]>
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 | 1037 | |
9d5e6a9f HD |
1038 | __update_lru_size(lruvec, lru, nr_pages); |
1039 | ||
3f58a829 MK |
1040 | if (mem_cgroup_disabled()) |
1041 | return; | |
1042 | ||
fa9add64 HD |
1043 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1044 | lru_size = mz->lru_size + lru; | |
ca707239 HD |
1045 | empty = list_empty(lruvec->lists + lru); |
1046 | ||
1047 | if (nr_pages < 0) | |
1048 | *lru_size += nr_pages; | |
1049 | ||
1050 | size = *lru_size; | |
1051 | if (WARN_ONCE(size < 0 || empty != !size, | |
1052 | "%s(%p, %d, %d): lru_size %ld but %sempty\n", | |
1053 | __func__, lruvec, lru, nr_pages, size, empty ? "" : "not ")) { | |
1054 | VM_BUG_ON(1); | |
1055 | *lru_size = 0; | |
1056 | } | |
1057 | ||
1058 | if (nr_pages > 0) | |
1059 | *lru_size += nr_pages; | |
08e552c6 | 1060 | } |
544122e5 | 1061 | |
2314b42d | 1062 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1063 | { |
2314b42d | 1064 | struct mem_cgroup *task_memcg; |
158e0a2d | 1065 | struct task_struct *p; |
ffbdccf5 | 1066 | bool ret; |
4c4a2214 | 1067 | |
158e0a2d | 1068 | p = find_lock_task_mm(task); |
de077d22 | 1069 | if (p) { |
2314b42d | 1070 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1071 | task_unlock(p); |
1072 | } else { | |
1073 | /* | |
1074 | * All threads may have already detached their mm's, but the oom | |
1075 | * killer still needs to detect if they have already been oom | |
1076 | * killed to prevent needlessly killing additional tasks. | |
1077 | */ | |
ffbdccf5 | 1078 | rcu_read_lock(); |
2314b42d JW |
1079 | task_memcg = mem_cgroup_from_task(task); |
1080 | css_get(&task_memcg->css); | |
ffbdccf5 | 1081 | rcu_read_unlock(); |
de077d22 | 1082 | } |
2314b42d JW |
1083 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1084 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1085 | return ret; |
1086 | } | |
1087 | ||
19942822 | 1088 | /** |
9d11ea9f | 1089 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1090 | * @memcg: the memory cgroup |
19942822 | 1091 | * |
9d11ea9f | 1092 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1093 | * pages. |
19942822 | 1094 | */ |
c0ff4b85 | 1095 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1096 | { |
3e32cb2e JW |
1097 | unsigned long margin = 0; |
1098 | unsigned long count; | |
1099 | unsigned long limit; | |
9d11ea9f | 1100 | |
3e32cb2e | 1101 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1102 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1103 | if (count < limit) |
1104 | margin = limit - count; | |
1105 | ||
7941d214 | 1106 | if (do_memsw_account()) { |
3e32cb2e | 1107 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1108 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1109 | if (count <= limit) |
1110 | margin = min(margin, limit - count); | |
1111 | } | |
1112 | ||
1113 | return margin; | |
19942822 JW |
1114 | } |
1115 | ||
32047e2a | 1116 | /* |
bdcbb659 | 1117 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1118 | * |
bdcbb659 QH |
1119 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1120 | * moving cgroups. This is for waiting at high-memory pressure | |
1121 | * caused by "move". | |
32047e2a | 1122 | */ |
c0ff4b85 | 1123 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1124 | { |
2bd9bb20 KH |
1125 | struct mem_cgroup *from; |
1126 | struct mem_cgroup *to; | |
4b534334 | 1127 | bool ret = false; |
2bd9bb20 KH |
1128 | /* |
1129 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1130 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1131 | */ | |
1132 | spin_lock(&mc.lock); | |
1133 | from = mc.from; | |
1134 | to = mc.to; | |
1135 | if (!from) | |
1136 | goto unlock; | |
3e92041d | 1137 | |
2314b42d JW |
1138 | ret = mem_cgroup_is_descendant(from, memcg) || |
1139 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1140 | unlock: |
1141 | spin_unlock(&mc.lock); | |
4b534334 KH |
1142 | return ret; |
1143 | } | |
1144 | ||
c0ff4b85 | 1145 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1146 | { |
1147 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1148 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1149 | DEFINE_WAIT(wait); |
1150 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1151 | /* moving charge context might have finished. */ | |
1152 | if (mc.moving_task) | |
1153 | schedule(); | |
1154 | finish_wait(&mc.waitq, &wait); | |
1155 | return true; | |
1156 | } | |
1157 | } | |
1158 | return false; | |
1159 | } | |
1160 | ||
58cf188e | 1161 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1162 | /** |
58cf188e | 1163 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1164 | * @memcg: The memory cgroup that went over limit |
1165 | * @p: Task that is going to be killed | |
1166 | * | |
1167 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1168 | * enabled | |
1169 | */ | |
1170 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1171 | { | |
58cf188e SZ |
1172 | struct mem_cgroup *iter; |
1173 | unsigned int i; | |
e222432b | 1174 | |
e222432b BS |
1175 | rcu_read_lock(); |
1176 | ||
2415b9f5 BV |
1177 | if (p) { |
1178 | pr_info("Task in "); | |
1179 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1180 | pr_cont(" killed as a result of limit of "); | |
1181 | } else { | |
1182 | pr_info("Memory limit reached of cgroup "); | |
1183 | } | |
1184 | ||
e61734c5 | 1185 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1186 | pr_cont("\n"); |
e222432b | 1187 | |
e222432b BS |
1188 | rcu_read_unlock(); |
1189 | ||
3e32cb2e JW |
1190 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1191 | K((u64)page_counter_read(&memcg->memory)), | |
1192 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1193 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1194 | K((u64)page_counter_read(&memcg->memsw)), | |
1195 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1196 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1197 | K((u64)page_counter_read(&memcg->kmem)), | |
1198 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1199 | |
1200 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1201 | pr_info("Memory cgroup stats for "); |
1202 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1203 | pr_cont(":"); |
1204 | ||
1205 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
37e84351 | 1206 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
58cf188e | 1207 | continue; |
484ebb3b | 1208 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1209 | K(mem_cgroup_read_stat(iter, i))); |
1210 | } | |
1211 | ||
1212 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1213 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1214 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1215 | ||
1216 | pr_cont("\n"); | |
1217 | } | |
e222432b BS |
1218 | } |
1219 | ||
81d39c20 KH |
1220 | /* |
1221 | * This function returns the number of memcg under hierarchy tree. Returns | |
1222 | * 1(self count) if no children. | |
1223 | */ | |
c0ff4b85 | 1224 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1225 | { |
1226 | int num = 0; | |
7d74b06f KH |
1227 | struct mem_cgroup *iter; |
1228 | ||
c0ff4b85 | 1229 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1230 | num++; |
81d39c20 KH |
1231 | return num; |
1232 | } | |
1233 | ||
a63d83f4 DR |
1234 | /* |
1235 | * Return the memory (and swap, if configured) limit for a memcg. | |
1236 | */ | |
3e32cb2e | 1237 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1238 | { |
3e32cb2e | 1239 | unsigned long limit; |
f3e8eb70 | 1240 | |
3e32cb2e | 1241 | limit = memcg->memory.limit; |
9a5a8f19 | 1242 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1243 | unsigned long memsw_limit; |
37e84351 | 1244 | unsigned long swap_limit; |
9a5a8f19 | 1245 | |
3e32cb2e | 1246 | memsw_limit = memcg->memsw.limit; |
37e84351 VD |
1247 | swap_limit = memcg->swap.limit; |
1248 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); | |
1249 | limit = min(limit + swap_limit, memsw_limit); | |
9a5a8f19 | 1250 | } |
9a5a8f19 | 1251 | return limit; |
a63d83f4 DR |
1252 | } |
1253 | ||
b6e6edcf | 1254 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1255 | int order) |
9cbb78bb | 1256 | { |
6e0fc46d DR |
1257 | struct oom_control oc = { |
1258 | .zonelist = NULL, | |
1259 | .nodemask = NULL, | |
1260 | .gfp_mask = gfp_mask, | |
1261 | .order = order, | |
6e0fc46d | 1262 | }; |
9cbb78bb DR |
1263 | struct mem_cgroup *iter; |
1264 | unsigned long chosen_points = 0; | |
1265 | unsigned long totalpages; | |
1266 | unsigned int points = 0; | |
1267 | struct task_struct *chosen = NULL; | |
1268 | ||
dc56401f JW |
1269 | mutex_lock(&oom_lock); |
1270 | ||
876aafbf | 1271 | /* |
465adcf1 DR |
1272 | * If current has a pending SIGKILL or is exiting, then automatically |
1273 | * select it. The goal is to allow it to allocate so that it may | |
1274 | * quickly exit and free its memory. | |
876aafbf | 1275 | */ |
d003f371 | 1276 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1277 | mark_oom_victim(current); |
3ef22dff | 1278 | try_oom_reaper(current); |
dc56401f | 1279 | goto unlock; |
876aafbf DR |
1280 | } |
1281 | ||
6e0fc46d | 1282 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1283 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1284 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1285 | struct css_task_iter it; |
9cbb78bb DR |
1286 | struct task_struct *task; |
1287 | ||
72ec7029 TH |
1288 | css_task_iter_start(&iter->css, &it); |
1289 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1290 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1291 | case OOM_SCAN_SELECT: |
1292 | if (chosen) | |
1293 | put_task_struct(chosen); | |
1294 | chosen = task; | |
1295 | chosen_points = ULONG_MAX; | |
1296 | get_task_struct(chosen); | |
1297 | /* fall through */ | |
1298 | case OOM_SCAN_CONTINUE: | |
1299 | continue; | |
1300 | case OOM_SCAN_ABORT: | |
72ec7029 | 1301 | css_task_iter_end(&it); |
9cbb78bb DR |
1302 | mem_cgroup_iter_break(memcg, iter); |
1303 | if (chosen) | |
1304 | put_task_struct(chosen); | |
dc56401f | 1305 | goto unlock; |
9cbb78bb DR |
1306 | case OOM_SCAN_OK: |
1307 | break; | |
1308 | }; | |
1309 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1310 | if (!points || points < chosen_points) |
1311 | continue; | |
1312 | /* Prefer thread group leaders for display purposes */ | |
1313 | if (points == chosen_points && | |
1314 | thread_group_leader(chosen)) | |
1315 | continue; | |
1316 | ||
1317 | if (chosen) | |
1318 | put_task_struct(chosen); | |
1319 | chosen = task; | |
1320 | chosen_points = points; | |
1321 | get_task_struct(chosen); | |
9cbb78bb | 1322 | } |
72ec7029 | 1323 | css_task_iter_end(&it); |
9cbb78bb DR |
1324 | } |
1325 | ||
dc56401f JW |
1326 | if (chosen) { |
1327 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1328 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1329 | "Memory cgroup out of memory"); | |
dc56401f JW |
1330 | } |
1331 | unlock: | |
1332 | mutex_unlock(&oom_lock); | |
b6e6edcf | 1333 | return chosen; |
9cbb78bb DR |
1334 | } |
1335 | ||
ae6e71d3 MC |
1336 | #if MAX_NUMNODES > 1 |
1337 | ||
4d0c066d KH |
1338 | /** |
1339 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1340 | * @memcg: the target memcg |
4d0c066d KH |
1341 | * @nid: the node ID to be checked. |
1342 | * @noswap : specify true here if the user wants flle only information. | |
1343 | * | |
1344 | * This function returns whether the specified memcg contains any | |
1345 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1346 | * pages in the node. | |
1347 | */ | |
c0ff4b85 | 1348 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1349 | int nid, bool noswap) |
1350 | { | |
c0ff4b85 | 1351 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1352 | return true; |
1353 | if (noswap || !total_swap_pages) | |
1354 | return false; | |
c0ff4b85 | 1355 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1356 | return true; |
1357 | return false; | |
1358 | ||
1359 | } | |
889976db YH |
1360 | |
1361 | /* | |
1362 | * Always updating the nodemask is not very good - even if we have an empty | |
1363 | * list or the wrong list here, we can start from some node and traverse all | |
1364 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1365 | * | |
1366 | */ | |
c0ff4b85 | 1367 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1368 | { |
1369 | int nid; | |
453a9bf3 KH |
1370 | /* |
1371 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1372 | * pagein/pageout changes since the last update. | |
1373 | */ | |
c0ff4b85 | 1374 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1375 | return; |
c0ff4b85 | 1376 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1377 | return; |
1378 | ||
889976db | 1379 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1380 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1381 | |
31aaea4a | 1382 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1383 | |
c0ff4b85 R |
1384 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1385 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1386 | } |
453a9bf3 | 1387 | |
c0ff4b85 R |
1388 | atomic_set(&memcg->numainfo_events, 0); |
1389 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1390 | } |
1391 | ||
1392 | /* | |
1393 | * Selecting a node where we start reclaim from. Because what we need is just | |
1394 | * reducing usage counter, start from anywhere is O,K. Considering | |
1395 | * memory reclaim from current node, there are pros. and cons. | |
1396 | * | |
1397 | * Freeing memory from current node means freeing memory from a node which | |
1398 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1399 | * hit limits, it will see a contention on a node. But freeing from remote | |
1400 | * node means more costs for memory reclaim because of memory latency. | |
1401 | * | |
1402 | * Now, we use round-robin. Better algorithm is welcomed. | |
1403 | */ | |
c0ff4b85 | 1404 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1405 | { |
1406 | int node; | |
1407 | ||
c0ff4b85 R |
1408 | mem_cgroup_may_update_nodemask(memcg); |
1409 | node = memcg->last_scanned_node; | |
889976db | 1410 | |
0edaf86c | 1411 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1412 | /* |
fda3d69b MH |
1413 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1414 | * last time it really checked all the LRUs due to rate limiting. | |
1415 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1416 | */ |
1417 | if (unlikely(node == MAX_NUMNODES)) | |
1418 | node = numa_node_id(); | |
1419 | ||
c0ff4b85 | 1420 | memcg->last_scanned_node = node; |
889976db YH |
1421 | return node; |
1422 | } | |
889976db | 1423 | #else |
c0ff4b85 | 1424 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1425 | { |
1426 | return 0; | |
1427 | } | |
1428 | #endif | |
1429 | ||
0608f43d AM |
1430 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1431 | struct zone *zone, | |
1432 | gfp_t gfp_mask, | |
1433 | unsigned long *total_scanned) | |
1434 | { | |
1435 | struct mem_cgroup *victim = NULL; | |
1436 | int total = 0; | |
1437 | int loop = 0; | |
1438 | unsigned long excess; | |
1439 | unsigned long nr_scanned; | |
1440 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1441 | .zone = zone, | |
1442 | .priority = 0, | |
1443 | }; | |
1444 | ||
3e32cb2e | 1445 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1446 | |
1447 | while (1) { | |
1448 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1449 | if (!victim) { | |
1450 | loop++; | |
1451 | if (loop >= 2) { | |
1452 | /* | |
1453 | * If we have not been able to reclaim | |
1454 | * anything, it might because there are | |
1455 | * no reclaimable pages under this hierarchy | |
1456 | */ | |
1457 | if (!total) | |
1458 | break; | |
1459 | /* | |
1460 | * We want to do more targeted reclaim. | |
1461 | * excess >> 2 is not to excessive so as to | |
1462 | * reclaim too much, nor too less that we keep | |
1463 | * coming back to reclaim from this cgroup | |
1464 | */ | |
1465 | if (total >= (excess >> 2) || | |
1466 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1467 | break; | |
1468 | } | |
1469 | continue; | |
1470 | } | |
0608f43d AM |
1471 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1472 | zone, &nr_scanned); | |
1473 | *total_scanned += nr_scanned; | |
3e32cb2e | 1474 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1475 | break; |
6d61ef40 | 1476 | } |
0608f43d AM |
1477 | mem_cgroup_iter_break(root_memcg, victim); |
1478 | return total; | |
6d61ef40 BS |
1479 | } |
1480 | ||
0056f4e6 JW |
1481 | #ifdef CONFIG_LOCKDEP |
1482 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1483 | .name = "memcg_oom_lock", | |
1484 | }; | |
1485 | #endif | |
1486 | ||
fb2a6fc5 JW |
1487 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1488 | ||
867578cb KH |
1489 | /* |
1490 | * Check OOM-Killer is already running under our hierarchy. | |
1491 | * If someone is running, return false. | |
1492 | */ | |
fb2a6fc5 | 1493 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1494 | { |
79dfdacc | 1495 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1496 | |
fb2a6fc5 JW |
1497 | spin_lock(&memcg_oom_lock); |
1498 | ||
9f3a0d09 | 1499 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1500 | if (iter->oom_lock) { |
79dfdacc MH |
1501 | /* |
1502 | * this subtree of our hierarchy is already locked | |
1503 | * so we cannot give a lock. | |
1504 | */ | |
79dfdacc | 1505 | failed = iter; |
9f3a0d09 JW |
1506 | mem_cgroup_iter_break(memcg, iter); |
1507 | break; | |
23751be0 JW |
1508 | } else |
1509 | iter->oom_lock = true; | |
7d74b06f | 1510 | } |
867578cb | 1511 | |
fb2a6fc5 JW |
1512 | if (failed) { |
1513 | /* | |
1514 | * OK, we failed to lock the whole subtree so we have | |
1515 | * to clean up what we set up to the failing subtree | |
1516 | */ | |
1517 | for_each_mem_cgroup_tree(iter, memcg) { | |
1518 | if (iter == failed) { | |
1519 | mem_cgroup_iter_break(memcg, iter); | |
1520 | break; | |
1521 | } | |
1522 | iter->oom_lock = false; | |
79dfdacc | 1523 | } |
0056f4e6 JW |
1524 | } else |
1525 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1526 | |
1527 | spin_unlock(&memcg_oom_lock); | |
1528 | ||
1529 | return !failed; | |
a636b327 | 1530 | } |
0b7f569e | 1531 | |
fb2a6fc5 | 1532 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1533 | { |
7d74b06f KH |
1534 | struct mem_cgroup *iter; |
1535 | ||
fb2a6fc5 | 1536 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1537 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1538 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1539 | iter->oom_lock = false; |
fb2a6fc5 | 1540 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1541 | } |
1542 | ||
c0ff4b85 | 1543 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1544 | { |
1545 | struct mem_cgroup *iter; | |
1546 | ||
c2b42d3c | 1547 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1548 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1549 | iter->under_oom++; |
1550 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1551 | } |
1552 | ||
c0ff4b85 | 1553 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1554 | { |
1555 | struct mem_cgroup *iter; | |
1556 | ||
867578cb KH |
1557 | /* |
1558 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1559 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1560 | */ |
c2b42d3c | 1561 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1562 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1563 | if (iter->under_oom > 0) |
1564 | iter->under_oom--; | |
1565 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1566 | } |
1567 | ||
867578cb KH |
1568 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1569 | ||
dc98df5a | 1570 | struct oom_wait_info { |
d79154bb | 1571 | struct mem_cgroup *memcg; |
dc98df5a KH |
1572 | wait_queue_t wait; |
1573 | }; | |
1574 | ||
1575 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1576 | unsigned mode, int sync, void *arg) | |
1577 | { | |
d79154bb HD |
1578 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1579 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1580 | struct oom_wait_info *oom_wait_info; |
1581 | ||
1582 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1583 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1584 | |
2314b42d JW |
1585 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1586 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1587 | return 0; |
dc98df5a KH |
1588 | return autoremove_wake_function(wait, mode, sync, arg); |
1589 | } | |
1590 | ||
c0ff4b85 | 1591 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1592 | { |
c2b42d3c TH |
1593 | /* |
1594 | * For the following lockless ->under_oom test, the only required | |
1595 | * guarantee is that it must see the state asserted by an OOM when | |
1596 | * this function is called as a result of userland actions | |
1597 | * triggered by the notification of the OOM. This is trivially | |
1598 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1599 | * triggering notification. | |
1600 | */ | |
1601 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1602 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1603 | } |
1604 | ||
3812c8c8 | 1605 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1606 | { |
626ebc41 | 1607 | if (!current->memcg_may_oom) |
3812c8c8 | 1608 | return; |
867578cb | 1609 | /* |
49426420 JW |
1610 | * We are in the middle of the charge context here, so we |
1611 | * don't want to block when potentially sitting on a callstack | |
1612 | * that holds all kinds of filesystem and mm locks. | |
1613 | * | |
1614 | * Also, the caller may handle a failed allocation gracefully | |
1615 | * (like optional page cache readahead) and so an OOM killer | |
1616 | * invocation might not even be necessary. | |
1617 | * | |
1618 | * That's why we don't do anything here except remember the | |
1619 | * OOM context and then deal with it at the end of the page | |
1620 | * fault when the stack is unwound, the locks are released, | |
1621 | * and when we know whether the fault was overall successful. | |
867578cb | 1622 | */ |
49426420 | 1623 | css_get(&memcg->css); |
626ebc41 TH |
1624 | current->memcg_in_oom = memcg; |
1625 | current->memcg_oom_gfp_mask = mask; | |
1626 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1627 | } |
1628 | ||
1629 | /** | |
1630 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1631 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1632 | * |
49426420 JW |
1633 | * This has to be called at the end of a page fault if the memcg OOM |
1634 | * handler was enabled. | |
3812c8c8 | 1635 | * |
49426420 | 1636 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1637 | * sleep on a waitqueue until the userspace task resolves the |
1638 | * situation. Sleeping directly in the charge context with all kinds | |
1639 | * of locks held is not a good idea, instead we remember an OOM state | |
1640 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1641 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1642 | * |
1643 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1644 | * completed, %false otherwise. |
3812c8c8 | 1645 | */ |
49426420 | 1646 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1647 | { |
626ebc41 | 1648 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1649 | struct oom_wait_info owait; |
49426420 | 1650 | bool locked; |
3812c8c8 JW |
1651 | |
1652 | /* OOM is global, do not handle */ | |
3812c8c8 | 1653 | if (!memcg) |
49426420 | 1654 | return false; |
3812c8c8 | 1655 | |
c32b3cbe | 1656 | if (!handle || oom_killer_disabled) |
49426420 | 1657 | goto cleanup; |
3812c8c8 JW |
1658 | |
1659 | owait.memcg = memcg; | |
1660 | owait.wait.flags = 0; | |
1661 | owait.wait.func = memcg_oom_wake_function; | |
1662 | owait.wait.private = current; | |
1663 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1664 | |
3812c8c8 | 1665 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1666 | mem_cgroup_mark_under_oom(memcg); |
1667 | ||
1668 | locked = mem_cgroup_oom_trylock(memcg); | |
1669 | ||
1670 | if (locked) | |
1671 | mem_cgroup_oom_notify(memcg); | |
1672 | ||
1673 | if (locked && !memcg->oom_kill_disable) { | |
1674 | mem_cgroup_unmark_under_oom(memcg); | |
1675 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1676 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1677 | current->memcg_oom_order); | |
49426420 | 1678 | } else { |
3812c8c8 | 1679 | schedule(); |
49426420 JW |
1680 | mem_cgroup_unmark_under_oom(memcg); |
1681 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1682 | } | |
1683 | ||
1684 | if (locked) { | |
fb2a6fc5 JW |
1685 | mem_cgroup_oom_unlock(memcg); |
1686 | /* | |
1687 | * There is no guarantee that an OOM-lock contender | |
1688 | * sees the wakeups triggered by the OOM kill | |
1689 | * uncharges. Wake any sleepers explicitely. | |
1690 | */ | |
1691 | memcg_oom_recover(memcg); | |
1692 | } | |
49426420 | 1693 | cleanup: |
626ebc41 | 1694 | current->memcg_in_oom = NULL; |
3812c8c8 | 1695 | css_put(&memcg->css); |
867578cb | 1696 | return true; |
0b7f569e KH |
1697 | } |
1698 | ||
d7365e78 | 1699 | /** |
81f8c3a4 JW |
1700 | * lock_page_memcg - lock a page->mem_cgroup binding |
1701 | * @page: the page | |
32047e2a | 1702 | * |
81f8c3a4 JW |
1703 | * This function protects unlocked LRU pages from being moved to |
1704 | * another cgroup and stabilizes their page->mem_cgroup binding. | |
d69b042f | 1705 | */ |
62cccb8c | 1706 | void lock_page_memcg(struct page *page) |
89c06bd5 KH |
1707 | { |
1708 | struct mem_cgroup *memcg; | |
6de22619 | 1709 | unsigned long flags; |
89c06bd5 | 1710 | |
6de22619 JW |
1711 | /* |
1712 | * The RCU lock is held throughout the transaction. The fast | |
1713 | * path can get away without acquiring the memcg->move_lock | |
1714 | * because page moving starts with an RCU grace period. | |
6de22619 | 1715 | */ |
d7365e78 JW |
1716 | rcu_read_lock(); |
1717 | ||
1718 | if (mem_cgroup_disabled()) | |
62cccb8c | 1719 | return; |
89c06bd5 | 1720 | again: |
1306a85a | 1721 | memcg = page->mem_cgroup; |
29833315 | 1722 | if (unlikely(!memcg)) |
62cccb8c | 1723 | return; |
d7365e78 | 1724 | |
bdcbb659 | 1725 | if (atomic_read(&memcg->moving_account) <= 0) |
62cccb8c | 1726 | return; |
89c06bd5 | 1727 | |
6de22619 | 1728 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1729 | if (memcg != page->mem_cgroup) { |
6de22619 | 1730 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1731 | goto again; |
1732 | } | |
6de22619 JW |
1733 | |
1734 | /* | |
1735 | * When charge migration first begins, we can have locked and | |
1736 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1737 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1738 | */ |
1739 | memcg->move_lock_task = current; | |
1740 | memcg->move_lock_flags = flags; | |
d7365e78 | 1741 | |
62cccb8c | 1742 | return; |
89c06bd5 | 1743 | } |
81f8c3a4 | 1744 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1745 | |
d7365e78 | 1746 | /** |
81f8c3a4 | 1747 | * unlock_page_memcg - unlock a page->mem_cgroup binding |
62cccb8c | 1748 | * @page: the page |
d7365e78 | 1749 | */ |
62cccb8c | 1750 | void unlock_page_memcg(struct page *page) |
89c06bd5 | 1751 | { |
62cccb8c JW |
1752 | struct mem_cgroup *memcg = page->mem_cgroup; |
1753 | ||
6de22619 JW |
1754 | if (memcg && memcg->move_lock_task == current) { |
1755 | unsigned long flags = memcg->move_lock_flags; | |
1756 | ||
1757 | memcg->move_lock_task = NULL; | |
1758 | memcg->move_lock_flags = 0; | |
1759 | ||
1760 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1761 | } | |
89c06bd5 | 1762 | |
d7365e78 | 1763 | rcu_read_unlock(); |
89c06bd5 | 1764 | } |
81f8c3a4 | 1765 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1766 | |
cdec2e42 KH |
1767 | /* |
1768 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1769 | * TODO: maybe necessary to use big numbers in big irons. | |
1770 | */ | |
7ec99d62 | 1771 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1772 | struct memcg_stock_pcp { |
1773 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1774 | unsigned int nr_pages; |
cdec2e42 | 1775 | struct work_struct work; |
26fe6168 | 1776 | unsigned long flags; |
a0db00fc | 1777 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1778 | }; |
1779 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1780 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1781 | |
a0956d54 SS |
1782 | /** |
1783 | * consume_stock: Try to consume stocked charge on this cpu. | |
1784 | * @memcg: memcg to consume from. | |
1785 | * @nr_pages: how many pages to charge. | |
1786 | * | |
1787 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1788 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1789 | * service an allocation will refill the stock. | |
1790 | * | |
1791 | * returns true if successful, false otherwise. | |
cdec2e42 | 1792 | */ |
a0956d54 | 1793 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1794 | { |
1795 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1796 | bool ret = false; |
cdec2e42 | 1797 | |
a0956d54 | 1798 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1799 | return ret; |
a0956d54 | 1800 | |
cdec2e42 | 1801 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1802 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1803 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1804 | ret = true; |
1805 | } | |
cdec2e42 KH |
1806 | put_cpu_var(memcg_stock); |
1807 | return ret; | |
1808 | } | |
1809 | ||
1810 | /* | |
3e32cb2e | 1811 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1812 | */ |
1813 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1814 | { | |
1815 | struct mem_cgroup *old = stock->cached; | |
1816 | ||
11c9ea4e | 1817 | if (stock->nr_pages) { |
3e32cb2e | 1818 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1819 | if (do_memsw_account()) |
3e32cb2e | 1820 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1821 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1822 | stock->nr_pages = 0; |
cdec2e42 KH |
1823 | } |
1824 | stock->cached = NULL; | |
cdec2e42 KH |
1825 | } |
1826 | ||
1827 | /* | |
1828 | * This must be called under preempt disabled or must be called by | |
1829 | * a thread which is pinned to local cpu. | |
1830 | */ | |
1831 | static void drain_local_stock(struct work_struct *dummy) | |
1832 | { | |
7c8e0181 | 1833 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1834 | drain_stock(stock); |
26fe6168 | 1835 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1836 | } |
1837 | ||
1838 | /* | |
3e32cb2e | 1839 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1840 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1841 | */ |
c0ff4b85 | 1842 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1843 | { |
1844 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1845 | ||
c0ff4b85 | 1846 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1847 | drain_stock(stock); |
c0ff4b85 | 1848 | stock->cached = memcg; |
cdec2e42 | 1849 | } |
11c9ea4e | 1850 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1851 | put_cpu_var(memcg_stock); |
1852 | } | |
1853 | ||
1854 | /* | |
c0ff4b85 | 1855 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1856 | * of the hierarchy under it. |
cdec2e42 | 1857 | */ |
6d3d6aa2 | 1858 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1859 | { |
26fe6168 | 1860 | int cpu, curcpu; |
d38144b7 | 1861 | |
6d3d6aa2 JW |
1862 | /* If someone's already draining, avoid adding running more workers. */ |
1863 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1864 | return; | |
cdec2e42 | 1865 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1866 | get_online_cpus(); |
5af12d0e | 1867 | curcpu = get_cpu(); |
cdec2e42 KH |
1868 | for_each_online_cpu(cpu) { |
1869 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1870 | struct mem_cgroup *memcg; |
26fe6168 | 1871 | |
c0ff4b85 R |
1872 | memcg = stock->cached; |
1873 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1874 | continue; |
2314b42d | 1875 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1876 | continue; |
d1a05b69 MH |
1877 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1878 | if (cpu == curcpu) | |
1879 | drain_local_stock(&stock->work); | |
1880 | else | |
1881 | schedule_work_on(cpu, &stock->work); | |
1882 | } | |
cdec2e42 | 1883 | } |
5af12d0e | 1884 | put_cpu(); |
f894ffa8 | 1885 | put_online_cpus(); |
9f50fad6 | 1886 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1887 | } |
1888 | ||
0db0628d | 1889 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1890 | unsigned long action, |
1891 | void *hcpu) | |
1892 | { | |
1893 | int cpu = (unsigned long)hcpu; | |
1894 | struct memcg_stock_pcp *stock; | |
1895 | ||
619d094b | 1896 | if (action == CPU_ONLINE) |
1489ebad | 1897 | return NOTIFY_OK; |
1489ebad | 1898 | |
d833049b | 1899 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1900 | return NOTIFY_OK; |
711d3d2c | 1901 | |
cdec2e42 KH |
1902 | stock = &per_cpu(memcg_stock, cpu); |
1903 | drain_stock(stock); | |
1904 | return NOTIFY_OK; | |
1905 | } | |
1906 | ||
f7e1cb6e JW |
1907 | static void reclaim_high(struct mem_cgroup *memcg, |
1908 | unsigned int nr_pages, | |
1909 | gfp_t gfp_mask) | |
1910 | { | |
1911 | do { | |
1912 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1913 | continue; | |
1914 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
1915 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
1916 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1917 | } | |
1918 | ||
1919 | static void high_work_func(struct work_struct *work) | |
1920 | { | |
1921 | struct mem_cgroup *memcg; | |
1922 | ||
1923 | memcg = container_of(work, struct mem_cgroup, high_work); | |
1924 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); | |
1925 | } | |
1926 | ||
b23afb93 TH |
1927 | /* |
1928 | * Scheduled by try_charge() to be executed from the userland return path | |
1929 | * and reclaims memory over the high limit. | |
1930 | */ | |
1931 | void mem_cgroup_handle_over_high(void) | |
1932 | { | |
1933 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1934 | struct mem_cgroup *memcg; |
b23afb93 TH |
1935 | |
1936 | if (likely(!nr_pages)) | |
1937 | return; | |
1938 | ||
f7e1cb6e JW |
1939 | memcg = get_mem_cgroup_from_mm(current->mm); |
1940 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1941 | css_put(&memcg->css); |
1942 | current->memcg_nr_pages_over_high = 0; | |
1943 | } | |
1944 | ||
00501b53 JW |
1945 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1946 | unsigned int nr_pages) | |
8a9f3ccd | 1947 | { |
7ec99d62 | 1948 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1949 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1950 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1951 | struct page_counter *counter; |
6539cc05 | 1952 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1953 | bool may_swap = true; |
1954 | bool drained = false; | |
a636b327 | 1955 | |
ce00a967 | 1956 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1957 | return 0; |
6539cc05 | 1958 | retry: |
b6b6cc72 | 1959 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1960 | return 0; |
8a9f3ccd | 1961 | |
7941d214 | 1962 | if (!do_memsw_account() || |
6071ca52 JW |
1963 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1964 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1965 | goto done_restock; |
7941d214 | 1966 | if (do_memsw_account()) |
3e32cb2e JW |
1967 | page_counter_uncharge(&memcg->memsw, batch); |
1968 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1969 | } else { |
3e32cb2e | 1970 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1971 | may_swap = false; |
3fbe7244 | 1972 | } |
7a81b88c | 1973 | |
6539cc05 JW |
1974 | if (batch > nr_pages) { |
1975 | batch = nr_pages; | |
1976 | goto retry; | |
1977 | } | |
6d61ef40 | 1978 | |
06b078fc JW |
1979 | /* |
1980 | * Unlike in global OOM situations, memcg is not in a physical | |
1981 | * memory shortage. Allow dying and OOM-killed tasks to | |
1982 | * bypass the last charges so that they can exit quickly and | |
1983 | * free their memory. | |
1984 | */ | |
1985 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
1986 | fatal_signal_pending(current) || | |
1987 | current->flags & PF_EXITING)) | |
10d53c74 | 1988 | goto force; |
06b078fc JW |
1989 | |
1990 | if (unlikely(task_in_memcg_oom(current))) | |
1991 | goto nomem; | |
1992 | ||
d0164adc | 1993 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 1994 | goto nomem; |
4b534334 | 1995 | |
241994ed JW |
1996 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
1997 | ||
b70a2a21 JW |
1998 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
1999 | gfp_mask, may_swap); | |
6539cc05 | 2000 | |
61e02c74 | 2001 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2002 | goto retry; |
28c34c29 | 2003 | |
b70a2a21 | 2004 | if (!drained) { |
6d3d6aa2 | 2005 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2006 | drained = true; |
2007 | goto retry; | |
2008 | } | |
2009 | ||
28c34c29 JW |
2010 | if (gfp_mask & __GFP_NORETRY) |
2011 | goto nomem; | |
6539cc05 JW |
2012 | /* |
2013 | * Even though the limit is exceeded at this point, reclaim | |
2014 | * may have been able to free some pages. Retry the charge | |
2015 | * before killing the task. | |
2016 | * | |
2017 | * Only for regular pages, though: huge pages are rather | |
2018 | * unlikely to succeed so close to the limit, and we fall back | |
2019 | * to regular pages anyway in case of failure. | |
2020 | */ | |
61e02c74 | 2021 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2022 | goto retry; |
2023 | /* | |
2024 | * At task move, charge accounts can be doubly counted. So, it's | |
2025 | * better to wait until the end of task_move if something is going on. | |
2026 | */ | |
2027 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2028 | goto retry; | |
2029 | ||
9b130619 JW |
2030 | if (nr_retries--) |
2031 | goto retry; | |
2032 | ||
06b078fc | 2033 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2034 | goto force; |
06b078fc | 2035 | |
6539cc05 | 2036 | if (fatal_signal_pending(current)) |
10d53c74 | 2037 | goto force; |
6539cc05 | 2038 | |
241994ed JW |
2039 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2040 | ||
3608de07 JM |
2041 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2042 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2043 | nomem: |
6d1fdc48 | 2044 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2045 | return -ENOMEM; |
10d53c74 TH |
2046 | force: |
2047 | /* | |
2048 | * The allocation either can't fail or will lead to more memory | |
2049 | * being freed very soon. Allow memory usage go over the limit | |
2050 | * temporarily by force charging it. | |
2051 | */ | |
2052 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2053 | if (do_memsw_account()) |
10d53c74 TH |
2054 | page_counter_charge(&memcg->memsw, nr_pages); |
2055 | css_get_many(&memcg->css, nr_pages); | |
2056 | ||
2057 | return 0; | |
6539cc05 JW |
2058 | |
2059 | done_restock: | |
e8ea14cc | 2060 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2061 | if (batch > nr_pages) |
2062 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2063 | |
241994ed | 2064 | /* |
b23afb93 TH |
2065 | * If the hierarchy is above the normal consumption range, schedule |
2066 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2067 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2068 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2069 | * not recorded as it most likely matches current's and won't | |
2070 | * change in the meantime. As high limit is checked again before | |
2071 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2072 | */ |
2073 | do { | |
b23afb93 | 2074 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2075 | /* Don't bother a random interrupted task */ |
2076 | if (in_interrupt()) { | |
2077 | schedule_work(&memcg->high_work); | |
2078 | break; | |
2079 | } | |
9516a18a | 2080 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2081 | set_notify_resume(current); |
2082 | break; | |
2083 | } | |
241994ed | 2084 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2085 | |
2086 | return 0; | |
7a81b88c | 2087 | } |
8a9f3ccd | 2088 | |
00501b53 | 2089 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2090 | { |
ce00a967 JW |
2091 | if (mem_cgroup_is_root(memcg)) |
2092 | return; | |
2093 | ||
3e32cb2e | 2094 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2095 | if (do_memsw_account()) |
3e32cb2e | 2096 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2097 | |
e8ea14cc | 2098 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2099 | } |
2100 | ||
0a31bc97 JW |
2101 | static void lock_page_lru(struct page *page, int *isolated) |
2102 | { | |
2103 | struct zone *zone = page_zone(page); | |
2104 | ||
2105 | spin_lock_irq(&zone->lru_lock); | |
2106 | if (PageLRU(page)) { | |
2107 | struct lruvec *lruvec; | |
2108 | ||
2109 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2110 | ClearPageLRU(page); | |
2111 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2112 | *isolated = 1; | |
2113 | } else | |
2114 | *isolated = 0; | |
2115 | } | |
2116 | ||
2117 | static void unlock_page_lru(struct page *page, int isolated) | |
2118 | { | |
2119 | struct zone *zone = page_zone(page); | |
2120 | ||
2121 | if (isolated) { | |
2122 | struct lruvec *lruvec; | |
2123 | ||
2124 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2125 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2126 | SetPageLRU(page); | |
2127 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2128 | } | |
2129 | spin_unlock_irq(&zone->lru_lock); | |
2130 | } | |
2131 | ||
00501b53 | 2132 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2133 | bool lrucare) |
7a81b88c | 2134 | { |
0a31bc97 | 2135 | int isolated; |
9ce70c02 | 2136 | |
1306a85a | 2137 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2138 | |
2139 | /* | |
2140 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2141 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2142 | */ | |
0a31bc97 JW |
2143 | if (lrucare) |
2144 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2145 | |
0a31bc97 JW |
2146 | /* |
2147 | * Nobody should be changing or seriously looking at | |
1306a85a | 2148 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2149 | * |
2150 | * - the page is uncharged | |
2151 | * | |
2152 | * - the page is off-LRU | |
2153 | * | |
2154 | * - an anonymous fault has exclusive page access, except for | |
2155 | * a locked page table | |
2156 | * | |
2157 | * - a page cache insertion, a swapin fault, or a migration | |
2158 | * have the page locked | |
2159 | */ | |
1306a85a | 2160 | page->mem_cgroup = memcg; |
9ce70c02 | 2161 | |
0a31bc97 JW |
2162 | if (lrucare) |
2163 | unlock_page_lru(page, isolated); | |
7a81b88c | 2164 | } |
66e1707b | 2165 | |
127424c8 | 2166 | #ifndef CONFIG_SLOB |
f3bb3043 | 2167 | static int memcg_alloc_cache_id(void) |
55007d84 | 2168 | { |
f3bb3043 VD |
2169 | int id, size; |
2170 | int err; | |
2171 | ||
dbcf73e2 | 2172 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2173 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2174 | if (id < 0) | |
2175 | return id; | |
55007d84 | 2176 | |
dbcf73e2 | 2177 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2178 | return id; |
2179 | ||
2180 | /* | |
2181 | * There's no space for the new id in memcg_caches arrays, | |
2182 | * so we have to grow them. | |
2183 | */ | |
05257a1a | 2184 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2185 | |
2186 | size = 2 * (id + 1); | |
55007d84 GC |
2187 | if (size < MEMCG_CACHES_MIN_SIZE) |
2188 | size = MEMCG_CACHES_MIN_SIZE; | |
2189 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2190 | size = MEMCG_CACHES_MAX_SIZE; | |
2191 | ||
f3bb3043 | 2192 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2193 | if (!err) |
2194 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2195 | if (!err) |
2196 | memcg_nr_cache_ids = size; | |
2197 | ||
2198 | up_write(&memcg_cache_ids_sem); | |
2199 | ||
f3bb3043 | 2200 | if (err) { |
dbcf73e2 | 2201 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2202 | return err; |
2203 | } | |
2204 | return id; | |
2205 | } | |
2206 | ||
2207 | static void memcg_free_cache_id(int id) | |
2208 | { | |
dbcf73e2 | 2209 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2210 | } |
2211 | ||
d5b3cf71 | 2212 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2213 | struct mem_cgroup *memcg; |
2214 | struct kmem_cache *cachep; | |
2215 | struct work_struct work; | |
2216 | }; | |
2217 | ||
d5b3cf71 | 2218 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2219 | { |
d5b3cf71 VD |
2220 | struct memcg_kmem_cache_create_work *cw = |
2221 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2222 | struct mem_cgroup *memcg = cw->memcg; |
2223 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2224 | |
d5b3cf71 | 2225 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2226 | |
5722d094 | 2227 | css_put(&memcg->css); |
d7f25f8a GC |
2228 | kfree(cw); |
2229 | } | |
2230 | ||
2231 | /* | |
2232 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2233 | */ |
d5b3cf71 VD |
2234 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2235 | struct kmem_cache *cachep) | |
d7f25f8a | 2236 | { |
d5b3cf71 | 2237 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2238 | |
776ed0f0 | 2239 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2240 | if (!cw) |
d7f25f8a | 2241 | return; |
8135be5a VD |
2242 | |
2243 | css_get(&memcg->css); | |
d7f25f8a GC |
2244 | |
2245 | cw->memcg = memcg; | |
2246 | cw->cachep = cachep; | |
d5b3cf71 | 2247 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2248 | |
d7f25f8a GC |
2249 | schedule_work(&cw->work); |
2250 | } | |
2251 | ||
d5b3cf71 VD |
2252 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2253 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2254 | { |
2255 | /* | |
2256 | * We need to stop accounting when we kmalloc, because if the | |
2257 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2258 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2259 | * |
2260 | * However, it is better to enclose the whole function. Depending on | |
2261 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2262 | * trigger an allocation. This too, will make us recurse. Because at | |
2263 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2264 | * the safest choice is to do it like this, wrapping the whole function. | |
2265 | */ | |
6f185c29 | 2266 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2267 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2268 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2269 | } |
c67a8a68 | 2270 | |
d7f25f8a GC |
2271 | /* |
2272 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2273 | * We try to use the current memcg's version of the cache. | |
2274 | * | |
2275 | * If the cache does not exist yet, if we are the first user of it, | |
2276 | * we either create it immediately, if possible, or create it asynchronously | |
2277 | * in a workqueue. | |
2278 | * In the latter case, we will let the current allocation go through with | |
2279 | * the original cache. | |
2280 | * | |
2281 | * Can't be called in interrupt context or from kernel threads. | |
2282 | * This function needs to be called with rcu_read_lock() held. | |
2283 | */ | |
230e9fc2 | 2284 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp) |
d7f25f8a GC |
2285 | { |
2286 | struct mem_cgroup *memcg; | |
959c8963 | 2287 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2288 | int kmemcg_id; |
d7f25f8a | 2289 | |
f7ce3190 | 2290 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2291 | |
230e9fc2 VD |
2292 | if (cachep->flags & SLAB_ACCOUNT) |
2293 | gfp |= __GFP_ACCOUNT; | |
2294 | ||
2295 | if (!(gfp & __GFP_ACCOUNT)) | |
2296 | return cachep; | |
2297 | ||
9d100c5e | 2298 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2299 | return cachep; |
2300 | ||
8135be5a | 2301 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2302 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2303 | if (kmemcg_id < 0) |
ca0dde97 | 2304 | goto out; |
d7f25f8a | 2305 | |
2a4db7eb | 2306 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2307 | if (likely(memcg_cachep)) |
2308 | return memcg_cachep; | |
ca0dde97 LZ |
2309 | |
2310 | /* | |
2311 | * If we are in a safe context (can wait, and not in interrupt | |
2312 | * context), we could be be predictable and return right away. | |
2313 | * This would guarantee that the allocation being performed | |
2314 | * already belongs in the new cache. | |
2315 | * | |
2316 | * However, there are some clashes that can arrive from locking. | |
2317 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2318 | * memcg_create_kmem_cache, this means no further allocation |
2319 | * could happen with the slab_mutex held. So it's better to | |
2320 | * defer everything. | |
ca0dde97 | 2321 | */ |
d5b3cf71 | 2322 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2323 | out: |
8135be5a | 2324 | css_put(&memcg->css); |
ca0dde97 | 2325 | return cachep; |
d7f25f8a | 2326 | } |
d7f25f8a | 2327 | |
8135be5a VD |
2328 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2329 | { | |
2330 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2331 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2332 | } |
2333 | ||
f3ccb2c4 VD |
2334 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
2335 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2336 | { |
f3ccb2c4 VD |
2337 | unsigned int nr_pages = 1 << order; |
2338 | struct page_counter *counter; | |
7ae1e1d0 GC |
2339 | int ret; |
2340 | ||
f3ccb2c4 | 2341 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2342 | if (ret) |
f3ccb2c4 | 2343 | return ret; |
52c29b04 JW |
2344 | |
2345 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2346 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2347 | cancel_charge(memcg, nr_pages); | |
2348 | return -ENOMEM; | |
7ae1e1d0 GC |
2349 | } |
2350 | ||
f3ccb2c4 | 2351 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2352 | |
f3ccb2c4 | 2353 | return 0; |
7ae1e1d0 GC |
2354 | } |
2355 | ||
f3ccb2c4 | 2356 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2357 | { |
f3ccb2c4 | 2358 | struct mem_cgroup *memcg; |
fcff7d7e | 2359 | int ret = 0; |
7ae1e1d0 | 2360 | |
f3ccb2c4 | 2361 | memcg = get_mem_cgroup_from_mm(current->mm); |
b6ecd2de | 2362 | if (!mem_cgroup_is_root(memcg)) |
fcff7d7e | 2363 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
7ae1e1d0 | 2364 | css_put(&memcg->css); |
d05e83a6 | 2365 | return ret; |
7ae1e1d0 GC |
2366 | } |
2367 | ||
d05e83a6 | 2368 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2369 | { |
1306a85a | 2370 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2371 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2372 | |
7ae1e1d0 GC |
2373 | if (!memcg) |
2374 | return; | |
2375 | ||
309381fe | 2376 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2377 | |
52c29b04 JW |
2378 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2379 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2380 | ||
f3ccb2c4 | 2381 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2382 | if (do_memsw_account()) |
f3ccb2c4 | 2383 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2384 | |
1306a85a | 2385 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2386 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2387 | } |
127424c8 | 2388 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2389 | |
ca3e0214 KH |
2390 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2391 | ||
ca3e0214 KH |
2392 | /* |
2393 | * Because tail pages are not marked as "used", set it. We're under | |
3ac808fd | 2394 | * zone->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2395 | */ |
e94c8a9c | 2396 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2397 | { |
e94c8a9c | 2398 | int i; |
ca3e0214 | 2399 | |
3d37c4a9 KH |
2400 | if (mem_cgroup_disabled()) |
2401 | return; | |
b070e65c | 2402 | |
29833315 | 2403 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2404 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2405 | |
1306a85a | 2406 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2407 | HPAGE_PMD_NR); |
ca3e0214 | 2408 | } |
12d27107 | 2409 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2410 | |
c255a458 | 2411 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2412 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2413 | bool charge) | |
d13d1443 | 2414 | { |
0a31bc97 JW |
2415 | int val = (charge) ? 1 : -1; |
2416 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2417 | } |
02491447 DN |
2418 | |
2419 | /** | |
2420 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2421 | * @entry: swap entry to be moved | |
2422 | * @from: mem_cgroup which the entry is moved from | |
2423 | * @to: mem_cgroup which the entry is moved to | |
2424 | * | |
2425 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2426 | * as the mem_cgroup's id of @from. | |
2427 | * | |
2428 | * Returns 0 on success, -EINVAL on failure. | |
2429 | * | |
3e32cb2e | 2430 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2431 | * both res and memsw, and called css_get(). |
2432 | */ | |
2433 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2434 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2435 | { |
2436 | unsigned short old_id, new_id; | |
2437 | ||
34c00c31 LZ |
2438 | old_id = mem_cgroup_id(from); |
2439 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2440 | |
2441 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2442 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2443 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2444 | return 0; |
2445 | } | |
2446 | return -EINVAL; | |
2447 | } | |
2448 | #else | |
2449 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2450 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2451 | { |
2452 | return -EINVAL; | |
2453 | } | |
8c7c6e34 | 2454 | #endif |
d13d1443 | 2455 | |
3e32cb2e | 2456 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2457 | |
d38d2a75 | 2458 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2459 | unsigned long limit) |
628f4235 | 2460 | { |
3e32cb2e JW |
2461 | unsigned long curusage; |
2462 | unsigned long oldusage; | |
2463 | bool enlarge = false; | |
81d39c20 | 2464 | int retry_count; |
3e32cb2e | 2465 | int ret; |
81d39c20 KH |
2466 | |
2467 | /* | |
2468 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2469 | * is depends on callers. We set our retry-count to be function | |
2470 | * of # of children which we should visit in this loop. | |
2471 | */ | |
3e32cb2e JW |
2472 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2473 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2474 | |
3e32cb2e | 2475 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2476 | |
3e32cb2e | 2477 | do { |
628f4235 KH |
2478 | if (signal_pending(current)) { |
2479 | ret = -EINTR; | |
2480 | break; | |
2481 | } | |
3e32cb2e JW |
2482 | |
2483 | mutex_lock(&memcg_limit_mutex); | |
2484 | if (limit > memcg->memsw.limit) { | |
2485 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2486 | ret = -EINVAL; |
628f4235 KH |
2487 | break; |
2488 | } | |
3e32cb2e JW |
2489 | if (limit > memcg->memory.limit) |
2490 | enlarge = true; | |
2491 | ret = page_counter_limit(&memcg->memory, limit); | |
2492 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2493 | |
2494 | if (!ret) | |
2495 | break; | |
2496 | ||
b70a2a21 JW |
2497 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2498 | ||
3e32cb2e | 2499 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2500 | /* Usage is reduced ? */ |
f894ffa8 | 2501 | if (curusage >= oldusage) |
81d39c20 KH |
2502 | retry_count--; |
2503 | else | |
2504 | oldusage = curusage; | |
3e32cb2e JW |
2505 | } while (retry_count); |
2506 | ||
3c11ecf4 KH |
2507 | if (!ret && enlarge) |
2508 | memcg_oom_recover(memcg); | |
14797e23 | 2509 | |
8c7c6e34 KH |
2510 | return ret; |
2511 | } | |
2512 | ||
338c8431 | 2513 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2514 | unsigned long limit) |
8c7c6e34 | 2515 | { |
3e32cb2e JW |
2516 | unsigned long curusage; |
2517 | unsigned long oldusage; | |
2518 | bool enlarge = false; | |
81d39c20 | 2519 | int retry_count; |
3e32cb2e | 2520 | int ret; |
8c7c6e34 | 2521 | |
81d39c20 | 2522 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2523 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2524 | mem_cgroup_count_children(memcg); | |
2525 | ||
2526 | oldusage = page_counter_read(&memcg->memsw); | |
2527 | ||
2528 | do { | |
8c7c6e34 KH |
2529 | if (signal_pending(current)) { |
2530 | ret = -EINTR; | |
2531 | break; | |
2532 | } | |
3e32cb2e JW |
2533 | |
2534 | mutex_lock(&memcg_limit_mutex); | |
2535 | if (limit < memcg->memory.limit) { | |
2536 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2537 | ret = -EINVAL; |
8c7c6e34 KH |
2538 | break; |
2539 | } | |
3e32cb2e JW |
2540 | if (limit > memcg->memsw.limit) |
2541 | enlarge = true; | |
2542 | ret = page_counter_limit(&memcg->memsw, limit); | |
2543 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2544 | |
2545 | if (!ret) | |
2546 | break; | |
2547 | ||
b70a2a21 JW |
2548 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2549 | ||
3e32cb2e | 2550 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2551 | /* Usage is reduced ? */ |
8c7c6e34 | 2552 | if (curusage >= oldusage) |
628f4235 | 2553 | retry_count--; |
81d39c20 KH |
2554 | else |
2555 | oldusage = curusage; | |
3e32cb2e JW |
2556 | } while (retry_count); |
2557 | ||
3c11ecf4 KH |
2558 | if (!ret && enlarge) |
2559 | memcg_oom_recover(memcg); | |
3e32cb2e | 2560 | |
628f4235 KH |
2561 | return ret; |
2562 | } | |
2563 | ||
0608f43d AM |
2564 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2565 | gfp_t gfp_mask, | |
2566 | unsigned long *total_scanned) | |
2567 | { | |
2568 | unsigned long nr_reclaimed = 0; | |
2569 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2570 | unsigned long reclaimed; | |
2571 | int loop = 0; | |
2572 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2573 | unsigned long excess; |
0608f43d AM |
2574 | unsigned long nr_scanned; |
2575 | ||
2576 | if (order > 0) | |
2577 | return 0; | |
2578 | ||
2579 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2580 | /* | |
2581 | * This loop can run a while, specially if mem_cgroup's continuously | |
2582 | * keep exceeding their soft limit and putting the system under | |
2583 | * pressure | |
2584 | */ | |
2585 | do { | |
2586 | if (next_mz) | |
2587 | mz = next_mz; | |
2588 | else | |
2589 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2590 | if (!mz) | |
2591 | break; | |
2592 | ||
2593 | nr_scanned = 0; | |
2594 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2595 | gfp_mask, &nr_scanned); | |
2596 | nr_reclaimed += reclaimed; | |
2597 | *total_scanned += nr_scanned; | |
0a31bc97 | 2598 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2599 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2600 | |
2601 | /* | |
2602 | * If we failed to reclaim anything from this memory cgroup | |
2603 | * it is time to move on to the next cgroup | |
2604 | */ | |
2605 | next_mz = NULL; | |
bc2f2e7f VD |
2606 | if (!reclaimed) |
2607 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2608 | ||
3e32cb2e | 2609 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2610 | /* |
2611 | * One school of thought says that we should not add | |
2612 | * back the node to the tree if reclaim returns 0. | |
2613 | * But our reclaim could return 0, simply because due | |
2614 | * to priority we are exposing a smaller subset of | |
2615 | * memory to reclaim from. Consider this as a longer | |
2616 | * term TODO. | |
2617 | */ | |
2618 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2619 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2620 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2621 | css_put(&mz->memcg->css); |
2622 | loop++; | |
2623 | /* | |
2624 | * Could not reclaim anything and there are no more | |
2625 | * mem cgroups to try or we seem to be looping without | |
2626 | * reclaiming anything. | |
2627 | */ | |
2628 | if (!nr_reclaimed && | |
2629 | (next_mz == NULL || | |
2630 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2631 | break; | |
2632 | } while (!nr_reclaimed); | |
2633 | if (next_mz) | |
2634 | css_put(&next_mz->memcg->css); | |
2635 | return nr_reclaimed; | |
2636 | } | |
2637 | ||
ea280e7b TH |
2638 | /* |
2639 | * Test whether @memcg has children, dead or alive. Note that this | |
2640 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2641 | * returns %true if there are child csses according to the cgroup | |
2642 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2643 | */ | |
b5f99b53 GC |
2644 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2645 | { | |
ea280e7b TH |
2646 | bool ret; |
2647 | ||
ea280e7b TH |
2648 | rcu_read_lock(); |
2649 | ret = css_next_child(NULL, &memcg->css); | |
2650 | rcu_read_unlock(); | |
2651 | return ret; | |
b5f99b53 GC |
2652 | } |
2653 | ||
c26251f9 MH |
2654 | /* |
2655 | * Reclaims as many pages from the given memcg as possible and moves | |
2656 | * the rest to the parent. | |
2657 | * | |
2658 | * Caller is responsible for holding css reference for memcg. | |
2659 | */ | |
2660 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2661 | { | |
2662 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2663 | |
c1e862c1 KH |
2664 | /* we call try-to-free pages for make this cgroup empty */ |
2665 | lru_add_drain_all(); | |
f817ed48 | 2666 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2667 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2668 | int progress; |
c1e862c1 | 2669 | |
c26251f9 MH |
2670 | if (signal_pending(current)) |
2671 | return -EINTR; | |
2672 | ||
b70a2a21 JW |
2673 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2674 | GFP_KERNEL, true); | |
c1e862c1 | 2675 | if (!progress) { |
f817ed48 | 2676 | nr_retries--; |
c1e862c1 | 2677 | /* maybe some writeback is necessary */ |
8aa7e847 | 2678 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2679 | } |
f817ed48 KH |
2680 | |
2681 | } | |
ab5196c2 MH |
2682 | |
2683 | return 0; | |
cc847582 KH |
2684 | } |
2685 | ||
6770c64e TH |
2686 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2687 | char *buf, size_t nbytes, | |
2688 | loff_t off) | |
c1e862c1 | 2689 | { |
6770c64e | 2690 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2691 | |
d8423011 MH |
2692 | if (mem_cgroup_is_root(memcg)) |
2693 | return -EINVAL; | |
6770c64e | 2694 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2695 | } |
2696 | ||
182446d0 TH |
2697 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2698 | struct cftype *cft) | |
18f59ea7 | 2699 | { |
182446d0 | 2700 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2701 | } |
2702 | ||
182446d0 TH |
2703 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2704 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2705 | { |
2706 | int retval = 0; | |
182446d0 | 2707 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2708 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2709 | |
567fb435 | 2710 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2711 | return 0; |
567fb435 | 2712 | |
18f59ea7 | 2713 | /* |
af901ca1 | 2714 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2715 | * in the child subtrees. If it is unset, then the change can |
2716 | * occur, provided the current cgroup has no children. | |
2717 | * | |
2718 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2719 | * set if there are no children. | |
2720 | */ | |
c0ff4b85 | 2721 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2722 | (val == 1 || val == 0)) { |
ea280e7b | 2723 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2724 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2725 | else |
2726 | retval = -EBUSY; | |
2727 | } else | |
2728 | retval = -EINVAL; | |
567fb435 | 2729 | |
18f59ea7 BS |
2730 | return retval; |
2731 | } | |
2732 | ||
72b54e73 | 2733 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
ce00a967 JW |
2734 | { |
2735 | struct mem_cgroup *iter; | |
72b54e73 | 2736 | int i; |
ce00a967 | 2737 | |
72b54e73 | 2738 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
ce00a967 | 2739 | |
72b54e73 VD |
2740 | for_each_mem_cgroup_tree(iter, memcg) { |
2741 | for (i = 0; i < MEMCG_NR_STAT; i++) | |
2742 | stat[i] += mem_cgroup_read_stat(iter, i); | |
2743 | } | |
ce00a967 JW |
2744 | } |
2745 | ||
72b54e73 | 2746 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
587d9f72 JW |
2747 | { |
2748 | struct mem_cgroup *iter; | |
72b54e73 | 2749 | int i; |
587d9f72 | 2750 | |
72b54e73 | 2751 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
587d9f72 | 2752 | |
72b54e73 VD |
2753 | for_each_mem_cgroup_tree(iter, memcg) { |
2754 | for (i = 0; i < MEMCG_NR_EVENTS; i++) | |
2755 | events[i] += mem_cgroup_read_events(iter, i); | |
2756 | } | |
587d9f72 JW |
2757 | } |
2758 | ||
6f646156 | 2759 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2760 | { |
72b54e73 | 2761 | unsigned long val = 0; |
ce00a967 | 2762 | |
3e32cb2e | 2763 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2764 | struct mem_cgroup *iter; |
2765 | ||
2766 | for_each_mem_cgroup_tree(iter, memcg) { | |
2767 | val += mem_cgroup_read_stat(iter, | |
2768 | MEM_CGROUP_STAT_CACHE); | |
2769 | val += mem_cgroup_read_stat(iter, | |
2770 | MEM_CGROUP_STAT_RSS); | |
2771 | if (swap) | |
2772 | val += mem_cgroup_read_stat(iter, | |
2773 | MEM_CGROUP_STAT_SWAP); | |
2774 | } | |
3e32cb2e | 2775 | } else { |
ce00a967 | 2776 | if (!swap) |
3e32cb2e | 2777 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2778 | else |
3e32cb2e | 2779 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2780 | } |
c12176d3 | 2781 | return val; |
ce00a967 JW |
2782 | } |
2783 | ||
3e32cb2e JW |
2784 | enum { |
2785 | RES_USAGE, | |
2786 | RES_LIMIT, | |
2787 | RES_MAX_USAGE, | |
2788 | RES_FAILCNT, | |
2789 | RES_SOFT_LIMIT, | |
2790 | }; | |
ce00a967 | 2791 | |
791badbd | 2792 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2793 | struct cftype *cft) |
8cdea7c0 | 2794 | { |
182446d0 | 2795 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2796 | struct page_counter *counter; |
af36f906 | 2797 | |
3e32cb2e | 2798 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2799 | case _MEM: |
3e32cb2e JW |
2800 | counter = &memcg->memory; |
2801 | break; | |
8c7c6e34 | 2802 | case _MEMSWAP: |
3e32cb2e JW |
2803 | counter = &memcg->memsw; |
2804 | break; | |
510fc4e1 | 2805 | case _KMEM: |
3e32cb2e | 2806 | counter = &memcg->kmem; |
510fc4e1 | 2807 | break; |
d55f90bf | 2808 | case _TCP: |
0db15298 | 2809 | counter = &memcg->tcpmem; |
d55f90bf | 2810 | break; |
8c7c6e34 KH |
2811 | default: |
2812 | BUG(); | |
8c7c6e34 | 2813 | } |
3e32cb2e JW |
2814 | |
2815 | switch (MEMFILE_ATTR(cft->private)) { | |
2816 | case RES_USAGE: | |
2817 | if (counter == &memcg->memory) | |
c12176d3 | 2818 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2819 | if (counter == &memcg->memsw) |
c12176d3 | 2820 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2821 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2822 | case RES_LIMIT: | |
2823 | return (u64)counter->limit * PAGE_SIZE; | |
2824 | case RES_MAX_USAGE: | |
2825 | return (u64)counter->watermark * PAGE_SIZE; | |
2826 | case RES_FAILCNT: | |
2827 | return counter->failcnt; | |
2828 | case RES_SOFT_LIMIT: | |
2829 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2830 | default: | |
2831 | BUG(); | |
2832 | } | |
8cdea7c0 | 2833 | } |
510fc4e1 | 2834 | |
127424c8 | 2835 | #ifndef CONFIG_SLOB |
567e9ab2 | 2836 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2837 | { |
d6441637 VD |
2838 | int memcg_id; |
2839 | ||
b313aeee VD |
2840 | if (cgroup_memory_nokmem) |
2841 | return 0; | |
2842 | ||
2a4db7eb | 2843 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2844 | BUG_ON(memcg->kmem_state); |
d6441637 | 2845 | |
f3bb3043 | 2846 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
2847 | if (memcg_id < 0) |
2848 | return memcg_id; | |
d6441637 | 2849 | |
ef12947c | 2850 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2851 | /* |
567e9ab2 | 2852 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2853 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2854 | * guarantee no one starts accounting before all call sites are |
2855 | * patched. | |
2856 | */ | |
900a38f0 | 2857 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2858 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
2859 | |
2860 | return 0; | |
d6441637 VD |
2861 | } |
2862 | ||
8e0a8912 JW |
2863 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2864 | { | |
2865 | struct cgroup_subsys_state *css; | |
2866 | struct mem_cgroup *parent, *child; | |
2867 | int kmemcg_id; | |
2868 | ||
2869 | if (memcg->kmem_state != KMEM_ONLINE) | |
2870 | return; | |
2871 | /* | |
2872 | * Clear the online state before clearing memcg_caches array | |
2873 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2874 | * guarantees that no cache will be created for this cgroup | |
2875 | * after we are done (see memcg_create_kmem_cache()). | |
2876 | */ | |
2877 | memcg->kmem_state = KMEM_ALLOCATED; | |
2878 | ||
2879 | memcg_deactivate_kmem_caches(memcg); | |
2880 | ||
2881 | kmemcg_id = memcg->kmemcg_id; | |
2882 | BUG_ON(kmemcg_id < 0); | |
2883 | ||
2884 | parent = parent_mem_cgroup(memcg); | |
2885 | if (!parent) | |
2886 | parent = root_mem_cgroup; | |
2887 | ||
2888 | /* | |
2889 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2890 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2891 | * to ones of the parent. After we have finished, all list_lrus | |
2892 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2893 | * ordering is imposed by list_lru_node->lock taken by | |
2894 | * memcg_drain_all_list_lrus(). | |
2895 | */ | |
2896 | css_for_each_descendant_pre(css, &memcg->css) { | |
2897 | child = mem_cgroup_from_css(css); | |
2898 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2899 | child->kmemcg_id = parent->kmemcg_id; | |
2900 | if (!memcg->use_hierarchy) | |
2901 | break; | |
2902 | } | |
2903 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
2904 | ||
2905 | memcg_free_cache_id(kmemcg_id); | |
2906 | } | |
2907 | ||
2908 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2909 | { | |
0b8f73e1 JW |
2910 | /* css_alloc() failed, offlining didn't happen */ |
2911 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
2912 | memcg_offline_kmem(memcg); | |
2913 | ||
8e0a8912 JW |
2914 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
2915 | memcg_destroy_kmem_caches(memcg); | |
2916 | static_branch_dec(&memcg_kmem_enabled_key); | |
2917 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2918 | } | |
8e0a8912 | 2919 | } |
d6441637 | 2920 | #else |
0b8f73e1 | 2921 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
2922 | { |
2923 | return 0; | |
2924 | } | |
2925 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
2926 | { | |
2927 | } | |
2928 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2929 | { | |
2930 | } | |
2931 | #endif /* !CONFIG_SLOB */ | |
2932 | ||
d6441637 | 2933 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2934 | unsigned long limit) |
d6441637 | 2935 | { |
b313aeee | 2936 | int ret; |
127424c8 JW |
2937 | |
2938 | mutex_lock(&memcg_limit_mutex); | |
127424c8 | 2939 | ret = page_counter_limit(&memcg->kmem, limit); |
127424c8 JW |
2940 | mutex_unlock(&memcg_limit_mutex); |
2941 | return ret; | |
d6441637 | 2942 | } |
510fc4e1 | 2943 | |
d55f90bf VD |
2944 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
2945 | { | |
2946 | int ret; | |
2947 | ||
2948 | mutex_lock(&memcg_limit_mutex); | |
2949 | ||
0db15298 | 2950 | ret = page_counter_limit(&memcg->tcpmem, limit); |
d55f90bf VD |
2951 | if (ret) |
2952 | goto out; | |
2953 | ||
0db15298 | 2954 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
2955 | /* |
2956 | * The active flag needs to be written after the static_key | |
2957 | * update. This is what guarantees that the socket activation | |
2958 | * function is the last one to run. See sock_update_memcg() for | |
2959 | * details, and note that we don't mark any socket as belonging | |
2960 | * to this memcg until that flag is up. | |
2961 | * | |
2962 | * We need to do this, because static_keys will span multiple | |
2963 | * sites, but we can't control their order. If we mark a socket | |
2964 | * as accounted, but the accounting functions are not patched in | |
2965 | * yet, we'll lose accounting. | |
2966 | * | |
2967 | * We never race with the readers in sock_update_memcg(), | |
2968 | * because when this value change, the code to process it is not | |
2969 | * patched in yet. | |
2970 | */ | |
2971 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 2972 | memcg->tcpmem_active = true; |
d55f90bf VD |
2973 | } |
2974 | out: | |
2975 | mutex_unlock(&memcg_limit_mutex); | |
2976 | return ret; | |
2977 | } | |
d55f90bf | 2978 | |
628f4235 KH |
2979 | /* |
2980 | * The user of this function is... | |
2981 | * RES_LIMIT. | |
2982 | */ | |
451af504 TH |
2983 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2984 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2985 | { |
451af504 | 2986 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2987 | unsigned long nr_pages; |
628f4235 KH |
2988 | int ret; |
2989 | ||
451af504 | 2990 | buf = strstrip(buf); |
650c5e56 | 2991 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
2992 | if (ret) |
2993 | return ret; | |
af36f906 | 2994 | |
3e32cb2e | 2995 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 2996 | case RES_LIMIT: |
4b3bde4c BS |
2997 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2998 | ret = -EINVAL; | |
2999 | break; | |
3000 | } | |
3e32cb2e JW |
3001 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3002 | case _MEM: | |
3003 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3004 | break; |
3e32cb2e JW |
3005 | case _MEMSWAP: |
3006 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3007 | break; |
3e32cb2e JW |
3008 | case _KMEM: |
3009 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3010 | break; | |
d55f90bf VD |
3011 | case _TCP: |
3012 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
3013 | break; | |
3e32cb2e | 3014 | } |
296c81d8 | 3015 | break; |
3e32cb2e JW |
3016 | case RES_SOFT_LIMIT: |
3017 | memcg->soft_limit = nr_pages; | |
3018 | ret = 0; | |
628f4235 KH |
3019 | break; |
3020 | } | |
451af504 | 3021 | return ret ?: nbytes; |
8cdea7c0 BS |
3022 | } |
3023 | ||
6770c64e TH |
3024 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3025 | size_t nbytes, loff_t off) | |
c84872e1 | 3026 | { |
6770c64e | 3027 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3028 | struct page_counter *counter; |
c84872e1 | 3029 | |
3e32cb2e JW |
3030 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3031 | case _MEM: | |
3032 | counter = &memcg->memory; | |
3033 | break; | |
3034 | case _MEMSWAP: | |
3035 | counter = &memcg->memsw; | |
3036 | break; | |
3037 | case _KMEM: | |
3038 | counter = &memcg->kmem; | |
3039 | break; | |
d55f90bf | 3040 | case _TCP: |
0db15298 | 3041 | counter = &memcg->tcpmem; |
d55f90bf | 3042 | break; |
3e32cb2e JW |
3043 | default: |
3044 | BUG(); | |
3045 | } | |
af36f906 | 3046 | |
3e32cb2e | 3047 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3048 | case RES_MAX_USAGE: |
3e32cb2e | 3049 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3050 | break; |
3051 | case RES_FAILCNT: | |
3e32cb2e | 3052 | counter->failcnt = 0; |
29f2a4da | 3053 | break; |
3e32cb2e JW |
3054 | default: |
3055 | BUG(); | |
29f2a4da | 3056 | } |
f64c3f54 | 3057 | |
6770c64e | 3058 | return nbytes; |
c84872e1 PE |
3059 | } |
3060 | ||
182446d0 | 3061 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3062 | struct cftype *cft) |
3063 | { | |
182446d0 | 3064 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3065 | } |
3066 | ||
02491447 | 3067 | #ifdef CONFIG_MMU |
182446d0 | 3068 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3069 | struct cftype *cft, u64 val) |
3070 | { | |
182446d0 | 3071 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3072 | |
1dfab5ab | 3073 | if (val & ~MOVE_MASK) |
7dc74be0 | 3074 | return -EINVAL; |
ee5e8472 | 3075 | |
7dc74be0 | 3076 | /* |
ee5e8472 GC |
3077 | * No kind of locking is needed in here, because ->can_attach() will |
3078 | * check this value once in the beginning of the process, and then carry | |
3079 | * on with stale data. This means that changes to this value will only | |
3080 | * affect task migrations starting after the change. | |
7dc74be0 | 3081 | */ |
c0ff4b85 | 3082 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3083 | return 0; |
3084 | } | |
02491447 | 3085 | #else |
182446d0 | 3086 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3087 | struct cftype *cft, u64 val) |
3088 | { | |
3089 | return -ENOSYS; | |
3090 | } | |
3091 | #endif | |
7dc74be0 | 3092 | |
406eb0c9 | 3093 | #ifdef CONFIG_NUMA |
2da8ca82 | 3094 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3095 | { |
25485de6 GT |
3096 | struct numa_stat { |
3097 | const char *name; | |
3098 | unsigned int lru_mask; | |
3099 | }; | |
3100 | ||
3101 | static const struct numa_stat stats[] = { | |
3102 | { "total", LRU_ALL }, | |
3103 | { "file", LRU_ALL_FILE }, | |
3104 | { "anon", LRU_ALL_ANON }, | |
3105 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3106 | }; | |
3107 | const struct numa_stat *stat; | |
406eb0c9 | 3108 | int nid; |
25485de6 | 3109 | unsigned long nr; |
2da8ca82 | 3110 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3111 | |
25485de6 GT |
3112 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3113 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3114 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3115 | for_each_node_state(nid, N_MEMORY) { | |
3116 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3117 | stat->lru_mask); | |
3118 | seq_printf(m, " N%d=%lu", nid, nr); | |
3119 | } | |
3120 | seq_putc(m, '\n'); | |
406eb0c9 | 3121 | } |
406eb0c9 | 3122 | |
071aee13 YH |
3123 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3124 | struct mem_cgroup *iter; | |
3125 | ||
3126 | nr = 0; | |
3127 | for_each_mem_cgroup_tree(iter, memcg) | |
3128 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3129 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3130 | for_each_node_state(nid, N_MEMORY) { | |
3131 | nr = 0; | |
3132 | for_each_mem_cgroup_tree(iter, memcg) | |
3133 | nr += mem_cgroup_node_nr_lru_pages( | |
3134 | iter, nid, stat->lru_mask); | |
3135 | seq_printf(m, " N%d=%lu", nid, nr); | |
3136 | } | |
3137 | seq_putc(m, '\n'); | |
406eb0c9 | 3138 | } |
406eb0c9 | 3139 | |
406eb0c9 YH |
3140 | return 0; |
3141 | } | |
3142 | #endif /* CONFIG_NUMA */ | |
3143 | ||
2da8ca82 | 3144 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3145 | { |
2da8ca82 | 3146 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3147 | unsigned long memory, memsw; |
af7c4b0e JW |
3148 | struct mem_cgroup *mi; |
3149 | unsigned int i; | |
406eb0c9 | 3150 | |
0ca44b14 GT |
3151 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3152 | MEM_CGROUP_STAT_NSTATS); | |
3153 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3154 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3155 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3156 | ||
af7c4b0e | 3157 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
7941d214 | 3158 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3159 | continue; |
484ebb3b | 3160 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3161 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3162 | } |
7b854121 | 3163 | |
af7c4b0e JW |
3164 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3165 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3166 | mem_cgroup_read_events(memcg, i)); | |
3167 | ||
3168 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3169 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3170 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3171 | ||
14067bb3 | 3172 | /* Hierarchical information */ |
3e32cb2e JW |
3173 | memory = memsw = PAGE_COUNTER_MAX; |
3174 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3175 | memory = min(memory, mi->memory.limit); | |
3176 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3177 | } |
3e32cb2e JW |
3178 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3179 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3180 | if (do_memsw_account()) |
3e32cb2e JW |
3181 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3182 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3183 | |
af7c4b0e | 3184 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3185 | unsigned long long val = 0; |
af7c4b0e | 3186 | |
7941d214 | 3187 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3188 | continue; |
af7c4b0e JW |
3189 | for_each_mem_cgroup_tree(mi, memcg) |
3190 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3191 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3192 | } |
3193 | ||
3194 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3195 | unsigned long long val = 0; | |
3196 | ||
3197 | for_each_mem_cgroup_tree(mi, memcg) | |
3198 | val += mem_cgroup_read_events(mi, i); | |
3199 | seq_printf(m, "total_%s %llu\n", | |
3200 | mem_cgroup_events_names[i], val); | |
3201 | } | |
3202 | ||
3203 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3204 | unsigned long long val = 0; | |
3205 | ||
3206 | for_each_mem_cgroup_tree(mi, memcg) | |
3207 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3208 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3209 | } |
14067bb3 | 3210 | |
7f016ee8 | 3211 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3212 | { |
3213 | int nid, zid; | |
3214 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3215 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3216 | unsigned long recent_rotated[2] = {0, 0}; |
3217 | unsigned long recent_scanned[2] = {0, 0}; | |
3218 | ||
3219 | for_each_online_node(nid) | |
3220 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3221 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3222 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3223 | |
89abfab1 HD |
3224 | recent_rotated[0] += rstat->recent_rotated[0]; |
3225 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3226 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3227 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3228 | } |
78ccf5b5 JW |
3229 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3230 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3231 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3232 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3233 | } |
3234 | #endif | |
3235 | ||
d2ceb9b7 KH |
3236 | return 0; |
3237 | } | |
3238 | ||
182446d0 TH |
3239 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3240 | struct cftype *cft) | |
a7885eb8 | 3241 | { |
182446d0 | 3242 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3243 | |
1f4c025b | 3244 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3245 | } |
3246 | ||
182446d0 TH |
3247 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3248 | struct cftype *cft, u64 val) | |
a7885eb8 | 3249 | { |
182446d0 | 3250 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3251 | |
3dae7fec | 3252 | if (val > 100) |
a7885eb8 KM |
3253 | return -EINVAL; |
3254 | ||
14208b0e | 3255 | if (css->parent) |
3dae7fec JW |
3256 | memcg->swappiness = val; |
3257 | else | |
3258 | vm_swappiness = val; | |
068b38c1 | 3259 | |
a7885eb8 KM |
3260 | return 0; |
3261 | } | |
3262 | ||
2e72b634 KS |
3263 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3264 | { | |
3265 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3266 | unsigned long usage; |
2e72b634 KS |
3267 | int i; |
3268 | ||
3269 | rcu_read_lock(); | |
3270 | if (!swap) | |
2c488db2 | 3271 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3272 | else |
2c488db2 | 3273 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3274 | |
3275 | if (!t) | |
3276 | goto unlock; | |
3277 | ||
ce00a967 | 3278 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3279 | |
3280 | /* | |
748dad36 | 3281 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3282 | * If it's not true, a threshold was crossed after last |
3283 | * call of __mem_cgroup_threshold(). | |
3284 | */ | |
5407a562 | 3285 | i = t->current_threshold; |
2e72b634 KS |
3286 | |
3287 | /* | |
3288 | * Iterate backward over array of thresholds starting from | |
3289 | * current_threshold and check if a threshold is crossed. | |
3290 | * If none of thresholds below usage is crossed, we read | |
3291 | * only one element of the array here. | |
3292 | */ | |
3293 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3294 | eventfd_signal(t->entries[i].eventfd, 1); | |
3295 | ||
3296 | /* i = current_threshold + 1 */ | |
3297 | i++; | |
3298 | ||
3299 | /* | |
3300 | * Iterate forward over array of thresholds starting from | |
3301 | * current_threshold+1 and check if a threshold is crossed. | |
3302 | * If none of thresholds above usage is crossed, we read | |
3303 | * only one element of the array here. | |
3304 | */ | |
3305 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3306 | eventfd_signal(t->entries[i].eventfd, 1); | |
3307 | ||
3308 | /* Update current_threshold */ | |
5407a562 | 3309 | t->current_threshold = i - 1; |
2e72b634 KS |
3310 | unlock: |
3311 | rcu_read_unlock(); | |
3312 | } | |
3313 | ||
3314 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3315 | { | |
ad4ca5f4 KS |
3316 | while (memcg) { |
3317 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3318 | if (do_memsw_account()) |
ad4ca5f4 KS |
3319 | __mem_cgroup_threshold(memcg, true); |
3320 | ||
3321 | memcg = parent_mem_cgroup(memcg); | |
3322 | } | |
2e72b634 KS |
3323 | } |
3324 | ||
3325 | static int compare_thresholds(const void *a, const void *b) | |
3326 | { | |
3327 | const struct mem_cgroup_threshold *_a = a; | |
3328 | const struct mem_cgroup_threshold *_b = b; | |
3329 | ||
2bff24a3 GT |
3330 | if (_a->threshold > _b->threshold) |
3331 | return 1; | |
3332 | ||
3333 | if (_a->threshold < _b->threshold) | |
3334 | return -1; | |
3335 | ||
3336 | return 0; | |
2e72b634 KS |
3337 | } |
3338 | ||
c0ff4b85 | 3339 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3340 | { |
3341 | struct mem_cgroup_eventfd_list *ev; | |
3342 | ||
2bcf2e92 MH |
3343 | spin_lock(&memcg_oom_lock); |
3344 | ||
c0ff4b85 | 3345 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3346 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3347 | |
3348 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3349 | return 0; |
3350 | } | |
3351 | ||
c0ff4b85 | 3352 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3353 | { |
7d74b06f KH |
3354 | struct mem_cgroup *iter; |
3355 | ||
c0ff4b85 | 3356 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3357 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3358 | } |
3359 | ||
59b6f873 | 3360 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3361 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3362 | { |
2c488db2 KS |
3363 | struct mem_cgroup_thresholds *thresholds; |
3364 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3365 | unsigned long threshold; |
3366 | unsigned long usage; | |
2c488db2 | 3367 | int i, size, ret; |
2e72b634 | 3368 | |
650c5e56 | 3369 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3370 | if (ret) |
3371 | return ret; | |
3372 | ||
3373 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3374 | |
05b84301 | 3375 | if (type == _MEM) { |
2c488db2 | 3376 | thresholds = &memcg->thresholds; |
ce00a967 | 3377 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3378 | } else if (type == _MEMSWAP) { |
2c488db2 | 3379 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3380 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3381 | } else |
2e72b634 KS |
3382 | BUG(); |
3383 | ||
2e72b634 | 3384 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3385 | if (thresholds->primary) |
2e72b634 KS |
3386 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3387 | ||
2c488db2 | 3388 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3389 | |
3390 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3391 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3392 | GFP_KERNEL); |
2c488db2 | 3393 | if (!new) { |
2e72b634 KS |
3394 | ret = -ENOMEM; |
3395 | goto unlock; | |
3396 | } | |
2c488db2 | 3397 | new->size = size; |
2e72b634 KS |
3398 | |
3399 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3400 | if (thresholds->primary) { |
3401 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3402 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3403 | } |
3404 | ||
2e72b634 | 3405 | /* Add new threshold */ |
2c488db2 KS |
3406 | new->entries[size - 1].eventfd = eventfd; |
3407 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3408 | |
3409 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3410 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3411 | compare_thresholds, NULL); |
3412 | ||
3413 | /* Find current threshold */ | |
2c488db2 | 3414 | new->current_threshold = -1; |
2e72b634 | 3415 | for (i = 0; i < size; i++) { |
748dad36 | 3416 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3417 | /* |
2c488db2 KS |
3418 | * new->current_threshold will not be used until |
3419 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3420 | * it here. |
3421 | */ | |
2c488db2 | 3422 | ++new->current_threshold; |
748dad36 SZ |
3423 | } else |
3424 | break; | |
2e72b634 KS |
3425 | } |
3426 | ||
2c488db2 KS |
3427 | /* Free old spare buffer and save old primary buffer as spare */ |
3428 | kfree(thresholds->spare); | |
3429 | thresholds->spare = thresholds->primary; | |
3430 | ||
3431 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3432 | |
907860ed | 3433 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3434 | synchronize_rcu(); |
3435 | ||
2e72b634 KS |
3436 | unlock: |
3437 | mutex_unlock(&memcg->thresholds_lock); | |
3438 | ||
3439 | return ret; | |
3440 | } | |
3441 | ||
59b6f873 | 3442 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3443 | struct eventfd_ctx *eventfd, const char *args) |
3444 | { | |
59b6f873 | 3445 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3446 | } |
3447 | ||
59b6f873 | 3448 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3449 | struct eventfd_ctx *eventfd, const char *args) |
3450 | { | |
59b6f873 | 3451 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3452 | } |
3453 | ||
59b6f873 | 3454 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3455 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3456 | { |
2c488db2 KS |
3457 | struct mem_cgroup_thresholds *thresholds; |
3458 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3459 | unsigned long usage; |
2c488db2 | 3460 | int i, j, size; |
2e72b634 KS |
3461 | |
3462 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3463 | |
3464 | if (type == _MEM) { | |
2c488db2 | 3465 | thresholds = &memcg->thresholds; |
ce00a967 | 3466 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3467 | } else if (type == _MEMSWAP) { |
2c488db2 | 3468 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3469 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3470 | } else |
2e72b634 KS |
3471 | BUG(); |
3472 | ||
371528ca AV |
3473 | if (!thresholds->primary) |
3474 | goto unlock; | |
3475 | ||
2e72b634 KS |
3476 | /* Check if a threshold crossed before removing */ |
3477 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3478 | ||
3479 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3480 | size = 0; |
3481 | for (i = 0; i < thresholds->primary->size; i++) { | |
3482 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3483 | size++; |
3484 | } | |
3485 | ||
2c488db2 | 3486 | new = thresholds->spare; |
907860ed | 3487 | |
2e72b634 KS |
3488 | /* Set thresholds array to NULL if we don't have thresholds */ |
3489 | if (!size) { | |
2c488db2 KS |
3490 | kfree(new); |
3491 | new = NULL; | |
907860ed | 3492 | goto swap_buffers; |
2e72b634 KS |
3493 | } |
3494 | ||
2c488db2 | 3495 | new->size = size; |
2e72b634 KS |
3496 | |
3497 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3498 | new->current_threshold = -1; |
3499 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3500 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3501 | continue; |
3502 | ||
2c488db2 | 3503 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3504 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3505 | /* |
2c488db2 | 3506 | * new->current_threshold will not be used |
2e72b634 KS |
3507 | * until rcu_assign_pointer(), so it's safe to increment |
3508 | * it here. | |
3509 | */ | |
2c488db2 | 3510 | ++new->current_threshold; |
2e72b634 KS |
3511 | } |
3512 | j++; | |
3513 | } | |
3514 | ||
907860ed | 3515 | swap_buffers: |
2c488db2 KS |
3516 | /* Swap primary and spare array */ |
3517 | thresholds->spare = thresholds->primary; | |
8c757763 | 3518 | |
2c488db2 | 3519 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3520 | |
907860ed | 3521 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3522 | synchronize_rcu(); |
6611d8d7 MC |
3523 | |
3524 | /* If all events are unregistered, free the spare array */ | |
3525 | if (!new) { | |
3526 | kfree(thresholds->spare); | |
3527 | thresholds->spare = NULL; | |
3528 | } | |
371528ca | 3529 | unlock: |
2e72b634 | 3530 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3531 | } |
c1e862c1 | 3532 | |
59b6f873 | 3533 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3534 | struct eventfd_ctx *eventfd) |
3535 | { | |
59b6f873 | 3536 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3537 | } |
3538 | ||
59b6f873 | 3539 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3540 | struct eventfd_ctx *eventfd) |
3541 | { | |
59b6f873 | 3542 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3543 | } |
3544 | ||
59b6f873 | 3545 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3546 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3547 | { |
9490ff27 | 3548 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3549 | |
9490ff27 KH |
3550 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3551 | if (!event) | |
3552 | return -ENOMEM; | |
3553 | ||
1af8efe9 | 3554 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3555 | |
3556 | event->eventfd = eventfd; | |
3557 | list_add(&event->list, &memcg->oom_notify); | |
3558 | ||
3559 | /* already in OOM ? */ | |
c2b42d3c | 3560 | if (memcg->under_oom) |
9490ff27 | 3561 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3562 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3563 | |
3564 | return 0; | |
3565 | } | |
3566 | ||
59b6f873 | 3567 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3568 | struct eventfd_ctx *eventfd) |
9490ff27 | 3569 | { |
9490ff27 | 3570 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3571 | |
1af8efe9 | 3572 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3573 | |
c0ff4b85 | 3574 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3575 | if (ev->eventfd == eventfd) { |
3576 | list_del(&ev->list); | |
3577 | kfree(ev); | |
3578 | } | |
3579 | } | |
3580 | ||
1af8efe9 | 3581 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3582 | } |
3583 | ||
2da8ca82 | 3584 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3585 | { |
2da8ca82 | 3586 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3587 | |
791badbd | 3588 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3589 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3590 | return 0; |
3591 | } | |
3592 | ||
182446d0 | 3593 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3594 | struct cftype *cft, u64 val) |
3595 | { | |
182446d0 | 3596 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3597 | |
3598 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3599 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3600 | return -EINVAL; |
3601 | ||
c0ff4b85 | 3602 | memcg->oom_kill_disable = val; |
4d845ebf | 3603 | if (!val) |
c0ff4b85 | 3604 | memcg_oom_recover(memcg); |
3dae7fec | 3605 | |
3c11ecf4 KH |
3606 | return 0; |
3607 | } | |
3608 | ||
52ebea74 TH |
3609 | #ifdef CONFIG_CGROUP_WRITEBACK |
3610 | ||
3611 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3612 | { | |
3613 | return &memcg->cgwb_list; | |
3614 | } | |
3615 | ||
841710aa TH |
3616 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3617 | { | |
3618 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3619 | } | |
3620 | ||
3621 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3622 | { | |
3623 | wb_domain_exit(&memcg->cgwb_domain); | |
3624 | } | |
3625 | ||
2529bb3a TH |
3626 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3627 | { | |
3628 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3629 | } | |
3630 | ||
841710aa TH |
3631 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3632 | { | |
3633 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3634 | ||
3635 | if (!memcg->css.parent) | |
3636 | return NULL; | |
3637 | ||
3638 | return &memcg->cgwb_domain; | |
3639 | } | |
3640 | ||
c2aa723a TH |
3641 | /** |
3642 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3643 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3644 | * @pfilepages: out parameter for number of file pages |
3645 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3646 | * @pdirty: out parameter for number of dirty pages |
3647 | * @pwriteback: out parameter for number of pages under writeback | |
3648 | * | |
c5edf9cd TH |
3649 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3650 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3651 | * is a bit more involved. | |
c2aa723a | 3652 | * |
c5edf9cd TH |
3653 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3654 | * headroom is calculated as the lowest headroom of itself and the | |
3655 | * ancestors. Note that this doesn't consider the actual amount of | |
3656 | * available memory in the system. The caller should further cap | |
3657 | * *@pheadroom accordingly. | |
c2aa723a | 3658 | */ |
c5edf9cd TH |
3659 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3660 | unsigned long *pheadroom, unsigned long *pdirty, | |
3661 | unsigned long *pwriteback) | |
c2aa723a TH |
3662 | { |
3663 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3664 | struct mem_cgroup *parent; | |
c2aa723a TH |
3665 | |
3666 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3667 | ||
3668 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3669 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3670 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3671 | (1 << LRU_ACTIVE_FILE)); | |
3672 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3673 | |
c2aa723a TH |
3674 | while ((parent = parent_mem_cgroup(memcg))) { |
3675 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3676 | unsigned long used = page_counter_read(&memcg->memory); | |
3677 | ||
c5edf9cd | 3678 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3679 | memcg = parent; |
3680 | } | |
c2aa723a TH |
3681 | } |
3682 | ||
841710aa TH |
3683 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3684 | ||
3685 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3686 | { | |
3687 | return 0; | |
3688 | } | |
3689 | ||
3690 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3691 | { | |
3692 | } | |
3693 | ||
2529bb3a TH |
3694 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3695 | { | |
3696 | } | |
3697 | ||
52ebea74 TH |
3698 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3699 | ||
3bc942f3 TH |
3700 | /* |
3701 | * DO NOT USE IN NEW FILES. | |
3702 | * | |
3703 | * "cgroup.event_control" implementation. | |
3704 | * | |
3705 | * This is way over-engineered. It tries to support fully configurable | |
3706 | * events for each user. Such level of flexibility is completely | |
3707 | * unnecessary especially in the light of the planned unified hierarchy. | |
3708 | * | |
3709 | * Please deprecate this and replace with something simpler if at all | |
3710 | * possible. | |
3711 | */ | |
3712 | ||
79bd9814 TH |
3713 | /* |
3714 | * Unregister event and free resources. | |
3715 | * | |
3716 | * Gets called from workqueue. | |
3717 | */ | |
3bc942f3 | 3718 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3719 | { |
3bc942f3 TH |
3720 | struct mem_cgroup_event *event = |
3721 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3722 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3723 | |
3724 | remove_wait_queue(event->wqh, &event->wait); | |
3725 | ||
59b6f873 | 3726 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3727 | |
3728 | /* Notify userspace the event is going away. */ | |
3729 | eventfd_signal(event->eventfd, 1); | |
3730 | ||
3731 | eventfd_ctx_put(event->eventfd); | |
3732 | kfree(event); | |
59b6f873 | 3733 | css_put(&memcg->css); |
79bd9814 TH |
3734 | } |
3735 | ||
3736 | /* | |
3737 | * Gets called on POLLHUP on eventfd when user closes it. | |
3738 | * | |
3739 | * Called with wqh->lock held and interrupts disabled. | |
3740 | */ | |
3bc942f3 TH |
3741 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3742 | int sync, void *key) | |
79bd9814 | 3743 | { |
3bc942f3 TH |
3744 | struct mem_cgroup_event *event = |
3745 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3746 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3747 | unsigned long flags = (unsigned long)key; |
3748 | ||
3749 | if (flags & POLLHUP) { | |
3750 | /* | |
3751 | * If the event has been detached at cgroup removal, we | |
3752 | * can simply return knowing the other side will cleanup | |
3753 | * for us. | |
3754 | * | |
3755 | * We can't race against event freeing since the other | |
3756 | * side will require wqh->lock via remove_wait_queue(), | |
3757 | * which we hold. | |
3758 | */ | |
fba94807 | 3759 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3760 | if (!list_empty(&event->list)) { |
3761 | list_del_init(&event->list); | |
3762 | /* | |
3763 | * We are in atomic context, but cgroup_event_remove() | |
3764 | * may sleep, so we have to call it in workqueue. | |
3765 | */ | |
3766 | schedule_work(&event->remove); | |
3767 | } | |
fba94807 | 3768 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3769 | } |
3770 | ||
3771 | return 0; | |
3772 | } | |
3773 | ||
3bc942f3 | 3774 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3775 | wait_queue_head_t *wqh, poll_table *pt) |
3776 | { | |
3bc942f3 TH |
3777 | struct mem_cgroup_event *event = |
3778 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3779 | |
3780 | event->wqh = wqh; | |
3781 | add_wait_queue(wqh, &event->wait); | |
3782 | } | |
3783 | ||
3784 | /* | |
3bc942f3 TH |
3785 | * DO NOT USE IN NEW FILES. |
3786 | * | |
79bd9814 TH |
3787 | * Parse input and register new cgroup event handler. |
3788 | * | |
3789 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3790 | * Interpretation of args is defined by control file implementation. | |
3791 | */ | |
451af504 TH |
3792 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3793 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3794 | { |
451af504 | 3795 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3796 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3797 | struct mem_cgroup_event *event; |
79bd9814 TH |
3798 | struct cgroup_subsys_state *cfile_css; |
3799 | unsigned int efd, cfd; | |
3800 | struct fd efile; | |
3801 | struct fd cfile; | |
fba94807 | 3802 | const char *name; |
79bd9814 TH |
3803 | char *endp; |
3804 | int ret; | |
3805 | ||
451af504 TH |
3806 | buf = strstrip(buf); |
3807 | ||
3808 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3809 | if (*endp != ' ') |
3810 | return -EINVAL; | |
451af504 | 3811 | buf = endp + 1; |
79bd9814 | 3812 | |
451af504 | 3813 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3814 | if ((*endp != ' ') && (*endp != '\0')) |
3815 | return -EINVAL; | |
451af504 | 3816 | buf = endp + 1; |
79bd9814 TH |
3817 | |
3818 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3819 | if (!event) | |
3820 | return -ENOMEM; | |
3821 | ||
59b6f873 | 3822 | event->memcg = memcg; |
79bd9814 | 3823 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3824 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3825 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3826 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3827 | |
3828 | efile = fdget(efd); | |
3829 | if (!efile.file) { | |
3830 | ret = -EBADF; | |
3831 | goto out_kfree; | |
3832 | } | |
3833 | ||
3834 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3835 | if (IS_ERR(event->eventfd)) { | |
3836 | ret = PTR_ERR(event->eventfd); | |
3837 | goto out_put_efile; | |
3838 | } | |
3839 | ||
3840 | cfile = fdget(cfd); | |
3841 | if (!cfile.file) { | |
3842 | ret = -EBADF; | |
3843 | goto out_put_eventfd; | |
3844 | } | |
3845 | ||
3846 | /* the process need read permission on control file */ | |
3847 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3848 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3849 | if (ret < 0) | |
3850 | goto out_put_cfile; | |
3851 | ||
fba94807 TH |
3852 | /* |
3853 | * Determine the event callbacks and set them in @event. This used | |
3854 | * to be done via struct cftype but cgroup core no longer knows | |
3855 | * about these events. The following is crude but the whole thing | |
3856 | * is for compatibility anyway. | |
3bc942f3 TH |
3857 | * |
3858 | * DO NOT ADD NEW FILES. | |
fba94807 | 3859 | */ |
b583043e | 3860 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3861 | |
3862 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3863 | event->register_event = mem_cgroup_usage_register_event; | |
3864 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3865 | } else if (!strcmp(name, "memory.oom_control")) { | |
3866 | event->register_event = mem_cgroup_oom_register_event; | |
3867 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3868 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3869 | event->register_event = vmpressure_register_event; | |
3870 | event->unregister_event = vmpressure_unregister_event; | |
3871 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3872 | event->register_event = memsw_cgroup_usage_register_event; |
3873 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3874 | } else { |
3875 | ret = -EINVAL; | |
3876 | goto out_put_cfile; | |
3877 | } | |
3878 | ||
79bd9814 | 3879 | /* |
b5557c4c TH |
3880 | * Verify @cfile should belong to @css. Also, remaining events are |
3881 | * automatically removed on cgroup destruction but the removal is | |
3882 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3883 | */ |
b583043e | 3884 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3885 | &memory_cgrp_subsys); |
79bd9814 | 3886 | ret = -EINVAL; |
5a17f543 | 3887 | if (IS_ERR(cfile_css)) |
79bd9814 | 3888 | goto out_put_cfile; |
5a17f543 TH |
3889 | if (cfile_css != css) { |
3890 | css_put(cfile_css); | |
79bd9814 | 3891 | goto out_put_cfile; |
5a17f543 | 3892 | } |
79bd9814 | 3893 | |
451af504 | 3894 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3895 | if (ret) |
3896 | goto out_put_css; | |
3897 | ||
3898 | efile.file->f_op->poll(efile.file, &event->pt); | |
3899 | ||
fba94807 TH |
3900 | spin_lock(&memcg->event_list_lock); |
3901 | list_add(&event->list, &memcg->event_list); | |
3902 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3903 | |
3904 | fdput(cfile); | |
3905 | fdput(efile); | |
3906 | ||
451af504 | 3907 | return nbytes; |
79bd9814 TH |
3908 | |
3909 | out_put_css: | |
b5557c4c | 3910 | css_put(css); |
79bd9814 TH |
3911 | out_put_cfile: |
3912 | fdput(cfile); | |
3913 | out_put_eventfd: | |
3914 | eventfd_ctx_put(event->eventfd); | |
3915 | out_put_efile: | |
3916 | fdput(efile); | |
3917 | out_kfree: | |
3918 | kfree(event); | |
3919 | ||
3920 | return ret; | |
3921 | } | |
3922 | ||
241994ed | 3923 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 3924 | { |
0eea1030 | 3925 | .name = "usage_in_bytes", |
8c7c6e34 | 3926 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 3927 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3928 | }, |
c84872e1 PE |
3929 | { |
3930 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3931 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 3932 | .write = mem_cgroup_reset, |
791badbd | 3933 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 3934 | }, |
8cdea7c0 | 3935 | { |
0eea1030 | 3936 | .name = "limit_in_bytes", |
8c7c6e34 | 3937 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 3938 | .write = mem_cgroup_write, |
791badbd | 3939 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3940 | }, |
296c81d8 BS |
3941 | { |
3942 | .name = "soft_limit_in_bytes", | |
3943 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 3944 | .write = mem_cgroup_write, |
791badbd | 3945 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 3946 | }, |
8cdea7c0 BS |
3947 | { |
3948 | .name = "failcnt", | |
8c7c6e34 | 3949 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 3950 | .write = mem_cgroup_reset, |
791badbd | 3951 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3952 | }, |
d2ceb9b7 KH |
3953 | { |
3954 | .name = "stat", | |
2da8ca82 | 3955 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 3956 | }, |
c1e862c1 KH |
3957 | { |
3958 | .name = "force_empty", | |
6770c64e | 3959 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 3960 | }, |
18f59ea7 BS |
3961 | { |
3962 | .name = "use_hierarchy", | |
3963 | .write_u64 = mem_cgroup_hierarchy_write, | |
3964 | .read_u64 = mem_cgroup_hierarchy_read, | |
3965 | }, | |
79bd9814 | 3966 | { |
3bc942f3 | 3967 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 3968 | .write = memcg_write_event_control, |
7dbdb199 | 3969 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 3970 | }, |
a7885eb8 KM |
3971 | { |
3972 | .name = "swappiness", | |
3973 | .read_u64 = mem_cgroup_swappiness_read, | |
3974 | .write_u64 = mem_cgroup_swappiness_write, | |
3975 | }, | |
7dc74be0 DN |
3976 | { |
3977 | .name = "move_charge_at_immigrate", | |
3978 | .read_u64 = mem_cgroup_move_charge_read, | |
3979 | .write_u64 = mem_cgroup_move_charge_write, | |
3980 | }, | |
9490ff27 KH |
3981 | { |
3982 | .name = "oom_control", | |
2da8ca82 | 3983 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 3984 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
3985 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
3986 | }, | |
70ddf637 AV |
3987 | { |
3988 | .name = "pressure_level", | |
70ddf637 | 3989 | }, |
406eb0c9 YH |
3990 | #ifdef CONFIG_NUMA |
3991 | { | |
3992 | .name = "numa_stat", | |
2da8ca82 | 3993 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
3994 | }, |
3995 | #endif | |
510fc4e1 GC |
3996 | { |
3997 | .name = "kmem.limit_in_bytes", | |
3998 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 3999 | .write = mem_cgroup_write, |
791badbd | 4000 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4001 | }, |
4002 | { | |
4003 | .name = "kmem.usage_in_bytes", | |
4004 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4005 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4006 | }, |
4007 | { | |
4008 | .name = "kmem.failcnt", | |
4009 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4010 | .write = mem_cgroup_reset, |
791badbd | 4011 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4012 | }, |
4013 | { | |
4014 | .name = "kmem.max_usage_in_bytes", | |
4015 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4016 | .write = mem_cgroup_reset, |
791badbd | 4017 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4018 | }, |
749c5415 GC |
4019 | #ifdef CONFIG_SLABINFO |
4020 | { | |
4021 | .name = "kmem.slabinfo", | |
b047501c VD |
4022 | .seq_start = slab_start, |
4023 | .seq_next = slab_next, | |
4024 | .seq_stop = slab_stop, | |
4025 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4026 | }, |
4027 | #endif | |
d55f90bf VD |
4028 | { |
4029 | .name = "kmem.tcp.limit_in_bytes", | |
4030 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4031 | .write = mem_cgroup_write, | |
4032 | .read_u64 = mem_cgroup_read_u64, | |
4033 | }, | |
4034 | { | |
4035 | .name = "kmem.tcp.usage_in_bytes", | |
4036 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4037 | .read_u64 = mem_cgroup_read_u64, | |
4038 | }, | |
4039 | { | |
4040 | .name = "kmem.tcp.failcnt", | |
4041 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4042 | .write = mem_cgroup_reset, | |
4043 | .read_u64 = mem_cgroup_read_u64, | |
4044 | }, | |
4045 | { | |
4046 | .name = "kmem.tcp.max_usage_in_bytes", | |
4047 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4048 | .write = mem_cgroup_reset, | |
4049 | .read_u64 = mem_cgroup_read_u64, | |
4050 | }, | |
6bc10349 | 4051 | { }, /* terminate */ |
af36f906 | 4052 | }; |
8c7c6e34 | 4053 | |
c0ff4b85 | 4054 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4055 | { |
4056 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4057 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4058 | int zone, tmp = node; |
1ecaab2b KH |
4059 | /* |
4060 | * This routine is called against possible nodes. | |
4061 | * But it's BUG to call kmalloc() against offline node. | |
4062 | * | |
4063 | * TODO: this routine can waste much memory for nodes which will | |
4064 | * never be onlined. It's better to use memory hotplug callback | |
4065 | * function. | |
4066 | */ | |
41e3355d KH |
4067 | if (!node_state(node, N_NORMAL_MEMORY)) |
4068 | tmp = -1; | |
17295c88 | 4069 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4070 | if (!pn) |
4071 | return 1; | |
1ecaab2b | 4072 | |
1ecaab2b KH |
4073 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4074 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4075 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4076 | mz->usage_in_excess = 0; |
4077 | mz->on_tree = false; | |
d79154bb | 4078 | mz->memcg = memcg; |
1ecaab2b | 4079 | } |
54f72fe0 | 4080 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4081 | return 0; |
4082 | } | |
4083 | ||
c0ff4b85 | 4084 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4085 | { |
54f72fe0 | 4086 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4087 | } |
4088 | ||
0b8f73e1 | 4089 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4090 | { |
c8b2a36f | 4091 | int node; |
59927fb9 | 4092 | |
0b8f73e1 | 4093 | memcg_wb_domain_exit(memcg); |
c8b2a36f GC |
4094 | for_each_node(node) |
4095 | free_mem_cgroup_per_zone_info(memcg, node); | |
c8b2a36f | 4096 | free_percpu(memcg->stat); |
8ff69e2c | 4097 | kfree(memcg); |
59927fb9 | 4098 | } |
3afe36b1 | 4099 | |
0b8f73e1 | 4100 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4101 | { |
d142e3e6 | 4102 | struct mem_cgroup *memcg; |
0b8f73e1 | 4103 | size_t size; |
6d12e2d8 | 4104 | int node; |
8cdea7c0 | 4105 | |
0b8f73e1 JW |
4106 | size = sizeof(struct mem_cgroup); |
4107 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4108 | ||
4109 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4110 | if (!memcg) |
0b8f73e1 JW |
4111 | return NULL; |
4112 | ||
4113 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); | |
4114 | if (!memcg->stat) | |
4115 | goto fail; | |
78fb7466 | 4116 | |
3ed28fa1 | 4117 | for_each_node(node) |
c0ff4b85 | 4118 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
0b8f73e1 | 4119 | goto fail; |
f64c3f54 | 4120 | |
0b8f73e1 JW |
4121 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4122 | goto fail; | |
28dbc4b6 | 4123 | |
f7e1cb6e | 4124 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4125 | memcg->last_scanned_node = MAX_NUMNODES; |
4126 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4127 | mutex_init(&memcg->thresholds_lock); |
4128 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4129 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4130 | INIT_LIST_HEAD(&memcg->event_list); |
4131 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4132 | memcg->socket_pressure = jiffies; |
127424c8 | 4133 | #ifndef CONFIG_SLOB |
900a38f0 | 4134 | memcg->kmemcg_id = -1; |
900a38f0 | 4135 | #endif |
52ebea74 TH |
4136 | #ifdef CONFIG_CGROUP_WRITEBACK |
4137 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4138 | #endif | |
0b8f73e1 JW |
4139 | return memcg; |
4140 | fail: | |
4141 | mem_cgroup_free(memcg); | |
4142 | return NULL; | |
d142e3e6 GC |
4143 | } |
4144 | ||
0b8f73e1 JW |
4145 | static struct cgroup_subsys_state * __ref |
4146 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4147 | { |
0b8f73e1 JW |
4148 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4149 | struct mem_cgroup *memcg; | |
4150 | long error = -ENOMEM; | |
d142e3e6 | 4151 | |
0b8f73e1 JW |
4152 | memcg = mem_cgroup_alloc(); |
4153 | if (!memcg) | |
4154 | return ERR_PTR(error); | |
d142e3e6 | 4155 | |
0b8f73e1 JW |
4156 | memcg->high = PAGE_COUNTER_MAX; |
4157 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4158 | if (parent) { | |
4159 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4160 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4161 | } | |
4162 | if (parent && parent->use_hierarchy) { | |
4163 | memcg->use_hierarchy = true; | |
3e32cb2e | 4164 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4165 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4166 | page_counter_init(&memcg->memsw, &parent->memsw); |
4167 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4168 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4169 | } else { |
3e32cb2e | 4170 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4171 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4172 | page_counter_init(&memcg->memsw, NULL); |
4173 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4174 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4175 | /* |
4176 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4177 | * much sense so let cgroup subsystem know about this | |
4178 | * unfortunate state in our controller. | |
4179 | */ | |
d142e3e6 | 4180 | if (parent != root_mem_cgroup) |
073219e9 | 4181 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4182 | } |
d6441637 | 4183 | |
0b8f73e1 JW |
4184 | /* The following stuff does not apply to the root */ |
4185 | if (!parent) { | |
4186 | root_mem_cgroup = memcg; | |
4187 | return &memcg->css; | |
4188 | } | |
4189 | ||
b313aeee | 4190 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4191 | if (error) |
4192 | goto fail; | |
127424c8 | 4193 | |
f7e1cb6e | 4194 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4195 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4196 | |
0b8f73e1 JW |
4197 | return &memcg->css; |
4198 | fail: | |
4199 | mem_cgroup_free(memcg); | |
4200 | return NULL; | |
4201 | } | |
4202 | ||
4203 | static int | |
4204 | mem_cgroup_css_online(struct cgroup_subsys_state *css) | |
4205 | { | |
4206 | if (css->id > MEM_CGROUP_ID_MAX) | |
4207 | return -ENOSPC; | |
2f7dd7a4 JW |
4208 | |
4209 | return 0; | |
8cdea7c0 BS |
4210 | } |
4211 | ||
eb95419b | 4212 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4213 | { |
eb95419b | 4214 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4215 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4216 | |
4217 | /* | |
4218 | * Unregister events and notify userspace. | |
4219 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4220 | * directory to avoid race between userspace and kernelspace. | |
4221 | */ | |
fba94807 TH |
4222 | spin_lock(&memcg->event_list_lock); |
4223 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4224 | list_del_init(&event->list); |
4225 | schedule_work(&event->remove); | |
4226 | } | |
fba94807 | 4227 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4228 | |
567e9ab2 | 4229 | memcg_offline_kmem(memcg); |
52ebea74 | 4230 | wb_memcg_offline(memcg); |
df878fb0 KH |
4231 | } |
4232 | ||
6df38689 VD |
4233 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4234 | { | |
4235 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4236 | ||
4237 | invalidate_reclaim_iterators(memcg); | |
4238 | } | |
4239 | ||
eb95419b | 4240 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4241 | { |
eb95419b | 4242 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4243 | |
f7e1cb6e | 4244 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4245 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4246 | |
0db15298 | 4247 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4248 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4249 | |
0b8f73e1 JW |
4250 | vmpressure_cleanup(&memcg->vmpressure); |
4251 | cancel_work_sync(&memcg->high_work); | |
4252 | mem_cgroup_remove_from_trees(memcg); | |
d886f4e4 | 4253 | memcg_free_kmem(memcg); |
0b8f73e1 | 4254 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4255 | } |
4256 | ||
1ced953b TH |
4257 | /** |
4258 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4259 | * @css: the target css | |
4260 | * | |
4261 | * Reset the states of the mem_cgroup associated with @css. This is | |
4262 | * invoked when the userland requests disabling on the default hierarchy | |
4263 | * but the memcg is pinned through dependency. The memcg should stop | |
4264 | * applying policies and should revert to the vanilla state as it may be | |
4265 | * made visible again. | |
4266 | * | |
4267 | * The current implementation only resets the essential configurations. | |
4268 | * This needs to be expanded to cover all the visible parts. | |
4269 | */ | |
4270 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4271 | { | |
4272 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4273 | ||
d334c9bc VD |
4274 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
4275 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); | |
4276 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); | |
4277 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); | |
4278 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
241994ed JW |
4279 | memcg->low = 0; |
4280 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4281 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4282 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4283 | } |
4284 | ||
02491447 | 4285 | #ifdef CONFIG_MMU |
7dc74be0 | 4286 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4287 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4288 | { |
05b84301 | 4289 | int ret; |
9476db97 | 4290 | |
d0164adc MG |
4291 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4292 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4293 | if (!ret) { |
854ffa8d | 4294 | mc.precharge += count; |
854ffa8d DN |
4295 | return ret; |
4296 | } | |
9476db97 JW |
4297 | |
4298 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4299 | while (count--) { |
00501b53 | 4300 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4301 | if (ret) |
38c5d72f | 4302 | return ret; |
854ffa8d | 4303 | mc.precharge++; |
9476db97 | 4304 | cond_resched(); |
854ffa8d | 4305 | } |
9476db97 | 4306 | return 0; |
4ffef5fe DN |
4307 | } |
4308 | ||
4309 | /** | |
8d32ff84 | 4310 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4311 | * @vma: the vma the pte to be checked belongs |
4312 | * @addr: the address corresponding to the pte to be checked | |
4313 | * @ptent: the pte to be checked | |
02491447 | 4314 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4315 | * |
4316 | * Returns | |
4317 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4318 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4319 | * move charge. if @target is not NULL, the page is stored in target->page | |
4320 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4321 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4322 | * target for charge migration. if @target is not NULL, the entry is stored | |
4323 | * in target->ent. | |
4ffef5fe DN |
4324 | * |
4325 | * Called with pte lock held. | |
4326 | */ | |
4ffef5fe DN |
4327 | union mc_target { |
4328 | struct page *page; | |
02491447 | 4329 | swp_entry_t ent; |
4ffef5fe DN |
4330 | }; |
4331 | ||
4ffef5fe | 4332 | enum mc_target_type { |
8d32ff84 | 4333 | MC_TARGET_NONE = 0, |
4ffef5fe | 4334 | MC_TARGET_PAGE, |
02491447 | 4335 | MC_TARGET_SWAP, |
4ffef5fe DN |
4336 | }; |
4337 | ||
90254a65 DN |
4338 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4339 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4340 | { |
90254a65 | 4341 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4342 | |
90254a65 DN |
4343 | if (!page || !page_mapped(page)) |
4344 | return NULL; | |
4345 | if (PageAnon(page)) { | |
1dfab5ab | 4346 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4347 | return NULL; |
1dfab5ab JW |
4348 | } else { |
4349 | if (!(mc.flags & MOVE_FILE)) | |
4350 | return NULL; | |
4351 | } | |
90254a65 DN |
4352 | if (!get_page_unless_zero(page)) |
4353 | return NULL; | |
4354 | ||
4355 | return page; | |
4356 | } | |
4357 | ||
4b91355e | 4358 | #ifdef CONFIG_SWAP |
90254a65 DN |
4359 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4360 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4361 | { | |
90254a65 DN |
4362 | struct page *page = NULL; |
4363 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4364 | ||
1dfab5ab | 4365 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4366 | return NULL; |
4b91355e KH |
4367 | /* |
4368 | * Because lookup_swap_cache() updates some statistics counter, | |
4369 | * we call find_get_page() with swapper_space directly. | |
4370 | */ | |
33806f06 | 4371 | page = find_get_page(swap_address_space(ent), ent.val); |
7941d214 | 4372 | if (do_memsw_account()) |
90254a65 DN |
4373 | entry->val = ent.val; |
4374 | ||
4375 | return page; | |
4376 | } | |
4b91355e KH |
4377 | #else |
4378 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4379 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4380 | { | |
4381 | return NULL; | |
4382 | } | |
4383 | #endif | |
90254a65 | 4384 | |
87946a72 DN |
4385 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4386 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4387 | { | |
4388 | struct page *page = NULL; | |
87946a72 DN |
4389 | struct address_space *mapping; |
4390 | pgoff_t pgoff; | |
4391 | ||
4392 | if (!vma->vm_file) /* anonymous vma */ | |
4393 | return NULL; | |
1dfab5ab | 4394 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4395 | return NULL; |
4396 | ||
87946a72 | 4397 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4398 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4399 | |
4400 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4401 | #ifdef CONFIG_SWAP |
4402 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4403 | if (shmem_mapping(mapping)) { |
4404 | page = find_get_entry(mapping, pgoff); | |
4405 | if (radix_tree_exceptional_entry(page)) { | |
4406 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4407 | if (do_memsw_account()) |
139b6a6f JW |
4408 | *entry = swp; |
4409 | page = find_get_page(swap_address_space(swp), swp.val); | |
4410 | } | |
4411 | } else | |
4412 | page = find_get_page(mapping, pgoff); | |
4413 | #else | |
4414 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4415 | #endif |
87946a72 DN |
4416 | return page; |
4417 | } | |
4418 | ||
b1b0deab CG |
4419 | /** |
4420 | * mem_cgroup_move_account - move account of the page | |
4421 | * @page: the page | |
4422 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4423 | * @from: mem_cgroup which the page is moved from. | |
4424 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4425 | * | |
3ac808fd | 4426 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4427 | * |
4428 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4429 | * from old cgroup. | |
4430 | */ | |
4431 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4432 | bool compound, |
b1b0deab CG |
4433 | struct mem_cgroup *from, |
4434 | struct mem_cgroup *to) | |
4435 | { | |
4436 | unsigned long flags; | |
f627c2f5 | 4437 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4438 | int ret; |
c4843a75 | 4439 | bool anon; |
b1b0deab CG |
4440 | |
4441 | VM_BUG_ON(from == to); | |
4442 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4443 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4444 | |
4445 | /* | |
6a93ca8f | 4446 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4447 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4448 | */ |
f627c2f5 | 4449 | ret = -EBUSY; |
b1b0deab CG |
4450 | if (!trylock_page(page)) |
4451 | goto out; | |
4452 | ||
4453 | ret = -EINVAL; | |
4454 | if (page->mem_cgroup != from) | |
4455 | goto out_unlock; | |
4456 | ||
c4843a75 GT |
4457 | anon = PageAnon(page); |
4458 | ||
b1b0deab CG |
4459 | spin_lock_irqsave(&from->move_lock, flags); |
4460 | ||
c4843a75 | 4461 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4462 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4463 | nr_pages); | |
4464 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4465 | nr_pages); | |
4466 | } | |
4467 | ||
c4843a75 GT |
4468 | /* |
4469 | * move_lock grabbed above and caller set from->moving_account, so | |
4470 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4471 | * So mapping should be stable for dirty pages. | |
4472 | */ | |
4473 | if (!anon && PageDirty(page)) { | |
4474 | struct address_space *mapping = page_mapping(page); | |
4475 | ||
4476 | if (mapping_cap_account_dirty(mapping)) { | |
4477 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4478 | nr_pages); | |
4479 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4480 | nr_pages); | |
4481 | } | |
4482 | } | |
4483 | ||
b1b0deab CG |
4484 | if (PageWriteback(page)) { |
4485 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4486 | nr_pages); | |
4487 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4488 | nr_pages); | |
4489 | } | |
4490 | ||
4491 | /* | |
4492 | * It is safe to change page->mem_cgroup here because the page | |
4493 | * is referenced, charged, and isolated - we can't race with | |
4494 | * uncharging, charging, migration, or LRU putback. | |
4495 | */ | |
4496 | ||
4497 | /* caller should have done css_get */ | |
4498 | page->mem_cgroup = to; | |
4499 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4500 | ||
4501 | ret = 0; | |
4502 | ||
4503 | local_irq_disable(); | |
f627c2f5 | 4504 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4505 | memcg_check_events(to, page); |
f627c2f5 | 4506 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4507 | memcg_check_events(from, page); |
4508 | local_irq_enable(); | |
4509 | out_unlock: | |
4510 | unlock_page(page); | |
4511 | out: | |
4512 | return ret; | |
4513 | } | |
4514 | ||
8d32ff84 | 4515 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4516 | unsigned long addr, pte_t ptent, union mc_target *target) |
4517 | { | |
4518 | struct page *page = NULL; | |
8d32ff84 | 4519 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4520 | swp_entry_t ent = { .val = 0 }; |
4521 | ||
4522 | if (pte_present(ptent)) | |
4523 | page = mc_handle_present_pte(vma, addr, ptent); | |
4524 | else if (is_swap_pte(ptent)) | |
4525 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4526 | else if (pte_none(ptent)) |
87946a72 | 4527 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4528 | |
4529 | if (!page && !ent.val) | |
8d32ff84 | 4530 | return ret; |
02491447 | 4531 | if (page) { |
02491447 | 4532 | /* |
0a31bc97 | 4533 | * Do only loose check w/o serialization. |
1306a85a | 4534 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4535 | * not under LRU exclusion. |
02491447 | 4536 | */ |
1306a85a | 4537 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4538 | ret = MC_TARGET_PAGE; |
4539 | if (target) | |
4540 | target->page = page; | |
4541 | } | |
4542 | if (!ret || !target) | |
4543 | put_page(page); | |
4544 | } | |
90254a65 DN |
4545 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4546 | if (ent.val && !ret && | |
34c00c31 | 4547 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4548 | ret = MC_TARGET_SWAP; |
4549 | if (target) | |
4550 | target->ent = ent; | |
4ffef5fe | 4551 | } |
4ffef5fe DN |
4552 | return ret; |
4553 | } | |
4554 | ||
12724850 NH |
4555 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4556 | /* | |
4557 | * We don't consider swapping or file mapped pages because THP does not | |
4558 | * support them for now. | |
4559 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4560 | */ | |
4561 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4562 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4563 | { | |
4564 | struct page *page = NULL; | |
12724850 NH |
4565 | enum mc_target_type ret = MC_TARGET_NONE; |
4566 | ||
4567 | page = pmd_page(pmd); | |
309381fe | 4568 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4569 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4570 | return ret; |
1306a85a | 4571 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4572 | ret = MC_TARGET_PAGE; |
4573 | if (target) { | |
4574 | get_page(page); | |
4575 | target->page = page; | |
4576 | } | |
4577 | } | |
4578 | return ret; | |
4579 | } | |
4580 | #else | |
4581 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4582 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4583 | { | |
4584 | return MC_TARGET_NONE; | |
4585 | } | |
4586 | #endif | |
4587 | ||
4ffef5fe DN |
4588 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4589 | unsigned long addr, unsigned long end, | |
4590 | struct mm_walk *walk) | |
4591 | { | |
26bcd64a | 4592 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4593 | pte_t *pte; |
4594 | spinlock_t *ptl; | |
4595 | ||
b6ec57f4 KS |
4596 | ptl = pmd_trans_huge_lock(pmd, vma); |
4597 | if (ptl) { | |
12724850 NH |
4598 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4599 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4600 | spin_unlock(ptl); |
1a5a9906 | 4601 | return 0; |
12724850 | 4602 | } |
03319327 | 4603 | |
45f83cef AA |
4604 | if (pmd_trans_unstable(pmd)) |
4605 | return 0; | |
4ffef5fe DN |
4606 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4607 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4608 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4609 | mc.precharge++; /* increment precharge temporarily */ |
4610 | pte_unmap_unlock(pte - 1, ptl); | |
4611 | cond_resched(); | |
4612 | ||
7dc74be0 DN |
4613 | return 0; |
4614 | } | |
4615 | ||
4ffef5fe DN |
4616 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4617 | { | |
4618 | unsigned long precharge; | |
4ffef5fe | 4619 | |
26bcd64a NH |
4620 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4621 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4622 | .mm = mm, | |
4623 | }; | |
dfe076b0 | 4624 | down_read(&mm->mmap_sem); |
26bcd64a | 4625 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4626 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4627 | |
4628 | precharge = mc.precharge; | |
4629 | mc.precharge = 0; | |
4630 | ||
4631 | return precharge; | |
4632 | } | |
4633 | ||
4ffef5fe DN |
4634 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4635 | { | |
dfe076b0 DN |
4636 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4637 | ||
4638 | VM_BUG_ON(mc.moving_task); | |
4639 | mc.moving_task = current; | |
4640 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4641 | } |
4642 | ||
dfe076b0 DN |
4643 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4644 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4645 | { |
2bd9bb20 KH |
4646 | struct mem_cgroup *from = mc.from; |
4647 | struct mem_cgroup *to = mc.to; | |
4648 | ||
4ffef5fe | 4649 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4650 | if (mc.precharge) { |
00501b53 | 4651 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4652 | mc.precharge = 0; |
4653 | } | |
4654 | /* | |
4655 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4656 | * we must uncharge here. | |
4657 | */ | |
4658 | if (mc.moved_charge) { | |
00501b53 | 4659 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4660 | mc.moved_charge = 0; |
4ffef5fe | 4661 | } |
483c30b5 DN |
4662 | /* we must fixup refcnts and charges */ |
4663 | if (mc.moved_swap) { | |
483c30b5 | 4664 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4665 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4666 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4667 | |
05b84301 | 4668 | /* |
3e32cb2e JW |
4669 | * we charged both to->memory and to->memsw, so we |
4670 | * should uncharge to->memory. | |
05b84301 | 4671 | */ |
ce00a967 | 4672 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4673 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4674 | ||
e8ea14cc | 4675 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4676 | |
4050377b | 4677 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4678 | mc.moved_swap = 0; |
4679 | } | |
dfe076b0 DN |
4680 | memcg_oom_recover(from); |
4681 | memcg_oom_recover(to); | |
4682 | wake_up_all(&mc.waitq); | |
4683 | } | |
4684 | ||
4685 | static void mem_cgroup_clear_mc(void) | |
4686 | { | |
264a0ae1 TH |
4687 | struct mm_struct *mm = mc.mm; |
4688 | ||
dfe076b0 DN |
4689 | /* |
4690 | * we must clear moving_task before waking up waiters at the end of | |
4691 | * task migration. | |
4692 | */ | |
4693 | mc.moving_task = NULL; | |
4694 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4695 | spin_lock(&mc.lock); |
4ffef5fe DN |
4696 | mc.from = NULL; |
4697 | mc.to = NULL; | |
264a0ae1 | 4698 | mc.mm = NULL; |
2bd9bb20 | 4699 | spin_unlock(&mc.lock); |
264a0ae1 TH |
4700 | |
4701 | mmput(mm); | |
4ffef5fe DN |
4702 | } |
4703 | ||
1f7dd3e5 | 4704 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4705 | { |
1f7dd3e5 | 4706 | struct cgroup_subsys_state *css; |
eed67d75 | 4707 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4708 | struct mem_cgroup *from; |
4530eddb | 4709 | struct task_struct *leader, *p; |
9f2115f9 | 4710 | struct mm_struct *mm; |
1dfab5ab | 4711 | unsigned long move_flags; |
9f2115f9 | 4712 | int ret = 0; |
7dc74be0 | 4713 | |
1f7dd3e5 TH |
4714 | /* charge immigration isn't supported on the default hierarchy */ |
4715 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4716 | return 0; |
4717 | ||
4530eddb TH |
4718 | /* |
4719 | * Multi-process migrations only happen on the default hierarchy | |
4720 | * where charge immigration is not used. Perform charge | |
4721 | * immigration if @tset contains a leader and whine if there are | |
4722 | * multiple. | |
4723 | */ | |
4724 | p = NULL; | |
1f7dd3e5 | 4725 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4726 | WARN_ON_ONCE(p); |
4727 | p = leader; | |
1f7dd3e5 | 4728 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4729 | } |
4730 | if (!p) | |
4731 | return 0; | |
4732 | ||
1f7dd3e5 TH |
4733 | /* |
4734 | * We are now commited to this value whatever it is. Changes in this | |
4735 | * tunable will only affect upcoming migrations, not the current one. | |
4736 | * So we need to save it, and keep it going. | |
4737 | */ | |
4738 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4739 | if (!move_flags) | |
4740 | return 0; | |
4741 | ||
9f2115f9 TH |
4742 | from = mem_cgroup_from_task(p); |
4743 | ||
4744 | VM_BUG_ON(from == memcg); | |
4745 | ||
4746 | mm = get_task_mm(p); | |
4747 | if (!mm) | |
4748 | return 0; | |
4749 | /* We move charges only when we move a owner of the mm */ | |
4750 | if (mm->owner == p) { | |
4751 | VM_BUG_ON(mc.from); | |
4752 | VM_BUG_ON(mc.to); | |
4753 | VM_BUG_ON(mc.precharge); | |
4754 | VM_BUG_ON(mc.moved_charge); | |
4755 | VM_BUG_ON(mc.moved_swap); | |
4756 | ||
4757 | spin_lock(&mc.lock); | |
264a0ae1 | 4758 | mc.mm = mm; |
9f2115f9 TH |
4759 | mc.from = from; |
4760 | mc.to = memcg; | |
4761 | mc.flags = move_flags; | |
4762 | spin_unlock(&mc.lock); | |
4763 | /* We set mc.moving_task later */ | |
4764 | ||
4765 | ret = mem_cgroup_precharge_mc(mm); | |
4766 | if (ret) | |
4767 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
4768 | } else { |
4769 | mmput(mm); | |
7dc74be0 DN |
4770 | } |
4771 | return ret; | |
4772 | } | |
4773 | ||
1f7dd3e5 | 4774 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4775 | { |
4e2f245d JW |
4776 | if (mc.to) |
4777 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4778 | } |
4779 | ||
4ffef5fe DN |
4780 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4781 | unsigned long addr, unsigned long end, | |
4782 | struct mm_walk *walk) | |
7dc74be0 | 4783 | { |
4ffef5fe | 4784 | int ret = 0; |
26bcd64a | 4785 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4786 | pte_t *pte; |
4787 | spinlock_t *ptl; | |
12724850 NH |
4788 | enum mc_target_type target_type; |
4789 | union mc_target target; | |
4790 | struct page *page; | |
4ffef5fe | 4791 | |
b6ec57f4 KS |
4792 | ptl = pmd_trans_huge_lock(pmd, vma); |
4793 | if (ptl) { | |
62ade86a | 4794 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4795 | spin_unlock(ptl); |
12724850 NH |
4796 | return 0; |
4797 | } | |
4798 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4799 | if (target_type == MC_TARGET_PAGE) { | |
4800 | page = target.page; | |
4801 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 4802 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 4803 | mc.from, mc.to)) { |
12724850 NH |
4804 | mc.precharge -= HPAGE_PMD_NR; |
4805 | mc.moved_charge += HPAGE_PMD_NR; | |
4806 | } | |
4807 | putback_lru_page(page); | |
4808 | } | |
4809 | put_page(page); | |
4810 | } | |
bf929152 | 4811 | spin_unlock(ptl); |
1a5a9906 | 4812 | return 0; |
12724850 NH |
4813 | } |
4814 | ||
45f83cef AA |
4815 | if (pmd_trans_unstable(pmd)) |
4816 | return 0; | |
4ffef5fe DN |
4817 | retry: |
4818 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4819 | for (; addr != end; addr += PAGE_SIZE) { | |
4820 | pte_t ptent = *(pte++); | |
02491447 | 4821 | swp_entry_t ent; |
4ffef5fe DN |
4822 | |
4823 | if (!mc.precharge) | |
4824 | break; | |
4825 | ||
8d32ff84 | 4826 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4827 | case MC_TARGET_PAGE: |
4828 | page = target.page; | |
53f9263b KS |
4829 | /* |
4830 | * We can have a part of the split pmd here. Moving it | |
4831 | * can be done but it would be too convoluted so simply | |
4832 | * ignore such a partial THP and keep it in original | |
4833 | * memcg. There should be somebody mapping the head. | |
4834 | */ | |
4835 | if (PageTransCompound(page)) | |
4836 | goto put; | |
4ffef5fe DN |
4837 | if (isolate_lru_page(page)) |
4838 | goto put; | |
f627c2f5 KS |
4839 | if (!mem_cgroup_move_account(page, false, |
4840 | mc.from, mc.to)) { | |
4ffef5fe | 4841 | mc.precharge--; |
854ffa8d DN |
4842 | /* we uncharge from mc.from later. */ |
4843 | mc.moved_charge++; | |
4ffef5fe DN |
4844 | } |
4845 | putback_lru_page(page); | |
8d32ff84 | 4846 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4847 | put_page(page); |
4848 | break; | |
02491447 DN |
4849 | case MC_TARGET_SWAP: |
4850 | ent = target.ent; | |
e91cbb42 | 4851 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4852 | mc.precharge--; |
483c30b5 DN |
4853 | /* we fixup refcnts and charges later. */ |
4854 | mc.moved_swap++; | |
4855 | } | |
02491447 | 4856 | break; |
4ffef5fe DN |
4857 | default: |
4858 | break; | |
4859 | } | |
4860 | } | |
4861 | pte_unmap_unlock(pte - 1, ptl); | |
4862 | cond_resched(); | |
4863 | ||
4864 | if (addr != end) { | |
4865 | /* | |
4866 | * We have consumed all precharges we got in can_attach(). | |
4867 | * We try charge one by one, but don't do any additional | |
4868 | * charges to mc.to if we have failed in charge once in attach() | |
4869 | * phase. | |
4870 | */ | |
854ffa8d | 4871 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4872 | if (!ret) |
4873 | goto retry; | |
4874 | } | |
4875 | ||
4876 | return ret; | |
4877 | } | |
4878 | ||
264a0ae1 | 4879 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 4880 | { |
26bcd64a NH |
4881 | struct mm_walk mem_cgroup_move_charge_walk = { |
4882 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 4883 | .mm = mc.mm, |
26bcd64a | 4884 | }; |
4ffef5fe DN |
4885 | |
4886 | lru_add_drain_all(); | |
312722cb | 4887 | /* |
81f8c3a4 JW |
4888 | * Signal lock_page_memcg() to take the memcg's move_lock |
4889 | * while we're moving its pages to another memcg. Then wait | |
4890 | * for already started RCU-only updates to finish. | |
312722cb JW |
4891 | */ |
4892 | atomic_inc(&mc.from->moving_account); | |
4893 | synchronize_rcu(); | |
dfe076b0 | 4894 | retry: |
264a0ae1 | 4895 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
4896 | /* |
4897 | * Someone who are holding the mmap_sem might be waiting in | |
4898 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4899 | * and retry. Because we cancel precharges, we might not be able | |
4900 | * to move enough charges, but moving charge is a best-effort | |
4901 | * feature anyway, so it wouldn't be a big problem. | |
4902 | */ | |
4903 | __mem_cgroup_clear_mc(); | |
4904 | cond_resched(); | |
4905 | goto retry; | |
4906 | } | |
26bcd64a NH |
4907 | /* |
4908 | * When we have consumed all precharges and failed in doing | |
4909 | * additional charge, the page walk just aborts. | |
4910 | */ | |
4911 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
264a0ae1 | 4912 | up_read(&mc.mm->mmap_sem); |
312722cb | 4913 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
4914 | } |
4915 | ||
264a0ae1 | 4916 | static void mem_cgroup_move_task(void) |
67e465a7 | 4917 | { |
264a0ae1 TH |
4918 | if (mc.to) { |
4919 | mem_cgroup_move_charge(); | |
a433658c | 4920 | mem_cgroup_clear_mc(); |
264a0ae1 | 4921 | } |
67e465a7 | 4922 | } |
5cfb80a7 | 4923 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 4924 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4925 | { |
4926 | return 0; | |
4927 | } | |
1f7dd3e5 | 4928 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4929 | { |
4930 | } | |
264a0ae1 | 4931 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
4932 | { |
4933 | } | |
4934 | #endif | |
67e465a7 | 4935 | |
f00baae7 TH |
4936 | /* |
4937 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
4938 | * to verify whether we're attached to the default hierarchy on each mount |
4939 | * attempt. | |
f00baae7 | 4940 | */ |
eb95419b | 4941 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
4942 | { |
4943 | /* | |
aa6ec29b | 4944 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
4945 | * guarantees that @root doesn't have any children, so turning it |
4946 | * on for the root memcg is enough. | |
4947 | */ | |
9e10a130 | 4948 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
4949 | root_mem_cgroup->use_hierarchy = true; |
4950 | else | |
4951 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
4952 | } |
4953 | ||
241994ed JW |
4954 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
4955 | struct cftype *cft) | |
4956 | { | |
f5fc3c5d JW |
4957 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4958 | ||
4959 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
4960 | } |
4961 | ||
4962 | static int memory_low_show(struct seq_file *m, void *v) | |
4963 | { | |
4964 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 4965 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
4966 | |
4967 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 4968 | seq_puts(m, "max\n"); |
241994ed JW |
4969 | else |
4970 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
4971 | ||
4972 | return 0; | |
4973 | } | |
4974 | ||
4975 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
4976 | char *buf, size_t nbytes, loff_t off) | |
4977 | { | |
4978 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
4979 | unsigned long low; | |
4980 | int err; | |
4981 | ||
4982 | buf = strstrip(buf); | |
d2973697 | 4983 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
4984 | if (err) |
4985 | return err; | |
4986 | ||
4987 | memcg->low = low; | |
4988 | ||
4989 | return nbytes; | |
4990 | } | |
4991 | ||
4992 | static int memory_high_show(struct seq_file *m, void *v) | |
4993 | { | |
4994 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 4995 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
4996 | |
4997 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 4998 | seq_puts(m, "max\n"); |
241994ed JW |
4999 | else |
5000 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5001 | ||
5002 | return 0; | |
5003 | } | |
5004 | ||
5005 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5006 | char *buf, size_t nbytes, loff_t off) | |
5007 | { | |
5008 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5009 | unsigned long nr_pages; |
241994ed JW |
5010 | unsigned long high; |
5011 | int err; | |
5012 | ||
5013 | buf = strstrip(buf); | |
d2973697 | 5014 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5015 | if (err) |
5016 | return err; | |
5017 | ||
5018 | memcg->high = high; | |
5019 | ||
588083bb JW |
5020 | nr_pages = page_counter_read(&memcg->memory); |
5021 | if (nr_pages > high) | |
5022 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5023 | GFP_KERNEL, true); | |
5024 | ||
2529bb3a | 5025 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5026 | return nbytes; |
5027 | } | |
5028 | ||
5029 | static int memory_max_show(struct seq_file *m, void *v) | |
5030 | { | |
5031 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5032 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5033 | |
5034 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5035 | seq_puts(m, "max\n"); |
241994ed JW |
5036 | else |
5037 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5038 | ||
5039 | return 0; | |
5040 | } | |
5041 | ||
5042 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5043 | char *buf, size_t nbytes, loff_t off) | |
5044 | { | |
5045 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5046 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5047 | bool drained = false; | |
241994ed JW |
5048 | unsigned long max; |
5049 | int err; | |
5050 | ||
5051 | buf = strstrip(buf); | |
d2973697 | 5052 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5053 | if (err) |
5054 | return err; | |
5055 | ||
b6e6edcf JW |
5056 | xchg(&memcg->memory.limit, max); |
5057 | ||
5058 | for (;;) { | |
5059 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5060 | ||
5061 | if (nr_pages <= max) | |
5062 | break; | |
5063 | ||
5064 | if (signal_pending(current)) { | |
5065 | err = -EINTR; | |
5066 | break; | |
5067 | } | |
5068 | ||
5069 | if (!drained) { | |
5070 | drain_all_stock(memcg); | |
5071 | drained = true; | |
5072 | continue; | |
5073 | } | |
5074 | ||
5075 | if (nr_reclaims) { | |
5076 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5077 | GFP_KERNEL, true)) | |
5078 | nr_reclaims--; | |
5079 | continue; | |
5080 | } | |
5081 | ||
5082 | mem_cgroup_events(memcg, MEMCG_OOM, 1); | |
5083 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) | |
5084 | break; | |
5085 | } | |
241994ed | 5086 | |
2529bb3a | 5087 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5088 | return nbytes; |
5089 | } | |
5090 | ||
5091 | static int memory_events_show(struct seq_file *m, void *v) | |
5092 | { | |
5093 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5094 | ||
5095 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5096 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5097 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5098 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5099 | ||
5100 | return 0; | |
5101 | } | |
5102 | ||
587d9f72 JW |
5103 | static int memory_stat_show(struct seq_file *m, void *v) |
5104 | { | |
5105 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
72b54e73 VD |
5106 | unsigned long stat[MEMCG_NR_STAT]; |
5107 | unsigned long events[MEMCG_NR_EVENTS]; | |
587d9f72 JW |
5108 | int i; |
5109 | ||
5110 | /* | |
5111 | * Provide statistics on the state of the memory subsystem as | |
5112 | * well as cumulative event counters that show past behavior. | |
5113 | * | |
5114 | * This list is ordered following a combination of these gradients: | |
5115 | * 1) generic big picture -> specifics and details | |
5116 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5117 | * | |
5118 | * Current memory state: | |
5119 | */ | |
5120 | ||
72b54e73 VD |
5121 | tree_stat(memcg, stat); |
5122 | tree_events(memcg, events); | |
5123 | ||
587d9f72 | 5124 | seq_printf(m, "anon %llu\n", |
72b54e73 | 5125 | (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE); |
587d9f72 | 5126 | seq_printf(m, "file %llu\n", |
72b54e73 | 5127 | (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE); |
12580e4b VD |
5128 | seq_printf(m, "kernel_stack %llu\n", |
5129 | (u64)stat[MEMCG_KERNEL_STACK] * PAGE_SIZE); | |
27ee57c9 VD |
5130 | seq_printf(m, "slab %llu\n", |
5131 | (u64)(stat[MEMCG_SLAB_RECLAIMABLE] + | |
5132 | stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5133 | seq_printf(m, "sock %llu\n", |
72b54e73 | 5134 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 JW |
5135 | |
5136 | seq_printf(m, "file_mapped %llu\n", | |
72b54e73 | 5137 | (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5138 | seq_printf(m, "file_dirty %llu\n", |
72b54e73 | 5139 | (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE); |
587d9f72 | 5140 | seq_printf(m, "file_writeback %llu\n", |
72b54e73 | 5141 | (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE); |
587d9f72 JW |
5142 | |
5143 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5144 | struct mem_cgroup *mi; | |
5145 | unsigned long val = 0; | |
5146 | ||
5147 | for_each_mem_cgroup_tree(mi, memcg) | |
5148 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
5149 | seq_printf(m, "%s %llu\n", | |
5150 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); | |
5151 | } | |
5152 | ||
27ee57c9 VD |
5153 | seq_printf(m, "slab_reclaimable %llu\n", |
5154 | (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE); | |
5155 | seq_printf(m, "slab_unreclaimable %llu\n", | |
5156 | (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE); | |
5157 | ||
587d9f72 JW |
5158 | /* Accumulated memory events */ |
5159 | ||
5160 | seq_printf(m, "pgfault %lu\n", | |
72b54e73 | 5161 | events[MEM_CGROUP_EVENTS_PGFAULT]); |
587d9f72 | 5162 | seq_printf(m, "pgmajfault %lu\n", |
72b54e73 | 5163 | events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
587d9f72 JW |
5164 | |
5165 | return 0; | |
5166 | } | |
5167 | ||
241994ed JW |
5168 | static struct cftype memory_files[] = { |
5169 | { | |
5170 | .name = "current", | |
f5fc3c5d | 5171 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5172 | .read_u64 = memory_current_read, |
5173 | }, | |
5174 | { | |
5175 | .name = "low", | |
5176 | .flags = CFTYPE_NOT_ON_ROOT, | |
5177 | .seq_show = memory_low_show, | |
5178 | .write = memory_low_write, | |
5179 | }, | |
5180 | { | |
5181 | .name = "high", | |
5182 | .flags = CFTYPE_NOT_ON_ROOT, | |
5183 | .seq_show = memory_high_show, | |
5184 | .write = memory_high_write, | |
5185 | }, | |
5186 | { | |
5187 | .name = "max", | |
5188 | .flags = CFTYPE_NOT_ON_ROOT, | |
5189 | .seq_show = memory_max_show, | |
5190 | .write = memory_max_write, | |
5191 | }, | |
5192 | { | |
5193 | .name = "events", | |
5194 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5195 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5196 | .seq_show = memory_events_show, |
5197 | }, | |
587d9f72 JW |
5198 | { |
5199 | .name = "stat", | |
5200 | .flags = CFTYPE_NOT_ON_ROOT, | |
5201 | .seq_show = memory_stat_show, | |
5202 | }, | |
241994ed JW |
5203 | { } /* terminate */ |
5204 | }; | |
5205 | ||
073219e9 | 5206 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5207 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5208 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5209 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5210 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5211 | .css_free = mem_cgroup_css_free, |
1ced953b | 5212 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5213 | .can_attach = mem_cgroup_can_attach, |
5214 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5215 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5216 | .bind = mem_cgroup_bind, |
241994ed JW |
5217 | .dfl_cftypes = memory_files, |
5218 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5219 | .early_init = 0, |
8cdea7c0 | 5220 | }; |
c077719b | 5221 | |
241994ed JW |
5222 | /** |
5223 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5224 | * @root: the highest ancestor to consider | |
5225 | * @memcg: the memory cgroup to check | |
5226 | * | |
5227 | * Returns %true if memory consumption of @memcg, and that of all | |
5228 | * configurable ancestors up to @root, is below the normal range. | |
5229 | */ | |
5230 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5231 | { | |
5232 | if (mem_cgroup_disabled()) | |
5233 | return false; | |
5234 | ||
5235 | /* | |
5236 | * The toplevel group doesn't have a configurable range, so | |
5237 | * it's never low when looked at directly, and it is not | |
5238 | * considered an ancestor when assessing the hierarchy. | |
5239 | */ | |
5240 | ||
5241 | if (memcg == root_mem_cgroup) | |
5242 | return false; | |
5243 | ||
4e54dede | 5244 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5245 | return false; |
5246 | ||
5247 | while (memcg != root) { | |
5248 | memcg = parent_mem_cgroup(memcg); | |
5249 | ||
5250 | if (memcg == root_mem_cgroup) | |
5251 | break; | |
5252 | ||
4e54dede | 5253 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5254 | return false; |
5255 | } | |
5256 | return true; | |
5257 | } | |
5258 | ||
00501b53 JW |
5259 | /** |
5260 | * mem_cgroup_try_charge - try charging a page | |
5261 | * @page: page to charge | |
5262 | * @mm: mm context of the victim | |
5263 | * @gfp_mask: reclaim mode | |
5264 | * @memcgp: charged memcg return | |
5265 | * | |
5266 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5267 | * pages according to @gfp_mask if necessary. | |
5268 | * | |
5269 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5270 | * Otherwise, an error code is returned. | |
5271 | * | |
5272 | * After page->mapping has been set up, the caller must finalize the | |
5273 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5274 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5275 | */ | |
5276 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5277 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5278 | bool compound) | |
00501b53 JW |
5279 | { |
5280 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5281 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5282 | int ret = 0; |
5283 | ||
5284 | if (mem_cgroup_disabled()) | |
5285 | goto out; | |
5286 | ||
5287 | if (PageSwapCache(page)) { | |
00501b53 JW |
5288 | /* |
5289 | * Every swap fault against a single page tries to charge the | |
5290 | * page, bail as early as possible. shmem_unuse() encounters | |
5291 | * already charged pages, too. The USED bit is protected by | |
5292 | * the page lock, which serializes swap cache removal, which | |
5293 | * in turn serializes uncharging. | |
5294 | */ | |
e993d905 | 5295 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5296 | if (page->mem_cgroup) |
00501b53 | 5297 | goto out; |
e993d905 | 5298 | |
37e84351 | 5299 | if (do_swap_account) { |
e993d905 VD |
5300 | swp_entry_t ent = { .val = page_private(page), }; |
5301 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5302 | ||
5303 | rcu_read_lock(); | |
5304 | memcg = mem_cgroup_from_id(id); | |
5305 | if (memcg && !css_tryget_online(&memcg->css)) | |
5306 | memcg = NULL; | |
5307 | rcu_read_unlock(); | |
5308 | } | |
00501b53 JW |
5309 | } |
5310 | ||
00501b53 JW |
5311 | if (!memcg) |
5312 | memcg = get_mem_cgroup_from_mm(mm); | |
5313 | ||
5314 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5315 | ||
5316 | css_put(&memcg->css); | |
00501b53 JW |
5317 | out: |
5318 | *memcgp = memcg; | |
5319 | return ret; | |
5320 | } | |
5321 | ||
5322 | /** | |
5323 | * mem_cgroup_commit_charge - commit a page charge | |
5324 | * @page: page to charge | |
5325 | * @memcg: memcg to charge the page to | |
5326 | * @lrucare: page might be on LRU already | |
5327 | * | |
5328 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5329 | * after page->mapping has been set up. This must happen atomically | |
5330 | * as part of the page instantiation, i.e. under the page table lock | |
5331 | * for anonymous pages, under the page lock for page and swap cache. | |
5332 | * | |
5333 | * In addition, the page must not be on the LRU during the commit, to | |
5334 | * prevent racing with task migration. If it might be, use @lrucare. | |
5335 | * | |
5336 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5337 | */ | |
5338 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5339 | bool lrucare, bool compound) |
00501b53 | 5340 | { |
f627c2f5 | 5341 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5342 | |
5343 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5344 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5345 | ||
5346 | if (mem_cgroup_disabled()) | |
5347 | return; | |
5348 | /* | |
5349 | * Swap faults will attempt to charge the same page multiple | |
5350 | * times. But reuse_swap_page() might have removed the page | |
5351 | * from swapcache already, so we can't check PageSwapCache(). | |
5352 | */ | |
5353 | if (!memcg) | |
5354 | return; | |
5355 | ||
6abb5a86 JW |
5356 | commit_charge(page, memcg, lrucare); |
5357 | ||
6abb5a86 | 5358 | local_irq_disable(); |
f627c2f5 | 5359 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5360 | memcg_check_events(memcg, page); |
5361 | local_irq_enable(); | |
00501b53 | 5362 | |
7941d214 | 5363 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5364 | swp_entry_t entry = { .val = page_private(page) }; |
5365 | /* | |
5366 | * The swap entry might not get freed for a long time, | |
5367 | * let's not wait for it. The page already received a | |
5368 | * memory+swap charge, drop the swap entry duplicate. | |
5369 | */ | |
5370 | mem_cgroup_uncharge_swap(entry); | |
5371 | } | |
5372 | } | |
5373 | ||
5374 | /** | |
5375 | * mem_cgroup_cancel_charge - cancel a page charge | |
5376 | * @page: page to charge | |
5377 | * @memcg: memcg to charge the page to | |
5378 | * | |
5379 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5380 | */ | |
f627c2f5 KS |
5381 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5382 | bool compound) | |
00501b53 | 5383 | { |
f627c2f5 | 5384 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5385 | |
5386 | if (mem_cgroup_disabled()) | |
5387 | return; | |
5388 | /* | |
5389 | * Swap faults will attempt to charge the same page multiple | |
5390 | * times. But reuse_swap_page() might have removed the page | |
5391 | * from swapcache already, so we can't check PageSwapCache(). | |
5392 | */ | |
5393 | if (!memcg) | |
5394 | return; | |
5395 | ||
00501b53 JW |
5396 | cancel_charge(memcg, nr_pages); |
5397 | } | |
5398 | ||
747db954 | 5399 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5400 | unsigned long nr_anon, unsigned long nr_file, |
5401 | unsigned long nr_huge, struct page *dummy_page) | |
5402 | { | |
18eca2e6 | 5403 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5404 | unsigned long flags; |
5405 | ||
ce00a967 | 5406 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 | 5407 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 5408 | if (do_memsw_account()) |
18eca2e6 | 5409 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 JW |
5410 | memcg_oom_recover(memcg); |
5411 | } | |
747db954 JW |
5412 | |
5413 | local_irq_save(flags); | |
5414 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5415 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5416 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5417 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5418 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5419 | memcg_check_events(memcg, dummy_page); |
5420 | local_irq_restore(flags); | |
e8ea14cc JW |
5421 | |
5422 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5423 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5424 | } |
5425 | ||
5426 | static void uncharge_list(struct list_head *page_list) | |
5427 | { | |
5428 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5429 | unsigned long nr_anon = 0; |
5430 | unsigned long nr_file = 0; | |
5431 | unsigned long nr_huge = 0; | |
5432 | unsigned long pgpgout = 0; | |
747db954 JW |
5433 | struct list_head *next; |
5434 | struct page *page; | |
5435 | ||
8b592656 JW |
5436 | /* |
5437 | * Note that the list can be a single page->lru; hence the | |
5438 | * do-while loop instead of a simple list_for_each_entry(). | |
5439 | */ | |
747db954 JW |
5440 | next = page_list->next; |
5441 | do { | |
5442 | unsigned int nr_pages = 1; | |
747db954 JW |
5443 | |
5444 | page = list_entry(next, struct page, lru); | |
5445 | next = page->lru.next; | |
5446 | ||
5447 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5448 | VM_BUG_ON_PAGE(page_count(page), page); | |
5449 | ||
1306a85a | 5450 | if (!page->mem_cgroup) |
747db954 JW |
5451 | continue; |
5452 | ||
5453 | /* | |
5454 | * Nobody should be changing or seriously looking at | |
1306a85a | 5455 | * page->mem_cgroup at this point, we have fully |
29833315 | 5456 | * exclusive access to the page. |
747db954 JW |
5457 | */ |
5458 | ||
1306a85a | 5459 | if (memcg != page->mem_cgroup) { |
747db954 | 5460 | if (memcg) { |
18eca2e6 JW |
5461 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5462 | nr_huge, page); | |
5463 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5464 | } |
1306a85a | 5465 | memcg = page->mem_cgroup; |
747db954 JW |
5466 | } |
5467 | ||
5468 | if (PageTransHuge(page)) { | |
5469 | nr_pages <<= compound_order(page); | |
5470 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5471 | nr_huge += nr_pages; | |
5472 | } | |
5473 | ||
5474 | if (PageAnon(page)) | |
5475 | nr_anon += nr_pages; | |
5476 | else | |
5477 | nr_file += nr_pages; | |
5478 | ||
1306a85a | 5479 | page->mem_cgroup = NULL; |
747db954 JW |
5480 | |
5481 | pgpgout++; | |
5482 | } while (next != page_list); | |
5483 | ||
5484 | if (memcg) | |
18eca2e6 JW |
5485 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5486 | nr_huge, page); | |
747db954 JW |
5487 | } |
5488 | ||
0a31bc97 JW |
5489 | /** |
5490 | * mem_cgroup_uncharge - uncharge a page | |
5491 | * @page: page to uncharge | |
5492 | * | |
5493 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5494 | * mem_cgroup_commit_charge(). | |
5495 | */ | |
5496 | void mem_cgroup_uncharge(struct page *page) | |
5497 | { | |
0a31bc97 JW |
5498 | if (mem_cgroup_disabled()) |
5499 | return; | |
5500 | ||
747db954 | 5501 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5502 | if (!page->mem_cgroup) |
0a31bc97 JW |
5503 | return; |
5504 | ||
747db954 JW |
5505 | INIT_LIST_HEAD(&page->lru); |
5506 | uncharge_list(&page->lru); | |
5507 | } | |
0a31bc97 | 5508 | |
747db954 JW |
5509 | /** |
5510 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5511 | * @page_list: list of pages to uncharge | |
5512 | * | |
5513 | * Uncharge a list of pages previously charged with | |
5514 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5515 | */ | |
5516 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5517 | { | |
5518 | if (mem_cgroup_disabled()) | |
5519 | return; | |
0a31bc97 | 5520 | |
747db954 JW |
5521 | if (!list_empty(page_list)) |
5522 | uncharge_list(page_list); | |
0a31bc97 JW |
5523 | } |
5524 | ||
5525 | /** | |
6a93ca8f JW |
5526 | * mem_cgroup_migrate - charge a page's replacement |
5527 | * @oldpage: currently circulating page | |
5528 | * @newpage: replacement page | |
0a31bc97 | 5529 | * |
6a93ca8f JW |
5530 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
5531 | * be uncharged upon free. | |
0a31bc97 JW |
5532 | * |
5533 | * Both pages must be locked, @newpage->mapping must be set up. | |
5534 | */ | |
6a93ca8f | 5535 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5536 | { |
29833315 | 5537 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
5538 | unsigned int nr_pages; |
5539 | bool compound; | |
0a31bc97 JW |
5540 | |
5541 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5542 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5543 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5544 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5545 | newpage); | |
0a31bc97 JW |
5546 | |
5547 | if (mem_cgroup_disabled()) | |
5548 | return; | |
5549 | ||
5550 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5551 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5552 | return; |
5553 | ||
45637bab | 5554 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5555 | memcg = oldpage->mem_cgroup; |
29833315 | 5556 | if (!memcg) |
0a31bc97 JW |
5557 | return; |
5558 | ||
44b7a8d3 JW |
5559 | /* Force-charge the new page. The old one will be freed soon */ |
5560 | compound = PageTransHuge(newpage); | |
5561 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
5562 | ||
5563 | page_counter_charge(&memcg->memory, nr_pages); | |
5564 | if (do_memsw_account()) | |
5565 | page_counter_charge(&memcg->memsw, nr_pages); | |
5566 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 5567 | |
9cf7666a | 5568 | commit_charge(newpage, memcg, false); |
44b7a8d3 JW |
5569 | |
5570 | local_irq_disable(); | |
5571 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); | |
5572 | memcg_check_events(memcg, newpage); | |
5573 | local_irq_enable(); | |
0a31bc97 JW |
5574 | } |
5575 | ||
ef12947c | 5576 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5577 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5578 | ||
5579 | void sock_update_memcg(struct sock *sk) | |
5580 | { | |
5581 | struct mem_cgroup *memcg; | |
5582 | ||
5583 | /* Socket cloning can throw us here with sk_cgrp already | |
5584 | * filled. It won't however, necessarily happen from | |
5585 | * process context. So the test for root memcg given | |
5586 | * the current task's memcg won't help us in this case. | |
5587 | * | |
5588 | * Respecting the original socket's memcg is a better | |
5589 | * decision in this case. | |
5590 | */ | |
5591 | if (sk->sk_memcg) { | |
5592 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
5593 | css_get(&sk->sk_memcg->css); | |
5594 | return; | |
5595 | } | |
5596 | ||
5597 | rcu_read_lock(); | |
5598 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5599 | if (memcg == root_mem_cgroup) |
5600 | goto out; | |
0db15298 | 5601 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 5602 | goto out; |
f7e1cb6e | 5603 | if (css_tryget_online(&memcg->css)) |
11092087 | 5604 | sk->sk_memcg = memcg; |
f7e1cb6e | 5605 | out: |
11092087 JW |
5606 | rcu_read_unlock(); |
5607 | } | |
5608 | EXPORT_SYMBOL(sock_update_memcg); | |
5609 | ||
5610 | void sock_release_memcg(struct sock *sk) | |
5611 | { | |
5612 | WARN_ON(!sk->sk_memcg); | |
5613 | css_put(&sk->sk_memcg->css); | |
5614 | } | |
5615 | ||
5616 | /** | |
5617 | * mem_cgroup_charge_skmem - charge socket memory | |
5618 | * @memcg: memcg to charge | |
5619 | * @nr_pages: number of pages to charge | |
5620 | * | |
5621 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5622 | * @memcg's configured limit, %false if the charge had to be forced. | |
5623 | */ | |
5624 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5625 | { | |
f7e1cb6e | 5626 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5627 | |
f7e1cb6e | 5628 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5629 | struct page_counter *fail; |
f7e1cb6e | 5630 | |
0db15298 JW |
5631 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
5632 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
5633 | return true; |
5634 | } | |
0db15298 JW |
5635 | page_counter_charge(&memcg->tcpmem, nr_pages); |
5636 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 5637 | return false; |
11092087 | 5638 | } |
d886f4e4 | 5639 | |
f7e1cb6e JW |
5640 | /* Don't block in the packet receive path */ |
5641 | if (in_softirq()) | |
5642 | gfp_mask = GFP_NOWAIT; | |
5643 | ||
b2807f07 JW |
5644 | this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5645 | ||
f7e1cb6e JW |
5646 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
5647 | return true; | |
5648 | ||
5649 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5650 | return false; |
5651 | } | |
5652 | ||
5653 | /** | |
5654 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5655 | * @memcg - memcg to uncharge | |
5656 | * @nr_pages - number of pages to uncharge | |
5657 | */ | |
5658 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5659 | { | |
f7e1cb6e | 5660 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5661 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
5662 | return; |
5663 | } | |
d886f4e4 | 5664 | |
b2807f07 JW |
5665 | this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5666 | ||
f7e1cb6e JW |
5667 | page_counter_uncharge(&memcg->memory, nr_pages); |
5668 | css_put_many(&memcg->css, nr_pages); | |
11092087 JW |
5669 | } |
5670 | ||
f7e1cb6e JW |
5671 | static int __init cgroup_memory(char *s) |
5672 | { | |
5673 | char *token; | |
5674 | ||
5675 | while ((token = strsep(&s, ",")) != NULL) { | |
5676 | if (!*token) | |
5677 | continue; | |
5678 | if (!strcmp(token, "nosocket")) | |
5679 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5680 | if (!strcmp(token, "nokmem")) |
5681 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5682 | } |
5683 | return 0; | |
5684 | } | |
5685 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5686 | |
2d11085e | 5687 | /* |
1081312f MH |
5688 | * subsys_initcall() for memory controller. |
5689 | * | |
5690 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5691 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5692 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5693 | * be initialized from here. | |
2d11085e MH |
5694 | */ |
5695 | static int __init mem_cgroup_init(void) | |
5696 | { | |
95a045f6 JW |
5697 | int cpu, node; |
5698 | ||
2d11085e | 5699 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5700 | |
5701 | for_each_possible_cpu(cpu) | |
5702 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5703 | drain_local_stock); | |
5704 | ||
5705 | for_each_node(node) { | |
5706 | struct mem_cgroup_tree_per_node *rtpn; | |
5707 | int zone; | |
5708 | ||
5709 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5710 | node_online(node) ? node : NUMA_NO_NODE); | |
5711 | ||
5712 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5713 | struct mem_cgroup_tree_per_zone *rtpz; | |
5714 | ||
5715 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5716 | rtpz->rb_root = RB_ROOT; | |
5717 | spin_lock_init(&rtpz->lock); | |
5718 | } | |
5719 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5720 | } | |
5721 | ||
2d11085e MH |
5722 | return 0; |
5723 | } | |
5724 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5725 | |
5726 | #ifdef CONFIG_MEMCG_SWAP | |
5727 | /** | |
5728 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5729 | * @page: page whose memsw charge to transfer | |
5730 | * @entry: swap entry to move the charge to | |
5731 | * | |
5732 | * Transfer the memsw charge of @page to @entry. | |
5733 | */ | |
5734 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5735 | { | |
5736 | struct mem_cgroup *memcg; | |
5737 | unsigned short oldid; | |
5738 | ||
5739 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5740 | VM_BUG_ON_PAGE(page_count(page), page); | |
5741 | ||
7941d214 | 5742 | if (!do_memsw_account()) |
21afa38e JW |
5743 | return; |
5744 | ||
5745 | memcg = page->mem_cgroup; | |
5746 | ||
5747 | /* Readahead page, never charged */ | |
5748 | if (!memcg) | |
5749 | return; | |
5750 | ||
5751 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5752 | VM_BUG_ON_PAGE(oldid, page); | |
5753 | mem_cgroup_swap_statistics(memcg, true); | |
5754 | ||
5755 | page->mem_cgroup = NULL; | |
5756 | ||
5757 | if (!mem_cgroup_is_root(memcg)) | |
5758 | page_counter_uncharge(&memcg->memory, 1); | |
5759 | ||
ce9ce665 SAS |
5760 | /* |
5761 | * Interrupts should be disabled here because the caller holds the | |
5762 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5763 | * important here to have the interrupts disabled because it is the | |
5764 | * only synchronisation we have for udpating the per-CPU variables. | |
5765 | */ | |
5766 | VM_BUG_ON(!irqs_disabled()); | |
f627c2f5 | 5767 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
21afa38e JW |
5768 | memcg_check_events(memcg, page); |
5769 | } | |
5770 | ||
37e84351 VD |
5771 | /* |
5772 | * mem_cgroup_try_charge_swap - try charging a swap entry | |
5773 | * @page: page being added to swap | |
5774 | * @entry: swap entry to charge | |
5775 | * | |
5776 | * Try to charge @entry to the memcg that @page belongs to. | |
5777 | * | |
5778 | * Returns 0 on success, -ENOMEM on failure. | |
5779 | */ | |
5780 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
5781 | { | |
5782 | struct mem_cgroup *memcg; | |
5783 | struct page_counter *counter; | |
5784 | unsigned short oldid; | |
5785 | ||
5786 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
5787 | return 0; | |
5788 | ||
5789 | memcg = page->mem_cgroup; | |
5790 | ||
5791 | /* Readahead page, never charged */ | |
5792 | if (!memcg) | |
5793 | return 0; | |
5794 | ||
5795 | if (!mem_cgroup_is_root(memcg) && | |
5796 | !page_counter_try_charge(&memcg->swap, 1, &counter)) | |
5797 | return -ENOMEM; | |
5798 | ||
5799 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5800 | VM_BUG_ON_PAGE(oldid, page); | |
5801 | mem_cgroup_swap_statistics(memcg, true); | |
5802 | ||
5803 | css_get(&memcg->css); | |
5804 | return 0; | |
5805 | } | |
5806 | ||
21afa38e JW |
5807 | /** |
5808 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5809 | * @entry: swap entry to uncharge | |
5810 | * | |
37e84351 | 5811 | * Drop the swap charge associated with @entry. |
21afa38e JW |
5812 | */ |
5813 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5814 | { | |
5815 | struct mem_cgroup *memcg; | |
5816 | unsigned short id; | |
5817 | ||
37e84351 | 5818 | if (!do_swap_account) |
21afa38e JW |
5819 | return; |
5820 | ||
5821 | id = swap_cgroup_record(entry, 0); | |
5822 | rcu_read_lock(); | |
adbe427b | 5823 | memcg = mem_cgroup_from_id(id); |
21afa38e | 5824 | if (memcg) { |
37e84351 VD |
5825 | if (!mem_cgroup_is_root(memcg)) { |
5826 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5827 | page_counter_uncharge(&memcg->swap, 1); | |
5828 | else | |
5829 | page_counter_uncharge(&memcg->memsw, 1); | |
5830 | } | |
21afa38e JW |
5831 | mem_cgroup_swap_statistics(memcg, false); |
5832 | css_put(&memcg->css); | |
5833 | } | |
5834 | rcu_read_unlock(); | |
5835 | } | |
5836 | ||
d8b38438 VD |
5837 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
5838 | { | |
5839 | long nr_swap_pages = get_nr_swap_pages(); | |
5840 | ||
5841 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5842 | return nr_swap_pages; | |
5843 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5844 | nr_swap_pages = min_t(long, nr_swap_pages, | |
5845 | READ_ONCE(memcg->swap.limit) - | |
5846 | page_counter_read(&memcg->swap)); | |
5847 | return nr_swap_pages; | |
5848 | } | |
5849 | ||
5ccc5aba VD |
5850 | bool mem_cgroup_swap_full(struct page *page) |
5851 | { | |
5852 | struct mem_cgroup *memcg; | |
5853 | ||
5854 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
5855 | ||
5856 | if (vm_swap_full()) | |
5857 | return true; | |
5858 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5859 | return false; | |
5860 | ||
5861 | memcg = page->mem_cgroup; | |
5862 | if (!memcg) | |
5863 | return false; | |
5864 | ||
5865 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5866 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) | |
5867 | return true; | |
5868 | ||
5869 | return false; | |
5870 | } | |
5871 | ||
21afa38e JW |
5872 | /* for remember boot option*/ |
5873 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5874 | static int really_do_swap_account __initdata = 1; | |
5875 | #else | |
5876 | static int really_do_swap_account __initdata; | |
5877 | #endif | |
5878 | ||
5879 | static int __init enable_swap_account(char *s) | |
5880 | { | |
5881 | if (!strcmp(s, "1")) | |
5882 | really_do_swap_account = 1; | |
5883 | else if (!strcmp(s, "0")) | |
5884 | really_do_swap_account = 0; | |
5885 | return 1; | |
5886 | } | |
5887 | __setup("swapaccount=", enable_swap_account); | |
5888 | ||
37e84351 VD |
5889 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
5890 | struct cftype *cft) | |
5891 | { | |
5892 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5893 | ||
5894 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
5895 | } | |
5896 | ||
5897 | static int swap_max_show(struct seq_file *m, void *v) | |
5898 | { | |
5899 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5900 | unsigned long max = READ_ONCE(memcg->swap.limit); | |
5901 | ||
5902 | if (max == PAGE_COUNTER_MAX) | |
5903 | seq_puts(m, "max\n"); | |
5904 | else | |
5905 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5906 | ||
5907 | return 0; | |
5908 | } | |
5909 | ||
5910 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
5911 | char *buf, size_t nbytes, loff_t off) | |
5912 | { | |
5913 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5914 | unsigned long max; | |
5915 | int err; | |
5916 | ||
5917 | buf = strstrip(buf); | |
5918 | err = page_counter_memparse(buf, "max", &max); | |
5919 | if (err) | |
5920 | return err; | |
5921 | ||
5922 | mutex_lock(&memcg_limit_mutex); | |
5923 | err = page_counter_limit(&memcg->swap, max); | |
5924 | mutex_unlock(&memcg_limit_mutex); | |
5925 | if (err) | |
5926 | return err; | |
5927 | ||
5928 | return nbytes; | |
5929 | } | |
5930 | ||
5931 | static struct cftype swap_files[] = { | |
5932 | { | |
5933 | .name = "swap.current", | |
5934 | .flags = CFTYPE_NOT_ON_ROOT, | |
5935 | .read_u64 = swap_current_read, | |
5936 | }, | |
5937 | { | |
5938 | .name = "swap.max", | |
5939 | .flags = CFTYPE_NOT_ON_ROOT, | |
5940 | .seq_show = swap_max_show, | |
5941 | .write = swap_max_write, | |
5942 | }, | |
5943 | { } /* terminate */ | |
5944 | }; | |
5945 | ||
21afa38e JW |
5946 | static struct cftype memsw_cgroup_files[] = { |
5947 | { | |
5948 | .name = "memsw.usage_in_bytes", | |
5949 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5950 | .read_u64 = mem_cgroup_read_u64, | |
5951 | }, | |
5952 | { | |
5953 | .name = "memsw.max_usage_in_bytes", | |
5954 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5955 | .write = mem_cgroup_reset, | |
5956 | .read_u64 = mem_cgroup_read_u64, | |
5957 | }, | |
5958 | { | |
5959 | .name = "memsw.limit_in_bytes", | |
5960 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5961 | .write = mem_cgroup_write, | |
5962 | .read_u64 = mem_cgroup_read_u64, | |
5963 | }, | |
5964 | { | |
5965 | .name = "memsw.failcnt", | |
5966 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5967 | .write = mem_cgroup_reset, | |
5968 | .read_u64 = mem_cgroup_read_u64, | |
5969 | }, | |
5970 | { }, /* terminate */ | |
5971 | }; | |
5972 | ||
5973 | static int __init mem_cgroup_swap_init(void) | |
5974 | { | |
5975 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5976 | do_swap_account = 1; | |
37e84351 VD |
5977 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
5978 | swap_files)); | |
21afa38e JW |
5979 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5980 | memsw_cgroup_files)); | |
5981 | } | |
5982 | return 0; | |
5983 | } | |
5984 | subsys_initcall(mem_cgroup_swap_init); | |
5985 | ||
5986 | #endif /* CONFIG_MEMCG_SWAP */ |