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