |
881 | ||
9f3a0d09 | 882 | return memcg; |
14067bb3 | 883 | } |
7d74b06f | 884 | |
5660048c JW |
885 | /** |
886 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
887 | * @root: hierarchy root | |
888 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
889 | */ | |
890 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
891 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
892 | { |
893 | if (!root) | |
894 | root = root_mem_cgroup; | |
895 | if (prev && prev != root) | |
896 | css_put(&prev->css); | |
897 | } | |
7d74b06f | 898 | |
6df38689 VD |
899 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
900 | { | |
901 | struct mem_cgroup *memcg = dead_memcg; | |
902 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
903 | struct mem_cgroup_per_node *mz; |
904 | int nid; | |
6df38689 VD |
905 | int i; |
906 | ||
907 | while ((memcg = parent_mem_cgroup(memcg))) { | |
908 | for_each_node(nid) { | |
ef8f2327 MG |
909 | mz = mem_cgroup_nodeinfo(memcg, nid); |
910 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
911 | iter = &mz->iter[i]; | |
912 | cmpxchg(&iter->position, | |
913 | dead_memcg, NULL); | |
6df38689 VD |
914 | } |
915 | } | |
916 | } | |
917 | } | |
918 | ||
9f3a0d09 JW |
919 | /* |
920 | * Iteration constructs for visiting all cgroups (under a tree). If | |
921 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
922 | * be used for reference counting. | |
923 | */ | |
924 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 925 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 926 | iter != NULL; \ |
527a5ec9 | 927 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 928 | |
9f3a0d09 | 929 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 930 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 931 | iter != NULL; \ |
527a5ec9 | 932 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 933 | |
7c5f64f8 VD |
934 | /** |
935 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
936 | * @memcg: hierarchy root | |
937 | * @fn: function to call for each task | |
938 | * @arg: argument passed to @fn | |
939 | * | |
940 | * This function iterates over tasks attached to @memcg or to any of its | |
941 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
942 | * value, the function breaks the iteration loop and returns the value. | |
943 | * Otherwise, it will iterate over all tasks and return 0. | |
944 | * | |
945 | * This function must not be called for the root memory cgroup. | |
946 | */ | |
947 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
948 | int (*fn)(struct task_struct *, void *), void *arg) | |
949 | { | |
950 | struct mem_cgroup *iter; | |
951 | int ret = 0; | |
952 | ||
953 | BUG_ON(memcg == root_mem_cgroup); | |
954 | ||
955 | for_each_mem_cgroup_tree(iter, memcg) { | |
956 | struct css_task_iter it; | |
957 | struct task_struct *task; | |
958 | ||
959 | css_task_iter_start(&iter->css, &it); | |
960 | while (!ret && (task = css_task_iter_next(&it))) | |
961 | ret = fn(task, arg); | |
962 | css_task_iter_end(&it); | |
963 | if (ret) { | |
964 | mem_cgroup_iter_break(memcg, iter); | |
965 | break; | |
966 | } | |
967 | } | |
968 | return ret; | |
969 | } | |
970 | ||
925b7673 | 971 | /** |
dfe0e773 | 972 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 973 | * @page: the page |
fa9add64 | 974 | * @zone: zone of the page |
dfe0e773 JW |
975 | * |
976 | * This function is only safe when following the LRU page isolation | |
977 | * and putback protocol: the LRU lock must be held, and the page must | |
978 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 979 | */ |
599d0c95 | 980 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 981 | { |
ef8f2327 | 982 | struct mem_cgroup_per_node *mz; |
925b7673 | 983 | struct mem_cgroup *memcg; |
bea8c150 | 984 | struct lruvec *lruvec; |
6d12e2d8 | 985 | |
bea8c150 | 986 | if (mem_cgroup_disabled()) { |
599d0c95 | 987 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
988 | goto out; |
989 | } | |
925b7673 | 990 | |
1306a85a | 991 | memcg = page->mem_cgroup; |
7512102c | 992 | /* |
dfe0e773 | 993 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 994 | * possibly migrated - before they are charged. |
7512102c | 995 | */ |
29833315 JW |
996 | if (!memcg) |
997 | memcg = root_mem_cgroup; | |
7512102c | 998 | |
ef8f2327 | 999 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1000 | lruvec = &mz->lruvec; |
1001 | out: | |
1002 | /* | |
1003 | * Since a node can be onlined after the mem_cgroup was created, | |
1004 | * we have to be prepared to initialize lruvec->zone here; | |
1005 | * and if offlined then reonlined, we need to reinitialize it. | |
1006 | */ | |
599d0c95 MG |
1007 | if (unlikely(lruvec->pgdat != pgdat)) |
1008 | lruvec->pgdat = pgdat; | |
bea8c150 | 1009 | return lruvec; |
08e552c6 | 1010 | } |
b69408e8 | 1011 | |
925b7673 | 1012 | /** |
fa9add64 HD |
1013 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1014 | * @lruvec: mem_cgroup per zone lru vector | |
1015 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1016 | * @zid: zone id of the accounted pages |
fa9add64 | 1017 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1018 | * |
ca707239 HD |
1019 | * This function must be called under lru_lock, just before a page is added |
1020 | * to or just after a page is removed from an lru list (that ordering being | |
1021 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1022 | */ |
fa9add64 | 1023 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1024 | int zid, int nr_pages) |
3f58a829 | 1025 | { |
ef8f2327 | 1026 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1027 | unsigned long *lru_size; |
ca707239 | 1028 | long size; |
3f58a829 MK |
1029 | |
1030 | if (mem_cgroup_disabled()) | |
1031 | return; | |
1032 | ||
ef8f2327 | 1033 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1034 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1035 | |
1036 | if (nr_pages < 0) | |
1037 | *lru_size += nr_pages; | |
1038 | ||
1039 | size = *lru_size; | |
b4536f0c MH |
1040 | if (WARN_ONCE(size < 0, |
1041 | "%s(%p, %d, %d): lru_size %ld\n", | |
1042 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1043 | VM_BUG_ON(1); |
1044 | *lru_size = 0; | |
1045 | } | |
1046 | ||
1047 | if (nr_pages > 0) | |
1048 | *lru_size += nr_pages; | |
08e552c6 | 1049 | } |
544122e5 | 1050 | |
2314b42d | 1051 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1052 | { |
2314b42d | 1053 | struct mem_cgroup *task_memcg; |
158e0a2d | 1054 | struct task_struct *p; |
ffbdccf5 | 1055 | bool ret; |
4c4a2214 | 1056 | |
158e0a2d | 1057 | p = find_lock_task_mm(task); |
de077d22 | 1058 | if (p) { |
2314b42d | 1059 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1060 | task_unlock(p); |
1061 | } else { | |
1062 | /* | |
1063 | * All threads may have already detached their mm's, but the oom | |
1064 | * killer still needs to detect if they have already been oom | |
1065 | * killed to prevent needlessly killing additional tasks. | |
1066 | */ | |
ffbdccf5 | 1067 | rcu_read_lock(); |
2314b42d JW |
1068 | task_memcg = mem_cgroup_from_task(task); |
1069 | css_get(&task_memcg->css); | |
ffbdccf5 | 1070 | rcu_read_unlock(); |
de077d22 | 1071 | } |
2314b42d JW |
1072 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1073 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1074 | return ret; |
1075 | } | |
1076 | ||
19942822 | 1077 | /** |
9d11ea9f | 1078 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1079 | * @memcg: the memory cgroup |
19942822 | 1080 | * |
9d11ea9f | 1081 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1082 | * pages. |
19942822 | 1083 | */ |
c0ff4b85 | 1084 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1085 | { |
3e32cb2e JW |
1086 | unsigned long margin = 0; |
1087 | unsigned long count; | |
1088 | unsigned long limit; | |
9d11ea9f | 1089 | |
3e32cb2e | 1090 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1091 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1092 | if (count < limit) |
1093 | margin = limit - count; | |
1094 | ||
7941d214 | 1095 | if (do_memsw_account()) { |
3e32cb2e | 1096 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1097 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1098 | if (count <= limit) |
1099 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1100 | else |
1101 | margin = 0; | |
3e32cb2e JW |
1102 | } |
1103 | ||
1104 | return margin; | |
19942822 JW |
1105 | } |
1106 | ||
32047e2a | 1107 | /* |
bdcbb659 | 1108 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1109 | * |
bdcbb659 QH |
1110 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1111 | * moving cgroups. This is for waiting at high-memory pressure | |
1112 | * caused by "move". | |
32047e2a | 1113 | */ |
c0ff4b85 | 1114 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1115 | { |
2bd9bb20 KH |
1116 | struct mem_cgroup *from; |
1117 | struct mem_cgroup *to; | |
4b534334 | 1118 | bool ret = false; |
2bd9bb20 KH |
1119 | /* |
1120 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1121 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1122 | */ | |
1123 | spin_lock(&mc.lock); | |
1124 | from = mc.from; | |
1125 | to = mc.to; | |
1126 | if (!from) | |
1127 | goto unlock; | |
3e92041d | 1128 | |
2314b42d JW |
1129 | ret = mem_cgroup_is_descendant(from, memcg) || |
1130 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1131 | unlock: |
1132 | spin_unlock(&mc.lock); | |
4b534334 KH |
1133 | return ret; |
1134 | } | |
1135 | ||
c0ff4b85 | 1136 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1137 | { |
1138 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1139 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1140 | DEFINE_WAIT(wait); |
1141 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1142 | /* moving charge context might have finished. */ | |
1143 | if (mc.moving_task) | |
1144 | schedule(); | |
1145 | finish_wait(&mc.waitq, &wait); | |
1146 | return true; | |
1147 | } | |
1148 | } | |
1149 | return false; | |
1150 | } | |
1151 | ||
58cf188e | 1152 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1153 | /** |
58cf188e | 1154 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1155 | * @memcg: The memory cgroup that went over limit |
1156 | * @p: Task that is going to be killed | |
1157 | * | |
1158 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1159 | * enabled | |
1160 | */ | |
1161 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1162 | { | |
58cf188e SZ |
1163 | struct mem_cgroup *iter; |
1164 | unsigned int i; | |
e222432b | 1165 | |
e222432b BS |
1166 | rcu_read_lock(); |
1167 | ||
2415b9f5 BV |
1168 | if (p) { |
1169 | pr_info("Task in "); | |
1170 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1171 | pr_cont(" killed as a result of limit of "); | |
1172 | } else { | |
1173 | pr_info("Memory limit reached of cgroup "); | |
1174 | } | |
1175 | ||
e61734c5 | 1176 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1177 | pr_cont("\n"); |
e222432b | 1178 | |
e222432b BS |
1179 | rcu_read_unlock(); |
1180 | ||
3e32cb2e JW |
1181 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1182 | K((u64)page_counter_read(&memcg->memory)), | |
1183 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1184 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1185 | K((u64)page_counter_read(&memcg->memsw)), | |
1186 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1187 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1188 | K((u64)page_counter_read(&memcg->kmem)), | |
1189 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1190 | |
1191 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1192 | pr_info("Memory cgroup stats for "); |
1193 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1194 | pr_cont(":"); |
1195 | ||
1196 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
37e84351 | 1197 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
58cf188e | 1198 | continue; |
484ebb3b | 1199 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1200 | K(mem_cgroup_read_stat(iter, i))); |
1201 | } | |
1202 | ||
1203 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1204 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1205 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1206 | ||
1207 | pr_cont("\n"); | |
1208 | } | |
e222432b BS |
1209 | } |
1210 | ||
81d39c20 KH |
1211 | /* |
1212 | * This function returns the number of memcg under hierarchy tree. Returns | |
1213 | * 1(self count) if no children. | |
1214 | */ | |
c0ff4b85 | 1215 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1216 | { |
1217 | int num = 0; | |
7d74b06f KH |
1218 | struct mem_cgroup *iter; |
1219 | ||
c0ff4b85 | 1220 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1221 | num++; |
81d39c20 KH |
1222 | return num; |
1223 | } | |
1224 | ||
a63d83f4 DR |
1225 | /* |
1226 | * Return the memory (and swap, if configured) limit for a memcg. | |
1227 | */ | |
7c5f64f8 | 1228 | unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1229 | { |
3e32cb2e | 1230 | unsigned long limit; |
f3e8eb70 | 1231 | |
3e32cb2e | 1232 | limit = memcg->memory.limit; |
9a5a8f19 | 1233 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1234 | unsigned long memsw_limit; |
37e84351 | 1235 | unsigned long swap_limit; |
9a5a8f19 | 1236 | |
3e32cb2e | 1237 | memsw_limit = memcg->memsw.limit; |
37e84351 VD |
1238 | swap_limit = memcg->swap.limit; |
1239 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); | |
1240 | limit = min(limit + swap_limit, memsw_limit); | |
9a5a8f19 | 1241 | } |
9a5a8f19 | 1242 | return limit; |
a63d83f4 DR |
1243 | } |
1244 | ||
b6e6edcf | 1245 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1246 | int order) |
9cbb78bb | 1247 | { |
6e0fc46d DR |
1248 | struct oom_control oc = { |
1249 | .zonelist = NULL, | |
1250 | .nodemask = NULL, | |
2a966b77 | 1251 | .memcg = memcg, |
6e0fc46d DR |
1252 | .gfp_mask = gfp_mask, |
1253 | .order = order, | |
6e0fc46d | 1254 | }; |
7c5f64f8 | 1255 | bool ret; |
9cbb78bb | 1256 | |
dc56401f | 1257 | mutex_lock(&oom_lock); |
7c5f64f8 | 1258 | ret = out_of_memory(&oc); |
dc56401f | 1259 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1260 | return ret; |
9cbb78bb DR |
1261 | } |
1262 | ||
ae6e71d3 MC |
1263 | #if MAX_NUMNODES > 1 |
1264 | ||
4d0c066d KH |
1265 | /** |
1266 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1267 | * @memcg: the target memcg |
4d0c066d KH |
1268 | * @nid: the node ID to be checked. |
1269 | * @noswap : specify true here if the user wants flle only information. | |
1270 | * | |
1271 | * This function returns whether the specified memcg contains any | |
1272 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1273 | * pages in the node. | |
1274 | */ | |
c0ff4b85 | 1275 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1276 | int nid, bool noswap) |
1277 | { | |
c0ff4b85 | 1278 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1279 | return true; |
1280 | if (noswap || !total_swap_pages) | |
1281 | return false; | |
c0ff4b85 | 1282 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1283 | return true; |
1284 | return false; | |
1285 | ||
1286 | } | |
889976db YH |
1287 | |
1288 | /* | |
1289 | * Always updating the nodemask is not very good - even if we have an empty | |
1290 | * list or the wrong list here, we can start from some node and traverse all | |
1291 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1292 | * | |
1293 | */ | |
c0ff4b85 | 1294 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1295 | { |
1296 | int nid; | |
453a9bf3 KH |
1297 | /* |
1298 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1299 | * pagein/pageout changes since the last update. | |
1300 | */ | |
c0ff4b85 | 1301 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1302 | return; |
c0ff4b85 | 1303 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1304 | return; |
1305 | ||
889976db | 1306 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1307 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1308 | |
31aaea4a | 1309 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1310 | |
c0ff4b85 R |
1311 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1312 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1313 | } |
453a9bf3 | 1314 | |
c0ff4b85 R |
1315 | atomic_set(&memcg->numainfo_events, 0); |
1316 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1317 | } |
1318 | ||
1319 | /* | |
1320 | * Selecting a node where we start reclaim from. Because what we need is just | |
1321 | * reducing usage counter, start from anywhere is O,K. Considering | |
1322 | * memory reclaim from current node, there are pros. and cons. | |
1323 | * | |
1324 | * Freeing memory from current node means freeing memory from a node which | |
1325 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1326 | * hit limits, it will see a contention on a node. But freeing from remote | |
1327 | * node means more costs for memory reclaim because of memory latency. | |
1328 | * | |
1329 | * Now, we use round-robin. Better algorithm is welcomed. | |
1330 | */ | |
c0ff4b85 | 1331 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1332 | { |
1333 | int node; | |
1334 | ||
c0ff4b85 R |
1335 | mem_cgroup_may_update_nodemask(memcg); |
1336 | node = memcg->last_scanned_node; | |
889976db | 1337 | |
0edaf86c | 1338 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1339 | /* |
fda3d69b MH |
1340 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1341 | * last time it really checked all the LRUs due to rate limiting. | |
1342 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1343 | */ |
1344 | if (unlikely(node == MAX_NUMNODES)) | |
1345 | node = numa_node_id(); | |
1346 | ||
c0ff4b85 | 1347 | memcg->last_scanned_node = node; |
889976db YH |
1348 | return node; |
1349 | } | |
889976db | 1350 | #else |
c0ff4b85 | 1351 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1352 | { |
1353 | return 0; | |
1354 | } | |
1355 | #endif | |
1356 | ||
0608f43d | 1357 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1358 | pg_data_t *pgdat, |
0608f43d AM |
1359 | gfp_t gfp_mask, |
1360 | unsigned long *total_scanned) | |
1361 | { | |
1362 | struct mem_cgroup *victim = NULL; | |
1363 | int total = 0; | |
1364 | int loop = 0; | |
1365 | unsigned long excess; | |
1366 | unsigned long nr_scanned; | |
1367 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1368 | .pgdat = pgdat, |
0608f43d AM |
1369 | .priority = 0, |
1370 | }; | |
1371 | ||
3e32cb2e | 1372 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1373 | |
1374 | while (1) { | |
1375 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1376 | if (!victim) { | |
1377 | loop++; | |
1378 | if (loop >= 2) { | |
1379 | /* | |
1380 | * If we have not been able to reclaim | |
1381 | * anything, it might because there are | |
1382 | * no reclaimable pages under this hierarchy | |
1383 | */ | |
1384 | if (!total) | |
1385 | break; | |
1386 | /* | |
1387 | * We want to do more targeted reclaim. | |
1388 | * excess >> 2 is not to excessive so as to | |
1389 | * reclaim too much, nor too less that we keep | |
1390 | * coming back to reclaim from this cgroup | |
1391 | */ | |
1392 | if (total >= (excess >> 2) || | |
1393 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1394 | break; | |
1395 | } | |
1396 | continue; | |
1397 | } | |
a9dd0a83 | 1398 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1399 | pgdat, &nr_scanned); |
0608f43d | 1400 | *total_scanned += nr_scanned; |
3e32cb2e | 1401 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1402 | break; |
6d61ef40 | 1403 | } |
0608f43d AM |
1404 | mem_cgroup_iter_break(root_memcg, victim); |
1405 | return total; | |
6d61ef40 BS |
1406 | } |
1407 | ||
0056f4e6 JW |
1408 | #ifdef CONFIG_LOCKDEP |
1409 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1410 | .name = "memcg_oom_lock", | |
1411 | }; | |
1412 | #endif | |
1413 | ||
fb2a6fc5 JW |
1414 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1415 | ||
867578cb KH |
1416 | /* |
1417 | * Check OOM-Killer is already running under our hierarchy. | |
1418 | * If someone is running, return false. | |
1419 | */ | |
fb2a6fc5 | 1420 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1421 | { |
79dfdacc | 1422 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1423 | |
fb2a6fc5 JW |
1424 | spin_lock(&memcg_oom_lock); |
1425 | ||
9f3a0d09 | 1426 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1427 | if (iter->oom_lock) { |
79dfdacc MH |
1428 | /* |
1429 | * this subtree of our hierarchy is already locked | |
1430 | * so we cannot give a lock. | |
1431 | */ | |
79dfdacc | 1432 | failed = iter; |
9f3a0d09 JW |
1433 | mem_cgroup_iter_break(memcg, iter); |
1434 | break; | |
23751be0 JW |
1435 | } else |
1436 | iter->oom_lock = true; | |
7d74b06f | 1437 | } |
867578cb | 1438 | |
fb2a6fc5 JW |
1439 | if (failed) { |
1440 | /* | |
1441 | * OK, we failed to lock the whole subtree so we have | |
1442 | * to clean up what we set up to the failing subtree | |
1443 | */ | |
1444 | for_each_mem_cgroup_tree(iter, memcg) { | |
1445 | if (iter == failed) { | |
1446 | mem_cgroup_iter_break(memcg, iter); | |
1447 | break; | |
1448 | } | |
1449 | iter->oom_lock = false; | |
79dfdacc | 1450 | } |
0056f4e6 JW |
1451 | } else |
1452 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1453 | |
1454 | spin_unlock(&memcg_oom_lock); | |
1455 | ||
1456 | return !failed; | |
a636b327 | 1457 | } |
0b7f569e | 1458 | |
fb2a6fc5 | 1459 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1460 | { |
7d74b06f KH |
1461 | struct mem_cgroup *iter; |
1462 | ||
fb2a6fc5 | 1463 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1464 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1465 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1466 | iter->oom_lock = false; |
fb2a6fc5 | 1467 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1468 | } |
1469 | ||
c0ff4b85 | 1470 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1471 | { |
1472 | struct mem_cgroup *iter; | |
1473 | ||
c2b42d3c | 1474 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1475 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1476 | iter->under_oom++; |
1477 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1478 | } |
1479 | ||
c0ff4b85 | 1480 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1481 | { |
1482 | struct mem_cgroup *iter; | |
1483 | ||
867578cb KH |
1484 | /* |
1485 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1486 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1487 | */ |
c2b42d3c | 1488 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1489 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1490 | if (iter->under_oom > 0) |
1491 | iter->under_oom--; | |
1492 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1493 | } |
1494 | ||
867578cb KH |
1495 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1496 | ||
dc98df5a | 1497 | struct oom_wait_info { |
d79154bb | 1498 | struct mem_cgroup *memcg; |
dc98df5a KH |
1499 | wait_queue_t wait; |
1500 | }; | |
1501 | ||
1502 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1503 | unsigned mode, int sync, void *arg) | |
1504 | { | |
d79154bb HD |
1505 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1506 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1507 | struct oom_wait_info *oom_wait_info; |
1508 | ||
1509 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1510 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1511 | |
2314b42d JW |
1512 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1513 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1514 | return 0; |
dc98df5a KH |
1515 | return autoremove_wake_function(wait, mode, sync, arg); |
1516 | } | |
1517 | ||
c0ff4b85 | 1518 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1519 | { |
c2b42d3c TH |
1520 | /* |
1521 | * For the following lockless ->under_oom test, the only required | |
1522 | * guarantee is that it must see the state asserted by an OOM when | |
1523 | * this function is called as a result of userland actions | |
1524 | * triggered by the notification of the OOM. This is trivially | |
1525 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1526 | * triggering notification. | |
1527 | */ | |
1528 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1529 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1530 | } |
1531 | ||
3812c8c8 | 1532 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1533 | { |
d0db7afa | 1534 | if (!current->memcg_may_oom) |
3812c8c8 | 1535 | return; |
867578cb | 1536 | /* |
49426420 JW |
1537 | * We are in the middle of the charge context here, so we |
1538 | * don't want to block when potentially sitting on a callstack | |
1539 | * that holds all kinds of filesystem and mm locks. | |
1540 | * | |
1541 | * Also, the caller may handle a failed allocation gracefully | |
1542 | * (like optional page cache readahead) and so an OOM killer | |
1543 | * invocation might not even be necessary. | |
1544 | * | |
1545 | * That's why we don't do anything here except remember the | |
1546 | * OOM context and then deal with it at the end of the page | |
1547 | * fault when the stack is unwound, the locks are released, | |
1548 | * and when we know whether the fault was overall successful. | |
867578cb | 1549 | */ |
49426420 | 1550 | css_get(&memcg->css); |
626ebc41 TH |
1551 | current->memcg_in_oom = memcg; |
1552 | current->memcg_oom_gfp_mask = mask; | |
1553 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1554 | } |
1555 | ||
1556 | /** | |
1557 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1558 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1559 | * |
49426420 JW |
1560 | * This has to be called at the end of a page fault if the memcg OOM |
1561 | * handler was enabled. | |
3812c8c8 | 1562 | * |
49426420 | 1563 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1564 | * sleep on a waitqueue until the userspace task resolves the |
1565 | * situation. Sleeping directly in the charge context with all kinds | |
1566 | * of locks held is not a good idea, instead we remember an OOM state | |
1567 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1568 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1569 | * |
1570 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1571 | * completed, %false otherwise. |
3812c8c8 | 1572 | */ |
49426420 | 1573 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1574 | { |
626ebc41 | 1575 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1576 | struct oom_wait_info owait; |
49426420 | 1577 | bool locked; |
3812c8c8 JW |
1578 | |
1579 | /* OOM is global, do not handle */ | |
3812c8c8 | 1580 | if (!memcg) |
49426420 | 1581 | return false; |
3812c8c8 | 1582 | |
7c5f64f8 | 1583 | if (!handle) |
49426420 | 1584 | goto cleanup; |
3812c8c8 JW |
1585 | |
1586 | owait.memcg = memcg; | |
1587 | owait.wait.flags = 0; | |
1588 | owait.wait.func = memcg_oom_wake_function; | |
1589 | owait.wait.private = current; | |
1590 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1591 | |
3812c8c8 | 1592 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1593 | mem_cgroup_mark_under_oom(memcg); |
1594 | ||
1595 | locked = mem_cgroup_oom_trylock(memcg); | |
1596 | ||
1597 | if (locked) | |
1598 | mem_cgroup_oom_notify(memcg); | |
1599 | ||
1600 | if (locked && !memcg->oom_kill_disable) { | |
1601 | mem_cgroup_unmark_under_oom(memcg); | |
1602 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1603 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1604 | current->memcg_oom_order); | |
49426420 | 1605 | } else { |
3812c8c8 | 1606 | schedule(); |
49426420 JW |
1607 | mem_cgroup_unmark_under_oom(memcg); |
1608 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1609 | } | |
1610 | ||
1611 | if (locked) { | |
fb2a6fc5 JW |
1612 | mem_cgroup_oom_unlock(memcg); |
1613 | /* | |
1614 | * There is no guarantee that an OOM-lock contender | |
1615 | * sees the wakeups triggered by the OOM kill | |
1616 | * uncharges. Wake any sleepers explicitely. | |
1617 | */ | |
1618 | memcg_oom_recover(memcg); | |
1619 | } | |
49426420 | 1620 | cleanup: |
626ebc41 | 1621 | current->memcg_in_oom = NULL; |
3812c8c8 | 1622 | css_put(&memcg->css); |
867578cb | 1623 | return true; |
0b7f569e KH |
1624 | } |
1625 | ||
d7365e78 | 1626 | /** |
81f8c3a4 JW |
1627 | * lock_page_memcg - lock a page->mem_cgroup binding |
1628 | * @page: the page | |
32047e2a | 1629 | * |
81f8c3a4 JW |
1630 | * This function protects unlocked LRU pages from being moved to |
1631 | * another cgroup and stabilizes their page->mem_cgroup binding. | |
d69b042f | 1632 | */ |
62cccb8c | 1633 | void lock_page_memcg(struct page *page) |
89c06bd5 KH |
1634 | { |
1635 | struct mem_cgroup *memcg; | |
6de22619 | 1636 | unsigned long flags; |
89c06bd5 | 1637 | |
6de22619 JW |
1638 | /* |
1639 | * The RCU lock is held throughout the transaction. The fast | |
1640 | * path can get away without acquiring the memcg->move_lock | |
1641 | * because page moving starts with an RCU grace period. | |
6de22619 | 1642 | */ |
d7365e78 JW |
1643 | rcu_read_lock(); |
1644 | ||
1645 | if (mem_cgroup_disabled()) | |
62cccb8c | 1646 | return; |
89c06bd5 | 1647 | again: |
1306a85a | 1648 | memcg = page->mem_cgroup; |
29833315 | 1649 | if (unlikely(!memcg)) |
62cccb8c | 1650 | return; |
d7365e78 | 1651 | |
bdcbb659 | 1652 | if (atomic_read(&memcg->moving_account) <= 0) |
62cccb8c | 1653 | return; |
89c06bd5 | 1654 | |
6de22619 | 1655 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1656 | if (memcg != page->mem_cgroup) { |
6de22619 | 1657 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1658 | goto again; |
1659 | } | |
6de22619 JW |
1660 | |
1661 | /* | |
1662 | * When charge migration first begins, we can have locked and | |
1663 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1664 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1665 | */ |
1666 | memcg->move_lock_task = current; | |
1667 | memcg->move_lock_flags = flags; | |
d7365e78 | 1668 | |
62cccb8c | 1669 | return; |
89c06bd5 | 1670 | } |
81f8c3a4 | 1671 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1672 | |
d7365e78 | 1673 | /** |
81f8c3a4 | 1674 | * unlock_page_memcg - unlock a page->mem_cgroup binding |
62cccb8c | 1675 | * @page: the page |
d7365e78 | 1676 | */ |
62cccb8c | 1677 | void unlock_page_memcg(struct page *page) |
89c06bd5 | 1678 | { |
62cccb8c JW |
1679 | struct mem_cgroup *memcg = page->mem_cgroup; |
1680 | ||
6de22619 JW |
1681 | if (memcg && memcg->move_lock_task == current) { |
1682 | unsigned long flags = memcg->move_lock_flags; | |
1683 | ||
1684 | memcg->move_lock_task = NULL; | |
1685 | memcg->move_lock_flags = 0; | |
1686 | ||
1687 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1688 | } | |
89c06bd5 | 1689 | |
d7365e78 | 1690 | rcu_read_unlock(); |
89c06bd5 | 1691 | } |
81f8c3a4 | 1692 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1693 | |
cdec2e42 KH |
1694 | /* |
1695 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1696 | * TODO: maybe necessary to use big numbers in big irons. | |
1697 | */ | |
7ec99d62 | 1698 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1699 | struct memcg_stock_pcp { |
1700 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1701 | unsigned int nr_pages; |
cdec2e42 | 1702 | struct work_struct work; |
26fe6168 | 1703 | unsigned long flags; |
a0db00fc | 1704 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1705 | }; |
1706 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1707 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1708 | |
a0956d54 SS |
1709 | /** |
1710 | * consume_stock: Try to consume stocked charge on this cpu. | |
1711 | * @memcg: memcg to consume from. | |
1712 | * @nr_pages: how many pages to charge. | |
1713 | * | |
1714 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1715 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1716 | * service an allocation will refill the stock. | |
1717 | * | |
1718 | * returns true if successful, false otherwise. | |
cdec2e42 | 1719 | */ |
a0956d54 | 1720 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1721 | { |
1722 | struct memcg_stock_pcp *stock; | |
db2ba40c | 1723 | unsigned long flags; |
3e32cb2e | 1724 | bool ret = false; |
cdec2e42 | 1725 | |
a0956d54 | 1726 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1727 | return ret; |
a0956d54 | 1728 | |
db2ba40c JW |
1729 | local_irq_save(flags); |
1730 | ||
1731 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 1732 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1733 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1734 | ret = true; |
1735 | } | |
db2ba40c JW |
1736 | |
1737 | local_irq_restore(flags); | |
1738 | ||
cdec2e42 KH |
1739 | return ret; |
1740 | } | |
1741 | ||
1742 | /* | |
3e32cb2e | 1743 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1744 | */ |
1745 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1746 | { | |
1747 | struct mem_cgroup *old = stock->cached; | |
1748 | ||
11c9ea4e | 1749 | if (stock->nr_pages) { |
3e32cb2e | 1750 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1751 | if (do_memsw_account()) |
3e32cb2e | 1752 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1753 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1754 | stock->nr_pages = 0; |
cdec2e42 KH |
1755 | } |
1756 | stock->cached = NULL; | |
cdec2e42 KH |
1757 | } |
1758 | ||
cdec2e42 KH |
1759 | static void drain_local_stock(struct work_struct *dummy) |
1760 | { | |
db2ba40c JW |
1761 | struct memcg_stock_pcp *stock; |
1762 | unsigned long flags; | |
1763 | ||
1764 | local_irq_save(flags); | |
1765 | ||
1766 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 1767 | drain_stock(stock); |
26fe6168 | 1768 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
1769 | |
1770 | local_irq_restore(flags); | |
cdec2e42 KH |
1771 | } |
1772 | ||
1773 | /* | |
3e32cb2e | 1774 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1775 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1776 | */ |
c0ff4b85 | 1777 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 1778 | { |
db2ba40c JW |
1779 | struct memcg_stock_pcp *stock; |
1780 | unsigned long flags; | |
1781 | ||
1782 | local_irq_save(flags); | |
cdec2e42 | 1783 | |
db2ba40c | 1784 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 1785 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1786 | drain_stock(stock); |
c0ff4b85 | 1787 | stock->cached = memcg; |
cdec2e42 | 1788 | } |
11c9ea4e | 1789 | stock->nr_pages += nr_pages; |
db2ba40c JW |
1790 | |
1791 | local_irq_restore(flags); | |
cdec2e42 KH |
1792 | } |
1793 | ||
1794 | /* | |
c0ff4b85 | 1795 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1796 | * of the hierarchy under it. |
cdec2e42 | 1797 | */ |
6d3d6aa2 | 1798 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1799 | { |
26fe6168 | 1800 | int cpu, curcpu; |
d38144b7 | 1801 | |
6d3d6aa2 JW |
1802 | /* If someone's already draining, avoid adding running more workers. */ |
1803 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1804 | return; | |
cdec2e42 | 1805 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1806 | get_online_cpus(); |
5af12d0e | 1807 | curcpu = get_cpu(); |
cdec2e42 KH |
1808 | for_each_online_cpu(cpu) { |
1809 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1810 | struct mem_cgroup *memcg; |
26fe6168 | 1811 | |
c0ff4b85 R |
1812 | memcg = stock->cached; |
1813 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1814 | continue; |
2314b42d | 1815 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1816 | continue; |
d1a05b69 MH |
1817 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1818 | if (cpu == curcpu) | |
1819 | drain_local_stock(&stock->work); | |
1820 | else | |
1821 | schedule_work_on(cpu, &stock->work); | |
1822 | } | |
cdec2e42 | 1823 | } |
5af12d0e | 1824 | put_cpu(); |
f894ffa8 | 1825 | put_online_cpus(); |
9f50fad6 | 1826 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1827 | } |
1828 | ||
308167fc | 1829 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 1830 | { |
cdec2e42 KH |
1831 | struct memcg_stock_pcp *stock; |
1832 | ||
cdec2e42 KH |
1833 | stock = &per_cpu(memcg_stock, cpu); |
1834 | drain_stock(stock); | |
308167fc | 1835 | return 0; |
cdec2e42 KH |
1836 | } |
1837 | ||
f7e1cb6e JW |
1838 | static void reclaim_high(struct mem_cgroup *memcg, |
1839 | unsigned int nr_pages, | |
1840 | gfp_t gfp_mask) | |
1841 | { | |
1842 | do { | |
1843 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1844 | continue; | |
1845 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
1846 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
1847 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1848 | } | |
1849 | ||
1850 | static void high_work_func(struct work_struct *work) | |
1851 | { | |
1852 | struct mem_cgroup *memcg; | |
1853 | ||
1854 | memcg = container_of(work, struct mem_cgroup, high_work); | |
1855 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); | |
1856 | } | |
1857 | ||
b23afb93 TH |
1858 | /* |
1859 | * Scheduled by try_charge() to be executed from the userland return path | |
1860 | * and reclaims memory over the high limit. | |
1861 | */ | |
1862 | void mem_cgroup_handle_over_high(void) | |
1863 | { | |
1864 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1865 | struct mem_cgroup *memcg; |
b23afb93 TH |
1866 | |
1867 | if (likely(!nr_pages)) | |
1868 | return; | |
1869 | ||
f7e1cb6e JW |
1870 | memcg = get_mem_cgroup_from_mm(current->mm); |
1871 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1872 | css_put(&memcg->css); |
1873 | current->memcg_nr_pages_over_high = 0; | |
1874 | } | |
1875 | ||
00501b53 JW |
1876 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1877 | unsigned int nr_pages) | |
8a9f3ccd | 1878 | { |
7ec99d62 | 1879 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1880 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1881 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1882 | struct page_counter *counter; |
6539cc05 | 1883 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1884 | bool may_swap = true; |
1885 | bool drained = false; | |
a636b327 | 1886 | |
ce00a967 | 1887 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1888 | return 0; |
6539cc05 | 1889 | retry: |
b6b6cc72 | 1890 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1891 | return 0; |
8a9f3ccd | 1892 | |
7941d214 | 1893 | if (!do_memsw_account() || |
6071ca52 JW |
1894 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1895 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1896 | goto done_restock; |
7941d214 | 1897 | if (do_memsw_account()) |
3e32cb2e JW |
1898 | page_counter_uncharge(&memcg->memsw, batch); |
1899 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1900 | } else { |
3e32cb2e | 1901 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1902 | may_swap = false; |
3fbe7244 | 1903 | } |
7a81b88c | 1904 | |
6539cc05 JW |
1905 | if (batch > nr_pages) { |
1906 | batch = nr_pages; | |
1907 | goto retry; | |
1908 | } | |
6d61ef40 | 1909 | |
06b078fc JW |
1910 | /* |
1911 | * Unlike in global OOM situations, memcg is not in a physical | |
1912 | * memory shortage. Allow dying and OOM-killed tasks to | |
1913 | * bypass the last charges so that they can exit quickly and | |
1914 | * free their memory. | |
1915 | */ | |
1916 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
1917 | fatal_signal_pending(current) || | |
1918 | current->flags & PF_EXITING)) | |
10d53c74 | 1919 | goto force; |
06b078fc | 1920 | |
89a28483 JW |
1921 | /* |
1922 | * Prevent unbounded recursion when reclaim operations need to | |
1923 | * allocate memory. This might exceed the limits temporarily, | |
1924 | * but we prefer facilitating memory reclaim and getting back | |
1925 | * under the limit over triggering OOM kills in these cases. | |
1926 | */ | |
1927 | if (unlikely(current->flags & PF_MEMALLOC)) | |
1928 | goto force; | |
1929 | ||
06b078fc JW |
1930 | if (unlikely(task_in_memcg_oom(current))) |
1931 | goto nomem; | |
1932 | ||
d0164adc | 1933 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 1934 | goto nomem; |
4b534334 | 1935 | |
241994ed JW |
1936 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
1937 | ||
b70a2a21 JW |
1938 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
1939 | gfp_mask, may_swap); | |
6539cc05 | 1940 | |
61e02c74 | 1941 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 1942 | goto retry; |
28c34c29 | 1943 | |
b70a2a21 | 1944 | if (!drained) { |
6d3d6aa2 | 1945 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
1946 | drained = true; |
1947 | goto retry; | |
1948 | } | |
1949 | ||
28c34c29 JW |
1950 | if (gfp_mask & __GFP_NORETRY) |
1951 | goto nomem; | |
6539cc05 JW |
1952 | /* |
1953 | * Even though the limit is exceeded at this point, reclaim | |
1954 | * may have been able to free some pages. Retry the charge | |
1955 | * before killing the task. | |
1956 | * | |
1957 | * Only for regular pages, though: huge pages are rather | |
1958 | * unlikely to succeed so close to the limit, and we fall back | |
1959 | * to regular pages anyway in case of failure. | |
1960 | */ | |
61e02c74 | 1961 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
1962 | goto retry; |
1963 | /* | |
1964 | * At task move, charge accounts can be doubly counted. So, it's | |
1965 | * better to wait until the end of task_move if something is going on. | |
1966 | */ | |
1967 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
1968 | goto retry; | |
1969 | ||
9b130619 JW |
1970 | if (nr_retries--) |
1971 | goto retry; | |
1972 | ||
06b078fc | 1973 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 1974 | goto force; |
06b078fc | 1975 | |
6539cc05 | 1976 | if (fatal_signal_pending(current)) |
10d53c74 | 1977 | goto force; |
6539cc05 | 1978 | |
241994ed JW |
1979 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
1980 | ||
3608de07 JM |
1981 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
1982 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 1983 | nomem: |
6d1fdc48 | 1984 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 1985 | return -ENOMEM; |
10d53c74 TH |
1986 | force: |
1987 | /* | |
1988 | * The allocation either can't fail or will lead to more memory | |
1989 | * being freed very soon. Allow memory usage go over the limit | |
1990 | * temporarily by force charging it. | |
1991 | */ | |
1992 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 1993 | if (do_memsw_account()) |
10d53c74 TH |
1994 | page_counter_charge(&memcg->memsw, nr_pages); |
1995 | css_get_many(&memcg->css, nr_pages); | |
1996 | ||
1997 | return 0; | |
6539cc05 JW |
1998 | |
1999 | done_restock: | |
e8ea14cc | 2000 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2001 | if (batch > nr_pages) |
2002 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2003 | |
241994ed | 2004 | /* |
b23afb93 TH |
2005 | * If the hierarchy is above the normal consumption range, schedule |
2006 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2007 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2008 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2009 | * not recorded as it most likely matches current's and won't | |
2010 | * change in the meantime. As high limit is checked again before | |
2011 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2012 | */ |
2013 | do { | |
b23afb93 | 2014 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2015 | /* Don't bother a random interrupted task */ |
2016 | if (in_interrupt()) { | |
2017 | schedule_work(&memcg->high_work); | |
2018 | break; | |
2019 | } | |
9516a18a | 2020 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2021 | set_notify_resume(current); |
2022 | break; | |
2023 | } | |
241994ed | 2024 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2025 | |
2026 | return 0; | |
7a81b88c | 2027 | } |
8a9f3ccd | 2028 | |
00501b53 | 2029 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2030 | { |
ce00a967 JW |
2031 | if (mem_cgroup_is_root(memcg)) |
2032 | return; | |
2033 | ||
3e32cb2e | 2034 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2035 | if (do_memsw_account()) |
3e32cb2e | 2036 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2037 | |
e8ea14cc | 2038 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2039 | } |
2040 | ||
0a31bc97 JW |
2041 | static void lock_page_lru(struct page *page, int *isolated) |
2042 | { | |
2043 | struct zone *zone = page_zone(page); | |
2044 | ||
a52633d8 | 2045 | spin_lock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2046 | if (PageLRU(page)) { |
2047 | struct lruvec *lruvec; | |
2048 | ||
599d0c95 | 2049 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2050 | ClearPageLRU(page); |
2051 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2052 | *isolated = 1; | |
2053 | } else | |
2054 | *isolated = 0; | |
2055 | } | |
2056 | ||
2057 | static void unlock_page_lru(struct page *page, int isolated) | |
2058 | { | |
2059 | struct zone *zone = page_zone(page); | |
2060 | ||
2061 | if (isolated) { | |
2062 | struct lruvec *lruvec; | |
2063 | ||
599d0c95 | 2064 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2065 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2066 | SetPageLRU(page); | |
2067 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2068 | } | |
a52633d8 | 2069 | spin_unlock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2070 | } |
2071 | ||
00501b53 | 2072 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2073 | bool lrucare) |
7a81b88c | 2074 | { |
0a31bc97 | 2075 | int isolated; |
9ce70c02 | 2076 | |
1306a85a | 2077 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2078 | |
2079 | /* | |
2080 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2081 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2082 | */ | |
0a31bc97 JW |
2083 | if (lrucare) |
2084 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2085 | |
0a31bc97 JW |
2086 | /* |
2087 | * Nobody should be changing or seriously looking at | |
1306a85a | 2088 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2089 | * |
2090 | * - the page is uncharged | |
2091 | * | |
2092 | * - the page is off-LRU | |
2093 | * | |
2094 | * - an anonymous fault has exclusive page access, except for | |
2095 | * a locked page table | |
2096 | * | |
2097 | * - a page cache insertion, a swapin fault, or a migration | |
2098 | * have the page locked | |
2099 | */ | |
1306a85a | 2100 | page->mem_cgroup = memcg; |
9ce70c02 | 2101 | |
0a31bc97 JW |
2102 | if (lrucare) |
2103 | unlock_page_lru(page, isolated); | |
7a81b88c | 2104 | } |
66e1707b | 2105 | |
127424c8 | 2106 | #ifndef CONFIG_SLOB |
f3bb3043 | 2107 | static int memcg_alloc_cache_id(void) |
55007d84 | 2108 | { |
f3bb3043 VD |
2109 | int id, size; |
2110 | int err; | |
2111 | ||
dbcf73e2 | 2112 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2113 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2114 | if (id < 0) | |
2115 | return id; | |
55007d84 | 2116 | |
dbcf73e2 | 2117 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2118 | return id; |
2119 | ||
2120 | /* | |
2121 | * There's no space for the new id in memcg_caches arrays, | |
2122 | * so we have to grow them. | |
2123 | */ | |
05257a1a | 2124 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2125 | |
2126 | size = 2 * (id + 1); | |
55007d84 GC |
2127 | if (size < MEMCG_CACHES_MIN_SIZE) |
2128 | size = MEMCG_CACHES_MIN_SIZE; | |
2129 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2130 | size = MEMCG_CACHES_MAX_SIZE; | |
2131 | ||
f3bb3043 | 2132 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2133 | if (!err) |
2134 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2135 | if (!err) |
2136 | memcg_nr_cache_ids = size; | |
2137 | ||
2138 | up_write(&memcg_cache_ids_sem); | |
2139 | ||
f3bb3043 | 2140 | if (err) { |
dbcf73e2 | 2141 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2142 | return err; |
2143 | } | |
2144 | return id; | |
2145 | } | |
2146 | ||
2147 | static void memcg_free_cache_id(int id) | |
2148 | { | |
dbcf73e2 | 2149 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2150 | } |
2151 | ||
d5b3cf71 | 2152 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2153 | struct mem_cgroup *memcg; |
2154 | struct kmem_cache *cachep; | |
2155 | struct work_struct work; | |
2156 | }; | |
2157 | ||
d5b3cf71 | 2158 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2159 | { |
d5b3cf71 VD |
2160 | struct memcg_kmem_cache_create_work *cw = |
2161 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2162 | struct mem_cgroup *memcg = cw->memcg; |
2163 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2164 | |
d5b3cf71 | 2165 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2166 | |
5722d094 | 2167 | css_put(&memcg->css); |
d7f25f8a GC |
2168 | kfree(cw); |
2169 | } | |
2170 | ||
2171 | /* | |
2172 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2173 | */ |
d5b3cf71 VD |
2174 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2175 | struct kmem_cache *cachep) | |
d7f25f8a | 2176 | { |
d5b3cf71 | 2177 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2178 | |
776ed0f0 | 2179 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2180 | if (!cw) |
d7f25f8a | 2181 | return; |
8135be5a VD |
2182 | |
2183 | css_get(&memcg->css); | |
d7f25f8a GC |
2184 | |
2185 | cw->memcg = memcg; | |
2186 | cw->cachep = cachep; | |
d5b3cf71 | 2187 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2188 | |
17cc4dfe | 2189 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2190 | } |
2191 | ||
d5b3cf71 VD |
2192 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2193 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2194 | { |
2195 | /* | |
2196 | * We need to stop accounting when we kmalloc, because if the | |
2197 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2198 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2199 | * |
2200 | * However, it is better to enclose the whole function. Depending on | |
2201 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2202 | * trigger an allocation. This too, will make us recurse. Because at | |
2203 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2204 | * the safest choice is to do it like this, wrapping the whole function. | |
2205 | */ | |
6f185c29 | 2206 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2207 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2208 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2209 | } |
c67a8a68 | 2210 | |
45264778 VD |
2211 | static inline bool memcg_kmem_bypass(void) |
2212 | { | |
2213 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2214 | return true; | |
2215 | return false; | |
2216 | } | |
2217 | ||
2218 | /** | |
2219 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2220 | * @cachep: the original global kmem cache | |
2221 | * | |
d7f25f8a GC |
2222 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2223 | * We try to use the current memcg's version of the cache. | |
2224 | * | |
45264778 VD |
2225 | * If the cache does not exist yet, if we are the first user of it, we |
2226 | * create it asynchronously in a workqueue and let the current allocation | |
2227 | * go through with the original cache. | |
d7f25f8a | 2228 | * |
45264778 VD |
2229 | * This function takes a reference to the cache it returns to assure it |
2230 | * won't get destroyed while we are working with it. Once the caller is | |
2231 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2232 | * reference. | |
d7f25f8a | 2233 | */ |
45264778 | 2234 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2235 | { |
2236 | struct mem_cgroup *memcg; | |
959c8963 | 2237 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2238 | int kmemcg_id; |
d7f25f8a | 2239 | |
f7ce3190 | 2240 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2241 | |
45264778 | 2242 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2243 | return cachep; |
2244 | ||
9d100c5e | 2245 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2246 | return cachep; |
2247 | ||
8135be5a | 2248 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2249 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2250 | if (kmemcg_id < 0) |
ca0dde97 | 2251 | goto out; |
d7f25f8a | 2252 | |
2a4db7eb | 2253 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2254 | if (likely(memcg_cachep)) |
2255 | return memcg_cachep; | |
ca0dde97 LZ |
2256 | |
2257 | /* | |
2258 | * If we are in a safe context (can wait, and not in interrupt | |
2259 | * context), we could be be predictable and return right away. | |
2260 | * This would guarantee that the allocation being performed | |
2261 | * already belongs in the new cache. | |
2262 | * | |
2263 | * However, there are some clashes that can arrive from locking. | |
2264 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2265 | * memcg_create_kmem_cache, this means no further allocation |
2266 | * could happen with the slab_mutex held. So it's better to | |
2267 | * defer everything. | |
ca0dde97 | 2268 | */ |
d5b3cf71 | 2269 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2270 | out: |
8135be5a | 2271 | css_put(&memcg->css); |
ca0dde97 | 2272 | return cachep; |
d7f25f8a | 2273 | } |
d7f25f8a | 2274 | |
45264778 VD |
2275 | /** |
2276 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2277 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2278 | */ | |
2279 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2280 | { |
2281 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2282 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2283 | } |
2284 | ||
45264778 VD |
2285 | /** |
2286 | * memcg_kmem_charge: charge a kmem page | |
2287 | * @page: page to charge | |
2288 | * @gfp: reclaim mode | |
2289 | * @order: allocation order | |
2290 | * @memcg: memory cgroup to charge | |
2291 | * | |
2292 | * Returns 0 on success, an error code on failure. | |
2293 | */ | |
2294 | int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, | |
2295 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2296 | { |
f3ccb2c4 VD |
2297 | unsigned int nr_pages = 1 << order; |
2298 | struct page_counter *counter; | |
7ae1e1d0 GC |
2299 | int ret; |
2300 | ||
f3ccb2c4 | 2301 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2302 | if (ret) |
f3ccb2c4 | 2303 | return ret; |
52c29b04 JW |
2304 | |
2305 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2306 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2307 | cancel_charge(memcg, nr_pages); | |
2308 | return -ENOMEM; | |
7ae1e1d0 GC |
2309 | } |
2310 | ||
f3ccb2c4 | 2311 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2312 | |
f3ccb2c4 | 2313 | return 0; |
7ae1e1d0 GC |
2314 | } |
2315 | ||
45264778 VD |
2316 | /** |
2317 | * memcg_kmem_charge: charge a kmem page to the current memory cgroup | |
2318 | * @page: page to charge | |
2319 | * @gfp: reclaim mode | |
2320 | * @order: allocation order | |
2321 | * | |
2322 | * Returns 0 on success, an error code on failure. | |
2323 | */ | |
2324 | int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) | |
7ae1e1d0 | 2325 | { |
f3ccb2c4 | 2326 | struct mem_cgroup *memcg; |
fcff7d7e | 2327 | int ret = 0; |
7ae1e1d0 | 2328 | |
45264778 VD |
2329 | if (memcg_kmem_bypass()) |
2330 | return 0; | |
2331 | ||
f3ccb2c4 | 2332 | memcg = get_mem_cgroup_from_mm(current->mm); |
c4159a75 | 2333 | if (!mem_cgroup_is_root(memcg)) { |
45264778 | 2334 | ret = memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2335 | if (!ret) |
2336 | __SetPageKmemcg(page); | |
2337 | } | |
7ae1e1d0 | 2338 | css_put(&memcg->css); |
d05e83a6 | 2339 | return ret; |
7ae1e1d0 | 2340 | } |
45264778 VD |
2341 | /** |
2342 | * memcg_kmem_uncharge: uncharge a kmem page | |
2343 | * @page: page to uncharge | |
2344 | * @order: allocation order | |
2345 | */ | |
2346 | void memcg_kmem_uncharge(struct page *page, int order) | |
7ae1e1d0 | 2347 | { |
1306a85a | 2348 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2349 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2350 | |
7ae1e1d0 GC |
2351 | if (!memcg) |
2352 | return; | |
2353 | ||
309381fe | 2354 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2355 | |
52c29b04 JW |
2356 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2357 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2358 | ||
f3ccb2c4 | 2359 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2360 | if (do_memsw_account()) |
f3ccb2c4 | 2361 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2362 | |
1306a85a | 2363 | page->mem_cgroup = NULL; |
c4159a75 VD |
2364 | |
2365 | /* slab pages do not have PageKmemcg flag set */ | |
2366 | if (PageKmemcg(page)) | |
2367 | __ClearPageKmemcg(page); | |
2368 | ||
f3ccb2c4 | 2369 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2370 | } |
127424c8 | 2371 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2372 | |
ca3e0214 KH |
2373 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2374 | ||
ca3e0214 KH |
2375 | /* |
2376 | * Because tail pages are not marked as "used", set it. We're under | |
a52633d8 | 2377 | * zone_lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2378 | */ |
e94c8a9c | 2379 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2380 | { |
e94c8a9c | 2381 | int i; |
ca3e0214 | 2382 | |
3d37c4a9 KH |
2383 | if (mem_cgroup_disabled()) |
2384 | return; | |
b070e65c | 2385 | |
29833315 | 2386 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2387 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2388 | |
1306a85a | 2389 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2390 | HPAGE_PMD_NR); |
ca3e0214 | 2391 | } |
12d27107 | 2392 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2393 | |
c255a458 | 2394 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2395 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2396 | bool charge) | |
d13d1443 | 2397 | { |
0a31bc97 JW |
2398 | int val = (charge) ? 1 : -1; |
2399 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2400 | } |
02491447 DN |
2401 | |
2402 | /** | |
2403 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2404 | * @entry: swap entry to be moved | |
2405 | * @from: mem_cgroup which the entry is moved from | |
2406 | * @to: mem_cgroup which the entry is moved to | |
2407 | * | |
2408 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2409 | * as the mem_cgroup's id of @from. | |
2410 | * | |
2411 | * Returns 0 on success, -EINVAL on failure. | |
2412 | * | |
3e32cb2e | 2413 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2414 | * both res and memsw, and called css_get(). |
2415 | */ | |
2416 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2417 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2418 | { |
2419 | unsigned short old_id, new_id; | |
2420 | ||
34c00c31 LZ |
2421 | old_id = mem_cgroup_id(from); |
2422 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2423 | |
2424 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2425 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2426 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2427 | return 0; |
2428 | } | |
2429 | return -EINVAL; | |
2430 | } | |
2431 | #else | |
2432 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2433 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2434 | { |
2435 | return -EINVAL; | |
2436 | } | |
8c7c6e34 | 2437 | #endif |
d13d1443 | 2438 | |
3e32cb2e | 2439 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2440 | |
d38d2a75 | 2441 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2442 | unsigned long limit) |
628f4235 | 2443 | { |
3e32cb2e JW |
2444 | unsigned long curusage; |
2445 | unsigned long oldusage; | |
2446 | bool enlarge = false; | |
81d39c20 | 2447 | int retry_count; |
3e32cb2e | 2448 | int ret; |
81d39c20 KH |
2449 | |
2450 | /* | |
2451 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2452 | * is depends on callers. We set our retry-count to be function | |
2453 | * of # of children which we should visit in this loop. | |
2454 | */ | |
3e32cb2e JW |
2455 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2456 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2457 | |
3e32cb2e | 2458 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2459 | |
3e32cb2e | 2460 | do { |
628f4235 KH |
2461 | if (signal_pending(current)) { |
2462 | ret = -EINTR; | |
2463 | break; | |
2464 | } | |
3e32cb2e JW |
2465 | |
2466 | mutex_lock(&memcg_limit_mutex); | |
2467 | if (limit > memcg->memsw.limit) { | |
2468 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2469 | ret = -EINVAL; |
628f4235 KH |
2470 | break; |
2471 | } | |
3e32cb2e JW |
2472 | if (limit > memcg->memory.limit) |
2473 | enlarge = true; | |
2474 | ret = page_counter_limit(&memcg->memory, limit); | |
2475 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2476 | |
2477 | if (!ret) | |
2478 | break; | |
2479 | ||
b70a2a21 JW |
2480 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2481 | ||
3e32cb2e | 2482 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2483 | /* Usage is reduced ? */ |
f894ffa8 | 2484 | if (curusage >= oldusage) |
81d39c20 KH |
2485 | retry_count--; |
2486 | else | |
2487 | oldusage = curusage; | |
3e32cb2e JW |
2488 | } while (retry_count); |
2489 | ||
3c11ecf4 KH |
2490 | if (!ret && enlarge) |
2491 | memcg_oom_recover(memcg); | |
14797e23 | 2492 | |
8c7c6e34 KH |
2493 | return ret; |
2494 | } | |
2495 | ||
338c8431 | 2496 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2497 | unsigned long limit) |
8c7c6e34 | 2498 | { |
3e32cb2e JW |
2499 | unsigned long curusage; |
2500 | unsigned long oldusage; | |
2501 | bool enlarge = false; | |
81d39c20 | 2502 | int retry_count; |
3e32cb2e | 2503 | int ret; |
8c7c6e34 | 2504 | |
81d39c20 | 2505 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2506 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2507 | mem_cgroup_count_children(memcg); | |
2508 | ||
2509 | oldusage = page_counter_read(&memcg->memsw); | |
2510 | ||
2511 | do { | |
8c7c6e34 KH |
2512 | if (signal_pending(current)) { |
2513 | ret = -EINTR; | |
2514 | break; | |
2515 | } | |
3e32cb2e JW |
2516 | |
2517 | mutex_lock(&memcg_limit_mutex); | |
2518 | if (limit < memcg->memory.limit) { | |
2519 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2520 | ret = -EINVAL; |
8c7c6e34 KH |
2521 | break; |
2522 | } | |
3e32cb2e JW |
2523 | if (limit > memcg->memsw.limit) |
2524 | enlarge = true; | |
2525 | ret = page_counter_limit(&memcg->memsw, limit); | |
2526 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2527 | |
2528 | if (!ret) | |
2529 | break; | |
2530 | ||
b70a2a21 JW |
2531 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2532 | ||
3e32cb2e | 2533 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2534 | /* Usage is reduced ? */ |
8c7c6e34 | 2535 | if (curusage >= oldusage) |
628f4235 | 2536 | retry_count--; |
81d39c20 KH |
2537 | else |
2538 | oldusage = curusage; | |
3e32cb2e JW |
2539 | } while (retry_count); |
2540 | ||
3c11ecf4 KH |
2541 | if (!ret && enlarge) |
2542 | memcg_oom_recover(memcg); | |
3e32cb2e | 2543 | |
628f4235 KH |
2544 | return ret; |
2545 | } | |
2546 | ||
ef8f2327 | 2547 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2548 | gfp_t gfp_mask, |
2549 | unsigned long *total_scanned) | |
2550 | { | |
2551 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2552 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2553 | unsigned long reclaimed; |
2554 | int loop = 0; | |
ef8f2327 | 2555 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2556 | unsigned long excess; |
0608f43d AM |
2557 | unsigned long nr_scanned; |
2558 | ||
2559 | if (order > 0) | |
2560 | return 0; | |
2561 | ||
ef8f2327 | 2562 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2563 | |
2564 | /* | |
2565 | * Do not even bother to check the largest node if the root | |
2566 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2567 | * are acceptable as soft limit is best effort anyway. | |
2568 | */ | |
bfc7228b | 2569 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2570 | return 0; |
2571 | ||
0608f43d AM |
2572 | /* |
2573 | * This loop can run a while, specially if mem_cgroup's continuously | |
2574 | * keep exceeding their soft limit and putting the system under | |
2575 | * pressure | |
2576 | */ | |
2577 | do { | |
2578 | if (next_mz) | |
2579 | mz = next_mz; | |
2580 | else | |
2581 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2582 | if (!mz) | |
2583 | break; | |
2584 | ||
2585 | nr_scanned = 0; | |
ef8f2327 | 2586 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2587 | gfp_mask, &nr_scanned); |
2588 | nr_reclaimed += reclaimed; | |
2589 | *total_scanned += nr_scanned; | |
0a31bc97 | 2590 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2591 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2592 | |
2593 | /* | |
2594 | * If we failed to reclaim anything from this memory cgroup | |
2595 | * it is time to move on to the next cgroup | |
2596 | */ | |
2597 | next_mz = NULL; | |
bc2f2e7f VD |
2598 | if (!reclaimed) |
2599 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2600 | ||
3e32cb2e | 2601 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2602 | /* |
2603 | * One school of thought says that we should not add | |
2604 | * back the node to the tree if reclaim returns 0. | |
2605 | * But our reclaim could return 0, simply because due | |
2606 | * to priority we are exposing a smaller subset of | |
2607 | * memory to reclaim from. Consider this as a longer | |
2608 | * term TODO. | |
2609 | */ | |
2610 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2611 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2612 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2613 | css_put(&mz->memcg->css); |
2614 | loop++; | |
2615 | /* | |
2616 | * Could not reclaim anything and there are no more | |
2617 | * mem cgroups to try or we seem to be looping without | |
2618 | * reclaiming anything. | |
2619 | */ | |
2620 | if (!nr_reclaimed && | |
2621 | (next_mz == NULL || | |
2622 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2623 | break; | |
2624 | } while (!nr_reclaimed); | |
2625 | if (next_mz) | |
2626 | css_put(&next_mz->memcg->css); | |
2627 | return nr_reclaimed; | |
2628 | } | |
2629 | ||
ea280e7b TH |
2630 | /* |
2631 | * Test whether @memcg has children, dead or alive. Note that this | |
2632 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2633 | * returns %true if there are child csses according to the cgroup | |
2634 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2635 | */ | |
b5f99b53 GC |
2636 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2637 | { | |
ea280e7b TH |
2638 | bool ret; |
2639 | ||
ea280e7b TH |
2640 | rcu_read_lock(); |
2641 | ret = css_next_child(NULL, &memcg->css); | |
2642 | rcu_read_unlock(); | |
2643 | return ret; | |
b5f99b53 GC |
2644 | } |
2645 | ||
c26251f9 | 2646 | /* |
51038171 | 2647 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2648 | * |
2649 | * Caller is responsible for holding css reference for memcg. | |
2650 | */ | |
2651 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2652 | { | |
2653 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2654 | |
c1e862c1 KH |
2655 | /* we call try-to-free pages for make this cgroup empty */ |
2656 | lru_add_drain_all(); | |
f817ed48 | 2657 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2658 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2659 | int progress; |
c1e862c1 | 2660 | |
c26251f9 MH |
2661 | if (signal_pending(current)) |
2662 | return -EINTR; | |
2663 | ||
b70a2a21 JW |
2664 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2665 | GFP_KERNEL, true); | |
c1e862c1 | 2666 | if (!progress) { |
f817ed48 | 2667 | nr_retries--; |
c1e862c1 | 2668 | /* maybe some writeback is necessary */ |
8aa7e847 | 2669 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2670 | } |
f817ed48 KH |
2671 | |
2672 | } | |
ab5196c2 MH |
2673 | |
2674 | return 0; | |
cc847582 KH |
2675 | } |
2676 | ||
6770c64e TH |
2677 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2678 | char *buf, size_t nbytes, | |
2679 | loff_t off) | |
c1e862c1 | 2680 | { |
6770c64e | 2681 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2682 | |
d8423011 MH |
2683 | if (mem_cgroup_is_root(memcg)) |
2684 | return -EINVAL; | |
6770c64e | 2685 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2686 | } |
2687 | ||
182446d0 TH |
2688 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2689 | struct cftype *cft) | |
18f59ea7 | 2690 | { |
182446d0 | 2691 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2692 | } |
2693 | ||
182446d0 TH |
2694 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2695 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2696 | { |
2697 | int retval = 0; | |
182446d0 | 2698 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2699 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2700 | |
567fb435 | 2701 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2702 | return 0; |
567fb435 | 2703 | |
18f59ea7 | 2704 | /* |
af901ca1 | 2705 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2706 | * in the child subtrees. If it is unset, then the change can |
2707 | * occur, provided the current cgroup has no children. | |
2708 | * | |
2709 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2710 | * set if there are no children. | |
2711 | */ | |
c0ff4b85 | 2712 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2713 | (val == 1 || val == 0)) { |
ea280e7b | 2714 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2715 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2716 | else |
2717 | retval = -EBUSY; | |
2718 | } else | |
2719 | retval = -EINVAL; | |
567fb435 | 2720 | |
18f59ea7 BS |
2721 | return retval; |
2722 | } | |
2723 | ||
72b54e73 | 2724 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
ce00a967 JW |
2725 | { |
2726 | struct mem_cgroup *iter; | |
72b54e73 | 2727 | int i; |
ce00a967 | 2728 | |
72b54e73 | 2729 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
ce00a967 | 2730 | |
72b54e73 VD |
2731 | for_each_mem_cgroup_tree(iter, memcg) { |
2732 | for (i = 0; i < MEMCG_NR_STAT; i++) | |
2733 | stat[i] += mem_cgroup_read_stat(iter, i); | |
2734 | } | |
ce00a967 JW |
2735 | } |
2736 | ||
72b54e73 | 2737 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
587d9f72 JW |
2738 | { |
2739 | struct mem_cgroup *iter; | |
72b54e73 | 2740 | int i; |
587d9f72 | 2741 | |
72b54e73 | 2742 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
587d9f72 | 2743 | |
72b54e73 VD |
2744 | for_each_mem_cgroup_tree(iter, memcg) { |
2745 | for (i = 0; i < MEMCG_NR_EVENTS; i++) | |
2746 | events[i] += mem_cgroup_read_events(iter, i); | |
2747 | } | |
587d9f72 JW |
2748 | } |
2749 | ||
6f646156 | 2750 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2751 | { |
72b54e73 | 2752 | unsigned long val = 0; |
ce00a967 | 2753 | |
3e32cb2e | 2754 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2755 | struct mem_cgroup *iter; |
2756 | ||
2757 | for_each_mem_cgroup_tree(iter, memcg) { | |
2758 | val += mem_cgroup_read_stat(iter, | |
2759 | MEM_CGROUP_STAT_CACHE); | |
2760 | val += mem_cgroup_read_stat(iter, | |
2761 | MEM_CGROUP_STAT_RSS); | |
2762 | if (swap) | |
2763 | val += mem_cgroup_read_stat(iter, | |
2764 | MEM_CGROUP_STAT_SWAP); | |
2765 | } | |
3e32cb2e | 2766 | } else { |
ce00a967 | 2767 | if (!swap) |
3e32cb2e | 2768 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2769 | else |
3e32cb2e | 2770 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2771 | } |
c12176d3 | 2772 | return val; |
ce00a967 JW |
2773 | } |
2774 | ||
3e32cb2e JW |
2775 | enum { |
2776 | RES_USAGE, | |
2777 | RES_LIMIT, | |
2778 | RES_MAX_USAGE, | |
2779 | RES_FAILCNT, | |
2780 | RES_SOFT_LIMIT, | |
2781 | }; | |
ce00a967 | 2782 | |
791badbd | 2783 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2784 | struct cftype *cft) |
8cdea7c0 | 2785 | { |
182446d0 | 2786 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2787 | struct page_counter *counter; |
af36f906 | 2788 | |
3e32cb2e | 2789 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2790 | case _MEM: |
3e32cb2e JW |
2791 | counter = &memcg->memory; |
2792 | break; | |
8c7c6e34 | 2793 | case _MEMSWAP: |
3e32cb2e JW |
2794 | counter = &memcg->memsw; |
2795 | break; | |
510fc4e1 | 2796 | case _KMEM: |
3e32cb2e | 2797 | counter = &memcg->kmem; |
510fc4e1 | 2798 | break; |
d55f90bf | 2799 | case _TCP: |
0db15298 | 2800 | counter = &memcg->tcpmem; |
d55f90bf | 2801 | break; |
8c7c6e34 KH |
2802 | default: |
2803 | BUG(); | |
8c7c6e34 | 2804 | } |
3e32cb2e JW |
2805 | |
2806 | switch (MEMFILE_ATTR(cft->private)) { | |
2807 | case RES_USAGE: | |
2808 | if (counter == &memcg->memory) | |
c12176d3 | 2809 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2810 | if (counter == &memcg->memsw) |
c12176d3 | 2811 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2812 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2813 | case RES_LIMIT: | |
2814 | return (u64)counter->limit * PAGE_SIZE; | |
2815 | case RES_MAX_USAGE: | |
2816 | return (u64)counter->watermark * PAGE_SIZE; | |
2817 | case RES_FAILCNT: | |
2818 | return counter->failcnt; | |
2819 | case RES_SOFT_LIMIT: | |
2820 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2821 | default: | |
2822 | BUG(); | |
2823 | } | |
8cdea7c0 | 2824 | } |
510fc4e1 | 2825 | |
127424c8 | 2826 | #ifndef CONFIG_SLOB |
567e9ab2 | 2827 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2828 | { |
d6441637 VD |
2829 | int memcg_id; |
2830 | ||
b313aeee VD |
2831 | if (cgroup_memory_nokmem) |
2832 | return 0; | |
2833 | ||
2a4db7eb | 2834 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2835 | BUG_ON(memcg->kmem_state); |
d6441637 | 2836 | |
f3bb3043 | 2837 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
2838 | if (memcg_id < 0) |
2839 | return memcg_id; | |
d6441637 | 2840 | |
ef12947c | 2841 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2842 | /* |
567e9ab2 | 2843 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2844 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2845 | * guarantee no one starts accounting before all call sites are |
2846 | * patched. | |
2847 | */ | |
900a38f0 | 2848 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2849 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 2850 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
2851 | |
2852 | return 0; | |
d6441637 VD |
2853 | } |
2854 | ||
8e0a8912 JW |
2855 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2856 | { | |
2857 | struct cgroup_subsys_state *css; | |
2858 | struct mem_cgroup *parent, *child; | |
2859 | int kmemcg_id; | |
2860 | ||
2861 | if (memcg->kmem_state != KMEM_ONLINE) | |
2862 | return; | |
2863 | /* | |
2864 | * Clear the online state before clearing memcg_caches array | |
2865 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2866 | * guarantees that no cache will be created for this cgroup | |
2867 | * after we are done (see memcg_create_kmem_cache()). | |
2868 | */ | |
2869 | memcg->kmem_state = KMEM_ALLOCATED; | |
2870 | ||
2871 | memcg_deactivate_kmem_caches(memcg); | |
2872 | ||
2873 | kmemcg_id = memcg->kmemcg_id; | |
2874 | BUG_ON(kmemcg_id < 0); | |
2875 | ||
2876 | parent = parent_mem_cgroup(memcg); | |
2877 | if (!parent) | |
2878 | parent = root_mem_cgroup; | |
2879 | ||
2880 | /* | |
2881 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2882 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2883 | * to ones of the parent. After we have finished, all list_lrus | |
2884 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2885 | * ordering is imposed by list_lru_node->lock taken by | |
2886 | * memcg_drain_all_list_lrus(). | |
2887 | */ | |
3a06bb78 | 2888 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
2889 | css_for_each_descendant_pre(css, &memcg->css) { |
2890 | child = mem_cgroup_from_css(css); | |
2891 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2892 | child->kmemcg_id = parent->kmemcg_id; | |
2893 | if (!memcg->use_hierarchy) | |
2894 | break; | |
2895 | } | |
3a06bb78 TH |
2896 | rcu_read_unlock(); |
2897 | ||
8e0a8912 JW |
2898 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
2899 | ||
2900 | memcg_free_cache_id(kmemcg_id); | |
2901 | } | |
2902 | ||
2903 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2904 | { | |
0b8f73e1 JW |
2905 | /* css_alloc() failed, offlining didn't happen */ |
2906 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
2907 | memcg_offline_kmem(memcg); | |
2908 | ||
8e0a8912 JW |
2909 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
2910 | memcg_destroy_kmem_caches(memcg); | |
2911 | static_branch_dec(&memcg_kmem_enabled_key); | |
2912 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2913 | } | |
8e0a8912 | 2914 | } |
d6441637 | 2915 | #else |
0b8f73e1 | 2916 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
2917 | { |
2918 | return 0; | |
2919 | } | |
2920 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
2921 | { | |
2922 | } | |
2923 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2924 | { | |
2925 | } | |
2926 | #endif /* !CONFIG_SLOB */ | |
2927 | ||
d6441637 | 2928 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2929 | unsigned long limit) |
d6441637 | 2930 | { |
b313aeee | 2931 | int ret; |
127424c8 JW |
2932 | |
2933 | mutex_lock(&memcg_limit_mutex); | |
127424c8 | 2934 | ret = page_counter_limit(&memcg->kmem, limit); |
127424c8 JW |
2935 | mutex_unlock(&memcg_limit_mutex); |
2936 | return ret; | |
d6441637 | 2937 | } |
510fc4e1 | 2938 | |
d55f90bf VD |
2939 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
2940 | { | |
2941 | int ret; | |
2942 | ||
2943 | mutex_lock(&memcg_limit_mutex); | |
2944 | ||
0db15298 | 2945 | ret = page_counter_limit(&memcg->tcpmem, limit); |
d55f90bf VD |
2946 | if (ret) |
2947 | goto out; | |
2948 | ||
0db15298 | 2949 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
2950 | /* |
2951 | * The active flag needs to be written after the static_key | |
2952 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
2953 | * function is the last one to run. See mem_cgroup_sk_alloc() |
2954 | * for details, and note that we don't mark any socket as | |
2955 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
2956 | * |
2957 | * We need to do this, because static_keys will span multiple | |
2958 | * sites, but we can't control their order. If we mark a socket | |
2959 | * as accounted, but the accounting functions are not patched in | |
2960 | * yet, we'll lose accounting. | |
2961 | * | |
2d758073 | 2962 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
2963 | * because when this value change, the code to process it is not |
2964 | * patched in yet. | |
2965 | */ | |
2966 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 2967 | memcg->tcpmem_active = true; |
d55f90bf VD |
2968 | } |
2969 | out: | |
2970 | mutex_unlock(&memcg_limit_mutex); | |
2971 | return ret; | |
2972 | } | |
d55f90bf | 2973 | |
628f4235 KH |
2974 | /* |
2975 | * The user of this function is... | |
2976 | * RES_LIMIT. | |
2977 | */ | |
451af504 TH |
2978 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2979 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2980 | { |
451af504 | 2981 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2982 | unsigned long nr_pages; |
628f4235 KH |
2983 | int ret; |
2984 | ||
451af504 | 2985 | buf = strstrip(buf); |
650c5e56 | 2986 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
2987 | if (ret) |
2988 | return ret; | |
af36f906 | 2989 | |
3e32cb2e | 2990 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 2991 | case RES_LIMIT: |
4b3bde4c BS |
2992 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2993 | ret = -EINVAL; | |
2994 | break; | |
2995 | } | |
3e32cb2e JW |
2996 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
2997 | case _MEM: | |
2998 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 2999 | break; |
3e32cb2e JW |
3000 | case _MEMSWAP: |
3001 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3002 | break; |
3e32cb2e JW |
3003 | case _KMEM: |
3004 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3005 | break; | |
d55f90bf VD |
3006 | case _TCP: |
3007 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
3008 | break; | |
3e32cb2e | 3009 | } |
296c81d8 | 3010 | break; |
3e32cb2e JW |
3011 | case RES_SOFT_LIMIT: |
3012 | memcg->soft_limit = nr_pages; | |
3013 | ret = 0; | |
628f4235 KH |
3014 | break; |
3015 | } | |
451af504 | 3016 | return ret ?: nbytes; |
8cdea7c0 BS |
3017 | } |
3018 | ||
6770c64e TH |
3019 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3020 | size_t nbytes, loff_t off) | |
c84872e1 | 3021 | { |
6770c64e | 3022 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3023 | struct page_counter *counter; |
c84872e1 | 3024 | |
3e32cb2e JW |
3025 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3026 | case _MEM: | |
3027 | counter = &memcg->memory; | |
3028 | break; | |
3029 | case _MEMSWAP: | |
3030 | counter = &memcg->memsw; | |
3031 | break; | |
3032 | case _KMEM: | |
3033 | counter = &memcg->kmem; | |
3034 | break; | |
d55f90bf | 3035 | case _TCP: |
0db15298 | 3036 | counter = &memcg->tcpmem; |
d55f90bf | 3037 | break; |
3e32cb2e JW |
3038 | default: |
3039 | BUG(); | |
3040 | } | |
af36f906 | 3041 | |
3e32cb2e | 3042 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3043 | case RES_MAX_USAGE: |
3e32cb2e | 3044 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3045 | break; |
3046 | case RES_FAILCNT: | |
3e32cb2e | 3047 | counter->failcnt = 0; |
29f2a4da | 3048 | break; |
3e32cb2e JW |
3049 | default: |
3050 | BUG(); | |
29f2a4da | 3051 | } |
f64c3f54 | 3052 | |
6770c64e | 3053 | return nbytes; |
c84872e1 PE |
3054 | } |
3055 | ||
182446d0 | 3056 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3057 | struct cftype *cft) |
3058 | { | |
182446d0 | 3059 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3060 | } |
3061 | ||
02491447 | 3062 | #ifdef CONFIG_MMU |
182446d0 | 3063 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3064 | struct cftype *cft, u64 val) |
3065 | { | |
182446d0 | 3066 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3067 | |
1dfab5ab | 3068 | if (val & ~MOVE_MASK) |
7dc74be0 | 3069 | return -EINVAL; |
ee5e8472 | 3070 | |
7dc74be0 | 3071 | /* |
ee5e8472 GC |
3072 | * No kind of locking is needed in here, because ->can_attach() will |
3073 | * check this value once in the beginning of the process, and then carry | |
3074 | * on with stale data. This means that changes to this value will only | |
3075 | * affect task migrations starting after the change. | |
7dc74be0 | 3076 | */ |
c0ff4b85 | 3077 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3078 | return 0; |
3079 | } | |
02491447 | 3080 | #else |
182446d0 | 3081 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3082 | struct cftype *cft, u64 val) |
3083 | { | |
3084 | return -ENOSYS; | |
3085 | } | |
3086 | #endif | |
7dc74be0 | 3087 | |
406eb0c9 | 3088 | #ifdef CONFIG_NUMA |
2da8ca82 | 3089 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3090 | { |
25485de6 GT |
3091 | struct numa_stat { |
3092 | const char *name; | |
3093 | unsigned int lru_mask; | |
3094 | }; | |
3095 | ||
3096 | static const struct numa_stat stats[] = { | |
3097 | { "total", LRU_ALL }, | |
3098 | { "file", LRU_ALL_FILE }, | |
3099 | { "anon", LRU_ALL_ANON }, | |
3100 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3101 | }; | |
3102 | const struct numa_stat *stat; | |
406eb0c9 | 3103 | int nid; |
25485de6 | 3104 | unsigned long nr; |
2da8ca82 | 3105 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3106 | |
25485de6 GT |
3107 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3108 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3109 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3110 | for_each_node_state(nid, N_MEMORY) { | |
3111 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3112 | stat->lru_mask); | |
3113 | seq_printf(m, " N%d=%lu", nid, nr); | |
3114 | } | |
3115 | seq_putc(m, '\n'); | |
406eb0c9 | 3116 | } |
406eb0c9 | 3117 | |
071aee13 YH |
3118 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3119 | struct mem_cgroup *iter; | |
3120 | ||
3121 | nr = 0; | |
3122 | for_each_mem_cgroup_tree(iter, memcg) | |
3123 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3124 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3125 | for_each_node_state(nid, N_MEMORY) { | |
3126 | nr = 0; | |
3127 | for_each_mem_cgroup_tree(iter, memcg) | |
3128 | nr += mem_cgroup_node_nr_lru_pages( | |
3129 | iter, nid, stat->lru_mask); | |
3130 | seq_printf(m, " N%d=%lu", nid, nr); | |
3131 | } | |
3132 | seq_putc(m, '\n'); | |
406eb0c9 | 3133 | } |
406eb0c9 | 3134 | |
406eb0c9 YH |
3135 | return 0; |
3136 | } | |
3137 | #endif /* CONFIG_NUMA */ | |
3138 | ||
2da8ca82 | 3139 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3140 | { |
2da8ca82 | 3141 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3142 | unsigned long memory, memsw; |
af7c4b0e JW |
3143 | struct mem_cgroup *mi; |
3144 | unsigned int i; | |
406eb0c9 | 3145 | |
0ca44b14 GT |
3146 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3147 | MEM_CGROUP_STAT_NSTATS); | |
3148 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3149 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3150 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3151 | ||
af7c4b0e | 3152 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
7941d214 | 3153 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3154 | continue; |
484ebb3b | 3155 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3156 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3157 | } |
7b854121 | 3158 | |
af7c4b0e JW |
3159 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3160 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3161 | mem_cgroup_read_events(memcg, i)); | |
3162 | ||
3163 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3164 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3165 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3166 | ||
14067bb3 | 3167 | /* Hierarchical information */ |
3e32cb2e JW |
3168 | memory = memsw = PAGE_COUNTER_MAX; |
3169 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3170 | memory = min(memory, mi->memory.limit); | |
3171 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3172 | } |
3e32cb2e JW |
3173 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3174 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3175 | if (do_memsw_account()) |
3e32cb2e JW |
3176 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3177 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3178 | |
af7c4b0e | 3179 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3180 | unsigned long long val = 0; |
af7c4b0e | 3181 | |
7941d214 | 3182 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3183 | continue; |
af7c4b0e JW |
3184 | for_each_mem_cgroup_tree(mi, memcg) |
3185 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3186 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3187 | } |
3188 | ||
3189 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3190 | unsigned long long val = 0; | |
3191 | ||
3192 | for_each_mem_cgroup_tree(mi, memcg) | |
3193 | val += mem_cgroup_read_events(mi, i); | |
3194 | seq_printf(m, "total_%s %llu\n", | |
3195 | mem_cgroup_events_names[i], val); | |
3196 | } | |
3197 | ||
3198 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3199 | unsigned long long val = 0; | |
3200 | ||
3201 | for_each_mem_cgroup_tree(mi, memcg) | |
3202 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3203 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3204 | } |
14067bb3 | 3205 | |
7f016ee8 | 3206 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3207 | { |
ef8f2327 MG |
3208 | pg_data_t *pgdat; |
3209 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3210 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3211 | unsigned long recent_rotated[2] = {0, 0}; |
3212 | unsigned long recent_scanned[2] = {0, 0}; | |
3213 | ||
ef8f2327 MG |
3214 | for_each_online_pgdat(pgdat) { |
3215 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3216 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3217 | |
ef8f2327 MG |
3218 | recent_rotated[0] += rstat->recent_rotated[0]; |
3219 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3220 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3221 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3222 | } | |
78ccf5b5 JW |
3223 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3224 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3225 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3226 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3227 | } |
3228 | #endif | |
3229 | ||
d2ceb9b7 KH |
3230 | return 0; |
3231 | } | |
3232 | ||
182446d0 TH |
3233 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3234 | struct cftype *cft) | |
a7885eb8 | 3235 | { |
182446d0 | 3236 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3237 | |
1f4c025b | 3238 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3239 | } |
3240 | ||
182446d0 TH |
3241 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3242 | struct cftype *cft, u64 val) | |
a7885eb8 | 3243 | { |
182446d0 | 3244 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3245 | |
3dae7fec | 3246 | if (val > 100) |
a7885eb8 KM |
3247 | return -EINVAL; |
3248 | ||
14208b0e | 3249 | if (css->parent) |
3dae7fec JW |
3250 | memcg->swappiness = val; |
3251 | else | |
3252 | vm_swappiness = val; | |
068b38c1 | 3253 | |
a7885eb8 KM |
3254 | return 0; |
3255 | } | |
3256 | ||
2e72b634 KS |
3257 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3258 | { | |
3259 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3260 | unsigned long usage; |
2e72b634 KS |
3261 | int i; |
3262 | ||
3263 | rcu_read_lock(); | |
3264 | if (!swap) | |
2c488db2 | 3265 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3266 | else |
2c488db2 | 3267 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3268 | |
3269 | if (!t) | |
3270 | goto unlock; | |
3271 | ||
ce00a967 | 3272 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3273 | |
3274 | /* | |
748dad36 | 3275 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3276 | * If it's not true, a threshold was crossed after last |
3277 | * call of __mem_cgroup_threshold(). | |
3278 | */ | |
5407a562 | 3279 | i = t->current_threshold; |
2e72b634 KS |
3280 | |
3281 | /* | |
3282 | * Iterate backward over array of thresholds starting from | |
3283 | * current_threshold and check if a threshold is crossed. | |
3284 | * If none of thresholds below usage is crossed, we read | |
3285 | * only one element of the array here. | |
3286 | */ | |
3287 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3288 | eventfd_signal(t->entries[i].eventfd, 1); | |
3289 | ||
3290 | /* i = current_threshold + 1 */ | |
3291 | i++; | |
3292 | ||
3293 | /* | |
3294 | * Iterate forward over array of thresholds starting from | |
3295 | * current_threshold+1 and check if a threshold is crossed. | |
3296 | * If none of thresholds above usage is crossed, we read | |
3297 | * only one element of the array here. | |
3298 | */ | |
3299 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3300 | eventfd_signal(t->entries[i].eventfd, 1); | |
3301 | ||
3302 | /* Update current_threshold */ | |
5407a562 | 3303 | t->current_threshold = i - 1; |
2e72b634 KS |
3304 | unlock: |
3305 | rcu_read_unlock(); | |
3306 | } | |
3307 | ||
3308 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3309 | { | |
ad4ca5f4 KS |
3310 | while (memcg) { |
3311 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3312 | if (do_memsw_account()) |
ad4ca5f4 KS |
3313 | __mem_cgroup_threshold(memcg, true); |
3314 | ||
3315 | memcg = parent_mem_cgroup(memcg); | |
3316 | } | |
2e72b634 KS |
3317 | } |
3318 | ||
3319 | static int compare_thresholds(const void *a, const void *b) | |
3320 | { | |
3321 | const struct mem_cgroup_threshold *_a = a; | |
3322 | const struct mem_cgroup_threshold *_b = b; | |
3323 | ||
2bff24a3 GT |
3324 | if (_a->threshold > _b->threshold) |
3325 | return 1; | |
3326 | ||
3327 | if (_a->threshold < _b->threshold) | |
3328 | return -1; | |
3329 | ||
3330 | return 0; | |
2e72b634 KS |
3331 | } |
3332 | ||
c0ff4b85 | 3333 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3334 | { |
3335 | struct mem_cgroup_eventfd_list *ev; | |
3336 | ||
2bcf2e92 MH |
3337 | spin_lock(&memcg_oom_lock); |
3338 | ||
c0ff4b85 | 3339 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3340 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3341 | |
3342 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3343 | return 0; |
3344 | } | |
3345 | ||
c0ff4b85 | 3346 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3347 | { |
7d74b06f KH |
3348 | struct mem_cgroup *iter; |
3349 | ||
c0ff4b85 | 3350 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3351 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3352 | } |
3353 | ||
59b6f873 | 3354 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3355 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3356 | { |
2c488db2 KS |
3357 | struct mem_cgroup_thresholds *thresholds; |
3358 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3359 | unsigned long threshold; |
3360 | unsigned long usage; | |
2c488db2 | 3361 | int i, size, ret; |
2e72b634 | 3362 | |
650c5e56 | 3363 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3364 | if (ret) |
3365 | return ret; | |
3366 | ||
3367 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3368 | |
05b84301 | 3369 | if (type == _MEM) { |
2c488db2 | 3370 | thresholds = &memcg->thresholds; |
ce00a967 | 3371 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3372 | } else if (type == _MEMSWAP) { |
2c488db2 | 3373 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3374 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3375 | } else |
2e72b634 KS |
3376 | BUG(); |
3377 | ||
2e72b634 | 3378 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3379 | if (thresholds->primary) |
2e72b634 KS |
3380 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3381 | ||
2c488db2 | 3382 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3383 | |
3384 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3385 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3386 | GFP_KERNEL); |
2c488db2 | 3387 | if (!new) { |
2e72b634 KS |
3388 | ret = -ENOMEM; |
3389 | goto unlock; | |
3390 | } | |
2c488db2 | 3391 | new->size = size; |
2e72b634 KS |
3392 | |
3393 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3394 | if (thresholds->primary) { |
3395 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3396 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3397 | } |
3398 | ||
2e72b634 | 3399 | /* Add new threshold */ |
2c488db2 KS |
3400 | new->entries[size - 1].eventfd = eventfd; |
3401 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3402 | |
3403 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3404 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3405 | compare_thresholds, NULL); |
3406 | ||
3407 | /* Find current threshold */ | |
2c488db2 | 3408 | new->current_threshold = -1; |
2e72b634 | 3409 | for (i = 0; i < size; i++) { |
748dad36 | 3410 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3411 | /* |
2c488db2 KS |
3412 | * new->current_threshold will not be used until |
3413 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3414 | * it here. |
3415 | */ | |
2c488db2 | 3416 | ++new->current_threshold; |
748dad36 SZ |
3417 | } else |
3418 | break; | |
2e72b634 KS |
3419 | } |
3420 | ||
2c488db2 KS |
3421 | /* Free old spare buffer and save old primary buffer as spare */ |
3422 | kfree(thresholds->spare); | |
3423 | thresholds->spare = thresholds->primary; | |
3424 | ||
3425 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3426 | |
907860ed | 3427 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3428 | synchronize_rcu(); |
3429 | ||
2e72b634 KS |
3430 | unlock: |
3431 | mutex_unlock(&memcg->thresholds_lock); | |
3432 | ||
3433 | return ret; | |
3434 | } | |
3435 | ||
59b6f873 | 3436 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3437 | struct eventfd_ctx *eventfd, const char *args) |
3438 | { | |
59b6f873 | 3439 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3440 | } |
3441 | ||
59b6f873 | 3442 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3443 | struct eventfd_ctx *eventfd, const char *args) |
3444 | { | |
59b6f873 | 3445 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3446 | } |
3447 | ||
59b6f873 | 3448 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3449 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3450 | { |
2c488db2 KS |
3451 | struct mem_cgroup_thresholds *thresholds; |
3452 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3453 | unsigned long usage; |
2c488db2 | 3454 | int i, j, size; |
2e72b634 KS |
3455 | |
3456 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3457 | |
3458 | if (type == _MEM) { | |
2c488db2 | 3459 | thresholds = &memcg->thresholds; |
ce00a967 | 3460 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3461 | } else if (type == _MEMSWAP) { |
2c488db2 | 3462 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3463 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3464 | } else |
2e72b634 KS |
3465 | BUG(); |
3466 | ||
371528ca AV |
3467 | if (!thresholds->primary) |
3468 | goto unlock; | |
3469 | ||
2e72b634 KS |
3470 | /* Check if a threshold crossed before removing */ |
3471 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3472 | ||
3473 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3474 | size = 0; |
3475 | for (i = 0; i < thresholds->primary->size; i++) { | |
3476 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3477 | size++; |
3478 | } | |
3479 | ||
2c488db2 | 3480 | new = thresholds->spare; |
907860ed | 3481 | |
2e72b634 KS |
3482 | /* Set thresholds array to NULL if we don't have thresholds */ |
3483 | if (!size) { | |
2c488db2 KS |
3484 | kfree(new); |
3485 | new = NULL; | |
907860ed | 3486 | goto swap_buffers; |
2e72b634 KS |
3487 | } |
3488 | ||
2c488db2 | 3489 | new->size = size; |
2e72b634 KS |
3490 | |
3491 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3492 | new->current_threshold = -1; |
3493 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3494 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3495 | continue; |
3496 | ||
2c488db2 | 3497 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3498 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3499 | /* |
2c488db2 | 3500 | * new->current_threshold will not be used |
2e72b634 KS |
3501 | * until rcu_assign_pointer(), so it's safe to increment |
3502 | * it here. | |
3503 | */ | |
2c488db2 | 3504 | ++new->current_threshold; |
2e72b634 KS |
3505 | } |
3506 | j++; | |
3507 | } | |
3508 | ||
907860ed | 3509 | swap_buffers: |
2c488db2 KS |
3510 | /* Swap primary and spare array */ |
3511 | thresholds->spare = thresholds->primary; | |
8c757763 | 3512 | |
2c488db2 | 3513 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3514 | |
907860ed | 3515 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3516 | synchronize_rcu(); |
6611d8d7 MC |
3517 | |
3518 | /* If all events are unregistered, free the spare array */ | |
3519 | if (!new) { | |
3520 | kfree(thresholds->spare); | |
3521 | thresholds->spare = NULL; | |
3522 | } | |
371528ca | 3523 | unlock: |
2e72b634 | 3524 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3525 | } |
c1e862c1 | 3526 | |
59b6f873 | 3527 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3528 | struct eventfd_ctx *eventfd) |
3529 | { | |
59b6f873 | 3530 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3531 | } |
3532 | ||
59b6f873 | 3533 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3534 | struct eventfd_ctx *eventfd) |
3535 | { | |
59b6f873 | 3536 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3537 | } |
3538 | ||
59b6f873 | 3539 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3540 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3541 | { |
9490ff27 | 3542 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3543 | |
9490ff27 KH |
3544 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3545 | if (!event) | |
3546 | return -ENOMEM; | |
3547 | ||
1af8efe9 | 3548 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3549 | |
3550 | event->eventfd = eventfd; | |
3551 | list_add(&event->list, &memcg->oom_notify); | |
3552 | ||
3553 | /* already in OOM ? */ | |
c2b42d3c | 3554 | if (memcg->under_oom) |
9490ff27 | 3555 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3556 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3557 | |
3558 | return 0; | |
3559 | } | |
3560 | ||
59b6f873 | 3561 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3562 | struct eventfd_ctx *eventfd) |
9490ff27 | 3563 | { |
9490ff27 | 3564 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3565 | |
1af8efe9 | 3566 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3567 | |
c0ff4b85 | 3568 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3569 | if (ev->eventfd == eventfd) { |
3570 | list_del(&ev->list); | |
3571 | kfree(ev); | |
3572 | } | |
3573 | } | |
3574 | ||
1af8efe9 | 3575 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3576 | } |
3577 | ||
2da8ca82 | 3578 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3579 | { |
2da8ca82 | 3580 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3581 | |
791badbd | 3582 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3583 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3584 | return 0; |
3585 | } | |
3586 | ||
182446d0 | 3587 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3588 | struct cftype *cft, u64 val) |
3589 | { | |
182446d0 | 3590 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3591 | |
3592 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3593 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3594 | return -EINVAL; |
3595 | ||
c0ff4b85 | 3596 | memcg->oom_kill_disable = val; |
4d845ebf | 3597 | if (!val) |
c0ff4b85 | 3598 | memcg_oom_recover(memcg); |
3dae7fec | 3599 | |
3c11ecf4 KH |
3600 | return 0; |
3601 | } | |
3602 | ||
52ebea74 TH |
3603 | #ifdef CONFIG_CGROUP_WRITEBACK |
3604 | ||
3605 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3606 | { | |
3607 | return &memcg->cgwb_list; | |
3608 | } | |
3609 | ||
841710aa TH |
3610 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3611 | { | |
3612 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3613 | } | |
3614 | ||
3615 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3616 | { | |
3617 | wb_domain_exit(&memcg->cgwb_domain); | |
3618 | } | |
3619 | ||
2529bb3a TH |
3620 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3621 | { | |
3622 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3623 | } | |
3624 | ||
841710aa TH |
3625 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3626 | { | |
3627 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3628 | ||
3629 | if (!memcg->css.parent) | |
3630 | return NULL; | |
3631 | ||
3632 | return &memcg->cgwb_domain; | |
3633 | } | |
3634 | ||
c2aa723a TH |
3635 | /** |
3636 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3637 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3638 | * @pfilepages: out parameter for number of file pages |
3639 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3640 | * @pdirty: out parameter for number of dirty pages |
3641 | * @pwriteback: out parameter for number of pages under writeback | |
3642 | * | |
c5edf9cd TH |
3643 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3644 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3645 | * is a bit more involved. | |
c2aa723a | 3646 | * |
c5edf9cd TH |
3647 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3648 | * headroom is calculated as the lowest headroom of itself and the | |
3649 | * ancestors. Note that this doesn't consider the actual amount of | |
3650 | * available memory in the system. The caller should further cap | |
3651 | * *@pheadroom accordingly. | |
c2aa723a | 3652 | */ |
c5edf9cd TH |
3653 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3654 | unsigned long *pheadroom, unsigned long *pdirty, | |
3655 | unsigned long *pwriteback) | |
c2aa723a TH |
3656 | { |
3657 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3658 | struct mem_cgroup *parent; | |
c2aa723a TH |
3659 | |
3660 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3661 | ||
3662 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3663 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3664 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3665 | (1 << LRU_ACTIVE_FILE)); | |
3666 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3667 | |
c2aa723a TH |
3668 | while ((parent = parent_mem_cgroup(memcg))) { |
3669 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3670 | unsigned long used = page_counter_read(&memcg->memory); | |
3671 | ||
c5edf9cd | 3672 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3673 | memcg = parent; |
3674 | } | |
c2aa723a TH |
3675 | } |
3676 | ||
841710aa TH |
3677 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3678 | ||
3679 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3680 | { | |
3681 | return 0; | |
3682 | } | |
3683 | ||
3684 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3685 | { | |
3686 | } | |
3687 | ||
2529bb3a TH |
3688 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3689 | { | |
3690 | } | |
3691 | ||
52ebea74 TH |
3692 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3693 | ||
3bc942f3 TH |
3694 | /* |
3695 | * DO NOT USE IN NEW FILES. | |
3696 | * | |
3697 | * "cgroup.event_control" implementation. | |
3698 | * | |
3699 | * This is way over-engineered. It tries to support fully configurable | |
3700 | * events for each user. Such level of flexibility is completely | |
3701 | * unnecessary especially in the light of the planned unified hierarchy. | |
3702 | * | |
3703 | * Please deprecate this and replace with something simpler if at all | |
3704 | * possible. | |
3705 | */ | |
3706 | ||
79bd9814 TH |
3707 | /* |
3708 | * Unregister event and free resources. | |
3709 | * | |
3710 | * Gets called from workqueue. | |
3711 | */ | |
3bc942f3 | 3712 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3713 | { |
3bc942f3 TH |
3714 | struct mem_cgroup_event *event = |
3715 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3716 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3717 | |
3718 | remove_wait_queue(event->wqh, &event->wait); | |
3719 | ||
59b6f873 | 3720 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3721 | |
3722 | /* Notify userspace the event is going away. */ | |
3723 | eventfd_signal(event->eventfd, 1); | |
3724 | ||
3725 | eventfd_ctx_put(event->eventfd); | |
3726 | kfree(event); | |
59b6f873 | 3727 | css_put(&memcg->css); |
79bd9814 TH |
3728 | } |
3729 | ||
3730 | /* | |
3731 | * Gets called on POLLHUP on eventfd when user closes it. | |
3732 | * | |
3733 | * Called with wqh->lock held and interrupts disabled. | |
3734 | */ | |
3bc942f3 TH |
3735 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3736 | int sync, void *key) | |
79bd9814 | 3737 | { |
3bc942f3 TH |
3738 | struct mem_cgroup_event *event = |
3739 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3740 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3741 | unsigned long flags = (unsigned long)key; |
3742 | ||
3743 | if (flags & POLLHUP) { | |
3744 | /* | |
3745 | * If the event has been detached at cgroup removal, we | |
3746 | * can simply return knowing the other side will cleanup | |
3747 | * for us. | |
3748 | * | |
3749 | * We can't race against event freeing since the other | |
3750 | * side will require wqh->lock via remove_wait_queue(), | |
3751 | * which we hold. | |
3752 | */ | |
fba94807 | 3753 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3754 | if (!list_empty(&event->list)) { |
3755 | list_del_init(&event->list); | |
3756 | /* | |
3757 | * We are in atomic context, but cgroup_event_remove() | |
3758 | * may sleep, so we have to call it in workqueue. | |
3759 | */ | |
3760 | schedule_work(&event->remove); | |
3761 | } | |
fba94807 | 3762 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3763 | } |
3764 | ||
3765 | return 0; | |
3766 | } | |
3767 | ||
3bc942f3 | 3768 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3769 | wait_queue_head_t *wqh, poll_table *pt) |
3770 | { | |
3bc942f3 TH |
3771 | struct mem_cgroup_event *event = |
3772 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3773 | |
3774 | event->wqh = wqh; | |
3775 | add_wait_queue(wqh, &event->wait); | |
3776 | } | |
3777 | ||
3778 | /* | |
3bc942f3 TH |
3779 | * DO NOT USE IN NEW FILES. |
3780 | * | |
79bd9814 TH |
3781 | * Parse input and register new cgroup event handler. |
3782 | * | |
3783 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3784 | * Interpretation of args is defined by control file implementation. | |
3785 | */ | |
451af504 TH |
3786 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3787 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3788 | { |
451af504 | 3789 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3790 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3791 | struct mem_cgroup_event *event; |
79bd9814 TH |
3792 | struct cgroup_subsys_state *cfile_css; |
3793 | unsigned int efd, cfd; | |
3794 | struct fd efile; | |
3795 | struct fd cfile; | |
fba94807 | 3796 | const char *name; |
79bd9814 TH |
3797 | char *endp; |
3798 | int ret; | |
3799 | ||
451af504 TH |
3800 | buf = strstrip(buf); |
3801 | ||
3802 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3803 | if (*endp != ' ') |
3804 | return -EINVAL; | |
451af504 | 3805 | buf = endp + 1; |
79bd9814 | 3806 | |
451af504 | 3807 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3808 | if ((*endp != ' ') && (*endp != '\0')) |
3809 | return -EINVAL; | |
451af504 | 3810 | buf = endp + 1; |
79bd9814 TH |
3811 | |
3812 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3813 | if (!event) | |
3814 | return -ENOMEM; | |
3815 | ||
59b6f873 | 3816 | event->memcg = memcg; |
79bd9814 | 3817 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3818 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3819 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3820 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3821 | |
3822 | efile = fdget(efd); | |
3823 | if (!efile.file) { | |
3824 | ret = -EBADF; | |
3825 | goto out_kfree; | |
3826 | } | |
3827 | ||
3828 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3829 | if (IS_ERR(event->eventfd)) { | |
3830 | ret = PTR_ERR(event->eventfd); | |
3831 | goto out_put_efile; | |
3832 | } | |
3833 | ||
3834 | cfile = fdget(cfd); | |
3835 | if (!cfile.file) { | |
3836 | ret = -EBADF; | |
3837 | goto out_put_eventfd; | |
3838 | } | |
3839 | ||
3840 | /* the process need read permission on control file */ | |
3841 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3842 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3843 | if (ret < 0) | |
3844 | goto out_put_cfile; | |
3845 | ||
fba94807 TH |
3846 | /* |
3847 | * Determine the event callbacks and set them in @event. This used | |
3848 | * to be done via struct cftype but cgroup core no longer knows | |
3849 | * about these events. The following is crude but the whole thing | |
3850 | * is for compatibility anyway. | |
3bc942f3 TH |
3851 | * |
3852 | * DO NOT ADD NEW FILES. | |
fba94807 | 3853 | */ |
b583043e | 3854 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3855 | |
3856 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3857 | event->register_event = mem_cgroup_usage_register_event; | |
3858 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3859 | } else if (!strcmp(name, "memory.oom_control")) { | |
3860 | event->register_event = mem_cgroup_oom_register_event; | |
3861 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3862 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3863 | event->register_event = vmpressure_register_event; | |
3864 | event->unregister_event = vmpressure_unregister_event; | |
3865 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3866 | event->register_event = memsw_cgroup_usage_register_event; |
3867 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3868 | } else { |
3869 | ret = -EINVAL; | |
3870 | goto out_put_cfile; | |
3871 | } | |
3872 | ||
79bd9814 | 3873 | /* |
b5557c4c TH |
3874 | * Verify @cfile should belong to @css. Also, remaining events are |
3875 | * automatically removed on cgroup destruction but the removal is | |
3876 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3877 | */ |
b583043e | 3878 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3879 | &memory_cgrp_subsys); |
79bd9814 | 3880 | ret = -EINVAL; |
5a17f543 | 3881 | if (IS_ERR(cfile_css)) |
79bd9814 | 3882 | goto out_put_cfile; |
5a17f543 TH |
3883 | if (cfile_css != css) { |
3884 | css_put(cfile_css); | |
79bd9814 | 3885 | goto out_put_cfile; |
5a17f543 | 3886 | } |
79bd9814 | 3887 | |
451af504 | 3888 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3889 | if (ret) |
3890 | goto out_put_css; | |
3891 | ||
3892 | efile.file->f_op->poll(efile.file, &event->pt); | |
3893 | ||
fba94807 TH |
3894 | spin_lock(&memcg->event_list_lock); |
3895 | list_add(&event->list, &memcg->event_list); | |
3896 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3897 | |
3898 | fdput(cfile); | |
3899 | fdput(efile); | |
3900 | ||
451af504 | 3901 | return nbytes; |
79bd9814 TH |
3902 | |
3903 | out_put_css: | |
b5557c4c | 3904 | css_put(css); |
79bd9814 TH |
3905 | out_put_cfile: |
3906 | fdput(cfile); | |
3907 | out_put_eventfd: | |
3908 | eventfd_ctx_put(event->eventfd); | |
3909 | out_put_efile: | |
3910 | fdput(efile); | |
3911 | out_kfree: | |
3912 | kfree(event); | |
3913 | ||
3914 | return ret; | |
3915 | } | |
3916 | ||
241994ed | 3917 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 3918 | { |
0eea1030 | 3919 | .name = "usage_in_bytes", |
8c7c6e34 | 3920 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 3921 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3922 | }, |
c84872e1 PE |
3923 | { |
3924 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3925 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 3926 | .write = mem_cgroup_reset, |
791badbd | 3927 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 3928 | }, |
8cdea7c0 | 3929 | { |
0eea1030 | 3930 | .name = "limit_in_bytes", |
8c7c6e34 | 3931 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 3932 | .write = mem_cgroup_write, |
791badbd | 3933 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3934 | }, |
296c81d8 BS |
3935 | { |
3936 | .name = "soft_limit_in_bytes", | |
3937 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 3938 | .write = mem_cgroup_write, |
791badbd | 3939 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 3940 | }, |
8cdea7c0 BS |
3941 | { |
3942 | .name = "failcnt", | |
8c7c6e34 | 3943 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 3944 | .write = mem_cgroup_reset, |
791badbd | 3945 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3946 | }, |
d2ceb9b7 KH |
3947 | { |
3948 | .name = "stat", | |
2da8ca82 | 3949 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 3950 | }, |
c1e862c1 KH |
3951 | { |
3952 | .name = "force_empty", | |
6770c64e | 3953 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 3954 | }, |
18f59ea7 BS |
3955 | { |
3956 | .name = "use_hierarchy", | |
3957 | .write_u64 = mem_cgroup_hierarchy_write, | |
3958 | .read_u64 = mem_cgroup_hierarchy_read, | |
3959 | }, | |
79bd9814 | 3960 | { |
3bc942f3 | 3961 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 3962 | .write = memcg_write_event_control, |
7dbdb199 | 3963 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 3964 | }, |
a7885eb8 KM |
3965 | { |
3966 | .name = "swappiness", | |
3967 | .read_u64 = mem_cgroup_swappiness_read, | |
3968 | .write_u64 = mem_cgroup_swappiness_write, | |
3969 | }, | |
7dc74be0 DN |
3970 | { |
3971 | .name = "move_charge_at_immigrate", | |
3972 | .read_u64 = mem_cgroup_move_charge_read, | |
3973 | .write_u64 = mem_cgroup_move_charge_write, | |
3974 | }, | |
9490ff27 KH |
3975 | { |
3976 | .name = "oom_control", | |
2da8ca82 | 3977 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 3978 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
3979 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
3980 | }, | |
70ddf637 AV |
3981 | { |
3982 | .name = "pressure_level", | |
70ddf637 | 3983 | }, |
406eb0c9 YH |
3984 | #ifdef CONFIG_NUMA |
3985 | { | |
3986 | .name = "numa_stat", | |
2da8ca82 | 3987 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
3988 | }, |
3989 | #endif | |
510fc4e1 GC |
3990 | { |
3991 | .name = "kmem.limit_in_bytes", | |
3992 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 3993 | .write = mem_cgroup_write, |
791badbd | 3994 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
3995 | }, |
3996 | { | |
3997 | .name = "kmem.usage_in_bytes", | |
3998 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 3999 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4000 | }, |
4001 | { | |
4002 | .name = "kmem.failcnt", | |
4003 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4004 | .write = mem_cgroup_reset, |
791badbd | 4005 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4006 | }, |
4007 | { | |
4008 | .name = "kmem.max_usage_in_bytes", | |
4009 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4010 | .write = mem_cgroup_reset, |
791badbd | 4011 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4012 | }, |
749c5415 GC |
4013 | #ifdef CONFIG_SLABINFO |
4014 | { | |
4015 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4016 | .seq_start = memcg_slab_start, |
4017 | .seq_next = memcg_slab_next, | |
4018 | .seq_stop = memcg_slab_stop, | |
b047501c | 4019 | .seq_show = memcg_slab_show, |
749c5415 GC |
4020 | }, |
4021 | #endif | |
d55f90bf VD |
4022 | { |
4023 | .name = "kmem.tcp.limit_in_bytes", | |
4024 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4025 | .write = mem_cgroup_write, | |
4026 | .read_u64 = mem_cgroup_read_u64, | |
4027 | }, | |
4028 | { | |
4029 | .name = "kmem.tcp.usage_in_bytes", | |
4030 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4031 | .read_u64 = mem_cgroup_read_u64, | |
4032 | }, | |
4033 | { | |
4034 | .name = "kmem.tcp.failcnt", | |
4035 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4036 | .write = mem_cgroup_reset, | |
4037 | .read_u64 = mem_cgroup_read_u64, | |
4038 | }, | |
4039 | { | |
4040 | .name = "kmem.tcp.max_usage_in_bytes", | |
4041 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4042 | .write = mem_cgroup_reset, | |
4043 | .read_u64 = mem_cgroup_read_u64, | |
4044 | }, | |
6bc10349 | 4045 | { }, /* terminate */ |
af36f906 | 4046 | }; |
8c7c6e34 | 4047 | |
73f576c0 JW |
4048 | /* |
4049 | * Private memory cgroup IDR | |
4050 | * | |
4051 | * Swap-out records and page cache shadow entries need to store memcg | |
4052 | * references in constrained space, so we maintain an ID space that is | |
4053 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4054 | * memory-controlled cgroups to 64k. | |
4055 | * | |
4056 | * However, there usually are many references to the oflline CSS after | |
4057 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4058 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4059 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4060 | * relatively small ID space and prevent the creation of new cgroups | |
4061 | * even when there are much fewer than 64k cgroups - possibly none. | |
4062 | * | |
4063 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4064 | * be freed and recycled when it's no longer needed, which is usually | |
4065 | * when the CSS is offlined. | |
4066 | * | |
4067 | * The only exception to that are records of swapped out tmpfs/shmem | |
4068 | * pages that need to be attributed to live ancestors on swapin. But | |
4069 | * those references are manageable from userspace. | |
4070 | */ | |
4071 | ||
4072 | static DEFINE_IDR(mem_cgroup_idr); | |
4073 | ||
615d66c3 | 4074 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4075 | { |
58fa2a55 | 4076 | VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0); |
615d66c3 | 4077 | atomic_add(n, &memcg->id.ref); |
73f576c0 JW |
4078 | } |
4079 | ||
615d66c3 | 4080 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4081 | { |
58fa2a55 | 4082 | VM_BUG_ON(atomic_read(&memcg->id.ref) < n); |
615d66c3 | 4083 | if (atomic_sub_and_test(n, &memcg->id.ref)) { |
73f576c0 JW |
4084 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
4085 | memcg->id.id = 0; | |
4086 | ||
4087 | /* Memcg ID pins CSS */ | |
4088 | css_put(&memcg->css); | |
4089 | } | |
4090 | } | |
4091 | ||
615d66c3 VD |
4092 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4093 | { | |
4094 | mem_cgroup_id_get_many(memcg, 1); | |
4095 | } | |
4096 | ||
4097 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4098 | { | |
4099 | mem_cgroup_id_put_many(memcg, 1); | |
4100 | } | |
4101 | ||
73f576c0 JW |
4102 | /** |
4103 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4104 | * @id: the memcg id to look up | |
4105 | * | |
4106 | * Caller must hold rcu_read_lock(). | |
4107 | */ | |
4108 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4109 | { | |
4110 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4111 | return idr_find(&mem_cgroup_idr, id); | |
4112 | } | |
4113 | ||
ef8f2327 | 4114 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4115 | { |
4116 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4117 | int tmp = node; |
1ecaab2b KH |
4118 | /* |
4119 | * This routine is called against possible nodes. | |
4120 | * But it's BUG to call kmalloc() against offline node. | |
4121 | * | |
4122 | * TODO: this routine can waste much memory for nodes which will | |
4123 | * never be onlined. It's better to use memory hotplug callback | |
4124 | * function. | |
4125 | */ | |
41e3355d KH |
4126 | if (!node_state(node, N_NORMAL_MEMORY)) |
4127 | tmp = -1; | |
17295c88 | 4128 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4129 | if (!pn) |
4130 | return 1; | |
1ecaab2b | 4131 | |
ef8f2327 MG |
4132 | lruvec_init(&pn->lruvec); |
4133 | pn->usage_in_excess = 0; | |
4134 | pn->on_tree = false; | |
4135 | pn->memcg = memcg; | |
4136 | ||
54f72fe0 | 4137 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4138 | return 0; |
4139 | } | |
4140 | ||
ef8f2327 | 4141 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4142 | { |
54f72fe0 | 4143 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4144 | } |
4145 | ||
40e952f9 | 4146 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4147 | { |
c8b2a36f | 4148 | int node; |
59927fb9 | 4149 | |
c8b2a36f | 4150 | for_each_node(node) |
ef8f2327 | 4151 | free_mem_cgroup_per_node_info(memcg, node); |
c8b2a36f | 4152 | free_percpu(memcg->stat); |
8ff69e2c | 4153 | kfree(memcg); |
59927fb9 | 4154 | } |
3afe36b1 | 4155 | |
40e952f9 TE |
4156 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4157 | { | |
4158 | memcg_wb_domain_exit(memcg); | |
4159 | __mem_cgroup_free(memcg); | |
4160 | } | |
4161 | ||
0b8f73e1 | 4162 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4163 | { |
d142e3e6 | 4164 | struct mem_cgroup *memcg; |
0b8f73e1 | 4165 | size_t size; |
6d12e2d8 | 4166 | int node; |
8cdea7c0 | 4167 | |
0b8f73e1 JW |
4168 | size = sizeof(struct mem_cgroup); |
4169 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4170 | ||
4171 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4172 | if (!memcg) |
0b8f73e1 JW |
4173 | return NULL; |
4174 | ||
73f576c0 JW |
4175 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4176 | 1, MEM_CGROUP_ID_MAX, | |
4177 | GFP_KERNEL); | |
4178 | if (memcg->id.id < 0) | |
4179 | goto fail; | |
4180 | ||
0b8f73e1 JW |
4181 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4182 | if (!memcg->stat) | |
4183 | goto fail; | |
78fb7466 | 4184 | |
3ed28fa1 | 4185 | for_each_node(node) |
ef8f2327 | 4186 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4187 | goto fail; |
f64c3f54 | 4188 | |
0b8f73e1 JW |
4189 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4190 | goto fail; | |
28dbc4b6 | 4191 | |
f7e1cb6e | 4192 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4193 | memcg->last_scanned_node = MAX_NUMNODES; |
4194 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4195 | mutex_init(&memcg->thresholds_lock); |
4196 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4197 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4198 | INIT_LIST_HEAD(&memcg->event_list); |
4199 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4200 | memcg->socket_pressure = jiffies; |
127424c8 | 4201 | #ifndef CONFIG_SLOB |
900a38f0 | 4202 | memcg->kmemcg_id = -1; |
900a38f0 | 4203 | #endif |
52ebea74 TH |
4204 | #ifdef CONFIG_CGROUP_WRITEBACK |
4205 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4206 | #endif | |
73f576c0 | 4207 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4208 | return memcg; |
4209 | fail: | |
73f576c0 JW |
4210 | if (memcg->id.id > 0) |
4211 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
40e952f9 | 4212 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4213 | return NULL; |
d142e3e6 GC |
4214 | } |
4215 | ||
0b8f73e1 JW |
4216 | static struct cgroup_subsys_state * __ref |
4217 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4218 | { |
0b8f73e1 JW |
4219 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4220 | struct mem_cgroup *memcg; | |
4221 | long error = -ENOMEM; | |
d142e3e6 | 4222 | |
0b8f73e1 JW |
4223 | memcg = mem_cgroup_alloc(); |
4224 | if (!memcg) | |
4225 | return ERR_PTR(error); | |
d142e3e6 | 4226 | |
0b8f73e1 JW |
4227 | memcg->high = PAGE_COUNTER_MAX; |
4228 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4229 | if (parent) { | |
4230 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4231 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4232 | } | |
4233 | if (parent && parent->use_hierarchy) { | |
4234 | memcg->use_hierarchy = true; | |
3e32cb2e | 4235 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4236 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4237 | page_counter_init(&memcg->memsw, &parent->memsw); |
4238 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4239 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4240 | } else { |
3e32cb2e | 4241 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4242 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4243 | page_counter_init(&memcg->memsw, NULL); |
4244 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4245 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4246 | /* |
4247 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4248 | * much sense so let cgroup subsystem know about this | |
4249 | * unfortunate state in our controller. | |
4250 | */ | |
d142e3e6 | 4251 | if (parent != root_mem_cgroup) |
073219e9 | 4252 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4253 | } |
d6441637 | 4254 | |
0b8f73e1 JW |
4255 | /* The following stuff does not apply to the root */ |
4256 | if (!parent) { | |
4257 | root_mem_cgroup = memcg; | |
4258 | return &memcg->css; | |
4259 | } | |
4260 | ||
b313aeee | 4261 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4262 | if (error) |
4263 | goto fail; | |
127424c8 | 4264 | |
f7e1cb6e | 4265 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4266 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4267 | |
0b8f73e1 JW |
4268 | return &memcg->css; |
4269 | fail: | |
4270 | mem_cgroup_free(memcg); | |
ea3a9645 | 4271 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4272 | } |
4273 | ||
73f576c0 | 4274 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4275 | { |
58fa2a55 VD |
4276 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4277 | ||
73f576c0 | 4278 | /* Online state pins memcg ID, memcg ID pins CSS */ |
58fa2a55 | 4279 | atomic_set(&memcg->id.ref, 1); |
73f576c0 | 4280 | css_get(css); |
2f7dd7a4 | 4281 | return 0; |
8cdea7c0 BS |
4282 | } |
4283 | ||
eb95419b | 4284 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4285 | { |
eb95419b | 4286 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4287 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4288 | |
4289 | /* | |
4290 | * Unregister events and notify userspace. | |
4291 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4292 | * directory to avoid race between userspace and kernelspace. | |
4293 | */ | |
fba94807 TH |
4294 | spin_lock(&memcg->event_list_lock); |
4295 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4296 | list_del_init(&event->list); |
4297 | schedule_work(&event->remove); | |
4298 | } | |
fba94807 | 4299 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4300 | |
567e9ab2 | 4301 | memcg_offline_kmem(memcg); |
52ebea74 | 4302 | wb_memcg_offline(memcg); |
73f576c0 JW |
4303 | |
4304 | mem_cgroup_id_put(memcg); | |
df878fb0 KH |
4305 | } |
4306 | ||
6df38689 VD |
4307 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4308 | { | |
4309 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4310 | ||
4311 | invalidate_reclaim_iterators(memcg); | |
4312 | } | |
4313 | ||
eb95419b | 4314 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4315 | { |
eb95419b | 4316 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4317 | |
f7e1cb6e | 4318 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4319 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4320 | |
0db15298 | 4321 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4322 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4323 | |
0b8f73e1 JW |
4324 | vmpressure_cleanup(&memcg->vmpressure); |
4325 | cancel_work_sync(&memcg->high_work); | |
4326 | mem_cgroup_remove_from_trees(memcg); | |
d886f4e4 | 4327 | memcg_free_kmem(memcg); |
0b8f73e1 | 4328 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4329 | } |
4330 | ||
1ced953b TH |
4331 | /** |
4332 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4333 | * @css: the target css | |
4334 | * | |
4335 | * Reset the states of the mem_cgroup associated with @css. This is | |
4336 | * invoked when the userland requests disabling on the default hierarchy | |
4337 | * but the memcg is pinned through dependency. The memcg should stop | |
4338 | * applying policies and should revert to the vanilla state as it may be | |
4339 | * made visible again. | |
4340 | * | |
4341 | * The current implementation only resets the essential configurations. | |
4342 | * This needs to be expanded to cover all the visible parts. | |
4343 | */ | |
4344 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4345 | { | |
4346 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4347 | ||
d334c9bc VD |
4348 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
4349 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); | |
4350 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); | |
4351 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); | |
4352 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
241994ed JW |
4353 | memcg->low = 0; |
4354 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4355 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4356 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4357 | } |
4358 | ||
02491447 | 4359 | #ifdef CONFIG_MMU |
7dc74be0 | 4360 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4361 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4362 | { |
05b84301 | 4363 | int ret; |
9476db97 | 4364 | |
d0164adc MG |
4365 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4366 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4367 | if (!ret) { |
854ffa8d | 4368 | mc.precharge += count; |
854ffa8d DN |
4369 | return ret; |
4370 | } | |
9476db97 | 4371 | |
3674534b | 4372 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4373 | while (count--) { |
3674534b | 4374 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4375 | if (ret) |
38c5d72f | 4376 | return ret; |
854ffa8d | 4377 | mc.precharge++; |
9476db97 | 4378 | cond_resched(); |
854ffa8d | 4379 | } |
9476db97 | 4380 | return 0; |
4ffef5fe DN |
4381 | } |
4382 | ||
4ffef5fe DN |
4383 | union mc_target { |
4384 | struct page *page; | |
02491447 | 4385 | swp_entry_t ent; |
4ffef5fe DN |
4386 | }; |
4387 | ||
4ffef5fe | 4388 | enum mc_target_type { |
8d32ff84 | 4389 | MC_TARGET_NONE = 0, |
4ffef5fe | 4390 | MC_TARGET_PAGE, |
02491447 | 4391 | MC_TARGET_SWAP, |
4ffef5fe DN |
4392 | }; |
4393 | ||
90254a65 DN |
4394 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4395 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4396 | { |
90254a65 | 4397 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4398 | |
90254a65 DN |
4399 | if (!page || !page_mapped(page)) |
4400 | return NULL; | |
4401 | if (PageAnon(page)) { | |
1dfab5ab | 4402 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4403 | return NULL; |
1dfab5ab JW |
4404 | } else { |
4405 | if (!(mc.flags & MOVE_FILE)) | |
4406 | return NULL; | |
4407 | } | |
90254a65 DN |
4408 | if (!get_page_unless_zero(page)) |
4409 | return NULL; | |
4410 | ||
4411 | return page; | |
4412 | } | |
4413 | ||
4b91355e | 4414 | #ifdef CONFIG_SWAP |
90254a65 | 4415 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4416 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4417 | { |
90254a65 DN |
4418 | struct page *page = NULL; |
4419 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4420 | ||
1dfab5ab | 4421 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4422 | return NULL; |
4b91355e KH |
4423 | /* |
4424 | * Because lookup_swap_cache() updates some statistics counter, | |
4425 | * we call find_get_page() with swapper_space directly. | |
4426 | */ | |
f6ab1f7f | 4427 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4428 | if (do_memsw_account()) |
90254a65 DN |
4429 | entry->val = ent.val; |
4430 | ||
4431 | return page; | |
4432 | } | |
4b91355e KH |
4433 | #else |
4434 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4435 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4436 | { |
4437 | return NULL; | |
4438 | } | |
4439 | #endif | |
90254a65 | 4440 | |
87946a72 DN |
4441 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4442 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4443 | { | |
4444 | struct page *page = NULL; | |
87946a72 DN |
4445 | struct address_space *mapping; |
4446 | pgoff_t pgoff; | |
4447 | ||
4448 | if (!vma->vm_file) /* anonymous vma */ | |
4449 | return NULL; | |
1dfab5ab | 4450 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4451 | return NULL; |
4452 | ||
87946a72 | 4453 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4454 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4455 | |
4456 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4457 | #ifdef CONFIG_SWAP |
4458 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4459 | if (shmem_mapping(mapping)) { |
4460 | page = find_get_entry(mapping, pgoff); | |
4461 | if (radix_tree_exceptional_entry(page)) { | |
4462 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4463 | if (do_memsw_account()) |
139b6a6f | 4464 | *entry = swp; |
f6ab1f7f HY |
4465 | page = find_get_page(swap_address_space(swp), |
4466 | swp_offset(swp)); | |
139b6a6f JW |
4467 | } |
4468 | } else | |
4469 | page = find_get_page(mapping, pgoff); | |
4470 | #else | |
4471 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4472 | #endif |
87946a72 DN |
4473 | return page; |
4474 | } | |
4475 | ||
b1b0deab CG |
4476 | /** |
4477 | * mem_cgroup_move_account - move account of the page | |
4478 | * @page: the page | |
25843c2b | 4479 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4480 | * @from: mem_cgroup which the page is moved from. |
4481 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4482 | * | |
3ac808fd | 4483 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4484 | * |
4485 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4486 | * from old cgroup. | |
4487 | */ | |
4488 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4489 | bool compound, |
b1b0deab CG |
4490 | struct mem_cgroup *from, |
4491 | struct mem_cgroup *to) | |
4492 | { | |
4493 | unsigned long flags; | |
f627c2f5 | 4494 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4495 | int ret; |
c4843a75 | 4496 | bool anon; |
b1b0deab CG |
4497 | |
4498 | VM_BUG_ON(from == to); | |
4499 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4500 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4501 | |
4502 | /* | |
6a93ca8f | 4503 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4504 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4505 | */ |
f627c2f5 | 4506 | ret = -EBUSY; |
b1b0deab CG |
4507 | if (!trylock_page(page)) |
4508 | goto out; | |
4509 | ||
4510 | ret = -EINVAL; | |
4511 | if (page->mem_cgroup != from) | |
4512 | goto out_unlock; | |
4513 | ||
c4843a75 GT |
4514 | anon = PageAnon(page); |
4515 | ||
b1b0deab CG |
4516 | spin_lock_irqsave(&from->move_lock, flags); |
4517 | ||
c4843a75 | 4518 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4519 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4520 | nr_pages); | |
4521 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4522 | nr_pages); | |
4523 | } | |
4524 | ||
c4843a75 GT |
4525 | /* |
4526 | * move_lock grabbed above and caller set from->moving_account, so | |
4527 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4528 | * So mapping should be stable for dirty pages. | |
4529 | */ | |
4530 | if (!anon && PageDirty(page)) { | |
4531 | struct address_space *mapping = page_mapping(page); | |
4532 | ||
4533 | if (mapping_cap_account_dirty(mapping)) { | |
4534 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4535 | nr_pages); | |
4536 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4537 | nr_pages); | |
4538 | } | |
4539 | } | |
4540 | ||
b1b0deab CG |
4541 | if (PageWriteback(page)) { |
4542 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4543 | nr_pages); | |
4544 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4545 | nr_pages); | |
4546 | } | |
4547 | ||
4548 | /* | |
4549 | * It is safe to change page->mem_cgroup here because the page | |
4550 | * is referenced, charged, and isolated - we can't race with | |
4551 | * uncharging, charging, migration, or LRU putback. | |
4552 | */ | |
4553 | ||
4554 | /* caller should have done css_get */ | |
4555 | page->mem_cgroup = to; | |
4556 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4557 | ||
4558 | ret = 0; | |
4559 | ||
4560 | local_irq_disable(); | |
f627c2f5 | 4561 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4562 | memcg_check_events(to, page); |
f627c2f5 | 4563 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4564 | memcg_check_events(from, page); |
4565 | local_irq_enable(); | |
4566 | out_unlock: | |
4567 | unlock_page(page); | |
4568 | out: | |
4569 | return ret; | |
4570 | } | |
4571 | ||
7cf7806c LR |
4572 | /** |
4573 | * get_mctgt_type - get target type of moving charge | |
4574 | * @vma: the vma the pte to be checked belongs | |
4575 | * @addr: the address corresponding to the pte to be checked | |
4576 | * @ptent: the pte to be checked | |
4577 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4578 | * | |
4579 | * Returns | |
4580 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4581 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4582 | * move charge. if @target is not NULL, the page is stored in target->page | |
4583 | * with extra refcnt got(Callers should handle it). | |
4584 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4585 | * target for charge migration. if @target is not NULL, the entry is stored | |
4586 | * in target->ent. | |
4587 | * | |
4588 | * Called with pte lock held. | |
4589 | */ | |
4590 | ||
8d32ff84 | 4591 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4592 | unsigned long addr, pte_t ptent, union mc_target *target) |
4593 | { | |
4594 | struct page *page = NULL; | |
8d32ff84 | 4595 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4596 | swp_entry_t ent = { .val = 0 }; |
4597 | ||
4598 | if (pte_present(ptent)) | |
4599 | page = mc_handle_present_pte(vma, addr, ptent); | |
4600 | else if (is_swap_pte(ptent)) | |
48406ef8 | 4601 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 4602 | else if (pte_none(ptent)) |
87946a72 | 4603 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4604 | |
4605 | if (!page && !ent.val) | |
8d32ff84 | 4606 | return ret; |
02491447 | 4607 | if (page) { |
02491447 | 4608 | /* |
0a31bc97 | 4609 | * Do only loose check w/o serialization. |
1306a85a | 4610 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4611 | * not under LRU exclusion. |
02491447 | 4612 | */ |
1306a85a | 4613 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4614 | ret = MC_TARGET_PAGE; |
4615 | if (target) | |
4616 | target->page = page; | |
4617 | } | |
4618 | if (!ret || !target) | |
4619 | put_page(page); | |
4620 | } | |
90254a65 DN |
4621 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4622 | if (ent.val && !ret && | |
34c00c31 | 4623 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4624 | ret = MC_TARGET_SWAP; |
4625 | if (target) | |
4626 | target->ent = ent; | |
4ffef5fe | 4627 | } |
4ffef5fe DN |
4628 | return ret; |
4629 | } | |
4630 | ||
12724850 NH |
4631 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4632 | /* | |
4633 | * We don't consider swapping or file mapped pages because THP does not | |
4634 | * support them for now. | |
4635 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4636 | */ | |
4637 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4638 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4639 | { | |
4640 | struct page *page = NULL; | |
12724850 NH |
4641 | enum mc_target_type ret = MC_TARGET_NONE; |
4642 | ||
4643 | page = pmd_page(pmd); | |
309381fe | 4644 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4645 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4646 | return ret; |
1306a85a | 4647 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4648 | ret = MC_TARGET_PAGE; |
4649 | if (target) { | |
4650 | get_page(page); | |
4651 | target->page = page; | |
4652 | } | |
4653 | } | |
4654 | return ret; | |
4655 | } | |
4656 | #else | |
4657 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4658 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4659 | { | |
4660 | return MC_TARGET_NONE; | |
4661 | } | |
4662 | #endif | |
4663 | ||
4ffef5fe DN |
4664 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4665 | unsigned long addr, unsigned long end, | |
4666 | struct mm_walk *walk) | |
4667 | { | |
26bcd64a | 4668 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4669 | pte_t *pte; |
4670 | spinlock_t *ptl; | |
4671 | ||
b6ec57f4 KS |
4672 | ptl = pmd_trans_huge_lock(pmd, vma); |
4673 | if (ptl) { | |
12724850 NH |
4674 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4675 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4676 | spin_unlock(ptl); |
1a5a9906 | 4677 | return 0; |
12724850 | 4678 | } |
03319327 | 4679 | |
45f83cef AA |
4680 | if (pmd_trans_unstable(pmd)) |
4681 | return 0; | |
4ffef5fe DN |
4682 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4683 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4684 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4685 | mc.precharge++; /* increment precharge temporarily */ |
4686 | pte_unmap_unlock(pte - 1, ptl); | |
4687 | cond_resched(); | |
4688 | ||
7dc74be0 DN |
4689 | return 0; |
4690 | } | |
4691 | ||
4ffef5fe DN |
4692 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4693 | { | |
4694 | unsigned long precharge; | |
4ffef5fe | 4695 | |
26bcd64a NH |
4696 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4697 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4698 | .mm = mm, | |
4699 | }; | |
dfe076b0 | 4700 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
4701 | walk_page_range(0, mm->highest_vm_end, |
4702 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 4703 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4704 | |
4705 | precharge = mc.precharge; | |
4706 | mc.precharge = 0; | |
4707 | ||
4708 | return precharge; | |
4709 | } | |
4710 | ||
4ffef5fe DN |
4711 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4712 | { | |
dfe076b0 DN |
4713 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4714 | ||
4715 | VM_BUG_ON(mc.moving_task); | |
4716 | mc.moving_task = current; | |
4717 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4718 | } |
4719 | ||
dfe076b0 DN |
4720 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4721 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4722 | { |
2bd9bb20 KH |
4723 | struct mem_cgroup *from = mc.from; |
4724 | struct mem_cgroup *to = mc.to; | |
4725 | ||
4ffef5fe | 4726 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4727 | if (mc.precharge) { |
00501b53 | 4728 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4729 | mc.precharge = 0; |
4730 | } | |
4731 | /* | |
4732 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4733 | * we must uncharge here. | |
4734 | */ | |
4735 | if (mc.moved_charge) { | |
00501b53 | 4736 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4737 | mc.moved_charge = 0; |
4ffef5fe | 4738 | } |
483c30b5 DN |
4739 | /* we must fixup refcnts and charges */ |
4740 | if (mc.moved_swap) { | |
483c30b5 | 4741 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4742 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4743 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4744 | |
615d66c3 VD |
4745 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
4746 | ||
05b84301 | 4747 | /* |
3e32cb2e JW |
4748 | * we charged both to->memory and to->memsw, so we |
4749 | * should uncharge to->memory. | |
05b84301 | 4750 | */ |
ce00a967 | 4751 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4752 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4753 | ||
615d66c3 VD |
4754 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
4755 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 4756 | |
483c30b5 DN |
4757 | mc.moved_swap = 0; |
4758 | } | |
dfe076b0 DN |
4759 | memcg_oom_recover(from); |
4760 | memcg_oom_recover(to); | |
4761 | wake_up_all(&mc.waitq); | |
4762 | } | |
4763 | ||
4764 | static void mem_cgroup_clear_mc(void) | |
4765 | { | |
264a0ae1 TH |
4766 | struct mm_struct *mm = mc.mm; |
4767 | ||
dfe076b0 DN |
4768 | /* |
4769 | * we must clear moving_task before waking up waiters at the end of | |
4770 | * task migration. | |
4771 | */ | |
4772 | mc.moving_task = NULL; | |
4773 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4774 | spin_lock(&mc.lock); |
4ffef5fe DN |
4775 | mc.from = NULL; |
4776 | mc.to = NULL; | |
264a0ae1 | 4777 | mc.mm = NULL; |
2bd9bb20 | 4778 | spin_unlock(&mc.lock); |
264a0ae1 TH |
4779 | |
4780 | mmput(mm); | |
4ffef5fe DN |
4781 | } |
4782 | ||
1f7dd3e5 | 4783 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4784 | { |
1f7dd3e5 | 4785 | struct cgroup_subsys_state *css; |
eed67d75 | 4786 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4787 | struct mem_cgroup *from; |
4530eddb | 4788 | struct task_struct *leader, *p; |
9f2115f9 | 4789 | struct mm_struct *mm; |
1dfab5ab | 4790 | unsigned long move_flags; |
9f2115f9 | 4791 | int ret = 0; |
7dc74be0 | 4792 | |
1f7dd3e5 TH |
4793 | /* charge immigration isn't supported on the default hierarchy */ |
4794 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4795 | return 0; |
4796 | ||
4530eddb TH |
4797 | /* |
4798 | * Multi-process migrations only happen on the default hierarchy | |
4799 | * where charge immigration is not used. Perform charge | |
4800 | * immigration if @tset contains a leader and whine if there are | |
4801 | * multiple. | |
4802 | */ | |
4803 | p = NULL; | |
1f7dd3e5 | 4804 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4805 | WARN_ON_ONCE(p); |
4806 | p = leader; | |
1f7dd3e5 | 4807 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4808 | } |
4809 | if (!p) | |
4810 | return 0; | |
4811 | ||
1f7dd3e5 TH |
4812 | /* |
4813 | * We are now commited to this value whatever it is. Changes in this | |
4814 | * tunable will only affect upcoming migrations, not the current one. | |
4815 | * So we need to save it, and keep it going. | |
4816 | */ | |
4817 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4818 | if (!move_flags) | |
4819 | return 0; | |
4820 | ||
9f2115f9 TH |
4821 | from = mem_cgroup_from_task(p); |
4822 | ||
4823 | VM_BUG_ON(from == memcg); | |
4824 | ||
4825 | mm = get_task_mm(p); | |
4826 | if (!mm) | |
4827 | return 0; | |
4828 | /* We move charges only when we move a owner of the mm */ | |
4829 | if (mm->owner == p) { | |
4830 | VM_BUG_ON(mc.from); | |
4831 | VM_BUG_ON(mc.to); | |
4832 | VM_BUG_ON(mc.precharge); | |
4833 | VM_BUG_ON(mc.moved_charge); | |
4834 | VM_BUG_ON(mc.moved_swap); | |
4835 | ||
4836 | spin_lock(&mc.lock); | |
264a0ae1 | 4837 | mc.mm = mm; |
9f2115f9 TH |
4838 | mc.from = from; |
4839 | mc.to = memcg; | |
4840 | mc.flags = move_flags; | |
4841 | spin_unlock(&mc.lock); | |
4842 | /* We set mc.moving_task later */ | |
4843 | ||
4844 | ret = mem_cgroup_precharge_mc(mm); | |
4845 | if (ret) | |
4846 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
4847 | } else { |
4848 | mmput(mm); | |
7dc74be0 DN |
4849 | } |
4850 | return ret; | |
4851 | } | |
4852 | ||
1f7dd3e5 | 4853 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4854 | { |
4e2f245d JW |
4855 | if (mc.to) |
4856 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4857 | } |
4858 | ||
4ffef5fe DN |
4859 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4860 | unsigned long addr, unsigned long end, | |
4861 | struct mm_walk *walk) | |
7dc74be0 | 4862 | { |
4ffef5fe | 4863 | int ret = 0; |
26bcd64a | 4864 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4865 | pte_t *pte; |
4866 | spinlock_t *ptl; | |
12724850 NH |
4867 | enum mc_target_type target_type; |
4868 | union mc_target target; | |
4869 | struct page *page; | |
4ffef5fe | 4870 | |
b6ec57f4 KS |
4871 | ptl = pmd_trans_huge_lock(pmd, vma); |
4872 | if (ptl) { | |
62ade86a | 4873 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4874 | spin_unlock(ptl); |
12724850 NH |
4875 | return 0; |
4876 | } | |
4877 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4878 | if (target_type == MC_TARGET_PAGE) { | |
4879 | page = target.page; | |
4880 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 4881 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 4882 | mc.from, mc.to)) { |
12724850 NH |
4883 | mc.precharge -= HPAGE_PMD_NR; |
4884 | mc.moved_charge += HPAGE_PMD_NR; | |
4885 | } | |
4886 | putback_lru_page(page); | |
4887 | } | |
4888 | put_page(page); | |
4889 | } | |
bf929152 | 4890 | spin_unlock(ptl); |
1a5a9906 | 4891 | return 0; |
12724850 NH |
4892 | } |
4893 | ||
45f83cef AA |
4894 | if (pmd_trans_unstable(pmd)) |
4895 | return 0; | |
4ffef5fe DN |
4896 | retry: |
4897 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4898 | for (; addr != end; addr += PAGE_SIZE) { | |
4899 | pte_t ptent = *(pte++); | |
02491447 | 4900 | swp_entry_t ent; |
4ffef5fe DN |
4901 | |
4902 | if (!mc.precharge) | |
4903 | break; | |
4904 | ||
8d32ff84 | 4905 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4906 | case MC_TARGET_PAGE: |
4907 | page = target.page; | |
53f9263b KS |
4908 | /* |
4909 | * We can have a part of the split pmd here. Moving it | |
4910 | * can be done but it would be too convoluted so simply | |
4911 | * ignore such a partial THP and keep it in original | |
4912 | * memcg. There should be somebody mapping the head. | |
4913 | */ | |
4914 | if (PageTransCompound(page)) | |
4915 | goto put; | |
4ffef5fe DN |
4916 | if (isolate_lru_page(page)) |
4917 | goto put; | |
f627c2f5 KS |
4918 | if (!mem_cgroup_move_account(page, false, |
4919 | mc.from, mc.to)) { | |
4ffef5fe | 4920 | mc.precharge--; |
854ffa8d DN |
4921 | /* we uncharge from mc.from later. */ |
4922 | mc.moved_charge++; | |
4ffef5fe DN |
4923 | } |
4924 | putback_lru_page(page); | |
8d32ff84 | 4925 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4926 | put_page(page); |
4927 | break; | |
02491447 DN |
4928 | case MC_TARGET_SWAP: |
4929 | ent = target.ent; | |
e91cbb42 | 4930 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4931 | mc.precharge--; |
483c30b5 DN |
4932 | /* we fixup refcnts and charges later. */ |
4933 | mc.moved_swap++; | |
4934 | } | |
02491447 | 4935 | break; |
4ffef5fe DN |
4936 | default: |
4937 | break; | |
4938 | } | |
4939 | } | |
4940 | pte_unmap_unlock(pte - 1, ptl); | |
4941 | cond_resched(); | |
4942 | ||
4943 | if (addr != end) { | |
4944 | /* | |
4945 | * We have consumed all precharges we got in can_attach(). | |
4946 | * We try charge one by one, but don't do any additional | |
4947 | * charges to mc.to if we have failed in charge once in attach() | |
4948 | * phase. | |
4949 | */ | |
854ffa8d | 4950 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4951 | if (!ret) |
4952 | goto retry; | |
4953 | } | |
4954 | ||
4955 | return ret; | |
4956 | } | |
4957 | ||
264a0ae1 | 4958 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 4959 | { |
26bcd64a NH |
4960 | struct mm_walk mem_cgroup_move_charge_walk = { |
4961 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 4962 | .mm = mc.mm, |
26bcd64a | 4963 | }; |
4ffef5fe DN |
4964 | |
4965 | lru_add_drain_all(); | |
312722cb | 4966 | /* |
81f8c3a4 JW |
4967 | * Signal lock_page_memcg() to take the memcg's move_lock |
4968 | * while we're moving its pages to another memcg. Then wait | |
4969 | * for already started RCU-only updates to finish. | |
312722cb JW |
4970 | */ |
4971 | atomic_inc(&mc.from->moving_account); | |
4972 | synchronize_rcu(); | |
dfe076b0 | 4973 | retry: |
264a0ae1 | 4974 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
4975 | /* |
4976 | * Someone who are holding the mmap_sem might be waiting in | |
4977 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4978 | * and retry. Because we cancel precharges, we might not be able | |
4979 | * to move enough charges, but moving charge is a best-effort | |
4980 | * feature anyway, so it wouldn't be a big problem. | |
4981 | */ | |
4982 | __mem_cgroup_clear_mc(); | |
4983 | cond_resched(); | |
4984 | goto retry; | |
4985 | } | |
26bcd64a NH |
4986 | /* |
4987 | * When we have consumed all precharges and failed in doing | |
4988 | * additional charge, the page walk just aborts. | |
4989 | */ | |
0247f3f4 JM |
4990 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
4991 | ||
264a0ae1 | 4992 | up_read(&mc.mm->mmap_sem); |
312722cb | 4993 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
4994 | } |
4995 | ||
264a0ae1 | 4996 | static void mem_cgroup_move_task(void) |
67e465a7 | 4997 | { |
264a0ae1 TH |
4998 | if (mc.to) { |
4999 | mem_cgroup_move_charge(); | |
a433658c | 5000 | mem_cgroup_clear_mc(); |
264a0ae1 | 5001 | } |
67e465a7 | 5002 | } |
5cfb80a7 | 5003 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5004 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5005 | { |
5006 | return 0; | |
5007 | } | |
1f7dd3e5 | 5008 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5009 | { |
5010 | } | |
264a0ae1 | 5011 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5012 | { |
5013 | } | |
5014 | #endif | |
67e465a7 | 5015 | |
f00baae7 TH |
5016 | /* |
5017 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5018 | * to verify whether we're attached to the default hierarchy on each mount |
5019 | * attempt. | |
f00baae7 | 5020 | */ |
eb95419b | 5021 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5022 | { |
5023 | /* | |
aa6ec29b | 5024 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5025 | * guarantees that @root doesn't have any children, so turning it |
5026 | * on for the root memcg is enough. | |
5027 | */ | |
9e10a130 | 5028 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5029 | root_mem_cgroup->use_hierarchy = true; |
5030 | else | |
5031 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5032 | } |
5033 | ||
241994ed JW |
5034 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5035 | struct cftype *cft) | |
5036 | { | |
f5fc3c5d JW |
5037 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5038 | ||
5039 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5040 | } |
5041 | ||
5042 | static int memory_low_show(struct seq_file *m, void *v) | |
5043 | { | |
5044 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5045 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5046 | |
5047 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5048 | seq_puts(m, "max\n"); |
241994ed JW |
5049 | else |
5050 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5051 | ||
5052 | return 0; | |
5053 | } | |
5054 | ||
5055 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5056 | char *buf, size_t nbytes, loff_t off) | |
5057 | { | |
5058 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5059 | unsigned long low; | |
5060 | int err; | |
5061 | ||
5062 | buf = strstrip(buf); | |
d2973697 | 5063 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5064 | if (err) |
5065 | return err; | |
5066 | ||
5067 | memcg->low = low; | |
5068 | ||
5069 | return nbytes; | |
5070 | } | |
5071 | ||
5072 | static int memory_high_show(struct seq_file *m, void *v) | |
5073 | { | |
5074 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5075 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5076 | |
5077 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5078 | seq_puts(m, "max\n"); |
241994ed JW |
5079 | else |
5080 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5081 | ||
5082 | return 0; | |
5083 | } | |
5084 | ||
5085 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5086 | char *buf, size_t nbytes, loff_t off) | |
5087 | { | |
5088 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5089 | unsigned long nr_pages; |
241994ed JW |
5090 | unsigned long high; |
5091 | int err; | |
5092 | ||
5093 | buf = strstrip(buf); | |
d2973697 | 5094 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5095 | if (err) |
5096 | return err; | |
5097 | ||
5098 | memcg->high = high; | |
5099 | ||
588083bb JW |
5100 | nr_pages = page_counter_read(&memcg->memory); |
5101 | if (nr_pages > high) | |
5102 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5103 | GFP_KERNEL, true); | |
5104 | ||
2529bb3a | 5105 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5106 | return nbytes; |
5107 | } | |
5108 | ||
5109 | static int memory_max_show(struct seq_file *m, void *v) | |
5110 | { | |
5111 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5112 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5113 | |
5114 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5115 | seq_puts(m, "max\n"); |
241994ed JW |
5116 | else |
5117 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5118 | ||
5119 | return 0; | |
5120 | } | |
5121 | ||
5122 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5123 | char *buf, size_t nbytes, loff_t off) | |
5124 | { | |
5125 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5126 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5127 | bool drained = false; | |
241994ed JW |
5128 | unsigned long max; |
5129 | int err; | |
5130 | ||
5131 | buf = strstrip(buf); | |
d2973697 | 5132 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5133 | if (err) |
5134 | return err; | |
5135 | ||
b6e6edcf JW |
5136 | xchg(&memcg->memory.limit, max); |
5137 | ||
5138 | for (;;) { | |
5139 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5140 | ||
5141 | if (nr_pages <= max) | |
5142 | break; | |
5143 | ||
5144 | if (signal_pending(current)) { | |
5145 | err = -EINTR; | |
5146 | break; | |
5147 | } | |
5148 | ||
5149 | if (!drained) { | |
5150 | drain_all_stock(memcg); | |
5151 | drained = true; | |
5152 | continue; | |
5153 | } | |
5154 | ||
5155 | if (nr_reclaims) { | |
5156 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5157 | GFP_KERNEL, true)) | |
5158 | nr_reclaims--; | |
5159 | continue; | |
5160 | } | |
5161 | ||
5162 | mem_cgroup_events(memcg, MEMCG_OOM, 1); | |
5163 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) | |
5164 | break; | |
5165 | } | |
241994ed | 5166 | |
2529bb3a | 5167 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5168 | return nbytes; |
5169 | } | |
5170 | ||
5171 | static int memory_events_show(struct seq_file *m, void *v) | |
5172 | { | |
5173 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5174 | ||
5175 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5176 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5177 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5178 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5179 | ||
5180 | return 0; | |
5181 | } | |
5182 | ||
587d9f72 JW |
5183 | static int memory_stat_show(struct seq_file *m, void *v) |
5184 | { | |
5185 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
72b54e73 VD |
5186 | unsigned long stat[MEMCG_NR_STAT]; |
5187 | unsigned long events[MEMCG_NR_EVENTS]; | |
587d9f72 JW |
5188 | int i; |
5189 | ||
5190 | /* | |
5191 | * Provide statistics on the state of the memory subsystem as | |
5192 | * well as cumulative event counters that show past behavior. | |
5193 | * | |
5194 | * This list is ordered following a combination of these gradients: | |
5195 | * 1) generic big picture -> specifics and details | |
5196 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5197 | * | |
5198 | * Current memory state: | |
5199 | */ | |
5200 | ||
72b54e73 VD |
5201 | tree_stat(memcg, stat); |
5202 | tree_events(memcg, events); | |
5203 | ||
587d9f72 | 5204 | seq_printf(m, "anon %llu\n", |
72b54e73 | 5205 | (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE); |
587d9f72 | 5206 | seq_printf(m, "file %llu\n", |
72b54e73 | 5207 | (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE); |
12580e4b | 5208 | seq_printf(m, "kernel_stack %llu\n", |
efdc9490 | 5209 | (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024); |
27ee57c9 VD |
5210 | seq_printf(m, "slab %llu\n", |
5211 | (u64)(stat[MEMCG_SLAB_RECLAIMABLE] + | |
5212 | stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5213 | seq_printf(m, "sock %llu\n", |
72b54e73 | 5214 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 | 5215 | |
9a4caf1e JW |
5216 | seq_printf(m, "shmem %llu\n", |
5217 | (u64)stat[MEM_CGROUP_STAT_SHMEM] * PAGE_SIZE); | |
587d9f72 | 5218 | seq_printf(m, "file_mapped %llu\n", |
72b54e73 | 5219 | (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5220 | seq_printf(m, "file_dirty %llu\n", |
72b54e73 | 5221 | (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE); |
587d9f72 | 5222 | seq_printf(m, "file_writeback %llu\n", |
72b54e73 | 5223 | (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE); |
587d9f72 JW |
5224 | |
5225 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5226 | struct mem_cgroup *mi; | |
5227 | unsigned long val = 0; | |
5228 | ||
5229 | for_each_mem_cgroup_tree(mi, memcg) | |
5230 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
5231 | seq_printf(m, "%s %llu\n", | |
5232 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); | |
5233 | } | |
5234 | ||
27ee57c9 VD |
5235 | seq_printf(m, "slab_reclaimable %llu\n", |
5236 | (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE); | |
5237 | seq_printf(m, "slab_unreclaimable %llu\n", | |
5238 | (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE); | |
5239 | ||
587d9f72 JW |
5240 | /* Accumulated memory events */ |
5241 | ||
5242 | seq_printf(m, "pgfault %lu\n", | |
72b54e73 | 5243 | events[MEM_CGROUP_EVENTS_PGFAULT]); |
587d9f72 | 5244 | seq_printf(m, "pgmajfault %lu\n", |
72b54e73 | 5245 | events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
587d9f72 JW |
5246 | |
5247 | return 0; | |
5248 | } | |
5249 | ||
241994ed JW |
5250 | static struct cftype memory_files[] = { |
5251 | { | |
5252 | .name = "current", | |
f5fc3c5d | 5253 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5254 | .read_u64 = memory_current_read, |
5255 | }, | |
5256 | { | |
5257 | .name = "low", | |
5258 | .flags = CFTYPE_NOT_ON_ROOT, | |
5259 | .seq_show = memory_low_show, | |
5260 | .write = memory_low_write, | |
5261 | }, | |
5262 | { | |
5263 | .name = "high", | |
5264 | .flags = CFTYPE_NOT_ON_ROOT, | |
5265 | .seq_show = memory_high_show, | |
5266 | .write = memory_high_write, | |
5267 | }, | |
5268 | { | |
5269 | .name = "max", | |
5270 | .flags = CFTYPE_NOT_ON_ROOT, | |
5271 | .seq_show = memory_max_show, | |
5272 | .write = memory_max_write, | |
5273 | }, | |
5274 | { | |
5275 | .name = "events", | |
5276 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5277 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5278 | .seq_show = memory_events_show, |
5279 | }, | |
587d9f72 JW |
5280 | { |
5281 | .name = "stat", | |
5282 | .flags = CFTYPE_NOT_ON_ROOT, | |
5283 | .seq_show = memory_stat_show, | |
5284 | }, | |
241994ed JW |
5285 | { } /* terminate */ |
5286 | }; | |
5287 | ||
073219e9 | 5288 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5289 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5290 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5291 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5292 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5293 | .css_free = mem_cgroup_css_free, |
1ced953b | 5294 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5295 | .can_attach = mem_cgroup_can_attach, |
5296 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5297 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5298 | .bind = mem_cgroup_bind, |
241994ed JW |
5299 | .dfl_cftypes = memory_files, |
5300 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5301 | .early_init = 0, |
8cdea7c0 | 5302 | }; |
c077719b | 5303 | |
241994ed JW |
5304 | /** |
5305 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5306 | * @root: the highest ancestor to consider | |
5307 | * @memcg: the memory cgroup to check | |
5308 | * | |
5309 | * Returns %true if memory consumption of @memcg, and that of all | |
5310 | * configurable ancestors up to @root, is below the normal range. | |
5311 | */ | |
5312 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5313 | { | |
5314 | if (mem_cgroup_disabled()) | |
5315 | return false; | |
5316 | ||
5317 | /* | |
5318 | * The toplevel group doesn't have a configurable range, so | |
5319 | * it's never low when looked at directly, and it is not | |
5320 | * considered an ancestor when assessing the hierarchy. | |
5321 | */ | |
5322 | ||
5323 | if (memcg == root_mem_cgroup) | |
5324 | return false; | |
5325 | ||
4e54dede | 5326 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5327 | return false; |
5328 | ||
5329 | while (memcg != root) { | |
5330 | memcg = parent_mem_cgroup(memcg); | |
5331 | ||
5332 | if (memcg == root_mem_cgroup) | |
5333 | break; | |
5334 | ||
4e54dede | 5335 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5336 | return false; |
5337 | } | |
5338 | return true; | |
5339 | } | |
5340 | ||
00501b53 JW |
5341 | /** |
5342 | * mem_cgroup_try_charge - try charging a page | |
5343 | * @page: page to charge | |
5344 | * @mm: mm context of the victim | |
5345 | * @gfp_mask: reclaim mode | |
5346 | * @memcgp: charged memcg return | |
25843c2b | 5347 | * @compound: charge the page as compound or small page |
00501b53 JW |
5348 | * |
5349 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5350 | * pages according to @gfp_mask if necessary. | |
5351 | * | |
5352 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5353 | * Otherwise, an error code is returned. | |
5354 | * | |
5355 | * After page->mapping has been set up, the caller must finalize the | |
5356 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5357 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5358 | */ | |
5359 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5360 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5361 | bool compound) | |
00501b53 JW |
5362 | { |
5363 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5364 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5365 | int ret = 0; |
5366 | ||
5367 | if (mem_cgroup_disabled()) | |
5368 | goto out; | |
5369 | ||
5370 | if (PageSwapCache(page)) { | |
00501b53 JW |
5371 | /* |
5372 | * Every swap fault against a single page tries to charge the | |
5373 | * page, bail as early as possible. shmem_unuse() encounters | |
5374 | * already charged pages, too. The USED bit is protected by | |
5375 | * the page lock, which serializes swap cache removal, which | |
5376 | * in turn serializes uncharging. | |
5377 | */ | |
e993d905 | 5378 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5379 | if (page->mem_cgroup) |
00501b53 | 5380 | goto out; |
e993d905 | 5381 | |
37e84351 | 5382 | if (do_swap_account) { |
e993d905 VD |
5383 | swp_entry_t ent = { .val = page_private(page), }; |
5384 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5385 | ||
5386 | rcu_read_lock(); | |
5387 | memcg = mem_cgroup_from_id(id); | |
5388 | if (memcg && !css_tryget_online(&memcg->css)) | |
5389 | memcg = NULL; | |
5390 | rcu_read_unlock(); | |
5391 | } | |
00501b53 JW |
5392 | } |
5393 | ||
00501b53 JW |
5394 | if (!memcg) |
5395 | memcg = get_mem_cgroup_from_mm(mm); | |
5396 | ||
5397 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5398 | ||
5399 | css_put(&memcg->css); | |
00501b53 JW |
5400 | out: |
5401 | *memcgp = memcg; | |
5402 | return ret; | |
5403 | } | |
5404 | ||
5405 | /** | |
5406 | * mem_cgroup_commit_charge - commit a page charge | |
5407 | * @page: page to charge | |
5408 | * @memcg: memcg to charge the page to | |
5409 | * @lrucare: page might be on LRU already | |
25843c2b | 5410 | * @compound: charge the page as compound or small page |
00501b53 JW |
5411 | * |
5412 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5413 | * after page->mapping has been set up. This must happen atomically | |
5414 | * as part of the page instantiation, i.e. under the page table lock | |
5415 | * for anonymous pages, under the page lock for page and swap cache. | |
5416 | * | |
5417 | * In addition, the page must not be on the LRU during the commit, to | |
5418 | * prevent racing with task migration. If it might be, use @lrucare. | |
5419 | * | |
5420 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5421 | */ | |
5422 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5423 | bool lrucare, bool compound) |
00501b53 | 5424 | { |
f627c2f5 | 5425 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5426 | |
5427 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5428 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5429 | ||
5430 | if (mem_cgroup_disabled()) | |
5431 | return; | |
5432 | /* | |
5433 | * Swap faults will attempt to charge the same page multiple | |
5434 | * times. But reuse_swap_page() might have removed the page | |
5435 | * from swapcache already, so we can't check PageSwapCache(). | |
5436 | */ | |
5437 | if (!memcg) | |
5438 | return; | |
5439 | ||
6abb5a86 JW |
5440 | commit_charge(page, memcg, lrucare); |
5441 | ||
6abb5a86 | 5442 | local_irq_disable(); |
f627c2f5 | 5443 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5444 | memcg_check_events(memcg, page); |
5445 | local_irq_enable(); | |
00501b53 | 5446 | |
7941d214 | 5447 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5448 | swp_entry_t entry = { .val = page_private(page) }; |
5449 | /* | |
5450 | * The swap entry might not get freed for a long time, | |
5451 | * let's not wait for it. The page already received a | |
5452 | * memory+swap charge, drop the swap entry duplicate. | |
5453 | */ | |
5454 | mem_cgroup_uncharge_swap(entry); | |
5455 | } | |
5456 | } | |
5457 | ||
5458 | /** | |
5459 | * mem_cgroup_cancel_charge - cancel a page charge | |
5460 | * @page: page to charge | |
5461 | * @memcg: memcg to charge the page to | |
25843c2b | 5462 | * @compound: charge the page as compound or small page |
00501b53 JW |
5463 | * |
5464 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5465 | */ | |
f627c2f5 KS |
5466 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5467 | bool compound) | |
00501b53 | 5468 | { |
f627c2f5 | 5469 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5470 | |
5471 | if (mem_cgroup_disabled()) | |
5472 | return; | |
5473 | /* | |
5474 | * Swap faults will attempt to charge the same page multiple | |
5475 | * times. But reuse_swap_page() might have removed the page | |
5476 | * from swapcache already, so we can't check PageSwapCache(). | |
5477 | */ | |
5478 | if (!memcg) | |
5479 | return; | |
5480 | ||
00501b53 JW |
5481 | cancel_charge(memcg, nr_pages); |
5482 | } | |
5483 | ||
747db954 | 5484 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 | 5485 | unsigned long nr_anon, unsigned long nr_file, |
9a4caf1e JW |
5486 | unsigned long nr_kmem, unsigned long nr_huge, |
5487 | unsigned long nr_shmem, struct page *dummy_page) | |
747db954 | 5488 | { |
5e8d35f8 | 5489 | unsigned long nr_pages = nr_anon + nr_file + nr_kmem; |
747db954 JW |
5490 | unsigned long flags; |
5491 | ||
ce00a967 | 5492 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 | 5493 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 5494 | if (do_memsw_account()) |
18eca2e6 | 5495 | page_counter_uncharge(&memcg->memsw, nr_pages); |
5e8d35f8 VD |
5496 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem) |
5497 | page_counter_uncharge(&memcg->kmem, nr_kmem); | |
ce00a967 JW |
5498 | memcg_oom_recover(memcg); |
5499 | } | |
747db954 JW |
5500 | |
5501 | local_irq_save(flags); | |
5502 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5503 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5504 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
9a4caf1e | 5505 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_SHMEM], nr_shmem); |
747db954 | 5506 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); |
18eca2e6 | 5507 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5508 | memcg_check_events(memcg, dummy_page); |
5509 | local_irq_restore(flags); | |
e8ea14cc JW |
5510 | |
5511 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5512 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5513 | } |
5514 | ||
5515 | static void uncharge_list(struct list_head *page_list) | |
5516 | { | |
5517 | struct mem_cgroup *memcg = NULL; | |
9a4caf1e | 5518 | unsigned long nr_shmem = 0; |
747db954 JW |
5519 | unsigned long nr_anon = 0; |
5520 | unsigned long nr_file = 0; | |
5521 | unsigned long nr_huge = 0; | |
5e8d35f8 | 5522 | unsigned long nr_kmem = 0; |
747db954 | 5523 | unsigned long pgpgout = 0; |
747db954 JW |
5524 | struct list_head *next; |
5525 | struct page *page; | |
5526 | ||
8b592656 JW |
5527 | /* |
5528 | * Note that the list can be a single page->lru; hence the | |
5529 | * do-while loop instead of a simple list_for_each_entry(). | |
5530 | */ | |
747db954 JW |
5531 | next = page_list->next; |
5532 | do { | |
747db954 JW |
5533 | page = list_entry(next, struct page, lru); |
5534 | next = page->lru.next; | |
5535 | ||
5536 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5537 | VM_BUG_ON_PAGE(page_count(page), page); | |
5538 | ||
1306a85a | 5539 | if (!page->mem_cgroup) |
747db954 JW |
5540 | continue; |
5541 | ||
5542 | /* | |
5543 | * Nobody should be changing or seriously looking at | |
1306a85a | 5544 | * page->mem_cgroup at this point, we have fully |
29833315 | 5545 | * exclusive access to the page. |
747db954 JW |
5546 | */ |
5547 | ||
1306a85a | 5548 | if (memcg != page->mem_cgroup) { |
747db954 | 5549 | if (memcg) { |
18eca2e6 | 5550 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
9a4caf1e JW |
5551 | nr_kmem, nr_huge, nr_shmem, page); |
5552 | pgpgout = nr_anon = nr_file = nr_kmem = 0; | |
5553 | nr_huge = nr_shmem = 0; | |
747db954 | 5554 | } |
1306a85a | 5555 | memcg = page->mem_cgroup; |
747db954 JW |
5556 | } |
5557 | ||
5e8d35f8 VD |
5558 | if (!PageKmemcg(page)) { |
5559 | unsigned int nr_pages = 1; | |
747db954 | 5560 | |
5e8d35f8 VD |
5561 | if (PageTransHuge(page)) { |
5562 | nr_pages <<= compound_order(page); | |
5e8d35f8 VD |
5563 | nr_huge += nr_pages; |
5564 | } | |
5565 | if (PageAnon(page)) | |
5566 | nr_anon += nr_pages; | |
9a4caf1e | 5567 | else { |
5e8d35f8 | 5568 | nr_file += nr_pages; |
9a4caf1e JW |
5569 | if (PageSwapBacked(page)) |
5570 | nr_shmem += nr_pages; | |
5571 | } | |
5e8d35f8 | 5572 | pgpgout++; |
c4159a75 | 5573 | } else { |
5e8d35f8 | 5574 | nr_kmem += 1 << compound_order(page); |
c4159a75 VD |
5575 | __ClearPageKmemcg(page); |
5576 | } | |
747db954 | 5577 | |
1306a85a | 5578 | page->mem_cgroup = NULL; |
747db954 JW |
5579 | } while (next != page_list); |
5580 | ||
5581 | if (memcg) | |
18eca2e6 | 5582 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
9a4caf1e | 5583 | nr_kmem, nr_huge, nr_shmem, page); |
747db954 JW |
5584 | } |
5585 | ||
0a31bc97 JW |
5586 | /** |
5587 | * mem_cgroup_uncharge - uncharge a page | |
5588 | * @page: page to uncharge | |
5589 | * | |
5590 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5591 | * mem_cgroup_commit_charge(). | |
5592 | */ | |
5593 | void mem_cgroup_uncharge(struct page *page) | |
5594 | { | |
0a31bc97 JW |
5595 | if (mem_cgroup_disabled()) |
5596 | return; | |
5597 | ||
747db954 | 5598 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5599 | if (!page->mem_cgroup) |
0a31bc97 JW |
5600 | return; |
5601 | ||
747db954 JW |
5602 | INIT_LIST_HEAD(&page->lru); |
5603 | uncharge_list(&page->lru); | |
5604 | } | |
0a31bc97 | 5605 | |
747db954 JW |
5606 | /** |
5607 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5608 | * @page_list: list of pages to uncharge | |
5609 | * | |
5610 | * Uncharge a list of pages previously charged with | |
5611 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5612 | */ | |
5613 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5614 | { | |
5615 | if (mem_cgroup_disabled()) | |
5616 | return; | |
0a31bc97 | 5617 | |
747db954 JW |
5618 | if (!list_empty(page_list)) |
5619 | uncharge_list(page_list); | |
0a31bc97 JW |
5620 | } |
5621 | ||
5622 | /** | |
6a93ca8f JW |
5623 | * mem_cgroup_migrate - charge a page's replacement |
5624 | * @oldpage: currently circulating page | |
5625 | * @newpage: replacement page | |
0a31bc97 | 5626 | * |
6a93ca8f JW |
5627 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
5628 | * be uncharged upon free. | |
0a31bc97 JW |
5629 | * |
5630 | * Both pages must be locked, @newpage->mapping must be set up. | |
5631 | */ | |
6a93ca8f | 5632 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5633 | { |
29833315 | 5634 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
5635 | unsigned int nr_pages; |
5636 | bool compound; | |
d93c4130 | 5637 | unsigned long flags; |
0a31bc97 JW |
5638 | |
5639 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5640 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5641 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5642 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5643 | newpage); | |
0a31bc97 JW |
5644 | |
5645 | if (mem_cgroup_disabled()) | |
5646 | return; | |
5647 | ||
5648 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5649 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5650 | return; |
5651 | ||
45637bab | 5652 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5653 | memcg = oldpage->mem_cgroup; |
29833315 | 5654 | if (!memcg) |
0a31bc97 JW |
5655 | return; |
5656 | ||
44b7a8d3 JW |
5657 | /* Force-charge the new page. The old one will be freed soon */ |
5658 | compound = PageTransHuge(newpage); | |
5659 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
5660 | ||
5661 | page_counter_charge(&memcg->memory, nr_pages); | |
5662 | if (do_memsw_account()) | |
5663 | page_counter_charge(&memcg->memsw, nr_pages); | |
5664 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 5665 | |
9cf7666a | 5666 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 5667 | |
d93c4130 | 5668 | local_irq_save(flags); |
44b7a8d3 JW |
5669 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
5670 | memcg_check_events(memcg, newpage); | |
d93c4130 | 5671 | local_irq_restore(flags); |
0a31bc97 JW |
5672 | } |
5673 | ||
ef12947c | 5674 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5675 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5676 | ||
2d758073 | 5677 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
5678 | { |
5679 | struct mem_cgroup *memcg; | |
5680 | ||
2d758073 JW |
5681 | if (!mem_cgroup_sockets_enabled) |
5682 | return; | |
5683 | ||
5684 | /* | |
5685 | * Socket cloning can throw us here with sk_memcg already | |
11092087 JW |
5686 | * filled. It won't however, necessarily happen from |
5687 | * process context. So the test for root memcg given | |
5688 | * the current task's memcg won't help us in this case. | |
5689 | * | |
5690 | * Respecting the original socket's memcg is a better | |
5691 | * decision in this case. | |
5692 | */ | |
5693 | if (sk->sk_memcg) { | |
5694 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
5695 | css_get(&sk->sk_memcg->css); | |
5696 | return; | |
5697 | } | |
5698 | ||
5699 | rcu_read_lock(); | |
5700 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5701 | if (memcg == root_mem_cgroup) |
5702 | goto out; | |
0db15298 | 5703 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 5704 | goto out; |
f7e1cb6e | 5705 | if (css_tryget_online(&memcg->css)) |
11092087 | 5706 | sk->sk_memcg = memcg; |
f7e1cb6e | 5707 | out: |
11092087 JW |
5708 | rcu_read_unlock(); |
5709 | } | |
11092087 | 5710 | |
2d758073 | 5711 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 5712 | { |
2d758073 JW |
5713 | if (sk->sk_memcg) |
5714 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
5715 | } |
5716 | ||
5717 | /** | |
5718 | * mem_cgroup_charge_skmem - charge socket memory | |
5719 | * @memcg: memcg to charge | |
5720 | * @nr_pages: number of pages to charge | |
5721 | * | |
5722 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5723 | * @memcg's configured limit, %false if the charge had to be forced. | |
5724 | */ | |
5725 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5726 | { | |
f7e1cb6e | 5727 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5728 | |
f7e1cb6e | 5729 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5730 | struct page_counter *fail; |
f7e1cb6e | 5731 | |
0db15298 JW |
5732 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
5733 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
5734 | return true; |
5735 | } | |
0db15298 JW |
5736 | page_counter_charge(&memcg->tcpmem, nr_pages); |
5737 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 5738 | return false; |
11092087 | 5739 | } |
d886f4e4 | 5740 | |
f7e1cb6e JW |
5741 | /* Don't block in the packet receive path */ |
5742 | if (in_softirq()) | |
5743 | gfp_mask = GFP_NOWAIT; | |
5744 | ||
b2807f07 JW |
5745 | this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5746 | ||
f7e1cb6e JW |
5747 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
5748 | return true; | |
5749 | ||
5750 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5751 | return false; |
5752 | } | |
5753 | ||
5754 | /** | |
5755 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5756 | * @memcg - memcg to uncharge | |
5757 | * @nr_pages - number of pages to uncharge | |
5758 | */ | |
5759 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5760 | { | |
f7e1cb6e | 5761 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5762 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
5763 | return; |
5764 | } | |
d886f4e4 | 5765 | |
b2807f07 JW |
5766 | this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5767 | ||
f7e1cb6e JW |
5768 | page_counter_uncharge(&memcg->memory, nr_pages); |
5769 | css_put_many(&memcg->css, nr_pages); | |
11092087 JW |
5770 | } |
5771 | ||
f7e1cb6e JW |
5772 | static int __init cgroup_memory(char *s) |
5773 | { | |
5774 | char *token; | |
5775 | ||
5776 | while ((token = strsep(&s, ",")) != NULL) { | |
5777 | if (!*token) | |
5778 | continue; | |
5779 | if (!strcmp(token, "nosocket")) | |
5780 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5781 | if (!strcmp(token, "nokmem")) |
5782 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5783 | } |
5784 | return 0; | |
5785 | } | |
5786 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5787 | |
2d11085e | 5788 | /* |
1081312f MH |
5789 | * subsys_initcall() for memory controller. |
5790 | * | |
308167fc SAS |
5791 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
5792 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
5793 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
5794 | * should be initialized from here. | |
2d11085e MH |
5795 | */ |
5796 | static int __init mem_cgroup_init(void) | |
5797 | { | |
95a045f6 JW |
5798 | int cpu, node; |
5799 | ||
13583c3d VD |
5800 | #ifndef CONFIG_SLOB |
5801 | /* | |
5802 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
5803 | * so use a workqueue with limited concurrency to avoid stalling |
5804 | * all worker threads in case lots of cgroups are created and | |
5805 | * destroyed simultaneously. | |
13583c3d | 5806 | */ |
17cc4dfe TH |
5807 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
5808 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
5809 | #endif |
5810 | ||
308167fc SAS |
5811 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
5812 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
5813 | |
5814 | for_each_possible_cpu(cpu) | |
5815 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5816 | drain_local_stock); | |
5817 | ||
5818 | for_each_node(node) { | |
5819 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
5820 | |
5821 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5822 | node_online(node) ? node : NUMA_NO_NODE); | |
5823 | ||
ef8f2327 MG |
5824 | rtpn->rb_root = RB_ROOT; |
5825 | spin_lock_init(&rtpn->lock); | |
95a045f6 JW |
5826 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
5827 | } | |
5828 | ||
2d11085e MH |
5829 | return 0; |
5830 | } | |
5831 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5832 | |
5833 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
5834 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
5835 | { | |
5836 | while (!atomic_inc_not_zero(&memcg->id.ref)) { | |
5837 | /* | |
5838 | * The root cgroup cannot be destroyed, so it's refcount must | |
5839 | * always be >= 1. | |
5840 | */ | |
5841 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
5842 | VM_BUG_ON(1); | |
5843 | break; | |
5844 | } | |
5845 | memcg = parent_mem_cgroup(memcg); | |
5846 | if (!memcg) | |
5847 | memcg = root_mem_cgroup; | |
5848 | } | |
5849 | return memcg; | |
5850 | } | |
5851 | ||
21afa38e JW |
5852 | /** |
5853 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5854 | * @page: page whose memsw charge to transfer | |
5855 | * @entry: swap entry to move the charge to | |
5856 | * | |
5857 | * Transfer the memsw charge of @page to @entry. | |
5858 | */ | |
5859 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5860 | { | |
1f47b61f | 5861 | struct mem_cgroup *memcg, *swap_memcg; |
21afa38e JW |
5862 | unsigned short oldid; |
5863 | ||
5864 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5865 | VM_BUG_ON_PAGE(page_count(page), page); | |
5866 | ||
7941d214 | 5867 | if (!do_memsw_account()) |
21afa38e JW |
5868 | return; |
5869 | ||
5870 | memcg = page->mem_cgroup; | |
5871 | ||
5872 | /* Readahead page, never charged */ | |
5873 | if (!memcg) | |
5874 | return; | |
5875 | ||
1f47b61f VD |
5876 | /* |
5877 | * In case the memcg owning these pages has been offlined and doesn't | |
5878 | * have an ID allocated to it anymore, charge the closest online | |
5879 | * ancestor for the swap instead and transfer the memory+swap charge. | |
5880 | */ | |
5881 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
5882 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg)); | |
21afa38e | 5883 | VM_BUG_ON_PAGE(oldid, page); |
1f47b61f | 5884 | mem_cgroup_swap_statistics(swap_memcg, true); |
21afa38e JW |
5885 | |
5886 | page->mem_cgroup = NULL; | |
5887 | ||
5888 | if (!mem_cgroup_is_root(memcg)) | |
5889 | page_counter_uncharge(&memcg->memory, 1); | |
5890 | ||
1f47b61f VD |
5891 | if (memcg != swap_memcg) { |
5892 | if (!mem_cgroup_is_root(swap_memcg)) | |
5893 | page_counter_charge(&swap_memcg->memsw, 1); | |
5894 | page_counter_uncharge(&memcg->memsw, 1); | |
5895 | } | |
5896 | ||
ce9ce665 SAS |
5897 | /* |
5898 | * Interrupts should be disabled here because the caller holds the | |
5899 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5900 | * important here to have the interrupts disabled because it is the | |
5901 | * only synchronisation we have for udpating the per-CPU variables. | |
5902 | */ | |
5903 | VM_BUG_ON(!irqs_disabled()); | |
f627c2f5 | 5904 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
21afa38e | 5905 | memcg_check_events(memcg, page); |
73f576c0 JW |
5906 | |
5907 | if (!mem_cgroup_is_root(memcg)) | |
5908 | css_put(&memcg->css); | |
21afa38e JW |
5909 | } |
5910 | ||
37e84351 VD |
5911 | /* |
5912 | * mem_cgroup_try_charge_swap - try charging a swap entry | |
5913 | * @page: page being added to swap | |
5914 | * @entry: swap entry to charge | |
5915 | * | |
5916 | * Try to charge @entry to the memcg that @page belongs to. | |
5917 | * | |
5918 | * Returns 0 on success, -ENOMEM on failure. | |
5919 | */ | |
5920 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
5921 | { | |
5922 | struct mem_cgroup *memcg; | |
5923 | struct page_counter *counter; | |
5924 | unsigned short oldid; | |
5925 | ||
5926 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
5927 | return 0; | |
5928 | ||
5929 | memcg = page->mem_cgroup; | |
5930 | ||
5931 | /* Readahead page, never charged */ | |
5932 | if (!memcg) | |
5933 | return 0; | |
5934 | ||
1f47b61f VD |
5935 | memcg = mem_cgroup_id_get_online(memcg); |
5936 | ||
37e84351 | 5937 | if (!mem_cgroup_is_root(memcg) && |
1f47b61f VD |
5938 | !page_counter_try_charge(&memcg->swap, 1, &counter)) { |
5939 | mem_cgroup_id_put(memcg); | |
37e84351 | 5940 | return -ENOMEM; |
1f47b61f | 5941 | } |
37e84351 VD |
5942 | |
5943 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5944 | VM_BUG_ON_PAGE(oldid, page); | |
5945 | mem_cgroup_swap_statistics(memcg, true); | |
5946 | ||
37e84351 VD |
5947 | return 0; |
5948 | } | |
5949 | ||
21afa38e JW |
5950 | /** |
5951 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5952 | * @entry: swap entry to uncharge | |
5953 | * | |
37e84351 | 5954 | * Drop the swap charge associated with @entry. |
21afa38e JW |
5955 | */ |
5956 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5957 | { | |
5958 | struct mem_cgroup *memcg; | |
5959 | unsigned short id; | |
5960 | ||
37e84351 | 5961 | if (!do_swap_account) |
21afa38e JW |
5962 | return; |
5963 | ||
5964 | id = swap_cgroup_record(entry, 0); | |
5965 | rcu_read_lock(); | |
adbe427b | 5966 | memcg = mem_cgroup_from_id(id); |
21afa38e | 5967 | if (memcg) { |
37e84351 VD |
5968 | if (!mem_cgroup_is_root(memcg)) { |
5969 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5970 | page_counter_uncharge(&memcg->swap, 1); | |
5971 | else | |
5972 | page_counter_uncharge(&memcg->memsw, 1); | |
5973 | } | |
21afa38e | 5974 | mem_cgroup_swap_statistics(memcg, false); |
73f576c0 | 5975 | mem_cgroup_id_put(memcg); |
21afa38e JW |
5976 | } |
5977 | rcu_read_unlock(); | |
5978 | } | |
5979 | ||
d8b38438 VD |
5980 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
5981 | { | |
5982 | long nr_swap_pages = get_nr_swap_pages(); | |
5983 | ||
5984 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5985 | return nr_swap_pages; | |
5986 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5987 | nr_swap_pages = min_t(long, nr_swap_pages, | |
5988 | READ_ONCE(memcg->swap.limit) - | |
5989 | page_counter_read(&memcg->swap)); | |
5990 | return nr_swap_pages; | |
5991 | } | |
5992 | ||
5ccc5aba VD |
5993 | bool mem_cgroup_swap_full(struct page *page) |
5994 | { | |
5995 | struct mem_cgroup *memcg; | |
5996 | ||
5997 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
5998 | ||
5999 | if (vm_swap_full()) | |
6000 | return true; | |
6001 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6002 | return false; | |
6003 | ||
6004 | memcg = page->mem_cgroup; | |
6005 | if (!memcg) | |
6006 | return false; | |
6007 | ||
6008 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6009 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) | |
6010 | return true; | |
6011 | ||
6012 | return false; | |
6013 | } | |
6014 | ||
21afa38e JW |
6015 | /* for remember boot option*/ |
6016 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6017 | static int really_do_swap_account __initdata = 1; | |
6018 | #else | |
6019 | static int really_do_swap_account __initdata; | |
6020 | #endif | |
6021 | ||
6022 | static int __init enable_swap_account(char *s) | |
6023 | { | |
6024 | if (!strcmp(s, "1")) | |
6025 | really_do_swap_account = 1; | |
6026 | else if (!strcmp(s, "0")) | |
6027 | really_do_swap_account = 0; | |
6028 | return 1; | |
6029 | } | |
6030 | __setup("swapaccount=", enable_swap_account); | |
6031 | ||
37e84351 VD |
6032 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6033 | struct cftype *cft) | |
6034 | { | |
6035 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6036 | ||
6037 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6038 | } | |
6039 | ||
6040 | static int swap_max_show(struct seq_file *m, void *v) | |
6041 | { | |
6042 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
6043 | unsigned long max = READ_ONCE(memcg->swap.limit); | |
6044 | ||
6045 | if (max == PAGE_COUNTER_MAX) | |
6046 | seq_puts(m, "max\n"); | |
6047 | else | |
6048 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
6049 | ||
6050 | return 0; | |
6051 | } | |
6052 | ||
6053 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6054 | char *buf, size_t nbytes, loff_t off) | |
6055 | { | |
6056 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6057 | unsigned long max; | |
6058 | int err; | |
6059 | ||
6060 | buf = strstrip(buf); | |
6061 | err = page_counter_memparse(buf, "max", &max); | |
6062 | if (err) | |
6063 | return err; | |
6064 | ||
6065 | mutex_lock(&memcg_limit_mutex); | |
6066 | err = page_counter_limit(&memcg->swap, max); | |
6067 | mutex_unlock(&memcg_limit_mutex); | |
6068 | if (err) | |
6069 | return err; | |
6070 | ||
6071 | return nbytes; | |
6072 | } | |
6073 | ||
6074 | static struct cftype swap_files[] = { | |
6075 | { | |
6076 | .name = "swap.current", | |
6077 | .flags = CFTYPE_NOT_ON_ROOT, | |
6078 | .read_u64 = swap_current_read, | |
6079 | }, | |
6080 | { | |
6081 | .name = "swap.max", | |
6082 | .flags = CFTYPE_NOT_ON_ROOT, | |
6083 | .seq_show = swap_max_show, | |
6084 | .write = swap_max_write, | |
6085 | }, | |
6086 | { } /* terminate */ | |
6087 | }; | |
6088 | ||
21afa38e JW |
6089 | static struct cftype memsw_cgroup_files[] = { |
6090 | { | |
6091 | .name = "memsw.usage_in_bytes", | |
6092 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6093 | .read_u64 = mem_cgroup_read_u64, | |
6094 | }, | |
6095 | { | |
6096 | .name = "memsw.max_usage_in_bytes", | |
6097 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6098 | .write = mem_cgroup_reset, | |
6099 | .read_u64 = mem_cgroup_read_u64, | |
6100 | }, | |
6101 | { | |
6102 | .name = "memsw.limit_in_bytes", | |
6103 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6104 | .write = mem_cgroup_write, | |
6105 | .read_u64 = mem_cgroup_read_u64, | |
6106 | }, | |
6107 | { | |
6108 | .name = "memsw.failcnt", | |
6109 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6110 | .write = mem_cgroup_reset, | |
6111 | .read_u64 = mem_cgroup_read_u64, | |
6112 | }, | |
6113 | { }, /* terminate */ | |
6114 | }; | |
6115 | ||
6116 | static int __init mem_cgroup_swap_init(void) | |
6117 | { | |
6118 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6119 | do_swap_account = 1; | |
37e84351 VD |
6120 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6121 | swap_files)); | |
21afa38e JW |
6122 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6123 | memsw_cgroup_files)); | |
6124 | } | |
6125 | return 0; | |
6126 | } | |
6127 | subsys_initcall(mem_cgroup_swap_init); | |
6128 | ||
6129 | #endif /* CONFIG_MEMCG_SWAP */ |