]>
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 */ |
264a0ae1 | 210 | struct mm_struct *mm; |
4ffef5fe DN |
211 | struct mem_cgroup *from; |
212 | struct mem_cgroup *to; | |
1dfab5ab | 213 | unsigned long flags; |
4ffef5fe | 214 | unsigned long precharge; |
854ffa8d | 215 | unsigned long moved_charge; |
483c30b5 | 216 | unsigned long moved_swap; |
8033b97c DN |
217 | struct task_struct *moving_task; /* a task moving charges */ |
218 | wait_queue_head_t waitq; /* a waitq for other context */ | |
219 | } mc = { | |
2bd9bb20 | 220 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
221 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
222 | }; | |
4ffef5fe | 223 | |
4e416953 BS |
224 | /* |
225 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
226 | * limit reclaim to prevent infinite loops, if they ever occur. | |
227 | */ | |
a0db00fc | 228 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 229 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 230 | |
217bc319 KH |
231 | enum charge_type { |
232 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 233 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 234 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 235 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
236 | NR_CHARGE_TYPE, |
237 | }; | |
238 | ||
8c7c6e34 | 239 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
240 | enum res_type { |
241 | _MEM, | |
242 | _MEMSWAP, | |
243 | _OOM_TYPE, | |
510fc4e1 | 244 | _KMEM, |
d55f90bf | 245 | _TCP, |
86ae53e1 GC |
246 | }; |
247 | ||
a0db00fc KS |
248 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
249 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 250 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
251 | /* Used for OOM nofiier */ |
252 | #define OOM_CONTROL (0) | |
8c7c6e34 | 253 | |
70ddf637 AV |
254 | /* Some nice accessors for the vmpressure. */ |
255 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
256 | { | |
257 | if (!memcg) | |
258 | memcg = root_mem_cgroup; | |
259 | return &memcg->vmpressure; | |
260 | } | |
261 | ||
262 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
263 | { | |
264 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
265 | } | |
266 | ||
7ffc0edc MH |
267 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
268 | { | |
269 | return (memcg == root_mem_cgroup); | |
270 | } | |
271 | ||
127424c8 | 272 | #ifndef CONFIG_SLOB |
55007d84 | 273 | /* |
f7ce3190 | 274 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
275 | * The main reason for not using cgroup id for this: |
276 | * this works better in sparse environments, where we have a lot of memcgs, | |
277 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
278 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
279 | * 200 entry array for that. | |
55007d84 | 280 | * |
dbcf73e2 VD |
281 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
282 | * will double each time we have to increase it. | |
55007d84 | 283 | */ |
dbcf73e2 VD |
284 | static DEFINE_IDA(memcg_cache_ida); |
285 | int memcg_nr_cache_ids; | |
749c5415 | 286 | |
05257a1a VD |
287 | /* Protects memcg_nr_cache_ids */ |
288 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
289 | ||
290 | void memcg_get_cache_ids(void) | |
291 | { | |
292 | down_read(&memcg_cache_ids_sem); | |
293 | } | |
294 | ||
295 | void memcg_put_cache_ids(void) | |
296 | { | |
297 | up_read(&memcg_cache_ids_sem); | |
298 | } | |
299 | ||
55007d84 GC |
300 | /* |
301 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
302 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
303 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
304 | * tunable, but that is strictly not necessary. | |
305 | * | |
b8627835 | 306 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
307 | * this constant directly from cgroup, but it is understandable that this is |
308 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 309 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
310 | * increase ours as well if it increases. |
311 | */ | |
312 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 313 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 314 | |
d7f25f8a GC |
315 | /* |
316 | * A lot of the calls to the cache allocation functions are expected to be | |
317 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
318 | * conditional to this static branch, we'll have to allow modules that does | |
319 | * kmem_cache_alloc and the such to see this symbol as well | |
320 | */ | |
ef12947c | 321 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 322 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 323 | |
127424c8 | 324 | #endif /* !CONFIG_SLOB */ |
a8964b9b | 325 | |
f64c3f54 | 326 | static struct mem_cgroup_per_zone * |
e231875b | 327 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 328 | { |
e231875b JZ |
329 | int nid = zone_to_nid(zone); |
330 | int zid = zone_idx(zone); | |
331 | ||
54f72fe0 | 332 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
333 | } |
334 | ||
ad7fa852 TH |
335 | /** |
336 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
337 | * @page: page of interest | |
338 | * | |
339 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
340 | * with @page is returned. The returned css remains associated with @page | |
341 | * until it is released. | |
342 | * | |
343 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
344 | * is returned. | |
ad7fa852 TH |
345 | */ |
346 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
347 | { | |
348 | struct mem_cgroup *memcg; | |
349 | ||
ad7fa852 TH |
350 | memcg = page->mem_cgroup; |
351 | ||
9e10a130 | 352 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
353 | memcg = root_mem_cgroup; |
354 | ||
ad7fa852 TH |
355 | return &memcg->css; |
356 | } | |
357 | ||
2fc04524 VD |
358 | /** |
359 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
360 | * @page: the page | |
361 | * | |
362 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
363 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
364 | * is safe to call this function without holding a reference to @page. | |
365 | * | |
366 | * Note, this function is inherently racy, because there is nothing to prevent | |
367 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
368 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
369 | * do not care (such as procfs interfaces). | |
370 | */ | |
371 | ino_t page_cgroup_ino(struct page *page) | |
372 | { | |
373 | struct mem_cgroup *memcg; | |
374 | unsigned long ino = 0; | |
375 | ||
376 | rcu_read_lock(); | |
377 | memcg = READ_ONCE(page->mem_cgroup); | |
378 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
379 | memcg = parent_mem_cgroup(memcg); | |
380 | if (memcg) | |
381 | ino = cgroup_ino(memcg->css.cgroup); | |
382 | rcu_read_unlock(); | |
383 | return ino; | |
384 | } | |
385 | ||
f64c3f54 | 386 | static struct mem_cgroup_per_zone * |
e231875b | 387 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 388 | { |
97a6c37b JW |
389 | int nid = page_to_nid(page); |
390 | int zid = page_zonenum(page); | |
f64c3f54 | 391 | |
e231875b | 392 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
393 | } |
394 | ||
bb4cc1a8 AM |
395 | static struct mem_cgroup_tree_per_zone * |
396 | soft_limit_tree_node_zone(int nid, int zid) | |
397 | { | |
398 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
399 | } | |
400 | ||
401 | static struct mem_cgroup_tree_per_zone * | |
402 | soft_limit_tree_from_page(struct page *page) | |
403 | { | |
404 | int nid = page_to_nid(page); | |
405 | int zid = page_zonenum(page); | |
406 | ||
407 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
408 | } | |
409 | ||
cf2c8127 JW |
410 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
411 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 412 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
413 | { |
414 | struct rb_node **p = &mctz->rb_root.rb_node; | |
415 | struct rb_node *parent = NULL; | |
416 | struct mem_cgroup_per_zone *mz_node; | |
417 | ||
418 | if (mz->on_tree) | |
419 | return; | |
420 | ||
421 | mz->usage_in_excess = new_usage_in_excess; | |
422 | if (!mz->usage_in_excess) | |
423 | return; | |
424 | while (*p) { | |
425 | parent = *p; | |
426 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
427 | tree_node); | |
428 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
429 | p = &(*p)->rb_left; | |
430 | /* | |
431 | * We can't avoid mem cgroups that are over their soft | |
432 | * limit by the same amount | |
433 | */ | |
434 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
435 | p = &(*p)->rb_right; | |
436 | } | |
437 | rb_link_node(&mz->tree_node, parent, p); | |
438 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
439 | mz->on_tree = true; | |
440 | } | |
441 | ||
cf2c8127 JW |
442 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
443 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
444 | { |
445 | if (!mz->on_tree) | |
446 | return; | |
447 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
448 | mz->on_tree = false; | |
449 | } | |
450 | ||
cf2c8127 JW |
451 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
452 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 453 | { |
0a31bc97 JW |
454 | unsigned long flags; |
455 | ||
456 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 457 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 458 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
459 | } |
460 | ||
3e32cb2e JW |
461 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
462 | { | |
463 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 464 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
465 | unsigned long excess = 0; |
466 | ||
467 | if (nr_pages > soft_limit) | |
468 | excess = nr_pages - soft_limit; | |
469 | ||
470 | return excess; | |
471 | } | |
bb4cc1a8 AM |
472 | |
473 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
474 | { | |
3e32cb2e | 475 | unsigned long excess; |
bb4cc1a8 AM |
476 | struct mem_cgroup_per_zone *mz; |
477 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 478 | |
e231875b | 479 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
480 | /* |
481 | * Necessary to update all ancestors when hierarchy is used. | |
482 | * because their event counter is not touched. | |
483 | */ | |
484 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 485 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 486 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
487 | /* |
488 | * We have to update the tree if mz is on RB-tree or | |
489 | * mem is over its softlimit. | |
490 | */ | |
491 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
492 | unsigned long flags; |
493 | ||
494 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
495 | /* if on-tree, remove it */ |
496 | if (mz->on_tree) | |
cf2c8127 | 497 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
498 | /* |
499 | * Insert again. mz->usage_in_excess will be updated. | |
500 | * If excess is 0, no tree ops. | |
501 | */ | |
cf2c8127 | 502 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 503 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
504 | } |
505 | } | |
506 | } | |
507 | ||
508 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
509 | { | |
bb4cc1a8 | 510 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
511 | struct mem_cgroup_per_zone *mz; |
512 | int nid, zid; | |
bb4cc1a8 | 513 | |
e231875b JZ |
514 | for_each_node(nid) { |
515 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
516 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
517 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 518 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
519 | } |
520 | } | |
521 | } | |
522 | ||
523 | static struct mem_cgroup_per_zone * | |
524 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
525 | { | |
526 | struct rb_node *rightmost = NULL; | |
527 | struct mem_cgroup_per_zone *mz; | |
528 | ||
529 | retry: | |
530 | mz = NULL; | |
531 | rightmost = rb_last(&mctz->rb_root); | |
532 | if (!rightmost) | |
533 | goto done; /* Nothing to reclaim from */ | |
534 | ||
535 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
536 | /* | |
537 | * Remove the node now but someone else can add it back, | |
538 | * we will to add it back at the end of reclaim to its correct | |
539 | * position in the tree. | |
540 | */ | |
cf2c8127 | 541 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 542 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 543 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
544 | goto retry; |
545 | done: | |
546 | return mz; | |
547 | } | |
548 | ||
549 | static struct mem_cgroup_per_zone * | |
550 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
551 | { | |
552 | struct mem_cgroup_per_zone *mz; | |
553 | ||
0a31bc97 | 554 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 555 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 556 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
557 | return mz; |
558 | } | |
559 | ||
711d3d2c | 560 | /* |
484ebb3b GT |
561 | * Return page count for single (non recursive) @memcg. |
562 | * | |
711d3d2c KH |
563 | * Implementation Note: reading percpu statistics for memcg. |
564 | * | |
565 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
566 | * synchronization to implement "quick" read. There are trade-off between | |
567 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 568 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
569 | * |
570 | * But this _read() function is used for user interface now. The user accounts | |
571 | * memory usage by memory cgroup and he _always_ requires exact value because | |
572 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
573 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
574 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
575 | * | |
576 | * If there are kernel internal actions which can make use of some not-exact | |
577 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 578 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
579 | * implemented. |
580 | */ | |
484ebb3b GT |
581 | static unsigned long |
582 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 583 | { |
7a159cc9 | 584 | long val = 0; |
c62b1a3b | 585 | int cpu; |
c62b1a3b | 586 | |
484ebb3b | 587 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 588 | for_each_possible_cpu(cpu) |
c0ff4b85 | 589 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
590 | /* |
591 | * Summing races with updates, so val may be negative. Avoid exposing | |
592 | * transient negative values. | |
593 | */ | |
594 | if (val < 0) | |
595 | val = 0; | |
c62b1a3b KH |
596 | return val; |
597 | } | |
598 | ||
c0ff4b85 | 599 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
600 | enum mem_cgroup_events_index idx) |
601 | { | |
602 | unsigned long val = 0; | |
603 | int cpu; | |
604 | ||
733a572e | 605 | for_each_possible_cpu(cpu) |
c0ff4b85 | 606 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
607 | return val; |
608 | } | |
609 | ||
c0ff4b85 | 610 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 611 | struct page *page, |
f627c2f5 | 612 | bool compound, int nr_pages) |
d52aa412 | 613 | { |
b2402857 KH |
614 | /* |
615 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
616 | * counted as CACHE even if it's on ANON LRU. | |
617 | */ | |
0a31bc97 | 618 | if (PageAnon(page)) |
b2402857 | 619 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 620 | nr_pages); |
d52aa412 | 621 | else |
b2402857 | 622 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 623 | nr_pages); |
55e462b0 | 624 | |
f627c2f5 KS |
625 | if (compound) { |
626 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
b070e65c DR |
627 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
628 | nr_pages); | |
f627c2f5 | 629 | } |
b070e65c | 630 | |
e401f176 KH |
631 | /* pagein of a big page is an event. So, ignore page size */ |
632 | if (nr_pages > 0) | |
c0ff4b85 | 633 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 634 | else { |
c0ff4b85 | 635 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
636 | nr_pages = -nr_pages; /* for event */ |
637 | } | |
e401f176 | 638 | |
13114716 | 639 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
640 | } |
641 | ||
0a6b76dd VD |
642 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
643 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 644 | { |
e231875b | 645 | unsigned long nr = 0; |
889976db YH |
646 | int zid; |
647 | ||
e231875b | 648 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 649 | |
e231875b JZ |
650 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
651 | struct mem_cgroup_per_zone *mz; | |
652 | enum lru_list lru; | |
653 | ||
654 | for_each_lru(lru) { | |
655 | if (!(BIT(lru) & lru_mask)) | |
656 | continue; | |
657 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
658 | nr += mz->lru_size[lru]; | |
659 | } | |
660 | } | |
661 | return nr; | |
889976db | 662 | } |
bb2a0de9 | 663 | |
c0ff4b85 | 664 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 665 | unsigned int lru_mask) |
6d12e2d8 | 666 | { |
e231875b | 667 | unsigned long nr = 0; |
889976db | 668 | int nid; |
6d12e2d8 | 669 | |
31aaea4a | 670 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
671 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
672 | return nr; | |
d52aa412 KH |
673 | } |
674 | ||
f53d7ce3 JW |
675 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
676 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
677 | { |
678 | unsigned long val, next; | |
679 | ||
13114716 | 680 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 681 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 682 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
683 | if ((long)next - (long)val < 0) { |
684 | switch (target) { | |
685 | case MEM_CGROUP_TARGET_THRESH: | |
686 | next = val + THRESHOLDS_EVENTS_TARGET; | |
687 | break; | |
bb4cc1a8 AM |
688 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
689 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
690 | break; | |
f53d7ce3 JW |
691 | case MEM_CGROUP_TARGET_NUMAINFO: |
692 | next = val + NUMAINFO_EVENTS_TARGET; | |
693 | break; | |
694 | default: | |
695 | break; | |
696 | } | |
697 | __this_cpu_write(memcg->stat->targets[target], next); | |
698 | return true; | |
7a159cc9 | 699 | } |
f53d7ce3 | 700 | return false; |
d2265e6f KH |
701 | } |
702 | ||
703 | /* | |
704 | * Check events in order. | |
705 | * | |
706 | */ | |
c0ff4b85 | 707 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
708 | { |
709 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
710 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
711 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 712 | bool do_softlimit; |
82b3f2a7 | 713 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 714 | |
bb4cc1a8 AM |
715 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
716 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
717 | #if MAX_NUMNODES > 1 |
718 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
719 | MEM_CGROUP_TARGET_NUMAINFO); | |
720 | #endif | |
c0ff4b85 | 721 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
722 | if (unlikely(do_softlimit)) |
723 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 724 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 725 | if (unlikely(do_numainfo)) |
c0ff4b85 | 726 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 727 | #endif |
0a31bc97 | 728 | } |
d2265e6f KH |
729 | } |
730 | ||
cf475ad2 | 731 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 732 | { |
31a78f23 BS |
733 | /* |
734 | * mm_update_next_owner() may clear mm->owner to NULL | |
735 | * if it races with swapoff, page migration, etc. | |
736 | * So this can be called with p == NULL. | |
737 | */ | |
738 | if (unlikely(!p)) | |
739 | return NULL; | |
740 | ||
073219e9 | 741 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 742 | } |
33398cf2 | 743 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 744 | |
df381975 | 745 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 746 | { |
c0ff4b85 | 747 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 748 | |
54595fe2 KH |
749 | rcu_read_lock(); |
750 | do { | |
6f6acb00 MH |
751 | /* |
752 | * Page cache insertions can happen withou an | |
753 | * actual mm context, e.g. during disk probing | |
754 | * on boot, loopback IO, acct() writes etc. | |
755 | */ | |
756 | if (unlikely(!mm)) | |
df381975 | 757 | memcg = root_mem_cgroup; |
6f6acb00 MH |
758 | else { |
759 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
760 | if (unlikely(!memcg)) | |
761 | memcg = root_mem_cgroup; | |
762 | } | |
ec903c0c | 763 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 764 | rcu_read_unlock(); |
c0ff4b85 | 765 | return memcg; |
54595fe2 KH |
766 | } |
767 | ||
5660048c JW |
768 | /** |
769 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
770 | * @root: hierarchy root | |
771 | * @prev: previously returned memcg, NULL on first invocation | |
772 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
773 | * | |
774 | * Returns references to children of the hierarchy below @root, or | |
775 | * @root itself, or %NULL after a full round-trip. | |
776 | * | |
777 | * Caller must pass the return value in @prev on subsequent | |
778 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
779 | * to cancel a hierarchy walk before the round-trip is complete. | |
780 | * | |
781 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
782 | * divide up the memcgs in the hierarchy among all concurrent | |
783 | * reclaimers operating on the same zone and priority. | |
784 | */ | |
694fbc0f | 785 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 786 | struct mem_cgroup *prev, |
694fbc0f | 787 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 788 | { |
33398cf2 | 789 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 790 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 791 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 792 | struct mem_cgroup *pos = NULL; |
711d3d2c | 793 | |
694fbc0f AM |
794 | if (mem_cgroup_disabled()) |
795 | return NULL; | |
5660048c | 796 | |
9f3a0d09 JW |
797 | if (!root) |
798 | root = root_mem_cgroup; | |
7d74b06f | 799 | |
9f3a0d09 | 800 | if (prev && !reclaim) |
5ac8fb31 | 801 | pos = prev; |
14067bb3 | 802 | |
9f3a0d09 JW |
803 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
804 | if (prev) | |
5ac8fb31 | 805 | goto out; |
694fbc0f | 806 | return root; |
9f3a0d09 | 807 | } |
14067bb3 | 808 | |
542f85f9 | 809 | rcu_read_lock(); |
5f578161 | 810 | |
5ac8fb31 JW |
811 | if (reclaim) { |
812 | struct mem_cgroup_per_zone *mz; | |
813 | ||
814 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
815 | iter = &mz->iter[reclaim->priority]; | |
816 | ||
817 | if (prev && reclaim->generation != iter->generation) | |
818 | goto out_unlock; | |
819 | ||
6df38689 | 820 | while (1) { |
4db0c3c2 | 821 | pos = READ_ONCE(iter->position); |
6df38689 VD |
822 | if (!pos || css_tryget(&pos->css)) |
823 | break; | |
5ac8fb31 | 824 | /* |
6df38689 VD |
825 | * css reference reached zero, so iter->position will |
826 | * be cleared by ->css_released. However, we should not | |
827 | * rely on this happening soon, because ->css_released | |
828 | * is called from a work queue, and by busy-waiting we | |
829 | * might block it. So we clear iter->position right | |
830 | * away. | |
5ac8fb31 | 831 | */ |
6df38689 VD |
832 | (void)cmpxchg(&iter->position, pos, NULL); |
833 | } | |
5ac8fb31 JW |
834 | } |
835 | ||
836 | if (pos) | |
837 | css = &pos->css; | |
838 | ||
839 | for (;;) { | |
840 | css = css_next_descendant_pre(css, &root->css); | |
841 | if (!css) { | |
842 | /* | |
843 | * Reclaimers share the hierarchy walk, and a | |
844 | * new one might jump in right at the end of | |
845 | * the hierarchy - make sure they see at least | |
846 | * one group and restart from the beginning. | |
847 | */ | |
848 | if (!prev) | |
849 | continue; | |
850 | break; | |
527a5ec9 | 851 | } |
7d74b06f | 852 | |
5ac8fb31 JW |
853 | /* |
854 | * Verify the css and acquire a reference. The root | |
855 | * is provided by the caller, so we know it's alive | |
856 | * and kicking, and don't take an extra reference. | |
857 | */ | |
858 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 859 | |
5ac8fb31 JW |
860 | if (css == &root->css) |
861 | break; | |
14067bb3 | 862 | |
0b8f73e1 JW |
863 | if (css_tryget(css)) |
864 | break; | |
9f3a0d09 | 865 | |
5ac8fb31 | 866 | memcg = NULL; |
9f3a0d09 | 867 | } |
5ac8fb31 JW |
868 | |
869 | if (reclaim) { | |
5ac8fb31 | 870 | /* |
6df38689 VD |
871 | * The position could have already been updated by a competing |
872 | * thread, so check that the value hasn't changed since we read | |
873 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 874 | */ |
6df38689 VD |
875 | (void)cmpxchg(&iter->position, pos, memcg); |
876 | ||
5ac8fb31 JW |
877 | if (pos) |
878 | css_put(&pos->css); | |
879 | ||
880 | if (!memcg) | |
881 | iter->generation++; | |
882 | else if (!prev) | |
883 | reclaim->generation = iter->generation; | |
9f3a0d09 | 884 | } |
5ac8fb31 | 885 | |
542f85f9 MH |
886 | out_unlock: |
887 | rcu_read_unlock(); | |
5ac8fb31 | 888 | out: |
c40046f3 MH |
889 | if (prev && prev != root) |
890 | css_put(&prev->css); | |
891 | ||
9f3a0d09 | 892 | return memcg; |
14067bb3 | 893 | } |
7d74b06f | 894 | |
5660048c JW |
895 | /** |
896 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
897 | * @root: hierarchy root | |
898 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
899 | */ | |
900 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
901 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
902 | { |
903 | if (!root) | |
904 | root = root_mem_cgroup; | |
905 | if (prev && prev != root) | |
906 | css_put(&prev->css); | |
907 | } | |
7d74b06f | 908 | |
6df38689 VD |
909 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
910 | { | |
911 | struct mem_cgroup *memcg = dead_memcg; | |
912 | struct mem_cgroup_reclaim_iter *iter; | |
913 | struct mem_cgroup_per_zone *mz; | |
914 | int nid, zid; | |
915 | int i; | |
916 | ||
917 | while ((memcg = parent_mem_cgroup(memcg))) { | |
918 | for_each_node(nid) { | |
919 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
920 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
921 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
922 | iter = &mz->iter[i]; | |
923 | cmpxchg(&iter->position, | |
924 | dead_memcg, NULL); | |
925 | } | |
926 | } | |
927 | } | |
928 | } | |
929 | } | |
930 | ||
9f3a0d09 JW |
931 | /* |
932 | * Iteration constructs for visiting all cgroups (under a tree). If | |
933 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
934 | * be used for reference counting. | |
935 | */ | |
936 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 937 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 938 | iter != NULL; \ |
527a5ec9 | 939 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 940 | |
9f3a0d09 | 941 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 942 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 943 | iter != NULL; \ |
527a5ec9 | 944 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 945 | |
925b7673 JW |
946 | /** |
947 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
948 | * @zone: zone of the wanted lruvec | |
fa9add64 | 949 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
950 | * |
951 | * Returns the lru list vector holding pages for the given @zone and | |
952 | * @mem. This can be the global zone lruvec, if the memory controller | |
953 | * is disabled. | |
954 | */ | |
955 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
956 | struct mem_cgroup *memcg) | |
957 | { | |
958 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 959 | struct lruvec *lruvec; |
925b7673 | 960 | |
bea8c150 HD |
961 | if (mem_cgroup_disabled()) { |
962 | lruvec = &zone->lruvec; | |
963 | goto out; | |
964 | } | |
925b7673 | 965 | |
e231875b | 966 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
967 | lruvec = &mz->lruvec; |
968 | out: | |
969 | /* | |
970 | * Since a node can be onlined after the mem_cgroup was created, | |
971 | * we have to be prepared to initialize lruvec->zone here; | |
972 | * and if offlined then reonlined, we need to reinitialize it. | |
973 | */ | |
974 | if (unlikely(lruvec->zone != zone)) | |
975 | lruvec->zone = zone; | |
976 | return lruvec; | |
925b7673 JW |
977 | } |
978 | ||
925b7673 | 979 | /** |
dfe0e773 | 980 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 981 | * @page: the page |
fa9add64 | 982 | * @zone: zone of the page |
dfe0e773 JW |
983 | * |
984 | * This function is only safe when following the LRU page isolation | |
985 | * and putback protocol: the LRU lock must be held, and the page must | |
986 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 987 | */ |
fa9add64 | 988 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 989 | { |
08e552c6 | 990 | struct mem_cgroup_per_zone *mz; |
925b7673 | 991 | struct mem_cgroup *memcg; |
bea8c150 | 992 | struct lruvec *lruvec; |
6d12e2d8 | 993 | |
bea8c150 HD |
994 | if (mem_cgroup_disabled()) { |
995 | lruvec = &zone->lruvec; | |
996 | goto out; | |
997 | } | |
925b7673 | 998 | |
1306a85a | 999 | memcg = page->mem_cgroup; |
7512102c | 1000 | /* |
dfe0e773 | 1001 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1002 | * possibly migrated - before they are charged. |
7512102c | 1003 | */ |
29833315 JW |
1004 | if (!memcg) |
1005 | memcg = root_mem_cgroup; | |
7512102c | 1006 | |
e231875b | 1007 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1008 | lruvec = &mz->lruvec; |
1009 | out: | |
1010 | /* | |
1011 | * Since a node can be onlined after the mem_cgroup was created, | |
1012 | * we have to be prepared to initialize lruvec->zone here; | |
1013 | * and if offlined then reonlined, we need to reinitialize it. | |
1014 | */ | |
1015 | if (unlikely(lruvec->zone != zone)) | |
1016 | lruvec->zone = zone; | |
1017 | return lruvec; | |
08e552c6 | 1018 | } |
b69408e8 | 1019 | |
925b7673 | 1020 | /** |
fa9add64 HD |
1021 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1022 | * @lruvec: mem_cgroup per zone lru vector | |
1023 | * @lru: index of lru list the page is sitting on | |
1024 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1025 | * |
ca707239 HD |
1026 | * This function must be called under lru_lock, just before a page is added |
1027 | * to or just after a page is removed from an lru list (that ordering being | |
1028 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1029 | */ |
fa9add64 HD |
1030 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1031 | int nr_pages) | |
3f58a829 MK |
1032 | { |
1033 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1034 | unsigned long *lru_size; |
ca707239 HD |
1035 | long size; |
1036 | bool empty; | |
3f58a829 MK |
1037 | |
1038 | if (mem_cgroup_disabled()) | |
1039 | return; | |
1040 | ||
fa9add64 HD |
1041 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1042 | lru_size = mz->lru_size + lru; | |
ca707239 HD |
1043 | empty = list_empty(lruvec->lists + lru); |
1044 | ||
1045 | if (nr_pages < 0) | |
1046 | *lru_size += nr_pages; | |
1047 | ||
1048 | size = *lru_size; | |
1049 | if (WARN_ONCE(size < 0 || empty != !size, | |
1050 | "%s(%p, %d, %d): lru_size %ld but %sempty\n", | |
1051 | __func__, lruvec, lru, nr_pages, size, empty ? "" : "not ")) { | |
1052 | VM_BUG_ON(1); | |
1053 | *lru_size = 0; | |
1054 | } | |
1055 | ||
1056 | if (nr_pages > 0) | |
1057 | *lru_size += nr_pages; | |
08e552c6 | 1058 | } |
544122e5 | 1059 | |
2314b42d | 1060 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1061 | { |
2314b42d | 1062 | struct mem_cgroup *task_memcg; |
158e0a2d | 1063 | struct task_struct *p; |
ffbdccf5 | 1064 | bool ret; |
4c4a2214 | 1065 | |
158e0a2d | 1066 | p = find_lock_task_mm(task); |
de077d22 | 1067 | if (p) { |
2314b42d | 1068 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1069 | task_unlock(p); |
1070 | } else { | |
1071 | /* | |
1072 | * All threads may have already detached their mm's, but the oom | |
1073 | * killer still needs to detect if they have already been oom | |
1074 | * killed to prevent needlessly killing additional tasks. | |
1075 | */ | |
ffbdccf5 | 1076 | rcu_read_lock(); |
2314b42d JW |
1077 | task_memcg = mem_cgroup_from_task(task); |
1078 | css_get(&task_memcg->css); | |
ffbdccf5 | 1079 | rcu_read_unlock(); |
de077d22 | 1080 | } |
2314b42d JW |
1081 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1082 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1083 | return ret; |
1084 | } | |
1085 | ||
19942822 | 1086 | /** |
9d11ea9f | 1087 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1088 | * @memcg: the memory cgroup |
19942822 | 1089 | * |
9d11ea9f | 1090 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1091 | * pages. |
19942822 | 1092 | */ |
c0ff4b85 | 1093 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1094 | { |
3e32cb2e JW |
1095 | unsigned long margin = 0; |
1096 | unsigned long count; | |
1097 | unsigned long limit; | |
9d11ea9f | 1098 | |
3e32cb2e | 1099 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1100 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1101 | if (count < limit) |
1102 | margin = limit - count; | |
1103 | ||
7941d214 | 1104 | if (do_memsw_account()) { |
3e32cb2e | 1105 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1106 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1107 | if (count <= limit) |
1108 | margin = min(margin, limit - count); | |
1109 | } | |
1110 | ||
1111 | return margin; | |
19942822 JW |
1112 | } |
1113 | ||
32047e2a | 1114 | /* |
bdcbb659 | 1115 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1116 | * |
bdcbb659 QH |
1117 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1118 | * moving cgroups. This is for waiting at high-memory pressure | |
1119 | * caused by "move". | |
32047e2a | 1120 | */ |
c0ff4b85 | 1121 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1122 | { |
2bd9bb20 KH |
1123 | struct mem_cgroup *from; |
1124 | struct mem_cgroup *to; | |
4b534334 | 1125 | bool ret = false; |
2bd9bb20 KH |
1126 | /* |
1127 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1128 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1129 | */ | |
1130 | spin_lock(&mc.lock); | |
1131 | from = mc.from; | |
1132 | to = mc.to; | |
1133 | if (!from) | |
1134 | goto unlock; | |
3e92041d | 1135 | |
2314b42d JW |
1136 | ret = mem_cgroup_is_descendant(from, memcg) || |
1137 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1138 | unlock: |
1139 | spin_unlock(&mc.lock); | |
4b534334 KH |
1140 | return ret; |
1141 | } | |
1142 | ||
c0ff4b85 | 1143 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1144 | { |
1145 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1146 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1147 | DEFINE_WAIT(wait); |
1148 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1149 | /* moving charge context might have finished. */ | |
1150 | if (mc.moving_task) | |
1151 | schedule(); | |
1152 | finish_wait(&mc.waitq, &wait); | |
1153 | return true; | |
1154 | } | |
1155 | } | |
1156 | return false; | |
1157 | } | |
1158 | ||
58cf188e | 1159 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1160 | /** |
58cf188e | 1161 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1162 | * @memcg: The memory cgroup that went over limit |
1163 | * @p: Task that is going to be killed | |
1164 | * | |
1165 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1166 | * enabled | |
1167 | */ | |
1168 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1169 | { | |
58cf188e SZ |
1170 | struct mem_cgroup *iter; |
1171 | unsigned int i; | |
e222432b | 1172 | |
e222432b BS |
1173 | rcu_read_lock(); |
1174 | ||
2415b9f5 BV |
1175 | if (p) { |
1176 | pr_info("Task in "); | |
1177 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1178 | pr_cont(" killed as a result of limit of "); | |
1179 | } else { | |
1180 | pr_info("Memory limit reached of cgroup "); | |
1181 | } | |
1182 | ||
e61734c5 | 1183 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1184 | pr_cont("\n"); |
e222432b | 1185 | |
e222432b BS |
1186 | rcu_read_unlock(); |
1187 | ||
3e32cb2e JW |
1188 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1189 | K((u64)page_counter_read(&memcg->memory)), | |
1190 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1191 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1192 | K((u64)page_counter_read(&memcg->memsw)), | |
1193 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1194 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1195 | K((u64)page_counter_read(&memcg->kmem)), | |
1196 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1197 | |
1198 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1199 | pr_info("Memory cgroup stats for "); |
1200 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1201 | pr_cont(":"); |
1202 | ||
1203 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
37e84351 | 1204 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
58cf188e | 1205 | continue; |
484ebb3b | 1206 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1207 | K(mem_cgroup_read_stat(iter, i))); |
1208 | } | |
1209 | ||
1210 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1211 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1212 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1213 | ||
1214 | pr_cont("\n"); | |
1215 | } | |
e222432b BS |
1216 | } |
1217 | ||
81d39c20 KH |
1218 | /* |
1219 | * This function returns the number of memcg under hierarchy tree. Returns | |
1220 | * 1(self count) if no children. | |
1221 | */ | |
c0ff4b85 | 1222 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1223 | { |
1224 | int num = 0; | |
7d74b06f KH |
1225 | struct mem_cgroup *iter; |
1226 | ||
c0ff4b85 | 1227 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1228 | num++; |
81d39c20 KH |
1229 | return num; |
1230 | } | |
1231 | ||
a63d83f4 DR |
1232 | /* |
1233 | * Return the memory (and swap, if configured) limit for a memcg. | |
1234 | */ | |
3e32cb2e | 1235 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1236 | { |
3e32cb2e | 1237 | unsigned long limit; |
f3e8eb70 | 1238 | |
3e32cb2e | 1239 | limit = memcg->memory.limit; |
9a5a8f19 | 1240 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1241 | unsigned long memsw_limit; |
37e84351 | 1242 | unsigned long swap_limit; |
9a5a8f19 | 1243 | |
3e32cb2e | 1244 | memsw_limit = memcg->memsw.limit; |
37e84351 VD |
1245 | swap_limit = memcg->swap.limit; |
1246 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); | |
1247 | limit = min(limit + swap_limit, memsw_limit); | |
9a5a8f19 | 1248 | } |
9a5a8f19 | 1249 | return limit; |
a63d83f4 DR |
1250 | } |
1251 | ||
b6e6edcf | 1252 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1253 | int order) |
9cbb78bb | 1254 | { |
6e0fc46d DR |
1255 | struct oom_control oc = { |
1256 | .zonelist = NULL, | |
1257 | .nodemask = NULL, | |
1258 | .gfp_mask = gfp_mask, | |
1259 | .order = order, | |
6e0fc46d | 1260 | }; |
9cbb78bb DR |
1261 | struct mem_cgroup *iter; |
1262 | unsigned long chosen_points = 0; | |
1263 | unsigned long totalpages; | |
1264 | unsigned int points = 0; | |
1265 | struct task_struct *chosen = NULL; | |
1266 | ||
dc56401f JW |
1267 | mutex_lock(&oom_lock); |
1268 | ||
876aafbf | 1269 | /* |
465adcf1 DR |
1270 | * If current has a pending SIGKILL or is exiting, then automatically |
1271 | * select it. The goal is to allow it to allocate so that it may | |
1272 | * quickly exit and free its memory. | |
876aafbf | 1273 | */ |
d003f371 | 1274 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1275 | mark_oom_victim(current); |
dc56401f | 1276 | goto unlock; |
876aafbf DR |
1277 | } |
1278 | ||
6e0fc46d | 1279 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1280 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1281 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1282 | struct css_task_iter it; |
9cbb78bb DR |
1283 | struct task_struct *task; |
1284 | ||
72ec7029 TH |
1285 | css_task_iter_start(&iter->css, &it); |
1286 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1287 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1288 | case OOM_SCAN_SELECT: |
1289 | if (chosen) | |
1290 | put_task_struct(chosen); | |
1291 | chosen = task; | |
1292 | chosen_points = ULONG_MAX; | |
1293 | get_task_struct(chosen); | |
1294 | /* fall through */ | |
1295 | case OOM_SCAN_CONTINUE: | |
1296 | continue; | |
1297 | case OOM_SCAN_ABORT: | |
72ec7029 | 1298 | css_task_iter_end(&it); |
9cbb78bb DR |
1299 | mem_cgroup_iter_break(memcg, iter); |
1300 | if (chosen) | |
1301 | put_task_struct(chosen); | |
dc56401f | 1302 | goto unlock; |
9cbb78bb DR |
1303 | case OOM_SCAN_OK: |
1304 | break; | |
1305 | }; | |
1306 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1307 | if (!points || points < chosen_points) |
1308 | continue; | |
1309 | /* Prefer thread group leaders for display purposes */ | |
1310 | if (points == chosen_points && | |
1311 | thread_group_leader(chosen)) | |
1312 | continue; | |
1313 | ||
1314 | if (chosen) | |
1315 | put_task_struct(chosen); | |
1316 | chosen = task; | |
1317 | chosen_points = points; | |
1318 | get_task_struct(chosen); | |
9cbb78bb | 1319 | } |
72ec7029 | 1320 | css_task_iter_end(&it); |
9cbb78bb DR |
1321 | } |
1322 | ||
dc56401f JW |
1323 | if (chosen) { |
1324 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1325 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1326 | "Memory cgroup out of memory"); | |
dc56401f JW |
1327 | } |
1328 | unlock: | |
1329 | mutex_unlock(&oom_lock); | |
b6e6edcf | 1330 | return chosen; |
9cbb78bb DR |
1331 | } |
1332 | ||
ae6e71d3 MC |
1333 | #if MAX_NUMNODES > 1 |
1334 | ||
4d0c066d KH |
1335 | /** |
1336 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1337 | * @memcg: the target memcg |
4d0c066d KH |
1338 | * @nid: the node ID to be checked. |
1339 | * @noswap : specify true here if the user wants flle only information. | |
1340 | * | |
1341 | * This function returns whether the specified memcg contains any | |
1342 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1343 | * pages in the node. | |
1344 | */ | |
c0ff4b85 | 1345 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1346 | int nid, bool noswap) |
1347 | { | |
c0ff4b85 | 1348 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1349 | return true; |
1350 | if (noswap || !total_swap_pages) | |
1351 | return false; | |
c0ff4b85 | 1352 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1353 | return true; |
1354 | return false; | |
1355 | ||
1356 | } | |
889976db YH |
1357 | |
1358 | /* | |
1359 | * Always updating the nodemask is not very good - even if we have an empty | |
1360 | * list or the wrong list here, we can start from some node and traverse all | |
1361 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1362 | * | |
1363 | */ | |
c0ff4b85 | 1364 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1365 | { |
1366 | int nid; | |
453a9bf3 KH |
1367 | /* |
1368 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1369 | * pagein/pageout changes since the last update. | |
1370 | */ | |
c0ff4b85 | 1371 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1372 | return; |
c0ff4b85 | 1373 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1374 | return; |
1375 | ||
889976db | 1376 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1377 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1378 | |
31aaea4a | 1379 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1380 | |
c0ff4b85 R |
1381 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1382 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1383 | } |
453a9bf3 | 1384 | |
c0ff4b85 R |
1385 | atomic_set(&memcg->numainfo_events, 0); |
1386 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1387 | } |
1388 | ||
1389 | /* | |
1390 | * Selecting a node where we start reclaim from. Because what we need is just | |
1391 | * reducing usage counter, start from anywhere is O,K. Considering | |
1392 | * memory reclaim from current node, there are pros. and cons. | |
1393 | * | |
1394 | * Freeing memory from current node means freeing memory from a node which | |
1395 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1396 | * hit limits, it will see a contention on a node. But freeing from remote | |
1397 | * node means more costs for memory reclaim because of memory latency. | |
1398 | * | |
1399 | * Now, we use round-robin. Better algorithm is welcomed. | |
1400 | */ | |
c0ff4b85 | 1401 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1402 | { |
1403 | int node; | |
1404 | ||
c0ff4b85 R |
1405 | mem_cgroup_may_update_nodemask(memcg); |
1406 | node = memcg->last_scanned_node; | |
889976db | 1407 | |
0edaf86c | 1408 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1409 | /* |
fda3d69b MH |
1410 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1411 | * last time it really checked all the LRUs due to rate limiting. | |
1412 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1413 | */ |
1414 | if (unlikely(node == MAX_NUMNODES)) | |
1415 | node = numa_node_id(); | |
1416 | ||
c0ff4b85 | 1417 | memcg->last_scanned_node = node; |
889976db YH |
1418 | return node; |
1419 | } | |
889976db | 1420 | #else |
c0ff4b85 | 1421 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1422 | { |
1423 | return 0; | |
1424 | } | |
1425 | #endif | |
1426 | ||
0608f43d AM |
1427 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1428 | struct zone *zone, | |
1429 | gfp_t gfp_mask, | |
1430 | unsigned long *total_scanned) | |
1431 | { | |
1432 | struct mem_cgroup *victim = NULL; | |
1433 | int total = 0; | |
1434 | int loop = 0; | |
1435 | unsigned long excess; | |
1436 | unsigned long nr_scanned; | |
1437 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1438 | .zone = zone, | |
1439 | .priority = 0, | |
1440 | }; | |
1441 | ||
3e32cb2e | 1442 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1443 | |
1444 | while (1) { | |
1445 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1446 | if (!victim) { | |
1447 | loop++; | |
1448 | if (loop >= 2) { | |
1449 | /* | |
1450 | * If we have not been able to reclaim | |
1451 | * anything, it might because there are | |
1452 | * no reclaimable pages under this hierarchy | |
1453 | */ | |
1454 | if (!total) | |
1455 | break; | |
1456 | /* | |
1457 | * We want to do more targeted reclaim. | |
1458 | * excess >> 2 is not to excessive so as to | |
1459 | * reclaim too much, nor too less that we keep | |
1460 | * coming back to reclaim from this cgroup | |
1461 | */ | |
1462 | if (total >= (excess >> 2) || | |
1463 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1464 | break; | |
1465 | } | |
1466 | continue; | |
1467 | } | |
0608f43d AM |
1468 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1469 | zone, &nr_scanned); | |
1470 | *total_scanned += nr_scanned; | |
3e32cb2e | 1471 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1472 | break; |
6d61ef40 | 1473 | } |
0608f43d AM |
1474 | mem_cgroup_iter_break(root_memcg, victim); |
1475 | return total; | |
6d61ef40 BS |
1476 | } |
1477 | ||
0056f4e6 JW |
1478 | #ifdef CONFIG_LOCKDEP |
1479 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1480 | .name = "memcg_oom_lock", | |
1481 | }; | |
1482 | #endif | |
1483 | ||
fb2a6fc5 JW |
1484 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1485 | ||
867578cb KH |
1486 | /* |
1487 | * Check OOM-Killer is already running under our hierarchy. | |
1488 | * If someone is running, return false. | |
1489 | */ | |
fb2a6fc5 | 1490 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1491 | { |
79dfdacc | 1492 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1493 | |
fb2a6fc5 JW |
1494 | spin_lock(&memcg_oom_lock); |
1495 | ||
9f3a0d09 | 1496 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1497 | if (iter->oom_lock) { |
79dfdacc MH |
1498 | /* |
1499 | * this subtree of our hierarchy is already locked | |
1500 | * so we cannot give a lock. | |
1501 | */ | |
79dfdacc | 1502 | failed = iter; |
9f3a0d09 JW |
1503 | mem_cgroup_iter_break(memcg, iter); |
1504 | break; | |
23751be0 JW |
1505 | } else |
1506 | iter->oom_lock = true; | |
7d74b06f | 1507 | } |
867578cb | 1508 | |
fb2a6fc5 JW |
1509 | if (failed) { |
1510 | /* | |
1511 | * OK, we failed to lock the whole subtree so we have | |
1512 | * to clean up what we set up to the failing subtree | |
1513 | */ | |
1514 | for_each_mem_cgroup_tree(iter, memcg) { | |
1515 | if (iter == failed) { | |
1516 | mem_cgroup_iter_break(memcg, iter); | |
1517 | break; | |
1518 | } | |
1519 | iter->oom_lock = false; | |
79dfdacc | 1520 | } |
0056f4e6 JW |
1521 | } else |
1522 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1523 | |
1524 | spin_unlock(&memcg_oom_lock); | |
1525 | ||
1526 | return !failed; | |
a636b327 | 1527 | } |
0b7f569e | 1528 | |
fb2a6fc5 | 1529 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1530 | { |
7d74b06f KH |
1531 | struct mem_cgroup *iter; |
1532 | ||
fb2a6fc5 | 1533 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1534 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1535 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1536 | iter->oom_lock = false; |
fb2a6fc5 | 1537 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1538 | } |
1539 | ||
c0ff4b85 | 1540 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1541 | { |
1542 | struct mem_cgroup *iter; | |
1543 | ||
c2b42d3c | 1544 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1545 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1546 | iter->under_oom++; |
1547 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1548 | } |
1549 | ||
c0ff4b85 | 1550 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1551 | { |
1552 | struct mem_cgroup *iter; | |
1553 | ||
867578cb KH |
1554 | /* |
1555 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1556 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1557 | */ |
c2b42d3c | 1558 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1559 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1560 | if (iter->under_oom > 0) |
1561 | iter->under_oom--; | |
1562 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1563 | } |
1564 | ||
867578cb KH |
1565 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1566 | ||
dc98df5a | 1567 | struct oom_wait_info { |
d79154bb | 1568 | struct mem_cgroup *memcg; |
dc98df5a KH |
1569 | wait_queue_t wait; |
1570 | }; | |
1571 | ||
1572 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1573 | unsigned mode, int sync, void *arg) | |
1574 | { | |
d79154bb HD |
1575 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1576 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1577 | struct oom_wait_info *oom_wait_info; |
1578 | ||
1579 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1580 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1581 | |
2314b42d JW |
1582 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1583 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1584 | return 0; |
dc98df5a KH |
1585 | return autoremove_wake_function(wait, mode, sync, arg); |
1586 | } | |
1587 | ||
c0ff4b85 | 1588 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1589 | { |
c2b42d3c TH |
1590 | /* |
1591 | * For the following lockless ->under_oom test, the only required | |
1592 | * guarantee is that it must see the state asserted by an OOM when | |
1593 | * this function is called as a result of userland actions | |
1594 | * triggered by the notification of the OOM. This is trivially | |
1595 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1596 | * triggering notification. | |
1597 | */ | |
1598 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1599 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1600 | } |
1601 | ||
3812c8c8 | 1602 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1603 | { |
626ebc41 | 1604 | if (!current->memcg_may_oom) |
3812c8c8 | 1605 | return; |
867578cb | 1606 | /* |
49426420 JW |
1607 | * We are in the middle of the charge context here, so we |
1608 | * don't want to block when potentially sitting on a callstack | |
1609 | * that holds all kinds of filesystem and mm locks. | |
1610 | * | |
1611 | * Also, the caller may handle a failed allocation gracefully | |
1612 | * (like optional page cache readahead) and so an OOM killer | |
1613 | * invocation might not even be necessary. | |
1614 | * | |
1615 | * That's why we don't do anything here except remember the | |
1616 | * OOM context and then deal with it at the end of the page | |
1617 | * fault when the stack is unwound, the locks are released, | |
1618 | * and when we know whether the fault was overall successful. | |
867578cb | 1619 | */ |
49426420 | 1620 | css_get(&memcg->css); |
626ebc41 TH |
1621 | current->memcg_in_oom = memcg; |
1622 | current->memcg_oom_gfp_mask = mask; | |
1623 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1624 | } |
1625 | ||
1626 | /** | |
1627 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1628 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1629 | * |
49426420 JW |
1630 | * This has to be called at the end of a page fault if the memcg OOM |
1631 | * handler was enabled. | |
3812c8c8 | 1632 | * |
49426420 | 1633 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1634 | * sleep on a waitqueue until the userspace task resolves the |
1635 | * situation. Sleeping directly in the charge context with all kinds | |
1636 | * of locks held is not a good idea, instead we remember an OOM state | |
1637 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1638 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1639 | * |
1640 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1641 | * completed, %false otherwise. |
3812c8c8 | 1642 | */ |
49426420 | 1643 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1644 | { |
626ebc41 | 1645 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1646 | struct oom_wait_info owait; |
49426420 | 1647 | bool locked; |
3812c8c8 JW |
1648 | |
1649 | /* OOM is global, do not handle */ | |
3812c8c8 | 1650 | if (!memcg) |
49426420 | 1651 | return false; |
3812c8c8 | 1652 | |
c32b3cbe | 1653 | if (!handle || oom_killer_disabled) |
49426420 | 1654 | goto cleanup; |
3812c8c8 JW |
1655 | |
1656 | owait.memcg = memcg; | |
1657 | owait.wait.flags = 0; | |
1658 | owait.wait.func = memcg_oom_wake_function; | |
1659 | owait.wait.private = current; | |
1660 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1661 | |
3812c8c8 | 1662 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1663 | mem_cgroup_mark_under_oom(memcg); |
1664 | ||
1665 | locked = mem_cgroup_oom_trylock(memcg); | |
1666 | ||
1667 | if (locked) | |
1668 | mem_cgroup_oom_notify(memcg); | |
1669 | ||
1670 | if (locked && !memcg->oom_kill_disable) { | |
1671 | mem_cgroup_unmark_under_oom(memcg); | |
1672 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1673 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1674 | current->memcg_oom_order); | |
49426420 | 1675 | } else { |
3812c8c8 | 1676 | schedule(); |
49426420 JW |
1677 | mem_cgroup_unmark_under_oom(memcg); |
1678 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1679 | } | |
1680 | ||
1681 | if (locked) { | |
fb2a6fc5 JW |
1682 | mem_cgroup_oom_unlock(memcg); |
1683 | /* | |
1684 | * There is no guarantee that an OOM-lock contender | |
1685 | * sees the wakeups triggered by the OOM kill | |
1686 | * uncharges. Wake any sleepers explicitely. | |
1687 | */ | |
1688 | memcg_oom_recover(memcg); | |
1689 | } | |
49426420 | 1690 | cleanup: |
626ebc41 | 1691 | current->memcg_in_oom = NULL; |
3812c8c8 | 1692 | css_put(&memcg->css); |
867578cb | 1693 | return true; |
0b7f569e KH |
1694 | } |
1695 | ||
d7365e78 | 1696 | /** |
81f8c3a4 JW |
1697 | * lock_page_memcg - lock a page->mem_cgroup binding |
1698 | * @page: the page | |
32047e2a | 1699 | * |
81f8c3a4 JW |
1700 | * This function protects unlocked LRU pages from being moved to |
1701 | * another cgroup and stabilizes their page->mem_cgroup binding. | |
d69b042f | 1702 | */ |
62cccb8c | 1703 | void lock_page_memcg(struct page *page) |
89c06bd5 KH |
1704 | { |
1705 | struct mem_cgroup *memcg; | |
6de22619 | 1706 | unsigned long flags; |
89c06bd5 | 1707 | |
6de22619 JW |
1708 | /* |
1709 | * The RCU lock is held throughout the transaction. The fast | |
1710 | * path can get away without acquiring the memcg->move_lock | |
1711 | * because page moving starts with an RCU grace period. | |
6de22619 | 1712 | */ |
d7365e78 JW |
1713 | rcu_read_lock(); |
1714 | ||
1715 | if (mem_cgroup_disabled()) | |
62cccb8c | 1716 | return; |
89c06bd5 | 1717 | again: |
1306a85a | 1718 | memcg = page->mem_cgroup; |
29833315 | 1719 | if (unlikely(!memcg)) |
62cccb8c | 1720 | return; |
d7365e78 | 1721 | |
bdcbb659 | 1722 | if (atomic_read(&memcg->moving_account) <= 0) |
62cccb8c | 1723 | return; |
89c06bd5 | 1724 | |
6de22619 | 1725 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1726 | if (memcg != page->mem_cgroup) { |
6de22619 | 1727 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1728 | goto again; |
1729 | } | |
6de22619 JW |
1730 | |
1731 | /* | |
1732 | * When charge migration first begins, we can have locked and | |
1733 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1734 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1735 | */ |
1736 | memcg->move_lock_task = current; | |
1737 | memcg->move_lock_flags = flags; | |
d7365e78 | 1738 | |
62cccb8c | 1739 | return; |
89c06bd5 | 1740 | } |
81f8c3a4 | 1741 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1742 | |
d7365e78 | 1743 | /** |
81f8c3a4 | 1744 | * unlock_page_memcg - unlock a page->mem_cgroup binding |
62cccb8c | 1745 | * @page: the page |
d7365e78 | 1746 | */ |
62cccb8c | 1747 | void unlock_page_memcg(struct page *page) |
89c06bd5 | 1748 | { |
62cccb8c JW |
1749 | struct mem_cgroup *memcg = page->mem_cgroup; |
1750 | ||
6de22619 JW |
1751 | if (memcg && memcg->move_lock_task == current) { |
1752 | unsigned long flags = memcg->move_lock_flags; | |
1753 | ||
1754 | memcg->move_lock_task = NULL; | |
1755 | memcg->move_lock_flags = 0; | |
1756 | ||
1757 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1758 | } | |
89c06bd5 | 1759 | |
d7365e78 | 1760 | rcu_read_unlock(); |
89c06bd5 | 1761 | } |
81f8c3a4 | 1762 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1763 | |
cdec2e42 KH |
1764 | /* |
1765 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1766 | * TODO: maybe necessary to use big numbers in big irons. | |
1767 | */ | |
7ec99d62 | 1768 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1769 | struct memcg_stock_pcp { |
1770 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1771 | unsigned int nr_pages; |
cdec2e42 | 1772 | struct work_struct work; |
26fe6168 | 1773 | unsigned long flags; |
a0db00fc | 1774 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1775 | }; |
1776 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1777 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1778 | |
a0956d54 SS |
1779 | /** |
1780 | * consume_stock: Try to consume stocked charge on this cpu. | |
1781 | * @memcg: memcg to consume from. | |
1782 | * @nr_pages: how many pages to charge. | |
1783 | * | |
1784 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1785 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1786 | * service an allocation will refill the stock. | |
1787 | * | |
1788 | * returns true if successful, false otherwise. | |
cdec2e42 | 1789 | */ |
a0956d54 | 1790 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1791 | { |
1792 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1793 | bool ret = false; |
cdec2e42 | 1794 | |
a0956d54 | 1795 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1796 | return ret; |
a0956d54 | 1797 | |
cdec2e42 | 1798 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1799 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1800 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1801 | ret = true; |
1802 | } | |
cdec2e42 KH |
1803 | put_cpu_var(memcg_stock); |
1804 | return ret; | |
1805 | } | |
1806 | ||
1807 | /* | |
3e32cb2e | 1808 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1809 | */ |
1810 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1811 | { | |
1812 | struct mem_cgroup *old = stock->cached; | |
1813 | ||
11c9ea4e | 1814 | if (stock->nr_pages) { |
3e32cb2e | 1815 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1816 | if (do_memsw_account()) |
3e32cb2e | 1817 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1818 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1819 | stock->nr_pages = 0; |
cdec2e42 KH |
1820 | } |
1821 | stock->cached = NULL; | |
cdec2e42 KH |
1822 | } |
1823 | ||
1824 | /* | |
1825 | * This must be called under preempt disabled or must be called by | |
1826 | * a thread which is pinned to local cpu. | |
1827 | */ | |
1828 | static void drain_local_stock(struct work_struct *dummy) | |
1829 | { | |
7c8e0181 | 1830 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1831 | drain_stock(stock); |
26fe6168 | 1832 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1833 | } |
1834 | ||
1835 | /* | |
3e32cb2e | 1836 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1837 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1838 | */ |
c0ff4b85 | 1839 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1840 | { |
1841 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1842 | ||
c0ff4b85 | 1843 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1844 | drain_stock(stock); |
c0ff4b85 | 1845 | stock->cached = memcg; |
cdec2e42 | 1846 | } |
11c9ea4e | 1847 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1848 | put_cpu_var(memcg_stock); |
1849 | } | |
1850 | ||
1851 | /* | |
c0ff4b85 | 1852 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1853 | * of the hierarchy under it. |
cdec2e42 | 1854 | */ |
6d3d6aa2 | 1855 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1856 | { |
26fe6168 | 1857 | int cpu, curcpu; |
d38144b7 | 1858 | |
6d3d6aa2 JW |
1859 | /* If someone's already draining, avoid adding running more workers. */ |
1860 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1861 | return; | |
cdec2e42 | 1862 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1863 | get_online_cpus(); |
5af12d0e | 1864 | curcpu = get_cpu(); |
cdec2e42 KH |
1865 | for_each_online_cpu(cpu) { |
1866 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1867 | struct mem_cgroup *memcg; |
26fe6168 | 1868 | |
c0ff4b85 R |
1869 | memcg = stock->cached; |
1870 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1871 | continue; |
2314b42d | 1872 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1873 | continue; |
d1a05b69 MH |
1874 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1875 | if (cpu == curcpu) | |
1876 | drain_local_stock(&stock->work); | |
1877 | else | |
1878 | schedule_work_on(cpu, &stock->work); | |
1879 | } | |
cdec2e42 | 1880 | } |
5af12d0e | 1881 | put_cpu(); |
f894ffa8 | 1882 | put_online_cpus(); |
9f50fad6 | 1883 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1884 | } |
1885 | ||
0db0628d | 1886 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1887 | unsigned long action, |
1888 | void *hcpu) | |
1889 | { | |
1890 | int cpu = (unsigned long)hcpu; | |
1891 | struct memcg_stock_pcp *stock; | |
1892 | ||
619d094b | 1893 | if (action == CPU_ONLINE) |
1489ebad | 1894 | return NOTIFY_OK; |
1489ebad | 1895 | |
d833049b | 1896 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1897 | return NOTIFY_OK; |
711d3d2c | 1898 | |
cdec2e42 KH |
1899 | stock = &per_cpu(memcg_stock, cpu); |
1900 | drain_stock(stock); | |
1901 | return NOTIFY_OK; | |
1902 | } | |
1903 | ||
f7e1cb6e JW |
1904 | static void reclaim_high(struct mem_cgroup *memcg, |
1905 | unsigned int nr_pages, | |
1906 | gfp_t gfp_mask) | |
1907 | { | |
1908 | do { | |
1909 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1910 | continue; | |
1911 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
1912 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
1913 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1914 | } | |
1915 | ||
1916 | static void high_work_func(struct work_struct *work) | |
1917 | { | |
1918 | struct mem_cgroup *memcg; | |
1919 | ||
1920 | memcg = container_of(work, struct mem_cgroup, high_work); | |
1921 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); | |
1922 | } | |
1923 | ||
b23afb93 TH |
1924 | /* |
1925 | * Scheduled by try_charge() to be executed from the userland return path | |
1926 | * and reclaims memory over the high limit. | |
1927 | */ | |
1928 | void mem_cgroup_handle_over_high(void) | |
1929 | { | |
1930 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1931 | struct mem_cgroup *memcg; |
b23afb93 TH |
1932 | |
1933 | if (likely(!nr_pages)) | |
1934 | return; | |
1935 | ||
f7e1cb6e JW |
1936 | memcg = get_mem_cgroup_from_mm(current->mm); |
1937 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1938 | css_put(&memcg->css); |
1939 | current->memcg_nr_pages_over_high = 0; | |
1940 | } | |
1941 | ||
00501b53 JW |
1942 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1943 | unsigned int nr_pages) | |
8a9f3ccd | 1944 | { |
7ec99d62 | 1945 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1946 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1947 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1948 | struct page_counter *counter; |
6539cc05 | 1949 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1950 | bool may_swap = true; |
1951 | bool drained = false; | |
a636b327 | 1952 | |
ce00a967 | 1953 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1954 | return 0; |
6539cc05 | 1955 | retry: |
b6b6cc72 | 1956 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1957 | return 0; |
8a9f3ccd | 1958 | |
7941d214 | 1959 | if (!do_memsw_account() || |
6071ca52 JW |
1960 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1961 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1962 | goto done_restock; |
7941d214 | 1963 | if (do_memsw_account()) |
3e32cb2e JW |
1964 | page_counter_uncharge(&memcg->memsw, batch); |
1965 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1966 | } else { |
3e32cb2e | 1967 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1968 | may_swap = false; |
3fbe7244 | 1969 | } |
7a81b88c | 1970 | |
6539cc05 JW |
1971 | if (batch > nr_pages) { |
1972 | batch = nr_pages; | |
1973 | goto retry; | |
1974 | } | |
6d61ef40 | 1975 | |
06b078fc JW |
1976 | /* |
1977 | * Unlike in global OOM situations, memcg is not in a physical | |
1978 | * memory shortage. Allow dying and OOM-killed tasks to | |
1979 | * bypass the last charges so that they can exit quickly and | |
1980 | * free their memory. | |
1981 | */ | |
1982 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
1983 | fatal_signal_pending(current) || | |
1984 | current->flags & PF_EXITING)) | |
10d53c74 | 1985 | goto force; |
06b078fc JW |
1986 | |
1987 | if (unlikely(task_in_memcg_oom(current))) | |
1988 | goto nomem; | |
1989 | ||
d0164adc | 1990 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 1991 | goto nomem; |
4b534334 | 1992 | |
241994ed JW |
1993 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
1994 | ||
b70a2a21 JW |
1995 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
1996 | gfp_mask, may_swap); | |
6539cc05 | 1997 | |
61e02c74 | 1998 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 1999 | goto retry; |
28c34c29 | 2000 | |
b70a2a21 | 2001 | if (!drained) { |
6d3d6aa2 | 2002 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2003 | drained = true; |
2004 | goto retry; | |
2005 | } | |
2006 | ||
28c34c29 JW |
2007 | if (gfp_mask & __GFP_NORETRY) |
2008 | goto nomem; | |
6539cc05 JW |
2009 | /* |
2010 | * Even though the limit is exceeded at this point, reclaim | |
2011 | * may have been able to free some pages. Retry the charge | |
2012 | * before killing the task. | |
2013 | * | |
2014 | * Only for regular pages, though: huge pages are rather | |
2015 | * unlikely to succeed so close to the limit, and we fall back | |
2016 | * to regular pages anyway in case of failure. | |
2017 | */ | |
61e02c74 | 2018 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2019 | goto retry; |
2020 | /* | |
2021 | * At task move, charge accounts can be doubly counted. So, it's | |
2022 | * better to wait until the end of task_move if something is going on. | |
2023 | */ | |
2024 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2025 | goto retry; | |
2026 | ||
9b130619 JW |
2027 | if (nr_retries--) |
2028 | goto retry; | |
2029 | ||
06b078fc | 2030 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2031 | goto force; |
06b078fc | 2032 | |
6539cc05 | 2033 | if (fatal_signal_pending(current)) |
10d53c74 | 2034 | goto force; |
6539cc05 | 2035 | |
241994ed JW |
2036 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2037 | ||
3608de07 JM |
2038 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2039 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2040 | nomem: |
6d1fdc48 | 2041 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2042 | return -ENOMEM; |
10d53c74 TH |
2043 | force: |
2044 | /* | |
2045 | * The allocation either can't fail or will lead to more memory | |
2046 | * being freed very soon. Allow memory usage go over the limit | |
2047 | * temporarily by force charging it. | |
2048 | */ | |
2049 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2050 | if (do_memsw_account()) |
10d53c74 TH |
2051 | page_counter_charge(&memcg->memsw, nr_pages); |
2052 | css_get_many(&memcg->css, nr_pages); | |
2053 | ||
2054 | return 0; | |
6539cc05 JW |
2055 | |
2056 | done_restock: | |
e8ea14cc | 2057 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2058 | if (batch > nr_pages) |
2059 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2060 | |
241994ed | 2061 | /* |
b23afb93 TH |
2062 | * If the hierarchy is above the normal consumption range, schedule |
2063 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2064 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2065 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2066 | * not recorded as it most likely matches current's and won't | |
2067 | * change in the meantime. As high limit is checked again before | |
2068 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2069 | */ |
2070 | do { | |
b23afb93 | 2071 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2072 | /* Don't bother a random interrupted task */ |
2073 | if (in_interrupt()) { | |
2074 | schedule_work(&memcg->high_work); | |
2075 | break; | |
2076 | } | |
9516a18a | 2077 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2078 | set_notify_resume(current); |
2079 | break; | |
2080 | } | |
241994ed | 2081 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2082 | |
2083 | return 0; | |
7a81b88c | 2084 | } |
8a9f3ccd | 2085 | |
00501b53 | 2086 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2087 | { |
ce00a967 JW |
2088 | if (mem_cgroup_is_root(memcg)) |
2089 | return; | |
2090 | ||
3e32cb2e | 2091 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2092 | if (do_memsw_account()) |
3e32cb2e | 2093 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2094 | |
e8ea14cc | 2095 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2096 | } |
2097 | ||
0a31bc97 JW |
2098 | static void lock_page_lru(struct page *page, int *isolated) |
2099 | { | |
2100 | struct zone *zone = page_zone(page); | |
2101 | ||
2102 | spin_lock_irq(&zone->lru_lock); | |
2103 | if (PageLRU(page)) { | |
2104 | struct lruvec *lruvec; | |
2105 | ||
2106 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2107 | ClearPageLRU(page); | |
2108 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2109 | *isolated = 1; | |
2110 | } else | |
2111 | *isolated = 0; | |
2112 | } | |
2113 | ||
2114 | static void unlock_page_lru(struct page *page, int isolated) | |
2115 | { | |
2116 | struct zone *zone = page_zone(page); | |
2117 | ||
2118 | if (isolated) { | |
2119 | struct lruvec *lruvec; | |
2120 | ||
2121 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2122 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2123 | SetPageLRU(page); | |
2124 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2125 | } | |
2126 | spin_unlock_irq(&zone->lru_lock); | |
2127 | } | |
2128 | ||
00501b53 | 2129 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2130 | bool lrucare) |
7a81b88c | 2131 | { |
0a31bc97 | 2132 | int isolated; |
9ce70c02 | 2133 | |
1306a85a | 2134 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2135 | |
2136 | /* | |
2137 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2138 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2139 | */ | |
0a31bc97 JW |
2140 | if (lrucare) |
2141 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2142 | |
0a31bc97 JW |
2143 | /* |
2144 | * Nobody should be changing or seriously looking at | |
1306a85a | 2145 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2146 | * |
2147 | * - the page is uncharged | |
2148 | * | |
2149 | * - the page is off-LRU | |
2150 | * | |
2151 | * - an anonymous fault has exclusive page access, except for | |
2152 | * a locked page table | |
2153 | * | |
2154 | * - a page cache insertion, a swapin fault, or a migration | |
2155 | * have the page locked | |
2156 | */ | |
1306a85a | 2157 | page->mem_cgroup = memcg; |
9ce70c02 | 2158 | |
0a31bc97 JW |
2159 | if (lrucare) |
2160 | unlock_page_lru(page, isolated); | |
7a81b88c | 2161 | } |
66e1707b | 2162 | |
127424c8 | 2163 | #ifndef CONFIG_SLOB |
f3bb3043 | 2164 | static int memcg_alloc_cache_id(void) |
55007d84 | 2165 | { |
f3bb3043 VD |
2166 | int id, size; |
2167 | int err; | |
2168 | ||
dbcf73e2 | 2169 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2170 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2171 | if (id < 0) | |
2172 | return id; | |
55007d84 | 2173 | |
dbcf73e2 | 2174 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2175 | return id; |
2176 | ||
2177 | /* | |
2178 | * There's no space for the new id in memcg_caches arrays, | |
2179 | * so we have to grow them. | |
2180 | */ | |
05257a1a | 2181 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2182 | |
2183 | size = 2 * (id + 1); | |
55007d84 GC |
2184 | if (size < MEMCG_CACHES_MIN_SIZE) |
2185 | size = MEMCG_CACHES_MIN_SIZE; | |
2186 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2187 | size = MEMCG_CACHES_MAX_SIZE; | |
2188 | ||
f3bb3043 | 2189 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2190 | if (!err) |
2191 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2192 | if (!err) |
2193 | memcg_nr_cache_ids = size; | |
2194 | ||
2195 | up_write(&memcg_cache_ids_sem); | |
2196 | ||
f3bb3043 | 2197 | if (err) { |
dbcf73e2 | 2198 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2199 | return err; |
2200 | } | |
2201 | return id; | |
2202 | } | |
2203 | ||
2204 | static void memcg_free_cache_id(int id) | |
2205 | { | |
dbcf73e2 | 2206 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2207 | } |
2208 | ||
d5b3cf71 | 2209 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2210 | struct mem_cgroup *memcg; |
2211 | struct kmem_cache *cachep; | |
2212 | struct work_struct work; | |
2213 | }; | |
2214 | ||
d5b3cf71 | 2215 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2216 | { |
d5b3cf71 VD |
2217 | struct memcg_kmem_cache_create_work *cw = |
2218 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2219 | struct mem_cgroup *memcg = cw->memcg; |
2220 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2221 | |
d5b3cf71 | 2222 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2223 | |
5722d094 | 2224 | css_put(&memcg->css); |
d7f25f8a GC |
2225 | kfree(cw); |
2226 | } | |
2227 | ||
2228 | /* | |
2229 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2230 | */ |
d5b3cf71 VD |
2231 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2232 | struct kmem_cache *cachep) | |
d7f25f8a | 2233 | { |
d5b3cf71 | 2234 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2235 | |
776ed0f0 | 2236 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2237 | if (!cw) |
d7f25f8a | 2238 | return; |
8135be5a VD |
2239 | |
2240 | css_get(&memcg->css); | |
d7f25f8a GC |
2241 | |
2242 | cw->memcg = memcg; | |
2243 | cw->cachep = cachep; | |
d5b3cf71 | 2244 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2245 | |
d7f25f8a GC |
2246 | schedule_work(&cw->work); |
2247 | } | |
2248 | ||
d5b3cf71 VD |
2249 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2250 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2251 | { |
2252 | /* | |
2253 | * We need to stop accounting when we kmalloc, because if the | |
2254 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2255 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2256 | * |
2257 | * However, it is better to enclose the whole function. Depending on | |
2258 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2259 | * trigger an allocation. This too, will make us recurse. Because at | |
2260 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2261 | * the safest choice is to do it like this, wrapping the whole function. | |
2262 | */ | |
6f185c29 | 2263 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2264 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2265 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2266 | } |
c67a8a68 | 2267 | |
d7f25f8a GC |
2268 | /* |
2269 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2270 | * We try to use the current memcg's version of the cache. | |
2271 | * | |
2272 | * If the cache does not exist yet, if we are the first user of it, | |
2273 | * we either create it immediately, if possible, or create it asynchronously | |
2274 | * in a workqueue. | |
2275 | * In the latter case, we will let the current allocation go through with | |
2276 | * the original cache. | |
2277 | * | |
2278 | * Can't be called in interrupt context or from kernel threads. | |
2279 | * This function needs to be called with rcu_read_lock() held. | |
2280 | */ | |
230e9fc2 | 2281 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp) |
d7f25f8a GC |
2282 | { |
2283 | struct mem_cgroup *memcg; | |
959c8963 | 2284 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2285 | int kmemcg_id; |
d7f25f8a | 2286 | |
f7ce3190 | 2287 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2288 | |
230e9fc2 VD |
2289 | if (cachep->flags & SLAB_ACCOUNT) |
2290 | gfp |= __GFP_ACCOUNT; | |
2291 | ||
2292 | if (!(gfp & __GFP_ACCOUNT)) | |
2293 | return cachep; | |
2294 | ||
9d100c5e | 2295 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2296 | return cachep; |
2297 | ||
8135be5a | 2298 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2299 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2300 | if (kmemcg_id < 0) |
ca0dde97 | 2301 | goto out; |
d7f25f8a | 2302 | |
2a4db7eb | 2303 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2304 | if (likely(memcg_cachep)) |
2305 | return memcg_cachep; | |
ca0dde97 LZ |
2306 | |
2307 | /* | |
2308 | * If we are in a safe context (can wait, and not in interrupt | |
2309 | * context), we could be be predictable and return right away. | |
2310 | * This would guarantee that the allocation being performed | |
2311 | * already belongs in the new cache. | |
2312 | * | |
2313 | * However, there are some clashes that can arrive from locking. | |
2314 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2315 | * memcg_create_kmem_cache, this means no further allocation |
2316 | * could happen with the slab_mutex held. So it's better to | |
2317 | * defer everything. | |
ca0dde97 | 2318 | */ |
d5b3cf71 | 2319 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2320 | out: |
8135be5a | 2321 | css_put(&memcg->css); |
ca0dde97 | 2322 | return cachep; |
d7f25f8a | 2323 | } |
d7f25f8a | 2324 | |
8135be5a VD |
2325 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2326 | { | |
2327 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2328 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2329 | } |
2330 | ||
f3ccb2c4 VD |
2331 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
2332 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2333 | { |
f3ccb2c4 VD |
2334 | unsigned int nr_pages = 1 << order; |
2335 | struct page_counter *counter; | |
7ae1e1d0 GC |
2336 | int ret; |
2337 | ||
f3ccb2c4 | 2338 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2339 | if (ret) |
f3ccb2c4 | 2340 | return ret; |
52c29b04 JW |
2341 | |
2342 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2343 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2344 | cancel_charge(memcg, nr_pages); | |
2345 | return -ENOMEM; | |
7ae1e1d0 GC |
2346 | } |
2347 | ||
f3ccb2c4 | 2348 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2349 | |
f3ccb2c4 | 2350 | return 0; |
7ae1e1d0 GC |
2351 | } |
2352 | ||
f3ccb2c4 | 2353 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2354 | { |
f3ccb2c4 | 2355 | struct mem_cgroup *memcg; |
fcff7d7e | 2356 | int ret = 0; |
7ae1e1d0 | 2357 | |
f3ccb2c4 | 2358 | memcg = get_mem_cgroup_from_mm(current->mm); |
b6ecd2de | 2359 | if (!mem_cgroup_is_root(memcg)) |
fcff7d7e | 2360 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
7ae1e1d0 | 2361 | css_put(&memcg->css); |
d05e83a6 | 2362 | return ret; |
7ae1e1d0 GC |
2363 | } |
2364 | ||
d05e83a6 | 2365 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2366 | { |
1306a85a | 2367 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2368 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2369 | |
7ae1e1d0 GC |
2370 | if (!memcg) |
2371 | return; | |
2372 | ||
309381fe | 2373 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2374 | |
52c29b04 JW |
2375 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2376 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2377 | ||
f3ccb2c4 | 2378 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2379 | if (do_memsw_account()) |
f3ccb2c4 | 2380 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2381 | |
1306a85a | 2382 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2383 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2384 | } |
127424c8 | 2385 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2386 | |
ca3e0214 KH |
2387 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2388 | ||
ca3e0214 KH |
2389 | /* |
2390 | * Because tail pages are not marked as "used", set it. We're under | |
3ac808fd | 2391 | * zone->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2392 | */ |
e94c8a9c | 2393 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2394 | { |
e94c8a9c | 2395 | int i; |
ca3e0214 | 2396 | |
3d37c4a9 KH |
2397 | if (mem_cgroup_disabled()) |
2398 | return; | |
b070e65c | 2399 | |
29833315 | 2400 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2401 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2402 | |
1306a85a | 2403 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2404 | HPAGE_PMD_NR); |
ca3e0214 | 2405 | } |
12d27107 | 2406 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2407 | |
c255a458 | 2408 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2409 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2410 | bool charge) | |
d13d1443 | 2411 | { |
0a31bc97 JW |
2412 | int val = (charge) ? 1 : -1; |
2413 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2414 | } |
02491447 DN |
2415 | |
2416 | /** | |
2417 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2418 | * @entry: swap entry to be moved | |
2419 | * @from: mem_cgroup which the entry is moved from | |
2420 | * @to: mem_cgroup which the entry is moved to | |
2421 | * | |
2422 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2423 | * as the mem_cgroup's id of @from. | |
2424 | * | |
2425 | * Returns 0 on success, -EINVAL on failure. | |
2426 | * | |
3e32cb2e | 2427 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2428 | * both res and memsw, and called css_get(). |
2429 | */ | |
2430 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2431 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2432 | { |
2433 | unsigned short old_id, new_id; | |
2434 | ||
34c00c31 LZ |
2435 | old_id = mem_cgroup_id(from); |
2436 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2437 | |
2438 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2439 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2440 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2441 | return 0; |
2442 | } | |
2443 | return -EINVAL; | |
2444 | } | |
2445 | #else | |
2446 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2447 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2448 | { |
2449 | return -EINVAL; | |
2450 | } | |
8c7c6e34 | 2451 | #endif |
d13d1443 | 2452 | |
3e32cb2e | 2453 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2454 | |
d38d2a75 | 2455 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2456 | unsigned long limit) |
628f4235 | 2457 | { |
3e32cb2e JW |
2458 | unsigned long curusage; |
2459 | unsigned long oldusage; | |
2460 | bool enlarge = false; | |
81d39c20 | 2461 | int retry_count; |
3e32cb2e | 2462 | int ret; |
81d39c20 KH |
2463 | |
2464 | /* | |
2465 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2466 | * is depends on callers. We set our retry-count to be function | |
2467 | * of # of children which we should visit in this loop. | |
2468 | */ | |
3e32cb2e JW |
2469 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2470 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2471 | |
3e32cb2e | 2472 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2473 | |
3e32cb2e | 2474 | do { |
628f4235 KH |
2475 | if (signal_pending(current)) { |
2476 | ret = -EINTR; | |
2477 | break; | |
2478 | } | |
3e32cb2e JW |
2479 | |
2480 | mutex_lock(&memcg_limit_mutex); | |
2481 | if (limit > memcg->memsw.limit) { | |
2482 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2483 | ret = -EINVAL; |
628f4235 KH |
2484 | break; |
2485 | } | |
3e32cb2e JW |
2486 | if (limit > memcg->memory.limit) |
2487 | enlarge = true; | |
2488 | ret = page_counter_limit(&memcg->memory, limit); | |
2489 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2490 | |
2491 | if (!ret) | |
2492 | break; | |
2493 | ||
b70a2a21 JW |
2494 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2495 | ||
3e32cb2e | 2496 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2497 | /* Usage is reduced ? */ |
f894ffa8 | 2498 | if (curusage >= oldusage) |
81d39c20 KH |
2499 | retry_count--; |
2500 | else | |
2501 | oldusage = curusage; | |
3e32cb2e JW |
2502 | } while (retry_count); |
2503 | ||
3c11ecf4 KH |
2504 | if (!ret && enlarge) |
2505 | memcg_oom_recover(memcg); | |
14797e23 | 2506 | |
8c7c6e34 KH |
2507 | return ret; |
2508 | } | |
2509 | ||
338c8431 | 2510 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2511 | unsigned long limit) |
8c7c6e34 | 2512 | { |
3e32cb2e JW |
2513 | unsigned long curusage; |
2514 | unsigned long oldusage; | |
2515 | bool enlarge = false; | |
81d39c20 | 2516 | int retry_count; |
3e32cb2e | 2517 | int ret; |
8c7c6e34 | 2518 | |
81d39c20 | 2519 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2520 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2521 | mem_cgroup_count_children(memcg); | |
2522 | ||
2523 | oldusage = page_counter_read(&memcg->memsw); | |
2524 | ||
2525 | do { | |
8c7c6e34 KH |
2526 | if (signal_pending(current)) { |
2527 | ret = -EINTR; | |
2528 | break; | |
2529 | } | |
3e32cb2e JW |
2530 | |
2531 | mutex_lock(&memcg_limit_mutex); | |
2532 | if (limit < memcg->memory.limit) { | |
2533 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2534 | ret = -EINVAL; |
8c7c6e34 KH |
2535 | break; |
2536 | } | |
3e32cb2e JW |
2537 | if (limit > memcg->memsw.limit) |
2538 | enlarge = true; | |
2539 | ret = page_counter_limit(&memcg->memsw, limit); | |
2540 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2541 | |
2542 | if (!ret) | |
2543 | break; | |
2544 | ||
b70a2a21 JW |
2545 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2546 | ||
3e32cb2e | 2547 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2548 | /* Usage is reduced ? */ |
8c7c6e34 | 2549 | if (curusage >= oldusage) |
628f4235 | 2550 | retry_count--; |
81d39c20 KH |
2551 | else |
2552 | oldusage = curusage; | |
3e32cb2e JW |
2553 | } while (retry_count); |
2554 | ||
3c11ecf4 KH |
2555 | if (!ret && enlarge) |
2556 | memcg_oom_recover(memcg); | |
3e32cb2e | 2557 | |
628f4235 KH |
2558 | return ret; |
2559 | } | |
2560 | ||
0608f43d AM |
2561 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2562 | gfp_t gfp_mask, | |
2563 | unsigned long *total_scanned) | |
2564 | { | |
2565 | unsigned long nr_reclaimed = 0; | |
2566 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2567 | unsigned long reclaimed; | |
2568 | int loop = 0; | |
2569 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2570 | unsigned long excess; |
0608f43d AM |
2571 | unsigned long nr_scanned; |
2572 | ||
2573 | if (order > 0) | |
2574 | return 0; | |
2575 | ||
2576 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2577 | /* | |
2578 | * This loop can run a while, specially if mem_cgroup's continuously | |
2579 | * keep exceeding their soft limit and putting the system under | |
2580 | * pressure | |
2581 | */ | |
2582 | do { | |
2583 | if (next_mz) | |
2584 | mz = next_mz; | |
2585 | else | |
2586 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2587 | if (!mz) | |
2588 | break; | |
2589 | ||
2590 | nr_scanned = 0; | |
2591 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2592 | gfp_mask, &nr_scanned); | |
2593 | nr_reclaimed += reclaimed; | |
2594 | *total_scanned += nr_scanned; | |
0a31bc97 | 2595 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2596 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2597 | |
2598 | /* | |
2599 | * If we failed to reclaim anything from this memory cgroup | |
2600 | * it is time to move on to the next cgroup | |
2601 | */ | |
2602 | next_mz = NULL; | |
bc2f2e7f VD |
2603 | if (!reclaimed) |
2604 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2605 | ||
3e32cb2e | 2606 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2607 | /* |
2608 | * One school of thought says that we should not add | |
2609 | * back the node to the tree if reclaim returns 0. | |
2610 | * But our reclaim could return 0, simply because due | |
2611 | * to priority we are exposing a smaller subset of | |
2612 | * memory to reclaim from. Consider this as a longer | |
2613 | * term TODO. | |
2614 | */ | |
2615 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2616 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2617 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2618 | css_put(&mz->memcg->css); |
2619 | loop++; | |
2620 | /* | |
2621 | * Could not reclaim anything and there are no more | |
2622 | * mem cgroups to try or we seem to be looping without | |
2623 | * reclaiming anything. | |
2624 | */ | |
2625 | if (!nr_reclaimed && | |
2626 | (next_mz == NULL || | |
2627 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2628 | break; | |
2629 | } while (!nr_reclaimed); | |
2630 | if (next_mz) | |
2631 | css_put(&next_mz->memcg->css); | |
2632 | return nr_reclaimed; | |
2633 | } | |
2634 | ||
ea280e7b TH |
2635 | /* |
2636 | * Test whether @memcg has children, dead or alive. Note that this | |
2637 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2638 | * returns %true if there are child csses according to the cgroup | |
2639 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2640 | */ | |
b5f99b53 GC |
2641 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2642 | { | |
ea280e7b TH |
2643 | bool ret; |
2644 | ||
ea280e7b TH |
2645 | rcu_read_lock(); |
2646 | ret = css_next_child(NULL, &memcg->css); | |
2647 | rcu_read_unlock(); | |
2648 | return ret; | |
b5f99b53 GC |
2649 | } |
2650 | ||
c26251f9 MH |
2651 | /* |
2652 | * Reclaims as many pages from the given memcg as possible and moves | |
2653 | * the rest to the parent. | |
2654 | * | |
2655 | * Caller is responsible for holding css reference for memcg. | |
2656 | */ | |
2657 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2658 | { | |
2659 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2660 | |
c1e862c1 KH |
2661 | /* we call try-to-free pages for make this cgroup empty */ |
2662 | lru_add_drain_all(); | |
f817ed48 | 2663 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2664 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2665 | int progress; |
c1e862c1 | 2666 | |
c26251f9 MH |
2667 | if (signal_pending(current)) |
2668 | return -EINTR; | |
2669 | ||
b70a2a21 JW |
2670 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2671 | GFP_KERNEL, true); | |
c1e862c1 | 2672 | if (!progress) { |
f817ed48 | 2673 | nr_retries--; |
c1e862c1 | 2674 | /* maybe some writeback is necessary */ |
8aa7e847 | 2675 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2676 | } |
f817ed48 KH |
2677 | |
2678 | } | |
ab5196c2 MH |
2679 | |
2680 | return 0; | |
cc847582 KH |
2681 | } |
2682 | ||
6770c64e TH |
2683 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2684 | char *buf, size_t nbytes, | |
2685 | loff_t off) | |
c1e862c1 | 2686 | { |
6770c64e | 2687 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2688 | |
d8423011 MH |
2689 | if (mem_cgroup_is_root(memcg)) |
2690 | return -EINVAL; | |
6770c64e | 2691 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2692 | } |
2693 | ||
182446d0 TH |
2694 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2695 | struct cftype *cft) | |
18f59ea7 | 2696 | { |
182446d0 | 2697 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2698 | } |
2699 | ||
182446d0 TH |
2700 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2701 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2702 | { |
2703 | int retval = 0; | |
182446d0 | 2704 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2705 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2706 | |
567fb435 | 2707 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2708 | return 0; |
567fb435 | 2709 | |
18f59ea7 | 2710 | /* |
af901ca1 | 2711 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2712 | * in the child subtrees. If it is unset, then the change can |
2713 | * occur, provided the current cgroup has no children. | |
2714 | * | |
2715 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2716 | * set if there are no children. | |
2717 | */ | |
c0ff4b85 | 2718 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2719 | (val == 1 || val == 0)) { |
ea280e7b | 2720 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2721 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2722 | else |
2723 | retval = -EBUSY; | |
2724 | } else | |
2725 | retval = -EINVAL; | |
567fb435 | 2726 | |
18f59ea7 BS |
2727 | return retval; |
2728 | } | |
2729 | ||
72b54e73 | 2730 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
ce00a967 JW |
2731 | { |
2732 | struct mem_cgroup *iter; | |
72b54e73 | 2733 | int i; |
ce00a967 | 2734 | |
72b54e73 | 2735 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
ce00a967 | 2736 | |
72b54e73 VD |
2737 | for_each_mem_cgroup_tree(iter, memcg) { |
2738 | for (i = 0; i < MEMCG_NR_STAT; i++) | |
2739 | stat[i] += mem_cgroup_read_stat(iter, i); | |
2740 | } | |
ce00a967 JW |
2741 | } |
2742 | ||
72b54e73 | 2743 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
587d9f72 JW |
2744 | { |
2745 | struct mem_cgroup *iter; | |
72b54e73 | 2746 | int i; |
587d9f72 | 2747 | |
72b54e73 | 2748 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
587d9f72 | 2749 | |
72b54e73 VD |
2750 | for_each_mem_cgroup_tree(iter, memcg) { |
2751 | for (i = 0; i < MEMCG_NR_EVENTS; i++) | |
2752 | events[i] += mem_cgroup_read_events(iter, i); | |
2753 | } | |
587d9f72 JW |
2754 | } |
2755 | ||
6f646156 | 2756 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2757 | { |
72b54e73 | 2758 | unsigned long val = 0; |
ce00a967 | 2759 | |
3e32cb2e | 2760 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2761 | struct mem_cgroup *iter; |
2762 | ||
2763 | for_each_mem_cgroup_tree(iter, memcg) { | |
2764 | val += mem_cgroup_read_stat(iter, | |
2765 | MEM_CGROUP_STAT_CACHE); | |
2766 | val += mem_cgroup_read_stat(iter, | |
2767 | MEM_CGROUP_STAT_RSS); | |
2768 | if (swap) | |
2769 | val += mem_cgroup_read_stat(iter, | |
2770 | MEM_CGROUP_STAT_SWAP); | |
2771 | } | |
3e32cb2e | 2772 | } else { |
ce00a967 | 2773 | if (!swap) |
3e32cb2e | 2774 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2775 | else |
3e32cb2e | 2776 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2777 | } |
c12176d3 | 2778 | return val; |
ce00a967 JW |
2779 | } |
2780 | ||
3e32cb2e JW |
2781 | enum { |
2782 | RES_USAGE, | |
2783 | RES_LIMIT, | |
2784 | RES_MAX_USAGE, | |
2785 | RES_FAILCNT, | |
2786 | RES_SOFT_LIMIT, | |
2787 | }; | |
ce00a967 | 2788 | |
791badbd | 2789 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2790 | struct cftype *cft) |
8cdea7c0 | 2791 | { |
182446d0 | 2792 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2793 | struct page_counter *counter; |
af36f906 | 2794 | |
3e32cb2e | 2795 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2796 | case _MEM: |
3e32cb2e JW |
2797 | counter = &memcg->memory; |
2798 | break; | |
8c7c6e34 | 2799 | case _MEMSWAP: |
3e32cb2e JW |
2800 | counter = &memcg->memsw; |
2801 | break; | |
510fc4e1 | 2802 | case _KMEM: |
3e32cb2e | 2803 | counter = &memcg->kmem; |
510fc4e1 | 2804 | break; |
d55f90bf | 2805 | case _TCP: |
0db15298 | 2806 | counter = &memcg->tcpmem; |
d55f90bf | 2807 | break; |
8c7c6e34 KH |
2808 | default: |
2809 | BUG(); | |
8c7c6e34 | 2810 | } |
3e32cb2e JW |
2811 | |
2812 | switch (MEMFILE_ATTR(cft->private)) { | |
2813 | case RES_USAGE: | |
2814 | if (counter == &memcg->memory) | |
c12176d3 | 2815 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2816 | if (counter == &memcg->memsw) |
c12176d3 | 2817 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2818 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2819 | case RES_LIMIT: | |
2820 | return (u64)counter->limit * PAGE_SIZE; | |
2821 | case RES_MAX_USAGE: | |
2822 | return (u64)counter->watermark * PAGE_SIZE; | |
2823 | case RES_FAILCNT: | |
2824 | return counter->failcnt; | |
2825 | case RES_SOFT_LIMIT: | |
2826 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2827 | default: | |
2828 | BUG(); | |
2829 | } | |
8cdea7c0 | 2830 | } |
510fc4e1 | 2831 | |
127424c8 | 2832 | #ifndef CONFIG_SLOB |
567e9ab2 | 2833 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2834 | { |
d6441637 VD |
2835 | int memcg_id; |
2836 | ||
b313aeee VD |
2837 | if (cgroup_memory_nokmem) |
2838 | return 0; | |
2839 | ||
2a4db7eb | 2840 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2841 | BUG_ON(memcg->kmem_state); |
d6441637 | 2842 | |
f3bb3043 | 2843 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
2844 | if (memcg_id < 0) |
2845 | return memcg_id; | |
d6441637 | 2846 | |
ef12947c | 2847 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2848 | /* |
567e9ab2 | 2849 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2850 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2851 | * guarantee no one starts accounting before all call sites are |
2852 | * patched. | |
2853 | */ | |
900a38f0 | 2854 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2855 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
2856 | |
2857 | return 0; | |
d6441637 VD |
2858 | } |
2859 | ||
8e0a8912 JW |
2860 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2861 | { | |
2862 | struct cgroup_subsys_state *css; | |
2863 | struct mem_cgroup *parent, *child; | |
2864 | int kmemcg_id; | |
2865 | ||
2866 | if (memcg->kmem_state != KMEM_ONLINE) | |
2867 | return; | |
2868 | /* | |
2869 | * Clear the online state before clearing memcg_caches array | |
2870 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2871 | * guarantees that no cache will be created for this cgroup | |
2872 | * after we are done (see memcg_create_kmem_cache()). | |
2873 | */ | |
2874 | memcg->kmem_state = KMEM_ALLOCATED; | |
2875 | ||
2876 | memcg_deactivate_kmem_caches(memcg); | |
2877 | ||
2878 | kmemcg_id = memcg->kmemcg_id; | |
2879 | BUG_ON(kmemcg_id < 0); | |
2880 | ||
2881 | parent = parent_mem_cgroup(memcg); | |
2882 | if (!parent) | |
2883 | parent = root_mem_cgroup; | |
2884 | ||
2885 | /* | |
2886 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2887 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2888 | * to ones of the parent. After we have finished, all list_lrus | |
2889 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2890 | * ordering is imposed by list_lru_node->lock taken by | |
2891 | * memcg_drain_all_list_lrus(). | |
2892 | */ | |
2893 | css_for_each_descendant_pre(css, &memcg->css) { | |
2894 | child = mem_cgroup_from_css(css); | |
2895 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2896 | child->kmemcg_id = parent->kmemcg_id; | |
2897 | if (!memcg->use_hierarchy) | |
2898 | break; | |
2899 | } | |
2900 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
2901 | ||
2902 | memcg_free_cache_id(kmemcg_id); | |
2903 | } | |
2904 | ||
2905 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2906 | { | |
0b8f73e1 JW |
2907 | /* css_alloc() failed, offlining didn't happen */ |
2908 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
2909 | memcg_offline_kmem(memcg); | |
2910 | ||
8e0a8912 JW |
2911 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
2912 | memcg_destroy_kmem_caches(memcg); | |
2913 | static_branch_dec(&memcg_kmem_enabled_key); | |
2914 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2915 | } | |
8e0a8912 | 2916 | } |
d6441637 | 2917 | #else |
0b8f73e1 | 2918 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
2919 | { |
2920 | return 0; | |
2921 | } | |
2922 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
2923 | { | |
2924 | } | |
2925 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2926 | { | |
2927 | } | |
2928 | #endif /* !CONFIG_SLOB */ | |
2929 | ||
d6441637 | 2930 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2931 | unsigned long limit) |
d6441637 | 2932 | { |
b313aeee | 2933 | int ret; |
127424c8 JW |
2934 | |
2935 | mutex_lock(&memcg_limit_mutex); | |
127424c8 | 2936 | ret = page_counter_limit(&memcg->kmem, limit); |
127424c8 JW |
2937 | mutex_unlock(&memcg_limit_mutex); |
2938 | return ret; | |
d6441637 | 2939 | } |
510fc4e1 | 2940 | |
d55f90bf VD |
2941 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
2942 | { | |
2943 | int ret; | |
2944 | ||
2945 | mutex_lock(&memcg_limit_mutex); | |
2946 | ||
0db15298 | 2947 | ret = page_counter_limit(&memcg->tcpmem, limit); |
d55f90bf VD |
2948 | if (ret) |
2949 | goto out; | |
2950 | ||
0db15298 | 2951 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
2952 | /* |
2953 | * The active flag needs to be written after the static_key | |
2954 | * update. This is what guarantees that the socket activation | |
2955 | * function is the last one to run. See sock_update_memcg() for | |
2956 | * details, and note that we don't mark any socket as belonging | |
2957 | * to this memcg until that flag is up. | |
2958 | * | |
2959 | * We need to do this, because static_keys will span multiple | |
2960 | * sites, but we can't control their order. If we mark a socket | |
2961 | * as accounted, but the accounting functions are not patched in | |
2962 | * yet, we'll lose accounting. | |
2963 | * | |
2964 | * We never race with the readers in sock_update_memcg(), | |
2965 | * because when this value change, the code to process it is not | |
2966 | * patched in yet. | |
2967 | */ | |
2968 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 2969 | memcg->tcpmem_active = true; |
d55f90bf VD |
2970 | } |
2971 | out: | |
2972 | mutex_unlock(&memcg_limit_mutex); | |
2973 | return ret; | |
2974 | } | |
d55f90bf | 2975 | |
628f4235 KH |
2976 | /* |
2977 | * The user of this function is... | |
2978 | * RES_LIMIT. | |
2979 | */ | |
451af504 TH |
2980 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2981 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2982 | { |
451af504 | 2983 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2984 | unsigned long nr_pages; |
628f4235 KH |
2985 | int ret; |
2986 | ||
451af504 | 2987 | buf = strstrip(buf); |
650c5e56 | 2988 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
2989 | if (ret) |
2990 | return ret; | |
af36f906 | 2991 | |
3e32cb2e | 2992 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 2993 | case RES_LIMIT: |
4b3bde4c BS |
2994 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2995 | ret = -EINVAL; | |
2996 | break; | |
2997 | } | |
3e32cb2e JW |
2998 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
2999 | case _MEM: | |
3000 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3001 | break; |
3e32cb2e JW |
3002 | case _MEMSWAP: |
3003 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3004 | break; |
3e32cb2e JW |
3005 | case _KMEM: |
3006 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3007 | break; | |
d55f90bf VD |
3008 | case _TCP: |
3009 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
3010 | break; | |
3e32cb2e | 3011 | } |
296c81d8 | 3012 | break; |
3e32cb2e JW |
3013 | case RES_SOFT_LIMIT: |
3014 | memcg->soft_limit = nr_pages; | |
3015 | ret = 0; | |
628f4235 KH |
3016 | break; |
3017 | } | |
451af504 | 3018 | return ret ?: nbytes; |
8cdea7c0 BS |
3019 | } |
3020 | ||
6770c64e TH |
3021 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3022 | size_t nbytes, loff_t off) | |
c84872e1 | 3023 | { |
6770c64e | 3024 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3025 | struct page_counter *counter; |
c84872e1 | 3026 | |
3e32cb2e JW |
3027 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3028 | case _MEM: | |
3029 | counter = &memcg->memory; | |
3030 | break; | |
3031 | case _MEMSWAP: | |
3032 | counter = &memcg->memsw; | |
3033 | break; | |
3034 | case _KMEM: | |
3035 | counter = &memcg->kmem; | |
3036 | break; | |
d55f90bf | 3037 | case _TCP: |
0db15298 | 3038 | counter = &memcg->tcpmem; |
d55f90bf | 3039 | break; |
3e32cb2e JW |
3040 | default: |
3041 | BUG(); | |
3042 | } | |
af36f906 | 3043 | |
3e32cb2e | 3044 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3045 | case RES_MAX_USAGE: |
3e32cb2e | 3046 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3047 | break; |
3048 | case RES_FAILCNT: | |
3e32cb2e | 3049 | counter->failcnt = 0; |
29f2a4da | 3050 | break; |
3e32cb2e JW |
3051 | default: |
3052 | BUG(); | |
29f2a4da | 3053 | } |
f64c3f54 | 3054 | |
6770c64e | 3055 | return nbytes; |
c84872e1 PE |
3056 | } |
3057 | ||
182446d0 | 3058 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3059 | struct cftype *cft) |
3060 | { | |
182446d0 | 3061 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3062 | } |
3063 | ||
02491447 | 3064 | #ifdef CONFIG_MMU |
182446d0 | 3065 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3066 | struct cftype *cft, u64 val) |
3067 | { | |
182446d0 | 3068 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3069 | |
1dfab5ab | 3070 | if (val & ~MOVE_MASK) |
7dc74be0 | 3071 | return -EINVAL; |
ee5e8472 | 3072 | |
7dc74be0 | 3073 | /* |
ee5e8472 GC |
3074 | * No kind of locking is needed in here, because ->can_attach() will |
3075 | * check this value once in the beginning of the process, and then carry | |
3076 | * on with stale data. This means that changes to this value will only | |
3077 | * affect task migrations starting after the change. | |
7dc74be0 | 3078 | */ |
c0ff4b85 | 3079 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3080 | return 0; |
3081 | } | |
02491447 | 3082 | #else |
182446d0 | 3083 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3084 | struct cftype *cft, u64 val) |
3085 | { | |
3086 | return -ENOSYS; | |
3087 | } | |
3088 | #endif | |
7dc74be0 | 3089 | |
406eb0c9 | 3090 | #ifdef CONFIG_NUMA |
2da8ca82 | 3091 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3092 | { |
25485de6 GT |
3093 | struct numa_stat { |
3094 | const char *name; | |
3095 | unsigned int lru_mask; | |
3096 | }; | |
3097 | ||
3098 | static const struct numa_stat stats[] = { | |
3099 | { "total", LRU_ALL }, | |
3100 | { "file", LRU_ALL_FILE }, | |
3101 | { "anon", LRU_ALL_ANON }, | |
3102 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3103 | }; | |
3104 | const struct numa_stat *stat; | |
406eb0c9 | 3105 | int nid; |
25485de6 | 3106 | unsigned long nr; |
2da8ca82 | 3107 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3108 | |
25485de6 GT |
3109 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3110 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3111 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3112 | for_each_node_state(nid, N_MEMORY) { | |
3113 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3114 | stat->lru_mask); | |
3115 | seq_printf(m, " N%d=%lu", nid, nr); | |
3116 | } | |
3117 | seq_putc(m, '\n'); | |
406eb0c9 | 3118 | } |
406eb0c9 | 3119 | |
071aee13 YH |
3120 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3121 | struct mem_cgroup *iter; | |
3122 | ||
3123 | nr = 0; | |
3124 | for_each_mem_cgroup_tree(iter, memcg) | |
3125 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3126 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3127 | for_each_node_state(nid, N_MEMORY) { | |
3128 | nr = 0; | |
3129 | for_each_mem_cgroup_tree(iter, memcg) | |
3130 | nr += mem_cgroup_node_nr_lru_pages( | |
3131 | iter, nid, stat->lru_mask); | |
3132 | seq_printf(m, " N%d=%lu", nid, nr); | |
3133 | } | |
3134 | seq_putc(m, '\n'); | |
406eb0c9 | 3135 | } |
406eb0c9 | 3136 | |
406eb0c9 YH |
3137 | return 0; |
3138 | } | |
3139 | #endif /* CONFIG_NUMA */ | |
3140 | ||
2da8ca82 | 3141 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3142 | { |
2da8ca82 | 3143 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3144 | unsigned long memory, memsw; |
af7c4b0e JW |
3145 | struct mem_cgroup *mi; |
3146 | unsigned int i; | |
406eb0c9 | 3147 | |
0ca44b14 GT |
3148 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3149 | MEM_CGROUP_STAT_NSTATS); | |
3150 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3151 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3152 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3153 | ||
af7c4b0e | 3154 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
7941d214 | 3155 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3156 | continue; |
484ebb3b | 3157 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3158 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3159 | } |
7b854121 | 3160 | |
af7c4b0e JW |
3161 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3162 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3163 | mem_cgroup_read_events(memcg, i)); | |
3164 | ||
3165 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3166 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3167 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3168 | ||
14067bb3 | 3169 | /* Hierarchical information */ |
3e32cb2e JW |
3170 | memory = memsw = PAGE_COUNTER_MAX; |
3171 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3172 | memory = min(memory, mi->memory.limit); | |
3173 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3174 | } |
3e32cb2e JW |
3175 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3176 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3177 | if (do_memsw_account()) |
3e32cb2e JW |
3178 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3179 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3180 | |
af7c4b0e | 3181 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3182 | unsigned long long val = 0; |
af7c4b0e | 3183 | |
7941d214 | 3184 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3185 | continue; |
af7c4b0e JW |
3186 | for_each_mem_cgroup_tree(mi, memcg) |
3187 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3188 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3189 | } |
3190 | ||
3191 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3192 | unsigned long long val = 0; | |
3193 | ||
3194 | for_each_mem_cgroup_tree(mi, memcg) | |
3195 | val += mem_cgroup_read_events(mi, i); | |
3196 | seq_printf(m, "total_%s %llu\n", | |
3197 | mem_cgroup_events_names[i], val); | |
3198 | } | |
3199 | ||
3200 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3201 | unsigned long long val = 0; | |
3202 | ||
3203 | for_each_mem_cgroup_tree(mi, memcg) | |
3204 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3205 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3206 | } |
14067bb3 | 3207 | |
7f016ee8 | 3208 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3209 | { |
3210 | int nid, zid; | |
3211 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3212 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3213 | unsigned long recent_rotated[2] = {0, 0}; |
3214 | unsigned long recent_scanned[2] = {0, 0}; | |
3215 | ||
3216 | for_each_online_node(nid) | |
3217 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3218 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3219 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3220 | |
89abfab1 HD |
3221 | recent_rotated[0] += rstat->recent_rotated[0]; |
3222 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3223 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3224 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3225 | } |
78ccf5b5 JW |
3226 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3227 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3228 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3229 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3230 | } |
3231 | #endif | |
3232 | ||
d2ceb9b7 KH |
3233 | return 0; |
3234 | } | |
3235 | ||
182446d0 TH |
3236 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3237 | struct cftype *cft) | |
a7885eb8 | 3238 | { |
182446d0 | 3239 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3240 | |
1f4c025b | 3241 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3242 | } |
3243 | ||
182446d0 TH |
3244 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3245 | struct cftype *cft, u64 val) | |
a7885eb8 | 3246 | { |
182446d0 | 3247 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3248 | |
3dae7fec | 3249 | if (val > 100) |
a7885eb8 KM |
3250 | return -EINVAL; |
3251 | ||
14208b0e | 3252 | if (css->parent) |
3dae7fec JW |
3253 | memcg->swappiness = val; |
3254 | else | |
3255 | vm_swappiness = val; | |
068b38c1 | 3256 | |
a7885eb8 KM |
3257 | return 0; |
3258 | } | |
3259 | ||
2e72b634 KS |
3260 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3261 | { | |
3262 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3263 | unsigned long usage; |
2e72b634 KS |
3264 | int i; |
3265 | ||
3266 | rcu_read_lock(); | |
3267 | if (!swap) | |
2c488db2 | 3268 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3269 | else |
2c488db2 | 3270 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3271 | |
3272 | if (!t) | |
3273 | goto unlock; | |
3274 | ||
ce00a967 | 3275 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3276 | |
3277 | /* | |
748dad36 | 3278 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3279 | * If it's not true, a threshold was crossed after last |
3280 | * call of __mem_cgroup_threshold(). | |
3281 | */ | |
5407a562 | 3282 | i = t->current_threshold; |
2e72b634 KS |
3283 | |
3284 | /* | |
3285 | * Iterate backward over array of thresholds starting from | |
3286 | * current_threshold and check if a threshold is crossed. | |
3287 | * If none of thresholds below usage is crossed, we read | |
3288 | * only one element of the array here. | |
3289 | */ | |
3290 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3291 | eventfd_signal(t->entries[i].eventfd, 1); | |
3292 | ||
3293 | /* i = current_threshold + 1 */ | |
3294 | i++; | |
3295 | ||
3296 | /* | |
3297 | * Iterate forward over array of thresholds starting from | |
3298 | * current_threshold+1 and check if a threshold is crossed. | |
3299 | * If none of thresholds above usage is crossed, we read | |
3300 | * only one element of the array here. | |
3301 | */ | |
3302 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3303 | eventfd_signal(t->entries[i].eventfd, 1); | |
3304 | ||
3305 | /* Update current_threshold */ | |
5407a562 | 3306 | t->current_threshold = i - 1; |
2e72b634 KS |
3307 | unlock: |
3308 | rcu_read_unlock(); | |
3309 | } | |
3310 | ||
3311 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3312 | { | |
ad4ca5f4 KS |
3313 | while (memcg) { |
3314 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3315 | if (do_memsw_account()) |
ad4ca5f4 KS |
3316 | __mem_cgroup_threshold(memcg, true); |
3317 | ||
3318 | memcg = parent_mem_cgroup(memcg); | |
3319 | } | |
2e72b634 KS |
3320 | } |
3321 | ||
3322 | static int compare_thresholds(const void *a, const void *b) | |
3323 | { | |
3324 | const struct mem_cgroup_threshold *_a = a; | |
3325 | const struct mem_cgroup_threshold *_b = b; | |
3326 | ||
2bff24a3 GT |
3327 | if (_a->threshold > _b->threshold) |
3328 | return 1; | |
3329 | ||
3330 | if (_a->threshold < _b->threshold) | |
3331 | return -1; | |
3332 | ||
3333 | return 0; | |
2e72b634 KS |
3334 | } |
3335 | ||
c0ff4b85 | 3336 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3337 | { |
3338 | struct mem_cgroup_eventfd_list *ev; | |
3339 | ||
2bcf2e92 MH |
3340 | spin_lock(&memcg_oom_lock); |
3341 | ||
c0ff4b85 | 3342 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3343 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3344 | |
3345 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3346 | return 0; |
3347 | } | |
3348 | ||
c0ff4b85 | 3349 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3350 | { |
7d74b06f KH |
3351 | struct mem_cgroup *iter; |
3352 | ||
c0ff4b85 | 3353 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3354 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3355 | } |
3356 | ||
59b6f873 | 3357 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3358 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3359 | { |
2c488db2 KS |
3360 | struct mem_cgroup_thresholds *thresholds; |
3361 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3362 | unsigned long threshold; |
3363 | unsigned long usage; | |
2c488db2 | 3364 | int i, size, ret; |
2e72b634 | 3365 | |
650c5e56 | 3366 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3367 | if (ret) |
3368 | return ret; | |
3369 | ||
3370 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3371 | |
05b84301 | 3372 | if (type == _MEM) { |
2c488db2 | 3373 | thresholds = &memcg->thresholds; |
ce00a967 | 3374 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3375 | } else if (type == _MEMSWAP) { |
2c488db2 | 3376 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3377 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3378 | } else |
2e72b634 KS |
3379 | BUG(); |
3380 | ||
2e72b634 | 3381 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3382 | if (thresholds->primary) |
2e72b634 KS |
3383 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3384 | ||
2c488db2 | 3385 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3386 | |
3387 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3388 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3389 | GFP_KERNEL); |
2c488db2 | 3390 | if (!new) { |
2e72b634 KS |
3391 | ret = -ENOMEM; |
3392 | goto unlock; | |
3393 | } | |
2c488db2 | 3394 | new->size = size; |
2e72b634 KS |
3395 | |
3396 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3397 | if (thresholds->primary) { |
3398 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3399 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3400 | } |
3401 | ||
2e72b634 | 3402 | /* Add new threshold */ |
2c488db2 KS |
3403 | new->entries[size - 1].eventfd = eventfd; |
3404 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3405 | |
3406 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3407 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3408 | compare_thresholds, NULL); |
3409 | ||
3410 | /* Find current threshold */ | |
2c488db2 | 3411 | new->current_threshold = -1; |
2e72b634 | 3412 | for (i = 0; i < size; i++) { |
748dad36 | 3413 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3414 | /* |
2c488db2 KS |
3415 | * new->current_threshold will not be used until |
3416 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3417 | * it here. |
3418 | */ | |
2c488db2 | 3419 | ++new->current_threshold; |
748dad36 SZ |
3420 | } else |
3421 | break; | |
2e72b634 KS |
3422 | } |
3423 | ||
2c488db2 KS |
3424 | /* Free old spare buffer and save old primary buffer as spare */ |
3425 | kfree(thresholds->spare); | |
3426 | thresholds->spare = thresholds->primary; | |
3427 | ||
3428 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3429 | |
907860ed | 3430 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3431 | synchronize_rcu(); |
3432 | ||
2e72b634 KS |
3433 | unlock: |
3434 | mutex_unlock(&memcg->thresholds_lock); | |
3435 | ||
3436 | return ret; | |
3437 | } | |
3438 | ||
59b6f873 | 3439 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3440 | struct eventfd_ctx *eventfd, const char *args) |
3441 | { | |
59b6f873 | 3442 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3443 | } |
3444 | ||
59b6f873 | 3445 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3446 | struct eventfd_ctx *eventfd, const char *args) |
3447 | { | |
59b6f873 | 3448 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3449 | } |
3450 | ||
59b6f873 | 3451 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3452 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3453 | { |
2c488db2 KS |
3454 | struct mem_cgroup_thresholds *thresholds; |
3455 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3456 | unsigned long usage; |
2c488db2 | 3457 | int i, j, size; |
2e72b634 KS |
3458 | |
3459 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3460 | |
3461 | if (type == _MEM) { | |
2c488db2 | 3462 | thresholds = &memcg->thresholds; |
ce00a967 | 3463 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3464 | } else if (type == _MEMSWAP) { |
2c488db2 | 3465 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3466 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3467 | } else |
2e72b634 KS |
3468 | BUG(); |
3469 | ||
371528ca AV |
3470 | if (!thresholds->primary) |
3471 | goto unlock; | |
3472 | ||
2e72b634 KS |
3473 | /* Check if a threshold crossed before removing */ |
3474 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3475 | ||
3476 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3477 | size = 0; |
3478 | for (i = 0; i < thresholds->primary->size; i++) { | |
3479 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3480 | size++; |
3481 | } | |
3482 | ||
2c488db2 | 3483 | new = thresholds->spare; |
907860ed | 3484 | |
2e72b634 KS |
3485 | /* Set thresholds array to NULL if we don't have thresholds */ |
3486 | if (!size) { | |
2c488db2 KS |
3487 | kfree(new); |
3488 | new = NULL; | |
907860ed | 3489 | goto swap_buffers; |
2e72b634 KS |
3490 | } |
3491 | ||
2c488db2 | 3492 | new->size = size; |
2e72b634 KS |
3493 | |
3494 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3495 | new->current_threshold = -1; |
3496 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3497 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3498 | continue; |
3499 | ||
2c488db2 | 3500 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3501 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3502 | /* |
2c488db2 | 3503 | * new->current_threshold will not be used |
2e72b634 KS |
3504 | * until rcu_assign_pointer(), so it's safe to increment |
3505 | * it here. | |
3506 | */ | |
2c488db2 | 3507 | ++new->current_threshold; |
2e72b634 KS |
3508 | } |
3509 | j++; | |
3510 | } | |
3511 | ||
907860ed | 3512 | swap_buffers: |
2c488db2 KS |
3513 | /* Swap primary and spare array */ |
3514 | thresholds->spare = thresholds->primary; | |
8c757763 | 3515 | |
2c488db2 | 3516 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3517 | |
907860ed | 3518 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3519 | synchronize_rcu(); |
6611d8d7 MC |
3520 | |
3521 | /* If all events are unregistered, free the spare array */ | |
3522 | if (!new) { | |
3523 | kfree(thresholds->spare); | |
3524 | thresholds->spare = NULL; | |
3525 | } | |
371528ca | 3526 | unlock: |
2e72b634 | 3527 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3528 | } |
c1e862c1 | 3529 | |
59b6f873 | 3530 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3531 | struct eventfd_ctx *eventfd) |
3532 | { | |
59b6f873 | 3533 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3534 | } |
3535 | ||
59b6f873 | 3536 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3537 | struct eventfd_ctx *eventfd) |
3538 | { | |
59b6f873 | 3539 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3540 | } |
3541 | ||
59b6f873 | 3542 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3543 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3544 | { |
9490ff27 | 3545 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3546 | |
9490ff27 KH |
3547 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3548 | if (!event) | |
3549 | return -ENOMEM; | |
3550 | ||
1af8efe9 | 3551 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3552 | |
3553 | event->eventfd = eventfd; | |
3554 | list_add(&event->list, &memcg->oom_notify); | |
3555 | ||
3556 | /* already in OOM ? */ | |
c2b42d3c | 3557 | if (memcg->under_oom) |
9490ff27 | 3558 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3559 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3560 | |
3561 | return 0; | |
3562 | } | |
3563 | ||
59b6f873 | 3564 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3565 | struct eventfd_ctx *eventfd) |
9490ff27 | 3566 | { |
9490ff27 | 3567 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3568 | |
1af8efe9 | 3569 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3570 | |
c0ff4b85 | 3571 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3572 | if (ev->eventfd == eventfd) { |
3573 | list_del(&ev->list); | |
3574 | kfree(ev); | |
3575 | } | |
3576 | } | |
3577 | ||
1af8efe9 | 3578 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3579 | } |
3580 | ||
2da8ca82 | 3581 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3582 | { |
2da8ca82 | 3583 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3584 | |
791badbd | 3585 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3586 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3587 | return 0; |
3588 | } | |
3589 | ||
182446d0 | 3590 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3591 | struct cftype *cft, u64 val) |
3592 | { | |
182446d0 | 3593 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3594 | |
3595 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3596 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3597 | return -EINVAL; |
3598 | ||
c0ff4b85 | 3599 | memcg->oom_kill_disable = val; |
4d845ebf | 3600 | if (!val) |
c0ff4b85 | 3601 | memcg_oom_recover(memcg); |
3dae7fec | 3602 | |
3c11ecf4 KH |
3603 | return 0; |
3604 | } | |
3605 | ||
52ebea74 TH |
3606 | #ifdef CONFIG_CGROUP_WRITEBACK |
3607 | ||
3608 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3609 | { | |
3610 | return &memcg->cgwb_list; | |
3611 | } | |
3612 | ||
841710aa TH |
3613 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3614 | { | |
3615 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3616 | } | |
3617 | ||
3618 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3619 | { | |
3620 | wb_domain_exit(&memcg->cgwb_domain); | |
3621 | } | |
3622 | ||
2529bb3a TH |
3623 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3624 | { | |
3625 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3626 | } | |
3627 | ||
841710aa TH |
3628 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3629 | { | |
3630 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3631 | ||
3632 | if (!memcg->css.parent) | |
3633 | return NULL; | |
3634 | ||
3635 | return &memcg->cgwb_domain; | |
3636 | } | |
3637 | ||
c2aa723a TH |
3638 | /** |
3639 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3640 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3641 | * @pfilepages: out parameter for number of file pages |
3642 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3643 | * @pdirty: out parameter for number of dirty pages |
3644 | * @pwriteback: out parameter for number of pages under writeback | |
3645 | * | |
c5edf9cd TH |
3646 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3647 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3648 | * is a bit more involved. | |
c2aa723a | 3649 | * |
c5edf9cd TH |
3650 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3651 | * headroom is calculated as the lowest headroom of itself and the | |
3652 | * ancestors. Note that this doesn't consider the actual amount of | |
3653 | * available memory in the system. The caller should further cap | |
3654 | * *@pheadroom accordingly. | |
c2aa723a | 3655 | */ |
c5edf9cd TH |
3656 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3657 | unsigned long *pheadroom, unsigned long *pdirty, | |
3658 | unsigned long *pwriteback) | |
c2aa723a TH |
3659 | { |
3660 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3661 | struct mem_cgroup *parent; | |
c2aa723a TH |
3662 | |
3663 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3664 | ||
3665 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3666 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3667 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3668 | (1 << LRU_ACTIVE_FILE)); | |
3669 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3670 | |
c2aa723a TH |
3671 | while ((parent = parent_mem_cgroup(memcg))) { |
3672 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3673 | unsigned long used = page_counter_read(&memcg->memory); | |
3674 | ||
c5edf9cd | 3675 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3676 | memcg = parent; |
3677 | } | |
c2aa723a TH |
3678 | } |
3679 | ||
841710aa TH |
3680 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3681 | ||
3682 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3683 | { | |
3684 | return 0; | |
3685 | } | |
3686 | ||
3687 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3688 | { | |
3689 | } | |
3690 | ||
2529bb3a TH |
3691 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3692 | { | |
3693 | } | |
3694 | ||
52ebea74 TH |
3695 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3696 | ||
3bc942f3 TH |
3697 | /* |
3698 | * DO NOT USE IN NEW FILES. | |
3699 | * | |
3700 | * "cgroup.event_control" implementation. | |
3701 | * | |
3702 | * This is way over-engineered. It tries to support fully configurable | |
3703 | * events for each user. Such level of flexibility is completely | |
3704 | * unnecessary especially in the light of the planned unified hierarchy. | |
3705 | * | |
3706 | * Please deprecate this and replace with something simpler if at all | |
3707 | * possible. | |
3708 | */ | |
3709 | ||
79bd9814 TH |
3710 | /* |
3711 | * Unregister event and free resources. | |
3712 | * | |
3713 | * Gets called from workqueue. | |
3714 | */ | |
3bc942f3 | 3715 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3716 | { |
3bc942f3 TH |
3717 | struct mem_cgroup_event *event = |
3718 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3719 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3720 | |
3721 | remove_wait_queue(event->wqh, &event->wait); | |
3722 | ||
59b6f873 | 3723 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3724 | |
3725 | /* Notify userspace the event is going away. */ | |
3726 | eventfd_signal(event->eventfd, 1); | |
3727 | ||
3728 | eventfd_ctx_put(event->eventfd); | |
3729 | kfree(event); | |
59b6f873 | 3730 | css_put(&memcg->css); |
79bd9814 TH |
3731 | } |
3732 | ||
3733 | /* | |
3734 | * Gets called on POLLHUP on eventfd when user closes it. | |
3735 | * | |
3736 | * Called with wqh->lock held and interrupts disabled. | |
3737 | */ | |
3bc942f3 TH |
3738 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3739 | int sync, void *key) | |
79bd9814 | 3740 | { |
3bc942f3 TH |
3741 | struct mem_cgroup_event *event = |
3742 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3743 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3744 | unsigned long flags = (unsigned long)key; |
3745 | ||
3746 | if (flags & POLLHUP) { | |
3747 | /* | |
3748 | * If the event has been detached at cgroup removal, we | |
3749 | * can simply return knowing the other side will cleanup | |
3750 | * for us. | |
3751 | * | |
3752 | * We can't race against event freeing since the other | |
3753 | * side will require wqh->lock via remove_wait_queue(), | |
3754 | * which we hold. | |
3755 | */ | |
fba94807 | 3756 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3757 | if (!list_empty(&event->list)) { |
3758 | list_del_init(&event->list); | |
3759 | /* | |
3760 | * We are in atomic context, but cgroup_event_remove() | |
3761 | * may sleep, so we have to call it in workqueue. | |
3762 | */ | |
3763 | schedule_work(&event->remove); | |
3764 | } | |
fba94807 | 3765 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3766 | } |
3767 | ||
3768 | return 0; | |
3769 | } | |
3770 | ||
3bc942f3 | 3771 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3772 | wait_queue_head_t *wqh, poll_table *pt) |
3773 | { | |
3bc942f3 TH |
3774 | struct mem_cgroup_event *event = |
3775 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3776 | |
3777 | event->wqh = wqh; | |
3778 | add_wait_queue(wqh, &event->wait); | |
3779 | } | |
3780 | ||
3781 | /* | |
3bc942f3 TH |
3782 | * DO NOT USE IN NEW FILES. |
3783 | * | |
79bd9814 TH |
3784 | * Parse input and register new cgroup event handler. |
3785 | * | |
3786 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3787 | * Interpretation of args is defined by control file implementation. | |
3788 | */ | |
451af504 TH |
3789 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3790 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3791 | { |
451af504 | 3792 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3793 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3794 | struct mem_cgroup_event *event; |
79bd9814 TH |
3795 | struct cgroup_subsys_state *cfile_css; |
3796 | unsigned int efd, cfd; | |
3797 | struct fd efile; | |
3798 | struct fd cfile; | |
fba94807 | 3799 | const char *name; |
79bd9814 TH |
3800 | char *endp; |
3801 | int ret; | |
3802 | ||
451af504 TH |
3803 | buf = strstrip(buf); |
3804 | ||
3805 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3806 | if (*endp != ' ') |
3807 | return -EINVAL; | |
451af504 | 3808 | buf = endp + 1; |
79bd9814 | 3809 | |
451af504 | 3810 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3811 | if ((*endp != ' ') && (*endp != '\0')) |
3812 | return -EINVAL; | |
451af504 | 3813 | buf = endp + 1; |
79bd9814 TH |
3814 | |
3815 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3816 | if (!event) | |
3817 | return -ENOMEM; | |
3818 | ||
59b6f873 | 3819 | event->memcg = memcg; |
79bd9814 | 3820 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3821 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3822 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3823 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3824 | |
3825 | efile = fdget(efd); | |
3826 | if (!efile.file) { | |
3827 | ret = -EBADF; | |
3828 | goto out_kfree; | |
3829 | } | |
3830 | ||
3831 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3832 | if (IS_ERR(event->eventfd)) { | |
3833 | ret = PTR_ERR(event->eventfd); | |
3834 | goto out_put_efile; | |
3835 | } | |
3836 | ||
3837 | cfile = fdget(cfd); | |
3838 | if (!cfile.file) { | |
3839 | ret = -EBADF; | |
3840 | goto out_put_eventfd; | |
3841 | } | |
3842 | ||
3843 | /* the process need read permission on control file */ | |
3844 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3845 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3846 | if (ret < 0) | |
3847 | goto out_put_cfile; | |
3848 | ||
fba94807 TH |
3849 | /* |
3850 | * Determine the event callbacks and set them in @event. This used | |
3851 | * to be done via struct cftype but cgroup core no longer knows | |
3852 | * about these events. The following is crude but the whole thing | |
3853 | * is for compatibility anyway. | |
3bc942f3 TH |
3854 | * |
3855 | * DO NOT ADD NEW FILES. | |
fba94807 | 3856 | */ |
b583043e | 3857 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3858 | |
3859 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3860 | event->register_event = mem_cgroup_usage_register_event; | |
3861 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3862 | } else if (!strcmp(name, "memory.oom_control")) { | |
3863 | event->register_event = mem_cgroup_oom_register_event; | |
3864 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3865 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3866 | event->register_event = vmpressure_register_event; | |
3867 | event->unregister_event = vmpressure_unregister_event; | |
3868 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3869 | event->register_event = memsw_cgroup_usage_register_event; |
3870 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3871 | } else { |
3872 | ret = -EINVAL; | |
3873 | goto out_put_cfile; | |
3874 | } | |
3875 | ||
79bd9814 | 3876 | /* |
b5557c4c TH |
3877 | * Verify @cfile should belong to @css. Also, remaining events are |
3878 | * automatically removed on cgroup destruction but the removal is | |
3879 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3880 | */ |
b583043e | 3881 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3882 | &memory_cgrp_subsys); |
79bd9814 | 3883 | ret = -EINVAL; |
5a17f543 | 3884 | if (IS_ERR(cfile_css)) |
79bd9814 | 3885 | goto out_put_cfile; |
5a17f543 TH |
3886 | if (cfile_css != css) { |
3887 | css_put(cfile_css); | |
79bd9814 | 3888 | goto out_put_cfile; |
5a17f543 | 3889 | } |
79bd9814 | 3890 | |
451af504 | 3891 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3892 | if (ret) |
3893 | goto out_put_css; | |
3894 | ||
3895 | efile.file->f_op->poll(efile.file, &event->pt); | |
3896 | ||
fba94807 TH |
3897 | spin_lock(&memcg->event_list_lock); |
3898 | list_add(&event->list, &memcg->event_list); | |
3899 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3900 | |
3901 | fdput(cfile); | |
3902 | fdput(efile); | |
3903 | ||
451af504 | 3904 | return nbytes; |
79bd9814 TH |
3905 | |
3906 | out_put_css: | |
b5557c4c | 3907 | css_put(css); |
79bd9814 TH |
3908 | out_put_cfile: |
3909 | fdput(cfile); | |
3910 | out_put_eventfd: | |
3911 | eventfd_ctx_put(event->eventfd); | |
3912 | out_put_efile: | |
3913 | fdput(efile); | |
3914 | out_kfree: | |
3915 | kfree(event); | |
3916 | ||
3917 | return ret; | |
3918 | } | |
3919 | ||
241994ed | 3920 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 3921 | { |
0eea1030 | 3922 | .name = "usage_in_bytes", |
8c7c6e34 | 3923 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 3924 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3925 | }, |
c84872e1 PE |
3926 | { |
3927 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3928 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 3929 | .write = mem_cgroup_reset, |
791badbd | 3930 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 3931 | }, |
8cdea7c0 | 3932 | { |
0eea1030 | 3933 | .name = "limit_in_bytes", |
8c7c6e34 | 3934 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 3935 | .write = mem_cgroup_write, |
791badbd | 3936 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3937 | }, |
296c81d8 BS |
3938 | { |
3939 | .name = "soft_limit_in_bytes", | |
3940 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 3941 | .write = mem_cgroup_write, |
791badbd | 3942 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 3943 | }, |
8cdea7c0 BS |
3944 | { |
3945 | .name = "failcnt", | |
8c7c6e34 | 3946 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 3947 | .write = mem_cgroup_reset, |
791badbd | 3948 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3949 | }, |
d2ceb9b7 KH |
3950 | { |
3951 | .name = "stat", | |
2da8ca82 | 3952 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 3953 | }, |
c1e862c1 KH |
3954 | { |
3955 | .name = "force_empty", | |
6770c64e | 3956 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 3957 | }, |
18f59ea7 BS |
3958 | { |
3959 | .name = "use_hierarchy", | |
3960 | .write_u64 = mem_cgroup_hierarchy_write, | |
3961 | .read_u64 = mem_cgroup_hierarchy_read, | |
3962 | }, | |
79bd9814 | 3963 | { |
3bc942f3 | 3964 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 3965 | .write = memcg_write_event_control, |
7dbdb199 | 3966 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 3967 | }, |
a7885eb8 KM |
3968 | { |
3969 | .name = "swappiness", | |
3970 | .read_u64 = mem_cgroup_swappiness_read, | |
3971 | .write_u64 = mem_cgroup_swappiness_write, | |
3972 | }, | |
7dc74be0 DN |
3973 | { |
3974 | .name = "move_charge_at_immigrate", | |
3975 | .read_u64 = mem_cgroup_move_charge_read, | |
3976 | .write_u64 = mem_cgroup_move_charge_write, | |
3977 | }, | |
9490ff27 KH |
3978 | { |
3979 | .name = "oom_control", | |
2da8ca82 | 3980 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 3981 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
3982 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
3983 | }, | |
70ddf637 AV |
3984 | { |
3985 | .name = "pressure_level", | |
70ddf637 | 3986 | }, |
406eb0c9 YH |
3987 | #ifdef CONFIG_NUMA |
3988 | { | |
3989 | .name = "numa_stat", | |
2da8ca82 | 3990 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
3991 | }, |
3992 | #endif | |
510fc4e1 GC |
3993 | { |
3994 | .name = "kmem.limit_in_bytes", | |
3995 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 3996 | .write = mem_cgroup_write, |
791badbd | 3997 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
3998 | }, |
3999 | { | |
4000 | .name = "kmem.usage_in_bytes", | |
4001 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4002 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4003 | }, |
4004 | { | |
4005 | .name = "kmem.failcnt", | |
4006 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4007 | .write = mem_cgroup_reset, |
791badbd | 4008 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4009 | }, |
4010 | { | |
4011 | .name = "kmem.max_usage_in_bytes", | |
4012 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4013 | .write = mem_cgroup_reset, |
791badbd | 4014 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4015 | }, |
749c5415 GC |
4016 | #ifdef CONFIG_SLABINFO |
4017 | { | |
4018 | .name = "kmem.slabinfo", | |
b047501c VD |
4019 | .seq_start = slab_start, |
4020 | .seq_next = slab_next, | |
4021 | .seq_stop = slab_stop, | |
4022 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4023 | }, |
4024 | #endif | |
d55f90bf VD |
4025 | { |
4026 | .name = "kmem.tcp.limit_in_bytes", | |
4027 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4028 | .write = mem_cgroup_write, | |
4029 | .read_u64 = mem_cgroup_read_u64, | |
4030 | }, | |
4031 | { | |
4032 | .name = "kmem.tcp.usage_in_bytes", | |
4033 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4034 | .read_u64 = mem_cgroup_read_u64, | |
4035 | }, | |
4036 | { | |
4037 | .name = "kmem.tcp.failcnt", | |
4038 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4039 | .write = mem_cgroup_reset, | |
4040 | .read_u64 = mem_cgroup_read_u64, | |
4041 | }, | |
4042 | { | |
4043 | .name = "kmem.tcp.max_usage_in_bytes", | |
4044 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4045 | .write = mem_cgroup_reset, | |
4046 | .read_u64 = mem_cgroup_read_u64, | |
4047 | }, | |
6bc10349 | 4048 | { }, /* terminate */ |
af36f906 | 4049 | }; |
8c7c6e34 | 4050 | |
c0ff4b85 | 4051 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4052 | { |
4053 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4054 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4055 | int zone, tmp = node; |
1ecaab2b KH |
4056 | /* |
4057 | * This routine is called against possible nodes. | |
4058 | * But it's BUG to call kmalloc() against offline node. | |
4059 | * | |
4060 | * TODO: this routine can waste much memory for nodes which will | |
4061 | * never be onlined. It's better to use memory hotplug callback | |
4062 | * function. | |
4063 | */ | |
41e3355d KH |
4064 | if (!node_state(node, N_NORMAL_MEMORY)) |
4065 | tmp = -1; | |
17295c88 | 4066 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4067 | if (!pn) |
4068 | return 1; | |
1ecaab2b | 4069 | |
1ecaab2b KH |
4070 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4071 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4072 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4073 | mz->usage_in_excess = 0; |
4074 | mz->on_tree = false; | |
d79154bb | 4075 | mz->memcg = memcg; |
1ecaab2b | 4076 | } |
54f72fe0 | 4077 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4078 | return 0; |
4079 | } | |
4080 | ||
c0ff4b85 | 4081 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4082 | { |
54f72fe0 | 4083 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4084 | } |
4085 | ||
0b8f73e1 | 4086 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4087 | { |
c8b2a36f | 4088 | int node; |
59927fb9 | 4089 | |
0b8f73e1 | 4090 | memcg_wb_domain_exit(memcg); |
c8b2a36f GC |
4091 | for_each_node(node) |
4092 | free_mem_cgroup_per_zone_info(memcg, node); | |
c8b2a36f | 4093 | free_percpu(memcg->stat); |
8ff69e2c | 4094 | kfree(memcg); |
59927fb9 | 4095 | } |
3afe36b1 | 4096 | |
0b8f73e1 | 4097 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4098 | { |
d142e3e6 | 4099 | struct mem_cgroup *memcg; |
0b8f73e1 | 4100 | size_t size; |
6d12e2d8 | 4101 | int node; |
8cdea7c0 | 4102 | |
0b8f73e1 JW |
4103 | size = sizeof(struct mem_cgroup); |
4104 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4105 | ||
4106 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4107 | if (!memcg) |
0b8f73e1 JW |
4108 | return NULL; |
4109 | ||
4110 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); | |
4111 | if (!memcg->stat) | |
4112 | goto fail; | |
78fb7466 | 4113 | |
3ed28fa1 | 4114 | for_each_node(node) |
c0ff4b85 | 4115 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
0b8f73e1 | 4116 | goto fail; |
f64c3f54 | 4117 | |
0b8f73e1 JW |
4118 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4119 | goto fail; | |
28dbc4b6 | 4120 | |
f7e1cb6e | 4121 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4122 | memcg->last_scanned_node = MAX_NUMNODES; |
4123 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4124 | mutex_init(&memcg->thresholds_lock); |
4125 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4126 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4127 | INIT_LIST_HEAD(&memcg->event_list); |
4128 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4129 | memcg->socket_pressure = jiffies; |
127424c8 | 4130 | #ifndef CONFIG_SLOB |
900a38f0 | 4131 | memcg->kmemcg_id = -1; |
900a38f0 | 4132 | #endif |
52ebea74 TH |
4133 | #ifdef CONFIG_CGROUP_WRITEBACK |
4134 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4135 | #endif | |
0b8f73e1 JW |
4136 | return memcg; |
4137 | fail: | |
4138 | mem_cgroup_free(memcg); | |
4139 | return NULL; | |
d142e3e6 GC |
4140 | } |
4141 | ||
0b8f73e1 JW |
4142 | static struct cgroup_subsys_state * __ref |
4143 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4144 | { |
0b8f73e1 JW |
4145 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4146 | struct mem_cgroup *memcg; | |
4147 | long error = -ENOMEM; | |
d142e3e6 | 4148 | |
0b8f73e1 JW |
4149 | memcg = mem_cgroup_alloc(); |
4150 | if (!memcg) | |
4151 | return ERR_PTR(error); | |
d142e3e6 | 4152 | |
0b8f73e1 JW |
4153 | memcg->high = PAGE_COUNTER_MAX; |
4154 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4155 | if (parent) { | |
4156 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4157 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4158 | } | |
4159 | if (parent && parent->use_hierarchy) { | |
4160 | memcg->use_hierarchy = true; | |
3e32cb2e | 4161 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4162 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4163 | page_counter_init(&memcg->memsw, &parent->memsw); |
4164 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4165 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4166 | } else { |
3e32cb2e | 4167 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4168 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4169 | page_counter_init(&memcg->memsw, NULL); |
4170 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4171 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4172 | /* |
4173 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4174 | * much sense so let cgroup subsystem know about this | |
4175 | * unfortunate state in our controller. | |
4176 | */ | |
d142e3e6 | 4177 | if (parent != root_mem_cgroup) |
073219e9 | 4178 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4179 | } |
d6441637 | 4180 | |
0b8f73e1 JW |
4181 | /* The following stuff does not apply to the root */ |
4182 | if (!parent) { | |
4183 | root_mem_cgroup = memcg; | |
4184 | return &memcg->css; | |
4185 | } | |
4186 | ||
b313aeee | 4187 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4188 | if (error) |
4189 | goto fail; | |
127424c8 | 4190 | |
f7e1cb6e | 4191 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4192 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4193 | |
0b8f73e1 JW |
4194 | return &memcg->css; |
4195 | fail: | |
4196 | mem_cgroup_free(memcg); | |
4197 | return NULL; | |
4198 | } | |
4199 | ||
4200 | static int | |
4201 | mem_cgroup_css_online(struct cgroup_subsys_state *css) | |
4202 | { | |
4203 | if (css->id > MEM_CGROUP_ID_MAX) | |
4204 | return -ENOSPC; | |
2f7dd7a4 JW |
4205 | |
4206 | return 0; | |
8cdea7c0 BS |
4207 | } |
4208 | ||
eb95419b | 4209 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4210 | { |
eb95419b | 4211 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4212 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4213 | |
4214 | /* | |
4215 | * Unregister events and notify userspace. | |
4216 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4217 | * directory to avoid race between userspace and kernelspace. | |
4218 | */ | |
fba94807 TH |
4219 | spin_lock(&memcg->event_list_lock); |
4220 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4221 | list_del_init(&event->list); |
4222 | schedule_work(&event->remove); | |
4223 | } | |
fba94807 | 4224 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4225 | |
567e9ab2 | 4226 | memcg_offline_kmem(memcg); |
52ebea74 | 4227 | wb_memcg_offline(memcg); |
df878fb0 KH |
4228 | } |
4229 | ||
6df38689 VD |
4230 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4231 | { | |
4232 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4233 | ||
4234 | invalidate_reclaim_iterators(memcg); | |
4235 | } | |
4236 | ||
eb95419b | 4237 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4238 | { |
eb95419b | 4239 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4240 | |
f7e1cb6e | 4241 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4242 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4243 | |
0db15298 | 4244 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4245 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4246 | |
0b8f73e1 JW |
4247 | vmpressure_cleanup(&memcg->vmpressure); |
4248 | cancel_work_sync(&memcg->high_work); | |
4249 | mem_cgroup_remove_from_trees(memcg); | |
d886f4e4 | 4250 | memcg_free_kmem(memcg); |
0b8f73e1 | 4251 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4252 | } |
4253 | ||
1ced953b TH |
4254 | /** |
4255 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4256 | * @css: the target css | |
4257 | * | |
4258 | * Reset the states of the mem_cgroup associated with @css. This is | |
4259 | * invoked when the userland requests disabling on the default hierarchy | |
4260 | * but the memcg is pinned through dependency. The memcg should stop | |
4261 | * applying policies and should revert to the vanilla state as it may be | |
4262 | * made visible again. | |
4263 | * | |
4264 | * The current implementation only resets the essential configurations. | |
4265 | * This needs to be expanded to cover all the visible parts. | |
4266 | */ | |
4267 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4268 | { | |
4269 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4270 | ||
d334c9bc VD |
4271 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
4272 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); | |
4273 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); | |
4274 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); | |
4275 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
241994ed JW |
4276 | memcg->low = 0; |
4277 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4278 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4279 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4280 | } |
4281 | ||
02491447 | 4282 | #ifdef CONFIG_MMU |
7dc74be0 | 4283 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4284 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4285 | { |
05b84301 | 4286 | int ret; |
9476db97 | 4287 | |
d0164adc MG |
4288 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4289 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4290 | if (!ret) { |
854ffa8d | 4291 | mc.precharge += count; |
854ffa8d DN |
4292 | return ret; |
4293 | } | |
9476db97 JW |
4294 | |
4295 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4296 | while (count--) { |
00501b53 | 4297 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4298 | if (ret) |
38c5d72f | 4299 | return ret; |
854ffa8d | 4300 | mc.precharge++; |
9476db97 | 4301 | cond_resched(); |
854ffa8d | 4302 | } |
9476db97 | 4303 | return 0; |
4ffef5fe DN |
4304 | } |
4305 | ||
4306 | /** | |
8d32ff84 | 4307 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4308 | * @vma: the vma the pte to be checked belongs |
4309 | * @addr: the address corresponding to the pte to be checked | |
4310 | * @ptent: the pte to be checked | |
02491447 | 4311 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4312 | * |
4313 | * Returns | |
4314 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4315 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4316 | * move charge. if @target is not NULL, the page is stored in target->page | |
4317 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4318 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4319 | * target for charge migration. if @target is not NULL, the entry is stored | |
4320 | * in target->ent. | |
4ffef5fe DN |
4321 | * |
4322 | * Called with pte lock held. | |
4323 | */ | |
4ffef5fe DN |
4324 | union mc_target { |
4325 | struct page *page; | |
02491447 | 4326 | swp_entry_t ent; |
4ffef5fe DN |
4327 | }; |
4328 | ||
4ffef5fe | 4329 | enum mc_target_type { |
8d32ff84 | 4330 | MC_TARGET_NONE = 0, |
4ffef5fe | 4331 | MC_TARGET_PAGE, |
02491447 | 4332 | MC_TARGET_SWAP, |
4ffef5fe DN |
4333 | }; |
4334 | ||
90254a65 DN |
4335 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4336 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4337 | { |
90254a65 | 4338 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4339 | |
90254a65 DN |
4340 | if (!page || !page_mapped(page)) |
4341 | return NULL; | |
4342 | if (PageAnon(page)) { | |
1dfab5ab | 4343 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4344 | return NULL; |
1dfab5ab JW |
4345 | } else { |
4346 | if (!(mc.flags & MOVE_FILE)) | |
4347 | return NULL; | |
4348 | } | |
90254a65 DN |
4349 | if (!get_page_unless_zero(page)) |
4350 | return NULL; | |
4351 | ||
4352 | return page; | |
4353 | } | |
4354 | ||
4b91355e | 4355 | #ifdef CONFIG_SWAP |
90254a65 DN |
4356 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4357 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4358 | { | |
90254a65 DN |
4359 | struct page *page = NULL; |
4360 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4361 | ||
1dfab5ab | 4362 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4363 | return NULL; |
4b91355e KH |
4364 | /* |
4365 | * Because lookup_swap_cache() updates some statistics counter, | |
4366 | * we call find_get_page() with swapper_space directly. | |
4367 | */ | |
33806f06 | 4368 | page = find_get_page(swap_address_space(ent), ent.val); |
7941d214 | 4369 | if (do_memsw_account()) |
90254a65 DN |
4370 | entry->val = ent.val; |
4371 | ||
4372 | return page; | |
4373 | } | |
4b91355e KH |
4374 | #else |
4375 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4376 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4377 | { | |
4378 | return NULL; | |
4379 | } | |
4380 | #endif | |
90254a65 | 4381 | |
87946a72 DN |
4382 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4383 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4384 | { | |
4385 | struct page *page = NULL; | |
87946a72 DN |
4386 | struct address_space *mapping; |
4387 | pgoff_t pgoff; | |
4388 | ||
4389 | if (!vma->vm_file) /* anonymous vma */ | |
4390 | return NULL; | |
1dfab5ab | 4391 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4392 | return NULL; |
4393 | ||
87946a72 | 4394 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4395 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4396 | |
4397 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4398 | #ifdef CONFIG_SWAP |
4399 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4400 | if (shmem_mapping(mapping)) { |
4401 | page = find_get_entry(mapping, pgoff); | |
4402 | if (radix_tree_exceptional_entry(page)) { | |
4403 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4404 | if (do_memsw_account()) |
139b6a6f JW |
4405 | *entry = swp; |
4406 | page = find_get_page(swap_address_space(swp), swp.val); | |
4407 | } | |
4408 | } else | |
4409 | page = find_get_page(mapping, pgoff); | |
4410 | #else | |
4411 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4412 | #endif |
87946a72 DN |
4413 | return page; |
4414 | } | |
4415 | ||
b1b0deab CG |
4416 | /** |
4417 | * mem_cgroup_move_account - move account of the page | |
4418 | * @page: the page | |
4419 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4420 | * @from: mem_cgroup which the page is moved from. | |
4421 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4422 | * | |
3ac808fd | 4423 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4424 | * |
4425 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4426 | * from old cgroup. | |
4427 | */ | |
4428 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4429 | bool compound, |
b1b0deab CG |
4430 | struct mem_cgroup *from, |
4431 | struct mem_cgroup *to) | |
4432 | { | |
4433 | unsigned long flags; | |
f627c2f5 | 4434 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4435 | int ret; |
c4843a75 | 4436 | bool anon; |
b1b0deab CG |
4437 | |
4438 | VM_BUG_ON(from == to); | |
4439 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4440 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4441 | |
4442 | /* | |
6a93ca8f | 4443 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4444 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4445 | */ |
f627c2f5 | 4446 | ret = -EBUSY; |
b1b0deab CG |
4447 | if (!trylock_page(page)) |
4448 | goto out; | |
4449 | ||
4450 | ret = -EINVAL; | |
4451 | if (page->mem_cgroup != from) | |
4452 | goto out_unlock; | |
4453 | ||
c4843a75 GT |
4454 | anon = PageAnon(page); |
4455 | ||
b1b0deab CG |
4456 | spin_lock_irqsave(&from->move_lock, flags); |
4457 | ||
c4843a75 | 4458 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4459 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4460 | nr_pages); | |
4461 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4462 | nr_pages); | |
4463 | } | |
4464 | ||
c4843a75 GT |
4465 | /* |
4466 | * move_lock grabbed above and caller set from->moving_account, so | |
4467 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4468 | * So mapping should be stable for dirty pages. | |
4469 | */ | |
4470 | if (!anon && PageDirty(page)) { | |
4471 | struct address_space *mapping = page_mapping(page); | |
4472 | ||
4473 | if (mapping_cap_account_dirty(mapping)) { | |
4474 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4475 | nr_pages); | |
4476 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4477 | nr_pages); | |
4478 | } | |
4479 | } | |
4480 | ||
b1b0deab CG |
4481 | if (PageWriteback(page)) { |
4482 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4483 | nr_pages); | |
4484 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4485 | nr_pages); | |
4486 | } | |
4487 | ||
4488 | /* | |
4489 | * It is safe to change page->mem_cgroup here because the page | |
4490 | * is referenced, charged, and isolated - we can't race with | |
4491 | * uncharging, charging, migration, or LRU putback. | |
4492 | */ | |
4493 | ||
4494 | /* caller should have done css_get */ | |
4495 | page->mem_cgroup = to; | |
4496 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4497 | ||
4498 | ret = 0; | |
4499 | ||
4500 | local_irq_disable(); | |
f627c2f5 | 4501 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4502 | memcg_check_events(to, page); |
f627c2f5 | 4503 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4504 | memcg_check_events(from, page); |
4505 | local_irq_enable(); | |
4506 | out_unlock: | |
4507 | unlock_page(page); | |
4508 | out: | |
4509 | return ret; | |
4510 | } | |
4511 | ||
8d32ff84 | 4512 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4513 | unsigned long addr, pte_t ptent, union mc_target *target) |
4514 | { | |
4515 | struct page *page = NULL; | |
8d32ff84 | 4516 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4517 | swp_entry_t ent = { .val = 0 }; |
4518 | ||
4519 | if (pte_present(ptent)) | |
4520 | page = mc_handle_present_pte(vma, addr, ptent); | |
4521 | else if (is_swap_pte(ptent)) | |
4522 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4523 | else if (pte_none(ptent)) |
87946a72 | 4524 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4525 | |
4526 | if (!page && !ent.val) | |
8d32ff84 | 4527 | return ret; |
02491447 | 4528 | if (page) { |
02491447 | 4529 | /* |
0a31bc97 | 4530 | * Do only loose check w/o serialization. |
1306a85a | 4531 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4532 | * not under LRU exclusion. |
02491447 | 4533 | */ |
1306a85a | 4534 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4535 | ret = MC_TARGET_PAGE; |
4536 | if (target) | |
4537 | target->page = page; | |
4538 | } | |
4539 | if (!ret || !target) | |
4540 | put_page(page); | |
4541 | } | |
90254a65 DN |
4542 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4543 | if (ent.val && !ret && | |
34c00c31 | 4544 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4545 | ret = MC_TARGET_SWAP; |
4546 | if (target) | |
4547 | target->ent = ent; | |
4ffef5fe | 4548 | } |
4ffef5fe DN |
4549 | return ret; |
4550 | } | |
4551 | ||
12724850 NH |
4552 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4553 | /* | |
4554 | * We don't consider swapping or file mapped pages because THP does not | |
4555 | * support them for now. | |
4556 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4557 | */ | |
4558 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4559 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4560 | { | |
4561 | struct page *page = NULL; | |
12724850 NH |
4562 | enum mc_target_type ret = MC_TARGET_NONE; |
4563 | ||
4564 | page = pmd_page(pmd); | |
309381fe | 4565 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4566 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4567 | return ret; |
1306a85a | 4568 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4569 | ret = MC_TARGET_PAGE; |
4570 | if (target) { | |
4571 | get_page(page); | |
4572 | target->page = page; | |
4573 | } | |
4574 | } | |
4575 | return ret; | |
4576 | } | |
4577 | #else | |
4578 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4579 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4580 | { | |
4581 | return MC_TARGET_NONE; | |
4582 | } | |
4583 | #endif | |
4584 | ||
4ffef5fe DN |
4585 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4586 | unsigned long addr, unsigned long end, | |
4587 | struct mm_walk *walk) | |
4588 | { | |
26bcd64a | 4589 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4590 | pte_t *pte; |
4591 | spinlock_t *ptl; | |
4592 | ||
b6ec57f4 KS |
4593 | ptl = pmd_trans_huge_lock(pmd, vma); |
4594 | if (ptl) { | |
12724850 NH |
4595 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4596 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4597 | spin_unlock(ptl); |
1a5a9906 | 4598 | return 0; |
12724850 | 4599 | } |
03319327 | 4600 | |
45f83cef AA |
4601 | if (pmd_trans_unstable(pmd)) |
4602 | return 0; | |
4ffef5fe DN |
4603 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4604 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4605 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4606 | mc.precharge++; /* increment precharge temporarily */ |
4607 | pte_unmap_unlock(pte - 1, ptl); | |
4608 | cond_resched(); | |
4609 | ||
7dc74be0 DN |
4610 | return 0; |
4611 | } | |
4612 | ||
4ffef5fe DN |
4613 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4614 | { | |
4615 | unsigned long precharge; | |
4ffef5fe | 4616 | |
26bcd64a NH |
4617 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4618 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4619 | .mm = mm, | |
4620 | }; | |
dfe076b0 | 4621 | down_read(&mm->mmap_sem); |
26bcd64a | 4622 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4623 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4624 | |
4625 | precharge = mc.precharge; | |
4626 | mc.precharge = 0; | |
4627 | ||
4628 | return precharge; | |
4629 | } | |
4630 | ||
4ffef5fe DN |
4631 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4632 | { | |
dfe076b0 DN |
4633 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4634 | ||
4635 | VM_BUG_ON(mc.moving_task); | |
4636 | mc.moving_task = current; | |
4637 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4638 | } |
4639 | ||
dfe076b0 DN |
4640 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4641 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4642 | { |
2bd9bb20 KH |
4643 | struct mem_cgroup *from = mc.from; |
4644 | struct mem_cgroup *to = mc.to; | |
4645 | ||
4ffef5fe | 4646 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4647 | if (mc.precharge) { |
00501b53 | 4648 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4649 | mc.precharge = 0; |
4650 | } | |
4651 | /* | |
4652 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4653 | * we must uncharge here. | |
4654 | */ | |
4655 | if (mc.moved_charge) { | |
00501b53 | 4656 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4657 | mc.moved_charge = 0; |
4ffef5fe | 4658 | } |
483c30b5 DN |
4659 | /* we must fixup refcnts and charges */ |
4660 | if (mc.moved_swap) { | |
483c30b5 | 4661 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4662 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4663 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4664 | |
05b84301 | 4665 | /* |
3e32cb2e JW |
4666 | * we charged both to->memory and to->memsw, so we |
4667 | * should uncharge to->memory. | |
05b84301 | 4668 | */ |
ce00a967 | 4669 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4670 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4671 | ||
e8ea14cc | 4672 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4673 | |
4050377b | 4674 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4675 | mc.moved_swap = 0; |
4676 | } | |
dfe076b0 DN |
4677 | memcg_oom_recover(from); |
4678 | memcg_oom_recover(to); | |
4679 | wake_up_all(&mc.waitq); | |
4680 | } | |
4681 | ||
4682 | static void mem_cgroup_clear_mc(void) | |
4683 | { | |
264a0ae1 TH |
4684 | struct mm_struct *mm = mc.mm; |
4685 | ||
dfe076b0 DN |
4686 | /* |
4687 | * we must clear moving_task before waking up waiters at the end of | |
4688 | * task migration. | |
4689 | */ | |
4690 | mc.moving_task = NULL; | |
4691 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4692 | spin_lock(&mc.lock); |
4ffef5fe DN |
4693 | mc.from = NULL; |
4694 | mc.to = NULL; | |
264a0ae1 | 4695 | mc.mm = NULL; |
2bd9bb20 | 4696 | spin_unlock(&mc.lock); |
264a0ae1 TH |
4697 | |
4698 | mmput(mm); | |
4ffef5fe DN |
4699 | } |
4700 | ||
1f7dd3e5 | 4701 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4702 | { |
1f7dd3e5 | 4703 | struct cgroup_subsys_state *css; |
eed67d75 | 4704 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4705 | struct mem_cgroup *from; |
4530eddb | 4706 | struct task_struct *leader, *p; |
9f2115f9 | 4707 | struct mm_struct *mm; |
1dfab5ab | 4708 | unsigned long move_flags; |
9f2115f9 | 4709 | int ret = 0; |
7dc74be0 | 4710 | |
1f7dd3e5 TH |
4711 | /* charge immigration isn't supported on the default hierarchy */ |
4712 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4713 | return 0; |
4714 | ||
4530eddb TH |
4715 | /* |
4716 | * Multi-process migrations only happen on the default hierarchy | |
4717 | * where charge immigration is not used. Perform charge | |
4718 | * immigration if @tset contains a leader and whine if there are | |
4719 | * multiple. | |
4720 | */ | |
4721 | p = NULL; | |
1f7dd3e5 | 4722 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4723 | WARN_ON_ONCE(p); |
4724 | p = leader; | |
1f7dd3e5 | 4725 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4726 | } |
4727 | if (!p) | |
4728 | return 0; | |
4729 | ||
1f7dd3e5 TH |
4730 | /* |
4731 | * We are now commited to this value whatever it is. Changes in this | |
4732 | * tunable will only affect upcoming migrations, not the current one. | |
4733 | * So we need to save it, and keep it going. | |
4734 | */ | |
4735 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4736 | if (!move_flags) | |
4737 | return 0; | |
4738 | ||
9f2115f9 TH |
4739 | from = mem_cgroup_from_task(p); |
4740 | ||
4741 | VM_BUG_ON(from == memcg); | |
4742 | ||
4743 | mm = get_task_mm(p); | |
4744 | if (!mm) | |
4745 | return 0; | |
4746 | /* We move charges only when we move a owner of the mm */ | |
4747 | if (mm->owner == p) { | |
4748 | VM_BUG_ON(mc.from); | |
4749 | VM_BUG_ON(mc.to); | |
4750 | VM_BUG_ON(mc.precharge); | |
4751 | VM_BUG_ON(mc.moved_charge); | |
4752 | VM_BUG_ON(mc.moved_swap); | |
4753 | ||
4754 | spin_lock(&mc.lock); | |
264a0ae1 | 4755 | mc.mm = mm; |
9f2115f9 TH |
4756 | mc.from = from; |
4757 | mc.to = memcg; | |
4758 | mc.flags = move_flags; | |
4759 | spin_unlock(&mc.lock); | |
4760 | /* We set mc.moving_task later */ | |
4761 | ||
4762 | ret = mem_cgroup_precharge_mc(mm); | |
4763 | if (ret) | |
4764 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
4765 | } else { |
4766 | mmput(mm); | |
7dc74be0 DN |
4767 | } |
4768 | return ret; | |
4769 | } | |
4770 | ||
1f7dd3e5 | 4771 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4772 | { |
4e2f245d JW |
4773 | if (mc.to) |
4774 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4775 | } |
4776 | ||
4ffef5fe DN |
4777 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4778 | unsigned long addr, unsigned long end, | |
4779 | struct mm_walk *walk) | |
7dc74be0 | 4780 | { |
4ffef5fe | 4781 | int ret = 0; |
26bcd64a | 4782 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4783 | pte_t *pte; |
4784 | spinlock_t *ptl; | |
12724850 NH |
4785 | enum mc_target_type target_type; |
4786 | union mc_target target; | |
4787 | struct page *page; | |
4ffef5fe | 4788 | |
b6ec57f4 KS |
4789 | ptl = pmd_trans_huge_lock(pmd, vma); |
4790 | if (ptl) { | |
62ade86a | 4791 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4792 | spin_unlock(ptl); |
12724850 NH |
4793 | return 0; |
4794 | } | |
4795 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4796 | if (target_type == MC_TARGET_PAGE) { | |
4797 | page = target.page; | |
4798 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 4799 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 4800 | mc.from, mc.to)) { |
12724850 NH |
4801 | mc.precharge -= HPAGE_PMD_NR; |
4802 | mc.moved_charge += HPAGE_PMD_NR; | |
4803 | } | |
4804 | putback_lru_page(page); | |
4805 | } | |
4806 | put_page(page); | |
4807 | } | |
bf929152 | 4808 | spin_unlock(ptl); |
1a5a9906 | 4809 | return 0; |
12724850 NH |
4810 | } |
4811 | ||
45f83cef AA |
4812 | if (pmd_trans_unstable(pmd)) |
4813 | return 0; | |
4ffef5fe DN |
4814 | retry: |
4815 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4816 | for (; addr != end; addr += PAGE_SIZE) { | |
4817 | pte_t ptent = *(pte++); | |
02491447 | 4818 | swp_entry_t ent; |
4ffef5fe DN |
4819 | |
4820 | if (!mc.precharge) | |
4821 | break; | |
4822 | ||
8d32ff84 | 4823 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4824 | case MC_TARGET_PAGE: |
4825 | page = target.page; | |
53f9263b KS |
4826 | /* |
4827 | * We can have a part of the split pmd here. Moving it | |
4828 | * can be done but it would be too convoluted so simply | |
4829 | * ignore such a partial THP and keep it in original | |
4830 | * memcg. There should be somebody mapping the head. | |
4831 | */ | |
4832 | if (PageTransCompound(page)) | |
4833 | goto put; | |
4ffef5fe DN |
4834 | if (isolate_lru_page(page)) |
4835 | goto put; | |
f627c2f5 KS |
4836 | if (!mem_cgroup_move_account(page, false, |
4837 | mc.from, mc.to)) { | |
4ffef5fe | 4838 | mc.precharge--; |
854ffa8d DN |
4839 | /* we uncharge from mc.from later. */ |
4840 | mc.moved_charge++; | |
4ffef5fe DN |
4841 | } |
4842 | putback_lru_page(page); | |
8d32ff84 | 4843 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4844 | put_page(page); |
4845 | break; | |
02491447 DN |
4846 | case MC_TARGET_SWAP: |
4847 | ent = target.ent; | |
e91cbb42 | 4848 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4849 | mc.precharge--; |
483c30b5 DN |
4850 | /* we fixup refcnts and charges later. */ |
4851 | mc.moved_swap++; | |
4852 | } | |
02491447 | 4853 | break; |
4ffef5fe DN |
4854 | default: |
4855 | break; | |
4856 | } | |
4857 | } | |
4858 | pte_unmap_unlock(pte - 1, ptl); | |
4859 | cond_resched(); | |
4860 | ||
4861 | if (addr != end) { | |
4862 | /* | |
4863 | * We have consumed all precharges we got in can_attach(). | |
4864 | * We try charge one by one, but don't do any additional | |
4865 | * charges to mc.to if we have failed in charge once in attach() | |
4866 | * phase. | |
4867 | */ | |
854ffa8d | 4868 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4869 | if (!ret) |
4870 | goto retry; | |
4871 | } | |
4872 | ||
4873 | return ret; | |
4874 | } | |
4875 | ||
264a0ae1 | 4876 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 4877 | { |
26bcd64a NH |
4878 | struct mm_walk mem_cgroup_move_charge_walk = { |
4879 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 4880 | .mm = mc.mm, |
26bcd64a | 4881 | }; |
4ffef5fe DN |
4882 | |
4883 | lru_add_drain_all(); | |
312722cb | 4884 | /* |
81f8c3a4 JW |
4885 | * Signal lock_page_memcg() to take the memcg's move_lock |
4886 | * while we're moving its pages to another memcg. Then wait | |
4887 | * for already started RCU-only updates to finish. | |
312722cb JW |
4888 | */ |
4889 | atomic_inc(&mc.from->moving_account); | |
4890 | synchronize_rcu(); | |
dfe076b0 | 4891 | retry: |
264a0ae1 | 4892 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
4893 | /* |
4894 | * Someone who are holding the mmap_sem might be waiting in | |
4895 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4896 | * and retry. Because we cancel precharges, we might not be able | |
4897 | * to move enough charges, but moving charge is a best-effort | |
4898 | * feature anyway, so it wouldn't be a big problem. | |
4899 | */ | |
4900 | __mem_cgroup_clear_mc(); | |
4901 | cond_resched(); | |
4902 | goto retry; | |
4903 | } | |
26bcd64a NH |
4904 | /* |
4905 | * When we have consumed all precharges and failed in doing | |
4906 | * additional charge, the page walk just aborts. | |
4907 | */ | |
4908 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
264a0ae1 | 4909 | up_read(&mc.mm->mmap_sem); |
312722cb | 4910 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
4911 | } |
4912 | ||
264a0ae1 | 4913 | static void mem_cgroup_move_task(void) |
67e465a7 | 4914 | { |
264a0ae1 TH |
4915 | if (mc.to) { |
4916 | mem_cgroup_move_charge(); | |
a433658c | 4917 | mem_cgroup_clear_mc(); |
264a0ae1 | 4918 | } |
67e465a7 | 4919 | } |
5cfb80a7 | 4920 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 4921 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4922 | { |
4923 | return 0; | |
4924 | } | |
1f7dd3e5 | 4925 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4926 | { |
4927 | } | |
264a0ae1 | 4928 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
4929 | { |
4930 | } | |
4931 | #endif | |
67e465a7 | 4932 | |
f00baae7 TH |
4933 | /* |
4934 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
4935 | * to verify whether we're attached to the default hierarchy on each mount |
4936 | * attempt. | |
f00baae7 | 4937 | */ |
eb95419b | 4938 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
4939 | { |
4940 | /* | |
aa6ec29b | 4941 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
4942 | * guarantees that @root doesn't have any children, so turning it |
4943 | * on for the root memcg is enough. | |
4944 | */ | |
9e10a130 | 4945 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
4946 | root_mem_cgroup->use_hierarchy = true; |
4947 | else | |
4948 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
4949 | } |
4950 | ||
241994ed JW |
4951 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
4952 | struct cftype *cft) | |
4953 | { | |
f5fc3c5d JW |
4954 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4955 | ||
4956 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
4957 | } |
4958 | ||
4959 | static int memory_low_show(struct seq_file *m, void *v) | |
4960 | { | |
4961 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 4962 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
4963 | |
4964 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 4965 | seq_puts(m, "max\n"); |
241994ed JW |
4966 | else |
4967 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
4968 | ||
4969 | return 0; | |
4970 | } | |
4971 | ||
4972 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
4973 | char *buf, size_t nbytes, loff_t off) | |
4974 | { | |
4975 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
4976 | unsigned long low; | |
4977 | int err; | |
4978 | ||
4979 | buf = strstrip(buf); | |
d2973697 | 4980 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
4981 | if (err) |
4982 | return err; | |
4983 | ||
4984 | memcg->low = low; | |
4985 | ||
4986 | return nbytes; | |
4987 | } | |
4988 | ||
4989 | static int memory_high_show(struct seq_file *m, void *v) | |
4990 | { | |
4991 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 4992 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
4993 | |
4994 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 4995 | seq_puts(m, "max\n"); |
241994ed JW |
4996 | else |
4997 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
4998 | ||
4999 | return 0; | |
5000 | } | |
5001 | ||
5002 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5003 | char *buf, size_t nbytes, loff_t off) | |
5004 | { | |
5005 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5006 | unsigned long nr_pages; |
241994ed JW |
5007 | unsigned long high; |
5008 | int err; | |
5009 | ||
5010 | buf = strstrip(buf); | |
d2973697 | 5011 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5012 | if (err) |
5013 | return err; | |
5014 | ||
5015 | memcg->high = high; | |
5016 | ||
588083bb JW |
5017 | nr_pages = page_counter_read(&memcg->memory); |
5018 | if (nr_pages > high) | |
5019 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5020 | GFP_KERNEL, true); | |
5021 | ||
2529bb3a | 5022 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5023 | return nbytes; |
5024 | } | |
5025 | ||
5026 | static int memory_max_show(struct seq_file *m, void *v) | |
5027 | { | |
5028 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5029 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5030 | |
5031 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5032 | seq_puts(m, "max\n"); |
241994ed JW |
5033 | else |
5034 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5035 | ||
5036 | return 0; | |
5037 | } | |
5038 | ||
5039 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5040 | char *buf, size_t nbytes, loff_t off) | |
5041 | { | |
5042 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5043 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5044 | bool drained = false; | |
241994ed JW |
5045 | unsigned long max; |
5046 | int err; | |
5047 | ||
5048 | buf = strstrip(buf); | |
d2973697 | 5049 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5050 | if (err) |
5051 | return err; | |
5052 | ||
b6e6edcf JW |
5053 | xchg(&memcg->memory.limit, max); |
5054 | ||
5055 | for (;;) { | |
5056 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5057 | ||
5058 | if (nr_pages <= max) | |
5059 | break; | |
5060 | ||
5061 | if (signal_pending(current)) { | |
5062 | err = -EINTR; | |
5063 | break; | |
5064 | } | |
5065 | ||
5066 | if (!drained) { | |
5067 | drain_all_stock(memcg); | |
5068 | drained = true; | |
5069 | continue; | |
5070 | } | |
5071 | ||
5072 | if (nr_reclaims) { | |
5073 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5074 | GFP_KERNEL, true)) | |
5075 | nr_reclaims--; | |
5076 | continue; | |
5077 | } | |
5078 | ||
5079 | mem_cgroup_events(memcg, MEMCG_OOM, 1); | |
5080 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) | |
5081 | break; | |
5082 | } | |
241994ed | 5083 | |
2529bb3a | 5084 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5085 | return nbytes; |
5086 | } | |
5087 | ||
5088 | static int memory_events_show(struct seq_file *m, void *v) | |
5089 | { | |
5090 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5091 | ||
5092 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5093 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5094 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5095 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5096 | ||
5097 | return 0; | |
5098 | } | |
5099 | ||
587d9f72 JW |
5100 | static int memory_stat_show(struct seq_file *m, void *v) |
5101 | { | |
5102 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
72b54e73 VD |
5103 | unsigned long stat[MEMCG_NR_STAT]; |
5104 | unsigned long events[MEMCG_NR_EVENTS]; | |
587d9f72 JW |
5105 | int i; |
5106 | ||
5107 | /* | |
5108 | * Provide statistics on the state of the memory subsystem as | |
5109 | * well as cumulative event counters that show past behavior. | |
5110 | * | |
5111 | * This list is ordered following a combination of these gradients: | |
5112 | * 1) generic big picture -> specifics and details | |
5113 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5114 | * | |
5115 | * Current memory state: | |
5116 | */ | |
5117 | ||
72b54e73 VD |
5118 | tree_stat(memcg, stat); |
5119 | tree_events(memcg, events); | |
5120 | ||
587d9f72 | 5121 | seq_printf(m, "anon %llu\n", |
72b54e73 | 5122 | (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE); |
587d9f72 | 5123 | seq_printf(m, "file %llu\n", |
72b54e73 | 5124 | (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE); |
12580e4b VD |
5125 | seq_printf(m, "kernel_stack %llu\n", |
5126 | (u64)stat[MEMCG_KERNEL_STACK] * PAGE_SIZE); | |
27ee57c9 VD |
5127 | seq_printf(m, "slab %llu\n", |
5128 | (u64)(stat[MEMCG_SLAB_RECLAIMABLE] + | |
5129 | stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5130 | seq_printf(m, "sock %llu\n", |
72b54e73 | 5131 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 JW |
5132 | |
5133 | seq_printf(m, "file_mapped %llu\n", | |
72b54e73 | 5134 | (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5135 | seq_printf(m, "file_dirty %llu\n", |
72b54e73 | 5136 | (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE); |
587d9f72 | 5137 | seq_printf(m, "file_writeback %llu\n", |
72b54e73 | 5138 | (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE); |
587d9f72 JW |
5139 | |
5140 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5141 | struct mem_cgroup *mi; | |
5142 | unsigned long val = 0; | |
5143 | ||
5144 | for_each_mem_cgroup_tree(mi, memcg) | |
5145 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
5146 | seq_printf(m, "%s %llu\n", | |
5147 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); | |
5148 | } | |
5149 | ||
27ee57c9 VD |
5150 | seq_printf(m, "slab_reclaimable %llu\n", |
5151 | (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE); | |
5152 | seq_printf(m, "slab_unreclaimable %llu\n", | |
5153 | (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE); | |
5154 | ||
587d9f72 JW |
5155 | /* Accumulated memory events */ |
5156 | ||
5157 | seq_printf(m, "pgfault %lu\n", | |
72b54e73 | 5158 | events[MEM_CGROUP_EVENTS_PGFAULT]); |
587d9f72 | 5159 | seq_printf(m, "pgmajfault %lu\n", |
72b54e73 | 5160 | events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
587d9f72 JW |
5161 | |
5162 | return 0; | |
5163 | } | |
5164 | ||
241994ed JW |
5165 | static struct cftype memory_files[] = { |
5166 | { | |
5167 | .name = "current", | |
f5fc3c5d | 5168 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5169 | .read_u64 = memory_current_read, |
5170 | }, | |
5171 | { | |
5172 | .name = "low", | |
5173 | .flags = CFTYPE_NOT_ON_ROOT, | |
5174 | .seq_show = memory_low_show, | |
5175 | .write = memory_low_write, | |
5176 | }, | |
5177 | { | |
5178 | .name = "high", | |
5179 | .flags = CFTYPE_NOT_ON_ROOT, | |
5180 | .seq_show = memory_high_show, | |
5181 | .write = memory_high_write, | |
5182 | }, | |
5183 | { | |
5184 | .name = "max", | |
5185 | .flags = CFTYPE_NOT_ON_ROOT, | |
5186 | .seq_show = memory_max_show, | |
5187 | .write = memory_max_write, | |
5188 | }, | |
5189 | { | |
5190 | .name = "events", | |
5191 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5192 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5193 | .seq_show = memory_events_show, |
5194 | }, | |
587d9f72 JW |
5195 | { |
5196 | .name = "stat", | |
5197 | .flags = CFTYPE_NOT_ON_ROOT, | |
5198 | .seq_show = memory_stat_show, | |
5199 | }, | |
241994ed JW |
5200 | { } /* terminate */ |
5201 | }; | |
5202 | ||
073219e9 | 5203 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5204 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5205 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5206 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5207 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5208 | .css_free = mem_cgroup_css_free, |
1ced953b | 5209 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5210 | .can_attach = mem_cgroup_can_attach, |
5211 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5212 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5213 | .bind = mem_cgroup_bind, |
241994ed JW |
5214 | .dfl_cftypes = memory_files, |
5215 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5216 | .early_init = 0, |
8cdea7c0 | 5217 | }; |
c077719b | 5218 | |
241994ed JW |
5219 | /** |
5220 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5221 | * @root: the highest ancestor to consider | |
5222 | * @memcg: the memory cgroup to check | |
5223 | * | |
5224 | * Returns %true if memory consumption of @memcg, and that of all | |
5225 | * configurable ancestors up to @root, is below the normal range. | |
5226 | */ | |
5227 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5228 | { | |
5229 | if (mem_cgroup_disabled()) | |
5230 | return false; | |
5231 | ||
5232 | /* | |
5233 | * The toplevel group doesn't have a configurable range, so | |
5234 | * it's never low when looked at directly, and it is not | |
5235 | * considered an ancestor when assessing the hierarchy. | |
5236 | */ | |
5237 | ||
5238 | if (memcg == root_mem_cgroup) | |
5239 | return false; | |
5240 | ||
4e54dede | 5241 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5242 | return false; |
5243 | ||
5244 | while (memcg != root) { | |
5245 | memcg = parent_mem_cgroup(memcg); | |
5246 | ||
5247 | if (memcg == root_mem_cgroup) | |
5248 | break; | |
5249 | ||
4e54dede | 5250 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5251 | return false; |
5252 | } | |
5253 | return true; | |
5254 | } | |
5255 | ||
00501b53 JW |
5256 | /** |
5257 | * mem_cgroup_try_charge - try charging a page | |
5258 | * @page: page to charge | |
5259 | * @mm: mm context of the victim | |
5260 | * @gfp_mask: reclaim mode | |
5261 | * @memcgp: charged memcg return | |
5262 | * | |
5263 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5264 | * pages according to @gfp_mask if necessary. | |
5265 | * | |
5266 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5267 | * Otherwise, an error code is returned. | |
5268 | * | |
5269 | * After page->mapping has been set up, the caller must finalize the | |
5270 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5271 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5272 | */ | |
5273 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5274 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5275 | bool compound) | |
00501b53 JW |
5276 | { |
5277 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5278 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5279 | int ret = 0; |
5280 | ||
5281 | if (mem_cgroup_disabled()) | |
5282 | goto out; | |
5283 | ||
5284 | if (PageSwapCache(page)) { | |
00501b53 JW |
5285 | /* |
5286 | * Every swap fault against a single page tries to charge the | |
5287 | * page, bail as early as possible. shmem_unuse() encounters | |
5288 | * already charged pages, too. The USED bit is protected by | |
5289 | * the page lock, which serializes swap cache removal, which | |
5290 | * in turn serializes uncharging. | |
5291 | */ | |
e993d905 | 5292 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5293 | if (page->mem_cgroup) |
00501b53 | 5294 | goto out; |
e993d905 | 5295 | |
37e84351 | 5296 | if (do_swap_account) { |
e993d905 VD |
5297 | swp_entry_t ent = { .val = page_private(page), }; |
5298 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5299 | ||
5300 | rcu_read_lock(); | |
5301 | memcg = mem_cgroup_from_id(id); | |
5302 | if (memcg && !css_tryget_online(&memcg->css)) | |
5303 | memcg = NULL; | |
5304 | rcu_read_unlock(); | |
5305 | } | |
00501b53 JW |
5306 | } |
5307 | ||
00501b53 JW |
5308 | if (!memcg) |
5309 | memcg = get_mem_cgroup_from_mm(mm); | |
5310 | ||
5311 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5312 | ||
5313 | css_put(&memcg->css); | |
00501b53 JW |
5314 | out: |
5315 | *memcgp = memcg; | |
5316 | return ret; | |
5317 | } | |
5318 | ||
5319 | /** | |
5320 | * mem_cgroup_commit_charge - commit a page charge | |
5321 | * @page: page to charge | |
5322 | * @memcg: memcg to charge the page to | |
5323 | * @lrucare: page might be on LRU already | |
5324 | * | |
5325 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5326 | * after page->mapping has been set up. This must happen atomically | |
5327 | * as part of the page instantiation, i.e. under the page table lock | |
5328 | * for anonymous pages, under the page lock for page and swap cache. | |
5329 | * | |
5330 | * In addition, the page must not be on the LRU during the commit, to | |
5331 | * prevent racing with task migration. If it might be, use @lrucare. | |
5332 | * | |
5333 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5334 | */ | |
5335 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5336 | bool lrucare, bool compound) |
00501b53 | 5337 | { |
f627c2f5 | 5338 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5339 | |
5340 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5341 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5342 | ||
5343 | if (mem_cgroup_disabled()) | |
5344 | return; | |
5345 | /* | |
5346 | * Swap faults will attempt to charge the same page multiple | |
5347 | * times. But reuse_swap_page() might have removed the page | |
5348 | * from swapcache already, so we can't check PageSwapCache(). | |
5349 | */ | |
5350 | if (!memcg) | |
5351 | return; | |
5352 | ||
6abb5a86 JW |
5353 | commit_charge(page, memcg, lrucare); |
5354 | ||
6abb5a86 | 5355 | local_irq_disable(); |
f627c2f5 | 5356 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5357 | memcg_check_events(memcg, page); |
5358 | local_irq_enable(); | |
00501b53 | 5359 | |
7941d214 | 5360 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5361 | swp_entry_t entry = { .val = page_private(page) }; |
5362 | /* | |
5363 | * The swap entry might not get freed for a long time, | |
5364 | * let's not wait for it. The page already received a | |
5365 | * memory+swap charge, drop the swap entry duplicate. | |
5366 | */ | |
5367 | mem_cgroup_uncharge_swap(entry); | |
5368 | } | |
5369 | } | |
5370 | ||
5371 | /** | |
5372 | * mem_cgroup_cancel_charge - cancel a page charge | |
5373 | * @page: page to charge | |
5374 | * @memcg: memcg to charge the page to | |
5375 | * | |
5376 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5377 | */ | |
f627c2f5 KS |
5378 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5379 | bool compound) | |
00501b53 | 5380 | { |
f627c2f5 | 5381 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5382 | |
5383 | if (mem_cgroup_disabled()) | |
5384 | return; | |
5385 | /* | |
5386 | * Swap faults will attempt to charge the same page multiple | |
5387 | * times. But reuse_swap_page() might have removed the page | |
5388 | * from swapcache already, so we can't check PageSwapCache(). | |
5389 | */ | |
5390 | if (!memcg) | |
5391 | return; | |
5392 | ||
00501b53 JW |
5393 | cancel_charge(memcg, nr_pages); |
5394 | } | |
5395 | ||
747db954 | 5396 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5397 | unsigned long nr_anon, unsigned long nr_file, |
5398 | unsigned long nr_huge, struct page *dummy_page) | |
5399 | { | |
18eca2e6 | 5400 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5401 | unsigned long flags; |
5402 | ||
ce00a967 | 5403 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 | 5404 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 5405 | if (do_memsw_account()) |
18eca2e6 | 5406 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 JW |
5407 | memcg_oom_recover(memcg); |
5408 | } | |
747db954 JW |
5409 | |
5410 | local_irq_save(flags); | |
5411 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5412 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5413 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5414 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5415 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5416 | memcg_check_events(memcg, dummy_page); |
5417 | local_irq_restore(flags); | |
e8ea14cc JW |
5418 | |
5419 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5420 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5421 | } |
5422 | ||
5423 | static void uncharge_list(struct list_head *page_list) | |
5424 | { | |
5425 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5426 | unsigned long nr_anon = 0; |
5427 | unsigned long nr_file = 0; | |
5428 | unsigned long nr_huge = 0; | |
5429 | unsigned long pgpgout = 0; | |
747db954 JW |
5430 | struct list_head *next; |
5431 | struct page *page; | |
5432 | ||
8b592656 JW |
5433 | /* |
5434 | * Note that the list can be a single page->lru; hence the | |
5435 | * do-while loop instead of a simple list_for_each_entry(). | |
5436 | */ | |
747db954 JW |
5437 | next = page_list->next; |
5438 | do { | |
5439 | unsigned int nr_pages = 1; | |
747db954 JW |
5440 | |
5441 | page = list_entry(next, struct page, lru); | |
5442 | next = page->lru.next; | |
5443 | ||
5444 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5445 | VM_BUG_ON_PAGE(page_count(page), page); | |
5446 | ||
1306a85a | 5447 | if (!page->mem_cgroup) |
747db954 JW |
5448 | continue; |
5449 | ||
5450 | /* | |
5451 | * Nobody should be changing or seriously looking at | |
1306a85a | 5452 | * page->mem_cgroup at this point, we have fully |
29833315 | 5453 | * exclusive access to the page. |
747db954 JW |
5454 | */ |
5455 | ||
1306a85a | 5456 | if (memcg != page->mem_cgroup) { |
747db954 | 5457 | if (memcg) { |
18eca2e6 JW |
5458 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5459 | nr_huge, page); | |
5460 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5461 | } |
1306a85a | 5462 | memcg = page->mem_cgroup; |
747db954 JW |
5463 | } |
5464 | ||
5465 | if (PageTransHuge(page)) { | |
5466 | nr_pages <<= compound_order(page); | |
5467 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5468 | nr_huge += nr_pages; | |
5469 | } | |
5470 | ||
5471 | if (PageAnon(page)) | |
5472 | nr_anon += nr_pages; | |
5473 | else | |
5474 | nr_file += nr_pages; | |
5475 | ||
1306a85a | 5476 | page->mem_cgroup = NULL; |
747db954 JW |
5477 | |
5478 | pgpgout++; | |
5479 | } while (next != page_list); | |
5480 | ||
5481 | if (memcg) | |
18eca2e6 JW |
5482 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5483 | nr_huge, page); | |
747db954 JW |
5484 | } |
5485 | ||
0a31bc97 JW |
5486 | /** |
5487 | * mem_cgroup_uncharge - uncharge a page | |
5488 | * @page: page to uncharge | |
5489 | * | |
5490 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5491 | * mem_cgroup_commit_charge(). | |
5492 | */ | |
5493 | void mem_cgroup_uncharge(struct page *page) | |
5494 | { | |
0a31bc97 JW |
5495 | if (mem_cgroup_disabled()) |
5496 | return; | |
5497 | ||
747db954 | 5498 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5499 | if (!page->mem_cgroup) |
0a31bc97 JW |
5500 | return; |
5501 | ||
747db954 JW |
5502 | INIT_LIST_HEAD(&page->lru); |
5503 | uncharge_list(&page->lru); | |
5504 | } | |
0a31bc97 | 5505 | |
747db954 JW |
5506 | /** |
5507 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5508 | * @page_list: list of pages to uncharge | |
5509 | * | |
5510 | * Uncharge a list of pages previously charged with | |
5511 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5512 | */ | |
5513 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5514 | { | |
5515 | if (mem_cgroup_disabled()) | |
5516 | return; | |
0a31bc97 | 5517 | |
747db954 JW |
5518 | if (!list_empty(page_list)) |
5519 | uncharge_list(page_list); | |
0a31bc97 JW |
5520 | } |
5521 | ||
5522 | /** | |
6a93ca8f JW |
5523 | * mem_cgroup_migrate - charge a page's replacement |
5524 | * @oldpage: currently circulating page | |
5525 | * @newpage: replacement page | |
0a31bc97 | 5526 | * |
6a93ca8f JW |
5527 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
5528 | * be uncharged upon free. | |
0a31bc97 JW |
5529 | * |
5530 | * Both pages must be locked, @newpage->mapping must be set up. | |
5531 | */ | |
6a93ca8f | 5532 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5533 | { |
29833315 | 5534 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
5535 | unsigned int nr_pages; |
5536 | bool compound; | |
0a31bc97 JW |
5537 | |
5538 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5539 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5540 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5541 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5542 | newpage); | |
0a31bc97 JW |
5543 | |
5544 | if (mem_cgroup_disabled()) | |
5545 | return; | |
5546 | ||
5547 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5548 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5549 | return; |
5550 | ||
45637bab | 5551 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5552 | memcg = oldpage->mem_cgroup; |
29833315 | 5553 | if (!memcg) |
0a31bc97 JW |
5554 | return; |
5555 | ||
44b7a8d3 JW |
5556 | /* Force-charge the new page. The old one will be freed soon */ |
5557 | compound = PageTransHuge(newpage); | |
5558 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
5559 | ||
5560 | page_counter_charge(&memcg->memory, nr_pages); | |
5561 | if (do_memsw_account()) | |
5562 | page_counter_charge(&memcg->memsw, nr_pages); | |
5563 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 5564 | |
9cf7666a | 5565 | commit_charge(newpage, memcg, false); |
44b7a8d3 JW |
5566 | |
5567 | local_irq_disable(); | |
5568 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); | |
5569 | memcg_check_events(memcg, newpage); | |
5570 | local_irq_enable(); | |
0a31bc97 JW |
5571 | } |
5572 | ||
ef12947c | 5573 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5574 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5575 | ||
5576 | void sock_update_memcg(struct sock *sk) | |
5577 | { | |
5578 | struct mem_cgroup *memcg; | |
5579 | ||
5580 | /* Socket cloning can throw us here with sk_cgrp already | |
5581 | * filled. It won't however, necessarily happen from | |
5582 | * process context. So the test for root memcg given | |
5583 | * the current task's memcg won't help us in this case. | |
5584 | * | |
5585 | * Respecting the original socket's memcg is a better | |
5586 | * decision in this case. | |
5587 | */ | |
5588 | if (sk->sk_memcg) { | |
5589 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
5590 | css_get(&sk->sk_memcg->css); | |
5591 | return; | |
5592 | } | |
5593 | ||
5594 | rcu_read_lock(); | |
5595 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5596 | if (memcg == root_mem_cgroup) |
5597 | goto out; | |
0db15298 | 5598 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 5599 | goto out; |
f7e1cb6e | 5600 | if (css_tryget_online(&memcg->css)) |
11092087 | 5601 | sk->sk_memcg = memcg; |
f7e1cb6e | 5602 | out: |
11092087 JW |
5603 | rcu_read_unlock(); |
5604 | } | |
5605 | EXPORT_SYMBOL(sock_update_memcg); | |
5606 | ||
5607 | void sock_release_memcg(struct sock *sk) | |
5608 | { | |
5609 | WARN_ON(!sk->sk_memcg); | |
5610 | css_put(&sk->sk_memcg->css); | |
5611 | } | |
5612 | ||
5613 | /** | |
5614 | * mem_cgroup_charge_skmem - charge socket memory | |
5615 | * @memcg: memcg to charge | |
5616 | * @nr_pages: number of pages to charge | |
5617 | * | |
5618 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5619 | * @memcg's configured limit, %false if the charge had to be forced. | |
5620 | */ | |
5621 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5622 | { | |
f7e1cb6e | 5623 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5624 | |
f7e1cb6e | 5625 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5626 | struct page_counter *fail; |
f7e1cb6e | 5627 | |
0db15298 JW |
5628 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
5629 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
5630 | return true; |
5631 | } | |
0db15298 JW |
5632 | page_counter_charge(&memcg->tcpmem, nr_pages); |
5633 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 5634 | return false; |
11092087 | 5635 | } |
d886f4e4 | 5636 | |
f7e1cb6e JW |
5637 | /* Don't block in the packet receive path */ |
5638 | if (in_softirq()) | |
5639 | gfp_mask = GFP_NOWAIT; | |
5640 | ||
b2807f07 JW |
5641 | this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5642 | ||
f7e1cb6e JW |
5643 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
5644 | return true; | |
5645 | ||
5646 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5647 | return false; |
5648 | } | |
5649 | ||
5650 | /** | |
5651 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5652 | * @memcg - memcg to uncharge | |
5653 | * @nr_pages - number of pages to uncharge | |
5654 | */ | |
5655 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5656 | { | |
f7e1cb6e | 5657 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5658 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
5659 | return; |
5660 | } | |
d886f4e4 | 5661 | |
b2807f07 JW |
5662 | this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5663 | ||
f7e1cb6e JW |
5664 | page_counter_uncharge(&memcg->memory, nr_pages); |
5665 | css_put_many(&memcg->css, nr_pages); | |
11092087 JW |
5666 | } |
5667 | ||
f7e1cb6e JW |
5668 | static int __init cgroup_memory(char *s) |
5669 | { | |
5670 | char *token; | |
5671 | ||
5672 | while ((token = strsep(&s, ",")) != NULL) { | |
5673 | if (!*token) | |
5674 | continue; | |
5675 | if (!strcmp(token, "nosocket")) | |
5676 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5677 | if (!strcmp(token, "nokmem")) |
5678 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5679 | } |
5680 | return 0; | |
5681 | } | |
5682 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5683 | |
2d11085e | 5684 | /* |
1081312f MH |
5685 | * subsys_initcall() for memory controller. |
5686 | * | |
5687 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5688 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5689 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5690 | * be initialized from here. | |
2d11085e MH |
5691 | */ |
5692 | static int __init mem_cgroup_init(void) | |
5693 | { | |
95a045f6 JW |
5694 | int cpu, node; |
5695 | ||
2d11085e | 5696 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5697 | |
5698 | for_each_possible_cpu(cpu) | |
5699 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5700 | drain_local_stock); | |
5701 | ||
5702 | for_each_node(node) { | |
5703 | struct mem_cgroup_tree_per_node *rtpn; | |
5704 | int zone; | |
5705 | ||
5706 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5707 | node_online(node) ? node : NUMA_NO_NODE); | |
5708 | ||
5709 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5710 | struct mem_cgroup_tree_per_zone *rtpz; | |
5711 | ||
5712 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5713 | rtpz->rb_root = RB_ROOT; | |
5714 | spin_lock_init(&rtpz->lock); | |
5715 | } | |
5716 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5717 | } | |
5718 | ||
2d11085e MH |
5719 | return 0; |
5720 | } | |
5721 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5722 | |
5723 | #ifdef CONFIG_MEMCG_SWAP | |
5724 | /** | |
5725 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5726 | * @page: page whose memsw charge to transfer | |
5727 | * @entry: swap entry to move the charge to | |
5728 | * | |
5729 | * Transfer the memsw charge of @page to @entry. | |
5730 | */ | |
5731 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5732 | { | |
5733 | struct mem_cgroup *memcg; | |
5734 | unsigned short oldid; | |
5735 | ||
5736 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5737 | VM_BUG_ON_PAGE(page_count(page), page); | |
5738 | ||
7941d214 | 5739 | if (!do_memsw_account()) |
21afa38e JW |
5740 | return; |
5741 | ||
5742 | memcg = page->mem_cgroup; | |
5743 | ||
5744 | /* Readahead page, never charged */ | |
5745 | if (!memcg) | |
5746 | return; | |
5747 | ||
5748 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5749 | VM_BUG_ON_PAGE(oldid, page); | |
5750 | mem_cgroup_swap_statistics(memcg, true); | |
5751 | ||
5752 | page->mem_cgroup = NULL; | |
5753 | ||
5754 | if (!mem_cgroup_is_root(memcg)) | |
5755 | page_counter_uncharge(&memcg->memory, 1); | |
5756 | ||
ce9ce665 SAS |
5757 | /* |
5758 | * Interrupts should be disabled here because the caller holds the | |
5759 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5760 | * important here to have the interrupts disabled because it is the | |
5761 | * only synchronisation we have for udpating the per-CPU variables. | |
5762 | */ | |
5763 | VM_BUG_ON(!irqs_disabled()); | |
f627c2f5 | 5764 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
21afa38e JW |
5765 | memcg_check_events(memcg, page); |
5766 | } | |
5767 | ||
37e84351 VD |
5768 | /* |
5769 | * mem_cgroup_try_charge_swap - try charging a swap entry | |
5770 | * @page: page being added to swap | |
5771 | * @entry: swap entry to charge | |
5772 | * | |
5773 | * Try to charge @entry to the memcg that @page belongs to. | |
5774 | * | |
5775 | * Returns 0 on success, -ENOMEM on failure. | |
5776 | */ | |
5777 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
5778 | { | |
5779 | struct mem_cgroup *memcg; | |
5780 | struct page_counter *counter; | |
5781 | unsigned short oldid; | |
5782 | ||
5783 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
5784 | return 0; | |
5785 | ||
5786 | memcg = page->mem_cgroup; | |
5787 | ||
5788 | /* Readahead page, never charged */ | |
5789 | if (!memcg) | |
5790 | return 0; | |
5791 | ||
5792 | if (!mem_cgroup_is_root(memcg) && | |
5793 | !page_counter_try_charge(&memcg->swap, 1, &counter)) | |
5794 | return -ENOMEM; | |
5795 | ||
5796 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5797 | VM_BUG_ON_PAGE(oldid, page); | |
5798 | mem_cgroup_swap_statistics(memcg, true); | |
5799 | ||
5800 | css_get(&memcg->css); | |
5801 | return 0; | |
5802 | } | |
5803 | ||
21afa38e JW |
5804 | /** |
5805 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5806 | * @entry: swap entry to uncharge | |
5807 | * | |
37e84351 | 5808 | * Drop the swap charge associated with @entry. |
21afa38e JW |
5809 | */ |
5810 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5811 | { | |
5812 | struct mem_cgroup *memcg; | |
5813 | unsigned short id; | |
5814 | ||
37e84351 | 5815 | if (!do_swap_account) |
21afa38e JW |
5816 | return; |
5817 | ||
5818 | id = swap_cgroup_record(entry, 0); | |
5819 | rcu_read_lock(); | |
adbe427b | 5820 | memcg = mem_cgroup_from_id(id); |
21afa38e | 5821 | if (memcg) { |
37e84351 VD |
5822 | if (!mem_cgroup_is_root(memcg)) { |
5823 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5824 | page_counter_uncharge(&memcg->swap, 1); | |
5825 | else | |
5826 | page_counter_uncharge(&memcg->memsw, 1); | |
5827 | } | |
21afa38e JW |
5828 | mem_cgroup_swap_statistics(memcg, false); |
5829 | css_put(&memcg->css); | |
5830 | } | |
5831 | rcu_read_unlock(); | |
5832 | } | |
5833 | ||
d8b38438 VD |
5834 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
5835 | { | |
5836 | long nr_swap_pages = get_nr_swap_pages(); | |
5837 | ||
5838 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5839 | return nr_swap_pages; | |
5840 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5841 | nr_swap_pages = min_t(long, nr_swap_pages, | |
5842 | READ_ONCE(memcg->swap.limit) - | |
5843 | page_counter_read(&memcg->swap)); | |
5844 | return nr_swap_pages; | |
5845 | } | |
5846 | ||
5ccc5aba VD |
5847 | bool mem_cgroup_swap_full(struct page *page) |
5848 | { | |
5849 | struct mem_cgroup *memcg; | |
5850 | ||
5851 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
5852 | ||
5853 | if (vm_swap_full()) | |
5854 | return true; | |
5855 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5856 | return false; | |
5857 | ||
5858 | memcg = page->mem_cgroup; | |
5859 | if (!memcg) | |
5860 | return false; | |
5861 | ||
5862 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5863 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) | |
5864 | return true; | |
5865 | ||
5866 | return false; | |
5867 | } | |
5868 | ||
21afa38e JW |
5869 | /* for remember boot option*/ |
5870 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5871 | static int really_do_swap_account __initdata = 1; | |
5872 | #else | |
5873 | static int really_do_swap_account __initdata; | |
5874 | #endif | |
5875 | ||
5876 | static int __init enable_swap_account(char *s) | |
5877 | { | |
5878 | if (!strcmp(s, "1")) | |
5879 | really_do_swap_account = 1; | |
5880 | else if (!strcmp(s, "0")) | |
5881 | really_do_swap_account = 0; | |
5882 | return 1; | |
5883 | } | |
5884 | __setup("swapaccount=", enable_swap_account); | |
5885 | ||
37e84351 VD |
5886 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
5887 | struct cftype *cft) | |
5888 | { | |
5889 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5890 | ||
5891 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
5892 | } | |
5893 | ||
5894 | static int swap_max_show(struct seq_file *m, void *v) | |
5895 | { | |
5896 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5897 | unsigned long max = READ_ONCE(memcg->swap.limit); | |
5898 | ||
5899 | if (max == PAGE_COUNTER_MAX) | |
5900 | seq_puts(m, "max\n"); | |
5901 | else | |
5902 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5903 | ||
5904 | return 0; | |
5905 | } | |
5906 | ||
5907 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
5908 | char *buf, size_t nbytes, loff_t off) | |
5909 | { | |
5910 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5911 | unsigned long max; | |
5912 | int err; | |
5913 | ||
5914 | buf = strstrip(buf); | |
5915 | err = page_counter_memparse(buf, "max", &max); | |
5916 | if (err) | |
5917 | return err; | |
5918 | ||
5919 | mutex_lock(&memcg_limit_mutex); | |
5920 | err = page_counter_limit(&memcg->swap, max); | |
5921 | mutex_unlock(&memcg_limit_mutex); | |
5922 | if (err) | |
5923 | return err; | |
5924 | ||
5925 | return nbytes; | |
5926 | } | |
5927 | ||
5928 | static struct cftype swap_files[] = { | |
5929 | { | |
5930 | .name = "swap.current", | |
5931 | .flags = CFTYPE_NOT_ON_ROOT, | |
5932 | .read_u64 = swap_current_read, | |
5933 | }, | |
5934 | { | |
5935 | .name = "swap.max", | |
5936 | .flags = CFTYPE_NOT_ON_ROOT, | |
5937 | .seq_show = swap_max_show, | |
5938 | .write = swap_max_write, | |
5939 | }, | |
5940 | { } /* terminate */ | |
5941 | }; | |
5942 | ||
21afa38e JW |
5943 | static struct cftype memsw_cgroup_files[] = { |
5944 | { | |
5945 | .name = "memsw.usage_in_bytes", | |
5946 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5947 | .read_u64 = mem_cgroup_read_u64, | |
5948 | }, | |
5949 | { | |
5950 | .name = "memsw.max_usage_in_bytes", | |
5951 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5952 | .write = mem_cgroup_reset, | |
5953 | .read_u64 = mem_cgroup_read_u64, | |
5954 | }, | |
5955 | { | |
5956 | .name = "memsw.limit_in_bytes", | |
5957 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5958 | .write = mem_cgroup_write, | |
5959 | .read_u64 = mem_cgroup_read_u64, | |
5960 | }, | |
5961 | { | |
5962 | .name = "memsw.failcnt", | |
5963 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5964 | .write = mem_cgroup_reset, | |
5965 | .read_u64 = mem_cgroup_read_u64, | |
5966 | }, | |
5967 | { }, /* terminate */ | |
5968 | }; | |
5969 | ||
5970 | static int __init mem_cgroup_swap_init(void) | |
5971 | { | |
5972 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5973 | do_swap_account = 1; | |
37e84351 VD |
5974 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
5975 | swap_files)); | |
21afa38e JW |
5976 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5977 | memsw_cgroup_files)); | |
5978 | } | |
5979 | return 0; | |
5980 | } | |
5981 | subsys_initcall(mem_cgroup_swap_init); | |
5982 | ||
5983 | #endif /* CONFIG_MEMCG_SWAP */ |