Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
4ffef5fe | 38 | #include <linux/hugetlb.h> |
d13d1443 | 39 | #include <linux/pagemap.h> |
d52aa412 | 40 | #include <linux/smp.h> |
8a9f3ccd | 41 | #include <linux/page-flags.h> |
66e1707b | 42 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
43 | #include <linux/bit_spinlock.h> |
44 | #include <linux/rcupdate.h> | |
e222432b | 45 | #include <linux/limits.h> |
b9e15baf | 46 | #include <linux/export.h> |
8c7c6e34 | 47 | #include <linux/mutex.h> |
bb4cc1a8 | 48 | #include <linux/rbtree.h> |
b6ac57d5 | 49 | #include <linux/slab.h> |
66e1707b | 50 | #include <linux/swap.h> |
02491447 | 51 | #include <linux/swapops.h> |
66e1707b | 52 | #include <linux/spinlock.h> |
2e72b634 | 53 | #include <linux/eventfd.h> |
79bd9814 | 54 | #include <linux/poll.h> |
2e72b634 | 55 | #include <linux/sort.h> |
66e1707b | 56 | #include <linux/fs.h> |
d2ceb9b7 | 57 | #include <linux/seq_file.h> |
70ddf637 | 58 | #include <linux/vmpressure.h> |
b69408e8 | 59 | #include <linux/mm_inline.h> |
5d1ea48b | 60 | #include <linux/swap_cgroup.h> |
cdec2e42 | 61 | #include <linux/cpu.h> |
158e0a2d | 62 | #include <linux/oom.h> |
0056f4e6 | 63 | #include <linux/lockdep.h> |
79bd9814 | 64 | #include <linux/file.h> |
b23afb93 | 65 | #include <linux/tracehook.h> |
08e552c6 | 66 | #include "internal.h" |
d1a4c0b3 | 67 | #include <net/sock.h> |
4bd2c1ee | 68 | #include <net/ip.h> |
f35c3a8e | 69 | #include "slab.h" |
8cdea7c0 | 70 | |
8697d331 BS |
71 | #include <asm/uaccess.h> |
72 | ||
cc8e970c KM |
73 | #include <trace/events/vmscan.h> |
74 | ||
073219e9 TH |
75 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
76 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 77 | |
7d828602 JW |
78 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
79 | ||
a181b0e8 | 80 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 81 | |
f7e1cb6e JW |
82 | /* Socket memory accounting disabled? */ |
83 | static bool cgroup_memory_nosocket; | |
84 | ||
04823c83 VD |
85 | /* Kernel memory accounting disabled? */ |
86 | static bool cgroup_memory_nokmem; | |
87 | ||
21afa38e | 88 | /* Whether the swap controller is active */ |
c255a458 | 89 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 90 | int do_swap_account __read_mostly; |
c077719b | 91 | #else |
a0db00fc | 92 | #define do_swap_account 0 |
c077719b KH |
93 | #endif |
94 | ||
7941d214 JW |
95 | /* Whether legacy memory+swap accounting is active */ |
96 | static bool do_memsw_account(void) | |
97 | { | |
98 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
99 | } | |
100 | ||
af7c4b0e JW |
101 | static const char * const mem_cgroup_stat_names[] = { |
102 | "cache", | |
103 | "rss", | |
b070e65c | 104 | "rss_huge", |
af7c4b0e | 105 | "mapped_file", |
c4843a75 | 106 | "dirty", |
3ea67d06 | 107 | "writeback", |
af7c4b0e JW |
108 | "swap", |
109 | }; | |
110 | ||
af7c4b0e JW |
111 | static const char * const mem_cgroup_events_names[] = { |
112 | "pgpgin", | |
113 | "pgpgout", | |
114 | "pgfault", | |
115 | "pgmajfault", | |
116 | }; | |
117 | ||
58cf188e SZ |
118 | static const char * const mem_cgroup_lru_names[] = { |
119 | "inactive_anon", | |
120 | "active_anon", | |
121 | "inactive_file", | |
122 | "active_file", | |
123 | "unevictable", | |
124 | }; | |
125 | ||
a0db00fc KS |
126 | #define THRESHOLDS_EVENTS_TARGET 128 |
127 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
128 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 129 | |
bb4cc1a8 AM |
130 | /* |
131 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
132 | * their hierarchy representation | |
133 | */ | |
134 | ||
135 | struct mem_cgroup_tree_per_zone { | |
136 | struct rb_root rb_root; | |
137 | spinlock_t lock; | |
138 | }; | |
139 | ||
140 | struct mem_cgroup_tree_per_node { | |
141 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
142 | }; | |
143 | ||
144 | struct mem_cgroup_tree { | |
145 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
146 | }; | |
147 | ||
148 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
149 | ||
9490ff27 KH |
150 | /* for OOM */ |
151 | struct mem_cgroup_eventfd_list { | |
152 | struct list_head list; | |
153 | struct eventfd_ctx *eventfd; | |
154 | }; | |
2e72b634 | 155 | |
79bd9814 TH |
156 | /* |
157 | * cgroup_event represents events which userspace want to receive. | |
158 | */ | |
3bc942f3 | 159 | struct mem_cgroup_event { |
79bd9814 | 160 | /* |
59b6f873 | 161 | * memcg which the event belongs to. |
79bd9814 | 162 | */ |
59b6f873 | 163 | struct mem_cgroup *memcg; |
79bd9814 TH |
164 | /* |
165 | * eventfd to signal userspace about the event. | |
166 | */ | |
167 | struct eventfd_ctx *eventfd; | |
168 | /* | |
169 | * Each of these stored in a list by the cgroup. | |
170 | */ | |
171 | struct list_head list; | |
fba94807 TH |
172 | /* |
173 | * register_event() callback will be used to add new userspace | |
174 | * waiter for changes related to this event. Use eventfd_signal() | |
175 | * on eventfd to send notification to userspace. | |
176 | */ | |
59b6f873 | 177 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 178 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
179 | /* |
180 | * unregister_event() callback will be called when userspace closes | |
181 | * the eventfd or on cgroup removing. This callback must be set, | |
182 | * if you want provide notification functionality. | |
183 | */ | |
59b6f873 | 184 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 185 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
186 | /* |
187 | * All fields below needed to unregister event when | |
188 | * userspace closes eventfd. | |
189 | */ | |
190 | poll_table pt; | |
191 | wait_queue_head_t *wqh; | |
192 | wait_queue_t wait; | |
193 | struct work_struct remove; | |
194 | }; | |
195 | ||
c0ff4b85 R |
196 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
197 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 198 | |
7dc74be0 DN |
199 | /* Stuffs for move charges at task migration. */ |
200 | /* | |
1dfab5ab | 201 | * Types of charges to be moved. |
7dc74be0 | 202 | */ |
1dfab5ab JW |
203 | #define MOVE_ANON 0x1U |
204 | #define MOVE_FILE 0x2U | |
205 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 206 | |
4ffef5fe DN |
207 | /* "mc" and its members are protected by cgroup_mutex */ |
208 | static struct move_charge_struct { | |
b1dd693e | 209 | spinlock_t lock; /* for from, to */ |
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 | ||
4219b2da LZ |
271 | /* |
272 | * We restrict the id in the range of [1, 65535], so it can fit into | |
273 | * an unsigned short. | |
274 | */ | |
275 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
276 | ||
34c00c31 LZ |
277 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
278 | { | |
15a4c835 | 279 | return memcg->css.id; |
34c00c31 LZ |
280 | } |
281 | ||
adbe427b VD |
282 | /* |
283 | * A helper function to get mem_cgroup from ID. must be called under | |
284 | * rcu_read_lock(). The caller is responsible for calling | |
285 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
286 | * refcnt from swap can be called against removed memcg.) | |
287 | */ | |
34c00c31 LZ |
288 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
289 | { | |
290 | struct cgroup_subsys_state *css; | |
291 | ||
7d699ddb | 292 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
293 | return mem_cgroup_from_css(css); |
294 | } | |
295 | ||
127424c8 | 296 | #ifndef CONFIG_SLOB |
55007d84 | 297 | /* |
f7ce3190 | 298 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
299 | * The main reason for not using cgroup id for this: |
300 | * this works better in sparse environments, where we have a lot of memcgs, | |
301 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
302 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
303 | * 200 entry array for that. | |
55007d84 | 304 | * |
dbcf73e2 VD |
305 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
306 | * will double each time we have to increase it. | |
55007d84 | 307 | */ |
dbcf73e2 VD |
308 | static DEFINE_IDA(memcg_cache_ida); |
309 | int memcg_nr_cache_ids; | |
749c5415 | 310 | |
05257a1a VD |
311 | /* Protects memcg_nr_cache_ids */ |
312 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
313 | ||
314 | void memcg_get_cache_ids(void) | |
315 | { | |
316 | down_read(&memcg_cache_ids_sem); | |
317 | } | |
318 | ||
319 | void memcg_put_cache_ids(void) | |
320 | { | |
321 | up_read(&memcg_cache_ids_sem); | |
322 | } | |
323 | ||
55007d84 GC |
324 | /* |
325 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
326 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
327 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
328 | * tunable, but that is strictly not necessary. | |
329 | * | |
b8627835 | 330 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
331 | * this constant directly from cgroup, but it is understandable that this is |
332 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 333 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
334 | * increase ours as well if it increases. |
335 | */ | |
336 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 337 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 338 | |
d7f25f8a GC |
339 | /* |
340 | * A lot of the calls to the cache allocation functions are expected to be | |
341 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
342 | * conditional to this static branch, we'll have to allow modules that does | |
343 | * kmem_cache_alloc and the such to see this symbol as well | |
344 | */ | |
ef12947c | 345 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 346 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 347 | |
127424c8 | 348 | #endif /* !CONFIG_SLOB */ |
a8964b9b | 349 | |
f64c3f54 | 350 | static struct mem_cgroup_per_zone * |
e231875b | 351 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 352 | { |
e231875b JZ |
353 | int nid = zone_to_nid(zone); |
354 | int zid = zone_idx(zone); | |
355 | ||
54f72fe0 | 356 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
357 | } |
358 | ||
ad7fa852 TH |
359 | /** |
360 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
361 | * @page: page of interest | |
362 | * | |
363 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
364 | * with @page is returned. The returned css remains associated with @page | |
365 | * until it is released. | |
366 | * | |
367 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
368 | * is returned. | |
ad7fa852 TH |
369 | */ |
370 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
371 | { | |
372 | struct mem_cgroup *memcg; | |
373 | ||
ad7fa852 TH |
374 | memcg = page->mem_cgroup; |
375 | ||
9e10a130 | 376 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
377 | memcg = root_mem_cgroup; |
378 | ||
ad7fa852 TH |
379 | return &memcg->css; |
380 | } | |
381 | ||
2fc04524 VD |
382 | /** |
383 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
384 | * @page: the page | |
385 | * | |
386 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
387 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
388 | * is safe to call this function without holding a reference to @page. | |
389 | * | |
390 | * Note, this function is inherently racy, because there is nothing to prevent | |
391 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
392 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
393 | * do not care (such as procfs interfaces). | |
394 | */ | |
395 | ino_t page_cgroup_ino(struct page *page) | |
396 | { | |
397 | struct mem_cgroup *memcg; | |
398 | unsigned long ino = 0; | |
399 | ||
400 | rcu_read_lock(); | |
401 | memcg = READ_ONCE(page->mem_cgroup); | |
402 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
403 | memcg = parent_mem_cgroup(memcg); | |
404 | if (memcg) | |
405 | ino = cgroup_ino(memcg->css.cgroup); | |
406 | rcu_read_unlock(); | |
407 | return ino; | |
408 | } | |
409 | ||
f64c3f54 | 410 | static struct mem_cgroup_per_zone * |
e231875b | 411 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 412 | { |
97a6c37b JW |
413 | int nid = page_to_nid(page); |
414 | int zid = page_zonenum(page); | |
f64c3f54 | 415 | |
e231875b | 416 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
417 | } |
418 | ||
bb4cc1a8 AM |
419 | static struct mem_cgroup_tree_per_zone * |
420 | soft_limit_tree_node_zone(int nid, int zid) | |
421 | { | |
422 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
423 | } | |
424 | ||
425 | static struct mem_cgroup_tree_per_zone * | |
426 | soft_limit_tree_from_page(struct page *page) | |
427 | { | |
428 | int nid = page_to_nid(page); | |
429 | int zid = page_zonenum(page); | |
430 | ||
431 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
432 | } | |
433 | ||
cf2c8127 JW |
434 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
435 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 436 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
437 | { |
438 | struct rb_node **p = &mctz->rb_root.rb_node; | |
439 | struct rb_node *parent = NULL; | |
440 | struct mem_cgroup_per_zone *mz_node; | |
441 | ||
442 | if (mz->on_tree) | |
443 | return; | |
444 | ||
445 | mz->usage_in_excess = new_usage_in_excess; | |
446 | if (!mz->usage_in_excess) | |
447 | return; | |
448 | while (*p) { | |
449 | parent = *p; | |
450 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
451 | tree_node); | |
452 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
453 | p = &(*p)->rb_left; | |
454 | /* | |
455 | * We can't avoid mem cgroups that are over their soft | |
456 | * limit by the same amount | |
457 | */ | |
458 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
459 | p = &(*p)->rb_right; | |
460 | } | |
461 | rb_link_node(&mz->tree_node, parent, p); | |
462 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
463 | mz->on_tree = true; | |
464 | } | |
465 | ||
cf2c8127 JW |
466 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
467 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
468 | { |
469 | if (!mz->on_tree) | |
470 | return; | |
471 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
472 | mz->on_tree = false; | |
473 | } | |
474 | ||
cf2c8127 JW |
475 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
476 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 477 | { |
0a31bc97 JW |
478 | unsigned long flags; |
479 | ||
480 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 481 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 482 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
483 | } |
484 | ||
3e32cb2e JW |
485 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
486 | { | |
487 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 488 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
489 | unsigned long excess = 0; |
490 | ||
491 | if (nr_pages > soft_limit) | |
492 | excess = nr_pages - soft_limit; | |
493 | ||
494 | return excess; | |
495 | } | |
bb4cc1a8 AM |
496 | |
497 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
498 | { | |
3e32cb2e | 499 | unsigned long excess; |
bb4cc1a8 AM |
500 | struct mem_cgroup_per_zone *mz; |
501 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 502 | |
e231875b | 503 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
504 | /* |
505 | * Necessary to update all ancestors when hierarchy is used. | |
506 | * because their event counter is not touched. | |
507 | */ | |
508 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 509 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 510 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
511 | /* |
512 | * We have to update the tree if mz is on RB-tree or | |
513 | * mem is over its softlimit. | |
514 | */ | |
515 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
516 | unsigned long flags; |
517 | ||
518 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
519 | /* if on-tree, remove it */ |
520 | if (mz->on_tree) | |
cf2c8127 | 521 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
522 | /* |
523 | * Insert again. mz->usage_in_excess will be updated. | |
524 | * If excess is 0, no tree ops. | |
525 | */ | |
cf2c8127 | 526 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 527 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
528 | } |
529 | } | |
530 | } | |
531 | ||
532 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
533 | { | |
bb4cc1a8 | 534 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
535 | struct mem_cgroup_per_zone *mz; |
536 | int nid, zid; | |
bb4cc1a8 | 537 | |
e231875b JZ |
538 | for_each_node(nid) { |
539 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
540 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
541 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 542 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
543 | } |
544 | } | |
545 | } | |
546 | ||
547 | static struct mem_cgroup_per_zone * | |
548 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
549 | { | |
550 | struct rb_node *rightmost = NULL; | |
551 | struct mem_cgroup_per_zone *mz; | |
552 | ||
553 | retry: | |
554 | mz = NULL; | |
555 | rightmost = rb_last(&mctz->rb_root); | |
556 | if (!rightmost) | |
557 | goto done; /* Nothing to reclaim from */ | |
558 | ||
559 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
560 | /* | |
561 | * Remove the node now but someone else can add it back, | |
562 | * we will to add it back at the end of reclaim to its correct | |
563 | * position in the tree. | |
564 | */ | |
cf2c8127 | 565 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 566 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 567 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
568 | goto retry; |
569 | done: | |
570 | return mz; | |
571 | } | |
572 | ||
573 | static struct mem_cgroup_per_zone * | |
574 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
575 | { | |
576 | struct mem_cgroup_per_zone *mz; | |
577 | ||
0a31bc97 | 578 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 579 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 580 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
581 | return mz; |
582 | } | |
583 | ||
711d3d2c | 584 | /* |
484ebb3b GT |
585 | * Return page count for single (non recursive) @memcg. |
586 | * | |
711d3d2c KH |
587 | * Implementation Note: reading percpu statistics for memcg. |
588 | * | |
589 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
590 | * synchronization to implement "quick" read. There are trade-off between | |
591 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 592 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
593 | * |
594 | * But this _read() function is used for user interface now. The user accounts | |
595 | * memory usage by memory cgroup and he _always_ requires exact value because | |
596 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
597 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
598 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
599 | * | |
600 | * If there are kernel internal actions which can make use of some not-exact | |
601 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 602 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
603 | * implemented. |
604 | */ | |
484ebb3b GT |
605 | static unsigned long |
606 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 607 | { |
7a159cc9 | 608 | long val = 0; |
c62b1a3b | 609 | int cpu; |
c62b1a3b | 610 | |
484ebb3b | 611 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 612 | for_each_possible_cpu(cpu) |
c0ff4b85 | 613 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
614 | /* |
615 | * Summing races with updates, so val may be negative. Avoid exposing | |
616 | * transient negative values. | |
617 | */ | |
618 | if (val < 0) | |
619 | val = 0; | |
c62b1a3b KH |
620 | return val; |
621 | } | |
622 | ||
c0ff4b85 | 623 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
624 | enum mem_cgroup_events_index idx) |
625 | { | |
626 | unsigned long val = 0; | |
627 | int cpu; | |
628 | ||
733a572e | 629 | for_each_possible_cpu(cpu) |
c0ff4b85 | 630 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
631 | return val; |
632 | } | |
633 | ||
c0ff4b85 | 634 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 635 | struct page *page, |
f627c2f5 | 636 | bool compound, int nr_pages) |
d52aa412 | 637 | { |
b2402857 KH |
638 | /* |
639 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
640 | * counted as CACHE even if it's on ANON LRU. | |
641 | */ | |
0a31bc97 | 642 | if (PageAnon(page)) |
b2402857 | 643 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 644 | nr_pages); |
d52aa412 | 645 | else |
b2402857 | 646 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 647 | nr_pages); |
55e462b0 | 648 | |
f627c2f5 KS |
649 | if (compound) { |
650 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
b070e65c DR |
651 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
652 | nr_pages); | |
f627c2f5 | 653 | } |
b070e65c | 654 | |
e401f176 KH |
655 | /* pagein of a big page is an event. So, ignore page size */ |
656 | if (nr_pages > 0) | |
c0ff4b85 | 657 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 658 | else { |
c0ff4b85 | 659 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
660 | nr_pages = -nr_pages; /* for event */ |
661 | } | |
e401f176 | 662 | |
13114716 | 663 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
664 | } |
665 | ||
e231875b JZ |
666 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
667 | int nid, | |
668 | unsigned int lru_mask) | |
bb2a0de9 | 669 | { |
e231875b | 670 | unsigned long nr = 0; |
889976db YH |
671 | int zid; |
672 | ||
e231875b | 673 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 674 | |
e231875b JZ |
675 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
676 | struct mem_cgroup_per_zone *mz; | |
677 | enum lru_list lru; | |
678 | ||
679 | for_each_lru(lru) { | |
680 | if (!(BIT(lru) & lru_mask)) | |
681 | continue; | |
682 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
683 | nr += mz->lru_size[lru]; | |
684 | } | |
685 | } | |
686 | return nr; | |
889976db | 687 | } |
bb2a0de9 | 688 | |
c0ff4b85 | 689 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 690 | unsigned int lru_mask) |
6d12e2d8 | 691 | { |
e231875b | 692 | unsigned long nr = 0; |
889976db | 693 | int nid; |
6d12e2d8 | 694 | |
31aaea4a | 695 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
696 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
697 | return nr; | |
d52aa412 KH |
698 | } |
699 | ||
f53d7ce3 JW |
700 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
701 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
702 | { |
703 | unsigned long val, next; | |
704 | ||
13114716 | 705 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 706 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 707 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
708 | if ((long)next - (long)val < 0) { |
709 | switch (target) { | |
710 | case MEM_CGROUP_TARGET_THRESH: | |
711 | next = val + THRESHOLDS_EVENTS_TARGET; | |
712 | break; | |
bb4cc1a8 AM |
713 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
714 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
715 | break; | |
f53d7ce3 JW |
716 | case MEM_CGROUP_TARGET_NUMAINFO: |
717 | next = val + NUMAINFO_EVENTS_TARGET; | |
718 | break; | |
719 | default: | |
720 | break; | |
721 | } | |
722 | __this_cpu_write(memcg->stat->targets[target], next); | |
723 | return true; | |
7a159cc9 | 724 | } |
f53d7ce3 | 725 | return false; |
d2265e6f KH |
726 | } |
727 | ||
728 | /* | |
729 | * Check events in order. | |
730 | * | |
731 | */ | |
c0ff4b85 | 732 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
733 | { |
734 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
735 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
736 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 737 | bool do_softlimit; |
82b3f2a7 | 738 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 739 | |
bb4cc1a8 AM |
740 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
741 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
742 | #if MAX_NUMNODES > 1 |
743 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
744 | MEM_CGROUP_TARGET_NUMAINFO); | |
745 | #endif | |
c0ff4b85 | 746 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
747 | if (unlikely(do_softlimit)) |
748 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 749 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 750 | if (unlikely(do_numainfo)) |
c0ff4b85 | 751 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 752 | #endif |
0a31bc97 | 753 | } |
d2265e6f KH |
754 | } |
755 | ||
cf475ad2 | 756 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 757 | { |
31a78f23 BS |
758 | /* |
759 | * mm_update_next_owner() may clear mm->owner to NULL | |
760 | * if it races with swapoff, page migration, etc. | |
761 | * So this can be called with p == NULL. | |
762 | */ | |
763 | if (unlikely(!p)) | |
764 | return NULL; | |
765 | ||
073219e9 | 766 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 767 | } |
33398cf2 | 768 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 769 | |
df381975 | 770 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 771 | { |
c0ff4b85 | 772 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 773 | |
54595fe2 KH |
774 | rcu_read_lock(); |
775 | do { | |
6f6acb00 MH |
776 | /* |
777 | * Page cache insertions can happen withou an | |
778 | * actual mm context, e.g. during disk probing | |
779 | * on boot, loopback IO, acct() writes etc. | |
780 | */ | |
781 | if (unlikely(!mm)) | |
df381975 | 782 | memcg = root_mem_cgroup; |
6f6acb00 MH |
783 | else { |
784 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
785 | if (unlikely(!memcg)) | |
786 | memcg = root_mem_cgroup; | |
787 | } | |
ec903c0c | 788 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 789 | rcu_read_unlock(); |
c0ff4b85 | 790 | return memcg; |
54595fe2 KH |
791 | } |
792 | ||
5660048c JW |
793 | /** |
794 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
795 | * @root: hierarchy root | |
796 | * @prev: previously returned memcg, NULL on first invocation | |
797 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
798 | * | |
799 | * Returns references to children of the hierarchy below @root, or | |
800 | * @root itself, or %NULL after a full round-trip. | |
801 | * | |
802 | * Caller must pass the return value in @prev on subsequent | |
803 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
804 | * to cancel a hierarchy walk before the round-trip is complete. | |
805 | * | |
806 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
807 | * divide up the memcgs in the hierarchy among all concurrent | |
808 | * reclaimers operating on the same zone and priority. | |
809 | */ | |
694fbc0f | 810 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 811 | struct mem_cgroup *prev, |
694fbc0f | 812 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 813 | { |
33398cf2 | 814 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 815 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 816 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 817 | struct mem_cgroup *pos = NULL; |
711d3d2c | 818 | |
694fbc0f AM |
819 | if (mem_cgroup_disabled()) |
820 | return NULL; | |
5660048c | 821 | |
9f3a0d09 JW |
822 | if (!root) |
823 | root = root_mem_cgroup; | |
7d74b06f | 824 | |
9f3a0d09 | 825 | if (prev && !reclaim) |
5ac8fb31 | 826 | pos = prev; |
14067bb3 | 827 | |
9f3a0d09 JW |
828 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
829 | if (prev) | |
5ac8fb31 | 830 | goto out; |
694fbc0f | 831 | return root; |
9f3a0d09 | 832 | } |
14067bb3 | 833 | |
542f85f9 | 834 | rcu_read_lock(); |
5f578161 | 835 | |
5ac8fb31 JW |
836 | if (reclaim) { |
837 | struct mem_cgroup_per_zone *mz; | |
838 | ||
839 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
840 | iter = &mz->iter[reclaim->priority]; | |
841 | ||
842 | if (prev && reclaim->generation != iter->generation) | |
843 | goto out_unlock; | |
844 | ||
6df38689 | 845 | while (1) { |
4db0c3c2 | 846 | pos = READ_ONCE(iter->position); |
6df38689 VD |
847 | if (!pos || css_tryget(&pos->css)) |
848 | break; | |
5ac8fb31 | 849 | /* |
6df38689 VD |
850 | * css reference reached zero, so iter->position will |
851 | * be cleared by ->css_released. However, we should not | |
852 | * rely on this happening soon, because ->css_released | |
853 | * is called from a work queue, and by busy-waiting we | |
854 | * might block it. So we clear iter->position right | |
855 | * away. | |
5ac8fb31 | 856 | */ |
6df38689 VD |
857 | (void)cmpxchg(&iter->position, pos, NULL); |
858 | } | |
5ac8fb31 JW |
859 | } |
860 | ||
861 | if (pos) | |
862 | css = &pos->css; | |
863 | ||
864 | for (;;) { | |
865 | css = css_next_descendant_pre(css, &root->css); | |
866 | if (!css) { | |
867 | /* | |
868 | * Reclaimers share the hierarchy walk, and a | |
869 | * new one might jump in right at the end of | |
870 | * the hierarchy - make sure they see at least | |
871 | * one group and restart from the beginning. | |
872 | */ | |
873 | if (!prev) | |
874 | continue; | |
875 | break; | |
527a5ec9 | 876 | } |
7d74b06f | 877 | |
5ac8fb31 JW |
878 | /* |
879 | * Verify the css and acquire a reference. The root | |
880 | * is provided by the caller, so we know it's alive | |
881 | * and kicking, and don't take an extra reference. | |
882 | */ | |
883 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 884 | |
5ac8fb31 JW |
885 | if (css == &root->css) |
886 | break; | |
14067bb3 | 887 | |
0b8f73e1 JW |
888 | if (css_tryget(css)) |
889 | break; | |
9f3a0d09 | 890 | |
5ac8fb31 | 891 | memcg = NULL; |
9f3a0d09 | 892 | } |
5ac8fb31 JW |
893 | |
894 | if (reclaim) { | |
5ac8fb31 | 895 | /* |
6df38689 VD |
896 | * The position could have already been updated by a competing |
897 | * thread, so check that the value hasn't changed since we read | |
898 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 899 | */ |
6df38689 VD |
900 | (void)cmpxchg(&iter->position, pos, memcg); |
901 | ||
5ac8fb31 JW |
902 | if (pos) |
903 | css_put(&pos->css); | |
904 | ||
905 | if (!memcg) | |
906 | iter->generation++; | |
907 | else if (!prev) | |
908 | reclaim->generation = iter->generation; | |
9f3a0d09 | 909 | } |
5ac8fb31 | 910 | |
542f85f9 MH |
911 | out_unlock: |
912 | rcu_read_unlock(); | |
5ac8fb31 | 913 | out: |
c40046f3 MH |
914 | if (prev && prev != root) |
915 | css_put(&prev->css); | |
916 | ||
9f3a0d09 | 917 | return memcg; |
14067bb3 | 918 | } |
7d74b06f | 919 | |
5660048c JW |
920 | /** |
921 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
922 | * @root: hierarchy root | |
923 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
924 | */ | |
925 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
926 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
927 | { |
928 | if (!root) | |
929 | root = root_mem_cgroup; | |
930 | if (prev && prev != root) | |
931 | css_put(&prev->css); | |
932 | } | |
7d74b06f | 933 | |
6df38689 VD |
934 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
935 | { | |
936 | struct mem_cgroup *memcg = dead_memcg; | |
937 | struct mem_cgroup_reclaim_iter *iter; | |
938 | struct mem_cgroup_per_zone *mz; | |
939 | int nid, zid; | |
940 | int i; | |
941 | ||
942 | while ((memcg = parent_mem_cgroup(memcg))) { | |
943 | for_each_node(nid) { | |
944 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
945 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
946 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
947 | iter = &mz->iter[i]; | |
948 | cmpxchg(&iter->position, | |
949 | dead_memcg, NULL); | |
950 | } | |
951 | } | |
952 | } | |
953 | } | |
954 | } | |
955 | ||
9f3a0d09 JW |
956 | /* |
957 | * Iteration constructs for visiting all cgroups (under a tree). If | |
958 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
959 | * be used for reference counting. | |
960 | */ | |
961 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 962 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 963 | iter != NULL; \ |
527a5ec9 | 964 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 965 | |
9f3a0d09 | 966 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 967 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 968 | iter != NULL; \ |
527a5ec9 | 969 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 970 | |
925b7673 JW |
971 | /** |
972 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
973 | * @zone: zone of the wanted lruvec | |
fa9add64 | 974 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
975 | * |
976 | * Returns the lru list vector holding pages for the given @zone and | |
977 | * @mem. This can be the global zone lruvec, if the memory controller | |
978 | * is disabled. | |
979 | */ | |
980 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
981 | struct mem_cgroup *memcg) | |
982 | { | |
983 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 984 | struct lruvec *lruvec; |
925b7673 | 985 | |
bea8c150 HD |
986 | if (mem_cgroup_disabled()) { |
987 | lruvec = &zone->lruvec; | |
988 | goto out; | |
989 | } | |
925b7673 | 990 | |
e231875b | 991 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
992 | lruvec = &mz->lruvec; |
993 | out: | |
994 | /* | |
995 | * Since a node can be onlined after the mem_cgroup was created, | |
996 | * we have to be prepared to initialize lruvec->zone here; | |
997 | * and if offlined then reonlined, we need to reinitialize it. | |
998 | */ | |
999 | if (unlikely(lruvec->zone != zone)) | |
1000 | lruvec->zone = zone; | |
1001 | return lruvec; | |
925b7673 JW |
1002 | } |
1003 | ||
925b7673 | 1004 | /** |
dfe0e773 | 1005 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1006 | * @page: the page |
fa9add64 | 1007 | * @zone: zone of the page |
dfe0e773 JW |
1008 | * |
1009 | * This function is only safe when following the LRU page isolation | |
1010 | * and putback protocol: the LRU lock must be held, and the page must | |
1011 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1012 | */ |
fa9add64 | 1013 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1014 | { |
08e552c6 | 1015 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1016 | struct mem_cgroup *memcg; |
bea8c150 | 1017 | struct lruvec *lruvec; |
6d12e2d8 | 1018 | |
bea8c150 HD |
1019 | if (mem_cgroup_disabled()) { |
1020 | lruvec = &zone->lruvec; | |
1021 | goto out; | |
1022 | } | |
925b7673 | 1023 | |
1306a85a | 1024 | memcg = page->mem_cgroup; |
7512102c | 1025 | /* |
dfe0e773 | 1026 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1027 | * possibly migrated - before they are charged. |
7512102c | 1028 | */ |
29833315 JW |
1029 | if (!memcg) |
1030 | memcg = root_mem_cgroup; | |
7512102c | 1031 | |
e231875b | 1032 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1033 | lruvec = &mz->lruvec; |
1034 | out: | |
1035 | /* | |
1036 | * Since a node can be onlined after the mem_cgroup was created, | |
1037 | * we have to be prepared to initialize lruvec->zone here; | |
1038 | * and if offlined then reonlined, we need to reinitialize it. | |
1039 | */ | |
1040 | if (unlikely(lruvec->zone != zone)) | |
1041 | lruvec->zone = zone; | |
1042 | return lruvec; | |
08e552c6 | 1043 | } |
b69408e8 | 1044 | |
925b7673 | 1045 | /** |
fa9add64 HD |
1046 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1047 | * @lruvec: mem_cgroup per zone lru vector | |
1048 | * @lru: index of lru list the page is sitting on | |
1049 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1050 | * |
fa9add64 HD |
1051 | * This function must be called when a page is added to or removed from an |
1052 | * lru list. | |
3f58a829 | 1053 | */ |
fa9add64 HD |
1054 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1055 | int nr_pages) | |
3f58a829 MK |
1056 | { |
1057 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1058 | unsigned long *lru_size; |
3f58a829 MK |
1059 | |
1060 | if (mem_cgroup_disabled()) | |
1061 | return; | |
1062 | ||
fa9add64 HD |
1063 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1064 | lru_size = mz->lru_size + lru; | |
1065 | *lru_size += nr_pages; | |
1066 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1067 | } |
544122e5 | 1068 | |
2314b42d | 1069 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1070 | { |
2314b42d | 1071 | struct mem_cgroup *task_memcg; |
158e0a2d | 1072 | struct task_struct *p; |
ffbdccf5 | 1073 | bool ret; |
4c4a2214 | 1074 | |
158e0a2d | 1075 | p = find_lock_task_mm(task); |
de077d22 | 1076 | if (p) { |
2314b42d | 1077 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1078 | task_unlock(p); |
1079 | } else { | |
1080 | /* | |
1081 | * All threads may have already detached their mm's, but the oom | |
1082 | * killer still needs to detect if they have already been oom | |
1083 | * killed to prevent needlessly killing additional tasks. | |
1084 | */ | |
ffbdccf5 | 1085 | rcu_read_lock(); |
2314b42d JW |
1086 | task_memcg = mem_cgroup_from_task(task); |
1087 | css_get(&task_memcg->css); | |
ffbdccf5 | 1088 | rcu_read_unlock(); |
de077d22 | 1089 | } |
2314b42d JW |
1090 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1091 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1092 | return ret; |
1093 | } | |
1094 | ||
19942822 | 1095 | /** |
9d11ea9f | 1096 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1097 | * @memcg: the memory cgroup |
19942822 | 1098 | * |
9d11ea9f | 1099 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1100 | * pages. |
19942822 | 1101 | */ |
c0ff4b85 | 1102 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1103 | { |
3e32cb2e JW |
1104 | unsigned long margin = 0; |
1105 | unsigned long count; | |
1106 | unsigned long limit; | |
9d11ea9f | 1107 | |
3e32cb2e | 1108 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1109 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1110 | if (count < limit) |
1111 | margin = limit - count; | |
1112 | ||
7941d214 | 1113 | if (do_memsw_account()) { |
3e32cb2e | 1114 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1115 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1116 | if (count <= limit) |
1117 | margin = min(margin, limit - count); | |
1118 | } | |
1119 | ||
1120 | return margin; | |
19942822 JW |
1121 | } |
1122 | ||
32047e2a | 1123 | /* |
bdcbb659 | 1124 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1125 | * |
bdcbb659 QH |
1126 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1127 | * moving cgroups. This is for waiting at high-memory pressure | |
1128 | * caused by "move". | |
32047e2a | 1129 | */ |
c0ff4b85 | 1130 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1131 | { |
2bd9bb20 KH |
1132 | struct mem_cgroup *from; |
1133 | struct mem_cgroup *to; | |
4b534334 | 1134 | bool ret = false; |
2bd9bb20 KH |
1135 | /* |
1136 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1137 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1138 | */ | |
1139 | spin_lock(&mc.lock); | |
1140 | from = mc.from; | |
1141 | to = mc.to; | |
1142 | if (!from) | |
1143 | goto unlock; | |
3e92041d | 1144 | |
2314b42d JW |
1145 | ret = mem_cgroup_is_descendant(from, memcg) || |
1146 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1147 | unlock: |
1148 | spin_unlock(&mc.lock); | |
4b534334 KH |
1149 | return ret; |
1150 | } | |
1151 | ||
c0ff4b85 | 1152 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1153 | { |
1154 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1155 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1156 | DEFINE_WAIT(wait); |
1157 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1158 | /* moving charge context might have finished. */ | |
1159 | if (mc.moving_task) | |
1160 | schedule(); | |
1161 | finish_wait(&mc.waitq, &wait); | |
1162 | return true; | |
1163 | } | |
1164 | } | |
1165 | return false; | |
1166 | } | |
1167 | ||
58cf188e | 1168 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1169 | /** |
58cf188e | 1170 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1171 | * @memcg: The memory cgroup that went over limit |
1172 | * @p: Task that is going to be killed | |
1173 | * | |
1174 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1175 | * enabled | |
1176 | */ | |
1177 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1178 | { | |
e61734c5 | 1179 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1180 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1181 | struct mem_cgroup *iter; |
1182 | unsigned int i; | |
e222432b | 1183 | |
08088cb9 | 1184 | mutex_lock(&oom_info_lock); |
e222432b BS |
1185 | rcu_read_lock(); |
1186 | ||
2415b9f5 BV |
1187 | if (p) { |
1188 | pr_info("Task in "); | |
1189 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1190 | pr_cont(" killed as a result of limit of "); | |
1191 | } else { | |
1192 | pr_info("Memory limit reached of cgroup "); | |
1193 | } | |
1194 | ||
e61734c5 | 1195 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1196 | pr_cont("\n"); |
e222432b | 1197 | |
e222432b BS |
1198 | rcu_read_unlock(); |
1199 | ||
3e32cb2e JW |
1200 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1201 | K((u64)page_counter_read(&memcg->memory)), | |
1202 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1203 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1204 | K((u64)page_counter_read(&memcg->memsw)), | |
1205 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1206 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1207 | K((u64)page_counter_read(&memcg->kmem)), | |
1208 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1209 | |
1210 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1211 | pr_info("Memory cgroup stats for "); |
1212 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1213 | pr_cont(":"); |
1214 | ||
1215 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
37e84351 | 1216 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
58cf188e | 1217 | continue; |
484ebb3b | 1218 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1219 | K(mem_cgroup_read_stat(iter, i))); |
1220 | } | |
1221 | ||
1222 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1223 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1224 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1225 | ||
1226 | pr_cont("\n"); | |
1227 | } | |
08088cb9 | 1228 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1229 | } |
1230 | ||
81d39c20 KH |
1231 | /* |
1232 | * This function returns the number of memcg under hierarchy tree. Returns | |
1233 | * 1(self count) if no children. | |
1234 | */ | |
c0ff4b85 | 1235 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1236 | { |
1237 | int num = 0; | |
7d74b06f KH |
1238 | struct mem_cgroup *iter; |
1239 | ||
c0ff4b85 | 1240 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1241 | num++; |
81d39c20 KH |
1242 | return num; |
1243 | } | |
1244 | ||
a63d83f4 DR |
1245 | /* |
1246 | * Return the memory (and swap, if configured) limit for a memcg. | |
1247 | */ | |
3e32cb2e | 1248 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1249 | { |
3e32cb2e | 1250 | unsigned long limit; |
f3e8eb70 | 1251 | |
3e32cb2e | 1252 | limit = memcg->memory.limit; |
9a5a8f19 | 1253 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1254 | unsigned long memsw_limit; |
37e84351 | 1255 | unsigned long swap_limit; |
9a5a8f19 | 1256 | |
3e32cb2e | 1257 | memsw_limit = memcg->memsw.limit; |
37e84351 VD |
1258 | swap_limit = memcg->swap.limit; |
1259 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); | |
1260 | limit = min(limit + swap_limit, memsw_limit); | |
9a5a8f19 | 1261 | } |
9a5a8f19 | 1262 | return limit; |
a63d83f4 DR |
1263 | } |
1264 | ||
19965460 DR |
1265 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1266 | int order) | |
9cbb78bb | 1267 | { |
6e0fc46d DR |
1268 | struct oom_control oc = { |
1269 | .zonelist = NULL, | |
1270 | .nodemask = NULL, | |
1271 | .gfp_mask = gfp_mask, | |
1272 | .order = order, | |
6e0fc46d | 1273 | }; |
9cbb78bb DR |
1274 | struct mem_cgroup *iter; |
1275 | unsigned long chosen_points = 0; | |
1276 | unsigned long totalpages; | |
1277 | unsigned int points = 0; | |
1278 | struct task_struct *chosen = NULL; | |
1279 | ||
dc56401f JW |
1280 | mutex_lock(&oom_lock); |
1281 | ||
876aafbf | 1282 | /* |
465adcf1 DR |
1283 | * If current has a pending SIGKILL or is exiting, then automatically |
1284 | * select it. The goal is to allow it to allocate so that it may | |
1285 | * quickly exit and free its memory. | |
876aafbf | 1286 | */ |
d003f371 | 1287 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1288 | mark_oom_victim(current); |
dc56401f | 1289 | goto unlock; |
876aafbf DR |
1290 | } |
1291 | ||
6e0fc46d | 1292 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1293 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1294 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1295 | struct css_task_iter it; |
9cbb78bb DR |
1296 | struct task_struct *task; |
1297 | ||
72ec7029 TH |
1298 | css_task_iter_start(&iter->css, &it); |
1299 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1300 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1301 | case OOM_SCAN_SELECT: |
1302 | if (chosen) | |
1303 | put_task_struct(chosen); | |
1304 | chosen = task; | |
1305 | chosen_points = ULONG_MAX; | |
1306 | get_task_struct(chosen); | |
1307 | /* fall through */ | |
1308 | case OOM_SCAN_CONTINUE: | |
1309 | continue; | |
1310 | case OOM_SCAN_ABORT: | |
72ec7029 | 1311 | css_task_iter_end(&it); |
9cbb78bb DR |
1312 | mem_cgroup_iter_break(memcg, iter); |
1313 | if (chosen) | |
1314 | put_task_struct(chosen); | |
dc56401f | 1315 | goto unlock; |
9cbb78bb DR |
1316 | case OOM_SCAN_OK: |
1317 | break; | |
1318 | }; | |
1319 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1320 | if (!points || points < chosen_points) |
1321 | continue; | |
1322 | /* Prefer thread group leaders for display purposes */ | |
1323 | if (points == chosen_points && | |
1324 | thread_group_leader(chosen)) | |
1325 | continue; | |
1326 | ||
1327 | if (chosen) | |
1328 | put_task_struct(chosen); | |
1329 | chosen = task; | |
1330 | chosen_points = points; | |
1331 | get_task_struct(chosen); | |
9cbb78bb | 1332 | } |
72ec7029 | 1333 | css_task_iter_end(&it); |
9cbb78bb DR |
1334 | } |
1335 | ||
dc56401f JW |
1336 | if (chosen) { |
1337 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1338 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1339 | "Memory cgroup out of memory"); | |
dc56401f JW |
1340 | } |
1341 | unlock: | |
1342 | mutex_unlock(&oom_lock); | |
9cbb78bb DR |
1343 | } |
1344 | ||
ae6e71d3 MC |
1345 | #if MAX_NUMNODES > 1 |
1346 | ||
4d0c066d KH |
1347 | /** |
1348 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1349 | * @memcg: the target memcg |
4d0c066d KH |
1350 | * @nid: the node ID to be checked. |
1351 | * @noswap : specify true here if the user wants flle only information. | |
1352 | * | |
1353 | * This function returns whether the specified memcg contains any | |
1354 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1355 | * pages in the node. | |
1356 | */ | |
c0ff4b85 | 1357 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1358 | int nid, bool noswap) |
1359 | { | |
c0ff4b85 | 1360 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1361 | return true; |
1362 | if (noswap || !total_swap_pages) | |
1363 | return false; | |
c0ff4b85 | 1364 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1365 | return true; |
1366 | return false; | |
1367 | ||
1368 | } | |
889976db YH |
1369 | |
1370 | /* | |
1371 | * Always updating the nodemask is not very good - even if we have an empty | |
1372 | * list or the wrong list here, we can start from some node and traverse all | |
1373 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1374 | * | |
1375 | */ | |
c0ff4b85 | 1376 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1377 | { |
1378 | int nid; | |
453a9bf3 KH |
1379 | /* |
1380 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1381 | * pagein/pageout changes since the last update. | |
1382 | */ | |
c0ff4b85 | 1383 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1384 | return; |
c0ff4b85 | 1385 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1386 | return; |
1387 | ||
889976db | 1388 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1389 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1390 | |
31aaea4a | 1391 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1392 | |
c0ff4b85 R |
1393 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1394 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1395 | } |
453a9bf3 | 1396 | |
c0ff4b85 R |
1397 | atomic_set(&memcg->numainfo_events, 0); |
1398 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1399 | } |
1400 | ||
1401 | /* | |
1402 | * Selecting a node where we start reclaim from. Because what we need is just | |
1403 | * reducing usage counter, start from anywhere is O,K. Considering | |
1404 | * memory reclaim from current node, there are pros. and cons. | |
1405 | * | |
1406 | * Freeing memory from current node means freeing memory from a node which | |
1407 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1408 | * hit limits, it will see a contention on a node. But freeing from remote | |
1409 | * node means more costs for memory reclaim because of memory latency. | |
1410 | * | |
1411 | * Now, we use round-robin. Better algorithm is welcomed. | |
1412 | */ | |
c0ff4b85 | 1413 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1414 | { |
1415 | int node; | |
1416 | ||
c0ff4b85 R |
1417 | mem_cgroup_may_update_nodemask(memcg); |
1418 | node = memcg->last_scanned_node; | |
889976db | 1419 | |
c0ff4b85 | 1420 | node = next_node(node, memcg->scan_nodes); |
889976db | 1421 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1422 | node = first_node(memcg->scan_nodes); |
889976db YH |
1423 | /* |
1424 | * We call this when we hit limit, not when pages are added to LRU. | |
1425 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1426 | * memcg is too small and all pages are not on LRU. In that case, | |
1427 | * we use curret node. | |
1428 | */ | |
1429 | if (unlikely(node == MAX_NUMNODES)) | |
1430 | node = numa_node_id(); | |
1431 | ||
c0ff4b85 | 1432 | memcg->last_scanned_node = node; |
889976db YH |
1433 | return node; |
1434 | } | |
889976db | 1435 | #else |
c0ff4b85 | 1436 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1437 | { |
1438 | return 0; | |
1439 | } | |
1440 | #endif | |
1441 | ||
0608f43d AM |
1442 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1443 | struct zone *zone, | |
1444 | gfp_t gfp_mask, | |
1445 | unsigned long *total_scanned) | |
1446 | { | |
1447 | struct mem_cgroup *victim = NULL; | |
1448 | int total = 0; | |
1449 | int loop = 0; | |
1450 | unsigned long excess; | |
1451 | unsigned long nr_scanned; | |
1452 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1453 | .zone = zone, | |
1454 | .priority = 0, | |
1455 | }; | |
1456 | ||
3e32cb2e | 1457 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1458 | |
1459 | while (1) { | |
1460 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1461 | if (!victim) { | |
1462 | loop++; | |
1463 | if (loop >= 2) { | |
1464 | /* | |
1465 | * If we have not been able to reclaim | |
1466 | * anything, it might because there are | |
1467 | * no reclaimable pages under this hierarchy | |
1468 | */ | |
1469 | if (!total) | |
1470 | break; | |
1471 | /* | |
1472 | * We want to do more targeted reclaim. | |
1473 | * excess >> 2 is not to excessive so as to | |
1474 | * reclaim too much, nor too less that we keep | |
1475 | * coming back to reclaim from this cgroup | |
1476 | */ | |
1477 | if (total >= (excess >> 2) || | |
1478 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1479 | break; | |
1480 | } | |
1481 | continue; | |
1482 | } | |
0608f43d AM |
1483 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1484 | zone, &nr_scanned); | |
1485 | *total_scanned += nr_scanned; | |
3e32cb2e | 1486 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1487 | break; |
6d61ef40 | 1488 | } |
0608f43d AM |
1489 | mem_cgroup_iter_break(root_memcg, victim); |
1490 | return total; | |
6d61ef40 BS |
1491 | } |
1492 | ||
0056f4e6 JW |
1493 | #ifdef CONFIG_LOCKDEP |
1494 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1495 | .name = "memcg_oom_lock", | |
1496 | }; | |
1497 | #endif | |
1498 | ||
fb2a6fc5 JW |
1499 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1500 | ||
867578cb KH |
1501 | /* |
1502 | * Check OOM-Killer is already running under our hierarchy. | |
1503 | * If someone is running, return false. | |
1504 | */ | |
fb2a6fc5 | 1505 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1506 | { |
79dfdacc | 1507 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1508 | |
fb2a6fc5 JW |
1509 | spin_lock(&memcg_oom_lock); |
1510 | ||
9f3a0d09 | 1511 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1512 | if (iter->oom_lock) { |
79dfdacc MH |
1513 | /* |
1514 | * this subtree of our hierarchy is already locked | |
1515 | * so we cannot give a lock. | |
1516 | */ | |
79dfdacc | 1517 | failed = iter; |
9f3a0d09 JW |
1518 | mem_cgroup_iter_break(memcg, iter); |
1519 | break; | |
23751be0 JW |
1520 | } else |
1521 | iter->oom_lock = true; | |
7d74b06f | 1522 | } |
867578cb | 1523 | |
fb2a6fc5 JW |
1524 | if (failed) { |
1525 | /* | |
1526 | * OK, we failed to lock the whole subtree so we have | |
1527 | * to clean up what we set up to the failing subtree | |
1528 | */ | |
1529 | for_each_mem_cgroup_tree(iter, memcg) { | |
1530 | if (iter == failed) { | |
1531 | mem_cgroup_iter_break(memcg, iter); | |
1532 | break; | |
1533 | } | |
1534 | iter->oom_lock = false; | |
79dfdacc | 1535 | } |
0056f4e6 JW |
1536 | } else |
1537 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1538 | |
1539 | spin_unlock(&memcg_oom_lock); | |
1540 | ||
1541 | return !failed; | |
a636b327 | 1542 | } |
0b7f569e | 1543 | |
fb2a6fc5 | 1544 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1545 | { |
7d74b06f KH |
1546 | struct mem_cgroup *iter; |
1547 | ||
fb2a6fc5 | 1548 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1549 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1550 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1551 | iter->oom_lock = false; |
fb2a6fc5 | 1552 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1553 | } |
1554 | ||
c0ff4b85 | 1555 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1556 | { |
1557 | struct mem_cgroup *iter; | |
1558 | ||
c2b42d3c | 1559 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1560 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1561 | iter->under_oom++; |
1562 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1563 | } |
1564 | ||
c0ff4b85 | 1565 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1566 | { |
1567 | struct mem_cgroup *iter; | |
1568 | ||
867578cb KH |
1569 | /* |
1570 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1571 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1572 | */ |
c2b42d3c | 1573 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1574 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1575 | if (iter->under_oom > 0) |
1576 | iter->under_oom--; | |
1577 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1578 | } |
1579 | ||
867578cb KH |
1580 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1581 | ||
dc98df5a | 1582 | struct oom_wait_info { |
d79154bb | 1583 | struct mem_cgroup *memcg; |
dc98df5a KH |
1584 | wait_queue_t wait; |
1585 | }; | |
1586 | ||
1587 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1588 | unsigned mode, int sync, void *arg) | |
1589 | { | |
d79154bb HD |
1590 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1591 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1592 | struct oom_wait_info *oom_wait_info; |
1593 | ||
1594 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1595 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1596 | |
2314b42d JW |
1597 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1598 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1599 | return 0; |
dc98df5a KH |
1600 | return autoremove_wake_function(wait, mode, sync, arg); |
1601 | } | |
1602 | ||
c0ff4b85 | 1603 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1604 | { |
c2b42d3c TH |
1605 | /* |
1606 | * For the following lockless ->under_oom test, the only required | |
1607 | * guarantee is that it must see the state asserted by an OOM when | |
1608 | * this function is called as a result of userland actions | |
1609 | * triggered by the notification of the OOM. This is trivially | |
1610 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1611 | * triggering notification. | |
1612 | */ | |
1613 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1614 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1615 | } |
1616 | ||
3812c8c8 | 1617 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1618 | { |
626ebc41 | 1619 | if (!current->memcg_may_oom) |
3812c8c8 | 1620 | return; |
867578cb | 1621 | /* |
49426420 JW |
1622 | * We are in the middle of the charge context here, so we |
1623 | * don't want to block when potentially sitting on a callstack | |
1624 | * that holds all kinds of filesystem and mm locks. | |
1625 | * | |
1626 | * Also, the caller may handle a failed allocation gracefully | |
1627 | * (like optional page cache readahead) and so an OOM killer | |
1628 | * invocation might not even be necessary. | |
1629 | * | |
1630 | * That's why we don't do anything here except remember the | |
1631 | * OOM context and then deal with it at the end of the page | |
1632 | * fault when the stack is unwound, the locks are released, | |
1633 | * and when we know whether the fault was overall successful. | |
867578cb | 1634 | */ |
49426420 | 1635 | css_get(&memcg->css); |
626ebc41 TH |
1636 | current->memcg_in_oom = memcg; |
1637 | current->memcg_oom_gfp_mask = mask; | |
1638 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1639 | } |
1640 | ||
1641 | /** | |
1642 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1643 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1644 | * |
49426420 JW |
1645 | * This has to be called at the end of a page fault if the memcg OOM |
1646 | * handler was enabled. | |
3812c8c8 | 1647 | * |
49426420 | 1648 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1649 | * sleep on a waitqueue until the userspace task resolves the |
1650 | * situation. Sleeping directly in the charge context with all kinds | |
1651 | * of locks held is not a good idea, instead we remember an OOM state | |
1652 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1653 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1654 | * |
1655 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1656 | * completed, %false otherwise. |
3812c8c8 | 1657 | */ |
49426420 | 1658 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1659 | { |
626ebc41 | 1660 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1661 | struct oom_wait_info owait; |
49426420 | 1662 | bool locked; |
3812c8c8 JW |
1663 | |
1664 | /* OOM is global, do not handle */ | |
3812c8c8 | 1665 | if (!memcg) |
49426420 | 1666 | return false; |
3812c8c8 | 1667 | |
c32b3cbe | 1668 | if (!handle || oom_killer_disabled) |
49426420 | 1669 | goto cleanup; |
3812c8c8 JW |
1670 | |
1671 | owait.memcg = memcg; | |
1672 | owait.wait.flags = 0; | |
1673 | owait.wait.func = memcg_oom_wake_function; | |
1674 | owait.wait.private = current; | |
1675 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1676 | |
3812c8c8 | 1677 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1678 | mem_cgroup_mark_under_oom(memcg); |
1679 | ||
1680 | locked = mem_cgroup_oom_trylock(memcg); | |
1681 | ||
1682 | if (locked) | |
1683 | mem_cgroup_oom_notify(memcg); | |
1684 | ||
1685 | if (locked && !memcg->oom_kill_disable) { | |
1686 | mem_cgroup_unmark_under_oom(memcg); | |
1687 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1688 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1689 | current->memcg_oom_order); | |
49426420 | 1690 | } else { |
3812c8c8 | 1691 | schedule(); |
49426420 JW |
1692 | mem_cgroup_unmark_under_oom(memcg); |
1693 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1694 | } | |
1695 | ||
1696 | if (locked) { | |
fb2a6fc5 JW |
1697 | mem_cgroup_oom_unlock(memcg); |
1698 | /* | |
1699 | * There is no guarantee that an OOM-lock contender | |
1700 | * sees the wakeups triggered by the OOM kill | |
1701 | * uncharges. Wake any sleepers explicitely. | |
1702 | */ | |
1703 | memcg_oom_recover(memcg); | |
1704 | } | |
49426420 | 1705 | cleanup: |
626ebc41 | 1706 | current->memcg_in_oom = NULL; |
3812c8c8 | 1707 | css_put(&memcg->css); |
867578cb | 1708 | return true; |
0b7f569e KH |
1709 | } |
1710 | ||
d7365e78 JW |
1711 | /** |
1712 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1713 | * @page: page that is going to change accounted state | |
32047e2a | 1714 | * |
d7365e78 JW |
1715 | * This function must mark the beginning of an accounted page state |
1716 | * change to prevent double accounting when the page is concurrently | |
1717 | * being moved to another memcg: | |
32047e2a | 1718 | * |
6de22619 | 1719 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1720 | * if (TestClearPageState(page)) |
1721 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1722 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1723 | */ |
6de22619 | 1724 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1725 | { |
1726 | struct mem_cgroup *memcg; | |
6de22619 | 1727 | unsigned long flags; |
89c06bd5 | 1728 | |
6de22619 JW |
1729 | /* |
1730 | * The RCU lock is held throughout the transaction. The fast | |
1731 | * path can get away without acquiring the memcg->move_lock | |
1732 | * because page moving starts with an RCU grace period. | |
1733 | * | |
1734 | * The RCU lock also protects the memcg from being freed when | |
1735 | * the page state that is going to change is the only thing | |
1736 | * preventing the page from being uncharged. | |
1737 | * E.g. end-writeback clearing PageWriteback(), which allows | |
1738 | * migration to go ahead and uncharge the page before the | |
1739 | * account transaction might be complete. | |
1740 | */ | |
d7365e78 JW |
1741 | rcu_read_lock(); |
1742 | ||
1743 | if (mem_cgroup_disabled()) | |
1744 | return NULL; | |
89c06bd5 | 1745 | again: |
1306a85a | 1746 | memcg = page->mem_cgroup; |
29833315 | 1747 | if (unlikely(!memcg)) |
d7365e78 JW |
1748 | return NULL; |
1749 | ||
bdcbb659 | 1750 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 1751 | return memcg; |
89c06bd5 | 1752 | |
6de22619 | 1753 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1754 | if (memcg != page->mem_cgroup) { |
6de22619 | 1755 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1756 | goto again; |
1757 | } | |
6de22619 JW |
1758 | |
1759 | /* | |
1760 | * When charge migration first begins, we can have locked and | |
1761 | * unlocked page stat updates happening concurrently. Track | |
1762 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
1763 | */ | |
1764 | memcg->move_lock_task = current; | |
1765 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
1766 | |
1767 | return memcg; | |
89c06bd5 | 1768 | } |
c4843a75 | 1769 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
89c06bd5 | 1770 | |
d7365e78 JW |
1771 | /** |
1772 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
1773 | * @memcg: the memcg that was accounted against | |
d7365e78 | 1774 | */ |
6de22619 | 1775 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 1776 | { |
6de22619 JW |
1777 | if (memcg && memcg->move_lock_task == current) { |
1778 | unsigned long flags = memcg->move_lock_flags; | |
1779 | ||
1780 | memcg->move_lock_task = NULL; | |
1781 | memcg->move_lock_flags = 0; | |
1782 | ||
1783 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1784 | } | |
89c06bd5 | 1785 | |
d7365e78 | 1786 | rcu_read_unlock(); |
89c06bd5 | 1787 | } |
c4843a75 | 1788 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
89c06bd5 | 1789 | |
cdec2e42 KH |
1790 | /* |
1791 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1792 | * TODO: maybe necessary to use big numbers in big irons. | |
1793 | */ | |
7ec99d62 | 1794 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1795 | struct memcg_stock_pcp { |
1796 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1797 | unsigned int nr_pages; |
cdec2e42 | 1798 | struct work_struct work; |
26fe6168 | 1799 | unsigned long flags; |
a0db00fc | 1800 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1801 | }; |
1802 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1803 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1804 | |
a0956d54 SS |
1805 | /** |
1806 | * consume_stock: Try to consume stocked charge on this cpu. | |
1807 | * @memcg: memcg to consume from. | |
1808 | * @nr_pages: how many pages to charge. | |
1809 | * | |
1810 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1811 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1812 | * service an allocation will refill the stock. | |
1813 | * | |
1814 | * returns true if successful, false otherwise. | |
cdec2e42 | 1815 | */ |
a0956d54 | 1816 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1817 | { |
1818 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1819 | bool ret = false; |
cdec2e42 | 1820 | |
a0956d54 | 1821 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1822 | return ret; |
a0956d54 | 1823 | |
cdec2e42 | 1824 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1825 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1826 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1827 | ret = true; |
1828 | } | |
cdec2e42 KH |
1829 | put_cpu_var(memcg_stock); |
1830 | return ret; | |
1831 | } | |
1832 | ||
1833 | /* | |
3e32cb2e | 1834 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1835 | */ |
1836 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1837 | { | |
1838 | struct mem_cgroup *old = stock->cached; | |
1839 | ||
11c9ea4e | 1840 | if (stock->nr_pages) { |
3e32cb2e | 1841 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1842 | if (do_memsw_account()) |
3e32cb2e | 1843 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1844 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1845 | stock->nr_pages = 0; |
cdec2e42 KH |
1846 | } |
1847 | stock->cached = NULL; | |
cdec2e42 KH |
1848 | } |
1849 | ||
1850 | /* | |
1851 | * This must be called under preempt disabled or must be called by | |
1852 | * a thread which is pinned to local cpu. | |
1853 | */ | |
1854 | static void drain_local_stock(struct work_struct *dummy) | |
1855 | { | |
7c8e0181 | 1856 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1857 | drain_stock(stock); |
26fe6168 | 1858 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1859 | } |
1860 | ||
1861 | /* | |
3e32cb2e | 1862 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1863 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1864 | */ |
c0ff4b85 | 1865 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1866 | { |
1867 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1868 | ||
c0ff4b85 | 1869 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1870 | drain_stock(stock); |
c0ff4b85 | 1871 | stock->cached = memcg; |
cdec2e42 | 1872 | } |
11c9ea4e | 1873 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1874 | put_cpu_var(memcg_stock); |
1875 | } | |
1876 | ||
1877 | /* | |
c0ff4b85 | 1878 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1879 | * of the hierarchy under it. |
cdec2e42 | 1880 | */ |
6d3d6aa2 | 1881 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1882 | { |
26fe6168 | 1883 | int cpu, curcpu; |
d38144b7 | 1884 | |
6d3d6aa2 JW |
1885 | /* If someone's already draining, avoid adding running more workers. */ |
1886 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1887 | return; | |
cdec2e42 | 1888 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1889 | get_online_cpus(); |
5af12d0e | 1890 | curcpu = get_cpu(); |
cdec2e42 KH |
1891 | for_each_online_cpu(cpu) { |
1892 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1893 | struct mem_cgroup *memcg; |
26fe6168 | 1894 | |
c0ff4b85 R |
1895 | memcg = stock->cached; |
1896 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1897 | continue; |
2314b42d | 1898 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1899 | continue; |
d1a05b69 MH |
1900 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1901 | if (cpu == curcpu) | |
1902 | drain_local_stock(&stock->work); | |
1903 | else | |
1904 | schedule_work_on(cpu, &stock->work); | |
1905 | } | |
cdec2e42 | 1906 | } |
5af12d0e | 1907 | put_cpu(); |
f894ffa8 | 1908 | put_online_cpus(); |
9f50fad6 | 1909 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1910 | } |
1911 | ||
0db0628d | 1912 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1913 | unsigned long action, |
1914 | void *hcpu) | |
1915 | { | |
1916 | int cpu = (unsigned long)hcpu; | |
1917 | struct memcg_stock_pcp *stock; | |
1918 | ||
619d094b | 1919 | if (action == CPU_ONLINE) |
1489ebad | 1920 | return NOTIFY_OK; |
1489ebad | 1921 | |
d833049b | 1922 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1923 | return NOTIFY_OK; |
711d3d2c | 1924 | |
cdec2e42 KH |
1925 | stock = &per_cpu(memcg_stock, cpu); |
1926 | drain_stock(stock); | |
1927 | return NOTIFY_OK; | |
1928 | } | |
1929 | ||
f7e1cb6e JW |
1930 | static void reclaim_high(struct mem_cgroup *memcg, |
1931 | unsigned int nr_pages, | |
1932 | gfp_t gfp_mask) | |
1933 | { | |
1934 | do { | |
1935 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1936 | continue; | |
1937 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
1938 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
1939 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1940 | } | |
1941 | ||
1942 | static void high_work_func(struct work_struct *work) | |
1943 | { | |
1944 | struct mem_cgroup *memcg; | |
1945 | ||
1946 | memcg = container_of(work, struct mem_cgroup, high_work); | |
1947 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); | |
1948 | } | |
1949 | ||
b23afb93 TH |
1950 | /* |
1951 | * Scheduled by try_charge() to be executed from the userland return path | |
1952 | * and reclaims memory over the high limit. | |
1953 | */ | |
1954 | void mem_cgroup_handle_over_high(void) | |
1955 | { | |
1956 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1957 | struct mem_cgroup *memcg; |
b23afb93 TH |
1958 | |
1959 | if (likely(!nr_pages)) | |
1960 | return; | |
1961 | ||
f7e1cb6e JW |
1962 | memcg = get_mem_cgroup_from_mm(current->mm); |
1963 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1964 | css_put(&memcg->css); |
1965 | current->memcg_nr_pages_over_high = 0; | |
1966 | } | |
1967 | ||
00501b53 JW |
1968 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1969 | unsigned int nr_pages) | |
8a9f3ccd | 1970 | { |
7ec99d62 | 1971 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1972 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1973 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1974 | struct page_counter *counter; |
6539cc05 | 1975 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1976 | bool may_swap = true; |
1977 | bool drained = false; | |
a636b327 | 1978 | |
ce00a967 | 1979 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1980 | return 0; |
6539cc05 | 1981 | retry: |
b6b6cc72 | 1982 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1983 | return 0; |
8a9f3ccd | 1984 | |
7941d214 | 1985 | if (!do_memsw_account() || |
6071ca52 JW |
1986 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1987 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1988 | goto done_restock; |
7941d214 | 1989 | if (do_memsw_account()) |
3e32cb2e JW |
1990 | page_counter_uncharge(&memcg->memsw, batch); |
1991 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1992 | } else { |
3e32cb2e | 1993 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1994 | may_swap = false; |
3fbe7244 | 1995 | } |
7a81b88c | 1996 | |
6539cc05 JW |
1997 | if (batch > nr_pages) { |
1998 | batch = nr_pages; | |
1999 | goto retry; | |
2000 | } | |
6d61ef40 | 2001 | |
06b078fc JW |
2002 | /* |
2003 | * Unlike in global OOM situations, memcg is not in a physical | |
2004 | * memory shortage. Allow dying and OOM-killed tasks to | |
2005 | * bypass the last charges so that they can exit quickly and | |
2006 | * free their memory. | |
2007 | */ | |
2008 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2009 | fatal_signal_pending(current) || | |
2010 | current->flags & PF_EXITING)) | |
10d53c74 | 2011 | goto force; |
06b078fc JW |
2012 | |
2013 | if (unlikely(task_in_memcg_oom(current))) | |
2014 | goto nomem; | |
2015 | ||
d0164adc | 2016 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2017 | goto nomem; |
4b534334 | 2018 | |
241994ed JW |
2019 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2020 | ||
b70a2a21 JW |
2021 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2022 | gfp_mask, may_swap); | |
6539cc05 | 2023 | |
61e02c74 | 2024 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2025 | goto retry; |
28c34c29 | 2026 | |
b70a2a21 | 2027 | if (!drained) { |
6d3d6aa2 | 2028 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2029 | drained = true; |
2030 | goto retry; | |
2031 | } | |
2032 | ||
28c34c29 JW |
2033 | if (gfp_mask & __GFP_NORETRY) |
2034 | goto nomem; | |
6539cc05 JW |
2035 | /* |
2036 | * Even though the limit is exceeded at this point, reclaim | |
2037 | * may have been able to free some pages. Retry the charge | |
2038 | * before killing the task. | |
2039 | * | |
2040 | * Only for regular pages, though: huge pages are rather | |
2041 | * unlikely to succeed so close to the limit, and we fall back | |
2042 | * to regular pages anyway in case of failure. | |
2043 | */ | |
61e02c74 | 2044 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2045 | goto retry; |
2046 | /* | |
2047 | * At task move, charge accounts can be doubly counted. So, it's | |
2048 | * better to wait until the end of task_move if something is going on. | |
2049 | */ | |
2050 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2051 | goto retry; | |
2052 | ||
9b130619 JW |
2053 | if (nr_retries--) |
2054 | goto retry; | |
2055 | ||
06b078fc | 2056 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2057 | goto force; |
06b078fc | 2058 | |
6539cc05 | 2059 | if (fatal_signal_pending(current)) |
10d53c74 | 2060 | goto force; |
6539cc05 | 2061 | |
241994ed JW |
2062 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2063 | ||
3608de07 JM |
2064 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2065 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2066 | nomem: |
6d1fdc48 | 2067 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2068 | return -ENOMEM; |
10d53c74 TH |
2069 | force: |
2070 | /* | |
2071 | * The allocation either can't fail or will lead to more memory | |
2072 | * being freed very soon. Allow memory usage go over the limit | |
2073 | * temporarily by force charging it. | |
2074 | */ | |
2075 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2076 | if (do_memsw_account()) |
10d53c74 TH |
2077 | page_counter_charge(&memcg->memsw, nr_pages); |
2078 | css_get_many(&memcg->css, nr_pages); | |
2079 | ||
2080 | return 0; | |
6539cc05 JW |
2081 | |
2082 | done_restock: | |
e8ea14cc | 2083 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2084 | if (batch > nr_pages) |
2085 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2086 | |
241994ed | 2087 | /* |
b23afb93 TH |
2088 | * If the hierarchy is above the normal consumption range, schedule |
2089 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2090 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2091 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2092 | * not recorded as it most likely matches current's and won't | |
2093 | * change in the meantime. As high limit is checked again before | |
2094 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2095 | */ |
2096 | do { | |
b23afb93 | 2097 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2098 | /* Don't bother a random interrupted task */ |
2099 | if (in_interrupt()) { | |
2100 | schedule_work(&memcg->high_work); | |
2101 | break; | |
2102 | } | |
9516a18a | 2103 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2104 | set_notify_resume(current); |
2105 | break; | |
2106 | } | |
241994ed | 2107 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2108 | |
2109 | return 0; | |
7a81b88c | 2110 | } |
8a9f3ccd | 2111 | |
00501b53 | 2112 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2113 | { |
ce00a967 JW |
2114 | if (mem_cgroup_is_root(memcg)) |
2115 | return; | |
2116 | ||
3e32cb2e | 2117 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2118 | if (do_memsw_account()) |
3e32cb2e | 2119 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2120 | |
e8ea14cc | 2121 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2122 | } |
2123 | ||
0a31bc97 JW |
2124 | static void lock_page_lru(struct page *page, int *isolated) |
2125 | { | |
2126 | struct zone *zone = page_zone(page); | |
2127 | ||
2128 | spin_lock_irq(&zone->lru_lock); | |
2129 | if (PageLRU(page)) { | |
2130 | struct lruvec *lruvec; | |
2131 | ||
2132 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2133 | ClearPageLRU(page); | |
2134 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2135 | *isolated = 1; | |
2136 | } else | |
2137 | *isolated = 0; | |
2138 | } | |
2139 | ||
2140 | static void unlock_page_lru(struct page *page, int isolated) | |
2141 | { | |
2142 | struct zone *zone = page_zone(page); | |
2143 | ||
2144 | if (isolated) { | |
2145 | struct lruvec *lruvec; | |
2146 | ||
2147 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2148 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2149 | SetPageLRU(page); | |
2150 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2151 | } | |
2152 | spin_unlock_irq(&zone->lru_lock); | |
2153 | } | |
2154 | ||
00501b53 | 2155 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2156 | bool lrucare) |
7a81b88c | 2157 | { |
0a31bc97 | 2158 | int isolated; |
9ce70c02 | 2159 | |
1306a85a | 2160 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2161 | |
2162 | /* | |
2163 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2164 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2165 | */ | |
0a31bc97 JW |
2166 | if (lrucare) |
2167 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2168 | |
0a31bc97 JW |
2169 | /* |
2170 | * Nobody should be changing or seriously looking at | |
1306a85a | 2171 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2172 | * |
2173 | * - the page is uncharged | |
2174 | * | |
2175 | * - the page is off-LRU | |
2176 | * | |
2177 | * - an anonymous fault has exclusive page access, except for | |
2178 | * a locked page table | |
2179 | * | |
2180 | * - a page cache insertion, a swapin fault, or a migration | |
2181 | * have the page locked | |
2182 | */ | |
1306a85a | 2183 | page->mem_cgroup = memcg; |
9ce70c02 | 2184 | |
0a31bc97 JW |
2185 | if (lrucare) |
2186 | unlock_page_lru(page, isolated); | |
7a81b88c | 2187 | } |
66e1707b | 2188 | |
127424c8 | 2189 | #ifndef CONFIG_SLOB |
f3bb3043 | 2190 | static int memcg_alloc_cache_id(void) |
55007d84 | 2191 | { |
f3bb3043 VD |
2192 | int id, size; |
2193 | int err; | |
2194 | ||
dbcf73e2 | 2195 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2196 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2197 | if (id < 0) | |
2198 | return id; | |
55007d84 | 2199 | |
dbcf73e2 | 2200 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2201 | return id; |
2202 | ||
2203 | /* | |
2204 | * There's no space for the new id in memcg_caches arrays, | |
2205 | * so we have to grow them. | |
2206 | */ | |
05257a1a | 2207 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2208 | |
2209 | size = 2 * (id + 1); | |
55007d84 GC |
2210 | if (size < MEMCG_CACHES_MIN_SIZE) |
2211 | size = MEMCG_CACHES_MIN_SIZE; | |
2212 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2213 | size = MEMCG_CACHES_MAX_SIZE; | |
2214 | ||
f3bb3043 | 2215 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2216 | if (!err) |
2217 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2218 | if (!err) |
2219 | memcg_nr_cache_ids = size; | |
2220 | ||
2221 | up_write(&memcg_cache_ids_sem); | |
2222 | ||
f3bb3043 | 2223 | if (err) { |
dbcf73e2 | 2224 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2225 | return err; |
2226 | } | |
2227 | return id; | |
2228 | } | |
2229 | ||
2230 | static void memcg_free_cache_id(int id) | |
2231 | { | |
dbcf73e2 | 2232 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2233 | } |
2234 | ||
d5b3cf71 | 2235 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2236 | struct mem_cgroup *memcg; |
2237 | struct kmem_cache *cachep; | |
2238 | struct work_struct work; | |
2239 | }; | |
2240 | ||
d5b3cf71 | 2241 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2242 | { |
d5b3cf71 VD |
2243 | struct memcg_kmem_cache_create_work *cw = |
2244 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2245 | struct mem_cgroup *memcg = cw->memcg; |
2246 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2247 | |
d5b3cf71 | 2248 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2249 | |
5722d094 | 2250 | css_put(&memcg->css); |
d7f25f8a GC |
2251 | kfree(cw); |
2252 | } | |
2253 | ||
2254 | /* | |
2255 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2256 | */ |
d5b3cf71 VD |
2257 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2258 | struct kmem_cache *cachep) | |
d7f25f8a | 2259 | { |
d5b3cf71 | 2260 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2261 | |
776ed0f0 | 2262 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2263 | if (!cw) |
d7f25f8a | 2264 | return; |
8135be5a VD |
2265 | |
2266 | css_get(&memcg->css); | |
d7f25f8a GC |
2267 | |
2268 | cw->memcg = memcg; | |
2269 | cw->cachep = cachep; | |
d5b3cf71 | 2270 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2271 | |
d7f25f8a GC |
2272 | schedule_work(&cw->work); |
2273 | } | |
2274 | ||
d5b3cf71 VD |
2275 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2276 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2277 | { |
2278 | /* | |
2279 | * We need to stop accounting when we kmalloc, because if the | |
2280 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2281 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2282 | * |
2283 | * However, it is better to enclose the whole function. Depending on | |
2284 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2285 | * trigger an allocation. This too, will make us recurse. Because at | |
2286 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2287 | * the safest choice is to do it like this, wrapping the whole function. | |
2288 | */ | |
6f185c29 | 2289 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2290 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2291 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2292 | } |
c67a8a68 | 2293 | |
d7f25f8a GC |
2294 | /* |
2295 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2296 | * We try to use the current memcg's version of the cache. | |
2297 | * | |
2298 | * If the cache does not exist yet, if we are the first user of it, | |
2299 | * we either create it immediately, if possible, or create it asynchronously | |
2300 | * in a workqueue. | |
2301 | * In the latter case, we will let the current allocation go through with | |
2302 | * the original cache. | |
2303 | * | |
2304 | * Can't be called in interrupt context or from kernel threads. | |
2305 | * This function needs to be called with rcu_read_lock() held. | |
2306 | */ | |
230e9fc2 | 2307 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp) |
d7f25f8a GC |
2308 | { |
2309 | struct mem_cgroup *memcg; | |
959c8963 | 2310 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2311 | int kmemcg_id; |
d7f25f8a | 2312 | |
f7ce3190 | 2313 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2314 | |
230e9fc2 VD |
2315 | if (cachep->flags & SLAB_ACCOUNT) |
2316 | gfp |= __GFP_ACCOUNT; | |
2317 | ||
2318 | if (!(gfp & __GFP_ACCOUNT)) | |
2319 | return cachep; | |
2320 | ||
9d100c5e | 2321 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2322 | return cachep; |
2323 | ||
8135be5a | 2324 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2325 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2326 | if (kmemcg_id < 0) |
ca0dde97 | 2327 | goto out; |
d7f25f8a | 2328 | |
2a4db7eb | 2329 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2330 | if (likely(memcg_cachep)) |
2331 | return memcg_cachep; | |
ca0dde97 LZ |
2332 | |
2333 | /* | |
2334 | * If we are in a safe context (can wait, and not in interrupt | |
2335 | * context), we could be be predictable and return right away. | |
2336 | * This would guarantee that the allocation being performed | |
2337 | * already belongs in the new cache. | |
2338 | * | |
2339 | * However, there are some clashes that can arrive from locking. | |
2340 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2341 | * memcg_create_kmem_cache, this means no further allocation |
2342 | * could happen with the slab_mutex held. So it's better to | |
2343 | * defer everything. | |
ca0dde97 | 2344 | */ |
d5b3cf71 | 2345 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2346 | out: |
8135be5a | 2347 | css_put(&memcg->css); |
ca0dde97 | 2348 | return cachep; |
d7f25f8a | 2349 | } |
d7f25f8a | 2350 | |
8135be5a VD |
2351 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2352 | { | |
2353 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2354 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2355 | } |
2356 | ||
f3ccb2c4 VD |
2357 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
2358 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2359 | { |
f3ccb2c4 VD |
2360 | unsigned int nr_pages = 1 << order; |
2361 | struct page_counter *counter; | |
7ae1e1d0 GC |
2362 | int ret; |
2363 | ||
567e9ab2 | 2364 | if (!memcg_kmem_online(memcg)) |
d05e83a6 | 2365 | return 0; |
6d42c232 | 2366 | |
f3ccb2c4 | 2367 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2368 | if (ret) |
f3ccb2c4 | 2369 | return ret; |
52c29b04 JW |
2370 | |
2371 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2372 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2373 | cancel_charge(memcg, nr_pages); | |
2374 | return -ENOMEM; | |
7ae1e1d0 GC |
2375 | } |
2376 | ||
f3ccb2c4 | 2377 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2378 | |
f3ccb2c4 | 2379 | return 0; |
7ae1e1d0 GC |
2380 | } |
2381 | ||
f3ccb2c4 | 2382 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2383 | { |
f3ccb2c4 VD |
2384 | struct mem_cgroup *memcg; |
2385 | int ret; | |
7ae1e1d0 | 2386 | |
f3ccb2c4 VD |
2387 | memcg = get_mem_cgroup_from_mm(current->mm); |
2388 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); | |
7ae1e1d0 | 2389 | css_put(&memcg->css); |
d05e83a6 | 2390 | return ret; |
7ae1e1d0 GC |
2391 | } |
2392 | ||
d05e83a6 | 2393 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2394 | { |
1306a85a | 2395 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2396 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2397 | |
7ae1e1d0 GC |
2398 | if (!memcg) |
2399 | return; | |
2400 | ||
309381fe | 2401 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2402 | |
52c29b04 JW |
2403 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2404 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2405 | ||
f3ccb2c4 | 2406 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2407 | if (do_memsw_account()) |
f3ccb2c4 | 2408 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2409 | |
1306a85a | 2410 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2411 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2412 | } |
127424c8 | 2413 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2414 | |
ca3e0214 KH |
2415 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2416 | ||
ca3e0214 KH |
2417 | /* |
2418 | * Because tail pages are not marked as "used", set it. We're under | |
3ac808fd | 2419 | * zone->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2420 | */ |
e94c8a9c | 2421 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2422 | { |
e94c8a9c | 2423 | int i; |
ca3e0214 | 2424 | |
3d37c4a9 KH |
2425 | if (mem_cgroup_disabled()) |
2426 | return; | |
b070e65c | 2427 | |
29833315 | 2428 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2429 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2430 | |
1306a85a | 2431 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2432 | HPAGE_PMD_NR); |
ca3e0214 | 2433 | } |
12d27107 | 2434 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2435 | |
c255a458 | 2436 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2437 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2438 | bool charge) | |
d13d1443 | 2439 | { |
0a31bc97 JW |
2440 | int val = (charge) ? 1 : -1; |
2441 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2442 | } |
02491447 DN |
2443 | |
2444 | /** | |
2445 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2446 | * @entry: swap entry to be moved | |
2447 | * @from: mem_cgroup which the entry is moved from | |
2448 | * @to: mem_cgroup which the entry is moved to | |
2449 | * | |
2450 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2451 | * as the mem_cgroup's id of @from. | |
2452 | * | |
2453 | * Returns 0 on success, -EINVAL on failure. | |
2454 | * | |
3e32cb2e | 2455 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2456 | * both res and memsw, and called css_get(). |
2457 | */ | |
2458 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2459 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2460 | { |
2461 | unsigned short old_id, new_id; | |
2462 | ||
34c00c31 LZ |
2463 | old_id = mem_cgroup_id(from); |
2464 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2465 | |
2466 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2467 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2468 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2469 | return 0; |
2470 | } | |
2471 | return -EINVAL; | |
2472 | } | |
2473 | #else | |
2474 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2475 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2476 | { |
2477 | return -EINVAL; | |
2478 | } | |
8c7c6e34 | 2479 | #endif |
d13d1443 | 2480 | |
3e32cb2e | 2481 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2482 | |
d38d2a75 | 2483 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2484 | unsigned long limit) |
628f4235 | 2485 | { |
3e32cb2e JW |
2486 | unsigned long curusage; |
2487 | unsigned long oldusage; | |
2488 | bool enlarge = false; | |
81d39c20 | 2489 | int retry_count; |
3e32cb2e | 2490 | int ret; |
81d39c20 KH |
2491 | |
2492 | /* | |
2493 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2494 | * is depends on callers. We set our retry-count to be function | |
2495 | * of # of children which we should visit in this loop. | |
2496 | */ | |
3e32cb2e JW |
2497 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2498 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2499 | |
3e32cb2e | 2500 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2501 | |
3e32cb2e | 2502 | do { |
628f4235 KH |
2503 | if (signal_pending(current)) { |
2504 | ret = -EINTR; | |
2505 | break; | |
2506 | } | |
3e32cb2e JW |
2507 | |
2508 | mutex_lock(&memcg_limit_mutex); | |
2509 | if (limit > memcg->memsw.limit) { | |
2510 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2511 | ret = -EINVAL; |
628f4235 KH |
2512 | break; |
2513 | } | |
3e32cb2e JW |
2514 | if (limit > memcg->memory.limit) |
2515 | enlarge = true; | |
2516 | ret = page_counter_limit(&memcg->memory, limit); | |
2517 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2518 | |
2519 | if (!ret) | |
2520 | break; | |
2521 | ||
b70a2a21 JW |
2522 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2523 | ||
3e32cb2e | 2524 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2525 | /* Usage is reduced ? */ |
f894ffa8 | 2526 | if (curusage >= oldusage) |
81d39c20 KH |
2527 | retry_count--; |
2528 | else | |
2529 | oldusage = curusage; | |
3e32cb2e JW |
2530 | } while (retry_count); |
2531 | ||
3c11ecf4 KH |
2532 | if (!ret && enlarge) |
2533 | memcg_oom_recover(memcg); | |
14797e23 | 2534 | |
8c7c6e34 KH |
2535 | return ret; |
2536 | } | |
2537 | ||
338c8431 | 2538 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2539 | unsigned long limit) |
8c7c6e34 | 2540 | { |
3e32cb2e JW |
2541 | unsigned long curusage; |
2542 | unsigned long oldusage; | |
2543 | bool enlarge = false; | |
81d39c20 | 2544 | int retry_count; |
3e32cb2e | 2545 | int ret; |
8c7c6e34 | 2546 | |
81d39c20 | 2547 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2548 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2549 | mem_cgroup_count_children(memcg); | |
2550 | ||
2551 | oldusage = page_counter_read(&memcg->memsw); | |
2552 | ||
2553 | do { | |
8c7c6e34 KH |
2554 | if (signal_pending(current)) { |
2555 | ret = -EINTR; | |
2556 | break; | |
2557 | } | |
3e32cb2e JW |
2558 | |
2559 | mutex_lock(&memcg_limit_mutex); | |
2560 | if (limit < memcg->memory.limit) { | |
2561 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2562 | ret = -EINVAL; |
8c7c6e34 KH |
2563 | break; |
2564 | } | |
3e32cb2e JW |
2565 | if (limit > memcg->memsw.limit) |
2566 | enlarge = true; | |
2567 | ret = page_counter_limit(&memcg->memsw, limit); | |
2568 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2569 | |
2570 | if (!ret) | |
2571 | break; | |
2572 | ||
b70a2a21 JW |
2573 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2574 | ||
3e32cb2e | 2575 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2576 | /* Usage is reduced ? */ |
8c7c6e34 | 2577 | if (curusage >= oldusage) |
628f4235 | 2578 | retry_count--; |
81d39c20 KH |
2579 | else |
2580 | oldusage = curusage; | |
3e32cb2e JW |
2581 | } while (retry_count); |
2582 | ||
3c11ecf4 KH |
2583 | if (!ret && enlarge) |
2584 | memcg_oom_recover(memcg); | |
3e32cb2e | 2585 | |
628f4235 KH |
2586 | return ret; |
2587 | } | |
2588 | ||
0608f43d AM |
2589 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2590 | gfp_t gfp_mask, | |
2591 | unsigned long *total_scanned) | |
2592 | { | |
2593 | unsigned long nr_reclaimed = 0; | |
2594 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2595 | unsigned long reclaimed; | |
2596 | int loop = 0; | |
2597 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2598 | unsigned long excess; |
0608f43d AM |
2599 | unsigned long nr_scanned; |
2600 | ||
2601 | if (order > 0) | |
2602 | return 0; | |
2603 | ||
2604 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2605 | /* | |
2606 | * This loop can run a while, specially if mem_cgroup's continuously | |
2607 | * keep exceeding their soft limit and putting the system under | |
2608 | * pressure | |
2609 | */ | |
2610 | do { | |
2611 | if (next_mz) | |
2612 | mz = next_mz; | |
2613 | else | |
2614 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2615 | if (!mz) | |
2616 | break; | |
2617 | ||
2618 | nr_scanned = 0; | |
2619 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2620 | gfp_mask, &nr_scanned); | |
2621 | nr_reclaimed += reclaimed; | |
2622 | *total_scanned += nr_scanned; | |
0a31bc97 | 2623 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2624 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2625 | |
2626 | /* | |
2627 | * If we failed to reclaim anything from this memory cgroup | |
2628 | * it is time to move on to the next cgroup | |
2629 | */ | |
2630 | next_mz = NULL; | |
bc2f2e7f VD |
2631 | if (!reclaimed) |
2632 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2633 | ||
3e32cb2e | 2634 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2635 | /* |
2636 | * One school of thought says that we should not add | |
2637 | * back the node to the tree if reclaim returns 0. | |
2638 | * But our reclaim could return 0, simply because due | |
2639 | * to priority we are exposing a smaller subset of | |
2640 | * memory to reclaim from. Consider this as a longer | |
2641 | * term TODO. | |
2642 | */ | |
2643 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2644 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2645 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2646 | css_put(&mz->memcg->css); |
2647 | loop++; | |
2648 | /* | |
2649 | * Could not reclaim anything and there are no more | |
2650 | * mem cgroups to try or we seem to be looping without | |
2651 | * reclaiming anything. | |
2652 | */ | |
2653 | if (!nr_reclaimed && | |
2654 | (next_mz == NULL || | |
2655 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2656 | break; | |
2657 | } while (!nr_reclaimed); | |
2658 | if (next_mz) | |
2659 | css_put(&next_mz->memcg->css); | |
2660 | return nr_reclaimed; | |
2661 | } | |
2662 | ||
ea280e7b TH |
2663 | /* |
2664 | * Test whether @memcg has children, dead or alive. Note that this | |
2665 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2666 | * returns %true if there are child csses according to the cgroup | |
2667 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2668 | */ | |
b5f99b53 GC |
2669 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2670 | { | |
ea280e7b TH |
2671 | bool ret; |
2672 | ||
ea280e7b TH |
2673 | rcu_read_lock(); |
2674 | ret = css_next_child(NULL, &memcg->css); | |
2675 | rcu_read_unlock(); | |
2676 | return ret; | |
b5f99b53 GC |
2677 | } |
2678 | ||
c26251f9 MH |
2679 | /* |
2680 | * Reclaims as many pages from the given memcg as possible and moves | |
2681 | * the rest to the parent. | |
2682 | * | |
2683 | * Caller is responsible for holding css reference for memcg. | |
2684 | */ | |
2685 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2686 | { | |
2687 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2688 | |
c1e862c1 KH |
2689 | /* we call try-to-free pages for make this cgroup empty */ |
2690 | lru_add_drain_all(); | |
f817ed48 | 2691 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2692 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2693 | int progress; |
c1e862c1 | 2694 | |
c26251f9 MH |
2695 | if (signal_pending(current)) |
2696 | return -EINTR; | |
2697 | ||
b70a2a21 JW |
2698 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2699 | GFP_KERNEL, true); | |
c1e862c1 | 2700 | if (!progress) { |
f817ed48 | 2701 | nr_retries--; |
c1e862c1 | 2702 | /* maybe some writeback is necessary */ |
8aa7e847 | 2703 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2704 | } |
f817ed48 KH |
2705 | |
2706 | } | |
ab5196c2 MH |
2707 | |
2708 | return 0; | |
cc847582 KH |
2709 | } |
2710 | ||
6770c64e TH |
2711 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2712 | char *buf, size_t nbytes, | |
2713 | loff_t off) | |
c1e862c1 | 2714 | { |
6770c64e | 2715 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2716 | |
d8423011 MH |
2717 | if (mem_cgroup_is_root(memcg)) |
2718 | return -EINVAL; | |
6770c64e | 2719 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2720 | } |
2721 | ||
182446d0 TH |
2722 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2723 | struct cftype *cft) | |
18f59ea7 | 2724 | { |
182446d0 | 2725 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2726 | } |
2727 | ||
182446d0 TH |
2728 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2729 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2730 | { |
2731 | int retval = 0; | |
182446d0 | 2732 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2733 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2734 | |
567fb435 | 2735 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2736 | return 0; |
567fb435 | 2737 | |
18f59ea7 | 2738 | /* |
af901ca1 | 2739 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2740 | * in the child subtrees. If it is unset, then the change can |
2741 | * occur, provided the current cgroup has no children. | |
2742 | * | |
2743 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2744 | * set if there are no children. | |
2745 | */ | |
c0ff4b85 | 2746 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2747 | (val == 1 || val == 0)) { |
ea280e7b | 2748 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2749 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2750 | else |
2751 | retval = -EBUSY; | |
2752 | } else | |
2753 | retval = -EINVAL; | |
567fb435 | 2754 | |
18f59ea7 BS |
2755 | return retval; |
2756 | } | |
2757 | ||
3e32cb2e JW |
2758 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
2759 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
2760 | { |
2761 | struct mem_cgroup *iter; | |
484ebb3b | 2762 | unsigned long val = 0; |
ce00a967 | 2763 | |
ce00a967 JW |
2764 | for_each_mem_cgroup_tree(iter, memcg) |
2765 | val += mem_cgroup_read_stat(iter, idx); | |
2766 | ||
ce00a967 JW |
2767 | return val; |
2768 | } | |
2769 | ||
6f646156 | 2770 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2771 | { |
c12176d3 | 2772 | unsigned long val; |
ce00a967 | 2773 | |
3e32cb2e JW |
2774 | if (mem_cgroup_is_root(memcg)) { |
2775 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
2776 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
2777 | if (swap) | |
2778 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
2779 | } else { | |
ce00a967 | 2780 | if (!swap) |
3e32cb2e | 2781 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2782 | else |
3e32cb2e | 2783 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2784 | } |
c12176d3 | 2785 | return val; |
ce00a967 JW |
2786 | } |
2787 | ||
3e32cb2e JW |
2788 | enum { |
2789 | RES_USAGE, | |
2790 | RES_LIMIT, | |
2791 | RES_MAX_USAGE, | |
2792 | RES_FAILCNT, | |
2793 | RES_SOFT_LIMIT, | |
2794 | }; | |
ce00a967 | 2795 | |
791badbd | 2796 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2797 | struct cftype *cft) |
8cdea7c0 | 2798 | { |
182446d0 | 2799 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2800 | struct page_counter *counter; |
af36f906 | 2801 | |
3e32cb2e | 2802 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2803 | case _MEM: |
3e32cb2e JW |
2804 | counter = &memcg->memory; |
2805 | break; | |
8c7c6e34 | 2806 | case _MEMSWAP: |
3e32cb2e JW |
2807 | counter = &memcg->memsw; |
2808 | break; | |
510fc4e1 | 2809 | case _KMEM: |
3e32cb2e | 2810 | counter = &memcg->kmem; |
510fc4e1 | 2811 | break; |
d55f90bf | 2812 | case _TCP: |
0db15298 | 2813 | counter = &memcg->tcpmem; |
d55f90bf | 2814 | break; |
8c7c6e34 KH |
2815 | default: |
2816 | BUG(); | |
8c7c6e34 | 2817 | } |
3e32cb2e JW |
2818 | |
2819 | switch (MEMFILE_ATTR(cft->private)) { | |
2820 | case RES_USAGE: | |
2821 | if (counter == &memcg->memory) | |
c12176d3 | 2822 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2823 | if (counter == &memcg->memsw) |
c12176d3 | 2824 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2825 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2826 | case RES_LIMIT: | |
2827 | return (u64)counter->limit * PAGE_SIZE; | |
2828 | case RES_MAX_USAGE: | |
2829 | return (u64)counter->watermark * PAGE_SIZE; | |
2830 | case RES_FAILCNT: | |
2831 | return counter->failcnt; | |
2832 | case RES_SOFT_LIMIT: | |
2833 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2834 | default: | |
2835 | BUG(); | |
2836 | } | |
8cdea7c0 | 2837 | } |
510fc4e1 | 2838 | |
127424c8 | 2839 | #ifndef CONFIG_SLOB |
567e9ab2 | 2840 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2841 | { |
d6441637 VD |
2842 | int memcg_id; |
2843 | ||
2a4db7eb | 2844 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2845 | BUG_ON(memcg->kmem_state); |
d6441637 | 2846 | |
f3bb3043 | 2847 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
2848 | if (memcg_id < 0) |
2849 | return memcg_id; | |
d6441637 | 2850 | |
ef12947c | 2851 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2852 | /* |
567e9ab2 | 2853 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2854 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2855 | * guarantee no one starts accounting before all call sites are |
2856 | * patched. | |
2857 | */ | |
900a38f0 | 2858 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2859 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
2860 | |
2861 | return 0; | |
d6441637 VD |
2862 | } |
2863 | ||
0b8f73e1 JW |
2864 | static int memcg_propagate_kmem(struct mem_cgroup *parent, |
2865 | struct mem_cgroup *memcg) | |
510fc4e1 | 2866 | { |
55007d84 | 2867 | int ret = 0; |
55007d84 | 2868 | |
8c0145b6 | 2869 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 2870 | /* |
567e9ab2 JW |
2871 | * If the parent cgroup is not kmem-online now, it cannot be |
2872 | * onlined after this point, because it has at least one child | |
2873 | * already. | |
55007d84 | 2874 | */ |
04823c83 VD |
2875 | if (memcg_kmem_online(parent) || |
2876 | (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nokmem)) | |
567e9ab2 | 2877 | ret = memcg_online_kmem(memcg); |
8c0145b6 | 2878 | mutex_unlock(&memcg_limit_mutex); |
55007d84 | 2879 | return ret; |
510fc4e1 | 2880 | } |
8e0a8912 | 2881 | |
8e0a8912 JW |
2882 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2883 | { | |
2884 | struct cgroup_subsys_state *css; | |
2885 | struct mem_cgroup *parent, *child; | |
2886 | int kmemcg_id; | |
2887 | ||
2888 | if (memcg->kmem_state != KMEM_ONLINE) | |
2889 | return; | |
2890 | /* | |
2891 | * Clear the online state before clearing memcg_caches array | |
2892 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2893 | * guarantees that no cache will be created for this cgroup | |
2894 | * after we are done (see memcg_create_kmem_cache()). | |
2895 | */ | |
2896 | memcg->kmem_state = KMEM_ALLOCATED; | |
2897 | ||
2898 | memcg_deactivate_kmem_caches(memcg); | |
2899 | ||
2900 | kmemcg_id = memcg->kmemcg_id; | |
2901 | BUG_ON(kmemcg_id < 0); | |
2902 | ||
2903 | parent = parent_mem_cgroup(memcg); | |
2904 | if (!parent) | |
2905 | parent = root_mem_cgroup; | |
2906 | ||
2907 | /* | |
2908 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2909 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2910 | * to ones of the parent. After we have finished, all list_lrus | |
2911 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2912 | * ordering is imposed by list_lru_node->lock taken by | |
2913 | * memcg_drain_all_list_lrus(). | |
2914 | */ | |
2915 | css_for_each_descendant_pre(css, &memcg->css) { | |
2916 | child = mem_cgroup_from_css(css); | |
2917 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2918 | child->kmemcg_id = parent->kmemcg_id; | |
2919 | if (!memcg->use_hierarchy) | |
2920 | break; | |
2921 | } | |
2922 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
2923 | ||
2924 | memcg_free_cache_id(kmemcg_id); | |
2925 | } | |
2926 | ||
2927 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2928 | { | |
0b8f73e1 JW |
2929 | /* css_alloc() failed, offlining didn't happen */ |
2930 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
2931 | memcg_offline_kmem(memcg); | |
2932 | ||
8e0a8912 JW |
2933 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
2934 | memcg_destroy_kmem_caches(memcg); | |
2935 | static_branch_dec(&memcg_kmem_enabled_key); | |
2936 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2937 | } | |
8e0a8912 | 2938 | } |
d6441637 | 2939 | #else |
0b8f73e1 JW |
2940 | static int memcg_propagate_kmem(struct mem_cgroup *parent, struct mem_cgroup *memcg) |
2941 | { | |
2942 | return 0; | |
2943 | } | |
2944 | static int memcg_online_kmem(struct mem_cgroup *memcg) | |
127424c8 JW |
2945 | { |
2946 | return 0; | |
2947 | } | |
2948 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
2949 | { | |
2950 | } | |
2951 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2952 | { | |
2953 | } | |
2954 | #endif /* !CONFIG_SLOB */ | |
2955 | ||
d6441637 | 2956 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2957 | unsigned long limit) |
d6441637 | 2958 | { |
0b8f73e1 | 2959 | int ret = 0; |
127424c8 JW |
2960 | |
2961 | mutex_lock(&memcg_limit_mutex); | |
2962 | /* Top-level cgroup doesn't propagate from root */ | |
2963 | if (!memcg_kmem_online(memcg)) { | |
0b8f73e1 JW |
2964 | if (cgroup_is_populated(memcg->css.cgroup) || |
2965 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
2966 | ret = -EBUSY; | |
2967 | if (ret) | |
2968 | goto out; | |
127424c8 JW |
2969 | ret = memcg_online_kmem(memcg); |
2970 | if (ret) | |
2971 | goto out; | |
2972 | } | |
2973 | ret = page_counter_limit(&memcg->kmem, limit); | |
2974 | out: | |
2975 | mutex_unlock(&memcg_limit_mutex); | |
2976 | return ret; | |
d6441637 | 2977 | } |
510fc4e1 | 2978 | |
d55f90bf VD |
2979 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
2980 | { | |
2981 | int ret; | |
2982 | ||
2983 | mutex_lock(&memcg_limit_mutex); | |
2984 | ||
0db15298 | 2985 | ret = page_counter_limit(&memcg->tcpmem, limit); |
d55f90bf VD |
2986 | if (ret) |
2987 | goto out; | |
2988 | ||
0db15298 | 2989 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
2990 | /* |
2991 | * The active flag needs to be written after the static_key | |
2992 | * update. This is what guarantees that the socket activation | |
2993 | * function is the last one to run. See sock_update_memcg() for | |
2994 | * details, and note that we don't mark any socket as belonging | |
2995 | * to this memcg until that flag is up. | |
2996 | * | |
2997 | * We need to do this, because static_keys will span multiple | |
2998 | * sites, but we can't control their order. If we mark a socket | |
2999 | * as accounted, but the accounting functions are not patched in | |
3000 | * yet, we'll lose accounting. | |
3001 | * | |
3002 | * We never race with the readers in sock_update_memcg(), | |
3003 | * because when this value change, the code to process it is not | |
3004 | * patched in yet. | |
3005 | */ | |
3006 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3007 | memcg->tcpmem_active = true; |
d55f90bf VD |
3008 | } |
3009 | out: | |
3010 | mutex_unlock(&memcg_limit_mutex); | |
3011 | return ret; | |
3012 | } | |
d55f90bf | 3013 | |
628f4235 KH |
3014 | /* |
3015 | * The user of this function is... | |
3016 | * RES_LIMIT. | |
3017 | */ | |
451af504 TH |
3018 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3019 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3020 | { |
451af504 | 3021 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3022 | unsigned long nr_pages; |
628f4235 KH |
3023 | int ret; |
3024 | ||
451af504 | 3025 | buf = strstrip(buf); |
650c5e56 | 3026 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3027 | if (ret) |
3028 | return ret; | |
af36f906 | 3029 | |
3e32cb2e | 3030 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3031 | case RES_LIMIT: |
4b3bde4c BS |
3032 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3033 | ret = -EINVAL; | |
3034 | break; | |
3035 | } | |
3e32cb2e JW |
3036 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3037 | case _MEM: | |
3038 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3039 | break; |
3e32cb2e JW |
3040 | case _MEMSWAP: |
3041 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3042 | break; |
3e32cb2e JW |
3043 | case _KMEM: |
3044 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3045 | break; | |
d55f90bf VD |
3046 | case _TCP: |
3047 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
3048 | break; | |
3e32cb2e | 3049 | } |
296c81d8 | 3050 | break; |
3e32cb2e JW |
3051 | case RES_SOFT_LIMIT: |
3052 | memcg->soft_limit = nr_pages; | |
3053 | ret = 0; | |
628f4235 KH |
3054 | break; |
3055 | } | |
451af504 | 3056 | return ret ?: nbytes; |
8cdea7c0 BS |
3057 | } |
3058 | ||
6770c64e TH |
3059 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3060 | size_t nbytes, loff_t off) | |
c84872e1 | 3061 | { |
6770c64e | 3062 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3063 | struct page_counter *counter; |
c84872e1 | 3064 | |
3e32cb2e JW |
3065 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3066 | case _MEM: | |
3067 | counter = &memcg->memory; | |
3068 | break; | |
3069 | case _MEMSWAP: | |
3070 | counter = &memcg->memsw; | |
3071 | break; | |
3072 | case _KMEM: | |
3073 | counter = &memcg->kmem; | |
3074 | break; | |
d55f90bf | 3075 | case _TCP: |
0db15298 | 3076 | counter = &memcg->tcpmem; |
d55f90bf | 3077 | break; |
3e32cb2e JW |
3078 | default: |
3079 | BUG(); | |
3080 | } | |
af36f906 | 3081 | |
3e32cb2e | 3082 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3083 | case RES_MAX_USAGE: |
3e32cb2e | 3084 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3085 | break; |
3086 | case RES_FAILCNT: | |
3e32cb2e | 3087 | counter->failcnt = 0; |
29f2a4da | 3088 | break; |
3e32cb2e JW |
3089 | default: |
3090 | BUG(); | |
29f2a4da | 3091 | } |
f64c3f54 | 3092 | |
6770c64e | 3093 | return nbytes; |
c84872e1 PE |
3094 | } |
3095 | ||
182446d0 | 3096 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3097 | struct cftype *cft) |
3098 | { | |
182446d0 | 3099 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3100 | } |
3101 | ||
02491447 | 3102 | #ifdef CONFIG_MMU |
182446d0 | 3103 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3104 | struct cftype *cft, u64 val) |
3105 | { | |
182446d0 | 3106 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3107 | |
1dfab5ab | 3108 | if (val & ~MOVE_MASK) |
7dc74be0 | 3109 | return -EINVAL; |
ee5e8472 | 3110 | |
7dc74be0 | 3111 | /* |
ee5e8472 GC |
3112 | * No kind of locking is needed in here, because ->can_attach() will |
3113 | * check this value once in the beginning of the process, and then carry | |
3114 | * on with stale data. This means that changes to this value will only | |
3115 | * affect task migrations starting after the change. | |
7dc74be0 | 3116 | */ |
c0ff4b85 | 3117 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3118 | return 0; |
3119 | } | |
02491447 | 3120 | #else |
182446d0 | 3121 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3122 | struct cftype *cft, u64 val) |
3123 | { | |
3124 | return -ENOSYS; | |
3125 | } | |
3126 | #endif | |
7dc74be0 | 3127 | |
406eb0c9 | 3128 | #ifdef CONFIG_NUMA |
2da8ca82 | 3129 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3130 | { |
25485de6 GT |
3131 | struct numa_stat { |
3132 | const char *name; | |
3133 | unsigned int lru_mask; | |
3134 | }; | |
3135 | ||
3136 | static const struct numa_stat stats[] = { | |
3137 | { "total", LRU_ALL }, | |
3138 | { "file", LRU_ALL_FILE }, | |
3139 | { "anon", LRU_ALL_ANON }, | |
3140 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3141 | }; | |
3142 | const struct numa_stat *stat; | |
406eb0c9 | 3143 | int nid; |
25485de6 | 3144 | unsigned long nr; |
2da8ca82 | 3145 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3146 | |
25485de6 GT |
3147 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3148 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3149 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3150 | for_each_node_state(nid, N_MEMORY) { | |
3151 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3152 | stat->lru_mask); | |
3153 | seq_printf(m, " N%d=%lu", nid, nr); | |
3154 | } | |
3155 | seq_putc(m, '\n'); | |
406eb0c9 | 3156 | } |
406eb0c9 | 3157 | |
071aee13 YH |
3158 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3159 | struct mem_cgroup *iter; | |
3160 | ||
3161 | nr = 0; | |
3162 | for_each_mem_cgroup_tree(iter, memcg) | |
3163 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3164 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3165 | for_each_node_state(nid, N_MEMORY) { | |
3166 | nr = 0; | |
3167 | for_each_mem_cgroup_tree(iter, memcg) | |
3168 | nr += mem_cgroup_node_nr_lru_pages( | |
3169 | iter, nid, stat->lru_mask); | |
3170 | seq_printf(m, " N%d=%lu", nid, nr); | |
3171 | } | |
3172 | seq_putc(m, '\n'); | |
406eb0c9 | 3173 | } |
406eb0c9 | 3174 | |
406eb0c9 YH |
3175 | return 0; |
3176 | } | |
3177 | #endif /* CONFIG_NUMA */ | |
3178 | ||
2da8ca82 | 3179 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3180 | { |
2da8ca82 | 3181 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3182 | unsigned long memory, memsw; |
af7c4b0e JW |
3183 | struct mem_cgroup *mi; |
3184 | unsigned int i; | |
406eb0c9 | 3185 | |
0ca44b14 GT |
3186 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3187 | MEM_CGROUP_STAT_NSTATS); | |
3188 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3189 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3190 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3191 | ||
af7c4b0e | 3192 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
7941d214 | 3193 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3194 | continue; |
484ebb3b | 3195 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3196 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3197 | } |
7b854121 | 3198 | |
af7c4b0e JW |
3199 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3200 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3201 | mem_cgroup_read_events(memcg, i)); | |
3202 | ||
3203 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3204 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3205 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3206 | ||
14067bb3 | 3207 | /* Hierarchical information */ |
3e32cb2e JW |
3208 | memory = memsw = PAGE_COUNTER_MAX; |
3209 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3210 | memory = min(memory, mi->memory.limit); | |
3211 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3212 | } |
3e32cb2e JW |
3213 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3214 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3215 | if (do_memsw_account()) |
3e32cb2e JW |
3216 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3217 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3218 | |
af7c4b0e | 3219 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3220 | unsigned long long val = 0; |
af7c4b0e | 3221 | |
7941d214 | 3222 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3223 | continue; |
af7c4b0e JW |
3224 | for_each_mem_cgroup_tree(mi, memcg) |
3225 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3226 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3227 | } |
3228 | ||
3229 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3230 | unsigned long long val = 0; | |
3231 | ||
3232 | for_each_mem_cgroup_tree(mi, memcg) | |
3233 | val += mem_cgroup_read_events(mi, i); | |
3234 | seq_printf(m, "total_%s %llu\n", | |
3235 | mem_cgroup_events_names[i], val); | |
3236 | } | |
3237 | ||
3238 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3239 | unsigned long long val = 0; | |
3240 | ||
3241 | for_each_mem_cgroup_tree(mi, memcg) | |
3242 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3243 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3244 | } |
14067bb3 | 3245 | |
7f016ee8 | 3246 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3247 | { |
3248 | int nid, zid; | |
3249 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3250 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3251 | unsigned long recent_rotated[2] = {0, 0}; |
3252 | unsigned long recent_scanned[2] = {0, 0}; | |
3253 | ||
3254 | for_each_online_node(nid) | |
3255 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3256 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3257 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3258 | |
89abfab1 HD |
3259 | recent_rotated[0] += rstat->recent_rotated[0]; |
3260 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3261 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3262 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3263 | } |
78ccf5b5 JW |
3264 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3265 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3266 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3267 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3268 | } |
3269 | #endif | |
3270 | ||
d2ceb9b7 KH |
3271 | return 0; |
3272 | } | |
3273 | ||
182446d0 TH |
3274 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3275 | struct cftype *cft) | |
a7885eb8 | 3276 | { |
182446d0 | 3277 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3278 | |
1f4c025b | 3279 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3280 | } |
3281 | ||
182446d0 TH |
3282 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3283 | struct cftype *cft, u64 val) | |
a7885eb8 | 3284 | { |
182446d0 | 3285 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3286 | |
3dae7fec | 3287 | if (val > 100) |
a7885eb8 KM |
3288 | return -EINVAL; |
3289 | ||
14208b0e | 3290 | if (css->parent) |
3dae7fec JW |
3291 | memcg->swappiness = val; |
3292 | else | |
3293 | vm_swappiness = val; | |
068b38c1 | 3294 | |
a7885eb8 KM |
3295 | return 0; |
3296 | } | |
3297 | ||
2e72b634 KS |
3298 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3299 | { | |
3300 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3301 | unsigned long usage; |
2e72b634 KS |
3302 | int i; |
3303 | ||
3304 | rcu_read_lock(); | |
3305 | if (!swap) | |
2c488db2 | 3306 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3307 | else |
2c488db2 | 3308 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3309 | |
3310 | if (!t) | |
3311 | goto unlock; | |
3312 | ||
ce00a967 | 3313 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3314 | |
3315 | /* | |
748dad36 | 3316 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3317 | * If it's not true, a threshold was crossed after last |
3318 | * call of __mem_cgroup_threshold(). | |
3319 | */ | |
5407a562 | 3320 | i = t->current_threshold; |
2e72b634 KS |
3321 | |
3322 | /* | |
3323 | * Iterate backward over array of thresholds starting from | |
3324 | * current_threshold and check if a threshold is crossed. | |
3325 | * If none of thresholds below usage is crossed, we read | |
3326 | * only one element of the array here. | |
3327 | */ | |
3328 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3329 | eventfd_signal(t->entries[i].eventfd, 1); | |
3330 | ||
3331 | /* i = current_threshold + 1 */ | |
3332 | i++; | |
3333 | ||
3334 | /* | |
3335 | * Iterate forward over array of thresholds starting from | |
3336 | * current_threshold+1 and check if a threshold is crossed. | |
3337 | * If none of thresholds above usage is crossed, we read | |
3338 | * only one element of the array here. | |
3339 | */ | |
3340 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3341 | eventfd_signal(t->entries[i].eventfd, 1); | |
3342 | ||
3343 | /* Update current_threshold */ | |
5407a562 | 3344 | t->current_threshold = i - 1; |
2e72b634 KS |
3345 | unlock: |
3346 | rcu_read_unlock(); | |
3347 | } | |
3348 | ||
3349 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3350 | { | |
ad4ca5f4 KS |
3351 | while (memcg) { |
3352 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3353 | if (do_memsw_account()) |
ad4ca5f4 KS |
3354 | __mem_cgroup_threshold(memcg, true); |
3355 | ||
3356 | memcg = parent_mem_cgroup(memcg); | |
3357 | } | |
2e72b634 KS |
3358 | } |
3359 | ||
3360 | static int compare_thresholds(const void *a, const void *b) | |
3361 | { | |
3362 | const struct mem_cgroup_threshold *_a = a; | |
3363 | const struct mem_cgroup_threshold *_b = b; | |
3364 | ||
2bff24a3 GT |
3365 | if (_a->threshold > _b->threshold) |
3366 | return 1; | |
3367 | ||
3368 | if (_a->threshold < _b->threshold) | |
3369 | return -1; | |
3370 | ||
3371 | return 0; | |
2e72b634 KS |
3372 | } |
3373 | ||
c0ff4b85 | 3374 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3375 | { |
3376 | struct mem_cgroup_eventfd_list *ev; | |
3377 | ||
2bcf2e92 MH |
3378 | spin_lock(&memcg_oom_lock); |
3379 | ||
c0ff4b85 | 3380 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3381 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3382 | |
3383 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3384 | return 0; |
3385 | } | |
3386 | ||
c0ff4b85 | 3387 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3388 | { |
7d74b06f KH |
3389 | struct mem_cgroup *iter; |
3390 | ||
c0ff4b85 | 3391 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3392 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3393 | } |
3394 | ||
59b6f873 | 3395 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3396 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3397 | { |
2c488db2 KS |
3398 | struct mem_cgroup_thresholds *thresholds; |
3399 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3400 | unsigned long threshold; |
3401 | unsigned long usage; | |
2c488db2 | 3402 | int i, size, ret; |
2e72b634 | 3403 | |
650c5e56 | 3404 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3405 | if (ret) |
3406 | return ret; | |
3407 | ||
3408 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3409 | |
05b84301 | 3410 | if (type == _MEM) { |
2c488db2 | 3411 | thresholds = &memcg->thresholds; |
ce00a967 | 3412 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3413 | } else if (type == _MEMSWAP) { |
2c488db2 | 3414 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3415 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3416 | } else |
2e72b634 KS |
3417 | BUG(); |
3418 | ||
2e72b634 | 3419 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3420 | if (thresholds->primary) |
2e72b634 KS |
3421 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3422 | ||
2c488db2 | 3423 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3424 | |
3425 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3426 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3427 | GFP_KERNEL); |
2c488db2 | 3428 | if (!new) { |
2e72b634 KS |
3429 | ret = -ENOMEM; |
3430 | goto unlock; | |
3431 | } | |
2c488db2 | 3432 | new->size = size; |
2e72b634 KS |
3433 | |
3434 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3435 | if (thresholds->primary) { |
3436 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3437 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3438 | } |
3439 | ||
2e72b634 | 3440 | /* Add new threshold */ |
2c488db2 KS |
3441 | new->entries[size - 1].eventfd = eventfd; |
3442 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3443 | |
3444 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3445 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3446 | compare_thresholds, NULL); |
3447 | ||
3448 | /* Find current threshold */ | |
2c488db2 | 3449 | new->current_threshold = -1; |
2e72b634 | 3450 | for (i = 0; i < size; i++) { |
748dad36 | 3451 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3452 | /* |
2c488db2 KS |
3453 | * new->current_threshold will not be used until |
3454 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3455 | * it here. |
3456 | */ | |
2c488db2 | 3457 | ++new->current_threshold; |
748dad36 SZ |
3458 | } else |
3459 | break; | |
2e72b634 KS |
3460 | } |
3461 | ||
2c488db2 KS |
3462 | /* Free old spare buffer and save old primary buffer as spare */ |
3463 | kfree(thresholds->spare); | |
3464 | thresholds->spare = thresholds->primary; | |
3465 | ||
3466 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3467 | |
907860ed | 3468 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3469 | synchronize_rcu(); |
3470 | ||
2e72b634 KS |
3471 | unlock: |
3472 | mutex_unlock(&memcg->thresholds_lock); | |
3473 | ||
3474 | return ret; | |
3475 | } | |
3476 | ||
59b6f873 | 3477 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3478 | struct eventfd_ctx *eventfd, const char *args) |
3479 | { | |
59b6f873 | 3480 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3481 | } |
3482 | ||
59b6f873 | 3483 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3484 | struct eventfd_ctx *eventfd, const char *args) |
3485 | { | |
59b6f873 | 3486 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3487 | } |
3488 | ||
59b6f873 | 3489 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3490 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3491 | { |
2c488db2 KS |
3492 | struct mem_cgroup_thresholds *thresholds; |
3493 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3494 | unsigned long usage; |
2c488db2 | 3495 | int i, j, size; |
2e72b634 KS |
3496 | |
3497 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3498 | |
3499 | if (type == _MEM) { | |
2c488db2 | 3500 | thresholds = &memcg->thresholds; |
ce00a967 | 3501 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3502 | } else if (type == _MEMSWAP) { |
2c488db2 | 3503 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3504 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3505 | } else |
2e72b634 KS |
3506 | BUG(); |
3507 | ||
371528ca AV |
3508 | if (!thresholds->primary) |
3509 | goto unlock; | |
3510 | ||
2e72b634 KS |
3511 | /* Check if a threshold crossed before removing */ |
3512 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3513 | ||
3514 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3515 | size = 0; |
3516 | for (i = 0; i < thresholds->primary->size; i++) { | |
3517 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3518 | size++; |
3519 | } | |
3520 | ||
2c488db2 | 3521 | new = thresholds->spare; |
907860ed | 3522 | |
2e72b634 KS |
3523 | /* Set thresholds array to NULL if we don't have thresholds */ |
3524 | if (!size) { | |
2c488db2 KS |
3525 | kfree(new); |
3526 | new = NULL; | |
907860ed | 3527 | goto swap_buffers; |
2e72b634 KS |
3528 | } |
3529 | ||
2c488db2 | 3530 | new->size = size; |
2e72b634 KS |
3531 | |
3532 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3533 | new->current_threshold = -1; |
3534 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3535 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3536 | continue; |
3537 | ||
2c488db2 | 3538 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3539 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3540 | /* |
2c488db2 | 3541 | * new->current_threshold will not be used |
2e72b634 KS |
3542 | * until rcu_assign_pointer(), so it's safe to increment |
3543 | * it here. | |
3544 | */ | |
2c488db2 | 3545 | ++new->current_threshold; |
2e72b634 KS |
3546 | } |
3547 | j++; | |
3548 | } | |
3549 | ||
907860ed | 3550 | swap_buffers: |
2c488db2 KS |
3551 | /* Swap primary and spare array */ |
3552 | thresholds->spare = thresholds->primary; | |
8c757763 | 3553 | |
2c488db2 | 3554 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3555 | |
907860ed | 3556 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3557 | synchronize_rcu(); |
6611d8d7 MC |
3558 | |
3559 | /* If all events are unregistered, free the spare array */ | |
3560 | if (!new) { | |
3561 | kfree(thresholds->spare); | |
3562 | thresholds->spare = NULL; | |
3563 | } | |
371528ca | 3564 | unlock: |
2e72b634 | 3565 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3566 | } |
c1e862c1 | 3567 | |
59b6f873 | 3568 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3569 | struct eventfd_ctx *eventfd) |
3570 | { | |
59b6f873 | 3571 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3572 | } |
3573 | ||
59b6f873 | 3574 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3575 | struct eventfd_ctx *eventfd) |
3576 | { | |
59b6f873 | 3577 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3578 | } |
3579 | ||
59b6f873 | 3580 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3581 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3582 | { |
9490ff27 | 3583 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3584 | |
9490ff27 KH |
3585 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3586 | if (!event) | |
3587 | return -ENOMEM; | |
3588 | ||
1af8efe9 | 3589 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3590 | |
3591 | event->eventfd = eventfd; | |
3592 | list_add(&event->list, &memcg->oom_notify); | |
3593 | ||
3594 | /* already in OOM ? */ | |
c2b42d3c | 3595 | if (memcg->under_oom) |
9490ff27 | 3596 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3597 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3598 | |
3599 | return 0; | |
3600 | } | |
3601 | ||
59b6f873 | 3602 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3603 | struct eventfd_ctx *eventfd) |
9490ff27 | 3604 | { |
9490ff27 | 3605 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3606 | |
1af8efe9 | 3607 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3608 | |
c0ff4b85 | 3609 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3610 | if (ev->eventfd == eventfd) { |
3611 | list_del(&ev->list); | |
3612 | kfree(ev); | |
3613 | } | |
3614 | } | |
3615 | ||
1af8efe9 | 3616 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3617 | } |
3618 | ||
2da8ca82 | 3619 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3620 | { |
2da8ca82 | 3621 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3622 | |
791badbd | 3623 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3624 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3625 | return 0; |
3626 | } | |
3627 | ||
182446d0 | 3628 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3629 | struct cftype *cft, u64 val) |
3630 | { | |
182446d0 | 3631 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3632 | |
3633 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3634 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3635 | return -EINVAL; |
3636 | ||
c0ff4b85 | 3637 | memcg->oom_kill_disable = val; |
4d845ebf | 3638 | if (!val) |
c0ff4b85 | 3639 | memcg_oom_recover(memcg); |
3dae7fec | 3640 | |
3c11ecf4 KH |
3641 | return 0; |
3642 | } | |
3643 | ||
52ebea74 TH |
3644 | #ifdef CONFIG_CGROUP_WRITEBACK |
3645 | ||
3646 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3647 | { | |
3648 | return &memcg->cgwb_list; | |
3649 | } | |
3650 | ||
841710aa TH |
3651 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3652 | { | |
3653 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3654 | } | |
3655 | ||
3656 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3657 | { | |
3658 | wb_domain_exit(&memcg->cgwb_domain); | |
3659 | } | |
3660 | ||
2529bb3a TH |
3661 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3662 | { | |
3663 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3664 | } | |
3665 | ||
841710aa TH |
3666 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3667 | { | |
3668 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3669 | ||
3670 | if (!memcg->css.parent) | |
3671 | return NULL; | |
3672 | ||
3673 | return &memcg->cgwb_domain; | |
3674 | } | |
3675 | ||
c2aa723a TH |
3676 | /** |
3677 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3678 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3679 | * @pfilepages: out parameter for number of file pages |
3680 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3681 | * @pdirty: out parameter for number of dirty pages |
3682 | * @pwriteback: out parameter for number of pages under writeback | |
3683 | * | |
c5edf9cd TH |
3684 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3685 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3686 | * is a bit more involved. | |
c2aa723a | 3687 | * |
c5edf9cd TH |
3688 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3689 | * headroom is calculated as the lowest headroom of itself and the | |
3690 | * ancestors. Note that this doesn't consider the actual amount of | |
3691 | * available memory in the system. The caller should further cap | |
3692 | * *@pheadroom accordingly. | |
c2aa723a | 3693 | */ |
c5edf9cd TH |
3694 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3695 | unsigned long *pheadroom, unsigned long *pdirty, | |
3696 | unsigned long *pwriteback) | |
c2aa723a TH |
3697 | { |
3698 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3699 | struct mem_cgroup *parent; | |
c2aa723a TH |
3700 | |
3701 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3702 | ||
3703 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3704 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3705 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3706 | (1 << LRU_ACTIVE_FILE)); | |
3707 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3708 | |
c2aa723a TH |
3709 | while ((parent = parent_mem_cgroup(memcg))) { |
3710 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3711 | unsigned long used = page_counter_read(&memcg->memory); | |
3712 | ||
c5edf9cd | 3713 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3714 | memcg = parent; |
3715 | } | |
c2aa723a TH |
3716 | } |
3717 | ||
841710aa TH |
3718 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3719 | ||
3720 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3721 | { | |
3722 | return 0; | |
3723 | } | |
3724 | ||
3725 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3726 | { | |
3727 | } | |
3728 | ||
2529bb3a TH |
3729 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3730 | { | |
3731 | } | |
3732 | ||
52ebea74 TH |
3733 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3734 | ||
3bc942f3 TH |
3735 | /* |
3736 | * DO NOT USE IN NEW FILES. | |
3737 | * | |
3738 | * "cgroup.event_control" implementation. | |
3739 | * | |
3740 | * This is way over-engineered. It tries to support fully configurable | |
3741 | * events for each user. Such level of flexibility is completely | |
3742 | * unnecessary especially in the light of the planned unified hierarchy. | |
3743 | * | |
3744 | * Please deprecate this and replace with something simpler if at all | |
3745 | * possible. | |
3746 | */ | |
3747 | ||
79bd9814 TH |
3748 | /* |
3749 | * Unregister event and free resources. | |
3750 | * | |
3751 | * Gets called from workqueue. | |
3752 | */ | |
3bc942f3 | 3753 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3754 | { |
3bc942f3 TH |
3755 | struct mem_cgroup_event *event = |
3756 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3757 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3758 | |
3759 | remove_wait_queue(event->wqh, &event->wait); | |
3760 | ||
59b6f873 | 3761 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3762 | |
3763 | /* Notify userspace the event is going away. */ | |
3764 | eventfd_signal(event->eventfd, 1); | |
3765 | ||
3766 | eventfd_ctx_put(event->eventfd); | |
3767 | kfree(event); | |
59b6f873 | 3768 | css_put(&memcg->css); |
79bd9814 TH |
3769 | } |
3770 | ||
3771 | /* | |
3772 | * Gets called on POLLHUP on eventfd when user closes it. | |
3773 | * | |
3774 | * Called with wqh->lock held and interrupts disabled. | |
3775 | */ | |
3bc942f3 TH |
3776 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3777 | int sync, void *key) | |
79bd9814 | 3778 | { |
3bc942f3 TH |
3779 | struct mem_cgroup_event *event = |
3780 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3781 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3782 | unsigned long flags = (unsigned long)key; |
3783 | ||
3784 | if (flags & POLLHUP) { | |
3785 | /* | |
3786 | * If the event has been detached at cgroup removal, we | |
3787 | * can simply return knowing the other side will cleanup | |
3788 | * for us. | |
3789 | * | |
3790 | * We can't race against event freeing since the other | |
3791 | * side will require wqh->lock via remove_wait_queue(), | |
3792 | * which we hold. | |
3793 | */ | |
fba94807 | 3794 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3795 | if (!list_empty(&event->list)) { |
3796 | list_del_init(&event->list); | |
3797 | /* | |
3798 | * We are in atomic context, but cgroup_event_remove() | |
3799 | * may sleep, so we have to call it in workqueue. | |
3800 | */ | |
3801 | schedule_work(&event->remove); | |
3802 | } | |
fba94807 | 3803 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3804 | } |
3805 | ||
3806 | return 0; | |
3807 | } | |
3808 | ||
3bc942f3 | 3809 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3810 | wait_queue_head_t *wqh, poll_table *pt) |
3811 | { | |
3bc942f3 TH |
3812 | struct mem_cgroup_event *event = |
3813 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3814 | |
3815 | event->wqh = wqh; | |
3816 | add_wait_queue(wqh, &event->wait); | |
3817 | } | |
3818 | ||
3819 | /* | |
3bc942f3 TH |
3820 | * DO NOT USE IN NEW FILES. |
3821 | * | |
79bd9814 TH |
3822 | * Parse input and register new cgroup event handler. |
3823 | * | |
3824 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3825 | * Interpretation of args is defined by control file implementation. | |
3826 | */ | |
451af504 TH |
3827 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3828 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3829 | { |
451af504 | 3830 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3831 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3832 | struct mem_cgroup_event *event; |
79bd9814 TH |
3833 | struct cgroup_subsys_state *cfile_css; |
3834 | unsigned int efd, cfd; | |
3835 | struct fd efile; | |
3836 | struct fd cfile; | |
fba94807 | 3837 | const char *name; |
79bd9814 TH |
3838 | char *endp; |
3839 | int ret; | |
3840 | ||
451af504 TH |
3841 | buf = strstrip(buf); |
3842 | ||
3843 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3844 | if (*endp != ' ') |
3845 | return -EINVAL; | |
451af504 | 3846 | buf = endp + 1; |
79bd9814 | 3847 | |
451af504 | 3848 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3849 | if ((*endp != ' ') && (*endp != '\0')) |
3850 | return -EINVAL; | |
451af504 | 3851 | buf = endp + 1; |
79bd9814 TH |
3852 | |
3853 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3854 | if (!event) | |
3855 | return -ENOMEM; | |
3856 | ||
59b6f873 | 3857 | event->memcg = memcg; |
79bd9814 | 3858 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3859 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3860 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3861 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3862 | |
3863 | efile = fdget(efd); | |
3864 | if (!efile.file) { | |
3865 | ret = -EBADF; | |
3866 | goto out_kfree; | |
3867 | } | |
3868 | ||
3869 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3870 | if (IS_ERR(event->eventfd)) { | |
3871 | ret = PTR_ERR(event->eventfd); | |
3872 | goto out_put_efile; | |
3873 | } | |
3874 | ||
3875 | cfile = fdget(cfd); | |
3876 | if (!cfile.file) { | |
3877 | ret = -EBADF; | |
3878 | goto out_put_eventfd; | |
3879 | } | |
3880 | ||
3881 | /* the process need read permission on control file */ | |
3882 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3883 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3884 | if (ret < 0) | |
3885 | goto out_put_cfile; | |
3886 | ||
fba94807 TH |
3887 | /* |
3888 | * Determine the event callbacks and set them in @event. This used | |
3889 | * to be done via struct cftype but cgroup core no longer knows | |
3890 | * about these events. The following is crude but the whole thing | |
3891 | * is for compatibility anyway. | |
3bc942f3 TH |
3892 | * |
3893 | * DO NOT ADD NEW FILES. | |
fba94807 | 3894 | */ |
b583043e | 3895 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3896 | |
3897 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3898 | event->register_event = mem_cgroup_usage_register_event; | |
3899 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3900 | } else if (!strcmp(name, "memory.oom_control")) { | |
3901 | event->register_event = mem_cgroup_oom_register_event; | |
3902 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3903 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3904 | event->register_event = vmpressure_register_event; | |
3905 | event->unregister_event = vmpressure_unregister_event; | |
3906 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3907 | event->register_event = memsw_cgroup_usage_register_event; |
3908 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3909 | } else { |
3910 | ret = -EINVAL; | |
3911 | goto out_put_cfile; | |
3912 | } | |
3913 | ||
79bd9814 | 3914 | /* |
b5557c4c TH |
3915 | * Verify @cfile should belong to @css. Also, remaining events are |
3916 | * automatically removed on cgroup destruction but the removal is | |
3917 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3918 | */ |
b583043e | 3919 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3920 | &memory_cgrp_subsys); |
79bd9814 | 3921 | ret = -EINVAL; |
5a17f543 | 3922 | if (IS_ERR(cfile_css)) |
79bd9814 | 3923 | goto out_put_cfile; |
5a17f543 TH |
3924 | if (cfile_css != css) { |
3925 | css_put(cfile_css); | |
79bd9814 | 3926 | goto out_put_cfile; |
5a17f543 | 3927 | } |
79bd9814 | 3928 | |
451af504 | 3929 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3930 | if (ret) |
3931 | goto out_put_css; | |
3932 | ||
3933 | efile.file->f_op->poll(efile.file, &event->pt); | |
3934 | ||
fba94807 TH |
3935 | spin_lock(&memcg->event_list_lock); |
3936 | list_add(&event->list, &memcg->event_list); | |
3937 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3938 | |
3939 | fdput(cfile); | |
3940 | fdput(efile); | |
3941 | ||
451af504 | 3942 | return nbytes; |
79bd9814 TH |
3943 | |
3944 | out_put_css: | |
b5557c4c | 3945 | css_put(css); |
79bd9814 TH |
3946 | out_put_cfile: |
3947 | fdput(cfile); | |
3948 | out_put_eventfd: | |
3949 | eventfd_ctx_put(event->eventfd); | |
3950 | out_put_efile: | |
3951 | fdput(efile); | |
3952 | out_kfree: | |
3953 | kfree(event); | |
3954 | ||
3955 | return ret; | |
3956 | } | |
3957 | ||
241994ed | 3958 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 3959 | { |
0eea1030 | 3960 | .name = "usage_in_bytes", |
8c7c6e34 | 3961 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 3962 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3963 | }, |
c84872e1 PE |
3964 | { |
3965 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3966 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 3967 | .write = mem_cgroup_reset, |
791badbd | 3968 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 3969 | }, |
8cdea7c0 | 3970 | { |
0eea1030 | 3971 | .name = "limit_in_bytes", |
8c7c6e34 | 3972 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 3973 | .write = mem_cgroup_write, |
791badbd | 3974 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3975 | }, |
296c81d8 BS |
3976 | { |
3977 | .name = "soft_limit_in_bytes", | |
3978 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 3979 | .write = mem_cgroup_write, |
791badbd | 3980 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 3981 | }, |
8cdea7c0 BS |
3982 | { |
3983 | .name = "failcnt", | |
8c7c6e34 | 3984 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 3985 | .write = mem_cgroup_reset, |
791badbd | 3986 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3987 | }, |
d2ceb9b7 KH |
3988 | { |
3989 | .name = "stat", | |
2da8ca82 | 3990 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 3991 | }, |
c1e862c1 KH |
3992 | { |
3993 | .name = "force_empty", | |
6770c64e | 3994 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 3995 | }, |
18f59ea7 BS |
3996 | { |
3997 | .name = "use_hierarchy", | |
3998 | .write_u64 = mem_cgroup_hierarchy_write, | |
3999 | .read_u64 = mem_cgroup_hierarchy_read, | |
4000 | }, | |
79bd9814 | 4001 | { |
3bc942f3 | 4002 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4003 | .write = memcg_write_event_control, |
7dbdb199 | 4004 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4005 | }, |
a7885eb8 KM |
4006 | { |
4007 | .name = "swappiness", | |
4008 | .read_u64 = mem_cgroup_swappiness_read, | |
4009 | .write_u64 = mem_cgroup_swappiness_write, | |
4010 | }, | |
7dc74be0 DN |
4011 | { |
4012 | .name = "move_charge_at_immigrate", | |
4013 | .read_u64 = mem_cgroup_move_charge_read, | |
4014 | .write_u64 = mem_cgroup_move_charge_write, | |
4015 | }, | |
9490ff27 KH |
4016 | { |
4017 | .name = "oom_control", | |
2da8ca82 | 4018 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4019 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4020 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4021 | }, | |
70ddf637 AV |
4022 | { |
4023 | .name = "pressure_level", | |
70ddf637 | 4024 | }, |
406eb0c9 YH |
4025 | #ifdef CONFIG_NUMA |
4026 | { | |
4027 | .name = "numa_stat", | |
2da8ca82 | 4028 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4029 | }, |
4030 | #endif | |
510fc4e1 GC |
4031 | { |
4032 | .name = "kmem.limit_in_bytes", | |
4033 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4034 | .write = mem_cgroup_write, |
791badbd | 4035 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4036 | }, |
4037 | { | |
4038 | .name = "kmem.usage_in_bytes", | |
4039 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4040 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4041 | }, |
4042 | { | |
4043 | .name = "kmem.failcnt", | |
4044 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4045 | .write = mem_cgroup_reset, |
791badbd | 4046 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4047 | }, |
4048 | { | |
4049 | .name = "kmem.max_usage_in_bytes", | |
4050 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4051 | .write = mem_cgroup_reset, |
791badbd | 4052 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4053 | }, |
749c5415 GC |
4054 | #ifdef CONFIG_SLABINFO |
4055 | { | |
4056 | .name = "kmem.slabinfo", | |
b047501c VD |
4057 | .seq_start = slab_start, |
4058 | .seq_next = slab_next, | |
4059 | .seq_stop = slab_stop, | |
4060 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4061 | }, |
4062 | #endif | |
d55f90bf VD |
4063 | { |
4064 | .name = "kmem.tcp.limit_in_bytes", | |
4065 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4066 | .write = mem_cgroup_write, | |
4067 | .read_u64 = mem_cgroup_read_u64, | |
4068 | }, | |
4069 | { | |
4070 | .name = "kmem.tcp.usage_in_bytes", | |
4071 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4072 | .read_u64 = mem_cgroup_read_u64, | |
4073 | }, | |
4074 | { | |
4075 | .name = "kmem.tcp.failcnt", | |
4076 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4077 | .write = mem_cgroup_reset, | |
4078 | .read_u64 = mem_cgroup_read_u64, | |
4079 | }, | |
4080 | { | |
4081 | .name = "kmem.tcp.max_usage_in_bytes", | |
4082 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4083 | .write = mem_cgroup_reset, | |
4084 | .read_u64 = mem_cgroup_read_u64, | |
4085 | }, | |
6bc10349 | 4086 | { }, /* terminate */ |
af36f906 | 4087 | }; |
8c7c6e34 | 4088 | |
c0ff4b85 | 4089 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4090 | { |
4091 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4092 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4093 | int zone, tmp = node; |
1ecaab2b KH |
4094 | /* |
4095 | * This routine is called against possible nodes. | |
4096 | * But it's BUG to call kmalloc() against offline node. | |
4097 | * | |
4098 | * TODO: this routine can waste much memory for nodes which will | |
4099 | * never be onlined. It's better to use memory hotplug callback | |
4100 | * function. | |
4101 | */ | |
41e3355d KH |
4102 | if (!node_state(node, N_NORMAL_MEMORY)) |
4103 | tmp = -1; | |
17295c88 | 4104 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4105 | if (!pn) |
4106 | return 1; | |
1ecaab2b | 4107 | |
1ecaab2b KH |
4108 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4109 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4110 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4111 | mz->usage_in_excess = 0; |
4112 | mz->on_tree = false; | |
d79154bb | 4113 | mz->memcg = memcg; |
1ecaab2b | 4114 | } |
54f72fe0 | 4115 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4116 | return 0; |
4117 | } | |
4118 | ||
c0ff4b85 | 4119 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4120 | { |
54f72fe0 | 4121 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4122 | } |
4123 | ||
0b8f73e1 | 4124 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4125 | { |
c8b2a36f | 4126 | int node; |
59927fb9 | 4127 | |
0b8f73e1 | 4128 | memcg_wb_domain_exit(memcg); |
c8b2a36f GC |
4129 | for_each_node(node) |
4130 | free_mem_cgroup_per_zone_info(memcg, node); | |
c8b2a36f | 4131 | free_percpu(memcg->stat); |
8ff69e2c | 4132 | kfree(memcg); |
59927fb9 | 4133 | } |
3afe36b1 | 4134 | |
0b8f73e1 | 4135 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4136 | { |
d142e3e6 | 4137 | struct mem_cgroup *memcg; |
0b8f73e1 | 4138 | size_t size; |
6d12e2d8 | 4139 | int node; |
8cdea7c0 | 4140 | |
0b8f73e1 JW |
4141 | size = sizeof(struct mem_cgroup); |
4142 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4143 | ||
4144 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4145 | if (!memcg) |
0b8f73e1 JW |
4146 | return NULL; |
4147 | ||
4148 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); | |
4149 | if (!memcg->stat) | |
4150 | goto fail; | |
78fb7466 | 4151 | |
3ed28fa1 | 4152 | for_each_node(node) |
c0ff4b85 | 4153 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
0b8f73e1 | 4154 | goto fail; |
f64c3f54 | 4155 | |
0b8f73e1 JW |
4156 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4157 | goto fail; | |
28dbc4b6 | 4158 | |
f7e1cb6e | 4159 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4160 | memcg->last_scanned_node = MAX_NUMNODES; |
4161 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4162 | mutex_init(&memcg->thresholds_lock); |
4163 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4164 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4165 | INIT_LIST_HEAD(&memcg->event_list); |
4166 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4167 | memcg->socket_pressure = jiffies; |
127424c8 | 4168 | #ifndef CONFIG_SLOB |
900a38f0 | 4169 | memcg->kmemcg_id = -1; |
900a38f0 | 4170 | #endif |
52ebea74 TH |
4171 | #ifdef CONFIG_CGROUP_WRITEBACK |
4172 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4173 | #endif | |
0b8f73e1 JW |
4174 | return memcg; |
4175 | fail: | |
4176 | mem_cgroup_free(memcg); | |
4177 | return NULL; | |
d142e3e6 GC |
4178 | } |
4179 | ||
0b8f73e1 JW |
4180 | static struct cgroup_subsys_state * __ref |
4181 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4182 | { |
0b8f73e1 JW |
4183 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4184 | struct mem_cgroup *memcg; | |
4185 | long error = -ENOMEM; | |
d142e3e6 | 4186 | |
0b8f73e1 JW |
4187 | memcg = mem_cgroup_alloc(); |
4188 | if (!memcg) | |
4189 | return ERR_PTR(error); | |
d142e3e6 | 4190 | |
0b8f73e1 JW |
4191 | memcg->high = PAGE_COUNTER_MAX; |
4192 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4193 | if (parent) { | |
4194 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4195 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4196 | } | |
4197 | if (parent && parent->use_hierarchy) { | |
4198 | memcg->use_hierarchy = true; | |
3e32cb2e | 4199 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4200 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4201 | page_counter_init(&memcg->memsw, &parent->memsw); |
4202 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4203 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4204 | } else { |
3e32cb2e | 4205 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4206 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4207 | page_counter_init(&memcg->memsw, NULL); |
4208 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4209 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4210 | /* |
4211 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4212 | * much sense so let cgroup subsystem know about this | |
4213 | * unfortunate state in our controller. | |
4214 | */ | |
d142e3e6 | 4215 | if (parent != root_mem_cgroup) |
073219e9 | 4216 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4217 | } |
d6441637 | 4218 | |
0b8f73e1 JW |
4219 | /* The following stuff does not apply to the root */ |
4220 | if (!parent) { | |
4221 | root_mem_cgroup = memcg; | |
4222 | return &memcg->css; | |
4223 | } | |
4224 | ||
4225 | error = memcg_propagate_kmem(parent, memcg); | |
4226 | if (error) | |
4227 | goto fail; | |
127424c8 | 4228 | |
f7e1cb6e | 4229 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4230 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4231 | |
0b8f73e1 JW |
4232 | return &memcg->css; |
4233 | fail: | |
4234 | mem_cgroup_free(memcg); | |
4235 | return NULL; | |
4236 | } | |
4237 | ||
4238 | static int | |
4239 | mem_cgroup_css_online(struct cgroup_subsys_state *css) | |
4240 | { | |
4241 | if (css->id > MEM_CGROUP_ID_MAX) | |
4242 | return -ENOSPC; | |
2f7dd7a4 JW |
4243 | |
4244 | return 0; | |
8cdea7c0 BS |
4245 | } |
4246 | ||
eb95419b | 4247 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4248 | { |
eb95419b | 4249 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4250 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4251 | |
4252 | /* | |
4253 | * Unregister events and notify userspace. | |
4254 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4255 | * directory to avoid race between userspace and kernelspace. | |
4256 | */ | |
fba94807 TH |
4257 | spin_lock(&memcg->event_list_lock); |
4258 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4259 | list_del_init(&event->list); |
4260 | schedule_work(&event->remove); | |
4261 | } | |
fba94807 | 4262 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4263 | |
567e9ab2 | 4264 | memcg_offline_kmem(memcg); |
52ebea74 | 4265 | wb_memcg_offline(memcg); |
df878fb0 KH |
4266 | } |
4267 | ||
6df38689 VD |
4268 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4269 | { | |
4270 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4271 | ||
4272 | invalidate_reclaim_iterators(memcg); | |
4273 | } | |
4274 | ||
eb95419b | 4275 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4276 | { |
eb95419b | 4277 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4278 | |
f7e1cb6e | 4279 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4280 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4281 | |
0db15298 | 4282 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4283 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4284 | |
0b8f73e1 JW |
4285 | vmpressure_cleanup(&memcg->vmpressure); |
4286 | cancel_work_sync(&memcg->high_work); | |
4287 | mem_cgroup_remove_from_trees(memcg); | |
d886f4e4 | 4288 | memcg_free_kmem(memcg); |
0b8f73e1 | 4289 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4290 | } |
4291 | ||
1ced953b TH |
4292 | /** |
4293 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4294 | * @css: the target css | |
4295 | * | |
4296 | * Reset the states of the mem_cgroup associated with @css. This is | |
4297 | * invoked when the userland requests disabling on the default hierarchy | |
4298 | * but the memcg is pinned through dependency. The memcg should stop | |
4299 | * applying policies and should revert to the vanilla state as it may be | |
4300 | * made visible again. | |
4301 | * | |
4302 | * The current implementation only resets the essential configurations. | |
4303 | * This needs to be expanded to cover all the visible parts. | |
4304 | */ | |
4305 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4306 | { | |
4307 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4308 | ||
3e32cb2e JW |
4309 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4310 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4311 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4312 | memcg->low = 0; |
4313 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4314 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4315 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4316 | } |
4317 | ||
02491447 | 4318 | #ifdef CONFIG_MMU |
7dc74be0 | 4319 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4320 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4321 | { |
05b84301 | 4322 | int ret; |
9476db97 | 4323 | |
d0164adc MG |
4324 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4325 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4326 | if (!ret) { |
854ffa8d | 4327 | mc.precharge += count; |
854ffa8d DN |
4328 | return ret; |
4329 | } | |
9476db97 JW |
4330 | |
4331 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4332 | while (count--) { |
00501b53 | 4333 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4334 | if (ret) |
38c5d72f | 4335 | return ret; |
854ffa8d | 4336 | mc.precharge++; |
9476db97 | 4337 | cond_resched(); |
854ffa8d | 4338 | } |
9476db97 | 4339 | return 0; |
4ffef5fe DN |
4340 | } |
4341 | ||
4342 | /** | |
8d32ff84 | 4343 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4344 | * @vma: the vma the pte to be checked belongs |
4345 | * @addr: the address corresponding to the pte to be checked | |
4346 | * @ptent: the pte to be checked | |
02491447 | 4347 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4348 | * |
4349 | * Returns | |
4350 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4351 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4352 | * move charge. if @target is not NULL, the page is stored in target->page | |
4353 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4354 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4355 | * target for charge migration. if @target is not NULL, the entry is stored | |
4356 | * in target->ent. | |
4ffef5fe DN |
4357 | * |
4358 | * Called with pte lock held. | |
4359 | */ | |
4ffef5fe DN |
4360 | union mc_target { |
4361 | struct page *page; | |
02491447 | 4362 | swp_entry_t ent; |
4ffef5fe DN |
4363 | }; |
4364 | ||
4ffef5fe | 4365 | enum mc_target_type { |
8d32ff84 | 4366 | MC_TARGET_NONE = 0, |
4ffef5fe | 4367 | MC_TARGET_PAGE, |
02491447 | 4368 | MC_TARGET_SWAP, |
4ffef5fe DN |
4369 | }; |
4370 | ||
90254a65 DN |
4371 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4372 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4373 | { |
90254a65 | 4374 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4375 | |
90254a65 DN |
4376 | if (!page || !page_mapped(page)) |
4377 | return NULL; | |
4378 | if (PageAnon(page)) { | |
1dfab5ab | 4379 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4380 | return NULL; |
1dfab5ab JW |
4381 | } else { |
4382 | if (!(mc.flags & MOVE_FILE)) | |
4383 | return NULL; | |
4384 | } | |
90254a65 DN |
4385 | if (!get_page_unless_zero(page)) |
4386 | return NULL; | |
4387 | ||
4388 | return page; | |
4389 | } | |
4390 | ||
4b91355e | 4391 | #ifdef CONFIG_SWAP |
90254a65 DN |
4392 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4393 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4394 | { | |
90254a65 DN |
4395 | struct page *page = NULL; |
4396 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4397 | ||
1dfab5ab | 4398 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4399 | return NULL; |
4b91355e KH |
4400 | /* |
4401 | * Because lookup_swap_cache() updates some statistics counter, | |
4402 | * we call find_get_page() with swapper_space directly. | |
4403 | */ | |
33806f06 | 4404 | page = find_get_page(swap_address_space(ent), ent.val); |
7941d214 | 4405 | if (do_memsw_account()) |
90254a65 DN |
4406 | entry->val = ent.val; |
4407 | ||
4408 | return page; | |
4409 | } | |
4b91355e KH |
4410 | #else |
4411 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4412 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4413 | { | |
4414 | return NULL; | |
4415 | } | |
4416 | #endif | |
90254a65 | 4417 | |
87946a72 DN |
4418 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4419 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4420 | { | |
4421 | struct page *page = NULL; | |
87946a72 DN |
4422 | struct address_space *mapping; |
4423 | pgoff_t pgoff; | |
4424 | ||
4425 | if (!vma->vm_file) /* anonymous vma */ | |
4426 | return NULL; | |
1dfab5ab | 4427 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4428 | return NULL; |
4429 | ||
87946a72 | 4430 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4431 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4432 | |
4433 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4434 | #ifdef CONFIG_SWAP |
4435 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4436 | if (shmem_mapping(mapping)) { |
4437 | page = find_get_entry(mapping, pgoff); | |
4438 | if (radix_tree_exceptional_entry(page)) { | |
4439 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4440 | if (do_memsw_account()) |
139b6a6f JW |
4441 | *entry = swp; |
4442 | page = find_get_page(swap_address_space(swp), swp.val); | |
4443 | } | |
4444 | } else | |
4445 | page = find_get_page(mapping, pgoff); | |
4446 | #else | |
4447 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4448 | #endif |
87946a72 DN |
4449 | return page; |
4450 | } | |
4451 | ||
b1b0deab CG |
4452 | /** |
4453 | * mem_cgroup_move_account - move account of the page | |
4454 | * @page: the page | |
4455 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4456 | * @from: mem_cgroup which the page is moved from. | |
4457 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4458 | * | |
3ac808fd | 4459 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4460 | * |
4461 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4462 | * from old cgroup. | |
4463 | */ | |
4464 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4465 | bool compound, |
b1b0deab CG |
4466 | struct mem_cgroup *from, |
4467 | struct mem_cgroup *to) | |
4468 | { | |
4469 | unsigned long flags; | |
f627c2f5 | 4470 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4471 | int ret; |
c4843a75 | 4472 | bool anon; |
b1b0deab CG |
4473 | |
4474 | VM_BUG_ON(from == to); | |
4475 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4476 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4477 | |
4478 | /* | |
45637bab HD |
4479 | * Prevent mem_cgroup_replace_page() from looking at |
4480 | * page->mem_cgroup of its source page while we change it. | |
b1b0deab | 4481 | */ |
f627c2f5 | 4482 | ret = -EBUSY; |
b1b0deab CG |
4483 | if (!trylock_page(page)) |
4484 | goto out; | |
4485 | ||
4486 | ret = -EINVAL; | |
4487 | if (page->mem_cgroup != from) | |
4488 | goto out_unlock; | |
4489 | ||
c4843a75 GT |
4490 | anon = PageAnon(page); |
4491 | ||
b1b0deab CG |
4492 | spin_lock_irqsave(&from->move_lock, flags); |
4493 | ||
c4843a75 | 4494 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4495 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4496 | nr_pages); | |
4497 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4498 | nr_pages); | |
4499 | } | |
4500 | ||
c4843a75 GT |
4501 | /* |
4502 | * move_lock grabbed above and caller set from->moving_account, so | |
4503 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4504 | * So mapping should be stable for dirty pages. | |
4505 | */ | |
4506 | if (!anon && PageDirty(page)) { | |
4507 | struct address_space *mapping = page_mapping(page); | |
4508 | ||
4509 | if (mapping_cap_account_dirty(mapping)) { | |
4510 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4511 | nr_pages); | |
4512 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4513 | nr_pages); | |
4514 | } | |
4515 | } | |
4516 | ||
b1b0deab CG |
4517 | if (PageWriteback(page)) { |
4518 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4519 | nr_pages); | |
4520 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4521 | nr_pages); | |
4522 | } | |
4523 | ||
4524 | /* | |
4525 | * It is safe to change page->mem_cgroup here because the page | |
4526 | * is referenced, charged, and isolated - we can't race with | |
4527 | * uncharging, charging, migration, or LRU putback. | |
4528 | */ | |
4529 | ||
4530 | /* caller should have done css_get */ | |
4531 | page->mem_cgroup = to; | |
4532 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4533 | ||
4534 | ret = 0; | |
4535 | ||
4536 | local_irq_disable(); | |
f627c2f5 | 4537 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4538 | memcg_check_events(to, page); |
f627c2f5 | 4539 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4540 | memcg_check_events(from, page); |
4541 | local_irq_enable(); | |
4542 | out_unlock: | |
4543 | unlock_page(page); | |
4544 | out: | |
4545 | return ret; | |
4546 | } | |
4547 | ||
8d32ff84 | 4548 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4549 | unsigned long addr, pte_t ptent, union mc_target *target) |
4550 | { | |
4551 | struct page *page = NULL; | |
8d32ff84 | 4552 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4553 | swp_entry_t ent = { .val = 0 }; |
4554 | ||
4555 | if (pte_present(ptent)) | |
4556 | page = mc_handle_present_pte(vma, addr, ptent); | |
4557 | else if (is_swap_pte(ptent)) | |
4558 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4559 | else if (pte_none(ptent)) |
87946a72 | 4560 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4561 | |
4562 | if (!page && !ent.val) | |
8d32ff84 | 4563 | return ret; |
02491447 | 4564 | if (page) { |
02491447 | 4565 | /* |
0a31bc97 | 4566 | * Do only loose check w/o serialization. |
1306a85a | 4567 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4568 | * not under LRU exclusion. |
02491447 | 4569 | */ |
1306a85a | 4570 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4571 | ret = MC_TARGET_PAGE; |
4572 | if (target) | |
4573 | target->page = page; | |
4574 | } | |
4575 | if (!ret || !target) | |
4576 | put_page(page); | |
4577 | } | |
90254a65 DN |
4578 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4579 | if (ent.val && !ret && | |
34c00c31 | 4580 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4581 | ret = MC_TARGET_SWAP; |
4582 | if (target) | |
4583 | target->ent = ent; | |
4ffef5fe | 4584 | } |
4ffef5fe DN |
4585 | return ret; |
4586 | } | |
4587 | ||
12724850 NH |
4588 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4589 | /* | |
4590 | * We don't consider swapping or file mapped pages because THP does not | |
4591 | * support them for now. | |
4592 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4593 | */ | |
4594 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4595 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4596 | { | |
4597 | struct page *page = NULL; | |
12724850 NH |
4598 | enum mc_target_type ret = MC_TARGET_NONE; |
4599 | ||
4600 | page = pmd_page(pmd); | |
309381fe | 4601 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4602 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4603 | return ret; |
1306a85a | 4604 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4605 | ret = MC_TARGET_PAGE; |
4606 | if (target) { | |
4607 | get_page(page); | |
4608 | target->page = page; | |
4609 | } | |
4610 | } | |
4611 | return ret; | |
4612 | } | |
4613 | #else | |
4614 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4615 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4616 | { | |
4617 | return MC_TARGET_NONE; | |
4618 | } | |
4619 | #endif | |
4620 | ||
4ffef5fe DN |
4621 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4622 | unsigned long addr, unsigned long end, | |
4623 | struct mm_walk *walk) | |
4624 | { | |
26bcd64a | 4625 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4626 | pte_t *pte; |
4627 | spinlock_t *ptl; | |
4628 | ||
4b471e88 | 4629 | if (pmd_trans_huge_lock(pmd, vma, &ptl)) { |
12724850 NH |
4630 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4631 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4632 | spin_unlock(ptl); |
1a5a9906 | 4633 | return 0; |
12724850 | 4634 | } |
03319327 | 4635 | |
45f83cef AA |
4636 | if (pmd_trans_unstable(pmd)) |
4637 | return 0; | |
4ffef5fe DN |
4638 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4639 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4640 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4641 | mc.precharge++; /* increment precharge temporarily */ |
4642 | pte_unmap_unlock(pte - 1, ptl); | |
4643 | cond_resched(); | |
4644 | ||
7dc74be0 DN |
4645 | return 0; |
4646 | } | |
4647 | ||
4ffef5fe DN |
4648 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4649 | { | |
4650 | unsigned long precharge; | |
4ffef5fe | 4651 | |
26bcd64a NH |
4652 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4653 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4654 | .mm = mm, | |
4655 | }; | |
dfe076b0 | 4656 | down_read(&mm->mmap_sem); |
26bcd64a | 4657 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4658 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4659 | |
4660 | precharge = mc.precharge; | |
4661 | mc.precharge = 0; | |
4662 | ||
4663 | return precharge; | |
4664 | } | |
4665 | ||
4ffef5fe DN |
4666 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4667 | { | |
dfe076b0 DN |
4668 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4669 | ||
4670 | VM_BUG_ON(mc.moving_task); | |
4671 | mc.moving_task = current; | |
4672 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4673 | } |
4674 | ||
dfe076b0 DN |
4675 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4676 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4677 | { |
2bd9bb20 KH |
4678 | struct mem_cgroup *from = mc.from; |
4679 | struct mem_cgroup *to = mc.to; | |
4680 | ||
4ffef5fe | 4681 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4682 | if (mc.precharge) { |
00501b53 | 4683 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4684 | mc.precharge = 0; |
4685 | } | |
4686 | /* | |
4687 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4688 | * we must uncharge here. | |
4689 | */ | |
4690 | if (mc.moved_charge) { | |
00501b53 | 4691 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4692 | mc.moved_charge = 0; |
4ffef5fe | 4693 | } |
483c30b5 DN |
4694 | /* we must fixup refcnts and charges */ |
4695 | if (mc.moved_swap) { | |
483c30b5 | 4696 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4697 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4698 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4699 | |
05b84301 | 4700 | /* |
3e32cb2e JW |
4701 | * we charged both to->memory and to->memsw, so we |
4702 | * should uncharge to->memory. | |
05b84301 | 4703 | */ |
ce00a967 | 4704 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4705 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4706 | ||
e8ea14cc | 4707 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4708 | |
4050377b | 4709 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4710 | mc.moved_swap = 0; |
4711 | } | |
dfe076b0 DN |
4712 | memcg_oom_recover(from); |
4713 | memcg_oom_recover(to); | |
4714 | wake_up_all(&mc.waitq); | |
4715 | } | |
4716 | ||
4717 | static void mem_cgroup_clear_mc(void) | |
4718 | { | |
dfe076b0 DN |
4719 | /* |
4720 | * we must clear moving_task before waking up waiters at the end of | |
4721 | * task migration. | |
4722 | */ | |
4723 | mc.moving_task = NULL; | |
4724 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4725 | spin_lock(&mc.lock); |
4ffef5fe DN |
4726 | mc.from = NULL; |
4727 | mc.to = NULL; | |
2bd9bb20 | 4728 | spin_unlock(&mc.lock); |
4ffef5fe DN |
4729 | } |
4730 | ||
1f7dd3e5 | 4731 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4732 | { |
1f7dd3e5 | 4733 | struct cgroup_subsys_state *css; |
eed67d75 | 4734 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4735 | struct mem_cgroup *from; |
4530eddb | 4736 | struct task_struct *leader, *p; |
9f2115f9 | 4737 | struct mm_struct *mm; |
1dfab5ab | 4738 | unsigned long move_flags; |
9f2115f9 | 4739 | int ret = 0; |
7dc74be0 | 4740 | |
1f7dd3e5 TH |
4741 | /* charge immigration isn't supported on the default hierarchy */ |
4742 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4743 | return 0; |
4744 | ||
4530eddb TH |
4745 | /* |
4746 | * Multi-process migrations only happen on the default hierarchy | |
4747 | * where charge immigration is not used. Perform charge | |
4748 | * immigration if @tset contains a leader and whine if there are | |
4749 | * multiple. | |
4750 | */ | |
4751 | p = NULL; | |
1f7dd3e5 | 4752 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4753 | WARN_ON_ONCE(p); |
4754 | p = leader; | |
1f7dd3e5 | 4755 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4756 | } |
4757 | if (!p) | |
4758 | return 0; | |
4759 | ||
1f7dd3e5 TH |
4760 | /* |
4761 | * We are now commited to this value whatever it is. Changes in this | |
4762 | * tunable will only affect upcoming migrations, not the current one. | |
4763 | * So we need to save it, and keep it going. | |
4764 | */ | |
4765 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4766 | if (!move_flags) | |
4767 | return 0; | |
4768 | ||
9f2115f9 TH |
4769 | from = mem_cgroup_from_task(p); |
4770 | ||
4771 | VM_BUG_ON(from == memcg); | |
4772 | ||
4773 | mm = get_task_mm(p); | |
4774 | if (!mm) | |
4775 | return 0; | |
4776 | /* We move charges only when we move a owner of the mm */ | |
4777 | if (mm->owner == p) { | |
4778 | VM_BUG_ON(mc.from); | |
4779 | VM_BUG_ON(mc.to); | |
4780 | VM_BUG_ON(mc.precharge); | |
4781 | VM_BUG_ON(mc.moved_charge); | |
4782 | VM_BUG_ON(mc.moved_swap); | |
4783 | ||
4784 | spin_lock(&mc.lock); | |
4785 | mc.from = from; | |
4786 | mc.to = memcg; | |
4787 | mc.flags = move_flags; | |
4788 | spin_unlock(&mc.lock); | |
4789 | /* We set mc.moving_task later */ | |
4790 | ||
4791 | ret = mem_cgroup_precharge_mc(mm); | |
4792 | if (ret) | |
4793 | mem_cgroup_clear_mc(); | |
7dc74be0 | 4794 | } |
9f2115f9 | 4795 | mmput(mm); |
7dc74be0 DN |
4796 | return ret; |
4797 | } | |
4798 | ||
1f7dd3e5 | 4799 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4800 | { |
4e2f245d JW |
4801 | if (mc.to) |
4802 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4803 | } |
4804 | ||
4ffef5fe DN |
4805 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4806 | unsigned long addr, unsigned long end, | |
4807 | struct mm_walk *walk) | |
7dc74be0 | 4808 | { |
4ffef5fe | 4809 | int ret = 0; |
26bcd64a | 4810 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4811 | pte_t *pte; |
4812 | spinlock_t *ptl; | |
12724850 NH |
4813 | enum mc_target_type target_type; |
4814 | union mc_target target; | |
4815 | struct page *page; | |
4ffef5fe | 4816 | |
4b471e88 | 4817 | if (pmd_trans_huge_lock(pmd, vma, &ptl)) { |
62ade86a | 4818 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4819 | spin_unlock(ptl); |
12724850 NH |
4820 | return 0; |
4821 | } | |
4822 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4823 | if (target_type == MC_TARGET_PAGE) { | |
4824 | page = target.page; | |
4825 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 4826 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 4827 | mc.from, mc.to)) { |
12724850 NH |
4828 | mc.precharge -= HPAGE_PMD_NR; |
4829 | mc.moved_charge += HPAGE_PMD_NR; | |
4830 | } | |
4831 | putback_lru_page(page); | |
4832 | } | |
4833 | put_page(page); | |
4834 | } | |
bf929152 | 4835 | spin_unlock(ptl); |
1a5a9906 | 4836 | return 0; |
12724850 NH |
4837 | } |
4838 | ||
45f83cef AA |
4839 | if (pmd_trans_unstable(pmd)) |
4840 | return 0; | |
4ffef5fe DN |
4841 | retry: |
4842 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4843 | for (; addr != end; addr += PAGE_SIZE) { | |
4844 | pte_t ptent = *(pte++); | |
02491447 | 4845 | swp_entry_t ent; |
4ffef5fe DN |
4846 | |
4847 | if (!mc.precharge) | |
4848 | break; | |
4849 | ||
8d32ff84 | 4850 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4851 | case MC_TARGET_PAGE: |
4852 | page = target.page; | |
53f9263b KS |
4853 | /* |
4854 | * We can have a part of the split pmd here. Moving it | |
4855 | * can be done but it would be too convoluted so simply | |
4856 | * ignore such a partial THP and keep it in original | |
4857 | * memcg. There should be somebody mapping the head. | |
4858 | */ | |
4859 | if (PageTransCompound(page)) | |
4860 | goto put; | |
4ffef5fe DN |
4861 | if (isolate_lru_page(page)) |
4862 | goto put; | |
f627c2f5 KS |
4863 | if (!mem_cgroup_move_account(page, false, |
4864 | mc.from, mc.to)) { | |
4ffef5fe | 4865 | mc.precharge--; |
854ffa8d DN |
4866 | /* we uncharge from mc.from later. */ |
4867 | mc.moved_charge++; | |
4ffef5fe DN |
4868 | } |
4869 | putback_lru_page(page); | |
8d32ff84 | 4870 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4871 | put_page(page); |
4872 | break; | |
02491447 DN |
4873 | case MC_TARGET_SWAP: |
4874 | ent = target.ent; | |
e91cbb42 | 4875 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4876 | mc.precharge--; |
483c30b5 DN |
4877 | /* we fixup refcnts and charges later. */ |
4878 | mc.moved_swap++; | |
4879 | } | |
02491447 | 4880 | break; |
4ffef5fe DN |
4881 | default: |
4882 | break; | |
4883 | } | |
4884 | } | |
4885 | pte_unmap_unlock(pte - 1, ptl); | |
4886 | cond_resched(); | |
4887 | ||
4888 | if (addr != end) { | |
4889 | /* | |
4890 | * We have consumed all precharges we got in can_attach(). | |
4891 | * We try charge one by one, but don't do any additional | |
4892 | * charges to mc.to if we have failed in charge once in attach() | |
4893 | * phase. | |
4894 | */ | |
854ffa8d | 4895 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4896 | if (!ret) |
4897 | goto retry; | |
4898 | } | |
4899 | ||
4900 | return ret; | |
4901 | } | |
4902 | ||
4903 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4904 | { | |
26bcd64a NH |
4905 | struct mm_walk mem_cgroup_move_charge_walk = { |
4906 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4907 | .mm = mm, | |
4908 | }; | |
4ffef5fe DN |
4909 | |
4910 | lru_add_drain_all(); | |
312722cb JW |
4911 | /* |
4912 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
4913 | * move_lock while we're moving its pages to another memcg. | |
4914 | * Then wait for already started RCU-only updates to finish. | |
4915 | */ | |
4916 | atomic_inc(&mc.from->moving_account); | |
4917 | synchronize_rcu(); | |
dfe076b0 DN |
4918 | retry: |
4919 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
4920 | /* | |
4921 | * Someone who are holding the mmap_sem might be waiting in | |
4922 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4923 | * and retry. Because we cancel precharges, we might not be able | |
4924 | * to move enough charges, but moving charge is a best-effort | |
4925 | * feature anyway, so it wouldn't be a big problem. | |
4926 | */ | |
4927 | __mem_cgroup_clear_mc(); | |
4928 | cond_resched(); | |
4929 | goto retry; | |
4930 | } | |
26bcd64a NH |
4931 | /* |
4932 | * When we have consumed all precharges and failed in doing | |
4933 | * additional charge, the page walk just aborts. | |
4934 | */ | |
4935 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 4936 | up_read(&mm->mmap_sem); |
312722cb | 4937 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
4938 | } |
4939 | ||
1f7dd3e5 | 4940 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
67e465a7 | 4941 | { |
1f7dd3e5 TH |
4942 | struct cgroup_subsys_state *css; |
4943 | struct task_struct *p = cgroup_taskset_first(tset, &css); | |
a433658c | 4944 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 4945 | |
dfe076b0 | 4946 | if (mm) { |
a433658c KM |
4947 | if (mc.to) |
4948 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
4949 | mmput(mm); |
4950 | } | |
a433658c KM |
4951 | if (mc.to) |
4952 | mem_cgroup_clear_mc(); | |
67e465a7 | 4953 | } |
5cfb80a7 | 4954 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 4955 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4956 | { |
4957 | return 0; | |
4958 | } | |
1f7dd3e5 | 4959 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4960 | { |
4961 | } | |
1f7dd3e5 | 4962 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4963 | { |
4964 | } | |
4965 | #endif | |
67e465a7 | 4966 | |
f00baae7 TH |
4967 | /* |
4968 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
4969 | * to verify whether we're attached to the default hierarchy on each mount |
4970 | * attempt. | |
f00baae7 | 4971 | */ |
eb95419b | 4972 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
4973 | { |
4974 | /* | |
aa6ec29b | 4975 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
4976 | * guarantees that @root doesn't have any children, so turning it |
4977 | * on for the root memcg is enough. | |
4978 | */ | |
9e10a130 | 4979 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
4980 | root_mem_cgroup->use_hierarchy = true; |
4981 | else | |
4982 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
4983 | } |
4984 | ||
241994ed JW |
4985 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
4986 | struct cftype *cft) | |
4987 | { | |
f5fc3c5d JW |
4988 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4989 | ||
4990 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
4991 | } |
4992 | ||
4993 | static int memory_low_show(struct seq_file *m, void *v) | |
4994 | { | |
4995 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 4996 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
4997 | |
4998 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 4999 | seq_puts(m, "max\n"); |
241994ed JW |
5000 | else |
5001 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5002 | ||
5003 | return 0; | |
5004 | } | |
5005 | ||
5006 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5007 | char *buf, size_t nbytes, loff_t off) | |
5008 | { | |
5009 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5010 | unsigned long low; | |
5011 | int err; | |
5012 | ||
5013 | buf = strstrip(buf); | |
d2973697 | 5014 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5015 | if (err) |
5016 | return err; | |
5017 | ||
5018 | memcg->low = low; | |
5019 | ||
5020 | return nbytes; | |
5021 | } | |
5022 | ||
5023 | static int memory_high_show(struct seq_file *m, void *v) | |
5024 | { | |
5025 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5026 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5027 | |
5028 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5029 | seq_puts(m, "max\n"); |
241994ed JW |
5030 | else |
5031 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5032 | ||
5033 | return 0; | |
5034 | } | |
5035 | ||
5036 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5037 | char *buf, size_t nbytes, loff_t off) | |
5038 | { | |
5039 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5040 | unsigned long high; | |
5041 | int err; | |
5042 | ||
5043 | buf = strstrip(buf); | |
d2973697 | 5044 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5045 | if (err) |
5046 | return err; | |
5047 | ||
5048 | memcg->high = high; | |
5049 | ||
2529bb3a | 5050 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5051 | return nbytes; |
5052 | } | |
5053 | ||
5054 | static int memory_max_show(struct seq_file *m, void *v) | |
5055 | { | |
5056 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5057 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5058 | |
5059 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5060 | seq_puts(m, "max\n"); |
241994ed JW |
5061 | else |
5062 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5063 | ||
5064 | return 0; | |
5065 | } | |
5066 | ||
5067 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5068 | char *buf, size_t nbytes, loff_t off) | |
5069 | { | |
5070 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5071 | unsigned long max; | |
5072 | int err; | |
5073 | ||
5074 | buf = strstrip(buf); | |
d2973697 | 5075 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5076 | if (err) |
5077 | return err; | |
5078 | ||
5079 | err = mem_cgroup_resize_limit(memcg, max); | |
5080 | if (err) | |
5081 | return err; | |
5082 | ||
2529bb3a | 5083 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5084 | return nbytes; |
5085 | } | |
5086 | ||
5087 | static int memory_events_show(struct seq_file *m, void *v) | |
5088 | { | |
5089 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5090 | ||
5091 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5092 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5093 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5094 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5095 | ||
5096 | return 0; | |
5097 | } | |
5098 | ||
5099 | static struct cftype memory_files[] = { | |
5100 | { | |
5101 | .name = "current", | |
f5fc3c5d | 5102 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5103 | .read_u64 = memory_current_read, |
5104 | }, | |
5105 | { | |
5106 | .name = "low", | |
5107 | .flags = CFTYPE_NOT_ON_ROOT, | |
5108 | .seq_show = memory_low_show, | |
5109 | .write = memory_low_write, | |
5110 | }, | |
5111 | { | |
5112 | .name = "high", | |
5113 | .flags = CFTYPE_NOT_ON_ROOT, | |
5114 | .seq_show = memory_high_show, | |
5115 | .write = memory_high_write, | |
5116 | }, | |
5117 | { | |
5118 | .name = "max", | |
5119 | .flags = CFTYPE_NOT_ON_ROOT, | |
5120 | .seq_show = memory_max_show, | |
5121 | .write = memory_max_write, | |
5122 | }, | |
5123 | { | |
5124 | .name = "events", | |
5125 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5126 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5127 | .seq_show = memory_events_show, |
5128 | }, | |
5129 | { } /* terminate */ | |
5130 | }; | |
5131 | ||
073219e9 | 5132 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5133 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5134 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5135 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5136 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5137 | .css_free = mem_cgroup_css_free, |
1ced953b | 5138 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5139 | .can_attach = mem_cgroup_can_attach, |
5140 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5141 | .attach = mem_cgroup_move_task, |
f00baae7 | 5142 | .bind = mem_cgroup_bind, |
241994ed JW |
5143 | .dfl_cftypes = memory_files, |
5144 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5145 | .early_init = 0, |
8cdea7c0 | 5146 | }; |
c077719b | 5147 | |
241994ed JW |
5148 | /** |
5149 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5150 | * @root: the highest ancestor to consider | |
5151 | * @memcg: the memory cgroup to check | |
5152 | * | |
5153 | * Returns %true if memory consumption of @memcg, and that of all | |
5154 | * configurable ancestors up to @root, is below the normal range. | |
5155 | */ | |
5156 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5157 | { | |
5158 | if (mem_cgroup_disabled()) | |
5159 | return false; | |
5160 | ||
5161 | /* | |
5162 | * The toplevel group doesn't have a configurable range, so | |
5163 | * it's never low when looked at directly, and it is not | |
5164 | * considered an ancestor when assessing the hierarchy. | |
5165 | */ | |
5166 | ||
5167 | if (memcg == root_mem_cgroup) | |
5168 | return false; | |
5169 | ||
4e54dede | 5170 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5171 | return false; |
5172 | ||
5173 | while (memcg != root) { | |
5174 | memcg = parent_mem_cgroup(memcg); | |
5175 | ||
5176 | if (memcg == root_mem_cgroup) | |
5177 | break; | |
5178 | ||
4e54dede | 5179 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5180 | return false; |
5181 | } | |
5182 | return true; | |
5183 | } | |
5184 | ||
00501b53 JW |
5185 | /** |
5186 | * mem_cgroup_try_charge - try charging a page | |
5187 | * @page: page to charge | |
5188 | * @mm: mm context of the victim | |
5189 | * @gfp_mask: reclaim mode | |
5190 | * @memcgp: charged memcg return | |
5191 | * | |
5192 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5193 | * pages according to @gfp_mask if necessary. | |
5194 | * | |
5195 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5196 | * Otherwise, an error code is returned. | |
5197 | * | |
5198 | * After page->mapping has been set up, the caller must finalize the | |
5199 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5200 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5201 | */ | |
5202 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5203 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5204 | bool compound) | |
00501b53 JW |
5205 | { |
5206 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5207 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5208 | int ret = 0; |
5209 | ||
5210 | if (mem_cgroup_disabled()) | |
5211 | goto out; | |
5212 | ||
5213 | if (PageSwapCache(page)) { | |
00501b53 JW |
5214 | /* |
5215 | * Every swap fault against a single page tries to charge the | |
5216 | * page, bail as early as possible. shmem_unuse() encounters | |
5217 | * already charged pages, too. The USED bit is protected by | |
5218 | * the page lock, which serializes swap cache removal, which | |
5219 | * in turn serializes uncharging. | |
5220 | */ | |
e993d905 | 5221 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5222 | if (page->mem_cgroup) |
00501b53 | 5223 | goto out; |
e993d905 | 5224 | |
37e84351 | 5225 | if (do_swap_account) { |
e993d905 VD |
5226 | swp_entry_t ent = { .val = page_private(page), }; |
5227 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5228 | ||
5229 | rcu_read_lock(); | |
5230 | memcg = mem_cgroup_from_id(id); | |
5231 | if (memcg && !css_tryget_online(&memcg->css)) | |
5232 | memcg = NULL; | |
5233 | rcu_read_unlock(); | |
5234 | } | |
00501b53 JW |
5235 | } |
5236 | ||
00501b53 JW |
5237 | if (!memcg) |
5238 | memcg = get_mem_cgroup_from_mm(mm); | |
5239 | ||
5240 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5241 | ||
5242 | css_put(&memcg->css); | |
00501b53 JW |
5243 | out: |
5244 | *memcgp = memcg; | |
5245 | return ret; | |
5246 | } | |
5247 | ||
5248 | /** | |
5249 | * mem_cgroup_commit_charge - commit a page charge | |
5250 | * @page: page to charge | |
5251 | * @memcg: memcg to charge the page to | |
5252 | * @lrucare: page might be on LRU already | |
5253 | * | |
5254 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5255 | * after page->mapping has been set up. This must happen atomically | |
5256 | * as part of the page instantiation, i.e. under the page table lock | |
5257 | * for anonymous pages, under the page lock for page and swap cache. | |
5258 | * | |
5259 | * In addition, the page must not be on the LRU during the commit, to | |
5260 | * prevent racing with task migration. If it might be, use @lrucare. | |
5261 | * | |
5262 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5263 | */ | |
5264 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5265 | bool lrucare, bool compound) |
00501b53 | 5266 | { |
f627c2f5 | 5267 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5268 | |
5269 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5270 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5271 | ||
5272 | if (mem_cgroup_disabled()) | |
5273 | return; | |
5274 | /* | |
5275 | * Swap faults will attempt to charge the same page multiple | |
5276 | * times. But reuse_swap_page() might have removed the page | |
5277 | * from swapcache already, so we can't check PageSwapCache(). | |
5278 | */ | |
5279 | if (!memcg) | |
5280 | return; | |
5281 | ||
6abb5a86 JW |
5282 | commit_charge(page, memcg, lrucare); |
5283 | ||
6abb5a86 | 5284 | local_irq_disable(); |
f627c2f5 | 5285 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5286 | memcg_check_events(memcg, page); |
5287 | local_irq_enable(); | |
00501b53 | 5288 | |
7941d214 | 5289 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5290 | swp_entry_t entry = { .val = page_private(page) }; |
5291 | /* | |
5292 | * The swap entry might not get freed for a long time, | |
5293 | * let's not wait for it. The page already received a | |
5294 | * memory+swap charge, drop the swap entry duplicate. | |
5295 | */ | |
5296 | mem_cgroup_uncharge_swap(entry); | |
5297 | } | |
5298 | } | |
5299 | ||
5300 | /** | |
5301 | * mem_cgroup_cancel_charge - cancel a page charge | |
5302 | * @page: page to charge | |
5303 | * @memcg: memcg to charge the page to | |
5304 | * | |
5305 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5306 | */ | |
f627c2f5 KS |
5307 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5308 | bool compound) | |
00501b53 | 5309 | { |
f627c2f5 | 5310 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5311 | |
5312 | if (mem_cgroup_disabled()) | |
5313 | return; | |
5314 | /* | |
5315 | * Swap faults will attempt to charge the same page multiple | |
5316 | * times. But reuse_swap_page() might have removed the page | |
5317 | * from swapcache already, so we can't check PageSwapCache(). | |
5318 | */ | |
5319 | if (!memcg) | |
5320 | return; | |
5321 | ||
00501b53 JW |
5322 | cancel_charge(memcg, nr_pages); |
5323 | } | |
5324 | ||
747db954 | 5325 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5326 | unsigned long nr_anon, unsigned long nr_file, |
5327 | unsigned long nr_huge, struct page *dummy_page) | |
5328 | { | |
18eca2e6 | 5329 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5330 | unsigned long flags; |
5331 | ||
ce00a967 | 5332 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 | 5333 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 5334 | if (do_memsw_account()) |
18eca2e6 | 5335 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 JW |
5336 | memcg_oom_recover(memcg); |
5337 | } | |
747db954 JW |
5338 | |
5339 | local_irq_save(flags); | |
5340 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5341 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5342 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5343 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5344 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5345 | memcg_check_events(memcg, dummy_page); |
5346 | local_irq_restore(flags); | |
e8ea14cc JW |
5347 | |
5348 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5349 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5350 | } |
5351 | ||
5352 | static void uncharge_list(struct list_head *page_list) | |
5353 | { | |
5354 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5355 | unsigned long nr_anon = 0; |
5356 | unsigned long nr_file = 0; | |
5357 | unsigned long nr_huge = 0; | |
5358 | unsigned long pgpgout = 0; | |
747db954 JW |
5359 | struct list_head *next; |
5360 | struct page *page; | |
5361 | ||
5362 | next = page_list->next; | |
5363 | do { | |
5364 | unsigned int nr_pages = 1; | |
747db954 JW |
5365 | |
5366 | page = list_entry(next, struct page, lru); | |
5367 | next = page->lru.next; | |
5368 | ||
5369 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5370 | VM_BUG_ON_PAGE(page_count(page), page); | |
5371 | ||
1306a85a | 5372 | if (!page->mem_cgroup) |
747db954 JW |
5373 | continue; |
5374 | ||
5375 | /* | |
5376 | * Nobody should be changing or seriously looking at | |
1306a85a | 5377 | * page->mem_cgroup at this point, we have fully |
29833315 | 5378 | * exclusive access to the page. |
747db954 JW |
5379 | */ |
5380 | ||
1306a85a | 5381 | if (memcg != page->mem_cgroup) { |
747db954 | 5382 | if (memcg) { |
18eca2e6 JW |
5383 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5384 | nr_huge, page); | |
5385 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5386 | } |
1306a85a | 5387 | memcg = page->mem_cgroup; |
747db954 JW |
5388 | } |
5389 | ||
5390 | if (PageTransHuge(page)) { | |
5391 | nr_pages <<= compound_order(page); | |
5392 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5393 | nr_huge += nr_pages; | |
5394 | } | |
5395 | ||
5396 | if (PageAnon(page)) | |
5397 | nr_anon += nr_pages; | |
5398 | else | |
5399 | nr_file += nr_pages; | |
5400 | ||
1306a85a | 5401 | page->mem_cgroup = NULL; |
747db954 JW |
5402 | |
5403 | pgpgout++; | |
5404 | } while (next != page_list); | |
5405 | ||
5406 | if (memcg) | |
18eca2e6 JW |
5407 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5408 | nr_huge, page); | |
747db954 JW |
5409 | } |
5410 | ||
0a31bc97 JW |
5411 | /** |
5412 | * mem_cgroup_uncharge - uncharge a page | |
5413 | * @page: page to uncharge | |
5414 | * | |
5415 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5416 | * mem_cgroup_commit_charge(). | |
5417 | */ | |
5418 | void mem_cgroup_uncharge(struct page *page) | |
5419 | { | |
0a31bc97 JW |
5420 | if (mem_cgroup_disabled()) |
5421 | return; | |
5422 | ||
747db954 | 5423 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5424 | if (!page->mem_cgroup) |
0a31bc97 JW |
5425 | return; |
5426 | ||
747db954 JW |
5427 | INIT_LIST_HEAD(&page->lru); |
5428 | uncharge_list(&page->lru); | |
5429 | } | |
0a31bc97 | 5430 | |
747db954 JW |
5431 | /** |
5432 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5433 | * @page_list: list of pages to uncharge | |
5434 | * | |
5435 | * Uncharge a list of pages previously charged with | |
5436 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5437 | */ | |
5438 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5439 | { | |
5440 | if (mem_cgroup_disabled()) | |
5441 | return; | |
0a31bc97 | 5442 | |
747db954 JW |
5443 | if (!list_empty(page_list)) |
5444 | uncharge_list(page_list); | |
0a31bc97 JW |
5445 | } |
5446 | ||
5447 | /** | |
45637bab | 5448 | * mem_cgroup_replace_page - migrate a charge to another page |
0a31bc97 JW |
5449 | * @oldpage: currently charged page |
5450 | * @newpage: page to transfer the charge to | |
0a31bc97 JW |
5451 | * |
5452 | * Migrate the charge from @oldpage to @newpage. | |
5453 | * | |
5454 | * Both pages must be locked, @newpage->mapping must be set up. | |
25be6a65 | 5455 | * Either or both pages might be on the LRU already. |
0a31bc97 | 5456 | */ |
45637bab | 5457 | void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5458 | { |
29833315 | 5459 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
5460 | unsigned int nr_pages; |
5461 | bool compound; | |
0a31bc97 JW |
5462 | |
5463 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5464 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5465 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5466 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5467 | newpage); | |
0a31bc97 JW |
5468 | |
5469 | if (mem_cgroup_disabled()) | |
5470 | return; | |
5471 | ||
5472 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5473 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5474 | return; |
5475 | ||
45637bab | 5476 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5477 | memcg = oldpage->mem_cgroup; |
29833315 | 5478 | if (!memcg) |
0a31bc97 JW |
5479 | return; |
5480 | ||
44b7a8d3 JW |
5481 | /* Force-charge the new page. The old one will be freed soon */ |
5482 | compound = PageTransHuge(newpage); | |
5483 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
5484 | ||
5485 | page_counter_charge(&memcg->memory, nr_pages); | |
5486 | if (do_memsw_account()) | |
5487 | page_counter_charge(&memcg->memsw, nr_pages); | |
5488 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 5489 | |
45637bab | 5490 | commit_charge(newpage, memcg, true); |
44b7a8d3 JW |
5491 | |
5492 | local_irq_disable(); | |
5493 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); | |
5494 | memcg_check_events(memcg, newpage); | |
5495 | local_irq_enable(); | |
0a31bc97 JW |
5496 | } |
5497 | ||
ef12947c | 5498 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5499 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5500 | ||
5501 | void sock_update_memcg(struct sock *sk) | |
5502 | { | |
5503 | struct mem_cgroup *memcg; | |
5504 | ||
5505 | /* Socket cloning can throw us here with sk_cgrp already | |
5506 | * filled. It won't however, necessarily happen from | |
5507 | * process context. So the test for root memcg given | |
5508 | * the current task's memcg won't help us in this case. | |
5509 | * | |
5510 | * Respecting the original socket's memcg is a better | |
5511 | * decision in this case. | |
5512 | */ | |
5513 | if (sk->sk_memcg) { | |
5514 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
5515 | css_get(&sk->sk_memcg->css); | |
5516 | return; | |
5517 | } | |
5518 | ||
5519 | rcu_read_lock(); | |
5520 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5521 | if (memcg == root_mem_cgroup) |
5522 | goto out; | |
0db15298 | 5523 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 5524 | goto out; |
f7e1cb6e | 5525 | if (css_tryget_online(&memcg->css)) |
11092087 | 5526 | sk->sk_memcg = memcg; |
f7e1cb6e | 5527 | out: |
11092087 JW |
5528 | rcu_read_unlock(); |
5529 | } | |
5530 | EXPORT_SYMBOL(sock_update_memcg); | |
5531 | ||
5532 | void sock_release_memcg(struct sock *sk) | |
5533 | { | |
5534 | WARN_ON(!sk->sk_memcg); | |
5535 | css_put(&sk->sk_memcg->css); | |
5536 | } | |
5537 | ||
5538 | /** | |
5539 | * mem_cgroup_charge_skmem - charge socket memory | |
5540 | * @memcg: memcg to charge | |
5541 | * @nr_pages: number of pages to charge | |
5542 | * | |
5543 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5544 | * @memcg's configured limit, %false if the charge had to be forced. | |
5545 | */ | |
5546 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5547 | { | |
f7e1cb6e | 5548 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5549 | |
f7e1cb6e | 5550 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5551 | struct page_counter *fail; |
f7e1cb6e | 5552 | |
0db15298 JW |
5553 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
5554 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
5555 | return true; |
5556 | } | |
0db15298 JW |
5557 | page_counter_charge(&memcg->tcpmem, nr_pages); |
5558 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 5559 | return false; |
11092087 | 5560 | } |
d886f4e4 | 5561 | |
f7e1cb6e JW |
5562 | /* Don't block in the packet receive path */ |
5563 | if (in_softirq()) | |
5564 | gfp_mask = GFP_NOWAIT; | |
5565 | ||
5566 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) | |
5567 | return true; | |
5568 | ||
5569 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5570 | return false; |
5571 | } | |
5572 | ||
5573 | /** | |
5574 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5575 | * @memcg - memcg to uncharge | |
5576 | * @nr_pages - number of pages to uncharge | |
5577 | */ | |
5578 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5579 | { | |
f7e1cb6e | 5580 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5581 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
5582 | return; |
5583 | } | |
d886f4e4 | 5584 | |
f7e1cb6e JW |
5585 | page_counter_uncharge(&memcg->memory, nr_pages); |
5586 | css_put_many(&memcg->css, nr_pages); | |
11092087 JW |
5587 | } |
5588 | ||
f7e1cb6e JW |
5589 | static int __init cgroup_memory(char *s) |
5590 | { | |
5591 | char *token; | |
5592 | ||
5593 | while ((token = strsep(&s, ",")) != NULL) { | |
5594 | if (!*token) | |
5595 | continue; | |
5596 | if (!strcmp(token, "nosocket")) | |
5597 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5598 | if (!strcmp(token, "nokmem")) |
5599 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5600 | } |
5601 | return 0; | |
5602 | } | |
5603 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5604 | |
2d11085e | 5605 | /* |
1081312f MH |
5606 | * subsys_initcall() for memory controller. |
5607 | * | |
5608 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5609 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5610 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5611 | * be initialized from here. | |
2d11085e MH |
5612 | */ |
5613 | static int __init mem_cgroup_init(void) | |
5614 | { | |
95a045f6 JW |
5615 | int cpu, node; |
5616 | ||
2d11085e | 5617 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5618 | |
5619 | for_each_possible_cpu(cpu) | |
5620 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5621 | drain_local_stock); | |
5622 | ||
5623 | for_each_node(node) { | |
5624 | struct mem_cgroup_tree_per_node *rtpn; | |
5625 | int zone; | |
5626 | ||
5627 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5628 | node_online(node) ? node : NUMA_NO_NODE); | |
5629 | ||
5630 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5631 | struct mem_cgroup_tree_per_zone *rtpz; | |
5632 | ||
5633 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5634 | rtpz->rb_root = RB_ROOT; | |
5635 | spin_lock_init(&rtpz->lock); | |
5636 | } | |
5637 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5638 | } | |
5639 | ||
2d11085e MH |
5640 | return 0; |
5641 | } | |
5642 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5643 | |
5644 | #ifdef CONFIG_MEMCG_SWAP | |
5645 | /** | |
5646 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5647 | * @page: page whose memsw charge to transfer | |
5648 | * @entry: swap entry to move the charge to | |
5649 | * | |
5650 | * Transfer the memsw charge of @page to @entry. | |
5651 | */ | |
5652 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5653 | { | |
5654 | struct mem_cgroup *memcg; | |
5655 | unsigned short oldid; | |
5656 | ||
5657 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5658 | VM_BUG_ON_PAGE(page_count(page), page); | |
5659 | ||
7941d214 | 5660 | if (!do_memsw_account()) |
21afa38e JW |
5661 | return; |
5662 | ||
5663 | memcg = page->mem_cgroup; | |
5664 | ||
5665 | /* Readahead page, never charged */ | |
5666 | if (!memcg) | |
5667 | return; | |
5668 | ||
5669 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5670 | VM_BUG_ON_PAGE(oldid, page); | |
5671 | mem_cgroup_swap_statistics(memcg, true); | |
5672 | ||
5673 | page->mem_cgroup = NULL; | |
5674 | ||
5675 | if (!mem_cgroup_is_root(memcg)) | |
5676 | page_counter_uncharge(&memcg->memory, 1); | |
5677 | ||
ce9ce665 SAS |
5678 | /* |
5679 | * Interrupts should be disabled here because the caller holds the | |
5680 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5681 | * important here to have the interrupts disabled because it is the | |
5682 | * only synchronisation we have for udpating the per-CPU variables. | |
5683 | */ | |
5684 | VM_BUG_ON(!irqs_disabled()); | |
f627c2f5 | 5685 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
21afa38e JW |
5686 | memcg_check_events(memcg, page); |
5687 | } | |
5688 | ||
37e84351 VD |
5689 | /* |
5690 | * mem_cgroup_try_charge_swap - try charging a swap entry | |
5691 | * @page: page being added to swap | |
5692 | * @entry: swap entry to charge | |
5693 | * | |
5694 | * Try to charge @entry to the memcg that @page belongs to. | |
5695 | * | |
5696 | * Returns 0 on success, -ENOMEM on failure. | |
5697 | */ | |
5698 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
5699 | { | |
5700 | struct mem_cgroup *memcg; | |
5701 | struct page_counter *counter; | |
5702 | unsigned short oldid; | |
5703 | ||
5704 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
5705 | return 0; | |
5706 | ||
5707 | memcg = page->mem_cgroup; | |
5708 | ||
5709 | /* Readahead page, never charged */ | |
5710 | if (!memcg) | |
5711 | return 0; | |
5712 | ||
5713 | if (!mem_cgroup_is_root(memcg) && | |
5714 | !page_counter_try_charge(&memcg->swap, 1, &counter)) | |
5715 | return -ENOMEM; | |
5716 | ||
5717 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5718 | VM_BUG_ON_PAGE(oldid, page); | |
5719 | mem_cgroup_swap_statistics(memcg, true); | |
5720 | ||
5721 | css_get(&memcg->css); | |
5722 | return 0; | |
5723 | } | |
5724 | ||
21afa38e JW |
5725 | /** |
5726 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5727 | * @entry: swap entry to uncharge | |
5728 | * | |
37e84351 | 5729 | * Drop the swap charge associated with @entry. |
21afa38e JW |
5730 | */ |
5731 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5732 | { | |
5733 | struct mem_cgroup *memcg; | |
5734 | unsigned short id; | |
5735 | ||
37e84351 | 5736 | if (!do_swap_account) |
21afa38e JW |
5737 | return; |
5738 | ||
5739 | id = swap_cgroup_record(entry, 0); | |
5740 | rcu_read_lock(); | |
adbe427b | 5741 | memcg = mem_cgroup_from_id(id); |
21afa38e | 5742 | if (memcg) { |
37e84351 VD |
5743 | if (!mem_cgroup_is_root(memcg)) { |
5744 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5745 | page_counter_uncharge(&memcg->swap, 1); | |
5746 | else | |
5747 | page_counter_uncharge(&memcg->memsw, 1); | |
5748 | } | |
21afa38e JW |
5749 | mem_cgroup_swap_statistics(memcg, false); |
5750 | css_put(&memcg->css); | |
5751 | } | |
5752 | rcu_read_unlock(); | |
5753 | } | |
5754 | ||
d8b38438 VD |
5755 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
5756 | { | |
5757 | long nr_swap_pages = get_nr_swap_pages(); | |
5758 | ||
5759 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5760 | return nr_swap_pages; | |
5761 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5762 | nr_swap_pages = min_t(long, nr_swap_pages, | |
5763 | READ_ONCE(memcg->swap.limit) - | |
5764 | page_counter_read(&memcg->swap)); | |
5765 | return nr_swap_pages; | |
5766 | } | |
5767 | ||
5ccc5aba VD |
5768 | bool mem_cgroup_swap_full(struct page *page) |
5769 | { | |
5770 | struct mem_cgroup *memcg; | |
5771 | ||
5772 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
5773 | ||
5774 | if (vm_swap_full()) | |
5775 | return true; | |
5776 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5777 | return false; | |
5778 | ||
5779 | memcg = page->mem_cgroup; | |
5780 | if (!memcg) | |
5781 | return false; | |
5782 | ||
5783 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5784 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) | |
5785 | return true; | |
5786 | ||
5787 | return false; | |
5788 | } | |
5789 | ||
21afa38e JW |
5790 | /* for remember boot option*/ |
5791 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5792 | static int really_do_swap_account __initdata = 1; | |
5793 | #else | |
5794 | static int really_do_swap_account __initdata; | |
5795 | #endif | |
5796 | ||
5797 | static int __init enable_swap_account(char *s) | |
5798 | { | |
5799 | if (!strcmp(s, "1")) | |
5800 | really_do_swap_account = 1; | |
5801 | else if (!strcmp(s, "0")) | |
5802 | really_do_swap_account = 0; | |
5803 | return 1; | |
5804 | } | |
5805 | __setup("swapaccount=", enable_swap_account); | |
5806 | ||
37e84351 VD |
5807 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
5808 | struct cftype *cft) | |
5809 | { | |
5810 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5811 | ||
5812 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
5813 | } | |
5814 | ||
5815 | static int swap_max_show(struct seq_file *m, void *v) | |
5816 | { | |
5817 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5818 | unsigned long max = READ_ONCE(memcg->swap.limit); | |
5819 | ||
5820 | if (max == PAGE_COUNTER_MAX) | |
5821 | seq_puts(m, "max\n"); | |
5822 | else | |
5823 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5824 | ||
5825 | return 0; | |
5826 | } | |
5827 | ||
5828 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
5829 | char *buf, size_t nbytes, loff_t off) | |
5830 | { | |
5831 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5832 | unsigned long max; | |
5833 | int err; | |
5834 | ||
5835 | buf = strstrip(buf); | |
5836 | err = page_counter_memparse(buf, "max", &max); | |
5837 | if (err) | |
5838 | return err; | |
5839 | ||
5840 | mutex_lock(&memcg_limit_mutex); | |
5841 | err = page_counter_limit(&memcg->swap, max); | |
5842 | mutex_unlock(&memcg_limit_mutex); | |
5843 | if (err) | |
5844 | return err; | |
5845 | ||
5846 | return nbytes; | |
5847 | } | |
5848 | ||
5849 | static struct cftype swap_files[] = { | |
5850 | { | |
5851 | .name = "swap.current", | |
5852 | .flags = CFTYPE_NOT_ON_ROOT, | |
5853 | .read_u64 = swap_current_read, | |
5854 | }, | |
5855 | { | |
5856 | .name = "swap.max", | |
5857 | .flags = CFTYPE_NOT_ON_ROOT, | |
5858 | .seq_show = swap_max_show, | |
5859 | .write = swap_max_write, | |
5860 | }, | |
5861 | { } /* terminate */ | |
5862 | }; | |
5863 | ||
21afa38e JW |
5864 | static struct cftype memsw_cgroup_files[] = { |
5865 | { | |
5866 | .name = "memsw.usage_in_bytes", | |
5867 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5868 | .read_u64 = mem_cgroup_read_u64, | |
5869 | }, | |
5870 | { | |
5871 | .name = "memsw.max_usage_in_bytes", | |
5872 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5873 | .write = mem_cgroup_reset, | |
5874 | .read_u64 = mem_cgroup_read_u64, | |
5875 | }, | |
5876 | { | |
5877 | .name = "memsw.limit_in_bytes", | |
5878 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5879 | .write = mem_cgroup_write, | |
5880 | .read_u64 = mem_cgroup_read_u64, | |
5881 | }, | |
5882 | { | |
5883 | .name = "memsw.failcnt", | |
5884 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5885 | .write = mem_cgroup_reset, | |
5886 | .read_u64 = mem_cgroup_read_u64, | |
5887 | }, | |
5888 | { }, /* terminate */ | |
5889 | }; | |
5890 | ||
5891 | static int __init mem_cgroup_swap_init(void) | |
5892 | { | |
5893 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5894 | do_swap_account = 1; | |
37e84351 VD |
5895 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
5896 | swap_files)); | |
21afa38e JW |
5897 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5898 | memsw_cgroup_files)); | |
5899 | } | |
5900 | return 0; | |
5901 | } | |
5902 | subsys_initcall(mem_cgroup_swap_init); | |
5903 | ||
5904 | #endif /* CONFIG_MEMCG_SWAP */ |